Handbook of Biomedical Fluorescence

Handbook of Biomedical Fluorescence edited by Mary-Ann Mycek University of Michigan Ann Arbor, Michigan, U.S.A. Bria...

Author: Mary-Ann Mycek | Brian W. Pogue

196 downloads 2127 Views 30MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Handbook of

Biomedical Fluorescence edited by

Mary-Ann Mycek University of Michigan Ann Arbor, Michigan, U.S.A.

Brian W. Rogue Dartmouth College Hanover, New Hampshire, U.S.A.

MARCEL

MARCEL DEKKER, INC.

NEW YORK • BASEL

Library of Congress Cataloging-in-Publication Data A catalog record for this book is available from the Library of Congress. ISBN: 0-8247-0955-1 This book is printed on acid-free paper. Headquarters Marcel Dekker, Inc. 270 Madison Avenue, New York, NY 10016 tel: 212-696-9000; fax: 212-685-4540 Eastern Hemisphere Distribution Marcel Dekker AG Hutgasse 4, Postfach 812, CH-4001 Basel, Switzerland tel: 41-61-260-6300; fax: 41-61-260-6333 World Wide Web http://www.dekker.com The publisher offers discounts on this book when ordered in bulk quantities. For more information, write to Special Sales/Professional Marketing at the headquarters address above. Copyright © 2003 by Marcel Dekker, Inc. AH Rights Reserved. Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, microfilming, and recording, or by any information storage and retrieval system, without permission in writing from the publisher. Current printing (last digit): 10 9 8 7 6 5 4 3

2 1

PRINTED IN THE UNITED STATES OF AMERICA

Preface

The Handbook of Biomedical Fluorescence is designed to introduce, review, and highlight key basic scientific and clinical research developments within the last decade using fluorescence measurements to probe complex biological systems. The chapters present a detailed background into the basic biophysics, photochemistry, computational modeling, and biomedical applications of fluorescence, with an emphasis on the technical developments and applications that are pioneering in the field. The book fills the niche between basic biomolecular studies and clinical applications of fluorescence spectroscopy, explaining the basic understanding of the field at present and elaborating on several specific promising applications. The main focus is on how measurements from bulk tissues can be used for diagnosis in medicine. There is intentionally less emphasis on well-known in vitro cellular methods, such as basic microscopy or flow cytometry, as the primary goal is to examine where in vivo luminescence has made inroads into medicine. The book is divided into four sections covering: (I) the basic science describing the origin of fluorescence signals from biological media both from a photochemical view and with a computational modeling perspective, (II) the spectroscopic detection of fluorescence with microscopic spatial resolution, (III) methods using intrinsic biological fluorescence characteristics for disease detection in vivo, (IV) methods using exogenous agents for fluorescence and their biomedical applications. These four sections blend basic science with clinical science. This is an important pedagogical construct, which also serves to broaden the appeal of the book to basic science and clinical researchers, as well as to students. Hi

iv

Preface

In Part I, Chapter 1, the basic photophysical theory of fluorescence from molecules is outlined by Redmond. This theoretical approach has its roots in describing fluorescence in dilute nonscattering media, but can be applied to turbid media with the assistance of light transport modeling. The basics of three dominant approaches to light transport modeling in tissue are outlined in Chapters 2, 3, and 4, including diffusion theory, Monte Carlo modeling, and hybrid analytical/empirical approaches. In Chapter 2, Farrell and Patterson present a comprehensive review of fluorescence modeling from diffusion theory. In Chapter 3, Jacques discusses and presents a computational model, with computer code, for simulating fluorescent photon propagation in turbid media using the Monte Carlo approach. In Chapter 4, Georgakoudi et al. examine quantitative spectroscopy of intrinsic tissue fluorescence. Taken as a whole, Part I provides the basic building blocks to interpret fluorescence signals from bulk tissues. In Part II, microscopic measurements of fluorescence are examined as a way of exploring fluorescence signals which are minimally affected by the tissue light transport problem. Rajadhyaksha and Gonzalez describe pioneering work toward real-time confocal fluorescence microscopy in vivo in Chapter 5. In Chapter 6, So and colleagues discuss the increasingly important application and use of multiphoton microscopy for tissue studies. In Chapter 7, Urayama and Mycek describe the use of fluorescence lifetime imaging measurements, with an emphasis on applications to endogenous biological fluorescence in cells and tissues. Part III presents several key applications of endogenous fluorescence in clinical research starting off with a review of endogenous fluorophores by Richards-Kortum and colleagues in Chapter 8, followed in Chapter 9 by a review of the use of fluorescence to diagnose cervical dysplasia. The use of fluorescence in imaging and detection of skin cancer is reviewed by Zheng and MacAulay in Chapter 10. In Chapter 11, autofluorescence applications in lung cancer imaging and detection are reviewed by Wagnieres et al. provide a detailed description of an area at the leading edge of clinical fluorescence endoscopy. In Chapter 12, Marcu et al. discuss the use of timeresolved measurements for staging atherosclerotic lesions. In Part IV, key areas in the field of exogenous fluorophores are examined for their use in cancer research and diagnosis. Dewhirst and colleagues introduce the use of green fluorescence protein (GFP) for basic cancer tumor biology studies in Chapter 13. In Chapter 14, Sevick-Muraca et al. discuss the potential and current applications of fluorescence for imaging through deep tissues. Photodynamic therapy applications of fluorescence are presented by Wilson and colleagues in Chapter 15. In Chapter 16, Lange discusses applications of controlled drug delivery using photodynamic agents based on aminolevulinic acid. In Chapter 17, Wilson and Vinogradov

Preface

v

present phosphorescence measurements for monitoring oxygen as an example of monitoring physiologically relevant molecules in vivo. These five chapters highlight key areas of new developments and promising applications in the use of exogenous agent luminescence in medicine. Each of the four sections of the book is designed to build on the previous section, so that fundamental principles are presented first, followed by basic scientific research, and, finally, an examination of the leading clinical applications in the field. This approach allows the text to be accessible to readers with a wide range of expertise. The contributors are leading experts in the field, with the goal of reviewing their area of research along with the most important contributions of other researchers. The Handbook of Biomedical Fluorescence is meant for those working in such diverse fields as biology, chemistry, medicine, biomedical engineering, and physics. It is designed for research groups currently active in the field, for researchers in related disciplines, and for students in a biomedical optics graduate course. The book was supported in part by the National Science Foundation (BES-9977982, M.-A. M.), The Whitaker Foundation (M.-A.M.), and the National Institutes of Health (R01 CA78734 and PO1 CA84203 B.W.P.). Mary-Ann Mycek Brian W. Pogue

Contents

Preface Contributors

I.

Hi xi

Fluorescence in Biological Media: Theory and Simulation

1.

Introduction to Fluorescence and Photophysics Robert W. Redmond

1

2.

Diffusion Modeling of Fluorescence in Tissue Thomas J. Farrell and Michael S. Patterson

29

3.

Monte Carlo Simulations of Fluorescence in Turbid Media Steven L. Jacques

61

4.

Intrinsic Fluorescence Spectroscopy of Biological Tissue Irene Georgakoudi, Markus G. Miiller, and Michael S. Feld

109

II. 5.

Microspectrofluorimetry Real-Time In Vivo Confocal Fluorescence Microscopy Milind Rajadhyaksha and Salvador Gonzalez

143

VII

viii

Contents

6.

Two-Photon Microscopy of Tissues Peter T. C. So, Ki H. Kim, Lily Hsu, Chen Y. Dong, Peter Kaplan, Tom Hacewicz, Urs Greater, Nick Schlumpf, and Christof Buehler

7.

Fluorescence Lifetime Imaging Microscopy of Endogenous Biological Fluorescence Paul Uravama and Marv-Ann M\cek

III.

181

211

Endogenous Fluorescence Methods for In Vivo Disease Detection

8.

Survey of Endogenous Biological Fluorophores Rebecca Richards-Kortum, Rebekah Drezek, Konstantin Sokolov, Ina Pavlova, and Michele Pollen

237

9.

Cervical Dysplasia Diagnosis with Fluorescence Spectroscopy Rebecca Richards-Kortum, Rebekah Drezek, Karen Basen-Engquist, Scott B. Cantor, Urs Utzinger, Carrie Brookner, and Michele Pollen

265

10.

Fluorescence Spectroscopy and Imaging for Skin Cancer Detection and Evaluation Haishan Zeng and Calum MacAulay

11.

Lung Cancer Imaging with Fluorescence Endoscopy Georges Wagnieres, Annette McWil/iams, and Stephen Lam

12.

Time-Resolved Laser-Induced Fluorescence Spectroscopy for Staging Atherosclerotic Lesions Laura Marcu, Warren S. Grundfest, and Michael C. Fishbein

IV. 13.

315

361

397

Applications of Exogenous Fluorophores Applications of the Green Fluorescent Protein and Its Variants in Tumor Angiogenesis and Physiology Studies Chuan-Yuan Li, Yiting Cao, and Mark W. Devvhirst

431

Contents

ix

14.

Near-Infrared Imaging with Fluorescent Contrast Agents Eva M. Sevick-Muraca, Anuradha Godavarty, Jessica P. Houston, Alan B. Thompson, and Ranadhir Roy

445

15.

Fluorescence in Photodynamic Therapy Dosimetry Brian C. Wilson, Robert A. Weersink, and Lothar Lilge

529

16.

Controlled Drug Delivery in Photodynamic Therapy and Fluorescence-Based Diagnosis of Cancer Norbert Lange

563

Tissue Oxygen Measurements Using Phosphorescence Quenching David F. Wilson and Sergei A. Vinogradov

637

17.

Index

663

Contributors

Karen Basen-Engquist Department of Electrical and Computer Engineering, University of Texas at Austin, Austin, Texas, U.S.A. Carrie Brookner Department of Electrical and Computer Engineering, University of Texas at Austin, Austin, Texas, U.S.A. Christof Buehler

Paul Schiller Institut, Villigen, Switzerland

Scott B. Cantor Department of Electrical and Computer Engineering, University of Texas at Austin, Austin, Texas, U.S.A. Yiting Cao Department of Radiation Oncology, Duke University Medical Center, Durham, North Carolina, U.S.A. Mark W. Dewhirst Department of Radiation Oncology, Duke University Medical Center, Durham, North Carolina, U.S.A. Chen Y. Dong. pei, Taiwan

Department of Physics, National Taiwan University, Tai-

Rebekah Drezek Department of Electrical and Computer Engineering, University of Texas at Austin, Austin, Texas, U.S.A. Thomas J. Farrell Canada

Hamilton Regional Cancer Center, Hamilton, Ontario, xi

xii

Contributors

Michael S. Feld G. R. Harrison Spectroscopy Laboratory, Massachusetts Institute of Technology, Cambridge, Massachusetts, U.S.A. Michael C. Fishbein Department of Pathology and Laboratory Medicine, University of California at Los Angeles School of Medicine, Los Angeles, California, U.S.A. Michele Follen Department of Biomedical Engineering, University of Texas at Austin, Austin, Texas, U.S.A. Irene Georgakoudi G. R. Harrison Spectroscopy Laboratory, Massachusetts Institute of Technology, Cambridge, Massachusetts, U.S.A. Anuradha Godavarty The Photon Migration Laboratory, Texas A&M University, College Station, Texas, U.S.A. Salvador Gonzalez Harvard Medical School and Massachusetts General Hospital, Boston, Massachusetts, U.S.A. Urs Greuter


Warren S. Grundfest Biomedical Engineering Interdepartamental Program, University of California at Los Angeles, Los Angeles, California, U.S.A. Tom Hacewicz U.S.A.

Unilever Edgewater Laboratory, Edgewater, New Jersey,

Jessica P. Houston The Photon Migration Laboratory, Texas A&M University, College Station, Texas, U.S.A. Lily Hsu Department of Mechanical Engineering and Division of Biological Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts, U.S.A. Steven L. Jacques Oregon, U.S.A. Peter Kaplan U.S.A.

Oregon Health and Science University, Portland,

Unilever Edgewater Laboratory, Edgewater, New Jersey,

Contributors

xiii

Ki H. Kim Department of Mechanical Engineering and Division of Biological Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts, U.S.A. Stephen Lam Cancer Imaging Department, British Columbia Cancer Agency, Vancouver, British Columbia, Canada Norbert Lange Ecole de Pharmacie, Institut de Biopharmacie et Galenique, Universite de Lausanne, Lausanne, Switzerland Chuan-Yuan Li Department of Radiation Oncology, Duke University Medical Center, Durham, North Carolina, U.S.A. Lothar Lilge University of Toronto and Photonics Research Ontario, Toronto, Ontario, Canada Calum MacAulay Cancer Imaging Department, British Columbia Cancer Agency, Vancouver, British Columbia, Canada Laura Marcu Laser Research and Technology Development, Cedars-Sinai Medical Center and University of Southern California, Los Angeles, California, U.S.A. Annette McWilliams Cancer Imaging Department, British Columbia Cancer Agency, Vancouver, British Columbia, Canada Markus G. Miiller G. R. Harrison Spectroscopy Laboratory, Massachusetts Institute of Technology, Cambridge, Massachusetts, U.S.A. and Kirchhoff-Institut fur Physik, Universitat Heidelberg, Heidelberg, Germany Mary-Ann Mycek Department of Biomedical Engineering, University of Michigan, Ann Arbor, Michigan, U.S.A. Michael S. Patterson tario, Canada

Hamilton Regional Cancer Center, Hamilton, On-

Ina Pavlova Department of Biomedical Engineering, University of Texas at Austin, Austin, Texas, U.S.A. Milind Rajadhyaksha Center for Subsurface Imaging and Systems, Northeastern University, Boston, Massachusetts and Department of Medi-

xiv

Contributors

cine, Memorial Sloan-Kettering Cancer Center, New York, New York, U.S.A. Robert W. Redmond Wellman Lab of Photomedicine, Massachusetts General Hospital, Boston, Massachusetts, U.S.A. Rebecca Richards-Kortum Department of Electrical and Computer Engineering, University of Texas at Austin, Austin, Texas, U.S.A. Ranadhir Roy The Photon Migration Laboratory, Texas A&M University, College Station, Texas, U.S.A. Nick Schlumpf


Eva M. Sevick-Muraca The Photon Migration Laboratory, Texas A&M University, College Station, Texas, U.S.A. Peter T. C. So Department of Mechanical Engineering and Division of Biological Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts, U.S.A. Konstantin Sokolov Department of Electrical and Computer Engineering, University of Texas at Austin, Austin, Texas, U.S.A. Alan B. Thompson The Photon Migration Laboratory, Texas A&M University, College Station, Texas, U.S.A. Paul Urayama Department of Biomedical Engineering, University of Michigan, Ann Arbor, Michigan, U.S.A. Urs Utzinger Department of Electrical and Computer Engineering, University of Texas at Austin, Austin, Texas, U.S.A. Sergei A. Vinogradov Department of Biochemistry and Biophysics, University of Pennsylvania, Philadelphia, Pennsylvania, U.S.A. Georges Wagnieres LPAS/ENAC, Swiss Federal Institute of Technology, Lausanne, Switzerland Robert A. Weersink Canada

Photonics Research Ontario, Toronto, Ontario,

Contributors

xv

Brian C. Wilson Ontario Cancer Institute, University of Toronto, and Photonics Research Ontario, Toronto, Ontario, Canada David F. Wilson Department of Biochemistry and Biophysics, University of Pennsylvania, Philadelphia, Pennsylvania, U.S.A. Haishan Zeng Cancer Imaging Department, British Columbia Cancer Agency, Vancouver, British Columbia, Canada

Handbook of

Biomedical Fluorescence

1 Introduction to Fluorescence and Photophysics Robert W. Redmond Massachusetts General Hospital, Boston, Massachusetts, U.S.A.

1.

ABSORPTION AND FLUORESCENCE

Emission of light in the form of fluorescence often accompanies deactivation of an electronically excited species. Fluorescence is defined as the radiative transition between two electronic states of the same spin multiplicity. Most organic molecules have "paired" electrons in their ground state molecular orbital configuration. The spins are balanced (e.g., s, = +5 and s2 = —5, S = £s = 0) and the spin multiplicity (Ms = 2S + 1 = 1) is singlet. Alternatively, inversion of the spin of the excited electron results in the two unpaired electrons having the same spin orientation. The overall spin S is 1 (s, = +i, s2 = +5), the spin multiplicity (Ms = 2S + 1) is 3, and a triplet state results. Most commonly, fluorescence refers to singlet-singlet transitions, especially the transition between the lowest, or first, excited singlet state (SO and the ground state (S0). Other types of less common fluorescence processes do occur, such as that from the second (S2) excited singlet state, and the doublet-doublet fluorescence exhibited in the radiative relaxation between excited and ground state free radicals (one unpaired electron, S = 5, Ms = 2S 4- 1 = 2). However, this introduction will focus on S,-S0 fluorescence, which is by far the most common type.

2

Redmond

The relationship between absorbance and fluorescence can be illustrated using simple potential energy diagrams of the type shown in Fig. 1 that show the relative electronic and vibrational energy levels as a function of internuclear separation in the affected bond. The absorption process involves interaction of the molecule in the ground state with a photon to promote an electron from a lower energy to a higher energy molecular orbital. The absorbance (A) of a sample is proportional to the concentration (c, in molarity M; Beer's law) of absorbing species in a sample of pathlength € (cm) traversed by the light and is generally independent of the intensity of the excitation light (Lambert's law), although the latter may not hold under high-intensity laser irradiation. In transparent media the pathlength is simply the thickness of the sample, but it is more complex to determine in opaque or highly scattering materials. This results in the common expression of the Beer-Lambert law shown in Eq. (1). The molar absorption coefficient (e in M ' cm ') is the proportionality factor and its magnitude reflects the probability of the absorption of a photon of a given energy by the molecule. The absorption spectrum of a compound is constructed by plotting e as a function of excitation wavelength.

Energy

nuclear configuration FIGURE 1 Potential energy diagram showing absorption and emission transitions between vibrational sublevels in ground and electronically excited states.

Introduction to Fluorescence and Photophysics

A = sc€

3

(1)

The absorption process takes place on a time scale (-1CT15 sec) much faster than that of molecular vibration; thus, absorption occurs in a "vertical" manner, i.e., the internuclear geometry will be identical immediately before and after absorption to form the excited state. This is the Franck-Condon principle. In the excited state, the electron is promoted to an antibonding orbital such that the bond order is reduced, the atoms in the bond are less tightly held, and the equilibrium bond length is subsequently longer. This is shown as a displacement to the right of the excited state potential curve with respect to the ground state in Fig. 1. The vibrational level that is initially populated is that where vertical overlap at the energy of the absorbed photon occurs and is generally v' > 0. In Fig. 1 the absorption is shown to the v' = 3 level of the S, state. The simplest fluorescence in terms of photophysics would be the exact reverse of the absorption process, emitting light of a wavelength identical to that absorbed. This is termed "resonance" fluorescence. In Fig. 1 this would correspond to a transition from the v' = 3 level of the S, state back to the v = 0 level of the S0 state. Such fluorescence can be observed from atoms or molecular gases at very low pressures but is not usually apparent for larger molecules in condensed phases such as liquids and solids. This is due to the fact that vibrational deactivation, through intermolecular collisions, occurs more rapidly than the fluorescence emission process, with the result that fluorescence generally occurs from the lowest vibrational level (v' =0) of the electronic excited state. The transition that is associated with the emission of a photon is also so rapid that no change in nuclear configuration can occur during the process. The Franck-Condon principle again dictates that the vibrational level that is initially populated in the electronic ground state will be that which shows "vertical" overlap with the v' = 0 level of the S, state. This would be the v = 1 level in the example shown in Fig. 1. Thus, the energy of the emitted photon will be significantly lower than the absorbed photon and the fluorescence is red shifted with respect to absorption. Several other points are worth noting from this diagram. The energy level spacings between adjacent vibrational levels decreases with increasing energy, and a similar spacing of vibrational levels is often seen in the ground (S0) and excited (S,) states. This results in the "mirror image" relationship commonly observed between absorption and emission spectra. This is shown in more detail in Fig. 2. The electronic excited state energy of S, can be obtained from the transition between the v = 0 levels in both states and is termed the 0-0 transition. Due to slight changes in internuclear distances at equilibrium in both states, the 0-0 transition energy (E0.0) is often not identical in absorbance and fluorescence but can be estimated from the wave-

Redmond

v'=5 v'=4 v'=3

v'=l v'=0

v=5 v=3

v=l v=0

FIGURE 2 Possible absorption and emission transitions between vibrational levels in ground and excited states. Emission occurs at lower energy and is red shifted with respect to absorption bands.

length at which these two spectra overlap. This difference is also termed the Stokes shift. The Stokes shift and E00 estimation are shown in Fig. 3 for the example of the photodynamic therapy (PDT) agent, benzoporphyrin derivative monoacid ring A (BPDMA).

2.

DEACTIVATION OF THE ST STATE

Fluorescence is only one of the possible mechanisms by which an excited molecule can undergo relaxation to the ground state. The Jablonski diagram in Fig. 4 shows that there are a number of potential transitions open to the S, state after population by excitation and internal conversion from upper states. Here the formalism of a solid arrow depicting a radiative transition and a wavy arrow depicting a nonradiative transition has been followed. The molecule can undergo both nonradiative (internal conversion, ic) and radiative (fluorescence) relaxation to the ground state (S,,) or nonradiative transition (intersystem crossing, isc) to the lowest excited triplet state (T,).


absorption fluorescence

660

680

700

720

740

wavelength (ran) FIGURE 3 Lowest energy absorption band (So-S^ full line) and normalized fluorescence emission band (S^So, dashed line) for benzoporphyrin derivative monoacid ring in methanol. The energy of the transition between zero vibrational levels in each state (E0.0) is estimated from the intersection of the spectra. The difference in wavelength maxima corresponds to the Stokes shift.

2.1

Internal Conversion

The energy separation between consecutive singlet levels (S0, S,, S2, . . . Sn) tends to decrease with increasing electronic energy. Generally, the rate of radiationless transition from one state to another is inversely proportional to the energy separation (the "energy gap law") and nonradiative transitions between upper states (Sn to S n _,) occur rapidly to populate the lowest excited state, S,. These types of transitions are denoted internal conversion as they occur between states of the same spin multiplicity. For this reason, the fluorescence emission spectrum is typically independent of the excitation wavelength, e.g., a molecule that is directly excited into the S3 state by absorption of a higher energy photon will relax by internal conversion to the S, state before emission can occur. This is an example of Kasha's generalization that

Redmond

Energy

phosphorescence

FIGURE 4 Modified Jablonski diagram showing radiative (solid arrows) and nonradiative (wavy arrows) transitions between ground state (S0) and excited singlet (S2, ST) and triplet (T-,) states, ic, internal conversion; isc, intersystem crossing.

radiative processes or excited state reactions arise from the lowest electronically excited states (S, or T,). The energy gap between S, and S(l is considerably larger than between other adjacent states. Thus, the S, state lifetime is much longer than the higher singlet states, and radiative emission can effectively compete with nonradiative processes from this level. Only in atypical molecules like azulene, which has a large S 2 -S, gap, can fluorescence be observed from upper states. In fact, the large energy gap between S, and S() states is often sufficient to completely inhibit the internal conversion process between these states and gives rise to Ermolaev's generalization, expressed as

cj>, 4- <J>- =1 ^f

|

'

^tsc

l

(2}

\*~ I

where $,• and <J>isc are the quantum yields of fluorescence and intersystem crossing, respectively.


2.2

7

Intersystem Crossing

Selection rules dictate that transitions between states of like spin multiplicity (e.g., singlet-singlet) are allowed and will be much more probable that transitions between states of different multiplicity (e.g., singlet-triplet). Intersystem crossing (isc) is defined as a nonradiative transition between states of different multiplicity and occurs via inversion of the spin of the excited electron resulting in the two unpaired electrons having the same spin orientation. The overall spin S is 1 and the spin multiplicity is 3 to give a triplet state. Although transitions between states of different multiplicity are formally forbidden, these processes are often evident as mechanisms exist to relax this selection rule and allow intersystem crossing to occur. These include spin-orbit and vibronic coupling mechanisms that lessen the "pure" character of the initial and final states. One additional possibility is that photochemical reaction, rather than relaxation, may occur from the S, state. Thus, the efficiency of the fluorescence pathway depends on the value of the rate constant for fluorescence emission in relation to the rate constants for the other possible processes from S,. Direct absorption into the triplet state from the ground state singlet (S0) is spin forbidden, but the triplet excited state can be accessed indirectly through intersystem crossing from S, to T,. Intersystem crossing is also formally spin forbidden; the excited states in large organic molecules are not well described in terms of pure spin multiplicities but can be described in terms of mixed singlet and triplet state character. This lowers the importance of the spin selection rule. The reverse intersystem crossing (isc') from T| to S() is also formally spin forbidden. 3.

EXCITED STATE CONFIGURATION

At this point it is important to consider the nature of these transitions with respect to the types of electrons involved. Electronic excitation involves the promotion of an electron from a lower to higher energy orbital. These electrons can be of three basic types: a bonding electron originating from a single bond (denoted cr), a bonding electron originating from a double bond (denoted TT), and a nonbonding electron associated with an atom (denoted n). Thus, excited states can be characterized by the transitions involved, e.g. 7777-* signifies promotion of a TT electron from a bonding to antibonding orbital. Similarly, UTT* donates promotion of a nonbonding electron to an antibonding orbital. These two configurations are the most relevant for biological studies involving ultraviolet (UV) and visible light. The crcr* transition is a high-energy transition that requires very short UV wavelengths for initiation and does not have a role in biological photochemistry.

8

3.1

Redmond

Differences Between nir* and TTTT* States

The photophysical behavior of HTT* and 7777* states is quite distinct, often allowing straightforward characterization of state character. Of course, pure hydrocarbons do not have the possibility of nvr* states as they lack atoms with nonbonding electrons, such as O, N, S, etc. Molecules containing carbonyl groups (—C=O) are the most widely studied examples where nvr* states are important. Excited states of mr* and 7777* configuration typically differ in the following ways: • • •

Fluorescence is more probable from 7777* singlet states. Intersystem crossing to the triplet state is higher for mr* singlet states. Singlet and triplet lifetimes are longer for 7777* states.

These effects arise from a combination of the energy gap law, and El Sayed's rule that transitions involving a change of multiplicity between states of the same configuration are less probable. Thus, intersystem crossing between two 7777* or two n77* states is forbidden, but a transition between a n77* and a 7777* state is allowed. This is due to the fact that spin-orbit coupling between states of the same configuration cannot occur. For pure hydrocarbons there are obviously no n77* states available and singlet-triplet intersystem crossing is less efficient, with the result that fluorescence emission can be a significant process for molecular deactivation in 7777* states. This can be illustrated by the case of naphthalene in Fig. 5. Intersystem crossing in naphthalene takes place from the S, (7777*) state to a T, (7777*) state that is separated by a large energy gap of about 30 kcal/mol. The combination of a large energy gap and a transition between two 7777* states results in a low value for the rate constant for intersystem crossing (kise) of 106 sec ' and considerable fluorescence emission. Figure 5 also shows the situation for biacetyl, where the process occurs between two n77* states separated by an energy gap of only 6 kcal/mol. The process also occurs between two states of the same configuration but the energy gap is now much smaller. As a result, the rate constant (k isc ) is correspondingly higher than for naphthalene at about 7 X I07 sec ' and the fluorescence yield is negligible. Intersystem crossing is much more probable between n77* and 7777* states, according to El Sayed's rule, and the process is also favored by the typically smaller energy gap between Si and T, states in these compounds. Figure 6 shows the very favorable case of benzophenone. At first glance one would consider intersystem crossing to be similar to the biacetyl case, as both S, and T, states have 1177* configuration and a similar overall S,-T, energy gap. However, benzophenone has a T2 state of 7777* configuration


-

#^-0? "•••"•"""•>• k = lO^s-' S, (nit*)

*~1.0 ks?= 7x10' a-'

T, din")

<S>r=0.24

Aa:S,-T,)=7kcal'mol

Naphthalene

Biacetyl

FIGURE 5 Energy level diagram showing Sri", intersystem crossing process in naphthalene and biacetyl. Rates of intersystem crossing (kisc) and quantum yield of intersystem crossing (îsJ are dependent on the energy gap between ST and T-, levels (^(SI-TT)) and are higher for biacetyl as a result of the lower energy gap.

k. .^

5 kca]/mol

Benzophenone FIGURE 6 Energy level diagram showing S^T, intersystem crossing process in benzophenone. High values are exhibited for kisc and 4>isc due to small energy gap between states and different electronic configuration of the states involved.

10

Redmond

that is almost isoenergetic with the mr* S, state. Intersystem crossing takes place to the T2 state and is highly favored due to the very small energy gap and the different configurations of the states involved. The rate constant for intersystem crossing is about 10" sec ' in benzophenone. The T2 state then rapidly undergoes internal conversion to the nearby T, (HTT*) state. El-Sayed's rule also impacts the triplet state relaxation process where the lifetimes of nrr* triplet states are typically shorter than TTTT* triplets. Although the T, levels in mr* states are often at higher energy than TTTT* T, states, the rate constant for the T, to S,, intersystem crossing is generally greater for n-rr* triplets due to the influence of El-Sayed's rule and the increased spin-orbit coupling that relaxes the selection rules pertaining to radiationless transitions between states of different spin multiplicities. Thus, the triplet lifetime of naphthalene is in the millisecond range whereas that of benzophenone is only a few tens of microseconds. 4.

MEASUREMENT OF FLUORESCENCE

The basic measurement of fluorescence requires an light source that matches the absorption spectrum of the molecule under study and a detector that monitors the emitted fluorescence. These measurements can be carried out in a spectral or time-resolved manner. In the former, both fluorescence emission and excitation spectra are often measured. 4.1 4.1.1

Fluorescence Spectra Emission Spectra

A fluorescence emission spectrum is a plot of the magnitude of the emitted fluorescence as a function of its wavelength. To obtain a fluorescence emission spectrum the excitation light is of a fixed wavelength, often at a peak in the absorption spectrum of the sample, and the emitted fluorescence is scanned as a function of detection wavelength. In older fluorimeters the excitation source is generally a broad-band lamp and the excitation wavelength is selected by a monochromator (the excitation monochromator) between the lamp and the sample. The detection is normally performed perpendicular to the excitation beam to minimize stray excitation photons impinging the sensitive emission detector. In many biological or medical applications the apparatus is simplified by using a monochromatic laser as excitation source. Similarly, the emission wavelength is also selected using a second monochromator between sample and detector. This emission monochromator is scanned across the wavelength range of the emission to construct the spectrum. The detector in these instruments is typically a sensitive photomultiplier tube (PMT). In many modern devices the spectrum can be


11

obtained more rapidly using array detectors that provide the entire spectrum in a single measurement. The emitted light is first passed through a spectral dispersion optic and imaged on the detection array, with each element in the array detecting light of a different wavelength. 4.1.2

Excitation Spectra

The fluorescence excitation spectrum is generated under the reverse conditions, i.e., the monitored emission wavelength is held constant, usually at the emission maximum, and the excitation source is varied across the wavelength range that covers the absorption spectrum of the compound. As the signal intensity in the excitation spectrum reflects the variation in the photons absorbed as a function of wavelength, it should be identical to the absorption spectrum of the molecule responsible for emitting the fluorescence. Thus, a match of absorption and excitation spectra of a given sample confirms the presence of one and the same absorbing and fluorescing compound. The fluorescence excitation spectrum is particularly valuable in analyzing fluorescence arising from a mixture of absorbing species. An example of the relationship between absorption, fluorescence emission and excitation spectra is given in Fig. 7. The example shown is for the tetrapyrrolic compound deuteroporphyrin (DP), where a number of features can be identified. The absorption and fluorescence excitation spectra are always to the high-energy (blue) side of the emission spectrum. The reason for this is evident from the Jablonski diagram in Fig. 4, showing that fluorescence typically arises from the lowest vibrational level (v = 0) of the lowest electronically excited singlet state (S,). Thus, absorption of higher energy photons results in some energy loss through nonradiative processes (internal conversion and vibrational deactivation) prior to emission. The emitted photons are of lower energy than the absorbed photons, and the fluorescence emission spectrum is red shifted in comparison with the absorption spectrum. The absorption and fluorescence excitation spectra are very similar, demonstrating that the fluorescent species is the same as the absorbing species, which is expected for a pure compound in solution. A different situation is shown in Fig. 8 for DP in aqueous solution (PBS). In this case the fluorescence excitation and absorption spectra do not match. The fluorescence excitation and emission spectra are similar to those obtained in organic solvent, as shown in Fig. 7. The reason for this difference is that in water the DP exhibits self-association, and the aggregates that are formed have a different absorption spectrum and are nonfluorescent. This is evident from the broader nature of the Soret band in the near-UV and the presence of a shoulder on the blue side of the band where the aggregated species absorb. Fluorescence arises only from the fraction of the DP that is monomeric in solution, and the excitation spectrum only reports on this

Redmond

12

?50

400

450

500

550

600

650

700

750

wavelength (nm)

FIGURE 7 Absorption (full line), fluorescence emission (symbols), and fluorescence excitation spectra (dashed line) for 10 /xM deuteroporphyrin (DP) in methanol. Note the excellent spectral correlation between absorption and fluorescence excitation spectra.

fraction. In organic solution, the DP is monomeric; thus, the excitation and emission spectra in Figs. 7 and 8 are due to the same species. In biological or medical scenarios the samples interrogated are not necessarily transparent and fluorescence that comes from surface or subsurface fluorophores is detected. The apparatus used for these types of measurement can be simplified in many ways, and a typical device for such a measurement may use fiberoptic coupled excitation and emission delivery and a diode array detector for nonscanning spectral measurements. Thus, an emission spectrum can be generated from a single excitation event, e.g., a nanosecond laser pulse. Such techniques are widespread in the field of tissue fluorescence research and are described in detail elsewhere in this book. 4.2

Quantum Yield of Fluorescence, <E>f

In photochemistry or photophysics, the quantum yield of any given process is defined as the number of molecules that undergo the process divided by the total number of molecules excited. The quantum yield of fluorescence is therefore defined as follows:


13

,0 S3

0.2

0.1

0• 350

400

450

500

550

600

650

700

750

wavelength (nm) FIGURE 8 Absorption (full line), fluorescence emission (symbols), and fluorescence excitation spectra (dashed line) for 10 ^M deuteroporphyrin (DP) in PBS. Difference in absorption and fluorescence excitation spectra is due to the presence of absorbing but nonfluorescent aggregated species in PBS.

number of fluorescence photons emitted number of photons absorbed

(3)

In practice, however, it is tedious and complex to measure these values in an absolute sense and fluorescence quantum yields are typically determined by comparative actinometry using reference compounds for which 3>f has been previously determined with a high degree of accuracy. In such cases, the measurement is generally carried out using optically matched conditions such that both unknown and standard have the same absorbance at their excitation wavelength, i.e., the number of photons absorbed by both solutions is identical and the denominator in Eq. (3) is identical. Thus, measurement of the integrated fluorescence intensities (If) for both unknown (u) and standard (s) allows determination of the f of the unknown. For increased accuracy the integrated fluorescence emission is generally measured for a number of samples of varying and low absorbance (A • 0.05) for both unknown and standard. The slopes (Sf) of the plots of It- vs. A are then substituted into Eq. (4) below for the determination of <E>f. An example is

Redmond

14

shown in Fig. 9 where the , of BPDMA in methanol was determined to be 0.11 relative to a reference of merocyanine 540 (MC540, p(r, t)

(9)

The primary fluence rate is attenuated as light penetrates the tissue, and this can be described exactly using Beer's law as given below, in which s is the distance along the primary ray path into the tissue: dOp(r, t) = -/At'4>p(r, t) ds

(10)

Thus, a light beam incident on the tissue gives rise to a diffuse source distribution that diminishes in intensity exponentially with pathlength into the tissue. In some instances it is accurate and convenient to approximate this extended diffuse source distribution as a single point source at a distance of one transport mean free path (l//A t ') from the entrance to the tissue. The source of fluorescence photons is fluorophore molecules that have absorbed either primary or diffuse excitation light and have relaxed to the ground state via fluorescence emission. The rate of excitation of fluorophore molecules per unit volume is the product of the fluorophore absorption coefficient and the excitation fluence rate. The excited state lifetime may be comparable to the time over which the diffuse excitation fluence rate changes. The probability that a fluorescence photon will be emitted at a time t' following absorption is usually written as Y P = -exp~ t7T

(11)

T

where Y is the quantum yield for fluorescence emission and • is the excited state lifetime. The fluorescence source strength is obtained by integrating the product of the probability of fluorescence emission and fluorophore absorption over time: t

(4>p(r, t') + m (r, t)

c

at

(13a)

= V • D m V3> m (r, t) - /Lta.m$m(r, t)

(z, x

c

Dx

d

. 0 - /4,x$x(z, t) + A<x3>p,x(z, t)

dz

m

~ dzD"1 dz

c

r

Y T

a,d,X

o

41

(z

t') +

p

t)

_ ^

(Z,

(23a)

t)

3>x(z, t'))e- t>/r dt'

(23b)

where z is the depth into the tissue. 2.5.3 Time-Independent Diffusion Equation If the externally applied excitation light does not change with time, then the excitation light fluence rate as well as the fluorescence fluence rate will be time independent. Under these conditions, the diffusion equation is simplified by setting the time derivative of the energy fluence rate to zero. In addition, the lifetime of the excited state of the fluorophore can be ignored, and the time-independent diffusion equations can be written more simply as

2.5.4

V-D x VO x (r) - Aia,xOx(r) = -/xs'.xO>p.x(r)

(24a)

V-D m Vd> m (r) - Ma,mm(r) = -Yâ,d,x[^>p,x(r) + , then the excitation and fluorescence energy radiance field will have a sinusoidal time dependence and can be written as the product of spatially and time-dependent terms. Ox(r, t) = , r)eia)t

(25)

The diffusion equation for the excitation energy fluence becomes a function only of the spatial position: V-D x VO x (w, r) - (Ai a x - — J , r) - ( â,m - — I 4>*(x(w, r)] V C ' (21)

42

Farrell and Patterson

The fluorescence lifetime is easily incorporated by multiplying by the Fourier transform of the function l/rexp( — t/r). The fluorescence energy fluence rate is given as *(co,

1 —

r) -

/2-77COT

;

(28)

As a final note, the time-dependent diffusion equation can be solved by taking the Fourier transform of the solution (ct>, r) for all to. 3.

CLOSED-FORM ANALYTICAL SOLUTIONS

As stated earlier, there are only a few cases in which analytical solutions for the coupled diffusion equations exist. In this section we will present the solutions for the most important cases. 3.1

Time-Independent, One-Dimensional, Broad-Beam, Homogeneous-Medium Case

When the excitation light is at a very short wavelength, such as would be used for UV excitation fluorescence measurements, the excitation energy fluence can be described using the Beer-Lambert law. The source term for the fluorescence decreases exponentially with pathlength into tissue. If the tissue is homogeneous, and the light is broad beam and steady state, then we have achieved the simplest possible formulation of the diffusion equation for fluorescence energy fluence. D,n

d24> (z) '", - Mum3> m (z) = -PY/Aê"*-' ~

(29)

In this equation, P is the irradiance incident on the tissue surface, z is depth in the tissue, and the surface is assumed to be at z = 0. The solution of this differential equation is the sum of two exponential functions 4> m (z) = Ae Mm and B into Eq. (19) — the boundary condition — at the surface. As an example, if the relative refractive index, n reh is 1.00, the boundary condition is

Diffusion Modeling of Fluorescence in Tissue

43

3>(r, t) = 2DVO>(r, t)

(32)

and the following expression is derived: A + B = -2Dm(/XetY,mA + /< X B)

(33)

which can be used to find factor A. A = -B

1 + 2 m x 1 + 2 ^ ^'-

(34)

At longer wavelengths, where scattering is significant, the excitation light is the solution of the diffusion equation. d2Ox(z)

^~^ ~ M-ax^xC2) = "~AAsxPe~ Mcx/

DX

(35)

dz2

This is solved as above, i.e., Ox(z) = Ae~Mc"-*z + Be"*'"* 1 + 2Dx/xt'x

(36a)

_ JJL'^P

2D x /t e f f , x '

1

(36b)

Dx

with factors A and B determined as above. The diffusion equation for the fluorescence energy fluence is Dm

d2

(45)

where <J> xm is the solution for a point source at the emission wavelength. A relationship similar to this has been derived by others using heuristic arguments and by making the assumption that the scattering coefficients at the excitation and emission wavelengths are equal [12]. In that case, D^Dm and

Farrell and Patterson

46

(46)

The diffusion theory solution for the excitation and fluorescence energy fluence rate for a point source are shown in Fig. 4. Both results are normalized to unit point source power and are compared with data derived from a Monte Carlo simulation. Diffusion theory predicts the energy fluence rate very well at distances greater than 1 mm for both excitation and fluorescence but is much less accurate at closer distances. This is a consequence of the fact that at short distances the radiance rate is not linearly anisotropic and the basic assumptions required for the diffusion approximation to the Boltzman transport equation are not satisfied. 3.3

Frequency Domain Point Source Solution

The result for a steady-state point source is easily extended to a sinusoidally modulated point source. In this case, we want solutions to the diffusion 10

Excitation Energy Fluence Rate

0.01 Fluorescence Energy Fluence Rate

0.001 4

6

10

Depth (mm) FIGURE 4 Plots of excitation and fluorescence energy fluence rate as functions of radial distance for point source steady-state excitation irradiation in an infinite medium. Diffusion theory results are compared with Monte Carlo for the excitation energy fluence rate and the fluorescence energy fluence rate. Tissue optical properties in mm^ are /4,x = 1.4, /ia.x = 0.015, /xa,x,d = 0.005, /4,m = 1.0, /ia,m = 0.009, â,m,d = 0.001. The fluorescence quantum yield is 1. Diffusion theory and Monte Carlo are plotted as solid and dashed lines, respectively.

Diffusion Modeling of Fluorescence in Tissue

47

equation as written in Eq. (26). For the excitation energy fluence rate, the solution is

, p) =

, where

KX = \ -

-- - )

(47)

The energy fluence rate at the emission wavelength is similar to the steadystate solution: II

YP

1

/ (*~KmP

n Dx (KX - /c m ) \

p

f»" K xP\

p

(48)

and it can also be written as the sum of the two point source solutions:

As mentioned earlier, fluorescence lifetime can be incorporated into this solution by multiplication by the factor [1 — /2muT]/[l + (27TWT)2]. One interesting feature of the frequency domain solution [Eq. (49)] is the factor I/(K X — K^). Unless the scattering coefficients at the two wavelengths are the same, this factor will necessarily be a function of the modulation frequency, co. This implies that the time-resolved fluorescence energy fluence rate for a point source cannot be determined in closed form even when the fluorescence lifetime is zero. The solution in the time domain will be the convolution of the Fourier transform of I/(K X — K^) with the weighted sum of the time-resolved point source solutions at the excitation and emission wavelengths. We note for completeness that if /zs'.x = /x' m , then

UL YP 3>(p, t) = —^ (â,x

-

- [$x.m(p, t) - m(0, s) are then evaluated to determine CI>X(0, p) and Om(0, p), the excitation and emission fluence rates at the surface. Using the excitation fluence as an example, 4> x (0, p) = —' ^'

27T

'

'

im, and the focal point is at 0.3 cm depth. (Left top) Fluence rate of excitation Fx [W/cm2 per W incident] or [W/cm2/W] shown as isofluence rate contours spaced logarithmically. Contours for 0.01 and 0.1 W/cm2/W are labeled. The incident Gaussian excitation is shown as bold line. The escaping flux, J x (r) [W/ cm2/W], is also shown on linear scale. (Right top) Fluence rate of fluorescence, Ff [W/cm2/W] shown as isofluence rate contours. The escaping flux of fluorescence, J f (r) [W/cm2/W], is also shown. (Left bottom) The escaping fluxes J x (r) and J f (r) on semilog plot. (Right bottom) The ratio of escaping fluorescence to escaping excitation, J f (r)/J x (r). Optical properties of medium at the excitation wavelength are /uax = 1 crrT1, /ASX = 10 cm \ gx = 0.90; at the emission wavelength are /xaf = 0.2 cm'1, /ms1 = 5 cm 1, gf = 0.90; and for the fluorophore are sC = 0.1 cm~1, Y = 1.

70

Jacques

FIGURE 3 Simulations of a collimated uniform flat-field beam. The beam radius is 1 cm. (Left top) Fluence rate of excitation Fx [W/cm2/W] shown as isofluence rate contours spaced logarithmically. Contours for 0.01, 0.1, and 1 W/cm2/W are labeled. The incident beam of excitation is shown as group of vertical lines. The escaping flux, J x (r) [W/cm2/W], is also shown on linear scale. (Right top) Fluence rate of fluorescence, Ff [W/cm2/W], shown as isofluence rate contours. The escaping flux of fluorescence, J f (r) [W/cm2/W], is also shown. (Left bottom) The escaping fluxes J x (r) and J f (r) on semilog plot. (Right bottom) The ratio of escaping fluorescence to escaping excitation, J f (r)/J x (r). Optical properties of medium at the excitation wavelength are /jtax = 1 cm" 1 , /xsx = 100 cm" 1 , gx = 0.90; at the emission wavelength are /xaf = 0.2 cm"1, /Asf - 50 cm" 1 , gf = 0.90; and for the fluorophore are eC = 0.1 cm 1, Y = 1.

Fluorescence in Turbid Media

71

FIGURE 4 A fluorescent heterogeneity located off the central z axis. A 0.1-cm diameter excitation beam at r = 0 excited the fluorescence from a spherical object with a 100-/u,m radius, ehCh = 0.1 crrT1, and Yh = 1. The optical properties were the same as for Figs. 1 and 2. This figure shows the incremental fluorescence due to the heterogeneity that adds to the background fluorescence.

/ * mcfluor . c * / header declare subroutines main() { Set up variables and arrays . Specify USER CHOICES of program parameters .

72

Jacques Call Monte Carlo subroutine with 999 photons , once for excitation and once for fluorescence wavelengths . Make estimate of completion time for simulation.

5.1

Header

The initial header portion of mcfluor . c sets up the main () program. The initial #include commands incorporate supporting standard C program files that are needed by the program. The #define commands cause global substitutions of the first argument by the second argument, i.e., USERLABEL is replaced by the string x * an example Monte Carlo simulation' ' and BINS is replaced by the value 101. Thus, the user can conveniently change the number of bins used by arrays in the program and can specify a label that prints out during the run. The subroutine declarations inform main () of the availability of subroutines listed at the end of the mcfluor . c: me sub ( ) executes a modular Monte Carlo simulation, RFresnel () evaluates the value of internal reflectance at the tissue/ air surface, SaveFile() saves the escaping flux density J(r) and fluence rate F(z, r) to a file, RandomGen () is the random number generator, and four subroutines for allocation of memory: nrerror () used if error encountered during memory allocation, *AllocVector () allocates memory for a 1-D array, **AllocMatrix () allocates memory for a 2-D array,


73

FreeVector () frees the memory allocated for a 1-D array, FreeMatrix () frees the memory allocated for a 2-D array. 5.2 main() The main () program organizes the user's problem, calls the modular subroutine me sub (), interprets the results, and saves the results to files. The main () program begins with declarations of the variables used in main (). The first set of variables are USER CHOICES where the user can specify the optical properties of the medium at the excitation and fluorescence wavelengths, and the properties of the fluorophore. In particular, the product of the fluorophore's extinction coefficient and concentration, eC, are specified by the lumped parameter eC, which has the units of an absorption coefficient [cm"1]. The energy yield Y specifies the W of fluorescence per W of excitation absorbed by fluorophore [WAV] or [dimensionless]. The program is usually run with Y set to 1. Then if one wishes to know the results for a lower Y value, such as Y = 0.01, one simply multiplies the results for flux J, and fluence rate F, for fluorescence by that lower Y value. The parameter mcflag determines whether photons will be launched as a collimated flat-field beam (mcflag = 0), as an approximation to a focused Gaussian beam (mcflag = 1), or as an isotropic point source (mcflag > 2). For flat or Gaussian beams, the parameter radius determines the radius of the beam incident at the surface. In the case of the focused Gaussian beam at the surface of the medium, radius is the radial position where the beam irradiance drops to 1/e the central peak value. For the focused Gaussian beam, an additional pair of parameters describes the focus point under conditions of matched boundary conditions. The parameter waist is the 1/e radius of the Gaussian beam at the focus depth, and the parameter zfocus is the depth position of the focus, both for a matched boundary condition. However, when a Gaussian beam is launched into a tissue with a mismatched boundary condition, m c s u b ( ) will calculate the specular reflectance and the refraction of the beam at the external medium/internal medium interface. The parameters xs, ys, and zs are used when mcflag > 2 to describe the position of isotropic launching. When mcflag equals 0, 1, or 2, the subroutine mcsub () will print out progress reports to the user during the Monte Carlo simulation. If mcflag > 2, then mcsub () will behave as if mcflag = 2 but will omit any printouts, and mcfluor. c will later use mcflag = 3 when iteratively computing contributions due to distributed point sources of fluorescence. The number of photons to be launched by mcsub () is set by Nphotons. In summary,

74

Jacques

Collimated Focused Gaussian Isotropic point

mcflag

Radius

waist

zfocus

0 1 ^2

+ + -

+ -

+

xs

ys

zs

+

+

+

where + means "uses" and — means "ignores." The user chooses the number of photons to be launched by me sub () by specifying the parameter Nruns. The user chooses Nruns and mcmain () calculates the appropriate Nphotons requested of mcsub (). There are Nruns* 106 excitation photons launched, Nruns* 100 background fluorescent photons launched from each of (BINS-1) 2 bins, and Nruns*10 6 fluorescent photons launched for the heterogeneity. If Nruns = 1 and BINS = 101, there will be 10 6 excitation photons launched, and lOO(BINS-l) 2 = 106 background fluorescent photons launched, and 106 fluroescence photons launched for the heterogeneity. Choosing NRUNS = 10 will cause 10-fold more excitation and fluorescent photons to be launched. Hence, the user can conveniently vary the number of photons being launched, such as by choosing NRUNS = 0.1 for a quick initial run lasting 3 min, followed by NRUNS = 10 for a final run lasting 5 hr. The user also chooses the size of the bins for depth, dz, and radial position, dr, appropriate for the cylindrical coordinates used to record escaping flux density J(r) and fluence rate F(z, r). The values (BINS-1) *dz and (BINS-1) *dr specify the total depth and radial extent over which results are stored. The indices [ i z ] [ i r ] specify the particular z and r bins, respectively. The last bins, iz = BINS and ir = BINS, are used to collect any overflow consisting of photons that migrate beyond the extent of the bins. The remaining variables are determined by the program and the user need not specify their values. All units are in [cm] or variations, such as [cm" 1 ] or [cm2]. The declaration and memory allocation of the one-dimensional arrays (Jx, Jf, tempi) and the two-dimensional arrays (Fx, Ff, temp2) employ memory allocation subroutines described in Sec. 6.5. There is a printout of the USER CHOICES so that the user can be reminded of the conditions of a simulation when reviewing a printout. A final initialization step sets all arrays to zero. 5.3

Excitation

This part of the program considers the penetration of excitation light into the tissue or medium. The control parameters for the Monte Carlo subroutine to be used for the excitation were set in the USER CHOICES section above.


75

The me sub () routine is called. The arguments of the subroutine include the optical properties of the medium at the excitation wavelength (muax, musx, gx, nl, n2), the number of r and z bins and their bin sizes, as well as the number of photons to be launched (NR, NZ, dr, dz , Nphotons), the control parameters (mcflag, xs, ys, zs, radius, waist, z focus), and the pointers to arrays Jx[ir] and F x [ i z ] [ir] for escaping flux density J x (r) [W/cm2/W] and fluence rate Fx(z, r) [W/cm2/ W] for the excitation light that are returned by the subroutine. The subroutine me sub () records its results in cylindrically symmetrical radial coordinates of z and r. The results have been normalized by Nphotons, so that the same answer is obtained regardless of the value of Nphotons used, although for a low choice of Nphotons the results are more noisy. Finally, the flux density Jx[ir] andfluencerate F x [ i z ] [ir] are sent to the subroutine SaveFile() that saves the data along with the appropriate r and z positions of each bin based on the function arguments (NR, NZ, dr, dz). The parameter mcflag is also an argument of SaveFile () and specifies the name of the files to be saved. For the case of mcflag = 0 the saved files are called "JO . dat" and "FO .dat"; for the case of mcflag = 1 the saved files are called "Jl.dat" and "Fl.dat"; and so forth. The format of JO. dat and F O . d a t are discussed in Sec. 6.3. 5.4

Background Fluorescence

A medium often has some uniform background fluorophore concentration. The amount of fluorescence generated in each bin of medium is proportional to the local fluence rate of excitation. The next section of the program considers this background fluorescence. In the previous section, mcsub() computed the fluence rate of excitation, F x [ i z ] [ i r ] . The local generation of fluorescence in the bin [ i z ] [ir] equals F x [ i z ] [ir] *eC*Y, where eC is the product of the extinction coefficient e [cm"' M '] and the concentration C [M], and Y is the power yield [W emission per W excitation absorbed by fluorophore]. The factor F x [ i z ] [ir] *eC*Y is the power density of fluorescence [W/ cmYW] that acts as a local source. Because the Monte Carlo subroutine uses cylindrical symmetry, the launching of fluorescent photons from any point in an annular bin will have the same effect in terms of migration in r and z space. Hence, the total fluorescent source power associated with the bin [ i z ] [ir] is the local fluorescence power density multiplied by the volume of the annular bin. Each annular bin has a volume equal to 2*PI*r*dr*dz [cm3], where r = (ir - 0 . 5) *dr. Consequently, the total fluorescent source for each bin is F x [ i z ] [ i r ] * 2 * P I * ( i r - 0 . 5 ) * d r * d r * d z * e C * Y .

76

Jacques

The program uses these fluorescent sources associated with each bin [ i z ] [ir] to weight the results from launching fluorescent photons from the bin using me sub ( ) operating with the optical properties of the medium at the fluorescent wavelength, muaf, musf, gf. The program sets the control parameter mcflag to 3, which causes me sub () to launch an isotropic point source from position xs, ys, zs, just like setting mcflag to 2, but setting mcflag to 3 causes me sub () to avoid printing out progress reports to the user. The ACCUMULATION LOOP sequentially sets xs, ys, zs to each of the positions of the [ i z ] [ir] bins. Because of cylindrical symmetry, xs and zs are set to the r and z positions of the bin, respectively, and ys is always set to 0. The number of photons requested of me sub () is 100*Nruns photons per bin. The LOOP does not launch fluorescent photons due to excitation power deposited in the overflow bins iz = NZ, ir = NR. Thus, there is an error due to ignoring excitation that has migrated beyond the extent covered by the [ i z ] [ir] bins. The user should select values of NZ , NR, dz , dr such that nearly all the deposition of excitation light occurs within the bins so as to minimize excitation energy deposition in the overflow bins. For each [ i z ] [ir] bin, the subroutine m c s u b ( ) returns J(r) as tempi [ir] and F(z, r) as temp2 [ i z ] [ i r ] . These values are multiplied by the strength of the fluorescent power source in that bin, then accumulated in Jf [ir] and Qf [ i z ] [ir] : Jf[iir] += tempi[iir]*Fx[iz] [ir]*2*PI*(ir-0.5)*dr*dr*eC*Y; F f [ i i z ] [iir]

+= temp2[iiz] [iir]*Fx[iz] [ i r ] * 2 * P I * ( i r - 0 . 5 ) *dr*dr*dz*eC*Y;

After all the [ i z ] rows of bins have accumulated for a given [ ir ] column, a progress report may be printed out for the user. After completion of the accumulation of fluorescence generated by all bins, Jf [ ] and Ff [ ] [ ] are saved to the files J3.dat and F 3 . d a t using the S a v e F i l e ( ) subroutine. 5.5

Fluorescent Heterogeneity

A common fluorescence problem to be solved by Monte Carlo is the magnitude of fluorescent signal from a fluorescent heterogeneity within the tissue. A small heterogeneity can be modeled as a point source of fluorescent source whose power of fluorescence [W] is equal to the product of the local fluence rate of excitation F x [ i z ] [ i r ] , the extra amount of fluorophore absorption and fluorescent yield, specified by the product £rhChYh or heC*hY, due to the fluorophore heterogeneity above the background fluorescence, and the volume of the heterogeneity here specified by a spherical


77

volume of radius hrad. This fluorescent power will scale the impulse response to that small heterogeneity located at position (xh, yh, zh). The impulse response is acquired by a call to me sub () for an isotropic source located at the depth position of the heterogeneity, zs = zh, with xs and ys equal to 0, which returns J as tempi [ii] and F as temp2 [iiz] [ii]. In summary, the observed J and F due to the fluorescent heterogeneity is calculated: Jf[iir] =tempi[ii]*Fx[iz][ir]*temp4; Ff[iiz][iir] =temp2[iiz][ii]*Fx[iz][ir]*temp4; where temp4 = 4.0/3*PI*hrad*hrad*hrad*heC*hY; and [ ii ] denotes the radial distance from the heterogeneity to the point of observation. The point of observation is denoted by [iiz] and [ i i r ] . The position of the heterogeneity is denoted by [ i z ] and [ir]. If a larger and/or irregularly shaped fluorescent heterogeneity is required, then the user can superimpose the fluorescence from many small fluorophore heterogeneities to create the result for the larger or irregular heterogeneity. 6. 6.1

THE SUBROUTINES mcsubO

The Monte Carlo simulation is executed using a subroutine called me sub () that considers a semi-infinite medium with an upper surface boundary. The subroutine has the arguments: mua mus g nl n2

absorption coefficient scattering coefficient anisotropy of scattering refractive index of internal medium refractive index of external medium

NR NZ

number of r bins number of z bins

dr dz Nphotons mcflag xs , ys , zs radius

incremental size of r bins incremental size of 2r bins number of photons to be launched control of the launch as collimated, focused, or isotropic location of isotropic launching radius of collimated beam or 1/e radius of focused Gaussian beam

78

Jacques

waist zfocus *J **F

radius of Gaussian beam at the focal point depth position of the focal point for Gaussian beam pointer to one-dimensional array of flux density escaping at surface, J [ i r ] pointer to two-dimensional array of fluence rate, F [ i z ] [ir]

The subroutine follows an algorithm that is depicted in Fig. 5. The program begins with SETUP, declaring and initializing the various variables. In particular, J [ir] and F [ i z ] [ir] are set initially to values of zero. The LAUNCH do-loop proceeds to launch the requested number of photons, Nphotons. If a photon escapes at the surface or is terminated by the ROULETTE procedure, then a new photon is launched. Each photon launching sets the initial photon weight W to a value 1 . 0 - rsp, where rsp is the specular reflectance at the air/tissue surface. The photon launching is executed as either a collimated beam (mcflag = 0), a focused Gaussian beam (mcflag = 1), or an isotropic point source (mcflag > 2). 6.1.1

Launching Collimated Beam

For a collimated beam, the radial position r of launch is selected based on a random number, rnd, and is assigned to the coordinate x while y and z are assigned the value 0: x = radius*sqrt(rnd);

y = o; z = 0; where radius is the beam radius provided as an argument of me sub (). The photon trajectory is specified by the cosine of the angle of the trajectory relative to each of the x, y, and z axes, and these cosine (angle) values are called ux, uy, and uz, respectively. A collimated beam would have uz equal to 1, whereas ux and uy equal 0:

The value of the specular reflectance, rsp, is calculated based on the refractive indices nl and n2:

n, +


79

f Normalize return( J, F ) FIGURE 5 Algorithm for the Monte Carlo subroutine mcsub ()

6.1.2

Launching Focused Gaussian Beam

For a focused Gaussian beam, the radial position of launch is determined: x = radius*sqrt(-log(rnd)); Y = 0; z = 0;

80

Jacques

where radius is the 1/e radius of the Gaussian beam at the tissue surface. Note that log () is a base e logarithm function. The focus of the Gaussian beam for a matched boundary condition is specifed by zfocus and waist, which is the 1/e radius of the beam at z focus. The ratio waist/radius is used to scale the launch position x at the surface to yield a radial position xf ocus at the depth z focus, and the trajectory is oriented inward toward the central axis pointing to the position (xf ocus, 0, z focus): xfocus = x*waist/radius; The program then computes the trajectory required for launching at the surface at position ( x , 0 , 0 ) toward the focus at position (xfocus , 0 , zf ocus), characterized by ux, uy, uz. This trajectory is for the case of matched boundary conditions. temp = sqrt( (x - xfocus)*(x - xfocus) + zfocus^zfocus); sintheta =-(x - xfocus)/temp costheta = zfocus/temp; ux = sintheta; uy = 0.0 ; uz = costheta; This incident trajectory is subsequently modified if there is a mismatched boundary condition. The refractive indices of the internal medium (the tissue), nl, and the external medium (the air), n2, determine the amount of specular reflectance that occurs upon entry into the tissue and the refraction that changes the photon trajectory. The subroutine RFresnel () determines the rsp for the angle of launch that is selected. The cosine of the angle of transmission across the surface boundary into the tissue is returned as the variable uz whose address is denoted in the argument for Rf resnel () as &uz: rsp = RFresnel(n2, nl, costheta, &uz); sintheta = sqrt(1.0- uz*uz); ux = -sintheta; uy = 0.0; uz = costheta; Each rsp. focal of a

photon is launched at a different angle and experiences a different The refraction at the mismatched boundary where nl > n2 causes the point to move deeper into the tissue. Figure 6a illustrates the launching Gaussian beam that would focus at z focus = 0.0300 cm under


81

FIGURE 6 Illustration of launching photons as a focused Gaussian beam across air-tissue surface boundary (nl = 1.00, n2 = 1.33) into tissue with same optical properties as Fig. 2. Gaussian distribution on surface has 1/e radius of radius = 0.0300 cm, and Gaussian distribution at the focal point has 1/e radius of waist = 0.0030 cm, zf ocus = 0.0300 cm. (A) Ten initial launching trajectories for matched boundary condition are shown as lines. (B) After refraction at boundary, the depth of focus along z axis has increased from 0.0300 (vertical dashed line) to 0.0390 cm (peak value of F f (z)).

82

Jacques

matched boundary conditions, and Fig. 6b illustrates how the actual depth of focus has increased to 0.0390 cm due to refraction at the air/tissue surface. 6.1.3

Launching Isotropic Point Source

For an isotropic point source, the position of launch is specified by xs, ys, zs in the argument for me sub (). The trajectory is isotropic and so has no preferential direction, and is specified: cos the ta = 1.0-2. 0*RandomGen (1, 0,NULL) ; sintheta = sqrt(1.0- costheta*costheta) ; psi = 2.0*PI*RandomGen(l,0,NULL) ; cospsi = cos(psi) ; if (psi < PI) sinpsi = sqrt(1.0 - cospsi*cospsi) ; else sinpsi = -sqrt(1.0- cospsi*cospsi) ; ux = sintheta*cospsi; uy = sintheta*sinpsi; uz = costheta; The value sin(psi) is calculated as sinpsi = sqrt ( 1 . 0 cospsi*cospsi) since the sqrt ( ) function is faster than the s i n ( ) function. Because the launch point is within the tissue, there is no specular reflectance and rsp is set equal to zero. For each photon launched, the specular reflectance rsp is calculated by one of the three methods above. Then the initial weight W of that photon is set to 1. 0 - rsp. Hence, a total weight of W = 1.0 is delivered to the tissue but only 1 . 0 - rsp actually enters the tissue. The specular reflectance from each launching is accumulated as a total Rsptot: Rsptot += rsp;

The resulting J [ i r ] will not include specular reflectance. The photon's status is initiated as photon_status = ALIVE where ALIVE has a Boolean value of 1. Once a photon is launched, it enters the PROPAGATION CYCLE. The HOP section sets the stepsize s that the photon takes and updates the current photon position ( x , y , z) based on the current trajectory ( u x , u y , u z ) : s = -log(rnd)/mut; x + = s * ux ;

y += s*uy; z + = s *uz;


83

The ESCAPE? step asks if (z RFresnel ( n l , n2 , -uz, & u z l ) ) If yes, then the photon has escaped the tissue and its weight W is added to the current escaping flux J [ i r ] . The ESCAPE section resets the photon position, then takes a partial step size to reach the surface. The radial position r is calculated based on the x and y positions and the choice of ir is made by the equivalent of an absolute value function, ir = (long) ( r / d r ) +1, with the minimal value being 1 to indicate the first bin. Then the photon is terminated by setting photon_status = DEAD, where DEAD has a Boolean value of 0. The photon will bypass the following DROPSPIN-ROULETTE section and reach the end of the PROPAGATION CYCLE. Because the photon_status = DEAD a new photon is launched. If no, then the photon does not escape but is internally reflected, accomplished by setting z = -z. The photon_status remains ALIVE so the photon can enter the DROP-SPIN-ROULETTE section. If there is no escape, the photon enters the DROP section that causes the current weight W to decrement by an amount that depends on the albedo = mus/ (mua + mus) : absorb = W*(1 - albedo); W -= absorb; Then the value absorb is added to the current bin F [ iz ] [ ir ]. The SPIN section causes the trajectory of the photon to deviate by an angle theta specified as costheta = cos(theta) based on sampling the Henyey-Greenstein scattering function using a random number rnd: rnd = RandomGen(1,0,NULL) ; if (g== 0.0) costheta = 2.0*rnd - 1.0; else if (g == 1. 0) costheta = 1.0; else { temp = (1.0- g*g) / (1. 0 - g + 2*g*rnd) ,costheta = ( 1 . 0 + g*g - t e m p * t e m p ) / ( 2 . 0 * g ) ; }

Also, an azimuthal angle psi for the trajectory change is chosen: psi = 2 . 0 * P I * R a n d o m G e n ( l , 0 , N U L L ) ;

84

Jacques

These angles of deviation are used to calculate a new trajectory assigned to (ux, uy, uz). The ROULETTE section provides a means of terminating a photon based on absorption. A value THRESHOLD was set equal to le-4 at the beginning of the subroutine. If the weight W drops below THRESHOLD, then the photon is either terminated or its weight W is increased and propagation continues. A random number rnd is obtained and compared with a value CHANCE set to 0.1 in this program. The program asks if rnd < CHANCE, and if yes then the weight is increased by a factor I/ CHANCE or 10fold. Propagation continues and the photon returns to the top of the PROPAGATION CYCLE. If no, then the photon is terminated by setting photon_status = DEAD. This method statistically conserves photon energy but terminates photons 9 out of 10 times that W drops below THRESHOLD. At the end of the PROPAGATION CYCLE a new photon is launched. After the PROPAGATION CYCLE has launched Nphotons, the simulation is complete. The subroutine normalizes the values in J [ i r ] by the area of each [ir] bin to yield the escaping flux density [W/cm2/W]. The subroutine normalizes the values in F [ i z ] [ir] by each bin volume to yield the density of power deposition [W/cmYW], and further normalizes by the absorption coefficient mua [cm '] to yield the fluence rate [W/cnr/W]. The subroutine returns J [ i r ] and F [ i z ] [ir] to the calling program mcmain ( ) . At this time, the total weight associated with escaping flux is accumulated in the parameter temp. The subroutine normalizes temp by Nphotons to yield the total amount of escaping flux [WAV]. The subroutine also normalizes Rsptot by Nphotons to yield the value of specular reflectance rsp. In addition, the total amount of absorbed photon weight was accumulated by the parameter Atot and is now normalized by Nphotons to yield [WAV] of absorbed power. These totals for specular reflectance, absorbed power, and escaping flux are printed out in the progress report to the user, and their sum equals unity illustrating conservation of photon energy by the algorithm. These normalized parameters could easily be returned to mcmain () as values if the user modified me sub ( ) . 6.1.4

Testing me sub ()

To test me sub ( ) , the total escaping flux, J tot [WAV], was computed: J,o,

J[ ir ]2-n-(ir - 0-5) dr

for the optical properties yaa = 1 cm ', pis = 9 cm ', g = 0.0, yielding an


85

albedo = 0.90. The medium was semi-infinite and the surface boundary condition was matched. This problem has been simulated by Prahl (1988) [10] who reported in his table 3-1 the result to be 0.4149 and that this value matched the published value of van de Hulst (1980) [11] for this problem. The mcsub () was run 10 times with 10 different Seed values for initializing the random number generator (see Sec. 6.4) to yield a mean and standard deviation of 0.41507 ± 0.00018 (n = 10). Also, an example problem was run that could be compared with diffusion theory. The optical properties were mua = 1 cm"1, mus = 100 cirT1, g = 0.90. The refractive indices of the internal and external media were 1.33 and 1.00, respectively, simulating a water/air interface. An isotropic point source was launched at a depth of zs = I/ (mua + mus* (1-g) ) or 0.09091 cm. Figure 7 shows the comparison of the mcsub () result and diffusion theory as outlined by Parrel et al. [12] using the extrapolated boundary condition: 1 / /I l\exp(-r,/S) /I l\exp(-r 2 /5) F J(r) = — zo - + + (zo + 4AD) - + ^ \r, 8 r, \r2 8 r2

A=

0.668 + 0.063n + 0.710/n - 1.440n2 where rt is the total internal reflectance at the tissue surface and n is the ratio n2/nl. The difference between Monte Carlo and diffusion theory is characterized by the ratio (DT - MQ/MC where DT is the J(r) calculated by diffusion theory and MC is the J(r) calculated by mcsub ( ) . The ratio is in the range of —0.20 to 0.10, except near r = 0 where DT is far too low. This is the same behavior predicted by the MCML code [6]. 6.2

RFresnel ( )

The subroutine RFresnel ( ) was prepared by Lihong Wang as part of the MCML code [6]. The subroutine computes the Fresnel reflectance, rb at an interface between two media with refractive indices ni and nt, where ni

86

Jacques

FIGURE 7 Testing mcsub() versus diffusion theory. (Top) Escaping flux density J(r) versus radial position r. (Bottom) Ratio (DT - MQ/MC where DT is J(r) calculated by diffusion theory and MC is J(r) calculated by mcsub(). Optical properties: /xa = 1 cm' 1 , /us = 100 cm"1, g = 0.90, nn = 1.33, n2 = 1.00. Isotropic point source at 1 mean free path below surface, i.e., zs = 1/[/Aa + /u,s(1 - g)].

is the medium from which the incident photon arrives at angle 9, and nt is the medium into which the photon transmits at angle 0,: 1 /sin 2 (0, - 0,) tan 2 (0 ( - 0t) r, = - — 1 ; — 2 V s i n (0; + 0t) tan (0; + 0,)


87

where sin(0t) sin(0j)

rii nt

The angle of incidence is specifed as cal = cos(angle of incidence). In the subroutine, the angle of transmission is calculated based on Snell's law by the subroutine and the cos(angle of transmission) is returned as the value of the variable ca2_Ptr : RFresnel(ni, nt, cal, *ca2_Ptr) During photon launch as a focused Gaussian beam, the incident medium refractive index is assigned the value of the external medium (ni = n2) and the transmitted medium refractive index is assigned the value of the internal medium (nt = nl). The specular reflectance rsp and the cos(angle of transmission), uz, are calculated by the subroutine call: rsp = RFresnel(n2, nl, costheta, &uz); During photon propagation as photons attempt to escape the medium, they are tested for the occurrence of total internal reflectance. In this case, the incident medium refractive index is the value of the internal medium (ni = nl) and the transmitted medium refractive index is assigned the value of the external medium (nt = n2). The incident cos(angle) is the negative of the current value uz, which is negative because the photon is escaping, so -uz is positive. The transmitted angle uzl is not used, but it does specify the cos(angle of transmission) for the escaping photon and could be used to document the angle of escape. The test for photon escape is phrased: if (rnd > RFresnel ( n l , n2 , -uz , & u z l ) ) If true then there is escape, and if false then there is total internal reflectance. 6.3 SaveFilesO The SaveFilesO subroutine saves twofiles,J [ i r ] and F [ i z ] [ i r ] , along with the appropriate values of r [ ir ] and z [ iz ]. The names of the files depends on the values of the argument Nfile: SaveFile(*J, * * F , N f i l e , N R , N Z , d r , d z )

where NR and NZ are the number of bins and dr and dz are the incremental bin sizes. If Nfile equals 1, then the names of the files are Jl. dat and Fl .dat. If Nfile equals 2, then the names of the files are J2 .dat and F2 .dat, and so on.

88

Jacques

The values of r [ i r ] and z [ i r ] are calculated: r [ i r ] = (ir - 0 . 5 ) * d r z[iz] = (iz - 0 . 5 ) * d z which are the midpoints of each bin. The calculation for r [ir] is based on the expectation value for r within the [ ir ] bin, assuming that the general form of the J [ i r ] and F [ i z ] [ i r ] responses versus r is 1/r. Then the expectation value is: r . . — -/*«\ V/ ~

/-b

r

1 r

27ir dr

7r(b2 - a2) b + a /i 27r(b — a)\ ~~ 2

~~ o

277T dr

which is the midpoint of the [ ir ] bin. A more correct assignment of r [ ir ] would involve using the true behavior of J and F versus r, but this leads to iteratively reconsidering an assignment after an initial determination of the behavior. The user is better advised to simply run the me sub () with smaller values of dr and dz if the user wishes to refine the estimate of behavior J and F at small r. One caveat is that the bins near the central z axis are smaller and few photons are collected in such bins. Therefore, the data for J and F near the central z axis often may be noisy. The file J. dat is a file with two columns and NR rows. The first column is r [ i r ] . The second column is J [ i r ] . The file F. dat is a file with NR+1 columns and NZ + 1 rows. The first element (1, 1) is ignored and set to 0. The first column, rows 2 to NZ + 1, lists the values of z [ iz ]. The first row, columns 2 to NR + 1, lists the values of r [ i r ] . The remaining array, rows 2 to NZ + 1, columns 2 to NR + 1, holds the values of F [ i z ] [ i r ] . 6.4 RandomNumber () The random number generator subroutine was prepared by Lihong Wang as part of the MCML code [6]: RandomGen(Type, Seed, *Status) The generator is initialized by the call with Type set to 0 and Seed set to a long integer (0 < Seed < 32000), e.g., set equal to 1: RandomGen(0,1,NULL); Subsequently, a random number rnd is generated by the call with Type set to 1:


89

rnd = RandomGen(1,0,NULL); In some cases, such as when a command must evaluate log ( r n d ) , it is important to exclude the value rnd = 0. In such cases, the call is phrased: while ( ( r n d = R a n d o m G e n ( 1 , 0 , N U L L ) ) < = 0 . 0 ) ; 6.5

Memory Allocation Routines

The memory allocation routines were also prepared by Lihong Wang as part of the MCML code [6]. The routines are as follows: nrerror(error_text[ ] ) ; *AllocVector(nl, nh) ; **AllocMatrix(nrl, nrh, ncl, nch); FreeVector(*v, nl, nh) ; FreeMatrix (**m, nrl, nrh, ncl, nch); They are used in declaring one- and two-dimensional arrays at the beginning of the program mcmain () and in freeing the memory allocated for these arrays at the end of mcmain (). They allow the arrays to be addressed by indices ranging from 1 to NR and 1 to NZ. Their use is shown in mcmain ().

90 7.

Jacques LISTING OF mcfluor.c

înclude <stdio.h> înclude <stdlib.h> ttinclude <string.h> înclude <math.h> înclude

/**** USER CHOICES *+****/ ^define USER_LABEL ^define BINS~

"an example Monte Carlo simulation" 101

* DECLARE SUBROUTINES /* The Monte Carlo subroutine */ void mcsub (double mua, double mus, double g, double nl, double n2, long NR, long NZ, double dr, double dz, double Nphotons, int mcflag, double xs, double ys , double zs, double radius, double waist, double zfocus, double + J, double +*F) ; /* Computes internal reflectance at tissue/air interface */ double RFresnel (double nl, double n2, double cal, double *ca2_Ptr); /* Saves surface escape R(r) and fluence rate distribution F(z,r) */ void SaveFile (double *J, double **F, j nt Nfile, long NR, long NZ, double dr, double dz) ; /* Random number generator Initiate by RandomGen ( 0, 1 , NULL) Use as rnd = RandomGen ( 1, 0, NULL) */ double RandomGen ( char Type, long Seed, long *Status); /* Memory allocation routines * from MCML ver. 1.0, 1992 L. V. Wang, S. L. Jacques, * which are modified versions from Numerical Recipes in C. */ void nrerrorfchar error__text [ ] ) ; double *AllocVector (short nl, short nh) ; double **AllocMatrix ( short nrl, short nrh, short ncl, short nch); void FreeVector (double *v, short nl, short nh) ; void FreeMatrix (double **m, short nrl, short nrh, short ncl, short nch)


91

MAIN PROGRAM int main() {

'* USER CHOICES /* excitation * excitation absorption coeff. [cmA-l] */ double muax 1.0; musx 100.0; /* excitation scattering coeff. [cmA-l] */ double gx /* excitation anisotropy [dimensionless] */ 0.90; double /* refractive index of medium */ 1.33; double nl n2 = 1 00; /* refractive index outside medium */ double /* 0 = collimated, 1 = focused Gaussian, mcflag = 0; short 2 = isotropic pt */ used if mcflag = 0 or 1 */ double used if mcflag = 1 */ double used if mcflag = 1 */ double used if mcflag =2*1 double used if mcflag = 2 */ double double = 0.090909; /* used if mcflag = 2 */ /* background fluorescence */ double /* fluorescence absorption coeff. [cmA-l] */ /* fluorescence scattering coeff. [cmA-l] */ double /* fluorescence anisotropy [dimensionless] */ double /* ext. coeff. x cone of fluor [cmA-l] */ double double /* Energy yield for fluorescence [W/W] */ /* heterogeneity double xh =0.2; / * heterogeneity */ double yh = 0.15; / * heterogeneity */ double zh =0.3; / * heterogeneity */ double heC = 0 . 1 ; / * extra eC of heterogeneity */ double hY = 1 . 0 ; / * energy yield of heterogeneity */ double hrad = 0.01; / * radius of spherical heterogeneity */ /* other parameters */ double Nruns = 0.1; /* number photons launched = Nruns x Ie6 */ double dr = 0.0100; /* radial bin size [cm] */ dz /* depth bin size [cm] */ double /*****,.

char label[1]; double PI = 3.1415926; double Nphotons; long ir, iz, iir, iiz, ii; double temp, temp3, temp4, r, rl, r2; /* dummy variables */ double start_time, finish_timel, finish_time2, finish timeS; clock() */ double timeA, timeB; time_t now; double *Jx, *Jf, *templ; double **Fx, **Ff, **temp2; long NR = BINS; /* number of radial bins */ long NZ = BINS; /* number of depth bins */ Jx = AllocVector(1,BINS); Jf = AllocVector(1,BINS); tempi = AllocVector(1,BINS);

for

92 Fx Ff ternp2

Jacques = AllocMatrix (1, BINS, 1, BINS); /* for absorbed excitation */ = AllocMatrix (1, BINS, 1, BINS); /* for absorbed fluor */ = AllocMatrix ( 1, BINS, 1, BINS); /* dummy matrix */

strcpy(label, USER_LABEL); printf("\n|| printf ( "| | *s\n",label) ; printf ( "M

An") ; -\n\n")

start_time = clock (); now = time (NULL); printf ( "%s\n", ctime ( Snow) ) ; if (1) { /* Switch printout ON=1 or OFF=0 */ /* print out summary of parameters to user */ printf (" ----- USER CHOICES ----- \n"); printf ("EXCITATION\n") ; printf ( "muax f'l. 3f \n" ,muax) ; printf ( "musx £1.3f\n",musx); «1.3f\n",gx); printf ( "gx printf ( "nl «1.3f\n",nl); V,1.3f\n",n2) ; printf ("n2 printf ( "mcf lag Ad\n",mcflag); printf ( "radius rl.4f\n",radius); printf ( "waist = S1.4f\n'.waist); U.4f\n' ,radius); printf 'zfocus -1.4f\n' • x s ) ; printf 'xs printf 'ys U.4f\n' , ys ) ; printf ' zs •1.4f\n",zs) ; printf 'BACKGROUND FLUORESCENCE\n" ) ; %l . 3f \n",muaf ) ; printf("muaf -1. 3f \n",musf ) ; printf("musf printf("gf *1.3f\n",gf ) ; ?1.3f\n",eC) ; printf("eC %1.3f\n", Y) ; printf("Y printf ("FLUORESCENT HETEROGENEITY\n" ) ; printf ("xh = *1 . 4f \n", xh) ; °.1.4f\n",yh) ; printf("yh printf("zh •*1.4f\n",zh) ; *1.4f\n",heC) ; printf("heC printf("hY *1. 4f\n",hY) ; printf("hrad = * 1.4f\n",hrad) ; printf("OTHER\n"); printf("Nruns = *l.lf \n", Nruns); printf("dr = ^1.4f\n",dr); printf("dz = *1.4f\n",dz); printf (" \n\n" ) ;

/* Initialize arrays */ for (ir=l; ir:-—_?*•— -i

/^

6

* >.

c-

''^r

2

4

i^

•v

FIGURE 13 (a) In vivo intrinsic fluorescence EEM of non-dysplastic Barrett's esophagus site, (b) Residual features remaining after a linear combination of NAD(P)H and collagen EEMs (shown in Fig. 11a, b) have been subtracted from the EEM shown in (a), (c) In vivo intrinsic fluorescence EEM of cervical squamous metaplastic site. The corresponding residual is shown in (c).

Georgakoudi et al.

136

FIGURE 13

Continued

Intrinsic Fluorescence Spectroscopy

137

TABLE 1 Relative Collagen and NAD(P)H Contributions to Intrinsic Fluorescence EEMs of Different Tissue Types

Non-dysplastic Barrett's esophagus High-grade dysplasia in Barrett's esophagus Cervical squamous metaplasia Cervical high-grade squamous intraepithelial lesion

Collagen coefficient

NAD(P)H coefficient

44 4.2

3.6 6.8

2.3 1.8

0.2 3.9

an important role in processes that take place during significant tissue architectural changes, as in the case of wound healing and tissue regeneration [30]. Changes in the expression of matrix metalloproteinases (MMPs), which are types of collagenases, have been reported in dysplastic lesions in the cervix [31], bronchus [32], and oral cavity [33]. In addition, it has been shown that an increased level of cysteine and serine proteases, which are known to be MMP activators [34], is found in gastric and colorectal cancerous and precancerous lesions [35]. Furthermore, the presence of MMPs is essential during tumor invasion and metastasis [36]. It is possible that the decrease in collagen fluorescence observed within sites of dysplasia or of the dynamically changing transformation zone of the cervix can be attributed to the presence of collagenases degrading the fluorescing collagen crosslinks. 6.

CONCLUSIONS

In this chapter, we have presented a photon migration-based theoretical model for the extraction of intrinsic tissue fluorescence from a combination of fluorescence and reflectance measurements. This model is valid in wavelength regimes of substantial hemoglobin absorption. We have validated this model with experiments performed using physical tissue models (phantoms) and ex vivo tissues, representing a wide range of scattering and absorption coefficients. Furthermore, we have used this model to analyze tissue spectra acquired in vivo during asphyxiation of esophageal varices to determine the intrinsic fluorescence changes that take place during the induced change in the redox state of the tissue. The corresponding intrinsic fluorescence excitation-emission matrices were decomposed to yield the spectral signatures of two components that describe the observed fluorescence changes. The

138

Georgakoudi et al.

CD 0

C

(D

CD

3 O C/5

2

400

500

600

700

800

Wavelength (nm) gO.6 C

0

O $0.4

00.2 C

°v_ -*-•

-Eo.o 400

500

600

700

800

Wavelength (nm) FIGURE 14 Intrinsic fluorescence spectra at 337 nm excitation for (a) a non-dysplastic (solid line) and a high-grade dysplastic Barrett's esophagus site (dashed line), and (b) a cervical squamous metaplastic (solid line) and a high-grade squamous intraepithelial lesion site (dotted line).

fluorescence lineshapes of these components were attributed to collagen and NAD(P)H. We can use a linear combination of these collagen and NAD(P)H EEMs to describe with excellent agreement intrinsic fluorescence spectra acquired in vivo within different types of epithelial tissues. Thus, we can extract quantitative biochemical information on the relative contributions of individual tissue fluorophores to the bulk intrinsic tissue fluorescence. Fi-


139

nally, differences in the intrinsic fluorescence lineshape between dysplastic and non-dysplastic tissues can be attributed to specific biochemical changes. Inasmuch as this technique can be applied clinically in a noninvasive manner, it can serve as a guide to biopsy of invisible or ambiguous lesions in the uterine cervix and Barrett's esophagus. Similar changes are expected during the development of precancerous lesions in a number of other organs, such as colon, bladder, breast, lung, and the oral cavity. In addition to its diagnostic value, extraction of quantitative biochemical information from in vivo tissue intrinsic fluorescence spectra provides a unique opportunity to study tissue changes without the introduction of excision and processing artifacts. Thus, such studies could enhance our understanding of the biochemical relationships between the epithelium and the stroma during cancer development. Furthermore, detection of biochemical changes within tissue in its native state could be used as a novel and effective means of assessing in a noninvasive way the effects of chemopreventive/immunotherapeutic treatments for a number of tissues [37].

ACKNOWLEDGMENTS This work is the result of a highly collaborative research program between colleagues at the MIT Laser Biomedical Research Center, New England Medical Center (NEMC), Boston, Brigham and Women's Hospital (BWH), Boston and Children's Hospital, Boston. We acknowledge the contributions of Dr. Qingguo Zhang, Dr. Ramachandra Dasari, and Mr. Luis Galindo at MIT; Dr. Tulio Valdez and Dr. Stanley Shapsay at NEMC; Dr. Jacques Van Dam and Dr. Brian Jacobson at BWH; and Dr. Kamran Badizadegan at Children's Hospital. We also acknowledge financial support from NIH grants P41RR02594, CA53717, and CA72517. Irene Georgakoudi gratefully acknowledges support from an NIH NRSA fellowship.

REFERENCES 1. 2.

3. 4.

C Ince, JMCC Coremans, HA Bruining. In vivo NADH fluorescence. Adv Exp Med Biol 317:277-296, 1992. FF Jobsis, M O'Connor, A Vitale, H Vreman. Intracellular redox changes in functioning cerebral cortex I. Metabolic effects of epileptiform activity. J Neurophysiol 34:735-749, 1971. A Mayevsky, B Chance. Repetitive patterns of metabolic changes during cortical spreading depression of the awake rat. Brain Res 65:529-533, 1974. S Ji, B Chance, BH Stuart, R Nathan. Two-dimensional analysis of the redox state of the rat cerebral cortex in vivo by NADH fluorescence photography. Brain Res 119:357-373, 1977.

140 5.

Georgakoudi et al.

A Mayevsky, B Chance. Intracellular oxidation-reduction state measured in situ by a multichannel fiber-optic surface fluorometer. Science 217:537-540, 1982. 6. E Dora, L Gyulai, AGB Kovach. Determinants of brain activation-induced cortical NAD/NADH responses in vivo. Brain Res 299:61-72, 1984. 7. S Ji, B Chance, K Nishiki, T Smith, T Rich. Micro-light guides: a new method for measuring tissue fluorescence and reflectance. Am J Physiol 236:C144C156, 1979. 8. S Kobayashi, K Kaede, E Ogata. Optical consequences of blood substitution on tissue oxidation-reduction state of the rat kidney in situ. J Appl Physiol 31: 93-96, 1971. 9. G Renault, E Raynal, M Sinet, M Muffat-Joly, J-P Berthier, J Cornillault, B Godard, J-J Pocidalo. In situ double-beam NADH laser fluorimetry: a choice of a reference wavelength. Am J Physiol 246:H491-H499, 1984. 10. J Wu, MS Feld, RP Rava. Analytical model for extracting intrinsic fluorescence in turbid media. Appl Optics 32:3585-3595, 1993. 11. JC Finlay, DL Conover, EL Hull, TH Foster. Porphyrin bleaching and PDTinduced spectral changes are irradiance dependent in ALA-sensitized normal rat skin in vivo. Photochem Photobiol 73:54-63, 2001. 12. AJ Durkin, S Jaikumar, N Ramanujam, R Richards-Kortum. Relation between fluorescence-spectra of dilute and turbid samples. Appl Optics 33:414-423, 1994. 13. NN Zhadin, RR Alfano. Correction of the internal absorption effect in fluorescence emission and excitation spectra from absorbing and highly scattering media: theory and experiment. J Biomed Opt 3:171-186, 1998. 14. CM Gardner, SL Jacques, AJ Welch. Fluorescence spectroscopy of tissue: Recovery of intrinsic fluorescence from measured fluorescence. Appl Optics 35: 1780-1792. 15. MS Patterson, BW Pogue. Mathematical model for time-resolved and frequency-domain fluorescence spectroscopy in biological tissues. Applied Optics 33:1963-1974, 1994. 16. GM Wagnieres, WM Star, BC Wilson. In vivo fluorescence spectroscopy and imaging for oncological applications. Photochem Photobiol 68:603-632, 1998. 17. N Ramanujam. Fluorescence spectroscopy of neoplastic and non-neoplastic tissues. Neoplasia 2:89-117, 2000. 18. Q Zhang, MG Miiller, J Wu, MS Feld. Turbidity-free fluorescence spectroscopy of biological tissue. Optics Lett 25:1451-1453, 2000. 19. MG Miiller, I Georgakoudi, Q Zhang, J Wu, MS Feld. Intrinsic fluorescence spectroscopy in turbid media: disentangling effects of scattering and absorption. Appl Optics 40:4633-4646, 2001. 20. RA Zangaro, L Silveira, R Manoharan, G Zonios, I Itzkan, RR Dasari, J Van Dam, MS Feld. Rapid multiexcitation fluorescence spectroscopy system for in vivo tissue diagnosis. Appl Optics 35:5211-5219, 1996. 21. RM Cothren, GB Hayes, JR Kramer, BA Sacks, C Kittrell, MS Feld. A multifiber catheter with an optical shield for laser angiosurgery. Lasers Life Sci 1: 1-12, 1986. 22. JR Mourant, JP Freyer, AH Hielscher, AA Eick, D Shen, TM Johnson. Mech-


23.

24.

25.

26. 27.

28.

29.

30. 31.

32.

33.

34.

35.

141

anisms of light scattering from biological cells relevant to noninvasive opticaltissue diagnostics. Appl Optics 37:3586-3593, 1998. G Zonios, LT Perelman, V Backman, R Manoharan, M Fitzmaurice, J Van Dam, MS Feld. Diffuse reflectance Spectroscopy of human ademonatous colon polyps in vivo. Appl Optics 38:6628-6637, 1999. H Zeng, C MacAulay, DI McLean, B Palcic. Spectroscopic and microscopic characteristics of human skin autofluorescence emission. Photochem Photobiol 61:639-645, 1995. TJ Farrell, RP Hawkes, MS Patterson, BC Wilson. Modeling of photosensitizer fluorescence emission and photobleaching for photodynamic therapy dosimetry. Appl Optics 37:7168-7183, 1998. R Richards-Kortum, Sevick-Muraca. Quantitative optical Spectroscopy for tissue diagnosis. Annu Rev Phys Chem 47:555-606, 1996. I Georgakoudi, BC Jacobson, MG Muller, K Badizadegan, DL Carr-Locke, EE Sheets, CP Crum, RR Dasari, J Van Dam, MS Feld. NAD(P)H and collagen as quantitative fluorescent biomarkers of epithelial pre-cancerous changes. Cancer Res 62:682-687, 2002. R Richards-Kortum, RP Rava, J Baraga, M Fitzmaurice, J Kramer, MS Feld. In: R Pratesi, ed. Optronic Techniques in Diagnostic and Therapeutic Medicine. New York: Plenum Press, 1990. I Georgakoudi, B Jacobson, J Van Dam, V Backman, MB Wallace, MG Muller, Q Zhang, K Badizadegan, D Sun, GA Thomas, LT Perelman, MS Feld. Fluorescence, reflectance and light scattering Spectroscopy for evaluating dysplasia in patients with Barrett's esophagus. Gastroenterol 120:1620-1629, 2001. DL Ellis, IV Yannas. Recent advances in tissue synthesis in vivo by use of collagen-glycosaminoglycan copolymers. Biomaterials 17:291-299, 1996. A Talvensaari, M Apaja-Sarkkiner, M Hoyhtya, A Westerlund, U Puistola, T Turpeernniemi-Hujanen. Matrix metalloproteinase 2 immunoreactive protein appears early in cervical epithelium dedifferentiation. Gynecol Oncol 72:306311, 1999. FB Galateau-Dalle, RE Luna, K Horiba, MN Sheppard, T Hayashi, MV Fleming, TV Colby, W Bennett, CC Harris, WG Stetler-Stevenson, LA Liotta, VJ Ferrans, WD Travis. Matrix metalloproteinases and tissue inhibitors of metalloproteinases in bronchial squamous preinvasive lesions. Hum Pathol 31:296305, 2000. M Sutinen, T Kainulainen, T Hurskainen, E Vesterlund, JP Alexander, CM Overall, T Sorsa, T Salo. Expression of matrix metalloproteinases (MMP-1 and -2) and their inhibitors (TIMP-1, -2 and -3) in oral lichen planus, dysplasia, squamous cell carcinoma and lymph node metastasis. Br J Cancer 77:22392245, 1998. YA DeClerck. Interaction between tumour cells and stromal cells and proteolytic modification of the extracellular matrix by metalloproteinases in cancer. Eur J Cancer 36:1258-1268, 2000. L Herszenyi, M Plebani, P Carrara, M De Paoli, G Roveroni, R Cardin, F Foschia, Z Tulassay, R Naccarato, F Farinati. Proteases in gastrointestinal neoplastic diseases. Clin Chim Acta 291:171-187, 2000.

142 36.

37.

Georgakoudi et al. LA Liotta, K Tryggbason, S Garbisa, I Hart, CM Foltz, S Shafie. Metastatic potential correlated with enzymatic degradation of basement membrane collagen. Nature 284:67-68, 1980. G Kelloff, J Crowell, V Steele, R Lubet, C Boone, W Malone, E Hawk, R Lieberman, J Lawrence, L Kopelovich, I Ali, J Viner, C Sigman. Progress in cancer chemoprevention. Ann NY Acad Sci 889:1-13, 1999.

Real-Time In Vivo Confocal Fluorescence Microscopy Milind Rajadhyaksha Northeastern University, Boston, Massachusetts and Memorial Sloan-Kettering Cancer Center, New York, New York, U.S.A. Salvador Gonzalez Harvard Medical School and Massachusetts General Hospital, Boston, Massachusetts, U.S.A.

1.

INTRODUCTION

The confocal scanning microscope is well known for its ability to perform optical sectioning: a thin plane or section within a thick turbid medium is noninvasively imaged with high resolution and contrast [1]. Since its invention and development, confocal scanning microscopes have been extensively used in biomedicine for imaging human and animal tissues in vivo. Nuclear, cellular and morphological detail is imaged in living intact tissue without having to physically excise and prepare thin sections or cultures. By comparison, the conventional microscope images nuclear and cellular detail either within prepared thin sections or in culture (in vitro). Histology is a wellknown example: it involves the removal of tissue (biopsy), followed by fixing or freezing, cutting into thin sections with a microtome, staining with hematoxylin-eosin or other dyes to enhance contrast, and, finally, viewing with the microscope. 143

144

Rajadhyaksha and Gonzalez

Real-time confocal microscopes provide a noninvasive window into living tissue for basic and clinical research. Tissue can be imaged in its native state either in vivo or freshly excised (ex vivo) without the processing that is necessary for conventional microscopy. Dynamic processes can be noninvasively observed over an extended period of time. For example, the cellular and nuclear morphology of tissue, cell-to-cell interactions, wound healing and tissue regeneration, effects of ultraviolet light, responses to allergic or irritant agents, photoaging, microcirculation, fungal infections, and pharmacokinetics of drug delivery and cosmetics may be directly observed. Among the various types of human and animal tissues, the cornea, retina, skin, and oral mucosa are easily accessible and therefore have been most amenable to confocal microscopy. Nuclear, cellular, and morphological detail in epithelial and deeper connective layers, including microcirculation, has been imaged in such tissues with both white-light tandem scanning and confocal scanning laser microscopes. Confocal imaging may be potentially used in the clinic for visualizing lesions and their margins prior to biopsy, diagnosis of lesions without biopsy, or detection of margins either intrasurgically (to guide surgery) or in freshly excised tumors (to guide surgical pathology). Clinical applications include characterization of diseases, including assessment of tumor margins. A related application is rapid morphological examination in the operating room of surgical specimens, either intraoperatively or freshly excised, without the tissue preparation (fixing or freezing, sectioning, staining) that is necessary for conventional histology. Confocal imaging may be performed in either reflectance or fluorescence. In reflectance, the contrast arises from native variations in the refractive indices of organelles and microstructures within tissue. Extensive research based on confocal reflectance imaging in both animals and human tissues in vivo has been reported in the literature; in particular, a significant effort is currently in progress in dermatology to study human skin and skin disease. In fluorescence, the contrast is from exogenous fluorophores that are administered to label specific tissue microstructures. (Another source of contrast is autofluorescence—i.e., endogenous or native fluorescence— which is inherently weak but has more recently been employed in twophoton confocal microscopy. This is described in Chapter 6 and is not discussed here.) As in reflectance, there has been concomitant research based on confocal fluorescence imaging, especially in small animals. During the past decade, fluorescence studies have been performed on microcirculation in rat brain cortex [2-6], nuclear and cellular morphology in mouse skin [7], rabbit bone [8,9], rabbit cornea [10], rat bone [I I], and rat kidney [12]. Other examples, especially of functional imaging, include those of the epidermis in transgenic mice expressing green fluorescent protein [13], neurons

In Vivo Confocal Fluorescence Microscopy

145

in rat basal forebrain [14], gene expression in rat hippocampus [15], wound repair in rabbit cornea [16], angiogenesis and revascularization in transplanted pancreatic islets [17], and oligonucleotide uptake in skin keratinocytes [18]. Internal organ tissues in vivo and surgically exposed tissues in situ have also been imaged through excisions with the use of optical fiberbased confocal microscopes in endoscope configurations; tissues include colon and vas deferens [19-22] and peritoneum [23,24]. Human studies include evaluation of microvascular patterns in choroidal melanomas [25], and largely unpublished work on skin morphology by various industry research groups. The morphology and function in plants in vivo has also been of interest [26-29]. Typically, the fluorophore has been administered either topically or by intravenous injection. Compared to reflectance, the contrast produced by fluorescence can be much more sensitive and specific to tissue microstructure and function. Both the significant number of publications during the last few years and the recent development of commercial confocal fluorescence microscopes specifically for in vivo use clearly indicate the growing interest in fluorescence contrast. However, the full potential of fluorescence-based imaging for both animal and human applications is yet to be realized, probably because realtime confocal detection of fluorophores within tissue in vivo is challenging. Delivery of fluorophore molecules at a specific site followed by clearance of unbound molecules is not easy. Moreover, fluorescence emission is subject to fundamental limitations due to photobleaching or excited singlet (or triplet) state saturation, such that at low, nontoxic fluorophore concentrations the signals may be weak. Thus, one requires a careful understanding of the physics of fluorescence with detailed attention to the optimization of instrumentation design and imaging parameters. Brakenhoff and Visscher [30] provide an excellent introduction to this topic. Factors governing successful real-time imaging in vivo include choice of fluorophore and its photophysics, and confocal microscope design parameters such as choice of optics and objective lenses, scanning configuration, resolution, collection and transmission efficiency, frame rate or detector integration (pixel) time, chromatic aberrations within the optics, spherical aberrations within the tissue, scattering and absorbing properties of the tissue, and optoelectromechanical methods to stabilize tissue. With effort toward understanding and applying these factors, the full potential of confocal fluorescence imaging can be realized. In this chapter, we will quantitatively look at the possibilities (and limitations) and applications of real-time confocal fluorescence microscopy for basic research and clinical applications in animal and human tissues in vivo. The results of our analytical and experimental studies as well as those of other groups are presented, including a review of the literature.

146

2.


CONCEPTS OF CONFOCAL IMAGING

Over the last four decades since its invention, the theory and practice of confocal microscopy have been well established [31-34]. In this section, we provide a brief introducion to the main concepts of confocal imaging and list the original references that contain the detailed theory. 2.1

Optical Sectioning

A confocal microscope is made up of a small, bright source of light that illuminates a small three-dimensional spot within the object (Fig. 1). The illuminated spot is imaged onto a detector through a small aperture (pinhole). The source of light, illuminated spot, and detector aperture lie in optically conjugate focal planes, and hence this configuration is called "confocal." Because we illuminate only a single small spot at a time and image through a small aperture, the detector receives light only from the single illuminated spot that is in focus. Light from all other spots that are either axially or laterally away from focus is rejected (or spatially filtered) by the small aperture in front of the detector. The spot may be scanned in two dimensions to illuminate and image a plane. Thus, with a confocal microscope, we image the single specific plane that is in focus within the object. Thin slices or sections are noninvasively imaged; this is called optical sectioning. Illuminating with a small spot and imaging through a small detector aperture provides high axial resolution and high contrast (i.e., strong rejection of light from all out-of-focus planes that otherwise would cause blur). By comparison, a conventional microscope consists of a large source of light that illuminates a large spot (large volume) within the object, which is then imaged onto a detector through a large aperture. (When you directly view the object, the pupil in your eye is the large aperture.) The detector receives light from the plane that is in focus as well as from planes that are away from focus. Conventional microscopes, therefore, do not have axial resolution and lack optical sectioning capability (Fig. 2). Consequently, the object has to be physically cut into thin sections and stained before viewing. It is obvious from Fig. 1 that the confocal microscope is essentially a point illumination and point detection system. Information about a single spot within an object is useful in many situations, but often we want an image of a large portion of the object. To create an image, the illuminated spot is scanned over the desired field of view: we illuminate the desired area of interest on the object point by point in a two-dimensional matrix (raster) and then build up the image correspondingly point by point. Scanning is accomplished either by keeping the illuminated spot stationary and moving the object (known as object scanning) or by moving the illumination beam over the stationary object (known as beam scanning). Although Fig. la

147


illuminated spot

[— out-of-focus plane in-focus illuminated plane formed by the scanned spot r- out-of-focus plane

detector small, bright sour of light

small aperture condenser lens

objective lens object

(a) Transmittance configuration beamsplitter small, bright source of light

tissue in vivo (human being or animal)

detector with small aperture

(b) Reflectance or epitaxial configuration FIGURE 1 Optical sectioning with a confocal microscope is the imaging of the single specific plane that is in focus within the object (a). A small source of light illuminates a small three-dimensional spot within the object, which is imaged onto a detector through a small aperture (pinhole). The detector receives light only from the illuminated spot that is in focus (bold lines); light from all other spots that are either axially or laterally away from focus is spatially filtered by the aperture in front of the detector (dotted lines). The illuminated spot may be scanned in two dimensions to create an image of a plane; thus, a thin slice or section is noninvasively observed. Illuminating a small spot and imaging through a small detector aperture provides high axial resolution and high contrast (due to strong rejection of out-of-focus light that otherwise would cause blur). Figure (a) shows the transmittance configuration with the illumination light penetrating through the object and being detected on the other side. However, imaging an animal or human being requires detection of back-scattered light in the reflectance (or epitaxial) configuration, with the illumination and detection being on the same side of the tissue (b).

148


FIGURE 2 Human skin in vivo appears blurred with a conventional microscope (a) but sharp and well resolved with a confocal microscope (b). The confocal section shows a thin layer of basal cells of diameter —10 ^tm and —70 /xm deep within skin. A conventional microscope does not provide axial resolution and lacks optical sectioning capability because a large source of light illuminates a large spot (large volume) within the object, which is imaged onto a detector through a large aperture. The detector receives light from the plane that is in focus as well as from planes that are away from focus. (Imagine what would happen when the detector aperture in Fig. 1a is made large.) Consequently, the object has to be physically cut into thin sections and stained before viewing with a conventional microscope. (See color insert.)

shows light to transmit through the object, the illumination light will actually not penetrate through an animal or human being. Hence, we detect backscattered light, with the illumination and detection being on the same side of the object (Fig. Ib). Object scanning has several advantages. The optical system is simple, and because the beam is stationary and always on axis, we obtain diffractionlimited resolution and constant illumination across the entire object. The object can be moved to obtain as large a large field of view as necessary, independently of the objective lens specifications; thus, a large field of view is possible with a high magnification, high numerical aperture (NA) lens. High NAs are often required for adequate axial resolution (confocal sectioning), particularly to visualize nuclear and cellular detail. A high NA collects more of the fluorescent light and improves detection sensitivity. Moving an object by electromechanical means is slow, which provides longer detection times and, again, higher detection sensitivity. Unfortunately, these powerful advantages lead to one significant limitation: object scanning cannot be used for imaging animals or humans in vivo. With the recent advent of optical and nonoptical imaging modalities, in vivo imaging has rapidly gained importance for basic and clinical research in biomedicine. Real-time imaging of animal or human tissue in vivo requires fast-beam scanning. Here, the optics is complex and lossy. Because the beam is


149

scanned, off-axis aberrations and vignetting causes both the resolution and illumination to vary across the field of view. The scan angles and objective lens magnification define the field of view; thus, high-NA lenses with their correspondingly high magnifications result in rather small, limited fields of view. Fast scanning also results in short detection times and hence lower detection sensitivity. The message here, rather obviously, is that real-time confocal fluorescence imaging in vivo is challenging. 2.2

Reflectance Point Spread Function

The impulse response (i.e., the three-dimensional image of a point object) of a conventional microscope is mainly defined by the three-dimensional point spread function (PSF) of the objective lens that performs the imaging. In general, the condenser lens is used mainly to illuminate the object with a large source of light; since this lens is not involved with imaging the object, it makes a minor contribution to the resolution. The PSF of the objective lens with a clear, circular aperture is well known to be the Airy function in the lateral plane (the plane perpendicular to the optical axis) and a sine2 function in the axial plane (plane containing the optical axis) [35,36]: Lateral: L onv (v) = [2J,(v)/v]2

(1)

Axial: Iconv(u) = [sin(u/4)/(u/4)]2

(2)

where I denotes irradiance and J, is the well-known first-order Bessel function of the first kind. For a given objective lens, with NA = sin a, the optical units v and u relate to the physical coordinates r (radial distance from the optical axis) and z (axial distance from the focus) according to v = (277/A)r sin a and u = (877/A)z sin2(o!/2). [This, of course, assumes that the objective lens images through air. If the lens images through an immersion medium of refractive index n, such that NA = n sin a, then v = (27r/A)rn sin a and u = (877/A)zn sin2(a/2).] In a confocal microscope (Fig. 1), both the condenser lens and the objective lens perform imaging. Therefore, the impulse is defined by the product of the PSFs of the two lenses [31,32,34,37-40]: Confocal PSF = condenser lens PSF X objective lens PSF = illumination PSF X imaging PSF

(3)

Imaging tissue in vivo using the epi- or back-scattered configuration (Fig. Ib) means that we illuminate and image through the same lens: the condenser lens is the same as the objective lens (NAcondenser = NAohjective = NA). When we use a confocal microscope to look at a point object in reflectance, using either monochromatic or colored light, the imaging is at the same

150


wavelength(s) as the illumination (Aii, um , nali()n = A,mai,mg = A), such that the confocal PSF is Reflectance, lateral: I c o n l (v) = [2J,(v)/v] 4 Reflectance, axial: I c i m l (u) = [sin(u/4)/(u/4)]

(4) 4

(5)

assuming, again, clear, circular apertures for both lenses. The full width at half maximum of these PSFs defines the confocal resolution. Assuming incoherent, uniform illumination, the resolution of a confocal microscope is thus Reflectance, lateral: Ax = 0.46A/NA Reflectance, axial: Az = l.4nA/NA

(6)

2

(7)

The factors 0.46 and 1.4 vary depending on whether the illumination is incoherent or coherent, planar or spherical wave (uniform or Gaussian irradiance). The axial resolution in Eq. (7) defines the diffraction-limited thickness of the optical section that is imaged by the confocal microscope. For pinholes of diameter larger than about 20 optical units (i.e., v > 10), a purely geometrical optics perspective gives a reasonable estimate of the section thickness. Here, the section thickness may be determined from the detected signal drop-off as a function of distance z from the focal plane [41,42]. The detected signal drops off as C(z) = 1 / f l + (z/d p )M tan a] 2

(8)

where dp is the detector aperture (pinhole) diameter, M is the magnification, and a is the half-angle of the light cone defined by the NA. The full width at half maximum of C(z) gives us the geometrically defined section thickness to be [42,43] Az gL , (imctnca , - V2d p /(M tan a) 2.3

(9)

Fluorescence Point Spread Function

In fluorescence, the (incoherent) impulse response is very different because the imaging (i.e., fluorescence emission) wavelength is not the same as the illumination (i.e., excitation) wavelength [41,42,44-52]. In fact, the imaging wavelength is longer than the illumination wavelength by the Stokes shift (A imagin ,, > A inimiinatu)n ). The lateral and axial PSFs now depend on the ratio /3 = A.n^.^/A,,,,,,,,,,,;,,,,,,, [32,46]. The confocal PSF, for a point object, is now Fluorescence, lateral: I collt (v) = [2J l (v)/vl 2 [2J,(v/ ) S)/(v//3)] 2 2

Fluorescence, axial: I c o n l (u) = [sin(u/4)/(u/4)] [sin(u/4/3)(u/4 j 8)]

(10) 2

(11)

The optimum resolution is obtained when /3 —> 1, meaning that the


151

fluorescence imaging wavelength should be close to the illumination wavelength. Of course, Eqs. (10) and (11) assume the simple case of illumination and imaging at single wavelengths through narrow bandpass filters. Often, in reality, we may use a range of wavelengths with either short-pass or longpass filters. This is especially true for real-time in vivo imaging where the detected florescence signal from a fluorophore at a low, nontoxic concentration, deep within tissue may be weak, necessitating broad-band detection. Detecting a range of wavelengths increases the section thickness. In other words, the axial resolution (section thickness) is proportional to /3, and if (3 becomes very large (/3 —» °°), the sectioning degrades and the imaging becomes similar to that of an incoherent conventional microscope. Equations (10) and (11) assume the ideal condition of a point detector aperture (i.e., pinhole diameter is equal to the diffraction-limited lateral resolution). Since in reality the fluorescence may often be weak, one may have to increase pinhole diameter to increase signal-to-noise ratio while accepting an increase in section thickness [46,53-55]. The sectioning remains constant over a range of small pinhole diameters that are close to the diffraction limit, but then degrades (somewhat linearly) as the pinhole diameter increases; however, that range becomes small as (3 increases. With large pinholes, the sectioning can be defined on the basis of geometrical optics, as in Eqs. (8) and (9) [41,42]. Here the inverse square dependence of the detected signal on the sectioning implies that the total detected signal drops as 1/t for outof-focus layers of thickness t. This effect suppresses the background fluorescence.

The above discussion is only a summary of the main concepts of confocal imaging. The detailed theory is well explained in the original references that were mentioned. Our main concern here from an application perspective is detectability of signal to better understand the possibilities and limitations of fluorescence imaging in animals and humans. Indeed, detectability has been previously proposed as a criterion for the design and evaluation of confocal microscope performance [56,57]. 3.

DETECTABILITY OF FLUORESCENCE SIGNAL

Real-time high-resolution confocal fluorescence imaging in tissue in vivo is challenging. At fluorophore concentration that is low enough to be nontoxic to the tissue, the very small illuminated confocal spot may not contain enough fluorophore molecules to produce a strong fluorescence signal. Moreover, in real time, the detector may not have a sufficiently long integration time for each pixel to collect a sufficiently large number of fluorescent photons. For example, for a typically high NA of 0.9 (water immersion) and illumination wavelength A of 488 nm, the diffraction-limited resolution

152


is —0.3 fjim (lateral) and — I yum (axial or section thickness), and thus the imaged confocal spot is ~10 l3 ml. At a typical dosage of — I yuM, we might expect —60 molecules in the confocal spot, and at a typical fluorescence emission rate of — l() x photons per molecule-sec, with a detector integration time of —100 nsec (video rate), each pixel in the image would collect —600 photons. Underlying this signal may be a background noise of —200 photons, and therefore the quantum-limited signal-to-noise ratio [i.e., V(signal + background)] would be —28, and contrast (i.e., signal/background) —3. Fortunately, the picture (or image!) is not as gloomy as it looks at first glance. Our experimental experience has taught us that it is possible, by using reasonably good-yield fluorophores and with careful optimization of microscopic optics, to confocally image tissue in vivo at high resolution and in real time. 3.1

Singlet State Excitation

Any experiment must necessarily begin by estimating the detected signal (number of detected fluorescent photons) for our chosen fluorophore and imaging conditions, so that we understand our chances of success. We expect the detected number (S) of fluorescent photons in a pixel in the image to be: S = FNf,ensTUssueTopt,cstdclcclor

(12)

where F

= fluorescence emission rate (photons per fluorophore molecule per second) N = number of fluorophore molecules in the imaged confocal spot = fiens fraction of the fluorescence emission that is collected by the objective lens TUSSUC = fraction transmitted by tissue T0ptics = fraction transmitted by the confocal microscope optics tdoiccior = detector integration time (pixel time) (sec) For a fluorophore with an emission quantum efficiency 17, absorbance or excitation rate ka (photons/molecule-sec) and decay or emission rate k, (photons/molecule-sec), the fraction of molecules that are in the excited state is given by the ratio k u /(k a + k,). This assumes excitation from the ground state to the lowest singlet state, with the excited singlet state lifetime being Tf. Note that rr = 1/k,. (The triplet state is considered later.) Then, the fluorescence emission rate (photons/molecule-sec) is [58,59] F= Tjk,-k a /(k a + k f )

(13)


153

One can calculate the absorbance or excitation rate (photons/moleculesec) as ka = o-I/hf 2

(14) 2

where a (cm /molecule) is the absorbing cross-section, I (watts/cm ) is the irradiance, and hv (joules/photon) is the energy of a photon (Planck's constant, h = 6.626 X 10~34 joule-second, frequency of the illumination, v c/A, with speed of light c = 3 X 108 m/sec and A being the wavelength). For the excited singlet state condition, the maximal signal-to-noise ratio is obtained under steady state, when the excitation rate ka is equal to the emission rate kt. This condition, ka = kf, then defines the optimum illumination irradiance to be loptimum (watts/cm2) = kfhiVcr

(15)

When the actual illumination irradiance I is less than the optimum (i.e., I < loptimum), the fluorescence emission rate is proportional to I such that the detected signal increases with increased illumination irradiance. However, when the illumination irradiance exceeds the optimum (i.e., I > I op ti murn ), the excited states saturate; with almost all molecules now in the excited state and almost none in the ground state, increasing I does not further increase the signal but causes the background noise to linearly increase. This background noise originates from fluorophore molecules or autofluorescence from out-of-focus (i.e., outside the imaged confocal spot) regions of the tissue. Outside the confocal spot, the fluorophore molecules are still in their ground state and continue to fluoresce because the illumination irradiance there is still less than the optimum. 3.2

Triplet State Excitation

When a fluorophore has a high quantum efficiency for transition from singlet to triplet state (intersystem crossing), a large fraction of the illuminated molecules will be trapped in a long-lived triplet state; consequently, the fluorescence emission decreases. If i7isc is the intersystem crossing efficiency and TT is the triplet state lifetime, then the triplet state will fill up at a rate kT [58,59]: kT = 1/TT + i7Isckfka/(ka + k.)

(16)

In other words, the triplet state fills up in time constant l/k T . In this situation, the fluorescence emission rate is F = 7jk a /[l + k a (l/k f + T/,SCTT)]

(17)

The decrease in fluorescence emission, according to Eq. (17), will hap-

154


pen if the dwell time (i.e., illumination time) per spot exceeds the triplet state lifetime. 3.3

Photobleaching

Photobleaehing occurs when the triplet state in the fluorophore reacts with molecular oxygen, forming a nonfluorescing molecule. Fluorophore molecules will photobleach with a quantum efficiency 77,, and at a rate kh [58,59]: kh = 77 h k l k : /(k a + k, )

( 1 8)

This gives the time constant of photobleaching as r h = l/k h , which is the time after irradiation when only 37% (e ') of molecules remains. (According to Sandison et al. [59|, photobleaching tends to limit the fluorescence emission to a maximum of ~1CP photons/molecule-sec for some of the best fluorophores in living cells.) 3.4

Steady-State Maximum Signal

From Eq. (14), the optimal irradiance I (watts/cm 2 ) is

(19) where ka now depends on whether the excited fluorophore molecules are in the singlet or triplet state. For fast scanning such that the dwell time (i.e., illumination time) at each pixel is shorter than the time required for triplet state fill-up, the maximal signal is obtained under steady-state condition, when the excitation rate equals the emission rate: k., = k, = l/r r . For slow scanning (long dwell times) that allows the triplet state to fill up, the steadystate maximal signal is obtained when k a = I/(T, + TJ IS C T T)-

4.

PERFORMANCE OF A PROTOTYPE VIDEO-RATE CONFOCAL MICROSCOPE

We built a video-rate confocal scanning laser microscope for imaging skin and other tissues in vivo 160-62]. The microscope was originally designed for reflectance, but we modified and used it for fluorescence imaging of the stratum corneum in human skin and the epidermis and dermal microcirculation in small animals. These experiments were performed to investigate feasibility, and the results provide insight into the possibilities (and limitations) of real-time confocal fluorescence imaging in tissue in vivo. Calculated or experimentally measured values for the instrumentation parameters (N, f, ons , TlisM1L., T l)plics , tllL.tc,.lor) of Eq. (12) are shown.


4.1

155

Volume of the Imaged Confocal Spot

High-resolution, high-contrast imaging within living tissue demands water immersion objective lenses with high NA. To visualize nuclear and cellular detail in tissue and blood circulation, the NA must be higher than 0.7 [61]. We usually use an NA of 0.9. The imaged spot, defined by the focused cone of light that is within the confocal section (Fig. 3a), is of volume V = (7r/12)[(Az) 3 tan 2 0 + 6(Ax)(Az) 2 tan 9 + 12(Ax) 2 (Az)]

(20)

where Ax is the lateral resolution and Az is the section thickness (axial resolution) and 9 is the half-angle of the focused cone (NA = n sin 9 for immersion medium of refractive index n). Under ideal, diffraction-limited conditions, Ax and Az are defined by Eqs. (6) and (7). With blue illumination at 488 nm and a water immersion (n = 1.33) objective lens of NA 0.9, the calculated diffraction-limited lateral resolution is 0.3 yum and section thickness is 1.1 /mm. In reality, however, the section thickness depends on the detector aperture (pinhole) size [53], chromatic aberrations due to the optics [59,63,64], and spherical aberrations that occur when imaging deep within tissue [65,66]; at high NAs, the sectioning degrades by a factor of about 2 from the diffraction limit when imaging in skin [61,64,67]. We experimentally measured the lateral resolution (Ax) to be 0.5 /xm and section thickness (Az) to be 2 /xm at a depth of —100 /xm within living human skin when using a pinhole diameter of 10 resels (i.e., v = 15 optical units. One resel is equivalent to the lateral resolution Ax [34], and based on Eq. (6), v - 1.5 optical units when the pinhole diameter is equal to one resel.) Then, the imaged spot volume V is 6 X 10 12 ml. 4.2

Number of Fluorophore Molecules

The number (N) of fluorophore molecules in the imaged confocal spot is easily determined from the dosage (milligrams of fluorophore per kilogram of body weight), known blood volume (milliliters per kilogram of body weight), and the fact that 1 mole is equivalent to the molecular weight in grams and contains Avogadro's number (6 X 1023) of molecules: N = {dosage [mg/kg]/blood volume [ml/kg]} • {imaged spot volume V [ml]} {6 X 1023 [molecules/mole] /molecular weight [g/mole]} {10~ 3 [g/mg]} 4.3

(21)

Fluorescence Light Collected by Objective Lens

Assuming that the fluorescence emission is uniformly distributed in a spherical volume of 4?r steradians about the illuminated spot (Fig. 3a), the ob-


157

jective lens will collect fluorescent photons within the solid angle defined by the NA (NA = n sin 0). That solid angle is 2vr(l — cos 6); thus, the fraction of the fluorescence emission that is collected by the objective lens is flens = C/2)(l

-

COS 9}

(22)

For our water immersion (n = 1 .33), 0.9 NA lens, the fraction turns out to be 13%. 4.4

Fluorescence Light Transmission Through Living Tissue

The fluorescent photons collected by the objective lens must migrate first through the tissue and then through the microscope optics before reaching the detector. Visser et al. [68] report a detailed analytical model to account for and correct for loss of both excitation and fluorescence light in a turbid medium, for reconstructing three-dimensional stacks of confocal images. Based on analytical results from a finite-difference time-domain model [69,70] as well as our experimental data in human skin samples, we expect an exponential (Beer's law) loss of photons in the epidermis and dermis such that the fraction of fluorescent light transmitted by tissue is TIlssue = e~ M ' z

(23)

with )Ltt being the net extinction coefficient due to scattering and absorption, and z the depth of imaging. The values for ^tt reported in the literature vary widely, and hence we choose representative estimates: /JLt = 10 mm" 1 for human epidermis and jiit = 25 mm ' for human dermis at the blue wavelength 488 nm [71]. In the absence of data for small-animal tissues, we will use the same values for animal skin. Assuming illumination at 488 nm (which is the wavelength that we have used most), the measured maximal depth of imaging is —100 /mm. This includes microcirculation in the dermis that typically occurs between 50 and 100 ^tm depth. Given these parameters, the fraction (Ttissue) of fluorescent light transmitted by tissue is estimated to be 60-37% from a depth of 50-100 /mm in the epidermis and 8% from a depth of 100 fjum in the dermis. 4.5

Fluorescence Light Transmission and Detection by the Confocal Microscope Optics

Beyond the tissue, the fraction (Toptics) transmitted by the microscope optics was measured to be 50%. (The transmission of 50% is actually poorer than normal, mainly because our microscope is optimized for reflectance and longer near-infrared wavelengths. Moreover, the optics in the original pro-

158


totype was not of the best quality. A well-designed, well-optimized microscope will generally provide 65-80% transmission, as we have determined in a recent, new setup.) Finally, the fluorescent photons are detected over an integration time (t delcclol ) of 100 nsec/pixel when the imaging is at video rate (30 frames/sec). The above analyisis, from Eqs. (12) to (23), addresses the question of detectability: how many photons per pixel will we detect? This analysis is, of course, somewhat approximate. Also available is a more rigorous Monte Carlo simulation of detected fluorescence in a turbid medium [72]. Although approximate, the analysis has nevertheless proven to be quite useful for predicting detectability, as demonstrated by the experimental results shown below. 4.6

Quantum and Background Noise

Within the detected signal (S), there is quantum noise that varies as the square root of S (i.e., quantum noise = \/S). Moreover, underlying the detected signal will be background noise (B) due to fluorophore molecules that are outside the confocal section, tissue endogenous autofluorescence, Raman fluorescence, and back-scattered (reflected) light that may leak through inefficient filters. The detected signal-to-background ratio determines the contrast in an image, and the signal-to-noise ratio determines its useful information content or image quality [59,73-75]. Gan and Sheppard [56] and Sheppard et al. [57] have presented a thorough analysis of detectability in terms of signal-to-noise ratio, taking into consideration all sources of noise from quantum effects, optical instrumentation, and object background. In fact, they propose detectability to be a rigorous measure of confocal microscope performance. Furthermore, in practical applications involving nonideal objects, such as living tissues, the image contrast and signal-to-noise ratio may also limit the obtainable resolution, i.e., the resolution will likely not be diffraction limited as defined by Eqs. (6) and (7) [34,76]. When the detected signals are low, as may happen from small-object features, the resulting contrast may not be adequate to distinguish between closely spaced object features, leading to loss of resolution. 5.

DETECTABILITY OF COMMON FLUOROPHORES: ANALYTICAL AND EXPERIMENTAL EXAMPLES

Commonly used fluorophores are listed in Tsien and Waggoner [58] and Terasaki and Dailey [77], and Cullander [78] provides excellent practical advice on their usage for confocal imaging. (See also the references in all of these book chapters.) Several general references that list fluorophores and their properties are available [79-86]. The emission quantum yield 17 of


159

fluorophores strongly depends on the microenvironment and thus is often not known within tissues or blood. The best we can do is to assume the quantum yield of fluorophores in their solvents. Nevertheless, such calculations provide an initial estimate of signal level—and indeed whether one might expect a signal at all! We have found it useful to validate the math with calibrated experiments on excised tissue (ex vivo) or in vitro specimens; this helps to estimate the range of optimal instrumentation and imaging parameters for the subsequent in vivo experiments. Our experiments in confocal fluorescence have mostly been to image microvasculature and blood flow in small rats (weight —300 mg, blood volume —70 ml/kg) using a fluorophore dosage of —1 mg/kg, with a 60X, 0.9 NA water immersion objective lens, blue illumination wavelength of 488 nm, and pinhole diameter of 1 mm (i.e., 10 resels for which v = 15 optical units). The confocal spot volume is estimated to be 6 X 10~ 1 2 ml. For the calculations, we assume imaging of microcirculation in the dermis at a depth of 100 /xm, from which the fraction (Tllssue) of fluorescent light transmitted by tissue is 8%. The fraction (Topllcs) transmitted by our confocal microscope optics is 50% and the detector integration time (tdetector) at video rate is 100 nsec/pixel. 5.1

Fluorescein in Aqueous Solution

Fluorescein (molecular weight 376) in aqueous solution has the following properties: emission rate k, = 2.2 X 108 photons/molecule-sec, molar extinction coefficient e = 80,000 L/mole-cm, which is equivalent to an absorbing cross-section a = 3 X I0~' 6 cm2/molecule [a = (In 10)e], excited singlet state lifetime rt = 4.5 nsec, quantum yield rj — 0.9, intersystem crossing efficiency i7,sc = 0.03, triplet state lifetime TT = 1000 nsec, peak excitation wavelength Aex = 490 nm, and peak emission wavelength Aem = 515-520 nm. We have 1.4 X 105 molecules in the imaged confocal spot volume. For the excited singlet state condition, the maximal signal-to-noise ratio is obtained under steady state when the excitation rate ka is equal to the emission rate kr. The math, using Eqs. (13) to (23), then tells us to expect the detected signal to be 7200 photons/pixel or 28 nW. (At 488 nm and videorate pixel integration time of 100 nsec, 1000 photons is equivalent to 4 nW.) This detected fluorescence signal compares well with that detected in reflectance; when imaging skin in reflectance with 1 mW illumination at 1064 nm, we typically collect 1000-10,000 photons/pixel (i.e., 10-100 nW) relative to a background of 100-500 photons/pixel [61]. Figure 4 shows fluorescein-labeled microspheres in rat microcirculation; this demonstrates the rather well-known detectability of fluorescein within blood flow.

160


FIGURE 4 Video-rate confocal image of individual fluorescein-labeled microspheres (arrows) in the microcirculation of a Sprague-Dawley rat. Each microsphere is of diameter 0.5 /am, containing ~106 molecules. Objective lens 60x, 0.9 NA water immersion, pinhole diameter 10 resels, excitation at 488 nm, detection through a 520-nm long-pass filter, illumination power 1-5 mW on the skin surface.

The triplet state fills up at a rate kT = 4.3 X 10 6 per second, which means that it fills up in 232 nsec. At video rates the pixel time is 100 nsec; thus, one tends to ignore the partial fill-up of the triplet state. However, at slower frame rates, one cannot ignore the eventual triplet state fill-up that will reduce the fluorescence emission. In this situation, the maximum signalto-noise ratio is obtained when ka = l/[r, + i7,scrT]. Using Eq. (17), we now find the detected signal to be 950 photons/pixel. This represents an almost eightfold drop in signal, which, to some extent, can be compensated for with the slower frame rate (i.e., longer pixel time or detector integration time). Given that the triplet state lasts for 1000 nsec, longer detection time will certainly help. Fluorescein has a photobleaching quantum efficiency (i7h) of 3 X 10 \ which means that the photobleaching time is 300 jusec. Since this is much longer than the dwell time (i.e., illumination time) per pixel of 100 nsec for video-rate scanning, photobleaching does not occur. The optimum irradiance for maximum signal is then calculated to be 3 X I05 W/cm2 within the microvasculature, which, over a diffraction-limited spot diameter of 0.65 /am (for blue 488-nm argon-ion laser wavelength 488 nm, water immersion 0.9 NA objective lens), corresponds to I mW of illumination power. Assuming the blood vessels to be 100 yum deep in the dermis, which is the maximal


161

depth that can be seen with a video-rate confocal microscope with 488 nm, the maximum illumination power on the skin surface turns out to be 12 mW. This is over a large defocused spot of diameter of about 180 /zm, since the focused illumination spot is actually 100 /zm deep (Fig. 3b), and thus the illumination itself does not cause photodamage to the skin. There is experimental evidence, based on examination of the histology of imaged skin specimens, that shows no damage to the tissue morphology. Topically applied fluorescein and fluorescein compounds have proven to be useful for confocal imaging studies of human skin in vivo. Current areas of research include morphology and permeability of the corneocyte layers within the stratum corneum (Figs. 5-7).

FIGURE 5 Processes of cornification and desquamation, showing the highly ordered lattice of regular polygonal-shaped corneocytes in healthy skin (a) versus the disordered pattern of irregularly shaped corneocytes in unhealthy skin (b), as imaged with a tandem scanning confocal microscope, following topical application of fluorescein. The images are maximal intensity renderings of stacks taken at 2.5-|jim intervals between slices. The images were taken with a Nikon 0.75 NA multi-immersion objective set to glycerol (the immersion fluid was sunflower oil, which has the same refractive index as glycerol), detector slit width of 10 /Jim, at video rate (30 frames/sec), 0.5 /zm per pixel, and no averaging. The excitation wavelength was 488 nm and detection through a 500-nm long-pass filter. The fluorescence contrast is due to the inhomogeneous absorption of fluorescein; differences in shape, size, spatial distribution, and morphology of corneocytes have been quantitatively and qualitatively analyzed from such images. Significant variations in such features are correlated to skin conditions such as age, dryness and disease. (Courtesy of Daan Thorn-Leeson, Kumar Subramanyam, and K. P. Ananth, Unilever US Research, Edgewater, NJ.)

162


FIGURE 6 Confocal fluorescence study of the relative penetration of oiland water-soluble moisturizer/cleanser actives in human stratum corneum in vivo: the hydrophobic active labeled with fluorescein octadecyl ester (ODF) remains trapped in the fine wrinkles and grooves on the surface and is adsorbed in the superficial corneocyte layers (a), whereas the hydrophilic active labeled with fluorescein penetrates through the intercellular spaces to depths of 10-20 yam (b). The active formulations contained —200 ppm of the fluorophores, of which —0.5 ml was applied on a test area of ~3 cm, resulting in an estimated ~107 fluorophore molecules in the confocal spot. Objective lens 60x, 0.9 NA water immersion, pinhole diameter 10 resels, excitation at 488 nm, detection through a 520-nm long-pass filter, illumination power 1-5 mW on the skin, averaging of 2-4 video-rate frames. (Courtesy of Hung-Ta Chang, K. P. Ananth, Unilever US Research, Edgewater, NJ.)

5.2

Fluorescein Isothiocyanate Isomer I (FITC-lsomer I), Conjugated to a Polymer Carrier, in Phosphate-Buffered Saline (PBS) Solution

FITC-lsomer I (molecular weight 389) has the following properties: excited singlet state lifetime r, = 3.8 nsec (emission rate k, = 2.6 X I0x photons/ molecule-sec), molar extinction coefficient e = 68,000 L/mole-cm that is equivalent to an absorbing cross-section or = 2.6 X 10 l6 cmVmolecule, quantum yield 17 = 0.71, peak excitation wavelength Acx = 490 nm, and peak emission wavelength A cm = 519-525 nm. Here we have 1.3 X \(f molecules in the imaged confocal spot volume. Consequently, under excited singlet steady-state conditions (k a = k,), the detected maximal signal is 6400 photons/pixel. The optimal irradiance for maximal signal is calculated to be 4.1 X 10s W/cm 2 within the microvasculature, which is equivalent to 1.3 mW. Assuming, again, the blood vessels to be 100 /xm deep in the dermis, the optimal illumination power


163

FIGURE 7 Simultaneous fluorescence and reflectance stacks of human skin in vivo following topical application of octadecyl fluorescein (ODF), demonstrating the colocalization of the fluorescence signal in the context of the reflected image. The false-colored green areas indicate the three-dimensional distribution of the dye within the stratum corneum, overlying the false-colored red epidermis and dermis. The stacks were imaged with a Noran Oz super-video-rate laser confocal scanner equipped with a contact depth-scanning device. The resolution with high numerical NA objectives (NA = 1.3-1.4) is of the order of A/2 laterally and about A in the axial or depth direction. The instrument can vary its scan rate from 30 frames/sec up to 480 frames/sec with a reduced format, and the scanner is designed to image simultaneously in fluorescence and reflectance. (Courtesy of Ricardo Toledo-Crow, Noran Inc., Middleton, Wl.) (See color insert.)

on the surface of the skin is 16 mW over the defocused spot of diameter about 180 /xm. FITC-Isomer I is certainly observable at video rate with a confocal microscope, as demonstrated by experimental results showing microvasculature and blood flow in Sprague-Dawley rats (Fig. 8). 5.3

Rhodamine B Sulfonyl Chloride, Conjugated to a Polymer Carrier, in PBS Solution

Rhodamine B sulfonyl chloride (molecular weight 577) has the following properties: excited singlet state lifetime r, = 1.0 nsec (i.e., emission rate kf = 109 photons/molecule-sec), molar extinction coefficient s = 93,000 L/mole-cm that is equivalent to an absorbing cross-section a = 3.6 X 10 16 cmVmolecule, quantum yield 17 = 0.25, peak excitation wavelength Aex = 577 nm, and peak emission wavelength Aem = 590 nm.

164


FIGURE 8 Video-rate confocal fluorescence image of microcirculation in the ear of a Sprague-Dawley rat using FITC-lsomer I conjugated to a polymer carrier in PBS solution. Imaging parameters: 60x, 0.9 NA water immersion objective lens, excitation wavelength 488 nm, detection through 520-nm long-pass filter, illumination power 10-20 mW on the skin surface, depth —100 ^m, dosage —1.2 mg/kg injected in the femoral vein (toxic dosage —100 mg/kg). Dark-appearing circulating blood cells were seen within the bright plasma in real time. (Courtesy of Mark Henrichs, Nycomed-Amersham, Wayne, PA.)

In this case, we have 9.1 X I04 molecules in the imaged confocal spot volume, and again, assuming excited singlet steady-state conditions, the detected maximal signal is determined to be 5800 photons/pixel. The optimal irradiance is 11 X 10" W/cm2 within the microvasculature, which corresponds to 3.8 mW, and the optimal illumination power on the skin surface turns out to be 47 mW. Compared with FITC-lsomer I, slightly less fluorescence is expected from rhodamine B sulfonyl chloride. However, the experimental images indicate a somewhat poor signal, probably because the illumination wavelength was far away from that for peak absorbance (Fig. 9). Moreover, the dosage was two orders of magnitude less than (and unnecessarily too far below) the toxic level. This is a good example of detected signal and image quality being unnecessarily compromised because of suboptimal experimental parameters. 5.4

Red-Shifted Green Fluorescent Protein (GFP) in PBS and Protein Solution

Red-shifted GFP (molecular weight 27,000) in PBS-and-protein solution has the following properties: excited singlet state lifetime r, = 3 nsec, emission rate k, = 2.1 X 10X photons/molecule-sec, molar extinction coefficient s 39,200 L/mole-cm that is equivalent to an absorbing cross-section or = 1.5


165

FIGURE 9 Video-rate confocal fluorescence image of microcirculation in the ear of a Sprague-Dawley rat using rhodamine B sulfonyl chloride conjugated to a polymer carrier in PBS solution. Imaging parameters: 60x, 0.9 NA water immersion objective lens, excitation wavelength 488 nm, detection through 540-nm long-pass filter, illumination power 3060 mW on the skin surface, depth —100 ^m, dosage —1.6 mg/kg injected in the femoral vein (toxic dosage —300 mg/kg). The detected signal was poor, probably because the illumination wavelength was far away from the 577 nm for peak absorbance. Moreover, the dosage was unnecessarily far below the toxic level. This is a good example of how detected signal and image quality was unnecessarily compromised because of suboptimal experimental parameters. (Courtesy of Mark Henrichs, Nycomed-Amersham, Wayne, PA.)

X 10 '6 cm2/molecule, quantum yield r\ = 0.66, peak excitation wavelength Aex = 490 nm, and peak emission wavelength Aem = 510 cm. The expressed amount of GFP in the tissue is usually not known, but assuming the concentration to be in the range 1 fjM to 1 mM, we expect 3.6 X 103 to 3.6 X 106 molecules in the imaged confocal spot volume. (This range of concentration was assumed on the basis of measured expression in cultured cells [87].) Assuming, again, excited singlet steady-state conditions (ka = k f ), the detected signal is expected to be 1.3 X 102 to 1.3 X 105 photons/pixel when detecting at 100 jjum depth in the dermis. For the more likely experiment of imaging cells within the superficial epidermis, at a depth of, say, 50 /am, the detected signal will be in the range of 103-106 photons/pixel. GFP is well known to be a stable fluorophore, with a photobleaching rate (kb) that is about sevenfold smaller than that of fluorescein. This gives kb ~ 470 per second which means that the photobleaching time is —2100 /Asec. Since this is much longer than the dwell time (i.e., illumination time) per pixel of 100 nsec for video-rate scanning, photobleaching does not occur. The optimal irradiance for maximal signal is then calculated to be 5.7 X 10s W/cm2 that, over the diffraction-limited spot diameter of

166


0.65 /mm, is equivalent to 1.9 mW. Under these conditions, the expression of GFP in the keratinocytes within the epidermis of a mouse was experimentally observed with about 2 mW of illumination (Fig. 10). 5.5

Benzoporphyrin Derivative-Monoacid Ring A (BPD-MA) in Methanol

BPD-MA (molecular weight 720) in methanol is a photodynamic drug with the following properties: emission rate k, = 1.9 X 108 photons/molecule-sec, molar extinction coefficient e — 15,000-32,000 L/mole-cm at wavelengths 488-690 nm (equivalent absorbing cross-section a - 1.2 X 10 l6 cnrr/molecule at 690 nm), excited singlet state lifetime r, = 5.2 nsec, quantum yield r\ = 0.05, intersystem crossing efficiency i7isc = 0.8, triplet state lifetime rr = 25 îsec, peak excitation wavelength in the red Aex = 690 nm, and peak emission wavelength Acm = 695 nm. The typical dosage in a small animal, such as a rat, is 0.25-4.0 mg/ kg. Here we assume 1 mg/kg, which yields 7.2 X 104 molecules in the confocal spot volume. Under excited singlet steady-state conditions, the maximal detected signal is only 200 photons/pixel. Increasing the dosage to 4 mg/kg yields 800 photons/pixel. BPD-MA has a high intersystem crossing efficiency, and hence the triplet state will fill up at a fairly fast rate (kT = 7.6 X 104 per second), within only 13 nsec. Since the triplet state fills up much faster than, and subsequently has a lifetime (25 /usec) much longer

FIGURE 10 Video-rate confocal image showing the expression of GFP in keratinocytes under the control of VEGF promoter in the epidermis of a transgenic mouse in vivo. Objective lens 60x, 0.9 NA water immersion, pinhole diameter 10 resets, excitation at 488 nm, detection through a 510 ± 5 nm bandpass filter, illumination power 2 mW on the skin, no frame averaging. (Courtesy of Ramnik Xavier, Department of Molecular Biology, Massachusetts General Hospital, Boston, MA.)


167

than, the video-rate pixel time of 100 nsec, the fluorescence emission ceases. Our experimental attempts have, in fact, confirmed this prediction; BPDMA could not be confocally imaged at high resolution and at video rate in rat tumor microvasculature. These experimental examples show that the approximate analysis based on Eqs. (12) to (23) provides a reasonably good prediction of detectability of a given fluorophore within tissue in vivo when imaging in real time with a confocal microscope. Conversely, one can use the analysis to choose fluorophores and optimize the microscope and design experiments for various applications. 6.

INSTRUMENTATION TRADE-OFFS AND OPTIMIZATION FOR SIGNAL DETECTABILITY

The analysis and experimental results lead to an obvious conclusion: detection of fluorescence in a point-scanning confocal microscope may be limited by small imaged spot volumes and short pixel times when imaging at very high resolution (micrometer level) and very high speed such as video rate (100 nsec). Increasing the laser illumination power increases the fluorescence emission signal until either the excited singlet or triplet state is saturated or photobleaching occurs. Photodamage to the tissue also limits the maximal illumination power that may be used. However, increased fluorescence detectability is possible via trade-offs in optical design and imaging parameters. One obvious factor is the careful choice of high-quality optics to maximize light throughput; two especially important components are (1) objective lenses (their NAs vs transmission) to maximize light collection and transmission, and (b) excitation and detection filters (their wavelength bandwidths vs transmission) to maximize light detection. In general, but especially for fluorophores with small Stokes shifts, the detection bandwidth should be as wide as possible even if it means using shorter wavelengths that are away from the peak absorbance wavelength. The reduced absorbance may usually be compensated for with increased illumination power, until the limit of excited state saturation, photobleaching, or photodamage. For improving detectability and signal-to-noise ratio, other factors to trade off are resolution that relates to the imaged spot volume, imaging speed (frame rate) that relates to the detector integration time (pixel time), point versus line scanning, and mechanical tissue-to-microscope coupling. 6.1

Resolution and Imaged Confocal Spot Volume

The resolution and imaged volume is defined by the objective lens NA and detector aperture (pinhole) diameter. High lateral resolution (small illumi-

168


nation spot diameter, small Ax) and high axial resolution (thin section, small Az) results in the detection of a small number of fluorophore molecules within a small imaged confocal volume [Eq. (20)]. Although the above analysis and experimental results are for the high NA of 0.9, a lower NA is often acceptable in many living tissue experiments. For nuclear and cellular detail, NAs as low as 0.7 may be used [21,61,62], which provide diffractionlimited sectioning of about 2 fjum at 488 nm with a water immersion objective lens (this will degrade to approximately 3- to 5-yttm deep within tissue due to spherical aberration and if large pinholes are used). In rabbit corneal wounds, NA of 0.6 with measured ~9 ptm sectioning was adequate for observing cellular structure [16]. Tissue architecture, microvasculature, and blood flow may be imaged with NAs as low as 0.3 (diffraction-limited sectioning — 10 /xm). Such low NAs have been used to visualize microcirculation in rat brain [2], mouse skin [7], and rat colon [20-22]. Even NA on the order of 0.1 has proven useful to image the pharmacokinetics of drug delivery and distribution in relatively thick sections in rat tumors. Dry objective lenses and pinhole diameters of 5-15 resels (or v = 8-25 optical units) provided sectioning of about 300 ûn to 1 mm (both calculated and experimentally measured) that were quite useful to investigate photodynamic drugs such as BPD-MA (Fig. 11) and chlorin-e6 [88], among others.

FIGURE 11 Confocal fluorescence investigation of the pharmacokinetics of liposomal BPD-MA drug delivery in a Lewis lung tumor on a mouse. The BPD-MA is fully within the microvasculature immediately after the intravenous injection (left, at 0 min), but later localizes in the endothelium, and 2 hr later leaks out into the surrounding tumor tissue (right, at 120 min). Imaging parameters: 2.5x, 0.07 NA dry objective lens, pinhole diameter 5-15 resels (v = 8-25 optical units, with measured sectioning 300 /am to 1 mm), excitation wavelengths 458-528 nm (argon ion multilines), detection through 560-nm long-pass filter, averaging 24 video-rate frames, illumination power 10-20 mW on the skin surface, depth —100 [jirr\, dosage 4 mg/kg injected intravenously.


169

Here the trade-off is that lower NAs image larger spots and thicker sections, and thus an increased confocal volume [Eq. (20)], but the collection efficiency decreases [Eq. (22)]. The reduced collection efficiency is, to some extent, compensated for by the higher transmission of objective lenses that have low NAs. Since both sectioning [Eq. (7)] and collection efficiency [Eq. (22)] increase quadratically with NA, whereas the sectioning decreases somewhat linearly with pinhole diameter [Eq. (9); 41,53], a better option is to use a somewhat higher NA than necessary (say, 0.5-0.3 instead of 0.3 — 0.1) to improve collection efficiency and then use a larger pinhole to increase the imaged volume while sacrificing some of the unnecessary sectioning and lateral resolution. 6.2

Frame Rate and Detection Integration Time (Pixel Time)

Although we built a video-rate (30 frames/sec) confocal microscope, our experience in both animal and human studies clearly indicates that video rate is, in fact, often not necessary for in vivo imaging. Imaging at onefourth to one-third video rate (i.e., 7-10 frames/sec) is adequate, provided tissue motion is minimized and the imaging site kept stable relative to the objective lens. Consequently, 3—4 video-rate frames may be integrated with a corresponding increase in detected fluorescence signal. The quantum-limited signal-to-noise ratio would decrease as the square root of the detection integration time [equivalently, \/(number of integrated frames)]. Slow frame rates of 1-16 frames/sec over fields of view varying from 48 X 32 to 768 X 512 pixels have been reported for viewing microcirculation in rat brain cortex [3-6], in mouse skin [7], and in rat colon [2022]. The resulting detector integration times are long, in the range 1-40 yasec/pixel. Of course, the increase in signal detectability is obtained at a price: there is a corresponding loss of temporal resolution. Some of the temporal resolution may be regained by imaging a single line instead of a full frame; for example, Dirnagl et al. [5] and Villringer et al. [6] have employed a method of sequentially scanning a single line, of duration 2 msec, that is oriented to be parallel to a blood vessel so as to observe moving cells within. The sequence of captured lines is then displayed as a spacetime plot from which average blood flow is computed. More recently, line scanning over full fields of view has been effectively used to image morphology and microcirculation. 6.3

Line Scanning

Longer detector integration times (pixel times) can be obtained at the expense of resolution with a line-scanning confocal microscope, in which an

170


illumination line is scanned [89-93]. Consider a frame to consist of 500 lines with each line containing 500 pixels: with point scanning, for videorate imaging (30 frames/sec), the pixel time is —100 nsec, but when a line is scanned at 30 Hz, assuming non-interlaced frames, we obtain video-rate imaging with pixel time of —65 yiisec. Compared with point scanning, there is loss of circular symmetry and lateral and axial (sectioning) resolution along the line [32J, but the sectioning is adequate to visualize nuclear and cellular detail in various tissues. The sectioning is especially well suited for imaging gross tissue morphology, including microvasculature and blood flow. The advantages of a line scanner are that it is simple to build and optimize, has higher light throughput, and uses a standard array detector such as a charge-coupled device (CCD) camera. The detectability in terms of signal-to-noise ratio of a line scanner, in comparison with a point scanner, is described by Sheppard et al. [57]. A line-scanning fiber-based confocal microendoscope has demonstrated visualization of peritoneum and other tissues through abdominal incisions in mice, at 4 frames/sec, with measured lateral resolution of ~3 /xm and section thickness of —10 /^m [23,24]. Lin recently built a video-rate line scanner, adapted from Brakenhoff et al.'s bilateral design [91,92], to visualize microcirculation, including rolling, adhesion, and leukocyte-endothelium interaction in mice [Charles Lin, Dermatology-Wellman Laboratories, Massachusetts General Hospital, personal communication, 2002]. Koester [89,90] developed line scanning with a split objective lens aperture: half of the aperture is used to illuminate and the other half to detect, such that the illumination and detection paths are at an angle to each other, and their intersection precisely defines the confocal section thickness. Although this design has produced remarkable images of nuclear and cellular detail in corneal tissue in vivo in reflectance, it has yet to be used in fluorescence (to the best of the authors' knowledge); however, the method has been used to measure the diffusion of fluorophores across excised rabbit cornea specimens [94]. 6.4

Wide-Field Imaging Methods

To obtain detection times that are even longer than those from line scanning, it may be possible to implement wide-field imaging methods that provide optical sectioning. When used at video rate (30 frames/sec), the detection integration time would be the standrd 33 msec per frame. Two techniques, one based on deconvolution and the other on structured illumination, have been successfully used for confocal fluorescence imaging of cell cultures in vitro and may perhaps find use in tissue in vivo (at least in superficial layers, if not deeper layers).


171

Deconvolution microscopy is a technique in which a stack of blurred images obtained with a conventional microscope are deconvolved with the microscope's PSF to obtain deblurred images of the in-focus planes [95,96]. The images may be recorded in real time but the subsequent deconvolution may require seconds to hours. (This processing time has been reducing and will surely reduce more with the availability of increasingly faster computing power.) Thus, morphology may be visualized, but dynamic events such as blood flow cannot. An interesting possibility is to use confocal reflectance imaging with a wavelength that does not excite the fluorophore, to view and locate the site of interest, followed by recording stacks of fluorescence images for deconvolution. The structured illumination technique is an innovative technique developed by Wilson and his group [97,98] to obtain confocal sectioning with a conventional microscope that is used in its standard configuration to image a wide field of view. Optical sectioning is obtained from a conventional microscope based on the fact that all nonzero spatial frequencies in the object transfer function attenuate with defocus (only the zero spatial frequency does not). By superimposing a specific spatial frequency on the object, one obtains contrast attenuation with defocus (i.e., sectioning). Wilson et al. [98] have presented methods to superimpose a single spatial frequency grid pattern onto cells in culture, and subsequent removal of the grid pattern, to obtain sectioned fluorescence images of the cells. The superimposition of such a grid pattern witin the superficial layers of tissue and imaging with a conventional microscope may provide sectioning in the widefield-of-view images. 6.5

Mechanical Tissue-to-Microscope Coupling

To facilitate slower imaging speeds (low frame rates), the tissue motion relative to the objective lens must be minimized. Fixtures have been devised to mechanically couple the tissue to the objective lens that first precisely locate the site to be imaged and then keep it stable. One that works particularly well on skin is a contact device consisting of a ring-and-template that is first attached to the tissue (with medical-grade adhesive) and then locks into a housing around the objective lens [61,62]; this was modified and effectively used to stabilize other tissues in vivo such as human oral mucosa [99] and, in small animals, intrasurgically exposed bladder [100], pancreas [101], liver, kidney, and other organs. A similar contact device was developed and used for skin by Corcuff et al. [102]. Another method that works well for stabilizing the cornea employs special objective lenses with a dipping cone as the contact element [90]. The dipping cones have been designed such that the curvature of their contact end-surface matches that of the cor-

172


neal surface; thus, the dipping cone stabilizes the corneal surface by gently pressing against it. Other methods include coverslip-glass windows and specialized fixtures that are implanted in the tissue; examples include those in the cranium to observe microvascular blood flow in the brain cortex [3] and in bone [8j. Boyde et al. [8] also designed objective lenses with a nose cone that engaged into a conical receptacle in the implanted fixtures. The design of fixtures and objective lenses that are specialized to stabilize different tissues thus becomes an important factor for real-time in vivo confocal imaging. Blurring due to tissue motion may also be minimized by triggering frame capture from the motion due to respiration or pulse. However, this distorts the confocal plane, especially when imaging at slow frame rates, since the tissue moves during the frame capture. Despite the distortion, this triggering method has proven useful; Villringer et al. [3,4] have managed to capture up to 10 single frames over 10 respiration cycles (each cycle 0.75-1 sec long), and then sum the frames with fairly minimal blurring in rat brain cortex. The distortion may be minimized by using faster frame rates and compensating for the loss of signal by summing a large number of frames. 7.

SUMMARY AND CONCLUSIONS

With careful analysis and attention to experimental detail, it is often possible to detect fluorescence in living tissue in real time (including video rate in many cases) with a confocal microscope. We expect significantly improved detectability and higher quality imaging than that presented in this chapter if well-designed microscopes are developed and the experimental parameters are well-optimized. One must use an analytical model for in vivo confocal fluorescence imaging studies; the analysis may be either approximate (as shown here) or more detailed (that can be developed from the available literature). The analysis allows one to design the confocal optics for maximal light throughput and optimal imaging parameters. Application-specific choices must be made for fluorophore and its concentration, resolution (imaged volume), frame rate (pixel time), and scanning method (resolution vs pixel time). The analytical results should be validated with calibration experiments on excised tissue (ex vivo) or in vitro specimens; this is often useful to estimate optimal instrumentation and imaging parameters for the in vivo experiments. For in vivo applications, imaging at video rate is not always necessary; imaging at much slower speeds (say, one-fourth or onethird video rate or 7-10 frames/sec) may be performed, provided the tissue is well stabilized. The relative motion between the tissue and the objective lens must be minimized to within about ±1 cell (or better); this requires


173

specially designed objective lenses and lens-to-tissue contact devices that are specific to the shape and structure of the tissue. The question of fluorophore-induced toxicity and the mechanism and extent of photodamage in tissues in vivo during confocal imaging is largely unanswered; this calls for a detailed investigation of the effects of illumination and fluorophore on living tissue morphology and function. Visible illumination wavelengths have limited the maximal depth of imaging to 100-200 îm in tissues. Deeper imaging, down to 200-500 îm, should be possible with the advent of near-infrared fluorophores that would allow the use of longer illumination wavelengths. Confocal fluorescence imaging of animal and human tissues in vivo remains to be fully exploited for basic and clinical research.

REFERENCES This list of references has been selected as representative of the literature and is by no means exhaustive. For an exhaustive list, we recommend a MEDLINE and INSPEC search using the keywords confocal fluorescence microscopy, imaging, real time, video rate, in vivo, animal, human, living tissue, skin, blood, tumor, etc. The abbreviation HBCM is used for: JB Pawley, ed. Handbook of Biological Confocal Microscopy. New York: Plenum Press. (This is reference 33.) Two editions of this book are referenced: revised ed., 1990 and 2nd ed., 1995. 1.

M Minsky. Microscopy Apparatus, US Patent 3013467 (filed 7 Nov 1957), 1961. 2. A Villringer, RL Haberl, U Dirnagl, F Anneser, M Verst, KM Einhaupl. Confocal laser microscopy to study microcirculation on the rat brain surface in vivo. Brain Res 504:159-160, 1989. 3. A Villringer, U Dirnagl, A Them, L Schurer, F Krombach, KM Einhaupl. Imaging of leukocytes within the rat brain cortex in vivo. Microvasc Res 42: 305-315, 1991. 4. U Dirnagl, A Villringer, R Gebhardt, RL Haberl, P Schmiedek, KM Einhaupl. Three-dimensional reconstruction of the rat brain cortical microcirculation in vivo. J Cerebr Blood Flow Metab 11:353-360, 1991. 5. U Dirnagl, A Villringer, KM Einhaupl. In-vivo confocal scanning laser microscopy of the cerebral microcirculation. J Microsc 165:147-157, 1992. 6. A Villringer, A Them, U Lindauer, K Einhaupl, U Dirnagl. Capillary perfusion of the rat brain cortex—an in vivo microscopical study. Circ Res 75:55-62, 1994. 7. LJ Bussau, LT Vo, PM Delaney, GD Papworth, DH Barkla, RG King. Fiber optic confocal imaging (FOCI) of keratinocytes, blood vessels and nerves in hairless mouse skin in vivo. J Anat 192:187-194, 1998. 8. A Boyde, LA Wolfe, M Maly, SJ Jones. Vital confocal microscopy in bone. Scanning 17:72-85, 1995.

174


9.

A Boyde, SJ Jones, ML Taylor, LA Wolfe, TF Watson. Fluorescence in the tandem scanning microscope. J Microsc 157:39-49, 1990. H Ichijima, WM Petroll, JV Jester, HD Cavanagh. Confocal microscopic studies of living rabbit cornea treated with benzalkonium chloride. Cornea 11: 221-225, 1992. SJ Jones, ML Taylor. Confocal fluorescence microscopy: some applications in bone cell biology. J Microsc 158:249-259, 1990. A Boyde, G Capasso, RJ Unwin. Conventional and confocal epi-reflection and fluorescence microscopy of the rat kidney in vivo. Exp Nephrol 6:398408, 1998. J Kishimoto. R Ehama, Y Ge, T Kobayashi, T Nishiyama, M Detmar, RE Burgeson. In vivo detection of human vascular endothelial growth factor promoter activity in transgenic mouse skin. Am J Pathol 157:103-110. 2000. J Kacza, J Grosche, J Seeger, K Brauer, G Bruckner, W Hartig. Laser scanning and electron microscopic evidence for rapid and specific in vivo labelling of cholinergic neurons in the rat basal forebrain with fluorochromated antibodies. Brain Res 867:232-238, 2000. SE Ilyn, MC Flynn, CR Plata-Salaman. Fiber-optic monitoring coupled with confocal microscopy for imaging gene expression in vitro and in vivo. J Neurosci Meth 108:91-96, 2001. T Moller-Pedersen, HF Li, WM Petroll, HD Cavanagh, JV Jester. Confocal microscopic characterization of wound repair after photorefractive keratectomy. Invest Ophthalmol Vis Sci 39:487-501, 1998. FA Merchant, SJ Aggarwal, KR Diller, AC Bovik. In-vivo analysis of angiogenesis and revascularization of transplanted pancreatic islets using confocal microscopy. J Microsc 176:262-275, 1994. PJ White, RD Fogarty, IJ Liepe, PM Delaney, GA Werther, CJ Wraight. Live confocal microscopy of oligonucleotide uptake by keratinocytes in human skin grafts on nude mice. J Invest Dermatol 112:887-892, 1999. PM Delaney, MR Harris, RG King. Fiber-optic laser scanning confocal microscope suitable for fluorescence imaging. Appl Opt 33:573-577, 1994. PM Delaney, RG King, JR Lambert, MR Harris. Fiber optic confocal imaging (FOCI) for subsurface microscopy of the colon in vivo. J Anat 184:157-160, 1994. GD Papworth, PM Delaney, LJ Bussau, LT Vo, RG King. In vivo fiber optic confocal imaging of microvasculature and nerves in the rat vas deferens and colon. J Anat 192:489-495, 1998. W McLaren, P Anikijenko, D Barkla, TP Delaney, R King. In vivo detection of experimental ulcerative colitis in rats using fiberoptic confocal imaging (FOCI). Dig Dis Sci 46:2263-2276, 2001. YS Sabharwal, AR Rouse, L Donaldson, MF Hopkins, AF Gmitro. Slit-scanning confocal microendoscope for high-resolution in vivo imaging. Appl Opt 38:7133-7144, 1999. AF Gmitro, AR Rouse. YS Sabharwal. In situ optical biopsy with a confocal microendoscope. Proceedings of the 22nd Annual EMBS International Conference, Chicago. 2000, pp 1040-1042.

10.

11. 12.

13.

14.

15.

16.

17.

18.

19. 20.

21.

22.

23.

24.

In Vivo Confocal Fluorescence Microscopy 25.

26. 27.

28.

29.

30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41.

42.

43.

44. 45.

175

AJ Mueller, WR Freeman, R Folberg, DU Bartsch, A Scheider, U Schaller, A Kampik. Evaluation of microvascularization pattern visibility in human choroidal melanomas: comparison of confocal fluorescein with indocyanine green angiography. Graefes Arch Clin Exp Ophthalmol 237:448-456, 1999. I Foissner, D Wendehenne, C Langebartels, J Durner. In vivo imaging of an elicitor-induced nitric oxide burst in tobacco. Plant J 23:817-824, 2000. JP Verbelen, S Kerstens. Polarization confocal microscopy and congo red fluorescence: a simple and rapid method to determine the mean cellulose fibril orientation in plants. J Microsc 198:101-107, 2000. B Hedtke, M Meixner, S Gillandt, E Richter, T Borner, A Weihe. Green fluorescent protein as a marker to investigate targeting of organellar RNA polymerases of higher plants in vivo. Plant J 17:557-561, 1999. RH Kohler, WR Zipfel, WW Webb, MR Hanson. The green fluorescent protein as a marker to visualize plant mitochondria in vivo. Plant J 11:613-621, 1997. GJ Brakenhoff, K Visscher. Chapter 21, Real-time stereo (3D) confocal microscopy. In: HBCM, 1995, pp 355-362. T Wilson, CJR Sheppard. Theory and Practice of Scanning Optical Microscopy. New York: Academic Press, 1984. T Wilson, ed. Confocal Microscopy. San Diego: Academic Press, 1990. JB Pawley, ed. Handbook of Biological Confocal Microscopy, 2nd ed. New York: Plenum Press, 1995. RH Webb. Confocal optical microscopy. Rep Prog Phys 59:427-471, 1996. M Born, E Wolf. Principles of Optics, 6th ed. New York: Pergamon Press, 1991. JW Goodman. Introduction to Fourier Optics. New York: McGraw-Hill, 1968. CJR Sheppard, A Choudhury. Image formation in the scanning microscope. Opt Acta 24:1051-1073, 1977. GJ Brakenhoff, P Blom, P Barends. Confocal scanning light microscopy with high aperture immersion lenses. J Microsc 117:219-232, 1979. T Wilson, AR Carlini. Three-dimensional imaging in confocal imaging systems with finite sized detectors. J Microsc 149:51-66, 1988. CJR Sheppard. Super-resolution in confocal imaging. Optik 80:53-54, 1988. GJ Brakenhoff, HTM van der Voort, EA van Spronsen, N Nanninga. Threedimensional imaging in fluorescence by confocal scanning microscopy. J Microsc 153:151-159, 1989. HTM van der Voort, GJ Brakenhoff. Three-dimensional image formation in high-aperture fluorescence confocal microscopy: a numerical analysis. J Microsc 158:43-54, 1990. GJ Brakenhoff, K Visscher, HTM van der Voort. Chapter 8, Size and shape of the confocal spot: control and relation to 3D imaging and image processing. In: HBCM, 1990, pp 87-91. IJ Cox, CJR Sheppard, T Wilson. Super-resolution by confocal fluorescence microscopy. Optik 60:391-396, 1981. RW Wijnaendts-van-Resandt, HJB Marsman, R Kaplan, J Davoust, EHK Stel-

176

46. 47. 48. 49. 50. 51.

52.

53. 54. 55. 56. 57. 58. 59.

60.

61.

62.

63.

Rajadhyaksha and Gonzalez zer, R Strieker. Optical fluorescence microscopy in three dimensions: microtomoscopy. J Microsc 138:29-34, 1985. T Wilson. Optical sectioning in confocal fluorescent microscopes. J Microsc 154:143-156, 1989. CJR Sheppard. Axial resolution of confocal fluorescence microscopy. J Microsc 154:237-241, 1989. S Kimura, C Munakata. Depth resolution of the fluorescent confocal scanning microscope. Appl Opt 29:489-494, 1990. T Wilson, SJ Hewlett. Imaging in scanning microscopes with slit-shaped detectors. J Microsc 160:115-139, 1990. CJR Sheppard, M Gu. Imaging performance of confocal fluorescence microscopes with finite-sized source. J Mod Opt 41:1521-1530, 1994. O Nakamura. Three-dimensional imaging characteristics of laser scan fluorescence microscopy: two-photon excitation vs. single-photon excitation. Optik 93:39-42. 1993. V Drazic. Three-dimensional transfer function analysis of a confocal fluorescence microscope with a finite-sized source and detector. J Mod Opt 40:879887, 1993. T Wilson. Chapter 11, The role of the pinhole in confocal imaging systems. In: HBCM, 1995. pp 167-182. M Gu, CJR Sheppard. Effects of a finite-sized detector on the OTF of confocal fluorescent microscopy. Optik 89:65-69, 1991. M Gu, CJR Sheppard. Confocal fluorescent microscopy with a finite-sized circular detector. J Opt Soc Am A 9:151-153, 1992. XS Gan, CJR Sheppard. Detectability: a new criterion for evaluation of the confocal microscope. Scanning 15:187-192, 1993. CJR Sheppard, X Gan. M Gu, M Roy. Chapter 22, Signal-to-noise ratio in confocal microscopes. In: HBCM, 1995, pp 363-371. RY Tsien, A Waggoner. Chapter 16, Fluorophores for confocal microscopy— photophysics and photochemistry. In: HBCM, 1995, pp 267-279. DR Sandison, RM Williams, KS Wells, J Strickler, WW Webb. Chapter 3, Quantitative fluorescence confocal laser scanning microscopy (CLSM). In: HBCM, 1995. pp 39-53. M Rajadhyaksha, M Grossman, D Esterowitz, RH Webb, RR Anderson. In vivo confocal scanning laser microscopy of human skin: melanin provides strong contrast. J Invest Dermatol 104:946-952, 1995. M Rajadhyaksha, RR Anderson, RH Webb. Video-rate confocal scanning laser microscope for imaging human tissues in vivo. Appl Opt 38:2105-2115, 1999. M Rajadhyaksha, S Gonzalez, JM Zavislan, RR Anderson, RH Webb. In vivo confocal scanning laser microscopy of human skin. II: Advances in instrumentation and comparison with histology. J Invest Dermatol 113:293-303, 1999. HE Keller. Chapter 7, Objective lenses for confocal microscopy. In: HBCM, 1995. pp 111-126.

In Vivo Confocal Fluorescence Microscopy 64.

177

SW Hell, EHK Stelzer. Chapter 20, Lens aberrations in confocal fluorescence microscopy. In: HBCM, 1995, pp 347-354. 65. S Hell, G Reiner, C Cremer, EHK Stelzer. Aberrations in confocal fluorescence microscopy induced by mismatches in refractive index. J Microsc 169: 391-405, 1993. 66. H Jacobsen, SW Hell. Effect of the specimen refractive index on the imaging of a confocal fluorescence microscope employing high aperture oil immersion lenses. Bioimaging 3:39-47, 1995. 67. DS Wan, M Rajadhyaksha, RH Webb. Analysis of spherical aberration of a water immersion objective: application to specimens with refractive indices 1.33-1.40. J Microsc 197:274-284, 2000. 68. TD Visser, FCA Groen, GJ Brakenhoff. Absorption and scattering correction in fluorescence confocal microscopy. J Microsc 163:189-200, 1991. 69. AK Dunn. Light scattering properties of cells. PhD thesis, University of Texas, Austin, 1997. 70. AK Dunn, C Smithpeter, AJ Welch, R Richards-Kortum. Finite-difference time-domain simulation of light scattering from single cells. J Biomed Opt 2: 262-266, 1997. 71. WF Cheong, SA Prahl, AJ Welch. A review of the optical properties of biological tissue. IEEE J Quant Electr 26:2166-2185, 1990. 72. C Saloma, C Palmes-Saloma, H Kondoh. Site-specific confocal fluorescence imaging of biological microstructures in a turbid medium. Phys Med Biol 43: 1741-1759, 1998. 73. DR Sandison, WW Webb. Background rejection and signal-to-noise optimization in confocal and alternative fluorescence microscopes. Appl Opt 33: 603-615, 1994. 74. DR Sandison, DW Piston, RM Williams, WW Webb. Quantitative comparison of background rejection, signal-to-noise ratio, and resolution in confocal and full-field laser scanner microscopes. Appl Opt 34:3576-3588, 1995. 75. R Gauderon, CJR Sheppard. Effect of a finite-size pinhole on noise performance in single-, two-, and three-photon confocal fluorescence microscopy. Appl Opt 38:3562-3565, 1999. 76. EHK Stelzer. Contrast, resolution, pixelation, dynamic range and signal-tonoise ratio: fundamental limits to resolution in fluorescence light microscopy. J Microsc 189:15-24, 1998. 77. M Terasaki, ME Dailey. Chapter 19, Confocal microscopy of living cells. In: HBCM, 1995, pp 327-346. 78. C Cullander. Chapter 3, Fluorescent probes for confocal microscopy. Meth Mol Biol 122:59-73, 1999. 79. JC Scaiano, ed. CRC Handbook of Organic Photochemistry, Vol 1. Boca Raton, FL: CRC Press, 1989. 80. IB Berlman. Handbook of Fluorescence Spectra of Aromatic Molecules, 2nd ed. New York: Academic Press, 1971. 81. SL Murov, I Carmichael, GL Hug. Handbook of Photochemistry, 2nd ed. New York: Marcel Dekker, 1993.

178 82.


W Horspool, P Song. eds. CRC Handbook of Organic Photochemistry and Photobiology. Boca Raton, FL: CRC Press. 1995. 83. RY Tsien. Fluorescent probes of cell signaling. Annu Rev Neurosci 12:227— 253. 1989. 84. RY Tsien. Chapter 5, Fluorescent indicators of ion concentrations. Methods Cell Biol 30:127-156, 1989. 85. A Waggoner, R DeBiasio, P Conrad, GR Bright, L Ernst, K Ryan, M Nederlof, D Taylor. Chapter 17, Multiple spectral parameter imaging. Meth Cell Biol 30:449-478, 1989. 86. RF Chen, CH Scott. Atlas of fluorescence spectra and lifetimes of dyes attached to protein. Anal Lett 18:383-421, 1985. 87. B Herman. Fluorescence Microscopy, 2nd ed. New York: Springer-Verlag, 1998. p 64. 88. MR Hamblin. M Rajadhyaksha, T Momma, NS Soukos, T Hasan. In vivo fluorescence imaging of the transport of charged chlorin-e6 conjugates in a rat orthotopic prostate tumor. Br J Cancer 81:261-268, 1999. 89. CJ Koester. Scanning mirror microscope with optical sectioning characteristics: applications to ophthalmology. Appl Opt 19:1749-1757, 1980. 90. CJ Koester. JD Auran, HD Rosskothen, GJ Florakis, RB Tackaberry. Clinical microscopy of the cornea utilizing optical sectioning and a high numericalaperture objective. J Opt Soc Am A 10:1670-1679, 1993. 91. GJ Brakenhoff, J Visscher. Confocal imaging with bilateral scanning and array detectors. J Microsc 165:139-146, 1992. 92. GJ Brakenhoff, J Visscher. Imaging modes for bilateral confocal scanning microscopy. J Microsc 171:17-26, 1993. 93. WB Amos, JG White. Chapter 25, Direct view confocal imaging systems using a slit aperture. In: HBCM, 1995, pp 403-416. 94. SP Srinivas, DM Maurice. A microfluorometer for measuring diffusion of fluorophores across the cornea. IEEE Trans Biomed Eng 39:1283-1291, 1992. 95. WA Carrington, KE Fogarty, L Lifschitz, FS Fay. Chapter 14. Three-dimensional imaging on confocal and wide-field microscopes. In: HBCM, 1990, pp 151-161. 96. PJ Shaw. Chapter 23, Comparison of wide-field/deconvolution and confocal microscopy for 3D imaging. In: HBCM, 1995, pp 373-387. 97. MAA Neil, R Juskaitis, T Wilson. Method of obtaining optical sectioning by using structured light in a conventional microscope. Opt Lett 22:1905-1907, 1997. 98. MAA Neil, A Squire, R Juskaitis, PIH Bastiaens, T Wilson. Wide-field optically sectioning fluorescence microscopy with laser illumination. J Microsc 197:1-4. 2000. 99. WM White, M Rajadhyaksha, S Gonzalez, RL Fabian, RR Anderson. Noninvasive imaging of human oral mucosa in vivo by confocal reflectance microscopy. Laryngoscope 109:1709-1717, 1999. 100. F Koenig, S Gonzalez, WM White, M Lein, M Rajadhyaksha. Near-infrared


101.

102.

179

confocal laser scanning microscopy of bladder tissue in vivo. Urology 53: 853-857, 1999. T Keck, V Campo-Ruiz, AL Warshaw, RR Anderson, CF Castillo, S Gonzalez. Evaluation of morphology and microcirculation of the pancreas by ex vivo and in vivo reflectance confocal microscopy. Pancreatology 1:48-57, 2001. P Corcuff, C Chaussepied, G Madry, C Hadjur. Skin optics revisited by in vivo confocal microscopy: melanin and sun exposure. J Cosmet Sci 52:91102, 2001.

Two-Photon Microscopy of Tissues Peter T. C. So, Ki H. Kim, and Lily Hsu Massachusetts Institute of Technology, Cambridge, Massachusetts, U.S.A. Peter Kaplan and Tom Hacewicz Unilever Edgewater Laboratory, Edgewater, New Jersey, U.S.A. Chen Y. Dong National Taiwan University, Taipei, Taiwan Urs Greuter, Nick Schlumpf, and Christof Buehler Paul Schiller Institut, Villigen, Switzerland

1.

INTRODUCTION

Tissue physiology and pathology can be better understood using two-photon microscopy that allows tissue structure and biochemistry to be assayed on the subcellular level. Histological analysis provides high-resolution tissue characterization and is the only imaging method that can be applied to tissues of arbitrary thicknesses. While histology is a powerful and well-accepted approach, it is inherently two-dimensional (2-D) and often misses the three-dimensional (3-D) aspects of tissue architecture. Furthermore, histological imaging can only be applied to specimens after chemical fixation and processing; in vivo imaging of tissue physiology is impossible. Much 181

So et al.

182

difficulty has been overcome with the introduction of confocal microscopy and two-photon microscopy. 1.1

Overview of Confocal and Two-Photon 3-D Microscopy

Scanning confocal microscopy was invented in 1962 by Minsky demonstrating that 3-D resolved images can be obtained in translucent specimens without physical sectioning [1,2]. 3-D resolution in confocal microscopy is obtained by placing a pinhole aperture in the emission light path at a conjugate location of the focal volume in the specimen. Photons generated inside this volume will be focused at the pinhole aperture and can be transmitted to the detector. However, photons originated outside this focal volume will be defocused at the aperture plane and will be blocked. Confocal microscopy obtains 3-D resolution by limitation of the region of observation. Additional discussions on confocal microscopy can be found in a different chapter of this volume. An alternative 3-D imaging technology is two-photon excitation microscopy, which was introduced by Denk, Webb, and coworkers in 1990 [3]. The electronic transition of a fluorophore can be induced by the simultaneous absorption of two photons. These two photons, typically in the infrared spectral range, have energies approximately equal to half of the energetic difference between the ground and excited electronic states (Fig. la). Since the two-photon excitation probability is significantly less than the one-photon probability, two-photon excitation occurs with appreciable rates only in regions of high temporal and spatial photon concentration. The high spatial concentration of photons can be achieved by focusing the laser beam with a high numerical aperture (NA) objective to a diffraction-limited focus. The high temporal concentration of photons is made possible by the availability of high peak power mode-locked lasers. Depth discrimination is the most important feature of two-photon microscopy. For one-photon excitation in a spatially uniform fluorescent sample, equal fluorescence intensities are generated from each z section above and below the focal plane assuming negligible excitation attenuation. However, in the two-photon case more than 80% of the total fluorescence intensity comes from a l-/xm-thick region about the focal point for objectives with NA 1.25. Thus, 3-D images can be constructed as in confocal microscopy, but without a confocal pinhole. This depth discrimination effect of the two-photon excitation arises from the quadratic dependence of two-photon fluorescence intensity upon the excitation photon flux, which decreases rapidly away from the focal plane. For a 1.25 NA objective using excitation wavelength of 960 nm, the typical point spread function has full width at half maximum (FWHM) of 0.3 /xm in the radial direction and 0.9 /xm in the axial direction (Fig. I b ) .

183

Two-Photon Microscopy of Tissues

(a) first electronic excited state

N

wwv

./ emission photon WWV

one-photon excitation

fluorescence emissior

excitation photon

\

excitation photons

vibrational relaxation \ X

/ emission photon

^ r

two-photon excitation

wwv fluorescence emissior

electronic ground state

(b)

vibrational states

1.0 0.8 0.6 0.4 0.2 0.0 -0.6-0.4-0.2 0.0 0.2 0.4 0.6-1.5-1.0-0.5 0.0 0.5 1.0 1.5 Radi al Re s oluti on (urn)

Axi al Res oluti on (um)

FIGURE 1 (a) Jablonski diagrams of one-photon (left) and two-photon (right) excitation, (b) Lateral and axial point spread functions of twophoton microscopy measured using 960 nm excitation with 1.25 NA objective. Lateral FWHM is 0.3 ^m and axial FWHM is 0.9 ^m.

It is useful to compare and contrast confocal and two-photon microscopy techniques. Confocal microscopy has a number of strengths. (1) Confocal microscopy can be operated in two modes: reflected light and fluorescence. Two-photon microscopy, on the other hand, can only be operated in the fluorescence mode. For deep tissue imaging, the applicability of fluorescence confocal microscopy is limited due to the strong tissue absorption of the one-photon excitation wavelengths (ultraviolet and blue spectral

184

So et al.

range); tissue penetration of these wavelengths is limited to 20-40 While the development of new infrared fluorophores may changes this situation, the availability of infrared fluorophores is limited. On the other hand, reflected light confocal microscopy has a penetration depth similar to that of two-photon microscopy. (2) Confocal microscopy has superior resolution as compared with two-photon excitation. For the excitation of the same fluorophore, two-photon resolution is roughly half the one-photon confocal resolution. This lower spatial resolution is due to the use of longer wavelength light. While two-photon microscopy has lower intrinsic resolution and cannot operate in reflected light mode, two-photon microscopy has a number of advantages over confocal detection for biological specimens: (1) Confocal imaging achieves 3-D resolution by using a pinhole to reject out-of-focalplane signal. In contrast, two-photon excitation achieves a similar effect by limiting the excitation region to a submicrometer volume at the focal point. The ability to limit the region of excitation instead of the region of detection is important. Photodamage of biological specimens is restricted to the focal point. Since out-of-plane fluorophores are not excited, they are not subject to photobleaching. The minimally invasive nature of two-photon imaging can be best appreciated in a number of embryology studies. Previous work on long-term monitoring of C. elegans and hamster embryos using confocal microscopy failed because of photodamage-induced developmental arrest. However, recent two-photon microscopy studies indicate that the embryos of these organisms can be imaged repeatedly over the course of hours without observable damage [4-6]. For instance, a hamster embryo was imaged with two-photon microscopy and then reimplanted in the uterus, and it eventually developed into a healthy adult animal. (2) Two-photon excitation wavelengths are red shifted to approximately twice the one-photon excitation wavelengths. The significantly lower absorption and scattering coefficients ensure deeper tissue penetration. (3) The wide separation between the excitation and emission spectra ensures that the excitation light and the Raman scattering can be rejected without filtering out any of the fluorescence photons, resulting in sensitivity enhancement with better signal-to-background ratio. 1.2 Two-Photon Microscopy Instrumentation The design of a typical two-photon fluorescence microscope design is based on three basic components: a femtosecond laser system, high-throughput microscope optics, and high-sensitivity detection optoelectronics (Fig. 2). A femtosecond titanium-sapphire laser (Mira 900, Coherent Inc., Santa Clara, CA), pumped by a solid-state, frequency-doubled Nd:YVO4 laser (Verdi X,


185

Sample

PiezoZ-stage

Dichroic mirror Modified fluorescence microscope

FIGURE 2 Schematics of a typical two-photon microscope.

Coherent Inc., Santa Clara, CA), is the light source for two-photon excitation. The broad lasing range (about 700-1000 nm) of the titanium-sapphire laser enables it to be a versatile excitation source for a wide range of visible, fluorescent molecules. Prior to entering a modified fluorescent microscope (Axiovert 100TV, Carl Zeiss Inc., Thornwood, NY), the laser beam is deflected by a galvanometric x-y scanner (Model 6350, Cambridge Technology, Watertown, MA) which allows faster scanning of the laser focal point with a line scanning rate of about 500 Hz. A beam expander further expands the laser spot diameter to overfill the microscope objective's back aperture and results in diffraction-limited focusing. Fluorescence generated by the two-photon excitation is collected by the microscope objective in an epi-illuminated geometry. For deep tissue studies, water immersion high-NA objectives should be used to ensure index matching. A typical objective used this system is a Zeiss 63x C-Apochromat (NA 1.2) lens. The imaging depth is partly limited by the objectives' working distances, which are typically on the order of 200 ^tm. Longer working distances (up to about 1 mm) water immersion objectives are available but with lower NA. The imaging depth is controlled by the distance between

So et al.

186

the objective and the specimen, which is adjusted using a piezo objective positioner (P-721 PIFOC, Physik Instrumente, Germany). Piezo objective positioners provide high-speed positional control but have a limited displacement range of a couple hundred micrometers. Servo motor coupled to the height control of the microscope stage provides lower bandwidth control of the imaging depth but with centimeter range translation. The fluorescence passes through a dichroic mirror and additional optical filters before reaching the photomultiplier tube (PMT). Signal is processed by single photon counting electronics to ensure high-sensitivity detection. Current pulses generated by individual photons are amplified and separated from background noise via a photon discriminator. The number of photons collected at each pixel is counted by the digital interface electronics and is recorded by the data acquisition computer. The data acquisition computer also handled image archival and 3-D image rendering. 2.

FLUOROPHORES FOR TWO-PHOTON TISSUE IMAGING AND SPECTROSCOPY

An important advantage of two-photon imaging is its suitability for in vivo imaging. In vivo imaging poses severe limits on the choice of tissue-labeling procedures and probes. Fluorophores utilized in two-photon tissue imaging can be divided into two classes: endogenous and exogenous. 2.1

Tissue Imaging Based on Endogenous Fluorophores

Biological tissues often have endogenous fluorescent molecules that can be imaged based on two-photon excitation. Most proteins are fluorescent due to the presence of tryptophan and tyrosine. Two-photon-induced fluorescence from tryptophan and tyrosine in proteins has been investigated [7-9]. However, two-photon tissue imaging based on amino acid fluorescence is uncommon because these amino acids have one-photon absorption in the UV spectral range (250-300 nm). Two-photon excitation of these fluorophores requires femtosecond laser sources with emission in the range of 500-600 nm that is not easily available. The transmission of the UV emission from these fluorophores is also significantly attenuated by typical microscopes with glass optics. Finally, since the distribution of proteins with tryptophan and tyrosine residues is fairly uniform, these amino acids often do not provide useful contrast to produce images in which various tissue structures can be distinguished. Two-photon imaging of cellular structures is often based on the fluorescence of endogenous /3-nicotinamide adenine dinucleotide phosphate (NAD(P)H) [9]. The production of NAD(P)H is associated with the cellular


187

metabolism and the intracellular redox state [10]. Typically, free NAD(P)H absorbs in the UV region at about 340 nm and fluoresces with a maximum at 460 nm. Free NAD(P)H has a fluorescence decay time of about 400 psec. Bound NAD(P)H has a blue-shifted emission spectrum and a lifetime of 2 nsec. Typically, a higher metabolic rate generates a higher concentration of NAD(P)H and produces brighter images. NAD(P)H fluorescence has been used to monitor redox state in cornea [11] and skin [12]. Another class of fluorophores commonly present in cells that can be utilized for two-photon imaging is flavoprotein. Flavoproteins are present almost exclusively in cellular mitochondria. Flavoproteins have a one-photon excitation range around 450 nm and emission in the range of 550 nm. Two-photon excitation spectrum of flavin mononucleotide has been measured [13]. Two-photon microscopy can also image extracellular matrix structures in tissues. Two of the key components in the extracellular matrix are collagen and elastin. Tropocollagen and collagen fibers typically have absorption ranges of 280-300 nm and emission ranges of 350 to 400 nm [14-16]. Elastin has been observed to have a redder emission spectrum at 400-450 nm with excitation wavelengths of 340-370 nm [17-19]. Elastin fibers can be readily imaged by two-photon excitation in the wavelength range of about 800 nm. While collagen can also be excited, its excitation spectrum is similar to that of tryptophan and not readily accessible in most two-photon systems. However, collagen is efficient in second harmonic generation due to its chiral crystalline structure [20-26]. Since second harmonic generation is also a quadratic function of the excitation peak power, second harmonic signal from collagen is readily generated in a standard two-photon microscope. Since second harmonic generated light is scattered primarily in a forward direction, it is best detected in transmission geometry. Although transmission geometry is generally not practical for biomedical imaging of tissues, the multiple scattering characteristics of most tissues ensures that a significant level of second harmonic generated signal can be detected in an epi-illuminated geometry. In addition to natural fluorophores present in tissues, it has been shown that new fluorophores can also be produced by multiphoton-induced chemical reactions. The definitive study that demonstrate the utility of this approach is the conversion of a serotonin molecule to a fluorescent form based on multiphoton-induced reaction. The fluorescence product of serotonin is subsequently imaged for the study of neurotransmitter release processes [27]. 2.2 Tissue Imaging Based on Exogenous Fluorophores The possibility of studying 3-D tissue structures in vivo is responsible for the growing popularity of two-photon microscopy. While autofluorescent

So et al.

188

tissue structures can be readily studied, the study of nonfluorescent components is difficult. Furthermore, the power of fluorescent microscopy often derives from its ability to measure tissue biochemical content and process in situ. The ability of native endogenous fluorophores to monitor tissue biochemistry is mostly restricted to redox reactions. There is a need for developing tissue imaging-compatible fluorescent probes that can monitor tissue structures and biochemical processes. A major difficulty of developing these markers lies in the virtual impossibility of delivering exogenous probes uniformly into intact, in vivo tissues. One of the most important recent innovations in molecular biology that address this problem is the development of fluorescent protein technology. Fluorescent protein technology is a powerful technique to fluorescently label cells and tissues in vivo [28]. After the first introduction of green fluorescent proteins, fluorescence proteins of a variety of colors, including blue, cyan, yellow, and red, have been developed [29,30]. These fluorescent proteins have their origins in marine animals; however, these proteins can be effectively transfected into and expressed in mammalian cells with minimal cytotoxicity. More importantly, the vectors of these fluorescence proteins can be strategically placed into specific loci in the genome of the host cells, allowing the monitoring of the expression of specific genes. Vectors coding for a particular protein of interest can be covalently linked to a fluorescent protein, providing a method to monitor protein distribution and trafficking. Equally importantly, novel fluorescent proteins that are spectrally sensitive to their biochemical environment, such as calcium concentration, have been developed [31]. It is promising that other biochemically sensitive fluorescent proteins can be created [32]. Two-photon imaging parameters for a number of these fluorescent proteins have been optimized [33,34]. 3.

EFFECTS OF TISSUE OPTICAL PARAMETERS ON TWO-PHOTON MICROSCOPY

The optical property of tissues can be characterized by their scattering and absorption coefficients and the index of refraction distribution. Scattering and index of refraction variations limit laser power transmission and signal detection. These factors further cause image resolution degradation. Tissue absorption of infrared light can result in thermal and other optical damage that is particularly important for in vivo imaging. 3.1

Two-Photon Excitation Efficiency as a Function of Tissue Depth

The imaging depth of two-photon microscopy is limited by the efficacy of delivering femtosecond laser pulses to tissues. If one assumes that the image


189

point spread function is invariant with depth, an assumption that is valid for moderate depth on the order of a few hundred micrometers in typical tissues, excitation light penetration is limited by power delivery and pulse broadening. In terms of power delivery, average laser power as a function of depth is an exponential function of depth: 7(Z) = 7oe-[«-^+^k where 7(z) is the intensity at depth z, 70 is the intensity at tissue surface, ^ts is the scattering coefficient and /xa the absorption coefficient, and g is the average cosine of the scattering angle. The scattering coefficient, absorption coefficient, and average cosine for typical tissues are 20-200 mm"', 0.1-1 mm" 1 , and 0.7-0.9. Since the fluorescence signal depends quadratically on the excitation power, the fluorescence signal has a mean free path of 25250 /mm. Laser pulses, typically on the order of 150 fsec from titanium-sapphire lasers, can be broadened in the optical path and two-photon excitation efficiency decreases linearly with pulse width. Laser pulse dispersion in microscope objectives are significant and pulse width can easily broaden to beyond 500 fsec. Some improvement in power can be achieved by using dispersion compensation optics to maintain a short pulse width after the objective. However, even if pulse dispersion through the optics is corrected, further pulse dispersion may occur in the tissues due to scattering and index of refraction heterogeneity. The severity of the broadening effect as femtosecond pulses travel through tissues has not been extensively studied. 3.2

Fluorescence Signal Detection from Multiple Scattering Specimens

Similar to the attenuation of the excitation light, fluorescence signal emitted from the focal volume will be scattered and absorbed by the surrounding tissue. The absorption and scattering effects are minimized for longer emission wavelengths. The development of longer wavelength-emitting fluorophores is important for deep tissue imaging. As compared with confocal microscopy, two-photon detection is more efficient in detecting scattered photons. Since the mean free path of visible photons is less than 100 /xm in typical tissues, the emitted photon will encounter one or more scattering events for deep tissue imaging. However, many of these scattered photons can still be detected since photons are primarily forward scattered in most tissues. In confocal microscopy, these scattered photons are lost at the confocal aperture. However, in two-photon microscopy, if the trajectories of these scattered photons remain within the collection solid angle of the objective lens, they can be directed to the detection light path of the microscope. Since some of these scattered photons follow more divergent optical

So et al.

190

paths, maximizing the collection solid angles and physical sizes of the detection optical components can improve detected signal levels [35]. 3.3

Limitations of Image Resolution in Highly Scattering Medium with Heterogeneous Index of Refractions

Long-range propagation of light through tissues also may result in contrast and resolution losses. The two major factors responsible for resolution degradation are the scattering of the excitation and emission light, and index of refraction mismatches between the sample and the immersion media of the objective. Scattering in a tissue medium leads to degradation of the image due to the deflection of both the excitation and the emission photons. An excellent study investigating the resolution degradation in two-photon microscopy due to scattering was done by Dunn et al. [36]. With Monte Carlo simulations they were able to separate and quantify the reduction of intensity into effects due to either the scattering of the excitation light or scattering of the emission light. The simulations established that the loss of signal intensity was due more to the scattering of the excitation photons than the emission photons. More importantly, they concluded that the signal loss is the limiting factor in two-photon microscope image quality rather than resolution loss by showing that resolution did not degrade as the image depth was increased. Their simulation was supported with experimental point spread function (PSF) measurement, and that is in agreement with Centonze et al. [37J. Dong and coworkers (manuscript in preparation) measured two-photon microscopy PSF directly by imaging l()0-nm spheres in agarose gels with different concentration of scatterers. The concentration of scatterers had no effect on the lateral and axial FWHM of the PSF for both oil and water immersion objectives. Furthermore, the water immersion objective suffered no significant degradation on the lateral and axial FWHM of the PSF as the objective was focused more deeply into the sample. On the other hand, the oil immersion objective suffered significant degradation due the spherical aberration introduced by index of refraction mismatch of the sample, and the lateral and axial FWHM was broadened. One may conclude that for typical tissue thickness on the order of a few hundred micrometers, resolution loss due to scattering is insignificant but index refraction mismatch can be a major factor. The use of water immersion objectives is preferred in biological tissues with a relatively homogeneous index of refraction distribution. However, Dong et al. also investigated the performance of oil versus water immersion objectives for imaging more optically heterogeneous biological tissues such as excised human skin tissue. Human skin is an optically heterogeneous


191

tissue with significant index of refraction variations for different structural layers. It is found that there is significant aberration in the imaging of dermal structures for both oil and water immersion objectives. These objectives perform equivalently because neither can match the varying index of refraction in the skin and both suffer from spherical aberration. For optically heterogeneous samples, it is best to choose an immersion media that best matches the average index of the sample. 3.4

Photodamage Mechanisms in Tissues

Photodamage is an important consideration. One- and higher order photon absorption results in photostress to the biological specimen. However, since photon absorption is required to induce fluorescence, a certain degree of photostress is inevitable. Therefore, effective in vivo imaging involves a fine balance between optimization of fluorescence signal level and minimization of photo-induced damage. Using femtosecond infrared radiation, oxidative damage is a major photodamage mechanism resulting from two- and higher photon absorption process. The photodamage pathway is similar to that of UV irradiation. Endogenous and exogenous fluorophores act as photosensitizers in photo-oxidative processes [38,39], and the photoactivation of these fluorophores results in the formation of reactive oxygen species that trigger the subsequent biochemical damage cascade in cells. Flavin-containing oxidases have been identified as one of the primary endogenous targets for photodamage [40]. Current studies found that the degree of photodamage follows a quadratic dependence on excitation power, indicating that twophoton process is the primary damage mechanism [40-44]. Experiments have also been performed to measure the effect of laser pulse width on cell viability. Results indicate that the degree of photodamage is proportional to the two-photon excited fluorescence generated, independent of pulse width. Hence, using a shorter pulse width for more efficient two-photon excitation also produces greater photodamage. An important consequence is that both femtosecond and picosecond light sources are equally effective for twophoton imaging in the absence of infrared one-photon absorbers [42,44]. However, another study has demonstrated that photobleaching of fluorophores can result from higher order (three- or four-photon) processes and suggests that higher order photodamage mechanisms may be also present during the imaging of biological specimens [45]. Since these higher order processes produce potential specimen damage but do not contribute to generation of fluorescence signal, high-order absorption processes often should be minimized. In addition to oxidative damage, thermal damage is also a concern in two-photon microscopy. The thermal effect resulting from twophoton absorption by water has been estimated to be on the order of 1 mK

So et al.

192

for typical excitation power and is therefore negligible [46,47]. However, in the presence of a strong infrared absorber such as melanin [48,49], there can be appreciable heating due to one-photon absorption. Thermal damage has been observed in the basal layer of human skin in the presence of high average excitation power [50].

4.

RECENT APPLICATIONS OF TWO-PHOTON MICROSCOPY IN TISSUE IMAGING

Two-photon microscopy has found an increasingly wide range of applications in biology and medicine. The advantage of two-photon imaging has been well demonstrated in neurobiology, embryology, dermatology, and pancreatic physiology. The usefulness of two-photon microscopy in studying the physiology of other tissue types is starting to emerge. Two-photon microscopy provides an unprecedented opportunity for in vivo study of neuronal interactions. Two-photon microscopy provide 3-D mapping of neuron organization and assays neuron communications by monitoring action potentials, calcium waves, and neural transmitters. Many twophoton neural biology studies have focused on the remodeling of neuronal dendritic spines and the subsequent effects these changes have on memory and learning, or on the dynamics of calcium signal propagation [51-53]. Other studies focus on system level interactions of neurons [54-56]. Twophoton imaging also contributes to the study of neuronal hemodynamics. Denk et al. used a two-photon microscope to probe red blood cell motion in rat cortical capillaries through 600-^tm-thick tissue. In some cases, they were able to detect flow changes in response to externally applied stimuli, such as vibrassa, direct touch, or moderate shock [57]. Another emerging application of two-photon microscopy is for in vivo study of neural pathology. Hyman and coworkers applied this technology to study the formation of /3-amyloid plaques associated with Alzheimer's disease [58,59]. Embryology is another area where two-photon microscopy has shown tremendous promise. The study of in vivo embryology has been difficult due to the sensitivity of the specimen to photodamage. This problem can be mitigated by using two-photon microscopy. Since most embryos are relatively small, it is often possible to image the full three-dimensional structures of the organism during development. For nonmammalian species, Summers et al. found that in sea urchin the first cleavage of the developing embryo does not predict the direction of the ultimate bilateral axes in the organism [60]. In this study, two-photon microscopy is used to activate caged fluorophores to effectively identify progenitor cells and allows the tracing of the


193

developmental lineage of cells in a given tissue structure. Periasamy et al. applied two-photon microscopy to the study of Xenopus embryos [61]. For mammalian species, Squirrell et al. imaged mitochondria in hamster embryos, and noted that after 24 hours of two-photon imaging the blastocyst was still competent for development. In contrast, imaging using confocal microscopy can impair development in the same system after only a few hours [6]. Since the skin structures are relatively thin and physically accessible, two-photon microscopy has been used extensively in dermatology studies. Masters et al. employed two-photon imaging to make visible the autofluorescence of in vivo human skin down to a depth of 200 /xm [12]. Cellular strata in the epidermis, including corneum, spinosum, and basal layer, can be clearly resolved based on NAD(P)H fluorescence in the cytoplasm. NADP(H) fluorescence is seen to be more pronounced in the basal layer due to its relatively higher metabolic activity. Below the epidermis, the dermal structure is visible due to elastin fluorescence and second harmonic signal generated from the collagen matrix. Recently, two-photon microscopy has been further applied to study the transport properties of the skin [62,63] facilitating the development of transdermal drug delivery technology. Two-photon microscopy has also been applied to study pancreatic physiology and metabolism based on NAD(P)H autofluorescence. Bennet et al. used two-photon imaging demonstrating that /3 cells metabolism is relatively constant in an intact islet eliminating a proposed mechanism for insulin secretion disorder [64]. This group also studied the effect of intercellular communication in regulating /3-cell metabolism [65]. Other organs and tissues have been investigated using two-photon microscopy. Piston et al. investigated NAD(P)H metabolism in the cornea [11] allowing in vivo study of cornea! structure down to 400 ^m. Napadow et al. investigated the heterogeneous microscopic structure of mammalian tongue tissue using two-photon microscopy in conjunction with magnetic resonance imaging and thus obtain spatial information on two different length scales [66].

5.

INSTRUMENTATION FRONTIER FOR TWO-PHOTON TISSUE IMAGING

A number of technological advancement in two-photon microscopy promises to further broader its application in biomedicine. Some of the most promising areas are video rate two-photon imaging, two-photon 3-D cytometry, and two-photon spectral resolved imaging.

So et al.

194

5.1

Video Rate Two-Photon Imaging

Video rate microscopes include a broad class of instruments capable of achieving imaging speed of 30 frames per second and faster. High-speed imaging is probably one of the most important innovations in two-photon imaging. First, many interesting biological processes are fast. Video rate imaging allows these processes, including calcium signaling events, to be studied on the millisecond time scale [67]. Second, video rate imaging is critical for the imaging of patients in clinics and living animals in the laboratory. A difficulty in imaging living subjects is the presence of motion artifacts resulting from physiological noise originated from blood flow and breathing. Physiological noise is typically on the order of 1 Hz, and imaging at frame rate on the order of tens of Hz or faster reduces the contamination of 2-D images by motion artifacts. Third, video rate imaging generally improves experiment efficiency, allowing the study of larger specimen volumes. Imaging of large areas enables the study of tissue structures and processes with correlation length on the order of millimeters and centimeters such as blood vessel distribution. High-efficiency imaging is further the foundation for the development of two-photon 3-D image cytometry, which will be discussed in the following section. The first implementation of video rate two-photon microscopic systems is based on the line-scan approach [68,69]. While video rate imaging using the line-scan approach improves speed, it suffers from resolution degradation due to interference effects especially in the axial direction. A second method was subsequently developed based on the rapid scanning of a point focus using high-speed scanning optics such as resonance mirrors or polygonal rotating mirrors [67,70]. This method provides high-speed imaging without compromising resolution. However, since the pixel residence time is reduced, it is difficult to achieve good signal-to-noise ratio images. The most recent improvement is based on multifocal scanning. By creating an array of separated foci, one can achieve parallel imaging from multiple focal volumes similar to the linescanning approach [71,72]. However, by spatially and temporally separating these foci, it is possible to minimize or eliminate resolution degradation due to interference effects [73-75]. The main disadvantage of this approach is the need of using area detectors such as charge-coupled device (CCD) cameras which tends to be less sensitive and more expensive. A two-photon microscope based on a rotating polygon mirror is shown in Fig. 3. A femtosecond titanium-sapphire is used to induce two-photon fluorescence. The laser beam is rapidly raster-scanned across the sample plane by means of two different scanners. A fast-rotating polygonal mirror (Lincoln laser, Phoenix, AZ) accomplishes high-speed line scanning (x axis), and a slower galvanometer-driven scanner with 500 Hz bandwidth (Cam-


Dichoric mirror

195

Galvanometer-driven mirror (Y-direction scanner)

PiezoZ slage

Polygonal mirror (X-direction scanner) FIGURE 3 Schematics diagram of a video rate two-photon microscope based on a polygonal scanner.

bridge Technology, Watertown, MA) correspondingly deflects the line-scanning beam along the sample's y axis. The spinning disk of the polygonal mirror is composed of 50 aluminum-coated facets (2 mm 2 ), arranged contiguously around the perimeter of the disk. The facets repetitively deflect the laser beam over a specific angular range, correspondingly scanning a line 50 times per revolution. Rotation speeds can be 10,000, 15,000, 20,000, or 30,000 rpm. In the fastest mode, the scanning speed is 40 fjus per line allowing the acquisition of approximately one hundred 256 X 256 pixel images per second. The image acquisition rate is 100 times faster than conventional scanning systems. The rest of the microscope design is similar to that of typical two-photon microscopes. This system is capable of imaging tissue structures based on autofluorescence at video rate. Dermal structures in ex vivo human skin have been studied. The skin was previously frozen. We studied the collagen/elastin fiber structures in the dermal layer. One hundred images were taken at depths between 80 and 120 yam below the skin surface. The frame acquisition time was 90 msec and the whole stack was imaged with a data acquisition time of 9 sec. The collagen/elastin fibers can be clearly observed. Representative images of the collagen/elastin fiber structures are shown in Fig. 4. The results of these studies clearly demonstrate the potential power of video rate twophoton microscopy.

£

i "£ B.{?8 O)

c o Mo 'CO +_, CD -3 •£- N * £ » -C ,„ CD

73

CD

CD Q.

CD O

> CD £

co > i= O CD O XT D (_ C/3 +-1 OJ TO O

._ .E ^ (/) CD .c

O

3 T3 ^

£ 0 «J

- co .. C "3 O 0 -Q CN

"-

o

•-E Io § w

O M-

05 T3 CD

CD ° D3 J=

g

S.

x: c o>

fi|8

c 8§ •S S » C/)

1-

-M

CO

O

CO

=5 o S

x

3^-§ * >ro•uj o E ^ OC i£ •- 05 ^

O

U_

Q. CD .=

n>

CO

2 -c 2 c


5.2

197

3-D Image Cytometry

Cytometry is an analytical method capable of precisely quantifying functional states of individual cells by measuring their optical characteristics based on fluorescence or scattered light. Cytometry can be grouped into two categories—flow cytometery and image cytometry—based on the measurement method. Flow cytometry monitors the properties of cells carried through the detection area by a fluid stream. It has several unique advantages. The most important of these is the rapidity of this measurement scheme. With the throughput rate up to 100,000 cells per second, the analysis of a large cell population for the detection of a few rare cells is possible. Also, based on multiparametric analysis, it is well suited to identify and distinguish the properties of cell subpopulations. Cell sorting methods implemented with flow cytometry make possible physical selection of a specific cellular subpopulation for further analysis or clonal propagation. Image cytometry has been recently introduced as a complementary method for flow cytometry. This method images individual cells plated in a 2-D culture. Cellular morphology and biochemical states are typically quantified by fluorescence microscopy. Although the throughput rate of this method is lower (approximately 200 cells per second), it has several unique advantages. Individual cells of interest can be relocated so that they can be further analyzed. One key example is the capacity of this method to monitor the temporal evolution of a cellular subpopulation. Image cytometry also provides cellular structural information, such as the relative distribution of a fluorophores in the nucleus and in the cytoplasm with micrometer level resolution. 3-D image cytometry extends cytometric study to 3-D tissues in situ. In one implementation, a video rate two-photon scanning microscope is coupled to a computer-controlled specimen translation stage. A two-photon 3D image cytometer retains all the important attributes of two-photon scanning microscopy, such as the ability to resolve structures in deep tissue and the minimization of photodamage. Thus, the properties of cellular populations in various tissue specimens can be studied in situ. The availability of this new instrument opens new opportunities for novel biological studies. As a preliminary demonstration, this 3-D image cytometer quantitatively measures cell fractions based on their fluorescence properties. A mixture of 3T3 fibroblasts expressing cyan and yellow fluorescent proteins at mixing ranging from 1/10 up to 1/105 were studied in both 2-D and 3-D cell cultures. An excellent correlation between the expected mixing ratios and the measured ratios has been observed (Fig. 5). 5.3

Two-Photon Spectral Resolved Imaging

Spectral imaging is a powerful technique to study both structures and dynamic processes in living cells [76,77]. Multicolor microscopy also provides

So et al.

198

(b)

Dilution study in cell blocks 1000000 er

100000 o

•o

o 1

loooo 1000 100

10

0.1

11mill

0.1

1

i i mini iiiiniil iiiiiiiil

Minna

10 100 1000 10000 100000 Expected Diluation Ratio

FIGURE 5 (a) Three z sections of a cell block containing a mixture of cells expressing cyan (gray) and yellow (white) fluorescent proteins. The images are 120 x 120 ^tm2. The z-section separation is 5 /um. Two yellow fluorescent protein-expressing cells are seen among a collection of cyan-expressing cells, (b) The expected and measured mixing ratios of the two cell types are determined for mixing ratios over five orders of magnitude, demonstrating excellent correlation.


199

new opportunities for developing diagnostic tools for noninvasive tissue studies. Imaging spectroscopy can be used in cancer surgery to delineate remaining traces of malignant cells [78]. Furthermore, analysis techniques such as principal component analysis [79] or generalized eigenvalue analysis can resolve subpixel-size objects or disentangle multiple tissue constituents that spectrally and spatially overlap [80]. For example, cellular metabolism can be noninvasively monitored by redox fluorometry [10,81,82]. The increasing activity in developing multicolor imagers reflects the newly realized scientific and clinical potential of spectrally resolved diagnosis. Spectral imaging requires exciting spectrally distinct fluorophores and distinguishes the different color emissions. To excite different fluorophores, one-photon excitation often requires the use of multiple excitation light sources for different color fluorophores. Since two-photon excitation spectra of many common fluorophores are very broad, it is often possible to efficiently excite multiple fluorophores via two-photon absorption at a single wavelength [76]. The efficient detection of emission light from multiple fluorophores and the resolution of this signal into individual spectral components can be a more difficult problem. Traditionally, the dispersion of light has been achieved by means of multiposition filter wheels. Although easy to incorporate into a microscope, mechanical filter wheels suffer from low speed, are limited in their color palette, and are mechanically cumbersome. Wholeimage spectral resolution imagers, such as liquid-crystal tunable filter (LCTF 83,84]) and acousto-optical tunable filter (AOTF [85,86]), are excellent devices but often are high cost and often are not necessary for single-point scanning devices, such as two-photon microscopes. Furthermore, similar to filter wheels, out-of-band emission is lost in LCTF and AOTF systems. The most commonly used systems in two-photon microscopy are grating-based spectrographs that provide a full spectrum at their exit ports. In particular, spectrographs are well suited for raster-scanning systems such as two-photon microscopes because they allow the parallel acquisition of the light spectrum at each pixel. Gratings can disperse light ranging from soft X-rays to infrared rays. The spectral resolution can be well below 0.1 nm, and the transmission efficiency can easily exceed 80% (at the grating's blazing angle). Photodetector selection is a primary issue in designing a grating spectral imager. The most commonly utilized photodetectors for spectral imaging instruments are CCD cameras and intensified CCDs (ICCDs). While these CCD systems are excellent, they are often expensive, have lower readout speed, or are not single-photon sensitive. More recently, multianode PMTs have become available commercially. These multichannel devices are particularly promising detectors for imaging spectroscopy because they allow parallel and therefore photon-efficient readout. PMTs are single-photon sen-

200

So et al.

sitive and excellent detectors for measuring very weak (femtowatt intensities) light pulses. PMTs have good sensitivity in the blue-green spectral region, with a typical quantum efficiency of 20-40%. However, their efficiency can drop below 1% for wavelengths above 600 nm. PMTs are also detectors with very high dynamic range since they can be operated from DC up to pulse rates exceeding 106 cps. Multianode PMTs have the minor disadvantage of exhibiting slightly higher dark count rates, and there is some cross-talk (10%) between adjacent PMT channels. The schematic diagram of a two-photon multicolor microscope based on multianode PMT is shown in Fig. 6. The design of the multicolor microcope is similar to that of typical two-photon microscopes except that the fluorescence emission is descanned (signal is transmitted back through the galvanometric scanner). The emitted light is dispersed using a grating spectrograph (Oriel Instruments, Stratford, CT). The spectrograph is configured for a spectral resolution of approximately 0.5 nm. The dispersion is adjusted such that the each channel of the multianode PMT is illuminted by a spectral range of approximately 12 nm. The 16-channel multianode PMT (R59000U00-L16, Hamamatsu, Bridgewater, NY) spans a total spectral range of slightly over 200 nm. The PMT is mounted onto a socket assembly with integrated voltage divider network (E6736, Hamamatsu). The biasing voltage is 800 V. Readout of the 16-channel PMT is achieved by means of a custom built multichannel single-photon counter card (mC-PhCC). Using advanced electronic components and board design techniques the mC-PhCC allows high-speed single-photon counting (>106 cps) up to 18 PMT channels (single unit) and provides two alternative high-speed interfaces for data readout. Both interfaces sustain real-time data transfer of 16-color, 256 X 256 pixel images. This two-photon imaging spectrometer can be used to study fluorescence emission from tissues. A fresh ear punch from a transgenic mouse was generously provided by Dr. Bevin Engelward (Dept. of Biological Engineering, MIT, Cambridge, MA). The genetically engineered mouse expresses enhanced yellow fluorescent protein (EYFP). Each specimen was freshly prepared by mounting it in a hanging-drop microscope slide. The stratum corneum of the mouse ear punch was placed closest to the coverslip. The power and the wavelength of the excitation light were 30 mW and 940 nm, respectively. The acquired image cube covers a spectral range from about 480 to 580 nm in 7-nm intervals, with 512 X 512 pixel spatial dimensions. The scanner's sampling rate was 5 kHz, and the data acquisition time was 52 sec. Images were taken from an arbitrary selected plane within the skin's stratum corneum (surface layer). The penetration depth was on the order of 15 yum. A representative image of the two-photon excited fluorescence of YFP-expressing mouse skin is shown in Fig. 7a.


201

Q) Q. O O W O

Q. 6

E CO i_

D) CD

T3 O "+->

CD

E 0)

.c o O) (O LLI

cc

D

o

s

o

o o

£

o 0

o c>

0 O

c

O

5 " £

0

?

o

" O C C

g ">

o

O

0

0

*-

tr,

[•ire] Ai!Sua}U|

0

"*

o0 C

«0

.-.

o ^ ^ c o c s ^ o ^~

• n - e ] Ajjsueiui

(/)

C

3

o

"* «

E

«

O3

g ^ "

.E ^ CO

25

O_

O o

CH

S -9>

o

c C

£

CO

15 CO 1 ^ ° E 0

g «3 §

c

o o

S & 0

CD c

0

s

°o

E

ij CD Q. 0)

CD

CO

0

i —

>•

S

03

CD CD

•^

03

CD

^

C

"2

CO CD

E +-i

O

tr !

X 9 ^2 CD

^ —„:' ^: -tx^^ 03 ^~ O) 1

c

— 0

o

O /""

o "-^

T3 s= CD CD i_ "CD +-•

3

CO

O

CD

.E CD F •— • -C CD

E ^ '1 CD D3 ±iCO CD 03

CD

E S CD O

CD O CO

CD

O

ro

.E ^ CO

c 03

E w CD

CO

£ E ao

EZ •n ^

r^ O ~ 0 LU CC

03

D Q) ^ O > CD LL — i^


203

Both the polygonal-shaped keratinocytes and a vertically oriented hair can be clearly identified based on their morphology. As shown in Fig. 7b, the "single-point" fluorescence spectra derived from small image areas allows unraveling of the cell constituents' intrinsic fluorophores. The characteristic emission peak at a wavelength of about 530 nm confirms YFP to be the major fluorophore in the keratinocytes. The spectral signature of the hair is much broader and red shifted. We therefore conclude that fluorescence signal of hair is most likely attributable to the autofluorescence of keratin, the major constituent of hair. We also acquired two-photon-induced fluorescence images from deeper within the skin. Although the fluorescence signal decreased significantly, we still were able to resolve individual cells in the epidermis and cellular structures such as the collagen and elastin fibers in the dermal layer. The diagnostic value of two-photon imaging can be further enhanced by numerically extracting the fluorescence spectroscopy information in tissues. For tissues containing fluorophores with known spectra, linear spectral decomposition technique can be used to determine the concentration of each species. In the case where the fluorophore species are not known or the spectroscopic emission of the fluorophores is sensitive to the unknown tissue microenvironment, numerical techniques such as principle component PCA can be applied to identify the spectrum of each fluorescent biochemical specie and its relative concentration. This analysis provides important information about the biochemical makeup and the microenvironment of the tissues. As a demonstration of spectral decomposition methods, dermal tissue structures can be spectrally distinguished based on their endogenous fluorescence properties. Spectrally resolved 3-D image stacks were acquired from ex vivo human skin. These image stacks were acquired at excitation wavelengths ranging from 700 nm to 920 nm at 10 nm intervals. Endogenous fluorescent species with unique excitation action spectra can be extracted from this four dimensional image data set using principle component analysis methods [79,87]. Our current principal component analysis algorithm involves two key steps. First, the number of independent spectral components is estimated based on single value decomposition. Second, the spectral shapes and their concentration at each voxel of each fluorescent specie are refined by alternative least-squares optimization based on nonnegativity constraints. Based on this approach, we can identify approximately five independent endogenous fluorescent species in the skin. One of the fluorescence species clearly corresponds to melanin or its photoproducts. An image representing the distribution of this spectra component in the dermal-epidermal junction (DEJ) is shown in Fig. 8a. The distribution of this fluorescence species is coincident with the distribution of the melanin caps on the

204

So et al. 1.2

1 0.8

0.6; 0.4 0.2 0 -0.2 760

800

840

Wavelength

30

45

60

75

90

105

Depth (urn) (b)

FIGURE 8 (a) The distribution of fluorescent species corresponding to melanin at the dermal-epidermal junction, (b) Top panel: the fluorescence excitation action spectrum of this species. Bottom panel: the depth distribution of this specie.

basal cells. Furthermore, we can map the relative abundance of the fluorescence species in the tissue. As a function of depth, it is clear that this fluorescence specie is most abundant at about 40 /mi, the location of the DEJ (Fig. 8b); this is in good agreement with the expected melanin distribution.

6.

CONCLUSION

Two-photon microscopy is a powerful method to extract structural and biochemical information from in vivo tissues on the subcellular scale. With additional technological advances, it is expected that its usage in the biomedical field will continue to increase.

ACKNOWLEDGMENTS The authors acknowledge support from NTH grants R21/R33 CA84740-01 and P01 HL64858-01A1.


205

REFERENCES 1. 2. 3. 4. 5. 6. 7.

8. 9. 10.

11.

12. 13.

14. 15. 16. 17. 18. 19. 20. 21.

M Minsky. Microscopy Apparatus. 1961, USA. BR Masters. Selected papers on confocal microscopy. Bellingham: SPIE, 1996. W Denk, JH Strickler, WW Webb. Two-photon laser scanning fluorescence microscopy. Science 248(4951):73-76, 1990. WA Mohler, et al. Dynamics and ultrastructure of developmental cell fusions in the Caenorhabditis elegans hypodermis. Cur Biol 8(19):1087-1090, 1998. WA Mohler, JG White. Stereo-4-D reconstruction and animation from living fluorescent specimens. Biotechniques 24(6): 1006-1010, 1012, 1998. JM Squirrell, et al. Long-term two-photon fluorescence imaging of mammalian embryos without compromising viability. Nat Biotechnol 17(8):763-767, 1999. JR Lakowicz, I Gryczynski. Tryptophan fluorescence intensity and anisotropy decays of human serum albumin resulting from one-photon and two-photon excitation. Biophys Chem 45:1-6, 1992. JR Lakowicz, et al. Fluorescence anisotropy of tyrosine using one- and twophoton excitation. Biophys Chem 56:263-271, 1995. B Kierdaszuk, et al. Fluorescence of reduced nicotinamides using one- and two-photon excitation. Biophys Chem 62:1-13, 1996. BR Masters, B Chance. Redox confocal imaging: intrinsic fluorescent probes of cellular metabolism. In: WT Mason, ed. Fluorescent and Luminescent Probes for Biological Activity. London: Academic Press, 1999, pp 361-374. DW Piston, BR Masters, WW Webb. Three-dimensionally resolved NAD(P)H cellular metabolic redox imaging of the in situ cornea with two-photon excitation laser scanning microscopy. J Microsc 178(Pt l):20-27, 1995. BR Masters, PT So, E Gratton. Multiphoton excitation fluorescence microscopy and spectroscopy of in vivo human skin. Biophys J 72(6):2405-2412, 1997. C Xu, WW Webb. Measurement of two-photon excitation cross sections of molecular fluorophores with data from 690 to 1050 nm. J Opt Soc Am B 13(3): 481-491, 1996. FS LaBella, P Gerald. Structure of collagen from human tendon as influence by age and sex. J Gerontol 20:54-59, 1965. MK Dabbous. Inter- and intramolecula cross-linking in tyrosinase-treated tropocollagen. J Biol Chem 241:5307-5312, 1966. KC Hoerman, AY Balekjian. Some quantum aspects of collagen. Fed Proc 25: 1016-1021, 1966. FS LaBella. Studies on the soluble products released from purified elastic fibers by pancreatic elastase. Arch Biochm Biophys 93:72-79, 1961. J Thomas, DF Elsden, PSM. Degradation products from elastin. Nature 200: 651-652, 1963. FS LaBella, WG Lindsay. The structure of human aortic elastin as influence by age. J Gerontol 18:111-118, 1963. P Stoller, et al. Polarization-dependent optical second-harmonic imaging of a rat-tail tendon. J Biomed Opt 7(2):205-214, 2002. T Theodossiou, et al. Thermally induced irreversible conformational changes

206

22. 23. 24.

25.

26. 27. 28. 29. 30.

31. 32. 33.

34. 35. 36. 37.

38.

39. 40.

So et al. in collagen probed by optical second harmonic generation and laser-induced fluorescence. Lasers Med Sci 17(1):34-41, 2002. PJ Campagnola, et al. Second-harmonic imaging microscopy of living cells. J Biomed Opt 6(3):277-286, 2001. T Theodossiou, et al. Visual observation of infrared laser speckle patterns at half their fundamental wavelength. Lasers Med Sci 16(l):34-39, 2001. BM Kim, et al. Collagen structure and nonlinear susceptibility: effects of heat, glycation, and enzymatic cleavage on second harmonic signal intensity. Lasers Surg Med 27(4):329-335, 2000. I Freund, M Deutsch, A Sprecher. Connective tissue polarity. Optical secondharmonic microscopy, crossed-beam summation, and small-angle scattering in rat-tail tendon. Biophys J 50(4);693-712, 1986. S Roth, I Freund. Optical second-harmonic scattering in rat-tail tendon. Biopolymers 20(6): 1271-1290, 1981. JB Shear, C Xu, WW Webb. Multiphoton-excited visible emission by serotonin solutions. Photochem Photobiol 65(6):931-936, 1997. M Chalfie, et al. Green fluorescent protein as a marker for gene expression. Science 263:802-805, 1994. RY Tsien. The green fluorescent protein. Annu Rev Biochem 67:509-544, 1998. GS Baird, DA Zacharias, RY Tsien. Circular permutation and receptor insertion within green fluorescent proteins. Proc Natl Acad Sci USA 96(20): 1124111246, 1999. GJ Allen, et al. Cameleon calcium indicator reports cytoplasmic calcium dynamics in Arabidopsis guard cells. Plant J 19(6):735-747, 1999. M Zaccolo, et al. A genetically encoded, fluorescent indicator for cyclic AMP in living cells. Nat Cell Biol 2(l):25-29, 2000. KD Niswender, et al. Quantitative imaging of green fluorescent protein in cultured cells: comparison of microscopic techniques, use in fusion proteins and detection limits. J Microsc 180(Pt 2): 109-116, 1995. SM Potter, et al. Intravital imaging of green fluorescent protein using twophoton laser-scanning microscopy. Gene 173:25-31, 1996. M Oheim, et al. Two-photon microscopy in brain tissue: parameters influencing the imaging depth. J Neurosci Meth 111( l):29-37, 2001. AK Dunn, et al. Influence of optical properties on two-photon fluorescence imaging in turbid samples. Appl Opt 39:1194-1201, 2000. VE Centonze, JG White. Multiphoton excitation provides optical sections from deeper within scattering specimens than confocal imaging. Biophys J 75(4): 2015-2024, 1998. SM Keyse, RM Tyrrell. Induction of the heme oxygenase gene in human skin fibroblasts by hydrogen peroxide and UVA (365 nm) radiation: evidence for the involvement of the hydroxyl radical. Carcinogenesis 11(5):787-791, 1990. RM Tyrrell, SM Keyse. New trends in photobiology. The interaction of UVA radiation with cultured cells. J Photochem Photobiol B 4(4):349-361, 1990. PE Hockberger. et al. Activation of flavin-containing oxidases underlies light-


41. 42.

43.

44. 45. 46.

47. 48. 49.

50. 51. 52. 53. 54.

55. 56. 57.

58.

59.

207

induced production of H2O2 in mammalian cells. Proc Natl Acad Sci USA 96(11):6255-6260, 1999. K Konig, et al. Two-photon excited lifetime imaging of autofluorescence in cells during UVA and NIR photostress. J Microsc 183(Pt 3): 197-204, 1996. K Konig, et al. Pulse-length dependence of cellular response to intense nearinfrared laser pulses in multiphoton microscopes. Opt Lett 24(2): 113-115, 1999. Y Sako, et al. Comparison of two-photon excitation laser scanning microscopy with UV-confocal laser scanning microscopy in three-dimensional calcium imaging using the fluorescence indicator Indo-1. J Microsc 185(Pt 1):9-20, 1997. HJ Koester, et al. Ca2+ fluorescence imaging with pico- and femtosecond twophoton excitation: signal and photodamage. Biophys J 77(4):2226-2236, 1999. GH Patterson, DW Piston. Photobleaching in two-photon excitation microscopy. Biophys J 78(4):2159-2162, 2000. WJ Denk, DW Piston, WW Webb. Two-photon molecular excitation laserscanning microscopy. In: JB Pawley, ed. Handbook of Biological Confocal Microscopy. New York: Plenum Press, 1995, pp 445-458. A Schonle, HSW. Heating by absorption in the focus of an objective lens. Opt Lett 23(5):325-327, 1998. SL Jacques, et al. Controlled removal of human stratum corneum by pulsed laser. J Invest Dermatol 88(l):88-93, 1987. VK Pustovalov. Initiation of explosive boiling and optical breakdown as a result of the action of laser pulses on melanosome in pigmented biotissues. Kvantovaya Elektronika 22(11): 1091-1094, 1995. C Buehler, et al. Innovations in two-photon deep tissue microscopy. IEEE Eng Med Biol Mag 18(5):23-30, 1999. R Yuste, W Denk. Dendritic spines as basic functional units of neuronal integration. Nature 375(6533):682-684, 1995. K Svoboda, DW Tank, W Denk. Direct measurement of coupling between dendritic spines and shafts. Science 272(5262):716-719, 1996. K Svoboda, et al. In vivo dendritic calcium dynamics in neocortical pyramidal neurons. Nature 385(6612): 161-165, 1997. M Maletic-Savatic, R Malinow, K Svoboda. Rapid dendritic morphogenesis in CA1 hippocampal dendrites induced by synaptic activity. Science 283(5409): 1923-1927, 1999. CL Cox, et al. Action potentials reliably invade axonal arbors of rat neocortical neurons. Proc Natl Acad Sci USA 97(17):9724-9728, 2000. F Engert, T Bonhoeffer. Dendritic spine changes associated with hippocampal long-term synaptic plasticity. Nature 399(6731):66-70, 1999. D Kleinfeld, et al. Fluctuations and stimulus-induced changes in blood flow observed in individual capillaries in layers 2 through 4 of rat neocortex. Proc Natl Acad Sci USA 95(26): 15741-15746, 1998. RH Christie, et al. Growth arrest of individual senile plaques in a model of Alzheimer's disease observed by in vivo multiphoton microscopy. J Neurosci 21(3):858-864, 2001. BJ Bacskai, et al. Imaging of amyloid-beta deposits in brains of living mice

208

60.

61.

62.

63.

64. 65.

66.

67. 68. 69. 70. 71. 72. 73. 74. 75.

76.

So et al. permits direct observation of clearance of plaques with immunotherapy. Nat Med 7(3):369-372, 2001. RG Summers, et al. The orientation of first cleavage in the sea urchin embryo, Lytechinus variegatus, does not specify the axes of bilateral symmetry. Dev Biol 175(1): 177-183, 1996. A Periasamy, et al. An evaluation of two-photon excitation versus confocal and digital deconvolution fluorescence microscopy imaging in Xenopus morphogenesis. Microsc Res Tech 47(3): 172-181, 1999. B Yu. et al. In vitro visualization and quantification of oleic acid induced changes in transdermal transport using two-photon fluorescence microscopy. J Invest Dermatol 117(1): 16-25, 2001. BS Grewal, et al. Transdermal macromolecular delivery: real-time visualization of iontophoretic and chemically enhanced transport using two-photon excitation microscopy. Pharm Res 17(7):788-795, 2000. BD Bennett, et al. Quantitative subcellular imaging of glucose metabolism within intact pancreatic islets. J Biol Chem 271(7):3647-3651, 1996. DW Piston, et al. Adenovirus-mediated knockout of a conditional glucokinase gene in isolated pancreatic islets reveals an essential role for proximal metabolic coupling events in glucose-stimulated insulin secretion. J Biol Chem 274(2): 1000-1004, 1999. VJ Napadow. et al. Quantitative analysis of three-dimensional-resolved fiber architecture in heterogeneous skeletal muscle tissue using nmr and optical imaging methods. Biophys J 80(6):2968-2975, 2001. GY Fan, et al. Video-rate scanning two-photon excitation fluorescence microscopy and ratio imaging with cameleons. Biophys J 76(5):2412-2420, 1999. JB Guild, WW Webb. Line scanning microscopy with two-photon fluorescence excitation. Biophys J 68:290a, 1995. GJ Brakenhoff, et al. Real-time two-photon confocal microscopy using a femtosecond, amplified Ti:sapphire system. J Microsc 181(Pt 3):253-259, 1996. KH Kirn, C Buehler, PTC So. High-speed, two-photon scanning microscope. Appl Opt 38(28):6004-6()09, 1999. J Bewersdorf, R Pick, SW Hell. Mulitfocal multiphoton microscopy. Opt Lett 23:655-657, 1998. AH Buist, et al. Real time two-photon absorption microscopy using multipoint excitation. J Microsc 192:217-226, 1998. SW Hell, V Andresen. Space-multiplexed multifocal nonlinear microscopy. J Microsc 202(Pt 3):457-463, 2001. M Straub. et al. Live cell imaging by multifocal multiphoton microscopy. Eur J Cell Biol 79(10):726-734, 2000. A Egner, SW Hell. Time multiplexing and parallelization in multifocal multiphoton microscopy. J Opt Soc Am A Opt Image Sci Vis 17(7): 1192-1201, 2000. C Xu, et al. Multiphoton fluorescence excitation: new spectral windows for biological nonlinear microscopy. Proc Natl Acad Sci USA 93(20): 1076310768, 1996.

Two-Photon Microscopy of Tissues 77. 78.

79. 80. 81. 82.

83.

84. 85. 86. 87.

209

A Miyawaki, et al. Dynamic and quantitative Ca2+ measurements using improved cameleons. Proc Natl Acad Sci USA 96(5):2135-2140, 1999. A Gillenwater, et al. Noninvasive diagnosis of oral neoplasia based on fluorescence spectroscopy and native tissue autofluorescence. Arch Otolaryngol Head Neck Surg 124(11):1251-1258, 1998. JJ Andrew, TM Hancewicz. Rapid analysis of Raman image data using twoway multivariate curve resolution. Appl Spectrosc 52(6):797-807, 1998. L Laiho, et al. Two-photon 3-D mapping of tissue endogenous fluorescence species based on fluorescence emission. J Phys Chem (submitted). B Chance. Pyridine nucleotide as an indicator of the oxygen requirements for energy-linked functions of mitochondria. Circ Res Suppl 38:131-138, 1976. BR Masters. In vivo corneal redox fluorometry. In: BR Masters, ed. Noninvasive Diagostic Techniques in Ophthalmology. New York: Springer-Verlag, 1990, pp 223-247. HR Morris, CC Hoyt, PJ Treado. Imaging spectrometers for fluorescence and Raman microscopy—acoustooptic and liquid-crystal tunable filters. Appl Spectrosc 48(7):857-866, 1994. HR Morris, et al. Liquid crystal tunable filter Raman chemical imaging. Appl Spectrosc 50(6):805-811, 1996. ES Wachman, WH Niu, DL Farkas. Imaging acousto-optic tunable filter with 0.35-micrometer spatial resolution. Appl Opt 35(25):5220-5226, 1996. ES Wachman, WH Niu, DL Farkas. AOTF microscope for imaging with increased speed and spectral versatility. Biophys J 73(3): 1215-1222, 1997. TM Hancewicz, C Petty. Quantitative analysis of vitamin A using Fourier transform Raman spectroscopy. Spectrochim Acta A Mol Spectrosc 12:2193-2198, 1995.

Fluorescence Lifetime Imaging Microscopy of Endogenous Biological Fluorescence Paul Urayama and Mary-Ann Mycek University of Michigan, Ann Arbor, Michigan, U.S.A.

1.

INTRODUCTION

Methods of steady-state fluorescence microscopy are routinely employed for studies in cell biology to reveal information regarding cellular morphology, intracellular ion concentrations, protein binding, lipid content, and membrane status [1]. Fluorophore lifetimes offer an additional source of contrast in imaging applications because lifetimes are known to be highly sensitive to physical conditions in the fluorophore's local environment (temperature, pH, oxygen concentration, polarity, binding to macromolecules, ion concentration, and relaxation through quenching and through resonant energy transfer), while being generally independent of factors influencing fluorescence intensity (fluorophore concentration, photobleaching, artifacts arising from optical loss) [2,3] While exogenous fluorophores provide useful and specific tools for enhancing contrast, endogenous fluorophores—fluorescent biomolecules indigenous to biological cells and tissues—are of biomedical interest as potential probes of metabolic function, tissue morphology, and biomarkers of disease. Because exogenous agents are not employed, endogenous fluores211

212

Urayama and Mycek

cence methods raise no concerns regarding issues of contrast agent toxicity or delivery. Several chapters from this book highlight the utility of endogenous fluorescence for biomedical applications. Fluorescence lifetime information would complement minimally invasive steady-state endogenous fluorescence methods for disease detection and metabolic imaging (reviewed in [4]). Therefore, this chapter focuses on endogenous fluorescence lifetime imaging in biomedicine, with discussions on technology and examples of applications. We begin in Sec. 1 with an introduction to fundamental concepts in fluorescence lifetime spectroscopy and describe several useful biological and biomedical imaging implementations of the method, before focusing on endogenous biological fluorophores of interest. In Sec. 2, we describe both time- and frequency-domain approaches to fluorescence lifetime measurement, with an emphasis on strategies employed for microscopic imaging, including several approaches to imaging in turbid biological media via optical sectioning. In Sec. 3, we highlight several applications of endogenous fluorescence lifetime imaging, ranging from cellular biological studies to in vivo tissue and clinical endoscopic studies, before concluding in Sec. 4. 1.1

Fluorescence Lifetime and Imaging

Fluorescence lifetime imaging (FLI) and fluorescence lifetime imaging microscopy (FLIM) are methods for producing spatially resolved images of fluorescence lifetimes. A discussion of factors affecting fluorescence lifetime illustrates the physical nature of the probe and its usefulness as a source for contrast. A system with N fluorophores in the excited state depopulates stochastically, e.g., at a rate (I)

dt

for a two-state model, where F is the radiative decay rate and km is the nonradiative decay rate. The resulting decay is exponential in time with a lifetime r and quantum yield Q, given by

T

=

and

=

V /) sin (r) + D~ )S (r)

(17) (18)

Note the extraction of two "lifetimes," one related to phase r^ and one related to demodulation TM. For a single exponential decay, r = rv = TM. For multiexponential decays, TV < TM, and a treatment can be found in [2]. 2.3

Comparing Time- and Frequency-Domain FLIM

In principle, time- and frequency-domain methods are equivalent and are related by a Fourier transformation. Both TD and FD FLIM implementations report comparable lifetime resolution (minimal detectable lifetime) and discrimination (detectable differences between samples). In TD FLIM, Webb et al. [37] reported a temporal resolution of 100 ps and a lifetime discrimination of 10 ps with a reproducibility of 2—5%, using DASPI in varying viscosity solutions of ethanol and glycerol. The excitation source was an amplified Ti:sapphire laser, frequency doubled, for a pulse width of 10 ps. Detection was achieved with a time-gated, intensified CCD with a minimal gate width of 85 ps. Also available in the time domain are TCSPC FLIM systems with the ability to resolve decay components down to 30 ps [36]. In FD FLIM, Hanley et al. [38] reported a lifetime discrimination of 40-70 ps with a reproducibility of 100 mW), picosecond pulsed diode lasers, along with picosecond gated intensified CCD detectors, has eliminated major technological difficulties and is implemented in TD systems [37,41,42]. Presently, both methods have comparable temporal resolution and discrimination, and both benefit from rapidly advancing technologies. Therefore, the choice of FLIM implementation depends on the specific application, availability of experimental equipment, and the nature of the lifetime information to be extracted. An advantage of FD method is its ability to identify multiexponential decays, for which TV < rM. A particular lifetime may also be rejected from detection using a phase suppression technique [43]. On the other hand, TD methods have a greater temporal dynamic range and are better suited for detection of long lifetimes. A model-independent lifetime can also be extracted by use of stretched exponentials [32]. 2.4

Sectioning and Imaging in Turbid Media

In thick, turbid biological media, it can be critically important to selectively image a thin, buried section of the sample. Advanced methods exist to minimize contamination from out-of-plane signals, resulting in effective sectioning and improved spatial resolution. Sectioned images can be used subsequently for 3-D fluorescence lifetime imaging. Several strategies for FLIM sectioning are considered here. The first two—confocal and multiphoton microscopy, use point measurements suitable for raster scanning methods. The second two—structured illumination and spatial deconvolution—employ wide-field illumination. In confocal microscopy, only signal from the front focal point of the objective is returned to the detector, usually by means of a pinhole at the back focal point of the objective, resulting in axial resolution of down to 0.6 jam [44]. Time-resolved confocal systems have been developed using both TD and FD schemes. A TD system by Ghiggino et al. [34] used a fiberoptic as the light path for the excitation and return beams, with the fiber aperture acting as the pinhole. Output from a PMT or APD was sent to a TCSPC module. However, at the time, scanning with TCSPC was considered slow compared with gated or gain-modulated detection [9]. Recently, TCSPC methods have reached megahertz photon counting rates, allowing a fluorescence decay to be determined in a few milliseconds, thereby enabling TCSPC confocal FLIM systems capable of fast, single-molecule detection [35,36]. Another TD scheme was implemented by Buurman et al. [45] using a gated microchannel plate photomultiplier with lifetime determination rates of 40 ^s/pixel. Alternatively, a double-pulsed system based on a ratio measurement of integrated fluorescence intensities under one-pulse and two-

224

Urayama and Mycek

pulse excitation was constructed [46,47], eliminating the need for fast detection electronics. In FD, two detection schemes for confocal FLIM systems were implemented by Morgan et al. [48] using a modulated ICCD and an imaging photon detector with an external correlator synched to the modulation peak. Lock-in detection has also been used in FD FLIM to simultaneously image with a confocal microscope multiple fluorophores excited using simultaneous sources modulated at different frequencies [49]. Multiphoton microscopy is a method with inherent spatial sectioning [50]. Because the excitation light must be tightly focused onto a sample in order to offset effects of small nonlinear excitation cross-sections, out-offocal-plane excitation is eliminated and fluorescence is emitted only from the small focal volume. This natural sectioning results in resolution comparable to confocal without the need for a pinhole. So et al. [51] reported an axial point spread function of 0.9 îm (FWHM) using 960 nm excitation and a 1.25-NA objective. Multiphoton imaging also has the advantage of reduced photodamage, due to the limited excitation volume [50]. Two-photon FLIM has been implemented in the frequency domain [51,52] and time domain [53]. A common excitation source in two-photon systems is the Ti: sapphiare laser, useful for its high-power, ultrashort pulses (—100 fs), high repetition rate (—100 MHz), and tunability, with a typical spectral range of 700-1000 nm for two-photon excitation (corresponding to a one-photon excitation range of 350-500 nm). Optical sectioning is also possible with wide-field illumination via structured illumination [54]. In this method, the sample is incoherently illuminated by a source with a spatially varying intensity achieved using a mask, e.g., of the form s(r) = 1 + m cos(ftw + o .«> 'o (O +-• <x|- _

05

•D CD o cO

£CD 13 i-> oy "5) ° £«

1*1 .2 O 5= E " c

—

C

C 05 O

c?

o e= O

'-M O CD ^2

cj

00 " » CO -Q CD (- 05 •(-•'il CD CD o

w -c .E -C > rQ. O E CD ^ 13 1 P c ^-i-O.C ^

o r .t

E I & 8c co^ -^ 05 ^ 0 0^ °0 CO

05

T3 O v_ c^ » S

- E < •8 o5 z o 05 V 05 O CD t -^ O

E o ~. o> 0

C

_CO »

CO O Q.73

^ C/5 CO "J CD > C EC

+-i -f,

C

Ss|s

U. CD
-^ 0 ^-^

C/3 CO O

CO CO ^ -^

c 0

O

c c

'— — CO CO

O 3

03

U_

O

(A L , X , r, z, 0) in W/cm2. Acx is the excitation wavelength, while r, z, 6 represent local position in cylindrical coordinates. Obtaining the intrinsic fluorescence coefficient, )8fA ( , v , A(,,,,, z). The intrinsic fluorescence coefficient (3 is defined as the product of the absorption coefficient due to the fluorophore, ju,al1 (cm" 1 ), and the quantum yield Y (dimensionless) of fluorescence emission. Epithelial tissues like skin usually have a layered structure. Within a layer, (3 can be considered as constant; therefore, it can be denoted as a function of z, /3(A ex , A t , m , z). Acm is the wavelength of emitted fluorescence light. The product c£>/3 yields the density of fluorescence sources in units of W/cm\ Using an MSP, one can measure the relative /3 distribution inside the tissue. The relative fluorophore density, p(z), can be inferred from the fluorescence image obtained using a CCD camera attached to the MSP. The intrinsic spectra of different tissue layers measured with the MSP are nor-

Fluorescence Spectroscopy and Skin Cancer

327

malized to equivalent overall integral intensity and is denoted as Inorm(A ex , A em , z). Then /3 can be obtained using /3(Aex, A em , z) = p(z) X I norm (A ex , A em , z) 3.

4.

(1)

Calculation of escape function E(Xem, r, z). Once a fluorophore emits a fluorescence photon, that photon must successfully reach the surface and escape to be observed. The escape function E(Aem, r, z) is the surface distribution as a function of radial position (r) of escaping photons from a point source of fluorescence at depth z and radial position r = 0 within a tissue of thickness D. This function can be calculated by Monte Carlo simulation. The units of E are cm~'. Simulations were conducted for a series of depths (z) inside the tissue, using the optical properties for the emission wavelengths of interest. Calculation of the predicted fluorescence, F(Xex, Xem, r). The observed flux of escaping fluorescence F in unit of W/cm2 at the tissue surface is computed by the following convolution [9]:

(Aex, r', z' 6>')£(A ex , A em , z')

2rr' cos 8', z')r' dr' d0' dz'

(2)

The convolution in Eq. (2) can be implemented numerically using discrete values for O and E that were generated by the Monte Carlo simulation, and the experimentally determined (3. Monte Carlo simulations were conducted to generate the fluence distributions inside the model skin for the 442-nm excitation laser light and for the escape functions for 40 different source depths and 29 different emission wavelengths from 470 nm to 750 nm. In each simulation, 1 million photons were launched. The in vivo fluorescence spectra were measured using a wide-beam illumination (1 -cm-diameter beam) and a small pickup spot (3 mm diameter) at the center of the illumination field. Therefore, the excitation light distribution can be simplified to a function of z only, i.e., O(z). For a wide illumination beam, the fluorescence intensity will be the same at the tissue surface independent of position r. In Eq. (2), the fluorescence escape function E(Aem, r, z) can be integrated with respect to r and 0 first. E(A em , r, z)r dr d0 = E(A em , z)

328

Zeng and MacAulay

The contribution from a specific skin layer (from depth zl to depth z2) to the observed in vivo fluorescence spectrum can be calculated as follows: A cm ) =

0(z)/3(Aex, A cm , z)E(A cm , z) dz

(4)

Note that, within a skin layer, /3 is assumed to be independent of depth. The excitation wavelength A cx was also fixed at 442 nm. Substitute A em with A, and Eq. (4) becomes: /+'/2

O(z)E(A, z) dz

(5)

We denote the integral on the right side of Eq. (5) as the fluorescence detection efficiency, i7, avcr:/ i^ /2 , which represents the likelihood of obtaining an autofluorescence signal from a specific skin layer. 0 dz

(6)

It integrates the product of the excitation light distribution inside the tissue and the fluorescence escape efficiency. Given that only for the stratum corneum and the dermis is the intrinsic fluorescence coefficient nonzero. The reconstructed skin in vivo spectrum is F(A)

= j8 s .ra,un,C-«n, C um(A)T? S t r d t u I 1 1 rorn e u n,(A)

+

&le,™s(A)l7 d c r n l i s (A)

(7)

which is a linear combination of the product of intrinsic spectrum and the fluorescence detection efficiency of all the excited fluorophores. Figure 6 shows the excitation light distribution as a function of depth z. The incident power density was I W/cm2. It can be seen that very little light penetrates into the lower dermis (z > 570 /AITI). The stratum corneum and the papillary dermis contribute the most to the in vivo fluorescence signal. Figure 7 shows the calculated fluorescence escape efficiency E(A, z) as a function of wavelength for different depths z inside the tissue. Near the tissue surface, E(A, z) is weakly dependent on A and the curves are flat and horizontal indicating that the reabsorption and scattering of the tissue to fluorescence photons have minimal effect on the fluorescence escape function. As the fluorescence sources appear deeper inside the tissue, the reabsorption and scattering of the fluorescence photons by the tissue have more and more effect on the fluorescence escape function. E(A, z) curves become more and more tilted as z increases because both the absorption coefficient (/u,a) and the scattering coefficient (fjij of the skin tissue decrease with increasing wavelengths. The double absorption valleys of blood at 540 nm

329


1.700

0.000 0.00

0.05

0.06

FIGURE 6 The excitation light (442 nm) distribution as a function of depth z inside the skin tissue (infinite wide beam, normal incidence). The incident power density is 1 W/cm2. (From Ref. 20.)

and 580 nm also become more pronounced at greater tissue depths. Figure 8 shows the calculated fluorescence escape efficiency E(A, z) as a function of depth z inside the tissue at different wavelengths. The logarithm of E(A, z) decreases almost linearly with increasing depth within the tissue. The E(A, z) as a function of z curve decays faster at short wavelengths, indicating that fluorescence photons with short wavelengths have greater difficulty escaping from the tissue than photons with longer wavelengths. This is due to the variation in the absorption coefficient (û,a) and scattering coefficient (yu,s) of skin tissue, which decreases with increasing wavelengths. Figure 9 shows the calculated fluorescence detection efficiency, 17, as a function of wavelength for both the upper dermis and the stratum corneum. The effect of absorption by blood is clearly seen on the i7dermis curve, while the T7stratumcorneum curve is a. relatively flat horizontal line indicating that the absorption of melanin found in the epidermal layer and the absorption by blood in the dermal layer have little effect on the escape of the stratum corneum fluorescence due to the surface position of this thin tissue layer. The i7(A) curves represent how the intrinsic fluorescence spectra are distorted by tissue reabsorption and scattering. Figure 10 shows the calculated skin autofluorescence spectrum in com-

Zeng and MacAulay

330

1000 FIGURE 7 Fluorescence escape efficiency, E(A, z), as a function of wavelength at different depth z inside the tissue calculated by Monte Carlo simulation for the model skin. A simulation was done for the stratum corneum layer (10 /um thick) at z = 5 yum. A single simulation was done for the 80-/xm-thick epidermis (at z = 45 /am) because there are few fluorophores in this layer. Thirty-eight simulations were done for the dermis. From z = 90 /um to z = 270 yum, simulations were done at 10-/um intervals; from z = 270 /zm to z = 510 /u,m, at 15-|um intervals; from z = 510 /zm to z = 1000 /im, at 25-/um intervals. (From Ref. 20.)

parison with the experimental in vivo spectra. For different volunteers or different body locations on one volunteer, the in vivo skin autofluorescence intensity changed significantly (by as much as 100%), but the spectral maximum position (wavelength) did not vary significantly (515 ± 2 nm), while the details of the spectral shape did change as a consequence of tissue reabsorption of the fluorescence light and the variation in the amount of the various absorption chromophores at different body locations or from subject to subject. Figure 10 shows five experimental curves from two volunteers (one Asian, one Caucasian) and three different body locations (inner forearm: upper and lower positions, and hand dorsum). Each curve is normalized to have a maximal intensity of 100 counts. The blood absorption has much


331

200

Depth (um)

1000

FIGURE 8 Fluorescence escape efficiency, E(A, z), as a function of depth within the tissue at different wavelengths as calculated by Monte Carlo simulation for the model skin. The logarithm of E(A, z) decays almost linearly with increasing tissue depth z. (From Ref. 20.)

larger effect on the spectral shape than melanin does for fair-colored Caucasian and Asian volunteers. The effect of the absorption on fluorescence light by blood can be seen in both the experimental curves and the theoretically reconstructed curve. Below 530 nm and above 600 nm, the theoretical curve fits quite close to the experimental curves. In the two wavelength bands (470-530 nm, 600-750 nm), light absorption by blood is relatively small in comparison with the light absorption by tissue. However, big differences between the experimental curves and the theoretical curve exist over the strong hemoglobin absorption wavelength range (530-590 nm). This suggests that the theoretical method employed, the published skin optical properties, and the microscopic fluorescence properties determined by the MSP measurements are basically correct. However, the assumptions in the skin model for the blood content amount, its distribution in the tissue model, and oxyhemoglobin/hemoglobin ratio may not truly represent the actual situation in the in vivo skin. An additional factor is that the amount of blood and its oxygenation state in tissue may also be highly dynamic. The effect of blood on tissue optical properties and tissue optical behavior is an area in need of further investigation.

332

Zeng and MacAulay

Dermis

10

CXI

8

o

Stratum Corneum

450

500

550

600

650

700

750

Wavelength (nm) FIGURE 9 Fluorescence detection efficiency 17 as a function of wavelength for the dermis layer and the stratum corneum layer. (From Ref. 20.)

Theoretical Experimental

550

600

650

700

750

Wavelength (nm) FIGURE 10 Comparison of the reconstructed skin autofluorescence spectrum curve with the experimental in vivo data. (From Ref. 20.)


333

As noted in Fig. 5, the autofluorescence spectral shape of the in vivo skin is quite different from the intrinsic spectra of both the stratum corneum and the upper dermis. The in vivo spectrum is the intrinsic spectrum modified by tissue optical properties due to light-tissue interactions. These interactions include absorption and scattering, which determine the excitation light distribution inside the tissue and affect the escape process of the fluorescence photons. The modifications are represented by the fluorescence detection efficiency 17 as defined in Eq. (6). 17 can be calculated independent of the intrinsic spectra and represents the effects of tissue optics on fluorescence detection. The theoretical reconstruction correctly accounted for the light-tissue interactions (scattering, absorption, and regenerating of fluorescence photons). Therefore, outside of the strong blood absorption wavelength band, the theoretical curve is consistent with the experimental data. 2.4

Temporal Dynamics of In Vivo Skin Autofluorescence Emission Under Continuous Wave Laser Exposure

It is known that autofluorescence signal intensity decays as a result of light exposure in a process termed photobleaching or photodegradation [38-40]. During photobleaching, it is presumed that fluorophores undergo photochemical alteration to species with either lower fluorescence quantum yields or different excitation and emission wavelengths. Zeng et al. [21] quantitatively measured the temporal process of in vivo skin autofluorescence decay during continuous wave laser irradiation and analyzed its kinetics using the Monte Carlo simulation for light propagation in the skin. The nonuniform distribution of the excitation light in different skin layers and the different fractional contributions of different tissue layers to the total observed fluorescence signal were correlated with the measured kinetics. The study demonstrated that measured decay dynamics could be used to determine the fractional contributions of different skin layers to the total in vivo autofluorescence signals. Measurements of in vivo skin autofluorescence decay during continuous laser exposure were performed with a computerized autofluorescence and diffuse reflectance spectroanalyzer system as shown in Fig. 11. A 400jam-core-diameter fiber with microlens was used to conduct light from a 442-nm He-Cd laser for continuous illumination of selected skin sites. The microlens attached to the fiber was used to achieve uniform illumination. The excited, autofluorescence light was collected by a large-core (1 mm)diameter fiber for transmission to an optical multichannel analyzer (OMA) for spectral analysis. The two fibers were cocentered with a fiber holder that was adjusted to produce a 5° illumination angle and 10-mm-diameter illumination spot size on the skin, and a 30° detection angle and 3-mm detection

Zeng and MacAulay

334

Optical Multichannel Analyzer

-Fiber Holder

SKIN FIGURE 11 Spectroanalyzer system configuration for quantitative monitoring of laser-induced autofluorescence decay. (From Ref. 21.)

spot size. During continuous laser-induced autofluorescence decay experiments, the computer-controlled OMA was programmed to acquire the autofluorescence spectra at constant time intervals with a constant exposure duration of 5.51 sec. For each skin site, 120 spectra were obtained continuously over an 11-min time period and were stored as a single computer file for further processing. The integral fluorescence intensity over a 10-nm band at the spectral maximum position of 520 nm was calculated from the 120 spectra and is denoted as I(t). Nonlinear regression was used to map the time response of the decay dynamics. The autofluorescence decay processes during continuous He-Cd laser (442 nm) exposure on the inner forearm of five volunteers in vivo showed similar decay dynamics. The spectral shapes demonstrated no significant changes during the laser exposure. Nonlinear regression fitting of all measured autofluorescence decay curves revealed that decreases in autofluorescence follow a double exponential function: l(t) = a exp(-t/r,) + b exp(-t/r 2 ) + c

(8)

with a fast process (first term) and a slow process (second term). The time constants (r,, r2) of the two processes differed by an order of magnitude. Parameters a. b, c were normalized to satisfy the condition that 1(0) = 1 at t = 0, i.e.:


335

a + b + c=1

(9)

Figure 12 shows an example of such data. The measurements were made on the inner forearm of a volunteer of Asian ancestry under an illumination power density of 64 mW/cm2. The upper part of the graph shows the 120 original data points, the fitted double exponential decay curve (solid line), and the equation of the best fit. The parameters (a, b, c, r,, r2) and their standard deviations derived by the nonlinear regression fitting are given in the caption. The lower part of the graph shows the residual curve using a magnified y-axis scale. The residuals fluctuate randomly around zero with amplitudes of less than 0.012 supporting the selection of the double exponential function as the best fit to the experimental data. Figure 13 shows the autofluorescence decay curves at four different laser exposure intensities: 64 mW/cm2, 42 mW/cm2, 27 mW/cm2, and 3.8 mW/cm2. For the lowest exposure intensity used (3.8 mW/cm2), the fitted curve shows a slow single exponential decay, whereas the other three curves are best fitted with double exponential functions. Figures 14 and 15 illustrate the relationship between the decay parameters and exposure intensities: a and b tend to increase with increasing exposure intensity, whereas the converse is true of c; T, and r2 decrease when the exposure intensity increases.

c

CD

1.0 n n

l(t)=0.138 exp(-t/12) + 0.293 exp(-t/370) + 0.569

0.9 0.8 0.7 0.6

0.010 0.0

(D 01

-0.010

100

200

300

400

500

600

700

Time (seconds) FIGURE 12 A sample of in vivo skin autofluorescence decay during continuous laser exposure (location, inner forearm, exposure power, 64 mW/cm2). Fitted parameters: a = 0.138 ± 0.007, b = 0.293 ± 0.004, c = 0.569 ± 0.005, T, = 12 ± 1, r2 = 370 ± 14. (From Ref. 21.)

336

Zeng and MacAulay

1.0500 0.9375 c 0.8250 CD 0.7125 0.6000

0

100 200 300 400 500 600 700

Time (seconds) FIGURE 13 Autofluorescence intensity decays induced with different exposure intensities. (From Ref. 21.)

1.0 0.8 O

_ 0.6

CE

0.4 0.2

-0.0

0

10

20

30

40

50

60

70

Exposure Power Density (mW/cm ) FIGURE 14 Skin autofluorescence intensity decay parameters a, b, c change as a function of exposure intensity. (From Ref. 21.)

337


10 10

20

30

40

50

60

Exposure Power Density (mW/cm

70

2,

FIGURE 15 Skin autofluorescence intensity decay parameters T, and r2 change as a function of exposure intensity. (From Ref. 21.)

At the lowest exposure intensity (3.8 mW/cm2), r2 could not be measured with an 11-min exposure experiment. To assess the autofluorescence recovery process after laser exposure, autofluorescence emission spectra were monitored at 5-min intervals for 50 min immediately following termination of the laser exposure. During this period, no fluorescence recovery was noted. Autofluorescence images of the exposed skin site appeared as a dark circular spot. In contrast, white-light diffuse reflectance images showed no differences between the exposed skin site and the surrounding, unexposed sites. Sequential autofluorescence imaging after termination of the continuous laser exposure revealed that complete recovery of the autofluorescence took about 6 days. With time, the dark spot in the images became brighter until it reached the same intensity as the surrounding unexposed skin. The random pattern on images of recovery suggests that the autofluorescence recovery is not due to fluorophore diffusion from the unexposed to the exposed areas, which would have manifested as autofluorescence signal recovery initially at the boundary and filling in toward the center. It is more likely that a metabolically derived fluorophore-regenerating process is responsible for the observed autofluorescence recovery. From Monte Carlo simulation of excitation light distribution inside the skin tissue and fluorescence escape from the tissue as well as the fluorophore distribution shown in Fig. 4, the study concluded that only the stratum corneum and the upper dermis (depth M' •.1— -1

CO CO CJ O 1 i

+ +

C

[^ CO c N o o o r - c n o c o o ^ c D L o ^

z CO

^>

*

CO

C O C O - D C O C O C O C O C O C O C O C O u. (_) O CJ O C J ' F O U O O C J C J U O g

Q Q"

H—

-M

-f-

CD

"co

c o c o c o c o c o c o c o c o c o c o c o C D CO CO CO CO C O C D Q C O C O C O C D C D C D C D C D ^ CO CD CD CD Q. Q. a a c o a a a a a a o - o . ^ a. a. a. a CO CO c o c o - ^ c o c o c o c o c o c o c o c o ™ CO CO CO CO

c CO CD

0

CO CO

_c •? GO

>

X3

gco "CD CD

O ^

CO

QCO >

C

d o

Q-

—

CO

O

£ CO —

CO

CO

co 'eo

m Q. E O

CT) ^" ^— 0 LO CD CD ^t

CN|--r-~CNCNLOLOCNCO

5—

C

*

~

CO ^

T3 C CO •1—
Q. 1 C "t- ±1 ~ — fc

m Q) QJ , — TO w Q. a £ > > c > >J5

Fluorescence Spectroscopy and Atherosclerotic

401

tional features of atherosclerotic plaque are associated with plaque instability and rupture. These include thinning of fibrous cap, lack of smooth muscle cells, increased inflammation, breakdown of collagen by proteinases, and soft lipid-rich cores. Except for the size of the lipid core, the other features of instability are superficial and are confined to the luminal surface. Therefore, the evaluation of components of the fibrous cap is relevant and important. Conventional clinical techniques (angiography, angioscopy, intravascular ultrasound) can accurately estimate stenosis but do not provide an accurate analysis of the plaque composition itself [3,4,6]. Consequently, various techniques are currently under study as potential clinical tools for identifying plaque composition and cellularity. These include magnetic resonance imaging [6,8], optical spectroscopy (near-infrared [9], Raman [10], fluorescence [11,12]), and local thermography [13]. Optical spectroscopy is attractive for the characterization of atherosclerotic plaque composition because the biochemical and morphological changes in tissue induced by progression of disease generates changes in the optical characteristics of the tissue. Such optical properties can be investigated using fiberoptic catheters, thus allowing direct and rapid investigation of the luminal surface or fibrous cap. 1.3

Fluorescence Spectroscopy Research for Arterial Wall Characterization

Several research groups [14-19] have explored the potential of fluorescence spectroscopy for diagnosis of diseased arterial walls. Fluorescence from both endogenous fluorophores within the arterial wall (elastin, collagen, lipids, etc.) and exogenous chemical probes (e.g., hematoporphyrin derivative, photofrin) have been investigated. Both in vitro [14-17] and in vivo [18,19] studies have been reported. Spectroscopic investigations have primarily utilized a steady-state LIFS technique wherein various wavelengths are used for tissue excitation: 306-310 nm [14], 308 nm [21], 325 nm [17], 337 nm [22], 458 nm [23], and 476 nm [15]. Various computational techniques have been employed for analysis of arterial fluorescence emission and for sample classification. These include ratio of intensities at two emission wavelengths [14], spectral width at half the maximal intensity [17], total fluorescence intensity [23], fitting spectra to known biochemical emission spectra (i.e., collagen, elastin) [14], sampling emission intensities at equal wavelength interval across the measured spectral range [17], principal components analysis (PCA) and neural network [17,24], and discriminant analysis [18]. The results demonstrate that there is potential for the application of fluorescencebased techniques for discriminating between normal and advanced lesions

402

Marcu et al.

[25]. For clinical applications, these studies have introduced the idea of incorporating spectroscopic evaluation into clinical fiberoptic systems, initially to guide laser angioplasty and to evaluate the likelihood of restenosis and, more recently, to diagnose unstable lesions. As demonstrated by a few research groups, TR-LIFS can improve the specificity of fluorescence measurements in tissue and overcome some limitations of the steady-state spectroscopy [16,20,22]. This concept was first introduce by Baraga et al. [20] in 1989 and further investigated by Andersson-Engels et al. in the early 1990s [16,22]. These studies reported distinct fluorescence decay characteristics of normal arterial wall from fibroatherosclerotic plaque, demonstrated that the presence of blood in tissue does not change arterial wall fluorescence decay characteristics, and suggested that time-resolved measurements can enhance diagnosis through the discrimination of variations in human atherosclerotic lesions. Recent research also shows that information retrieved from time-resolved spectra of normal and diseased human arterial walls can be used for differentiation of different types of atherosclerotic plaques and specifically identification of lipid-rich lesions [12,26,27]. 2.

TIME-RESOLVED FLUORESCENCE SPECTROSCOPY FOR TISSUE DIAGNOSIS

TR-LIFS using a time-domain technique and ultraviolet (UV) light excitation (337 nm) can be applied for identification and staging atherosclerotic lesions and demonstrates potential for diagnosis of unstable lesions. This chapter provides an overview of (1) time-resolved fluorescence of arterial wall endogenous fluorescence; (2) time-resolved fluorescence of normal and diseased arterial walls (human aorta and coronary artery); and (3) correlation between the fluorescent signature of the fluorescent constituents of arterial wall and the time-resolved emission of the arterial walls (healthy or diseased). 2.1

Review of Time-Resolved Spectroscopy: Principles and Methods

Fluorescence measurements can be categorized as measurements that are either static (steady state or time integrated) or dynamic (time resolved). Steady-state techniques provide an "integrated" spectrum over time that gives information about fluorescence emission intensity and spectral distribution. Typically, the sample fluorescence is induced by a continuous light source and recorded with the help of a detector array such as an optical multichannel analyzer (OMA) or by a scanning monochromator connected


403

to a photomultiplier. Scanning the spectrum by means of a monochromator, though convenient, allows at a specific time only the light of a particular wavelength to be recorded. This disadvantage is compensated by the superior sensitivity of photodetectors and by their linear response, in contrast to the diode array used by an OMA system. Due to the relative simplicity of the instrumentation, measurements based on steady-state methods are the most common for biological tissues research. However, the time dependence of emission and the information contained therein are lost when only this method is used. Time-resolved techniques record the dynamically evolving fluorescence emission and thus provide deeper insight into the molecular species of the sample (e.g., the number of fluorescent species and their contribution to the overall emission), quenching processes, and/or changes in the local environment. Two methods of time-resolved measurements are often used: time domain and frequency domain. For time domain measurements, the sample fluorescence is induced by a short pulse of light (typically nanoseconds or shorter) and the emission is recorded with a fast and sensitive photodetector. The most common technique for time domain measurements is time-correlated single-photon counting (TCSPC). For the frequency domain, the sample is excited with intensity-modulated light. Typically, the intensity of the excitation light is varied (sine wave modulation) at a high frequency. These methods are described in detail by Lakowicz [28]. This chapter primarily describes theoretical and applied concepts to characterization of atherosclerotic lesions based on a time-resolved time domain method. The most important characteristics of a fluorophore (fluorescent molecule) for fluorescence measurements are the quantum yield and fluorescence lifetime. The quantum yield is expressed as the ratio of the number of photons emitted to the number of photons absorbed. The fluorescence, or radiative, lifetime is determined by the time the molecule spends in the excited state prior to decaying radiatively to the ground state. Direct measurements of fluorescence lifetime, T, is based on the assumption that this process follows first-order kinetics quantitatively described by equation: T — - T i, inp c

t/T

where !„ and It are the fluorescence intensities at times zero and T [28]. 2.2 Time-Resolved Fluorescence Spectroscopy of Tissue In contrast to fluorescent measurements in dilute solutions, fluorescence measurements in tissue, a highly scattering, absorbing, and heterogeneous medium, is a complex process. This process involves three major steps [29,30]: (1) penetration of the excitation light into the tissue; (2) absorption

404

Marcu et al.

of the excitation light by the fluorescent molecule(s) and conversion to fluorescent emission; and (3) propagation of the emission into tissue from the position of the fluorescent molecule(s) to the detection system. Both steps 1 and 3 are strongly dependent on the tissue optical properties (absorption, /Aa, and scattering, /Lts, coefficients). Consequently, the blood content and heterogeneous stucture of tissue affect these two steps in a nonlinear manner. Step 2 is directly related to the absorption properties of the fluorescent molecule (/x at ) and its quantum yield (). Fluorescence measurements, based on a steady-state (time-integrated) approach, record the intensity of the fluorescence of the fluorescent molecule^), which in turn is directly proportional to the concentration and quantum yield of the fluorescent molecule(s). However, absolute intensity measurements cannot be obtained from highly scattering and inhomogeneous medium such as tissue. This is due to such factors as optical properties of the tissue (ju,a, /zs) along the excitation and emission path, the excitationcollection geometry (optical assembly), and photobleaching of the emitting molecule(s). Moreover, several distinct molecules characterized by overlaping excitation and emission spectra can be excited simultaneously, thus making interpretation of the tissue emission difficult. In contrast, fluorescence measurements based on time-resolved methods are primarily related to the quantum yield or lifetime of the fluorescent molecule and are independent of the absolute emission intensity or concentration of the fluorescent molecule(s). Consequently, advantages for tissue characterization and in vivo diagnosis include the following: (1) Spectral overlap of endogenous fluorophores in tissue can be resolved. A change in the relative concentration of tissue fluorophores will result in changes of the fluorescence decay dynamics or lifetime. (2) As long as commensurable signal to noise is obtained, the measurement is independent of the presence of endogenous absorbers in tissue (hemoglobin) or of excitation-collection geometry (optical assembly). (3) As the lifetime defines the time available for the fluorophore to interact with its environment, the measurement is sensitive to microenvironmental parameters in tissue (pH, enzymatic activity, temperature) and thus may reflect metabolic and inflammatory activity at the tissue level. 2.3

Experimental Methods for TR-LIFS of Tissue

As an example of how a TR-LIFS technique can be used for characterization and staging of atherosclerotic lesions, in the following we describe an apparatus based on a time domain method and a data analysis technique for time-resolved fluorescence data. Although TCSPC is the prevailing method used by biochemists for time domain measurements, in this chapter we de-


405

scribe an alternative method utilizing a fast digitizer and gate detection for sampling the fluorescence pulse. This method has intrinsic advantages for in situ measurements of tissue, including direct recording of the time-resolved fluorescence emission pulse, suitability for fiberoptic systems, and applicability for use with low-repetition-rate lasers. 2.3.1

Instrumentation

A TR-LIFS apparatus using a fast digitizer and gated detection is shown in Fig. 2. The optical pulses of a nitrogen laser (337 nm, 3 nsec) are focused into a fiberoptic probe and directed to the sample (arterial specimens or fluorescent arterial constituents) from above. The resulting fluorescence emission is collected by a fiberoptic bundle, focused into a scanning monochromator, and detected by a gated multichannel plate photomultiplier tube (rise time: 0.3 nsec) placed at the monochromator exit slit. The photomultiplier output is amplified (rise time: 0.35 nsec; bandwidth: 1 GHz), and the

SAMPLE

FIGURE 2 Time-resolved laser-induced fluorescence spectroscopy experimental apparatus. PD 1 and PD 2, silicon photodetectors; PMT, multichannel plate photomultiplier tube; BS 1 and BS 2, beam splitters.

406

Marcu et al.

entire fluorescent pulse from a single excitation laser pulse is recorded with a digital oscilloscope (bandwidth: 500 MHz, sampling frequency: 2 Gsamples/sec or faster). A fraction of the excitation source output beam is monitored by two fast silicon detectors, for triggering gate and delay generators, which in turn gate the photomultiplier, and for triggering the oscilloscope to begin sweeping the photomultiplier output and to monitor laser pulse-topulse shape and energy variation. A personal computer is used to control data acquisition, data transfer from the oscilloscope, and monochromator wavelength scanning. A more detailed description of this apparatus can be found in [31]. 2.3.2

Data Analysis

The apparatus described above records the fluorophore emission at a number of wavelengths across the emission spectrum, so that a complete fluorophore emission spectrum can be obtained. A time-integrated spectrum (conventional emission spectrum or steady-state spectrum) is also obtained by integrating each measured fluorescence pulse as a function of time. Thus, the spectrum is characterized by discrete fluorescence intensities (I w] ) that show the variation of fluorescence intensity as a function of wavelength. The timeresolved spectrum is determined by deconvolving the excitation pulse from each measured fluorescence pulse using methods described below. Therefore, the time-resolved spectrum represents the intrinsic fluorescence decay as a function of time [fluorescence impulse response function (FIRF)] for the different wavelengths of emission. Generally, there are several distinct steps in the analysis of time-resolved data from a fluorescent system, including (1) determination of the FIRF of the fluorescent system, (2) identification of a set of fitting parameters that best describe the characteristics of the fluorescence decay, (3) application of statistical methods to evaluate the effect of various experimental conditions on the decay characteristics of the fluorescent system, and (4) classification of emission into qualitative (discrimination of tissue types based on subjective criteria) or quantitative (discrimination based on prediction of the compositional makeup of the tissue) analysis/interpretation categories of the fluorescent system. To address each step, various analytical tools can be utilized. Examples of analytical methods that were successfully employed for the analysis of time-resolved data obtained from arterial tissue follow. Computation of the Fluorescence Impulse Response Function. All of the information content in a single fluorescence decay experiment is contained in the FIRF, I,(T), which characterizes the fluorescent molecule and its environment. The measured fluorescence response pulse y(t) of a fluorophore is related to the excitation pulse x(t) by a convolution equation:


y(t) =

407

I t (r)x(t - T) dr Jo

Accurate deconvolution of FIRF is important for time-resolved fluorescence studies. Consequently, a variety of deconvolution techniques have been explored. These include, but are not limited to, least-squares iterative reconvolution method [32-34], method of moments [35], Laplace transform [36], Fourier transform [35], and the maximal entropy method [37]. These methods are thoroughly reviewed elsewhere [28,32]. Among these, the leastsquare iterative reconvolution is the most commonly used and reliable deconvolution technique. It has been proven mathematically that this method provides the best estimates for parameter values. Generally, the FIRF is a priori postulated as a sum of several exponential functions whose time constants were iteratively adjusted to minimize the residual difference between observed and reconstructed emission signals [28,32]. However, it has been shown that a fit to a priori multiexponential models is poorly suited for fluorescence decay analysis of complex biological samples [33]. The fluorescence decay of some biological systems may not follow a mutiexponential decay law. In this chapter, we present a method that does not require an a priori defined model; instead, FIRF is formulated as an expansion of discrete-time Laguerre polynomials (Laguerre expansion of kernels). Because the Laguerre basis is a complete family of functions, this approach has the advantage that it can be used to reconstruct a FIRF of arbitrary form. Moreover, the Laguerre basis appears to be a judicious solution for fluorescence decay analysis because of its exponential weighting function. The method of expansion of a system impulse response function on the orthonormal basis of discrete-time Laguerre functions was originally introduced by Marmarelis [38] to model nonlinear biological systems. Briefly, the convolution equation in discrete time becomes: y(n) =

If(m)x(n - m) m

The discrete FIRF, I f (m), is expanded over the discrete-time Laguerre basis {b,(m, a)}: I f (m) = 2_j Cjbj(m, a) j Where Cj are the expansion coefficients and bj(m, a) are the discrete-time orthonormal Laguerre functions. The discrete-time Laguerre parameter, a (0 < a. < 1), determines the Laguerre function rate of exponential asymptotic decline. The convolution equation is calculated using an initial guess for the a value and the measured laser pulse x(n). The optimal order j of the La-

408

Marcu et al.

guerre expansion can be determined for each fluorescent system by evaluating the sum of weighted residuals on a representative set of data. The parameters a and c, from Laguerre expansion are determined iteratively by least-squares minimization of the weighted square sum of the residuals 2'X

=

where wn is a statistical weighting factor. Detailed description of this method is reported elsewhere [39]. Deconvolution based on this technique can separate the computation of the FIRF from the analysis or modeling of the fluorescent system. This feature can facilitate an unconstrained interpretation of time-resolved fluorescence data from complex biological media such as arterial tissue. Analysis of the Fluorescence Decay. A commonly used set of fitting parameters to analyze fluorescence decay are (1) the average lifetime (r t ) and (2) the decay constants (r,, a,) obtained by approximating each FIRF with a multiexponential function. For instance, the lifetime can be estimated as the interpolated time at which the FIRF decays to 1/e of its maximum value, whereas for biexponential model parameters such as the time-decay constants, r, (fast-term) and r2 (slow-term), and the fractional contribution, ratio A: (A, = ^/(a, + a 2 ), i = 1, 2), of each decay component to the FIRF allow analysis of the decay dynamics. I,-(t) = a,e l/T| + a2e'/T: The meaning of these parameters depends on the system being studied. Their meaning also is different for a mixture of fluorophores or for one fluorophore. These parameters can also be used to analytically determine the average lifetime. This set of parameters can be used for both quantitative and qualitative analysis of the resulting data [28]. Statistical Analysis for Time-Resolved Data. Univariate statistical analysis is utilized to evaluate changes of the dynamics of fluorescence decay and fluorescence intensity at discrete wavelengths. For instance, common variance analysis, such as one- or two-way ANOVA, can be applied to the time-dependent parameters (r t , TJ, A,) to evaluate the effect of emission wavelength, components/specimens types, and changes of experimental conditions on the dynamics of fluorescence decay. When a significant effect is observed, differences among individual means can be assessed with a post hoc comparison test (e.g., Student-Newman-Keuls).


409

Classification Method. A multivariate statistical analysis, such as a stepwise discriminant analysis, can be employed for a qualitative classification of arterial samples [18,40]. This method allows the following: 1.

2.

3.

A determination of which combination of predictor variables (fluorescence emission parameters: time or amplitude constants, and fluorescence intensity at discrete wavelengths across emission spectrum) accounts for most of the differences in the average profiles of the groups, so that a set of canonical discriminant functions that provides the best discrimination between tissue specimens from different groups can be generated. A test of the validity of discriminant functions derived for the classification of tissue samples, and determination of the classification accuracy for a training set and a cross-validated set. For this purpose, the tissue samples are initially classified in several groups based on histopathological examination. Using "subjective" criteria, the histopathological analysis can account for either biochemical composition or structure/anatomy of the sample or both.

FLUORESCENCE EMISSION OF MAJOR CLASSES OF ENDOGENOUS FLUOROPHORES IN ARTERIAL WALL

The fluorescent signature of each fluorescent constituent for this method is related to the overall fluorescence emission of the arterial sample. An important step toward the development of fluorescence Spectroscopy techniques for diagnosis of tissue is the identification of the origin of tissue fluorescence in various situations (healthy or diseased) as characterized by the fluorescence of endogenous molecules. 3.1

Endogenous Fluorescent Constituents of the Arterial Wall

The fluorescence emission of the arterial wall, induced by UV light, has been attributed to several endogenous constituents. The fluorescence emission of structural proteins including elastin and collagen is primarily involved with the fluorescence emission of arterial wall. Early research [14,41] demonstrates that the fluorescence emission of normal coronary artery or aorta is primarily related emission of elastic fibers within internal elastic lamina or within media, respectively, whereas the fluorescence emission of advanced fibrous lesions (type Va or fibrous plaques) is associated with the emission of collagen (30—60% dry weight [3]). The atherosclerotic stages between normal and advanced fibrous lesions are characterized by accumulation within arterial intima of various

410

Marcu et al.

types and amounts of lipids (20-25% in type II, 35-30% in type III, -60% in type IV) that alter the intimal morphology [4]. Albeit reported by several studies, the contribution of lipids to arterial wall fluorescence is still not well understood. Several lipid components are reported to fluoresce and thus are likely to modulate the fluorescence emission of matrix structural proteins in arterial tissue. These include the following: 1. 2.

3.

4.

Blood-derived particles that undergo transendothelial diffusion (LDLs or VLDLs and their oxidative products) [21,42]. Cholesterol esters (cholesteryl oleate and linoleate). These lipids account for more than 60% of type II and III lesions lipid composition [43]. Cholesteryl oleate is the major lipid component of type II lesions and the oleate/total esters (oleate and linoleate) ratio decreases with lesion progression [43-45]. Free cholesterol that accumulates largely within the necrotic core of type IV lesions. The free/total cholesterol ratio increases with lesion progression. Lipopigment ceroids and carotenoids [15,46,47]. Ceroid, an endproduct of lipoprotein oxidation, is found in macrophage and in lesions with lipid-rich necrotic cores. Carotenoids are present in lipid-rich lesion cores.

Other constituents, such as glycosaminoglycans, tryptophan, and calcium, are also reported to fluoresce. However, glycosaminoglycans represent less than 2% of the organic matrix in normal arterial wall and about 0.4% of the organic matrix in atherosclerotic wall. Thus, it has been suggested that these components may not significantly contribute to the fluorescence of the arterial wall (.41]. Tryptophan fluorescence induced by excitation wavelengths below 310 nm strongly influences the emission of the arterial wall at wavelengths below 360 nm [14]. However, these have minimal influence on emission above 360 nm [14,22]. Calcium exhibits sharp fluorescence peaks in the range 350-650 nm upon 308 nm excitation [42]; however, its emission with longer excitation wavelengths (325 nm) was not reliably detected [41]. Table 1 outlines the main fluorescent constituents in normal and diseased arterial wall together with their emission characteristics. 3.2

Time-Resolved Fluorescence Emission Characteristics of Endogenous Fluorophores

The following presents a summary of the time-resolved fluorescence emission of the main structural proteins in arterial wall (elastin and distinct types of collagen) and lipids constituents (free cholesterol, esterified cholesterol, lipoproteins). A more detailed description of their emission characteristics


411

and experimental conditions is presented elsewhere [26,39,49,50]. The data were obtained from both dry and hydrated (0.9% saline solution) powder forms by utilizing the instrumental apparatus and the data analysis method described above. The time-resolved emission was recorded in the 360- to 510-nm spectral range (5-nm interval). To minimize the photobleaching of these endogenous fluorescent molecules, the energy used for sample excitation was less than 1 ^J/pulse. During a single measurement sequence the energy total fluence delivered to the sample was less than 1.2 mJ/mm2. Choosing an optimal energy level for tissue excitation is important for an accurate interpretation of fluorescent measurements in terms of tissue composition. To gain insights into these aspects, a comprehensive study of the photobleaching characteristics of the main fluorescent components of the arterial wall (elastin, collagen type I, and free cholesterol) during prolonged irradiation is provided in [31]. Representative time-resolved spectra and time-decay characteristics of structural proteins (elastin and types I and III collagen) are depicted in Fig. 3a-c. Unique time-integrated (Fig. 3d) and time-resolved (Fig. 3e, f) characteristics are observed for each endogenous fluorophore. The fluorescence emission of each component decayed at markedly different rates that are wavelength dependent such that the largest difference between decay rates was observed for wavelength smaller than 430 nm. Typically, the fluorescence of lipid compounds (fast-decay emission) could be easily distinguished from that of structural proteins (slow-decay emission). In addition, timeresolved parameters distinguish different types of cholesterols, different types of lipoproteins, and different types of collagen from each other [26,39,49]. The characteristics of the fluorescence decay (T,, r2, A,) at 390 nm emission for elastin, type I and III collagen, free and esterified cholesterol, and lipoproteins are summarized in Fig. 4. As shown in several studies, the hydration affected the emission of endogenous fluorophores [39,50]. For instance, the fluorescence decay of collagen and esterified cholesterol (oleate and linoleate) is markedly hydration dependent. Hydration shortened the decay constants of these components. This effect is primarily noted in the blue-shifted range of the emission spectrum. In summary, these studies show that TR-LIFS technique can characterize the main components of the arterial tissue matrix by generating distinct patterns of fluorescence emission for each component. The change of environmental conditions, such as the presence of water molecule, affects each component in a different manner. Also, the distinct fluorescence features of each constituent suggest that TR-LIFS technique can provide information about the biochemical components that contribute to the fluorescence of normal and diseased arterial wall.

412

Marcu et al. ,

C/3

• .•

"co _ ^03 DC

.1-1

C

-1

C LU

J

Q

n as

—to •— c/) ™. E .52 2

0 £ :n

"co

1__ _-_

_^

~T"

0^

£2,

1

.

—

— — • — UJ

3 — ,i- c\F

CM -J^ 4-j

e -s 0 «

s

5

CO Q Q3

— ro 00

OM *^

W

"

F_1

,

21

, 0^

•—' OO 00

" CD

2, " "co CO

,

,

,

,

OO

OO

00 >—< —: ^ . _• co —•

—

. , ^! in '—'in

03

ro

•

in

lii

CO

«

s

ljl!! ll

~

3 |„ - ^ 0 0 0 CO

O ^^^

•—

CO

CO

3

CO

CO

oom ^ oo co CD in co rô*-

mooo r^- ^~ oo ^j co oo co co «-OOCDCD

o o o o in ^~ r^ in co o in o o o o o CD 00 00

o in o o o

o

o o in in o oo oo r^ r- oo

o oo

ooo •

"o co

CO

~^

- P 03

M r -

a* ^! sl o f! en O —

,,

(/3

,

CD ^ 7

G1 -03

oo f ") 00 00

oo

o o in o in

00 00 00

00 CN CO ^

CO

00 00 00 00 00

oo o in in co i oo co in

OM O O O O

OO OO

£-

00

O

CD 0 Q. Q. CO

L_

03

C CO 03

in co

^ ~

T3

0 « < "7; C/3

0 V 0

L_

O

0

..

rs/ Q

0

00 •—

'

o c

q 06 m •— •

°0 ^S: r^ Lfj CD

(0 r-1

j [

00

° in °°

00 "" CD 00 00

CO

~

^r ; • * i% • .tt 3•» •• •

23

"*" O

CM

UL

%

"c cd c: E o

». *

"

if-

+- T3 M- 03

»^" *>

"

Q.

^ .2

»

* "•

o

£ C

* ®

ft!

Q "co *

c*

,e\i

ffi

0*

»

„

^.

e

*

P 2 c =5 1 -^

c en

'^

•

It

CD *

*

•
CM

' ~ > c ° F o^ '+-•

O

Z uojpun-j

ro

oo c .52

* * * * *

co 03

£

0 c 'w 24 hr 24 hr >24 hr >24 hr >24 hr >24 hr

Fluorescence intensity13 IX

35x 35x 35X 1X

NA 18x 20x

ln addition to the major emission maximum at 509 nm, wtGFP, GFPuv, EGFP, and dIEGFP also have a shoulder at 540 nm. The numbers in parentheses represent secondary excitation/emission wavelength. b Fluorescence intensity is measured as a relative to wild-type GFP.

Green Fluorescent Protein and Tumor Angiogenesis

433

mized for protein synthesis in the host. Many of the variants exhibited markedly improved properties over the wild-type protein. For example, an enhanced green fluorescent protein (EGFP) from Clontech Corporation (Palo Alto, CA) was synthesized 10 times more efficiently due to human-optimized codon usage. In addition, it fluoresces 35 times more intensively in comparison with wild-type protein. The end result is a new construct that is 350-fold more effective in comparison with the original protein [4]. Adding to the spectrum of GFP family are homologues from other coelenterates, such as Obelia and Phialidium, both hydrozoas, and Renilla reniformis, an anthozoa. Stratagene (La Jolla, CA) cloned and humanized Renilla GFP. It was claimed that R. reniformis GFP (hrGFP) is 2.5-fold more efficient in the generation of fluorescence than wild-type and red-shifted variants Aeqourea GFP. In addition, Renilla GFP is more stable in high and low pH and is more resistant to organic solvents or other denaturing reagents, which usually causes the quenching of GFP fluorescence in immunohistochemistry process. Moreover, Renilla GFP is said to be less toxic than Aeqourea GFP (Stratagene, http://www.stratagene.com/voll3 3/p85-87.htm). Another interesting protein was cloned from Discosoma sp., a sea animoe relative from the IndoPacific sea. It absorbs ultraviolet (UV) or blue light and emits bright red fluorescence. However, its mechanism for fluorescence is not fully understood as it bears little sequence homology with the GFP family of fluorescence proteins. Its red fluorescence makes it uniquely suited to be used in conjunction with members of the GFP family of fluorescent proteins in double- or triple-labeling experiments. In addition, Terskikh et al. [5] recently reported the identification of a mutant red fluorescent protein (E5 mutant from drFP583, Clontech, Palo Alto, CA). This mutant has a very exotic property in that it can autocatalyze and change the fluorescence properties of its chromophore over time in the presence of molecular oxygen. This leads to a red shift of its excitation peak. Therefore, the protein will initially fluoresce green and turn into red over time. This property make it possible to be used as a "molecular timer" of gene transcription in cell culture or living organisms. The extensive array of applications of the GFP proteins are possible because of the wide spectrum of cell/organism types in which they can be expressed. Furthermore, the expression results in fluorescence in targeted cells/organisms without any cofactors/substrates. For example, it has been shown that GFP and its variants have been widely expressed in other organisms, including transgenic mice [6-8], Xenopus [9,10], fish [11], Drosophila melanogaster [12,13], Caenorhabditis elegans [3,14], Saccharomyces cerevisiae [15,16], E. coli [3,17], bacteriophage [18], and viruses [19].

Li et al.

434

2.

METHODS FOR LABELING CELLS AND PROTEINS WITH GFP

There are many methods to label cells and proteins with GFP as it is a rather small protein (27 kD) with a robust three-dimensional structure that is quite resilient to external disturbances. This is reflected in the following two aspects: (1) It can be fused with many different proteins without losing fluorescence. (2) It is resistant to photobleaches (especially the red-shifted mutant versions), moderate oxidizing reagents, moderate reducing agents (2% /3-mercaptoethanol, 10 mM DTT). (3) It is quite stable at pH range of 7.011.5. It is sensitive to pH < 7. Therefore, it has been proposed for use as an in vivo, noninvasive pH meter. (4) It is stable when treated with mild chemical denaturants such as 1% sodium dodecyl sulfate or 8 M urea. It also retains fluorescence after fixation with formaldehyde or glutaldehyde, a very useful property in immunohistochemical applications. Most GFP applications so far involve the following two methods: 1. Direct transditction of the GFP gene into target cells with the GFP gene under the control of a constitutive or regulated promoter specific for the investigator's interest. For example, a "green mouse" was made by microinjecting GFP-expressing DNA fragment (with a strong, constitutively active promoter) into the pronuclei of single-cell mouse embryo [20]. The constitutively GFP-expressing mouse can be seen as a living GFP-labeled tissue bank providing all kinds of cells, tissues, and organs for homogeneous transplantation and tracing cell linages. On the other hand, regulated GFPexpressing transgenic mouse is an ideal model to study the effect of spatial/ temporal turn on/off status of different genes [8], the tracing of specific cell lineages [21], or effects of environmental factors on individual development [8]. In some cases, GFP coding sequence usually is inserted downstream of an internal ribosomal entry site (IRES), which is located 3' to a gene of interest. This allows the expression of target gene to be monitored both temporally and spatially by use of the GFP expression as a surrogate marker. This approach is especially powerful when combined with the transgenic approach [22-24]. 2. GFP coding sequence engineered into genes of interest by creation of in-frame fusions with target genes. This is possible thanks to the robust structure of the GFP gene. Under most circumstances, GFP-tagged proteins retain the bright green fluorescence of GFP, whose presence directly reflects the expression of its fusion partner. Most importantly, such fusions usually do not change the native distribution of the fusion partner in various cellular compartments. By use of this approach it is often possible to monitor many previously difficult-to-study proteins in unprecedented detail because it is now possible to follow the movement of individual proteins noninvasively


435

and continuously in a cell. This capability opened up new horizons for cell biology studies. This is especially true for fields such as membrane trafficking, a dynamic process that is difficult to study by immunofluorescence techniques [25]. It also revolutionized the study of another dynamic process, protein secretion. The use of GFP tag allows direct tracing of the budding, transport, and fusion process and makes it feasible to do time course assays. Presley and Scales visualized ER-to-Golgi transport using the GFP-tagged vesicular stomatitis virus G protein (VSVG) [26]. Recently, Stage-Zimmermann et al. visualized the nuclear export of the 60S ribosomal subunit by tagged ribosomal protein L l l b (Rplllb) with GFP. The binding, internalization, and separation of GnRHR membrane receptor and its agonist in living cells [27] were recently achieved by conjugating GFP to GnRHR. These studies are possible as long as the fusion does not affect the target gene structurally or functionally. In order to achieve this, GFP should be fused apart from the active or binding areas of the target protein to avoid interferences on three-dimensional folding and posttranslational process. Another very powerful application of the GFP family of proteins is the study of protein interaction by fluorescence energy resonance transfer (FRET). When two fluorescence proteins with different excitation-emission characteristics are close together, excitation of protein A leads to emission of light with a wavelength that coincides with the excitation wavelength of protein B, which makes protein B fluorescent. The intensity of the fluorescence from protein B should correlate with the degree of association of proteins A and B. The FRET reaction has been successfully used to measure the Ca2+ concentration in different subcellular organelles [28], to detect apoptosis [29], to establish a protease sensor [30], and many other applications where protein-protein interactions are involved. 3.

TUMOR CELL BIOLOGY

At the cellular level, the GFPs have greatly facilitated our understanding of tumor cell structures and signaling pathways. Many of the signal transduction proteins have now been examined in exquisite details with the help of the fluorescent proteins. An important question in cancer research is how to distinguish apoptosis and necrosis in tumor cells. DNA staining and TUNEL assay are established methods to show apoptosis in dead cells. However, they are not helpful in demonstrating the real-time apoptosis happening in living cells. One solution that has been tested is the observation of the fluorescence distribution of a GFP-tagged nuclear pore membrane protein. To achieve this, a neuroblastoma cell line was established that expresses a GFPPOM121 (a nuclear pore membrane protein) fusion protein [31]. It turned

Li et al.

436

out that GFP-POM121 diminished or lost its fluorescence in the apoptotic cells. In contrast, the intensity of this GFP fusion protein was unaffected in necrotic cells. Therefore, apoptosis and necrosis can be distinguished in living cells with this approach. Another example involves the evaluation of the activity and function of p53, an important tumor suppressor gene. Zhang et al. engineered a reporter GFP gene with p53-responsive sequences (p21 promoter and multiple copies of the PG13 enhancer that is derived from the consensus p53 binding sequence) to monitor wild-type p53 activity [32]. In stably transduced cells, up-regulated p53 gene activity was made apparent by the up-regulated GFP gene expression.

4.

TUMOR ANGIOGENESIS AND PHYSIOLOGY

Although there have been numerous studies of tumor development in vivo, there is still insufficient understanding of tumor physiology at the cellular level. The main reason is the lack of methods to follow tumor cells noninvasively in a manner similar to in vitro studies. The application of GFP promises to change the situation considerably. There is already a rich body of literature for the application of GFP in studies conducted in situ in animals. The advantage of GFP is obvious. It allows continuous or serial observation of tumor cell behavior in vivo in a noninvasive manner. GFP has been applied in many aspects of the tumor physiology studies. We will try to present a brief overview of several important areas. 4.1

Angiogenesis

Angiogenesis, the formation of new blood vessels from existing vessels, is key to tumor development and tumor invasion. It is now well established that angiogenesis is essential for tumor growth and development. It is intimately involved in tumor invasion (metastasis). However, it has been difficult to study the early stages of tumor angiogenesis because angiogenesis must be studied in vivo and there has been no available tool to visualize endothelial cells in vivo noninvasively. Most previous studies relied on immunohistological methods that characterize blood vessels by use of antibodies that specifically stain for endothelial cells. However, this approach only allows for angiogenesis to be observed when the tumor tissues are fixed. While many important discoveries were made by this approach, nothing is known at the earliest stages of tumor angiogenic process. Knowledge of the molecular mechanisms at these early stages of angiogenesis may provide important clues about how tumor angiogenesis initiates and how to stop tumor angiogenesis, an important antitumor strategy now taken by many investigators. The availability of GFP has significantly boosted studies in


437

this area. For example, tumor angiogenesis at the earliest stages of tumor growth was observed serially and noninvasively in a rodent dorsal skinfold chamber. In that study, the authors used GFP-labeled tumor cells in rodent dorsal skin window chamber models to noninvasively and serially observe (up to 4 weeks) tumor formation. For this purpose, a dorsal skinfold window chamber model (Fig. 1 A) created in rat or mouse was used to monitor tumor formation [33]. Utilizing this approach the authors serially followed tumor formation from the initial 20- to 50-cell transplant to tumors reaching 4-7

FIGURE 1 The growth of a tumor from single 4T1 cells in a syngeneic Balb/C mouse window chamber. (A) The dorsal skin window chamber model. (B) About 20 cells were injected in a Balb/C mouse window chamber and their growth was followed serially after the initial implantation. The red arrow in the day 2 panels indicates an elongated cell. Those in day 6 indicate dilated host vessels (in comparison with day 4). The arrows in day 8 panel indicate new microvessels. The pink ones point to tumor (localized in the yellow circle) microvessels, and the red ones point to dilated and/or tumor-induced vasculature outside the tumor. The size bars in day 0-8 panels represent 200 ^tm whereas that in day 20 represents 500

Li et al.

438

mm diameter. The results provided new insights into the complex interplay between tumor cells and host vasculature at these earliest stages of tumor growth. Four stages of early angiogenesis were seen to occur. (1) The initial orchestration of tumor angiogenesis involved migration of tumor cells toward existing vasculature before neovascularization. (2) Changes in surrounding microvessel structure, such as vasodilation and increased tortuosity, were seen at the 60- to 80-cell stage. (3) Clear demonstration of new vessel formation was seen at the 100- to 300-cell stage. (4) Both tumor lines developed intimate contact with developing neovasculature as the tumor continued to expand into surrounding normal tissue (Fig. IB). These results clearly demonstrated the power of GFP in facilitating noninvasive, cellularlevel studies of tumor angiogenesis with exquisite detail. Another example of GFP application in the study of tumor angiogenesis was carried out by Fukumura and colleagues. By use of a combination of a strain of transgenic mice that has a GFP gene under the control of the promoter for the vascular endothelial growth factor (VEGF) gene and the dorsal skinfold window chamber, they identified the predominant roles of stromal cells play in VEGF generation [34]. Finally, a transgenic mouse with the Tie2 gene promoter controlling the EGFP gene was created by Sato and colleagues [35]. This strain of mouse is unique in that all of the blood vessels uniformly express the GFP gene, making it very easy to clearly identify the blood vessels under a fluorescence microscope. Clearly the availability of this mouse strain will facilitate the study of angiogenesis by making available large quantities of endothelial cells that can easily be sorted by fluorescence-activated cell sorting. 4.2

Invasion and Micrometastasis

Invasion and metastasis are key steps in tumor development. However, earlystage invasion and micrometastasis are difficult to study. Attempts to study tumor cell migration were carried out many years ago with fluorescent dyelabeled tumor cells, and many important discoveries were made. However, the fluorescent dye approach can only be applied to study of tumor cell migration for hours at a time, limiting the kind of observation that can be carried out. In addition, radioactive labeling methods were used to observed the arrest of tumor cells in lung. However, it is difficult to study the process in detail at the cellular level. GFP provides a whole new way to carry out experiments in tumor invasion and metastasis studies. With GFP-transduced tumor cells, morphological changes such as polarization, cellular mobility in stroma, tumor cell distribution around blood vessel, and time-course intravasation and extravasations can all be observed noninvasively. By labeling different tumor cells with GFP |36| it is possible to compare the invasion


439

and metastasis behaviors of highly metastatic and poorly metastatic tumor cell lines [37]. GFP also brings some surprising results in metastasis research. For example, it was once believed that distal metastasis happened only after metastatic cells extravasated at distal sites. Al-Mehdi et al. reported that GFP-labeled transformed rat embryonic fibroblast and human fibrosarcoma cell lines metastasize to the lung by attachment to precapillary arterioles [38]. They clearly demonstrated that this was not due to trapping, since the vessels were larger than the diameter of the attached cells. Cells proliferated selectively inside vessels up to a size beyond which the vessel appeared to burst from pressure of the proliferating cell mass. Growth of micrometastases was not observed outside the vessels. This observation challenges the paradigms that hematogenous metastases are mainly formed from tumor cells trapped in capillaries and that growth can only occur once the cells have escaped from the vasculature. In a series of papers, Hoffman and colleagues traced the growth of GFP-labeled tumor cells with unprecedented detail [39-41]. In many instances, single cells or small colonies could be visualized. In addition, noninvasive, whole-body imaging of tumors, metastases, and angiogenesis in real time has been achieved [42]. 4.3

Hypoxia

Besides studying tumor growth and angiogenesis, GFP can also be used to study important problems in tumor physiology. With appropriate genetic manipulation, it is possible to visualize various aspects of important physiological conditions in tumors. For example, tumor hypoxia has long been studied as an important parameter in tumor physiology and therapy. Hypoxia has been identified as a major factor in determining radiation resistance of cancer cells. It has also been suggested as a major factor in tumor angiogenesis since it induces the production of VEGF genes. It has also been reported to be an important prognostic factor for several types of cancer. Therefore, there has been a recent surge of interest in the study of hypoxia in tumors. The use of hypoxia-inducible promoters and the GFP gene makes it possible to visualize hypoxia in tumors noninvasively and thereby provide an approach to study it in greater details in combination with other aspects of tumor development. Koshikawa created a transgenic tumor cell line in which the VEGF promoter is used to control the GFP gene. Up-regulation of GFP is observed under hypoxic condition [43]. Data from our group indicated that it is possible to visualize hypoxia in the dorsal skin window chamber by use of cells that have been artificially transduced with a reporter construct where the GFP gene is under the control of an enhanced hypoxiaresponsive promoter (Fig. 2A). In addition, it is possible to visualize activation of another promoter, hsp70 promotor for the heat-shock protein 70

Li et al.

440

FIGURE 2 Visualization of stress-inducible gene expression in a rat tumor window chamber. R3030Ac rat mammary tumor cells that had been stably tranduced with a GFP under the control of an hsp70 gene promoter was used to form a tumor in the window chamber. Ten days after implantation, the window chamber was observed using transmitted white light (A) and epifluorescence (B). Notice the pattern of green fluorescence in panel B, which is reflective of gene activation for hsp70. The size bar in panel B represents 1 mm. (See color insert.)

gene (Fig. 2B), whose activation can indicate a variety of stress signals such as heat, glucose deprivation, presence of extra free radicals, and hypoxia [44]. 4.4

pH

The protonation or deprotonation state of the chromophore of GFP is pH sensitive and can cause the shift of its excitation and emission spectral properties. The fluorescene intensity of GFP can also be affected by pH. This property has been used by a number of groups to measure intracellular pH noninvasively. As acidic pH is a common property of many tumors, GFP should also be very useful in this aspect of tumor physiology studies. Because heterologously synthesized GFP has no specific preference to intracellular organization, it is possible to target GFP to various subcellular organelles in a cell, thereby studying proteins in those organelles. In one study, investigators engineered GFP with various targeting of peptide sequences toward different organelles [45]. By pH titration and GFP quenching experiments plus in vivo pH calibration, they measured the subcellular pH values in different organelles such as mitochondria, Golgi body, and endoplasmic reticulum. Such intracellular pH indicators are very site specific, even at a subcellular level. They have relatively less toxicity in comparison with other chemical pH tracers. In addition, it allows direct lifetime pH measurement


441

over a period of time. This is difficult to achieve by the widely used microelectroprobe methods.

5.

CANCER GENE THERAPY

Gene therapy is a relatively new field in cancer research. With the completion of the human genome and rapid advances in genomics, there are a large number of candidate genes suitable for testing in cancer gene therapy studies. One of the biggest hurdles in cancer gene therapy is effective delivery of therapeutic genes to tumor cells. This is an important issue in cancer therapy because virtually all of the gene therapy vectors are bigger than traditional chemotherapeutic drugs. Various strategies have been taken to make gene therapy tumor-cell specific. One of the tools that should be very useful in facilitating the creation of more effective tumor cell targeting gene therapy vectors is an approach that allows the noninvasive, dynamic monitoring of gene transduction in vivo by various gene therapy vectors. Again, GFP is an ideal tool in this case. By use of a dorsal skinfold window chamber, we were able to monitor the activity of adenovirus in infecting a mammary cell carcinoma noninvasively over the course of days. In addition, long-term, noninvasive monitoring of gene expression is achieved in the retina [46] and in the brochial epithelium [47]. In summary, green fluorescent protein has already been used in a wide variety of biological research fields. The list of its applications is expected to grow in the foreseeable future.

REFERENCES 1.

2.

3. 4. 5.

6.

O Shimomura, FH Johnson, Y Saiga. Extraction, purification and properties of aequorin, a bioluminescent protein from the luminous hydromedusan, Aequorea. J Cell Comp Physiol 59:223-239, 1962. FH Johnson, O Shimomura, Y Saiga, LC Gershman, GT Reynolds, JR Waters. Quantum efficiency of cypridina luminescence, with a note on that of Aequorea. J Cell Comp Physiol 60:85-103, 1962. M Chalfie, Y Tu, G Euskirchen, WW Ward, DC Prasher. Green fluorescent protein as a marker for gene expression. Science 263:802-805, 1994. BP Cormack, RH Valdivia, S Falkow. FACS-optimized mutants of the green fluorescent protein (GFP). Gene 173:33-38, 1996. A Terskikh, A Fradkov, G Ermakova, A Zaraisky, P Tan, AV Kajava, X Zhao, S Lukyanov, M Matz, S Kim, I Weissman, P Siebert. "Fluorescent Tinier": protein that changes color with time [In Process Citation]. Science 290:15851588, 2000. M Ikawa, K Kominami, Y Yoshimura, K Tanaka, Y Nishimune, M Okabe. A

442

Li et al.

rapid and noninvasive selection of transgenic embryos before implantation using green fluorescent protein (GFP). FEES Lett 375:125-128, 1995. 7. YP Wu, E McMahon, MR Kraine, R Tisch, A Meyers, J Frelinger, GK Matsushima, K Suzuki. Distribution and characterization of GFP( + ) donor hematogenous cells in Twitcher mice after bone marrow transplantation. Am J Pathol 156:1849-1854, 2000. 8. S Kawamoto, H Niwa, F Tashiro, S Sano, G Kondoh, J Takeda, K Tabayashi, J Miyazaki. A novel reporter mouse strain that expresses enhanced green fluorescent protein upon Cre-mediated recombination. FEBS Lett 470:263-268, 2000. 9. D Tannahill, S Bray, WA Harris. A Drosophila E(spl) gene is "neurogenic" in Xenopus: a green fluorescent protein study. Dev Biol 168:694-697, 1995. 10. BM Tarn, OL Moritz, LB Hurd, DS Papermaster. Identification of an outer segment targeting signal in the COOH terminus of rhodopsin using transgenic Xenopus laevis. J Cell Biol 151:1369-1380, 2000. 11. A Amsterdam, S Lin, N Hopkins. The Aequorea victoria green fluorescent protein can be used as a reporter in live zebrafish embryos. Dev Biol 171:123129, 1995. 12. R Rizzuto, M Brini, P Pizzo, M Murgia, T Pozzan. Chimeric green fluorescent protein as a tool for visualizing subcellular organelles in living cells. Curr Biol 5:635-642, 1995. 13. VG Thackray, RH Young, JE Hooper, SK Nordeen. Estrogen agonist and antagonist action on the human estrogen receptor in Drosophila [In Process Citation]. Endocrinology 141:3912-3915. 2000. 14. CM Coburn, CI Bargmann. A putative cyclic nucleotide-gated channel is required for sensory development and function in C. elegans. Neuron 17:695706, 1996. 15. P De Wulf, L Brambilla, M Vanoni, D Porro, L Alberghina. Real-time flow cytometric quantification of GFP expression and gfp-fluorescence generation in saccharomyces cerevisiae [In Process Citation]. J Microbiol Meth 42:5764. 2000. 16. JR Damn, N Gomez-Ospina, M Winey, DJ Burke. The spindle checkpoint of saccharomyces cerevisiae responds to separable microtubule-dependent events [In Process Citation]. Curr Biol 10:1375-1378, 2000. 17. BJ Feilmeier. G Iseminger, D Schroeder, H Webber, GJ Phillips. Green fluorescent protein functions as a reporter for protein localization in Escherichia coli. J Bacteriol 182:4068-4076, 2000. 18. JM Mullaney, RB Thompson, Z Gryczynski. LW Black. Green fluorescent protein as a probe of rotational mobility within bacteriophage T4 [In Process Citation). J Virol Meth 88:35-40, 2000. 19. J Chen, Y Watanabe. N Sako, K Ohshima, Y Okada. Complete nucleotide sequence and synthesis of infectious in vitro transcripts from a full-length cDNA clone of a rakkyo strain of tobacco mosaic virus. Arch Virol 141:885900, 1996. 20. M Okabe, M Ikawa, K Kominami, T Nakanishi, Y Nishimune. "Green mice" as a source of ubiquitous green cells. FEBS Lett 407:313-319, 1997.

Green Fluorescent Protein and Tumor Angiogenesis 21.

22.

23.

24.

25. 26.

27.

28.

29.

30.

31.

32.

33.

34.

35.

443

M Sato, Y Yasuoka, H Kodama, T Watanabe, JI Miyazaki, M Kimura. New approach to cell lineage analysis in mammals using the Cre-loxP system. Mol Reprod Dev 56:34-44, 2000. XW Yang, C Wynder, ML Doughty, N Heintz. BAC-mediated gene-dosage analysis reveals a role for Ziprol (Ru49/Zfp38) in progenitor cell proliferation in cerebellum and skin. Nat Genet 22:327-335, 1999. XW Yang, P Model, N Heintz. Homologous recombination based modification in Escherichia coli and germline transmission in transgenic mice of a bacterial artificial chromosome. Nat Biotechnol 15:859-865, 1997. J Zuo, J Treadaway, TW Buckner, B Fritzsch. Visualization of alpha9 acetylcholine receptor expression in hair cells of transgenic mice containing a modified bacterial artificial chromosome. Proc Natl Acad Sci USA 96:1410014105, 1999. S Kupzig. Methods in Enzymology, Vol. 302. San Diego: Academic Press, 1999, pp 3-10. SJ Scales, R Pepperkok, TE Kreis. Visualization of ER-to-Golgi transport in living cells reveals a sequential mode of action for COPII and COPI. Cell 90: 1137-1148, 1997. A Cornea, JA Janovick, X Lin, PM Conn. Simultaneous and independent visualization of the gonadotropin-releasing hormone receptor and its ligand: evidence for independent processing and recycling in living cells [published erratum appears in Endocrinology 1999 Oct;140(10):4771]. Endocrinology 140: 4272-4280, 1999. A Miyawaki, J Llopis, R Heim, JM McCaffery, JA Adams, M Ikura, RY Tsien. Fluorescent indicators for Ca 24 based on green fluorescent proteins and calmodulin [see comments]. Nature 388:882-887, 1997. X Xu, AL Gerard, BC Huang, DC Anderson, DG Payan, Y Luo. Detection of programmed cell death using fluorescence energy transfer. Nucleic Acids Res 26:2034-2035, 1998. R Heim, RY Tsien. Engineering green fluorescent protein for improved brightness, longer wavelengths and fluorescence resonance energy transfer. Curr Biol 6:178-182, 1996. G Imreh, M Beckman, K Iverfeldt, E Hallberg. Noninvasive monitoring of apoptosis versus necrosis in a neuroblastoma cell line expressing a nuclear pore protein tagged with the green fluorescent protein. Exp Cell Res 238:371376, 1998. W Zhang, S Labrecque, E Azoulay, R Dudley, G Matlashewski. Development of a p53 responsive GFP reporter; identification of live cells with p53 activity. J Biotechnol 84:79-86, 2000. CY Li, S Shan, Q Huang, RD Braun, J Lanzen, K Hu, P Lin, MW Dewhirst. Initial stages of tumor cell-induced angiogenesis: evaluation via skin window chambers in rodent models. J Natl Cancer Inst 92:143-147, 2000. D Fukumura, R Xavier, T Sugiura, Y Chen, EC Park, N Lu, M Selig, G Nielsen, T Taksir, RK Jain, B Seed. Tumor induction of VEGF promoter activity in stromal cells. Cell 94:715-725, 1998. T Motoike, S Loughna, E Perens, BL Roman, W Liao, TC Chau, CD Rich-

444

36.

37.

38.

39.

40.

41.

42.

43.

44.

45. 46.

47.

Li et al.

ardson, T Kawate, J Kuno, BM Weinstein, DY Stainier, TN Sato. Universal GFP reporter for the study of vascular development. Genesis 28:75-81, 2000. JS Condeelis, J Wyckoff, JE Segall. Imaging of cancer invasion and metastasis using green fluorescent protein [In Process Citation!. Eur J Cancer 36:16711680, 2000. GN Naumov, SM Wilson, 1C MacDonald, EE Schmidt, VL Morris, AC Groom, RM Hoffman, AF Chambers. Cellular expression of green fluorescent protein, coupled with high-resolution in vivo videomicroscopy, to monitor steps in tumor metastasis. J Cell Sci 112:1835-1842, 1999. AB Al-Mehdi, K Tozawa, AB Fisher, L Shientag, A Lee, RJ Muschel. Intravascular origin of metastasis from the proliferation of endothelium-attached tumor cells: a new model for metastasis. Nat Med 6:100-102, 2000. T Chishima, Y Miyagi, X Wang, Y Tan, H Shimada, A Moossa, RM Hoffman. Visualization of the metastatic process by green fluorescent protein expression. Anticancer Res 17:2377-2384, 1997. M Yang, S Hasegawa, P Jiang, X Wang, Y Tan, T Chishima, H Shimada, AR Moossa, RM Hoffman. Widespread skeletal metastatic potential of human lung cancer revealed by green fluorescent protein expression. Cancer Res 58:42174221, 1998. M Yang, P Jiang, Z An, E Baranov, L Li, S Hasegawa, M Al-Tuwaijri, T Chishima, H Shimada, AR Moossa, RM Hoffman. Genetically fluorescent melanoma bone and organ metastasis models. Clin Cancer Res 5:3549-3559, 1999. M Yang, E Baranov. P Jiang, FX Sun, XM Li, L Li, S Hasegawa, M Bouvet, M Al-Tuwaijri, T Chishima, H Shimada, AR Moossa, S Penman, RM Hoffman. Whole-body optical imaging of green fluorescent protein-expressing tumors and metastases. Proc Natl Acad Sci USA 97:1206-1211, 2000. N Koshikawa, K Takenaga, M Tagawa, S Sakiyama. Therapeutic efficacy of the suicide gene driven by the promoter of vascular endothelial growth factor gene against hypoxic tumor cells. Cancer Res 60:2936-2941, 2000. A Wysocka, Z Krawczyk. Green fluorescent protein as a marker for monitoring activity of stress-inducible hsp70 rat gene promoter. Mol Cell Biochem 215: 153-156, 2000. M Kneen, J Farinas, Y Li, AS Verkman. Green fluorescent protein as a noninvasive intracellular pH indicator. Biophys J 74:1591-1599, 1998. F Rolling, WY Shen, NL Barnett, H Tabarias, Y Kanagasingam, I Constable, PE Rakoczy. Long-term real-time monitoring of adeno-associated virus-mediated gene expression in the rat retina. Clin Exp Ophthalmol 28:382-386, 2000. TR Flotte. SE Beck, K Chesnut, M Potter. A Poirier, S Zolotukhin. A fluorescence video endoscopy technique for detection of gene transfer and expression. Gene Ther 5:166-173, 1998.

14 Near-Infrared Imaging with Fluorescent Contrast Agents Eva M. Sevick-Muraca, Anuradha Godavarty, Jessica P. Houston, Alan B. Thompson, and Ranadhir Roy Texas A&M University, College Station, Texas, U.S.A.

1.

INTRODUCTION

In this chapter, the development of near-infrared (NIR) excitable contrast agents, the methods to excite and register the resulting fluorescence that originates from targeted tissues, and the mathematical approaches to perform three-dimensional tomographic reconstructions from the measured, re-emitted fluorescence are reviewed in the context of an evolving diagnostic imaging modality for the clinic. Specifically, whereas previous chapters in this volume deal with the measurement of fluorescence for probing surface and subsurface tissues either directly or through endoscopic examination, this chapter focuses on the use of exogenous fluorescent contrast agents that can be used for NIR fluorescence-enhanced optical imaging of deep tissues. The opportunity to develop an emission-based tomographic imaging modality without the use of radioisotopes is offered by NIR fluorescent agents. 1.1

What Is NIR Optical Imaging?

NIR fluorescence-enhanced optical imaging has evolved from NIR optical imaging techniques that depend on endogenous contrast, or the natural dif445

446

Sevick-Muraca et al.

ferences in tissue optical properties. As a consequence, NIR fluorescenceenhanced optical imaging can be explained in the context of NIR optical imaging. Briefly, NIR optical imaging takes advantage of the "therapeutic window" between 700 and 900 nm in which tissues exhibit low absorbance but high scattering capacity. As a result, light in this wavelength regime can scatter through several centimeters of tissues before being extinguished by absorption. Absorption occurs primarily from the tissue chromophores of oxy- and deoxyhemoglobin, fat, melanin, and water, whereas scattering is typically due to refractive index differences of extracellular and intracellular structures. Indeed, the propagation of NIR light through tissues is governed by its endogenous optical properties, which vary with degree of vascularity, fatty tissue content, water content, cell densities, etc. Over the past decade, an intense area of investigation has deployed measurements of NIR light to tomographically recover interior maps of endogenous optical properties from exterior measurements of NIR light that is transmitted between a single point source and a point detector located at the air-tissue interface [ 1 ] . The ability to detect diseased tissues with NIR light depends critically on the "optical contrast," or the consistent differences between the absorption and scattering properties of normal and diseased tissue volumes of interest, and the tomographic reconstruction algorithms needed to recover three-dimensional maps of absorption and scattering properties. Stunning results have been achieved in the area of optical mammography, which seeks to employ the absorption contrast owing to tumor angiogenesis, or the hypervascularization of tumor periphery, for tumor detection. While exciting and promising, the necessity of angiogenesis-mediated absorption contrast for diagnostic imaging limits the potential for using NIR techniques for other diagnostic applications, such as assessing sentinel lymph node staging, metastatic spread, and multifocality of breast disease. Such diagnostics could probably not be done with natural endogenous contrast. In addition, since the endogenous contrast owing to angiogenesis can be expected to be low in small lesions and to be nonspecific to cancer, there is certainly a limitation for NIR detection of nonpalpable disease in dense breast tissue. There is some evidence for scattering contrast between normal and diseased tissue, but little work has been done to solve the tomography problem based on scattering contrast alone [2]. Hence, the capability for high-resolution molecular imaging within tissues using unassisted NIR optical techniques is somewhat limited and can be expanded through the use of contrast-enhancing agents. 1.2

What Is NIR Fluorescence-Enhanced Optical Imaging/Tomography?

NIR fluorescence-enhanced optical imaging seeks to preserve the penetration depth (>l cm) of NIR optical imaging techniques, but provide added

Near-Infrared Imaging

447

contrast owing to targeting and reporting exogenous dyes that are excitable in the NIR region. NIR fluorescence-enhanced optical imaging involves three components: (1) introduction of the targeting and reporting fluorescent dye or contrast agent; (2) illumination of the tissue surface and collection of the fluorescence that is generated deeply within the tissue and propagates to the surface for detection (Fig. 1); and for certain measurement cases; (3) tomographic reconstruction for three-dimensional rendering of interior tissues. Tomographic imaging requires the solution of two problems: (1) the forward imaging problem, i.e., the prediction of the measurements of NIR/fluorescence light generation and propagation measured at the air-tissue interface given the three-dimensional interior optical property map; and (2) the inverse problem, i.e., the prediction of the three-dimensional optical property map given the measurements of NIR/fluorescence light propagation at the airtissue interface (Fig. 2). This chapter is organized to describe targeting and reporting fluorescent dyes, measurement geometries and the forward and inverse problems for tomographic reconstruction, along with their unique challenges and advantages. So as to properly characterize previous work in the proper context, we first describe the methods for measurement (Sec. 2), then describe the studies using a variety of fluorescent contrast agents (Sec. 3) and approaches for tomographic reconstructions (Sees. 4 and 5). To ensure completeness and reader comprehension, we first briefly review the essential

FIGURE 1 NIR fluorescence-enhanced optical imaging depends on illumination of the tissue surface with excitation light, propagation of the excitation light to the embedded fluorophore, generation of the emission light, propagation to the air-tissue interface, and collection of the fluorescence.

448


FIGURE 2 General illustration of (a) the forward imaging problem in which the tissue optical properties are known and used to predict the measurement and (b) the inverse imaging problem in which the measurements are used to obtain the unknown tissue optical properties. (From Ref. 1.)


449

aspects of the fluorescence decay process, as its parameters will be referred to through the remainder of the chapter. 1.3 The Fluorescence Decay Processes The basic principles behind fluorescence-enhanced NIR optical imaging and tomography focus first on the kinetics of fluorescence generation (Fig. 3). When a molecule of significant aromaticity absorbs light corresponding to a transitional energy level, it becomes activated into a "singlet" state from where it can relax radiatively, releasing light of lower energy (or higher wavelength) than the incident light. Typically, the rate of radiative decay, F, and the rate of nonradiative decay, knr, from the "singlet state" to the ground state is mediated by the local environment of the activated dye molecule. The fluorescent lifetime r (or the mean time that the fluorophore is in the activated singlet state) is influenced by the relative rates of decay as illustrated in Fig. 3. The parameter ris described as r= \I(Y + k nr ). The quantum efficiency of the fluorescent emission, $, is the fraction of excited dye mol-

FIGURE 3 The Jablonski diagram illustrating the activation of fluorophore to its single excited state and its nonradiative and radiative (fluorescence) relaxation to the ground state. The fluorescence lifetime r, is equivalent to the mean time that the fluorophore remains in its activated state and the quantum efficiency, c/>, is the proportion of relaxations that occur radiatively. In PDT agents, the singlet excited state can undergo "intersystem crossing" in which the spin state of the electron is flipped. Relaxation of the triplet excited state is forbidden until the electron spin state is reversed. The lifetimes of the triplet state are on the order of microseconds to milliseconds and are termed phosphorescence. (From Ref. 1.)

450


ecules, or activated fluorophores, which relax radiatively. It is also described by = r/(F + k n r ). As described later, the fluorescent decay process from relevant MR fluorophores is usually more complicated than the first-order decay from a single activated state that is depicted in Fig. 3. Depending on the targeting and reporting construct employed, the kinetics of the decay process are governed by multiple activated states, quenching, as well as fluorescence energy resonance transfer (FRET). In FRET, energy is nonradiatively transferred from a "donor" fluorophore that is optically excited to a nearby "acceptor" fluorophore that subsequently undergoes radiative relaxation. We refer to the fluorescent contrast agent as the entity containing the targeting and reporting constructs as well as one or more fluorochromes or fluorophores, which undergo the decay process described herein in order to provide the "beacon" signal for fluorescence-enhanced optical imaging. The intensity of fluorescent light that is detected at the tissue surface is dependent on (1) the amount of excitation light present within the tissue for excitation of the fluorophore, (2) the amount of fluorescent contrast agent present at the location of interest, (3) the decay kinetics which, in the case of a dye exhibiting first-order relaxation kinetics, influence the lifetime, r, and the quantum efficiency of the dye, 6 rejection of excitation light and to use a set of coupled interference filters for OD 4 rejection outside the bandwidth of the emission light. Nonetheless, for the plethora of model studies employing rat or mouse subcutaneous tumors whereby penetration depth is severely limited, the use of simple interference filters has proven the ability to provide spatial twodimensional discrimination of targeted tissue regions as summarized below. The challenges remain to extend fluorescence optical imaging and tomography to penetration depths that would be clinically relevant.

454

2.3


Continuous Wave or Intensity-Based Approaches

Typically, most demonstrations of fluorescence-enhanced imaging or spectroscopy involve time-invariant, continuous wave (CW) or intensity-based approaches. In these approaches, the incident excitation energy is constant over time scales of milliseconds, and the detected re-emitted fluorescent energy is likewise constant. The excitation light propagates through the scattering medium and as it propagates, it is exponentially attenuated relative to the incident light. When the attenuated excitation light encounters a fluorochrome, fluorescent light is generated which is further attenuated owing to the quantum efficiency and lifetime of the fluorochrome responsible for its generation. As the fluorescent light propagates to the air-tissue interface, it is further attenuated owing to the absorption and scattering properties of the intervening tissues. Specifically, the measurement of interest in CW techniques is emission intensity, or a least the intensity of light that is registered past the interference or holographic filters at the photodetector. Nearly all investigations indicative of fluorescent contrast-enhanced imaging focus on CW intensity, owing to the slight fluorescent photon budget from random media and the difficulty in registering measurements in time-dependent techniques. The first CW fluorescence enhanced optical spectroscopy/imaging studies focused on quantitation of PDT agents in tumor tissues. Using surface illumination and point, fiberoptic collection of re-emitted light from zinc phthalocyanine dye delivered in liposomes to a mouse tumor, Biolo et al. [5] showed the measurement of fluorescence for following the time course of therapeutic agent deposition in tumor tissues. Straight and coworkers [6] employed the same approach to assess PDT agent distribution but used a CCD camera to image the re-emitted fluorescence across an area of tissue following internal excitation with an interstitial optical fiber. To the authors' best knowledge, this was the first demonstration of use of a CCD camera for registration of fluorescence signals from a PDT or an exogenous diagnostic agent. Following this study, the use of a CCD or an image-intensified CCD (also called an ICCD) for area detection following point or planar source illumination predominated in the exogenous contrast agent studies that are reported in the literature. CW techniques are intrinsically unspecific to changes in fluorescent decay kinetics and concentration of fluorophore. For example, the fluorescent intensity from a nonscattering solution containing a fluorophore that exhibits a first-order relaxation process can be written as:

i.

log — = -/A:I|.

r

exp(-t/T) dt = -e[C]/r

(1)


455

where Imx represents the detected emission or incident excitation intensity; /xaf is the absorption cross section of the fluorophore; e is the extinction coefficient of the fluorophore; and [C] is the concentration of the fluorophore. What is notable for CW measurements of fluorescence is that one cannot distinguish between a change in the concentration of fluorophore and a change of the fluorescence decay kinetics (or lifetime). Indeed, fluorescence lifetime spectroscopy measurements with CW light focus on the measurement of the change in emission intensity in the presence or absence of a quencher while the concentration of the fluorophore is held constant. In tissues, the situation is far more complicated since the amount of fluorescent light collected depends not only on the concentration and the fluorescence decay kinetics, but also on optical properties of the tissues. The relationship between CW fluorescence measurements and the tissue optical properties will be discussed more fully in Sec. 4. 2.4

Time Domain Approaches

Figure 5 provides an illustration describing the propagation of a pulse of excitation light emanating from a point source into a random media mimicking the optical properties of the tissue. The impulse of excitation light propagates through the scattering media and as it does so becomes spatially and temporally broadened relative to the incident impulse of excitation light. When the broadened pulse of excitation light encounters a fluorochrome, a resulting fluorescence pulse is generated that is further attenuated and broadened owing to the quantum efficiency and lifetime of the fluorochrome responsible for its generation. As the generated fluorescence pulse propagates to the air-tissue interface, it is further attenuated and broadened owing to the absorption and scattering properties of the intervening tissues. One can see the advantages of time-dependent measurements for biomedical imaging and spectroscopy immediately by comparing CW and time domain measurements in nonscattering solutions. Consider the simple case of the time-dependent emission intensity generated from a dye exhibiting first-order relaxation following an incident impulse of excitation light: exp(-tVr) dt' = -e[C]

exp(-tVr) dt'

(2)

Jo

Upon examining the time dependence of the detected emission intensity, the decay kinetics can be obtained directly and independently from the concentration of fluorophore. As will be discussed in Sec. 4, the temporal

Source / Pulse

Time(IO- 1 ( l s)

Detected / Light

Time(10- 10 s) (e)

Time-gated Measurement

At

(0

Time(10- 1! 's)

FIGURE 5 Schematic of time domain measurement approaches used in NIR optical tomography. Time domain photon migration (TDPM) imaging approaches utilize an incident impulse of light that results in the propagation of the pulse throughout the tissue that attenuates as a function of distance from the source and time following its incident impulse. The detected pulse is measured as intensity versus time and represents the photon "time of flight." Panel (a) illustrates the light distribution in tissue from a pulse point source after 1 x 10 10 sec, (b) 25 x 10 10 sec, and (c) 150 x 10 10 sec following the incident impulse. The corresponding recorded data during the time intervals at the detector is illustrated in panels (d) through (f). A time gated illumination measurement is shown in panel (f) in which the integrated intensity measured within a specified window is measured. (From Ref. 1.)


457

discrimination of detected emission and excitation light may enable one to distinguish between concentration of fluorochrome, its decay properties, as well as the tissue optical properties. However, time domain approaches are difficult to implement practically using single-photon counting and gated integration techniques. In single-photon counting techniques (also called time-correlated photon counting techniques), the first fluorescent photon that reaches the detector following each incident impulse of excitation light is counted statistically in order to compile a distribution of photon "collection times" that collectively count the excitation and fluorescent photon "time of flight" as well as the fluorescence decay kinetics. These approaches are called single-photon counting methods and can only acquire single emission photon counts as a function of time for at most one out of every 10 or more pulses in order to statistically sample the entire photon time of flight distribution. Specifically, single-photon counting provides histograms of the first arrival photon collection times with temporal resolution that typically exceeds the detector response time. Gated integration methods may present faster data acquisition times as they involve collecting light in a small temporal window at varying times following an incident pulse of light. Briefly, the techniques consist of repeatedly measuring the intensity of light arriving between time t and t + At at a detector following an incident pulse of light (Fig. 5f). Upon moving the start time t, one can recover the distribution of photon time of flight. Regardless of whether time-gated or single-photon counting techniques are employed, the photon collection times are associated with the excitation photon time of flight to the fluorophore, the time associated with the kinetics of absorption and radiative relaxation, and the emission photon time of flight from the embedded fluorophore to the collection site on the air-tissue interface. The first time domain approach for assessing PDT agent distribution was used with a planar wave excitation following area, gated integration detection. The ICCD system was the area detection system of choice for Kohl et al. [7] as well as for Cubeddu et al. [8,9]. Their studies consisted of illuminating the tissue surface area overlying the tissue region of interest with a pulsed light source. The fluorescent light emanating from the tissue was captured by an ICCD that was gated with nanosecond resolution to register the later fluorescence generated from the long-lived, exogenous PDT agents and to reject on a temporal basis the native, short-lived fluorescence. In their studies, native, short-lived fluorescence resulting from excitation in the far-UV region contaminated the detected signals and could be removed by simple time gating. Gating also enabled further rejection of the multiply scattered excitation light, further enhancing SNR and overcoming the char-

458


acteristic identified above as the limitation governing the sensitivity and penetration depth for fluorescence-enhanced optical imaging. Regardless of the time domain approach used, whether single-photon counting or gated integration measurements, a significant number of pulses is needed to acquire the statistical photon count or intensity measurement of the delayed time window of interest. If accomplished successfully, time domain measurements can provide direct measurements of fluorescence lifetime shortening and lengthening independent of concentration of fluorescent yield. In contrast to CW measurements, time domain measurements can discriminate between changes in fluorochrome concentration and fluorescent decay kinetics. Unfortunately, the SNR of time domain approaches suffers notoriously in comparison with CW measurements (as well as in comparison with frequency domain photon migration (FDPM) approaches described below). The issue of reduced SNR is evident in the time domain studies by Ntziachristos et al. [10] who employed indocyanine green (ICG) as an intravenous, blood pooling, systemic contrast agent in breast cancer patients undergoing concomitant gadolinium-enhanced magnetic resonance imaging (MRI). In their studies that employed incident pulses of excitation light delivered to the tissue surface via a fiberoptic and collection from a fiberoptic collection, the detection of fluorescence signal via single-photon counting was not reported, presumably owing to a low SNR. Instead, they measured the excitation light signal for registration of contrast owing to the absorption of excitation light. Clearly, with a reduced average fluorescent photon budget, the further temporal discrimination of fluorescent photon arrival using point illumination and point detection may be fraught with SNR issues, especially when probing deep tissues within the human female breast. On the other hand, upon using gated integration of an ICCD with area illumination and area detection, the ability to assess the fluorescence owing to long-lived fluorophores in subsurface tissues has been clearly demonstrated by Cubeddu, Kohl, and their coworkers [7-9]. These limited studies point to the use of illumination and detection scenarios in time-dependent measurements of light propagation. 2.5

Frequency Domain Approaches

CW steady approaches are contrasted by time-dependent approaches, which can themselves be broken into time domain approaches of single-photon counting and of gated integration as described earlier, and frequency domain approaches. While time domain approaches focus on the emission intensity measurements as a function of time after incident impulse of excitation light, the frequency domain approaches provide the Fourier analogue to timedependent measurement with the added advantage that frequency domain


459

measurements represent steady-state measurements of a single frequency component. Briefly, frequency domain approaches consist of an incident source of excitation light, modulated at frequency, w. The "photon density wave" of excitation light propagates through the scattering medium and as it does so becomes amplitude attenuated and phase-shifted relative to the incident light (Fig. 6). When the modulated excitation light encounters a fluorochrome, a resulting fluorescence photon density wave is generated that is further attenuated and phase shifted owing to the quantum efficiency and lifetime of the fluorochrome responsible for its generation. As the fluorescent photon density wave propagates to the air-tissue interface, it is further attenuated and phase shifted owing to the absorption and scattering properties of the intervening tissues. Figure 6b illustrates detected phase delay and amplitude at some distance from an incident point source. Since the amplitude of the detected fluorescence is insensitive to the intensity owing to the ambient light, the approach has clear advantages for clinical application in a nonlight-tight environment. In addition, since frequency domain approaches offer a steady-state measurement of time-dependent light propagation processes, they have comparatively high SNR with respect to time domain approaches, and retain the signal dependency on lifetime that is otherwise missing in CW measurements. It is not surprising that the traditional point source and point detector tomographic reconstructions have been demonstrated using frequency domain techniques in large volumes with sparse measurements, whereas tomographic reconstructions with CW light has been limited to small volumes with dense measurements. Like time domain, frequency domain approaches can distinguish between the amount of dye and lifetime kinetics. In nonscattering solutions of a dye exhibiting first-order relaxation kinetics, the measurement of phase, 0, and amplitude modulation, I AC , divided by the average intensity, IDC, can be written in terms of the lifetime and the concentration of fluorophore [11]:

Of 6(io)\

>L = *tan-i/(wr);

a

= lAO

1 +

;

2

,_ (3)

V

Of course, more complicated expressions for dyes exhibiting higher order decay kinetics can be derived, yet all enable distinction of decay kinetics from fluorophore concentration. Typically, these measurements in nonscattering solutions are referred to as "phase modulation" measurements. As discussed in Sec. 4, when conducted in scattering solutions frequency domain measurements, or phase modulation measurements, enable discrimi-


460

- 8 E-06 - 7.E-06 - 6.E-06 - 5.E-06 8 1.5

3 O C/3

- 4.E-06 - 3.E-06

1 -

- 2.E-06 0.5

- 1 .E-06 - O.E + 00 (b)

time (ns)

FIGURE 6 Schematic of the frequency domain measurement approach used in NIR optical tomography. Frequency domain photon migration (FDPM) imaging consists of an incident, intensity-modulated light source that creates a "photon density wave" that propagates continuously throughout the tissue. Panel (a) is a depiction of light distribution in tissue due to a modulated source (exaggerated for purposes of illustration), and panel (b) illustrates the detected signal (solid line) in response to the source illumination (dotted line). The typical frequency domain quantities are the phase shift 0, the amplitude of each wave IAC, and the bias of each wave ct>DC. As shown in panel (b), the intensity wave that is detected some distance away from the source is amplitude attenuated and phase delayed relative to the source. (From Ref. 1.)


461

nation of fluorophore concentration, lifetime, as well as the tissue optical properties. In addition, the relationship between time and frequency domain measurements is given by the Fourier transform: Ix,m(t)exp(-io>t) dt

(4)

Here P(o>) is the power of the measurement at modulation frequency co, and its real and imaginary components can be used to determine the phase delay and amplitude attenuation directly (as done above for nonscattering measurements of a dye exhibiting single exponential decay kinetics). Frequency domain measurements for fluorescence-enhanced contrast imaging involve area illumination and area detection using a gain-modulated ICCD system, as well as modulated point source illumination via fiberoptics and point collection via fiberoptics. It is important to note that because an intensity-modulated component is measured, frequency domain approaches provide a simple means for ambient light rejection that is not afforded by CW or time domain approaches. The ability to monitor IAC without the influence of DC component (ambient light) illustrates the frequency-filter feature that, unlike in CW or time-dependent methods, frequency domain methods (and other associated methods employing a periodic input function) inherently employ for ambient light rejection. Since only the modulated signal is detected, the phase and modulation ratios are minimally corrupted by an increase in ambient light that manifests itself as a DC contribution in the detected signal. Since frequency domain approaches offer the greatest opportunities for biomedical imaging and spectroscopy using fluorescent contrast agents, the following section provides the description of two types of measurements: (1) heterodyned point source and point detection, and (2) homodyned area illumination and detection. 2.5.1

Point Source Illumination and Heterodyned Point Detection

Figure 7 is a schematic of the single-point source and point detector configuration employed for fluorescence-enhanced spectroscopy and imaging. We refer to the single-point source and point detection when used in imaging studies as single-pixel measurements. The incident source is typically intensity modulated at frequency, cu, by a master oscillator and its output is collected into a 1-mm-diameter single-mode or multimode fiberoptic and directed onto the air-tissue or phantom interface. The excitation and emission light, L, which reaches the collection fiber located a distance p away is

462


FIGURE 7

Schematic of heterodyned FDPM system.

detected by a second photodetector which is gain modulated (G) at the same frequency plus a small offset, o> + Ao>, by a slave oscillator that is phase locked to the master oscillator. The modulated light reaching the collection fiber has a phase delay, 0; average intensity, Lnc; and an amplitude intensity, LAC.

L = LIX + LAC cos(cot + 0}

(5)

The gain of the photodetector has an average, GDC, and an amplitude, GAC, at modulation frequency slightly offset from the source: G = GIX + GAC cos[(w + Acu)t]

(6)

The output of the photodetector is a mixed heterodyned signal, S, containing all the amplitude, DC, and phase information of the optical signal collected by the detector. S = L X G = L 1K .G iX . + L DC G AC cos[(o> + Aw)t] r

+ G I X -LAC cos(wt + 0) + I

+

C^

AC

"

Ac:

f~*

AC AC 9

cos(2o)t + Awt +6)

cos(Awt - 6) (7)


463

The signal is passed through a frequency filter that removes all high-frequency components except those at the small offset frequency, Ao>: T

S => LDCGDC +

(~* AC

AC

cos(Acot -

6)

(8)

that can be collected by standard data acquisition approaches. A neutral density filter at the photodetector is used to control the amount of excitation light (the emission light is generally an insignificant portion of the detected light) when FDPM measurements of excitation light are made. A combination of holographic filters [12], and interference filters ensures passage of the small component of the emission light with minimized excitation light leakage when emission measurements are to be made. As discussed later, the rejection of excitation light is crucial to the success of fluorescence-enhanced optical imaging with multiply scattered light. As shown in the sections below, all fluorescence-enhanced contrast imaging studies that seek to perform tomographic reconstructions involve the point source and point detector geometry, necessitating a large number of measurements at the tissue or phantom interface. In an effort to increase the speed of conducting FDPM measurements without sacrificing accuracy, our group in the Photon Migration Laboratory adapted the homodyned ICCD systems using area illumination and detection. 2.5.2

Area Illumination and Homodyned Area Detection

Figure 8 illustrates the ICCD homodyne detection system consisting of three major components, including (1) a CCD camera, which houses a multipixel array of photosensitive detectors; (2) a gain-modulated image intensifier, which as described below facilitates measurements of FDPM; and (3) oscillators that sinusoidally modulate the laser diode light source and the image intensifier's photocathode gain at the same frequency, ox A 10-MHz reference signal between the oscillators ensures that they operate at the same frequency with a constant phase difference. Emitted light from the tissue or phantom surface is imaged via a lens onto the photocathode of the image intensifier. As before, the light (L) that reaches the photocathode of the image intensifier has a phase delay, #(r); average intensity, LDC(r); and amplitude intensity, LAC(r), which may vary as a function of position on the sample and consequently across the photocathode face. L(r) = LDC(r) + LAC(r)cos[o>t + 6»(r)]

(9)

The gain of the image intensifer has an average, GDC; a possible phase delay owing to the instrument response time, 0inst; and an amplitude, GAC, at the modulation frequency as the source:


464

CCD camera intensifier 105 mrn i

AF lens

laser diode

50 mm «/

AF lens \

introduce phase delay rjd FIGURE 8

Schematic of the ICCD homodyne FDPM system in the PML.

G = GDC + GAC cos(cot + 0 illst )

(10)

The modulated gain is accomplished by modulating the potential between the photocathode, which converts the NIR photons into electrons, and the multichannel plate (MCP), which multiplies the electrons before they are focused onto the phosphor screen (Fig. 9). The resulting signal at the phosphor screen is a mixed homodyne signal (S), containing all the amplitude, DC, and phase information of the optical signal collected by the detector. S(r) = L(r) X G = L IX -(r)G IX - + L nc (r)G AC cos[cot + 0 instr ] G lx -L At -(r)cos(wt + 0(r))

L AC (r)G

cos(0 inst - 0(r))

cos(2wt + 0ins( + 0(r))

(11)

Yet, since the phosphor screen has response times on the order of submilliseconds, it acts as a low-pass filter so that the image transferred to the CCD camera is simply: S(r) = L(r) X G = L,x.(r)G

L Ar((r)G AC ' cos(0insl - 0(r))

(12)

Near-Infrared Imaging Image Intensifier multichannel photocathode plate

465

phosphor screen

65V

1000V 4000V adjustable GBS Micro Power Supply

FIGURE 9 Schematic of the image intensifier circuit and system used in the homodyne ICCD system. (From Ref. 13.)

The time-invariant but phase-sensitive image on the phosphor screen is then imaged onto the CCD using either a lens or fiber coupling [13-16]. Rapid multipixel FDPM data acquisition proceeds as follows. The phase of the photocathode modulation is stepped, or delayed, at regular intervals between 0 and 360° relative to the phase of the laser diode modulation. At each phase delay r]d, the CCD camera acquires a phase-sensitive image for a given exposure time (Fig. 10), which is on the order of milliseconds. A computer program then arranges the phase-sensitive images in the order acquired and performs a fast Fourier transform (FFT) to calculate modulation amplitude, IAC, and phase, 6, at each CCD pixel (i, j) using the following relationships.

466


FIGURE 10 The process in which the phase delay between the image intensifier and laser diode modulation is adjusted between 0 and 360° yielding phase-sensitive yet constant intensity images at the phosphor screen. Upon compiling the intensities at each pixel, the sine wave is reconstructed and the phase and amplitude attenutation is obtained from simple fast Fourier transform. (From Ref. 13.)

[{IMAG[I(f m a J.,|} 2 /IMAG[I(f m , x ) M = arctan '

^ (13)

(14)

where I(f) is the Fourier transform of the phase sensitive intensity data, I(i7 d ); IMAG[I(f max )] and REAL[I(f m;ix )] symbolize the imaginary and real components in the digital frequency spectrum that best describe the sinusoidal data; and N relates the number of phase delays between the gain modulation of the image intensifier and the incident light source. Area illumination is accomplished simply by expanding a modulated laser diode beam onto the surface of the phantom or tissue to be imaged. To date, there has been no attempt to employ area illumination and area


467

detection for tomographic reconstructions. However, the ICCD camera can be employed to rapidly conduct single-pixel measurements for tomographic reconstructions by simply using the ICCD camera to simultaneously measure the phase and amplitude of light collected by a number of fibers whose ends are affixed onto an interfacing plate that is focused on the photocathode of the image intensifier via a lens (Fig. 11). Owing to the advent of new CCDs with enhanced red sensitivity and back-illuminated chips, there is no real SNR advantage to using an image intensifier unless frequency domain measurements are to be conducted. 3.

FLUORESCENT CONTRAST AGENTS FOR NIR IMAGING

Table 1 provides a chronological listing of studies reported in the literature over the past decade, which involve a number of different fluorescent contrast agents [5-10,17-49]. While the studies have progressed from using photodynamic agents, freely diffusable agents such as indocyanine green, fluorochromes conjugated to monoclonal antibodies and their fragments, small-peptide targeting agents similar to those employed in nuclear imaging, and, finally, activatable and "reporting agents," the translation of these agents to human clinical studies has been limited. Furthermore, investigations have been largely confined to superficial or subcutaneous tumors where the true advantages of NIR fluorescent agents—i.e., deep tissue penetration

FIGURE 11 Adaptation of a number of single fibers collecting detected light for imaging by the ICCD system (depicted in Fig. 9).


468 h-

^ Comme

c/3 'c

c/>

•f-'

c 0 O>

< +-> CO CO i_

c OJ o co o>

o

'>

^ _22

c

0 ^

^ i- O

cH£-

c o

4—'

ro 3 .E S

05 .y

-S T3 w ro

O5

c

CD

CD

03

-

g r?

CD .—

as •£ o a

o c •+= ° .•t±

.. 0

.1 1 E-8

JP c E 'o5 ^

-0 § |

4->

(J

"o 03 c c «

£

^ m C

CD ^ ^~ O5

1 ~^ =

o o co

oo 5 5t -S •^

iZ ro

c

"CD

d} ^_

.E

O 03

—

CD *tr

Q)

st

b en c CO

D

^ CD 05 CD 05 CD C 0

O

,

CD

*-• o O

)

2

'["".

"CD o

CN

. CO CD CO CO CD D 0 0 «

^

c

05

.i —

CD

1° < fc

03

co .t!

•- b ^

C

CO CD

-L

.2 o !c o

C CD ico Q- O > 03 3 CJ ^ ^r O C/5

3

£ °E °-

o ^ S. '> ^ ^

^

«p - 05

a; "~7 F--1

"Hi — '— Q_ *~

CD

. r-,

— O)

469

| e c

CD

o

03

O3 •o d 0)

o c

I'E ^ E

>-8 ?

-0

cxI

2 ° p 5 i —

CD^^

0

^ . ^ O ' o ' O

^1C

OHt

1 "E 1

^^D."cD

^-DoE

-i — OT

fT) Ji; .— . ,— ,

4

o O^

oO'~, 7'.Ticu;~; -—^^

-c*j(j>O3cD^:oo'^

^CD ^

^

r-

C

1

s

+ 3^ ^ i--no'^; 1

2

. J- ^r ôOoc ^Dc -n

.5? IS

SI

4-1 -—CD O2 i0 _ C1 T3

as Ji'"" LU

-

un

OT

, 0

• CM 0

(•

U3 CD CD

m

^

"

8 'C*-CDO

si »o O •- en o

O

CM £i O Z *^

CN > 0

CT>

,X

uo

-;OT03 £2, 3 03 p 'S o CM r-

'cF o "c ~-~ o y — O 3 £i < CD CM CO

0)0

Peptide-cyar conjugate, i bocyanine < tricarbocya gated to oc an analogu somatostati

o > '>

ICG, ICG-sm conjugates cybesin

Indocyanine

.y "3 •§ E

E

co

c

~E

c o

r^

CO

03 T3 T3 C

•f: CD QCD 03 £ Q. CD

T >^. JDO)

=5£ 0) E O

E i. ^

o

r~^ 5 1

^ .a ^j«:

;

03

ro

cc

CD

si

Targeting to mouse tu mor lines expressing somatostatin recepto

6 tn

Area illumination and detection using CCD camera

c 5 O — ® ,- .E ^ •- en

a0 is O

CO

"cj D

O3 "D

"o > £•-

^ C co E

'c

0

0

O3 0 C CO

IZ

D CN CM

°l= 15

**

6 , si 2 0

"CD

"o

m

CM

H-

g

rn

E =5£ E "o

CO

c

_

?;

uj Q. •":

I!!

a

CM ~

^26 .J-

O3 0 ^^ 'E 03 —

"o E

Q. _-

c

C3 >- 3 C 0 ? — 'C T3 JD CD CD t

"E

CO

'E0

_^ Q.

< ro "5? 3r .E "o £

a jc QL !§

o f-~

c

i.-

'•§_ i S Js IE ~ ™O o >.co « » | £ Q.

X Q.-| 0 > C. T3 0 O3 '35 U T3 C >• Q.

C

o E 'eo

cz ^

, >I = -o0 H! > o » "5 g g

,_ oL C/3 0 ? ^ to f^ o£ «

0

r- c

o

^

CN

D.

CD

C 31 I

1 O O 1^ «- A CO

0

E E

oc

C c 0 !|

0

Q-

O ^-•

cence rescen

U3 ^ c

£

Demonstrated the targeting Mab-dye compi could be used to provide immunophotodiagnostic information thereby guiding therapeutic intervention Demonstrated fluorescence imaging in vivo

co

•^ « ?to S £ «S 0

0

Photobleaching kinetics of ALA-induced protoporphyrin measured

Measure MMP activity i vivo for directing the therapeutic use of proteinase inhibitors

X

,

•~ 2

Imaging light-emitting

c o ~o

c

J2 O3 O) ~^

-5. E

~1 "o

CD

E CO

D) O

•- o " *5 *~"

0

"E

in CM *- '

I

O

•a o

^ Q

CO

CD IT

^ Q

CC

^

0) SI

£— ^ *^

"0 ° 03

CD i" O -C O Q-

c o CD

"U CD f~ CO

cc

c

£ L-^ CO O3 0 -^ 3 = ^3 5

c

0

2,

0 p I—i

5

S E S II *- ~
-dPhje-Cys-Pbe-dTrp-Lys-Thr-Cys-Thr-COO' R3=-dPhe-Met-Phfi-dTrp-Lys-Thr-Met-Thr-COOFIGURE 20 Chemical structure of indodicarbocyanine (IDCC; n = 2) and indotricarbocyanine (ITCC; n = 3) dyes and the peptide conjugates somatostatin-14 (R1), octreotate (R2), and (M 2 M 7 )octreotate (R3). (Adapted from Ref. 43).

cathespin B or H, lysosomal proteases whose opportunity for activity may be enhanced in cancer cells, the fluorochromes become free and radiatively relax to produce fluorescence. In contrast to the small-peptide conjugated dyes, this system requires fluorochrome internalization. The pioneering work enabled detection of 10 /nmol of agent or 250 pmol of fluorochrome administered per tumor-bearing animal and represented the first time an optical contrast agent based on an internalization construct had been demonstrated. Along the same lines, another agent that reported on the basis of protease

NH—Transferrin or Human serum albumin

(CH=CH)3—CH

\ (CH2)4SO3

(CH2)4SO3 Na+

FIGURE 21 Chemical structure of indotricarbocyanine (ITCC) dye-labeled with transferrin or human serum albumin. (Adapted from Ref. 34.)


I

PEG

NH2

MPEG

.ys—Ls—llys-Lys—llys— _ys—iLys-L NH

487

NH

\

Enzymatic Cleavage

Lysr

MPEG

\

HfN

MPEG -t-^-

FIGURE 22 Near-infrared fluorescent probe (NIRF) attached to molecular beacon (copolymer poly-L-lysine protected by methoxypolyethylene glycol, MPEG). Through enzymatic cleavage, driven by tumor enzymes, the fluorochrome peptide substrate is released. (Adapted from Ref. 33.)

activity was developed using the same design principles. Bremer et al. [44] coupled matrix metalloproteinase-2 (MMP-2) peptide substrates onto a polylysine polymer backbone and further conjugated Cy5.5 onto the peptides (Fig. 23). The fluorochromes were sufficiently packed to be quenched upon activation. Upon action of the proteinase on the peptide, the Cy5.5 was freed and able to radiatively relax, reporting proteinase activity. MMPs are overexpressed in cancers and MMP-2 in particular has been identified as being responsible for the collagen IV degradation that is the major component of basement membranes. The MMP-2 activity is thought to be responsible for the pathogensis of cancer, including spread, metastasis, and angiogenesis. Using area illumination and detection, as little as 167 pmol per animal resulted in detected fluorescence to measure MMP activity in vivo for directing the therapeutic use of proteinase activity. 3.7

Combined Targeting and Reporting Dyes

Finally, Licha and coworkers [59] sought to combine fluorochrome targeting using membrane receptors, such as transferin or presumably the somatostatin and bombesin receptors, with acid-cleavable constructs that would enable internalization of the fluorochromes in the lysosomal compartments and recycling of the receptors. Such constructs to augment the accumulation and therefore concentrate the signal from the targeting fluorochrome would only be enhanced if the contrast agent has a long half-life in the circulation. Coupling these cyanine dyes to different acid-cleavable hydrazone links that were bound to peptides, proteins, and antibodies, this group also sought to

05 0

E .s

II o >O 15

57 ^ 8 cr _> CD c & 0

a

1 o

t

)

1

Pi

z

T

>

s

2 >

•43 o

'w Q.

1

0 x~

^0 LU

a. 2

*U " , ;>. u

o t fe u^

zl

J=K

o[ 0>
) + Ai axt' { r)' ' i «x v

-i

i

x ~z—T 2 /• \ 2 V(r)' = 0

(20) V '

496


Again, to solve Eq. (20) for the complex emission fluence, O m (r, a>) = IAC(?, w)exp[ — i 0 m ( r , co)], the complex excitation fluence, O x (f, &>) = I A c(r, o>)exp[ —i$"(r, w)], must first be computed from the following: V - ( D x ( r ) V < £ x ( r , a;)) -

x u. l x (r) + — c 0>"(r, w) + S (r, a>) = 0

L "

J

(21)

where the source of the excitation light is represented by a sine wave or any other periodic function. Here again, the optical properties at the excitation and emission wavelengths should be expected to differ, and the same boundary conditions that are used for solution of the complex excitation fluence are again used for solution of the complex emission fluence. The complex emission fluence, ^'"(r, o>), is measured at the air-tissue interface in terms of a phase delay, Om, and amplitude attenuation, I AC , relative to the incident excitation source. The coupled diffusion Eqs. (20) and (21) have an analytical solution when absorption, scattering, and fluorescent properties at excitation and emission wavelengths are spatially constant in a homogeneous scattering medium [75,76]. Using frequency-domain photon migration measurements at both the excitation and emission wavelengths, the ability to measure the single exponential lifetimes of indocyanine green (ICG) and 3,3'-diethylthiatricarbocyanine iodide (DTTCI) [75], rhodamine B [76] and mixtures of ICG and DTTCI [69] in tissue-like scattering media of intralipid has been demonstrated experimentally. Effective contrasts for frequency domain approaches are also limited, as are time domain approaches, to fluorescent rather than phosphorescent or long-lived compounds. This was demonstrated in the Photon Migration Laboratory by comparing contrast offered by Tris(2,2'-bipyridyl)dichlororuthenium(II), Ru(bpy)i', with a lifetime of 600 ns and ICG with a lifetime of 0.56 nsec. Using a single target with 100 fold greater concentration than the background in a phantom (see Fig. 25 for measurement geometry of the phantom), the phase and amplitude modulation contrast at each of the detectors could be seen when ICG was used (Fig. 26a); whereas no contrast was measured when ruthenium dye, Ru(bpy)^ 1 ', was used as the contrast agent (Fig. 26b). These results confirm computational predictions that effective contrast agents must possess shorter lifetimes than the time of flight of photon propagation [77]. 5.

FLUORESCENCE-ENHANCED NIR OPTICAL TOMOGRAPHY

There are few studies that successfully invert NIR tissue optical measurements to render images of exogenously contrasted tissue volumes of clinical


497

source

point :detectors FIGURE 25 Schematic of the phantom tests to show the change in emission phase measurements as a fluorescent and phosphorescently tagged, 10-mm diameter target was moved from the periphery toward the center of a 100-mm diameter cylindrical vessel.

relevance. Nonetheless, several investigations have employed synthetic datasets and phantom studies as outlined in Table 2 [49,70,78-96]. Basically, the approaches are similar to those used in NIR optical tomography work with the exception of three points: (1) owing to the low quantum yield of fluorescent dyes, the SNR for CW, time domain and frequency domain measurements is inarguably lower, potentially making it more difficult to successfully reconstruct images; (2) owing to the fluorescence lifetime delay, both time domain and frequency domain approaches have additional contrast in the time-dependent photon migration characteristics; and, finally, (3) owing to the ability to directly invert the fluorescence kinetic parameters of fluorescence lifetime and quantum efficiency, the technique can be used to perform quantitative imaging via dyes that report cancer [97]. The latter characteristic of fluorescence lifetime imaging reveals its similarity to MRI. In MRI, imaging is accomplished by monitoring the radio frequency signal arising from the relaxation of a magnetic dipole perturbed from its aligned state using a pulsed magnetic field. In fluorescence contrastenhanced optical tomography, imaging is accomplished by monitoring the emission signal arising from the electronic relaxation from an optically activated state to its ground state. Unfortunately, the emission light is multiply scattered, hence the resolution afforded by MRI is unlikely to be matched by contrast-enhanced optical tomography. Unlike contrast agents for conventional imaging modalities, optical contrast owing to fluorescent agents may be imparted in two ways: (1) through increased target:background concentration ratios, and (2) through alteration in the fluorescence decay kinetics upon partitioning within tissue regions of interest [97].

498


(a)

20 0

12

(b)

12

FIGURE 26 The phase contrast (determined from the phase measured in the presence and absence of a target) measured at the emission light


5.1

499

Approaches to the Inverse Imaging Problem

Attempts to solve the fluorescence-enhanced optical imaging problem have been made both by solving a formal inverse problem and by taking a less rigorous model-based approach. For example, localization of a fluorescent target has been demonstrated using localization techniques in which the strength of reradiating target is used to ascertain its central position in a background containing no fluorophore. Considering FDPM measurements as the source and detector positions scanned over the phantom surface, the center of a single fluorescent target could be accurately identified [78]. In another frequency domain system employing one or more laser diodes modulated 90° out of phase to one another, interfering excitation photon density waves are generated within the scattering medium. When an interference plane set up by two sources out of phase with one another are scanned across the volume by altering the relative source strengths or by phase delays, a "null" plane is generated in which there is no AC component to the excitation wave. Outside the scanning null plane, a fluorescent target reradiates modulated intensity wave at the emission wavelength. When present in the "null" plane, there is no AC component of the generated fluorescence [78,98]. In yet another approach, an analytical solution to the spherical propagation of emission light in scattering media is used to determine the x,y,z position of the point source of fluorescence in an otherwise nonfluorescent background probed by CW measurements [70]. In another approach, Wu and coworkers [79,83] developed a time domain system for assessing the position of a fluorescent target in turbid media by evaluating the earlyarriving photons to determine the origin of the fluorescence generation. Hull and coworkers similarly used spatially resolved CW measurements to determine the location of the fluorescent target in scattering media [86]. Unlike these studies, in the presence of background signal or when fluorescent decay kinetics is used to report and provide contrast, the full inverse imaging problem must be solved. The solution of a formal inverse problem requires the use of the appropriate mathematical models described in Sec. 4. Specifically, a guess of

for a 100:1 target-to-background ratio for (a) phosphorescent dye with lifetime of 1 msec and (b) fluorescent dye of lifetime of 1 nsec. The phase contrast is predicted from simulation of the target moving from the perimeter (10 mm) toward the center (50 mm) of a 10 cm diameter cylinder under conditions of maximal phase contrast, i.e., OOT = 1, and uniform lifetime. The detectors are located around half of the perimeter of the cylinder as described in Fig. 25.


500 0 •~ . CO

CD

CD

CD

CD

CD

TO r-

Q. ^3 j_, O CD t CD

CD \x .!_, Q. ~)

Q_

C CD

-C

CD

o 3

CO

c o CD DC CD O) CD

E H-

0 CD

Q.O 3 O 4_ O O ' CD -rj

Q. T 3 O ^ CM CJ ^ CD CD

iH

c

^- ~^

CD

CD CD

C

o o

o o

*; > -C 05 Q

O 0 -Q CD °

CO CD C

2

C L O .

>

4_t

O

CD t CD Q-

«~ chi CM

o

Q. T 3 0

Q. 3

^ 0

o "~

O 03 ^t 03 D_

CD -n

"—

^-- C O CD

CD *

CL

O 03 t 03 CL

o ^

CD '—'

01 oo ^ i—

CD _£Z Q.

LU

LU

CO^

C CD CD

CO

£ 3

.y

05

f^

CD ;

.y

,

P

Inversion formulatior

•escence-E

C

a> ~p:

Random wal theory

to Q

o

•— CD

E o

'c

Differential

O C

CO "^

E o

E o c CD "a.

Laplace transform Differential (conjugate gradient) Differential (Newton's ii erative method) Localization

CD

CD

0

Differential (NewtonRaphson)

-Q

CD Q.

Localization

C

o o

Localization

H—'

'

S CD C/5 CD — CD Q. CD 3 |8

O^CN

cn N

§-5E

O) 0

C

Nl O)

" co '— • -rj

CO CO ^-

jj.

'

CD _,

CD

2 0

5§llM

o 2 ° .5> < ^ 2 c

2

C CD O

CD

CD

C

Q.

CD

^

O jC

CD

CD _C Q.

TD

Q. X LU

-CD

if

— DQ CD ~O

TO

(D TO D

LO 05

O —:

0 CM

"CD

o

ci d CJ 0)

CD -D -D CD Q. D O

o

CD TO

o

CD

1 2§

_§!§

-I l^ 0 "O

~n •— "^ .. O) ^

1 .j. a.

o

^ o 1

2rp c w ^ > 5

Q. ^ p CM LU

'

.y -s-

*-* it "5

__^

III CD

o

^; >

E

CO 0 CO ^ ^ -P C ^ CO CD

,

T3 TO 2 C ° ^ CD £ O

o

'E

'~C

^3 V CO

c

d.

i C — i '-P CD CD c £ Q. -

^ 3 ® CT

o

X LU

CO

*- D 00 C

c Q.

^— CD ^" T3

X LU

O

«"ô

£ o CD

•£ 2

i §5 §

CO — CD"

CM

£ o

1 CJ 01 ^

cn co O ° c ^ E co wo CD -o o^ ^ .E ..1 d

CQ ~ .E c "° 2 03 0 LO 'o "O 'CO LO

method) Marquardt and Tikhonov regularization Differential

> c

CO

o ._

^c Q- c 'c °- o

CO

Ntziachristos ei

o? 7j • t— CD O

CO

truncated Newton's method Integral (normalized born expansion)

°

LO

co 7^'| c c" co

Q __

°-~

~

CM

CJ =

1

!

CO JZ

Q

°

11

cn

Q CO

CJ

TO CO C

^

? CO E CO CO O5 Q CD CM

o £ .CD, CD O c Q. =

> Q -D CD O u_ H= C

uous-wave

,-

c

^

CD~

Q.

—•

c E

'5 oc

LO LO -X. ,', CO 0

h 1—

O

TO

*— d o

^ CD

£ *~

O

m (rs, r d ) ci> x (r s , ? i f

( h

_ ,

dw•
n i (r s , r d )/ O x (r s , r,)] exp . Here, x (r s , r,.) represents the excitation fluence detected at a fixed reference position, r,, in response to excitation from the source position, r s . Even though the excitation fluence, ^Trs, r,), is updated during the iteration, the Jacobian matrix can be directly calculated from Eq. (36), and


509

the change in <J> x (r s , rr) is small compared to that in <J> m (r s , r r ). As the source term of the emission diffusion equation is modified after each iteration [Eq. (22)], changes in the emission fluence are greater than that of the excitation fluence. The same consequence can be inferred from the integral equation [Eq. (36)]. Moreover, the phase of the emission fluence is greater than that of the excitation fluence and the normalized fluence, mAI>x, maintains a high phase contrast. Owing to noise and the ill condition of the Jacobian matrix and for inverting the systems of equations, updating can be accomplished using Newton's method [107] with Marquardt-Levenburg parameters A:

(37) Using excitation referencing at a single reference point, Lee and SevickMuraca [96] reconstructed an 8 X 4 X 8 cm3 phantom containing a 1 X 1 X 1 cm3 target with 100-fold greater ICG concentration, by using 8 excitation sources, twenty four detection fibers for collecting excitation light, and two reference detection fibers (one on either side of the reflectance and transillumination measurements) for collecting excitation light. Figure 28a is the original map containing two-dimensional slices of the three-dimentional geometry that demark the heterogeneity placement, while Fig. 28b is the three-dimensional reconstructed image. The results in Fig. 28 represent reconstructions based on emission FDPM measurements relative to excitation FDPM measurements at a fixed reference position: Ntziachristos and Weissleder [95] successfully reconstructed two fluorescent targets in a 2.5-cm diameter, 2.5-cm-long cylindrical vessel containing ICG and Cy5.5, and using CW emission measurements referenced to excitation measurements at each of the 36 detector fibers as a result of point excitation at 24 source fibers. The high density of measurements for reconstruction of the small simulated tissue volume is troublesome for validity of the diffusion equation used in the forward solver, but is similar to that demonstrated by Yang and coworkers [49] who reconstructed ICG and DTTCI in similarly sized phantoms and mice, presumably from absolute FDPM measurements at the emission wavelength alone. It is noteworthy that the studies of the reconstruction presented above assumed that the absorption and scattering properties were known a priori. However, using differential approaches coupled with Bayesian reconstruction approaches (see below), Eppstein and coworkers [108] demonstrated the insensitivity of reconstructions to changes in endogenous optical properties. Using a synthetic 256-cm3 volume containing 0.125-cm3 targets with 10:1 contrast in absorption owing to fluorophore and surrounded on four

510


FIGURE 28 The reconstruction of piaxf using the excitation wave as a reference using the integral approach and Marquardt-Levenburg reconstruction. The image was required after 27 iterations with regularization parameters for AC ratio (ACR), AAC = 1.0, and for relative phase shift (RPS), A,, = 0.02 (a) optical property maps of true juaxf distribution (b) and reconstructed /uaxf distribution. Peak values of /uaxf reached 0.1205 cm 1 (c) Iteration vs. SSE.


511

sides with 68 sources and 408 detection fibers, Eppstein showed that when the absorption cross-section at the excitation wavelength, /zaxi, varied as much as 90% and was unmodeled while the scattering coefficient, ;ttsxl, varied 10% or less and was also unmodeled, the impact on the reconstruction was minimal or negligible. Similar results have recently been shown by Roy and Sevick-Muraca [ 109] who show unmodeled variations in all endogenous optical properties by as much as 50%, which did not impact reconstructions when emission FDPM measurements were individually self-referenced to excitation FDPM measurements, as was done with the CW measurements of Ntziachristos and Weissleder [95]. While it appears promising that fluorescence-enhanced optical tomography can be accomplished without much a priori information regarding the endogenous optical properties, these results are nonetheless on synthetic studies and need to be conducted on actual tissues of substantive and clinically relevant volumes for validation. 5.3

Differential Formulation of the Inverse Problem

A second approach of the full inverse imaging problem may be the differential formulation, but in reality, this time it is rewritten for measurement Z(r d > r s ), whether absolute, or relative to a reference measurement at the emission or excitation wavelength, or self-referenced relative to the excitation wavelength at each detector position, rd. We term this approach the differential formulation because a small change in the predicted measurements is directly expressed in terms of a small change in the optical properties, AX, using a Jacobian matrix, J, d(AZj)/3Xj. Considering the number of detectors to be M; then the error function is defined as the sum of square of errors between the measured and calculated values at detector i = 1..M:

F(X) = We refer to each f; as a residual and the gradients of the error function with respect to the property, X: VF(X) = 2JTf(X) |-

(39) M

V2F(X) = 2 JTJ + 2 L

i= l

-j

fi(X)V 2 f,(X)

(40) J

Consider the Taylor's expansion of function F around a small perturbation of optical properties, AX: F(X + AX) = F(X) + VF(X)-AX + - AX r - V 2 F(X)- A(X)

(41)

512


which can be expressed as: F(X + AX) = F(X) + 2JTf(X) • AX JTJ +

+ 2 • AX

f,(X)V 2 f,(X) ' AX

(42)

and the function to be minimized, <J>(AX), can be explicitly written: 4>(AX) = F(X + AX) - F(X) = 2JTf(X) • AX + 2- AX T J T J +

L

t=i

ft(X)V2f}(X)

J

-AX

For first order Newton's methods, the term 2-AXT[E!l1 f i (X)V 2 f i (X)j • AX is neglected and the Gauss-Newton Method becomes one of minimizing, VO(AX) => 0 = JTJ • AX + J T f(X)

(44)

JTJ AX = -J T f(X)

(45)

The Levenberg-Marquardt method of optimization becomes [JTJ + A I ] - A X = -J T f(X)

(46)

The gradient based truncated Newton's method is based on retaining the second-order terms such that Eq. (43) becomes [89]: VO>(AX) => 0 = J T f(X) +

JTJ + L

fi(X)V 2 f,(X) • AX i= l

J

or, alternatively, VO(AX) => 0 = VF(X) + V2F(X) • AX

(48)

Typically, the first order Newton's methods are employed with the exception of the work by Roy [110]. In Newton's methods, it is assumed that A = J • AX and the solution is found using one of the several optimization approaches. The Jacobian can be computed either directly from the stiffness matrices of the finite element formulation or, simply but more computationally time consuming, from backward, forward, or central differencing approaches that compute the differences in the values of Z(r d , r s ) with small differences in the parameter to be updated, X(r s ). The Gauss-Newton and the Levenberg-Marquardt algorithms performed poorly in a large residual problem. Since the inverse is highly nonlinear and ill conditioned due to the error in measurement data, the residual at the solution will be large. It seems reasonable, therefore, to consider the truncated Newton method.


513

For the truncated Newton's method, the additional computational cost of computing the Hessian [associated with V2F(X)] is assisted by reverse automatic differentiation [89,110]. Using synthetic data, Roy has shown the feasibility for using the technique for three-dimensional reconstruction of lifetime, T, and absorption coefficient îaxf changes in frustrum and slab geometries from synthetic data containing noise that mimics experimental data [93].

5.4

Regularization and Other Approaches for Parameter Updating

In both the integral and differential formulations of the inverse problem, the tissue to be imaged must be mathematically discretized into a series of nodes or volume elements (voxels) in order to solve these inverse problems. The unknowns of the inverse problems then comprise the optical properties at each node or voxel. The final image resolution is naturally related to the density nodes or voxels. However, the dimensionality of the imaging problem is directly related to the number of nodes and can easily exceed 10,000 unknowns for a three-dimensional image. In a problem of this scale, the calculation of Jacobian matrices and matrix inversions involved in updating the optical property map are computationally intensive and contribute to the long computing times required to reconstruct the image. The instability arises because the measurement noise in the data or errors associated with the validity of the diffusion approximation can result in large errors in the reconstructed image. One of the greatest challenges associated with fluorescence-enhanced tomography is the propagation of error. In comparison with absorption imaging based on measurements of excitation light, fluorescence measurements have a reduced signal level and SNR. Lee and Sevick-Muraca [111] measured the SNR for single-pixel excitation and emission frequency domain measurements at 100 MHz to be 55 and 35 dB, respectively. In addition to the reduced signal, the noise floor of emission measurements can be expected to be elevated when excitation light leakage constitutes an increased proportion of the detected signal. Consequently, for emission tomography measurements, excitation light leakage is crucial and interference filters that attenuate excitation light four orders of magnitude (i.e., filters of optical density 4) may be clearly insufficient. Excitation light leakage will be a significant problem when emission measurements are conducted in tissue regions in which the target is absent and fluorescent contrast agents are not activated. Unfortunately, this type of error is not present in synthetic studies and is undoubtedly underestimated in the vast proportion of tomographic investigations to date.

514

5.4.1


Regularization

Regularization is a mathematical tool used to stabilize the solution of the inverse problem and to make it more tolerant to measurement error. Regularization approaches will play an important role in the development of suitable algorithms for actual clinical screening. For example, when discretized the differential and integral general formulations result in a set of linear Newton's equations generally denoted by AY = Z, where Y are the unknown optical properties and Z are the measurements. This system is commonly solved in the least-squares sense where the object function Q = ||AY — Z||2 + A Y 2 is minimized, where A is called the regular!zation parameter. Minimization of this function results in Y = (A1 A + AI) 'ATZ. The regularization parameter is generally chosen either arbitrarily or by a Levenberg-Marquardt algorithm so that the object function is minimized [112]. Thus, the choice of regularization parameter is through a priori information and adds another degree of freedom to the inverse problem solution. While this section is not meant to be a mathematical treatise of inverse algorithm and regularization approaches, we nonetheless point out that in a recent work by Pogue and coworkers [100], a physically based rationale for empirically choosing a spatially varying regularization parameter is presented to improve image reconstruction. Bayesian Regularization. Eppstein and coworkers [87,88,91,108] use actual measurement error statistics to govern the choice of varying regularization parameters in their Kalman filter implementation to optical tomography. In their work, they developed novel Bayesian reconstruction technique, called APPRIZE (Automatic Progressive Parameter-Reducing Inverse Zonation and Estimation), specifically for groundwater problems and adapted them to fluorescence-enhanced optical tomography [104-106]. Unique components of the APPRIZE method are an approximate extended Kalman filter (AEKF), which employs measurement error and parameter uncertainty to regularize the inversion and compensate for spatial variability in SNR, and a unique approach to stabilize and accelerate convergence called data-driven zonation (DDZ). Using the notation [AX, f(X)] as described in Sec. 5.3, here the Newton's solution is formulated as [91]: AX = [[JT(Q + R) 'J + P;X'] ' • JT(Q + R ) - ' l - f ( X )

(49)

where Q is the system noise covariance which describes the inherent model mismatch between the forward model (the diffusion equation) and the actual physics of the problem; R is the covariance of the measurement error that is actually acquired in the measurement set; and Pxx is the recursively updated error covariance of the parameters, X, being estimated from the measurement error, f(X). The use of this spatially and dynamically variant co-


515

variance matrix results in the minimization of the variance of the estimated parameters taking into account the measurement and system error. The novel Bayesian minimal variance reconstruction algorithm compensates for the spatial variability in signal to noise ratio that must be expected to occur in actual NIR contrast-enhanced diagnostic medical imaging. Figure 29 illustrates the image reconstruction of 256-cm3 tissue-mimicking phantoms containing none (case 3), one (case 1), or two (case 2)

FIGURE 29 Image reconstruction with APPRIZE, a) The initial homogeneous estimate discretized onto the 9 x 1 7 x 1 7 grid used for the initial inversion iteration, and shown with the true locations of the 3 heterogeneities and the 50 detectors (small dots), b) Case 1: The reconstructed absorption due to the middle fluorescing heterogeneity, interpolated onto the 17 x 33 x 33 grid used for prediction, and shown with the locations of the 4 sources used (open circles), c) Case 2: The reconstructed absorption due to the top and bottom fluorescing heterogeneities shown with the locations of the 8 sources used (open circles), d) Case 3: The reconstructed absorption of a homogeneous phantom shown with the locations of the 4 sources used (open circles). Although the phantoms and reconstructions were actually 8 cm in the vertical dimension, only the center 4 vertical cm is shown here. (From Ref. 91.)

516


1-cm3 heterogeneities with 50- to 100-fold greater concentration of ICG dye over background levels. The spatial parameter estimate of absorption owing to the dye was reconstructed from only 160 to 296 surface reference measurements of emission light at 830 nm (as described in Sec. 5.2.2) in response to incident 785-nm excitation light modulated at 100 MHz. Measurement error of acquired fluence at fluorescent emission wavelengths is shown to be highly variable. Another important feature of the Bayesian APPRIZE algorithm is the use of DZZ. With DDZ, spatially adjacent voxels with similarly updated estimates are identified through cluster analysis and merged into larger stochastic parameter "zones" via random field union [113]. Thus, as the iterative process proceeds, the number of unknown parameters, X, decreases dramatically, and the size, shape, value, and covariance of the different "parameter zones" are simultaneously determined in a data-driven fashion. Other approaches to reduce the dimensionality of the problems involve concurrent NIR optical imaging with MRI [10,114,115] and ultrasonography [116] to compartmentalize tissue volumes and to reduce the number of parameters to be recovered in the optical image reconstruction. 5.4.2

Simply Bounded Constrained Optimization

Imposing restrictions on the ill-posed problem can transform it to a wellposed problem as discussed above. Regularization is one method to reduce the ill posedness of the problem [117]. In the optical tomography problem, its solution, i.e., the optical properties of tissue, must satisfy certain constraints, and imposing these conditions in itself can regularize or stabilize the problem. Imposing these constraints explicitly restricts the solution sets and can restore uniqueness. Provencher and Vogel [117] have suggested two techniques: (1) prior

FIGURE 30 Three-dimensional reconstruction from simply bound truncated Newton's method. Actual distribution of fluorophore absorption coefficient of background tissue variability of endogenous (50%) and exogenous (500%) properties, (b) Reconstructed fluorophore absorption coefficient of background tissue variability of endogenous (50%) and exogenous (500%) properties using relative measurement of the emission fluence with respect to the excitation fluence at the same detector point, s = 0.0001. (c) Reconstructed fluorophore absorption coefficient of background tissue variability of endogenous (50%) and exogenous (500%) properties using relative measurement of the emission fluence with respect to the excitation fluence at the same detector point. (From Ref. 109.)


517

518


knowledge and (2) parsimony for well posedness of the problem. The first condition requires that all prior physical knowledge about the solution be included in the model. The second condition protects against the introduction of nonphysical phenomena. Tikhonov and Arsenin [118] also suggested that, to obtain a unique and stable solution from the data, supplementary information should be used so that the inverse problem becomes well posed. The basic principle of using a priori knowledge of the properties of the inverse problem is to restrict the space of possible solutions so that the data uniquely determine a stable solution. In his work, Roy showed that the constrained optimization technique, which places simple bounds on a physical parameter to be estimated, may be more appropriate for solving the fluorescence-enhanced optical tomography problem [90]. Specifically, a range of fluorescent optical properties is physically defined for the problem and the recovered parameter, X, must always be positive. Specifically, he demonstrated use of the bounding parameter, £, not only as a means to regularize and accelerate convergence but as a means to set the level of optical property contrast that is to be reconstructed using referenced emission measurements [109]. Here the possible values of parameter estimates are stated to lie between upper and lower bounds. In the first pass of the iterative solution, the optical property map is recovered and parameter estimates that lie within the upper and lower bounds plus and minus a small bounding parameter, s, are recovered and held constant for the next iteration. In this manner, the number of unknowns decrease with iteration. Indeed, the value of the bounding parameter can be used to set the resolution and the performance of the tomographic image. For example, if the bounding parameter is large, then the tomographic image will "filter" out artifacts not associated with the target, whereas if the bounding parameter is small, then the tomographic image may sensitively capture artifacts and heterogeneity that is not necessarily associated with the target. Figure 30 illustrates the reconstruction using the simply bounded truncated Newton's method, which shows that as the bounding parameter is increased, the recovered image becomes less sensitive to the background "noise." 6.

SUMMARY: THE CHALLENGES FOR NIR FLUORESCENCE-ENHANCED IMAGING AND TOMOGRAPHY

In the earlier sections, an overview of the status of fluorescence-enhanced optical imaging was presented. The opportunity to develop an emissionbased tomographic imaging modality similar to that provided by nuclear imaging but without the use of radionucleotides is offered by NIR fluorescent agents. Yet the added challenge for NIR fluorescence-enhanced im-


519

aging over nuclear imaging is that, unlike nuclear techniques, an activating or excitation signal must first be delivered to the contrast agent before there is registration of the emission signal from the tissue. Preliminary data from animals (Table 1) and phantoms (not presented herein) suggest that penetration depth and sensitivity may very well be comparable to those of nuclear techniques. A side-by-side comparison of NIR fluorescence-enhanced imaging with nuclear imaging is needed before the comparative performance can be ascertained. Another opportunity for optical imaging is the potential for tomographic reconstruction and additional diagnostic information based on the fluorescence decay kinetics of smartly designed probes. Tomography of large tissue-simulating volume has been demonstrated from experimental data as well as synthetic data as reviewed herein (Table 2), albeit with the somewhat inconvenient point source and point detector geometries. The single point source and detector geometry is a throw-back to NIR optical tomography from endogenous contrast studies and may not be the appropriate geometry for fluorescence-enhanced optical imaging, especially when transillumination through large tissues is required. Nonetheless, the tomographic algorithms as reviewed in Sec. 5, are already established for these systems. The challenge for the future is to develop tomographic algorithms for illumination and detection that are clinically feasible and adaptable for hybrid, nuclear imaging. Although the area of NIR fluorescence-enhanced optical imaging is less than a decade old, the developments are apt to continue for the coming decade, hopefully resulting in an adjuvant tomographic imaging modality for nuclear imaging. ACKNOWLEDGMENTS The review is supported in part by the National Institutes of Health grants R01 CA67176 and R01CA88082 (P. I. Eppstein, University of Vermont) and the Texas Advanced Technology Research Program. REFERENCES 1.

2.

3.

DJ Hawrysz, EM Sevick-Muraca. Developments toward diagnostic breast cancer imaging using near-infrared optical measurements and fluorescent contrast agents. Neoplasia 2(5):388-417, 2002. BJ Tromberg, N Shah, R Lanning, A Cerussi, J Espinoza, T Phain, L Svaasand, J Butler. Non-invasive in vivo characterization of breast tumors using photon migration spectroscopy. Neoplasia 2:26-40, 2000. E Reddi, A Segalla, G Jori, PK Kerrigan, PA Lidell, AL Moore, TA Moore,

520

4. 5.

6.

7.

8.

9.

10.

11. 12.

13.

14. 15.

16.

17.

18.

Sevick-Muraca et al. D Gust. Carotenoporphyrins as selective photodiagnostic agents for tumors. Br J Cancer 69:40-45, 1994. T Takemura, S Nakajima, I Sakata. Tumor-localizing fluorescent diagnostic agents without phototoxicity. Photochem Photobiol 59:366-370, 1994. R Biolo, G Jori, JC Kennedy, P Nadeau, R Potteir, E Reddi, G Weagle. A comparison of fluorescence methods used in the pharmacokinetic studies of Zn(II) phthalocyanine in mice. Photochem Photobiol 53:113-118, 1991. RC Straight, RE Benner, RW McClane, PMNY Go, G Yoon, JA Dixon. Application of charge-coupled device technology for measurement of laser light and fluorescence distribution in tumors for photodynamic therapy. Photochem Photobiol 53:787-796, 1991. M Kohl, U Sukowski, B Ebert, J Neukammer, HH Rinneberg. Imaging of superficially growing tumors by delayed observation of laser-induced fluorescence. Proc SPIE 1881:206-221, 1993. R Cubeddu, G Canti, P Taroni, G Valentini. Time-gated fluorescence imaging for the diagnosis of tumors in a murine model. Photochem Photobiol 57:480485, 1993. R Cubeddu, G Canti, A Pifferi, P Taroni, G Valentini. Fluorescence lifetime imaging of experimental tumors in hematoporphyrin derivative-sensitized mice. Photochem Photobiol 66:229-236, 1997. V Ntziachristos, AG Yodh, M Schnall, B Chance. Concurrent MRI and diffuse optical tomography of the breast after indocyanine green enhancement. Proc Natl Acad Sci 97:2767-2772, 2000. JR Lakowicz. Principles of Fluorescence Spectroscopy. Plenum Press, New York, 1983. JP Houston, AB Thompson, M Gurfinkel, EM Sevick-Muraca. Sensitivity and penetration depth of NIR fluorescence-enhanced contrast imaging. Photochem Photobiol 2002 (submitted 07/02). AB Thompson, EM Sevick-Muraca. NIR fluorescence contrast enhanced imaging with ICCD homodyne detection: Measurement precision and accuracy. J Biomed Opt in press (2003). J Lakowicz, K Berndt. Lifetime-sensitive fluorescence imaging using an if phase-camera. Rev Sci Instrum 62:1727-1734, 1991. EM Sevick, JR Lakowicz, H Szmacinski, K Nowaczyk, M Johnson. Frequency-domain imaging of obscure absorbers: principles and applications. J Photochem Photobiol 16:169-185, 1992. JS Reynolds, TL Troy, EM Sevick-Muraca. Multipixel techniques for frequency-domain photon migration imaging. Biotechnol Progr 13(5):669-680, 1997. A Pelegrin, S Folli, F Buchegger, J-P Mach, G Wagnieres, H van den Bergh. Antibody-fluorescein conjugates for photoimmunodiagnosis of human colon carcinoma in nude mice. Cancer 67:2529-2537, 1991. S Folli, G Wagnieres, A Pelegrin. JM Calmes, D Braichotte, F Buchegger, Y Chalandon, N Hardman, DC Heusser, JC Givel, G Chapuis, A Chatelain, H van Den Bergh, JP Mach. Immunophotodiagnosis of colon carcinomas in


521

patients injected with fluoresceinated chimeric antibodies against carcinoembryonic antigen. Proc Natl Acad Sci 89:7973-7977, 1992. 19. S Folli, P Westermann, D Braichotte, A Pelegrin, G Wagnieres, H van den Bergh, JP Mach. Antibody-indocyanin conjugates for immunophotodetection of human squamous cell carcinoma in nude mice. Cancer Res 54:2643-2649, 1994. 20. MM Haglund, DW Hochman. AM Spence, MS Berger. Enhanced optical imaging of rat gliomas and tumor margins. Neurosurgery 35:930-939, 1994. 21. S Mordon, JM Devoisselle, V Maunoury. In vivo pH measurement and imaging of tumor tissue using a pH-sensitive fluorescent probe (5,6-carboxyfluorescein): instrumental and experimental studies. Photochem Photobiol 60: 274-279, 1994. 22. B Ballou, GW Fisher, AS Waggoner, DL Parkas, JM Reiland, R Jaffe, RB Mujumdar, SR Mujumdar, TR Hakala. Tumor labeling in vivo using cyanineconjugated monoclonal antibodies. Cancer Immunol Immunother 41:257263, 1995. 23. JM Devoisselle, S Soulie, SR Mordon, G Mestres, TMD Desmettre, H Maillols. Effect of indocyanine green formulation on blood clearance and in vivo fluorescence kinetic profile of skin. Proc SPIE 2627:100-108, 1995. 24. I Rokahr, S Andersson-Engels, S Svanberg, M-A D'Hallewin, L Baert, I Wang, K Svanberg. Optical detection of human urinary bladder carcinoma utilising tissue autofluorescence and protoporphyrin IX-induced fluorescence following low-dose ALA instillation. Proc SPIE 2627:2-12, 1995. 25. MM Haglund, MS Berger, DW Hochman. Enhanced optical imaging of human gliomas and tumor margins. Neurosurgery 38:308-316, 1996. 26. K Sakatani, M Kashiwasake-Jibu, Y Taka, S Wang, H Zuo, K Yamamoto, K Shimizu. Noninvasive optical imaging of the subarachnoid space and cerebrospinal fluid pathways based on near-infrared fluorescence. Journal of Neurosurgery 87:738-745, 1997. 27. D Neri. Targeting by affinity-matured recombinant antibody fragments of an angiogenisis-associated fibronectin isoform. Nat Biotechnol 15:1271-1275, 1997. 28. B Ballou, GW Fisher, J-S Deng, TR Hakala, M Srivastava, DL Farkas. Fluorochrome-labeled antibodies in vivo: assessment of tumor imaging using Cy3, Cy5, Cy5.5, and Cy7. Cancer Det Prev 22(3):251-257, 1998. 29. C Eker, S Montan, E Jaramillo, K Koizumi, C Rubio, S Andersson-Engels, K Svanberg, S Svanberg, P Slezak. Clinical spectral characterization of coIonic mucosal lesions using autofluorescence and d aminolevulinic acid sensitizations. Gut 44:511-518, 1999. 30. JS Reynolds, TL Troy, RH Mayer, AB Thompson, DJ Waters, KK Cornell, PW Snyder, EM Sevick-Muraca. Imaging of spontaneous canine mammary tumors using fluorescent contrast agents. Photochem Photobiol 70:87-94, 1999. 31. A Becker, K Licha, M Kresse, B Riefke, U Sukowski, B Ebert, H Rinneberg, W Semmler. Transferrin-mediated tumor delivery of contrast media for optical imaging and magnetic resonance imaging. Proc SPIE 3600:142-150, 1999.

522


32.

R Weissleder, CH Tung, U Mahmood, A Bogdanov Jr. In vivo imaging of tumors with protease-activated near-infrared fluorescent probes. Nat Biotechnol 17:375-378, 1999. U Mahmood, C-H Tung. A Bogdanov, R Weissleder. Near-infrared optical imaging of protease activity for tumor detection. Radiology 21(3):866-870, 1999. A Becker, B Riefke, B Ebert, U Sukowski, H Rinneberg, W Semmler, K Licha. Macromolecular contrast agents for optical imaging of tumors: comparison of indotricarbocyanine-labeled human serum albumin and transferrin. Photochem Photobiol 72:234-241, 2000. M Gurfinkel, AB Thomson, W Ralston, TL Troy, AL Moore, TA Moore, JD Gust, D Tatman, JS Reynolds. B Muggenburg, K Nikula, R Pandey, RH Mayer, DJ Hawrysz, EM Sevick-Muraca. Pharmacokinetics of ICG and HPPH-car for the detection of normal and tumor tissue using fluorescence, near-infrared reflectance imaging: A case study. Photochem Photobiol 72:94102, 2000. K Licha, B Riefke, V Ntziachristos, A Becker, B Chance, W Semmler. Hydrophilic cyanine dyes as contrast agents for near-infrared tumor imaging: synthesis, photophysical properties and spectroscopic in vivo charactarization. Photochem Photobiol 72:392-398, 2000. M Yang, E Baranov, F Sun. X Li, L Li, S Hasegawa. M Bouvet, M AlTuwaijri, T Chishima, H Shimada, A Moossa, S Penman, R Hoffman. Wholebody optical imaging of green fluorescent protein-expressing tumors and metastases. PNAS 97:1206-1211, 2000. RM Hoffman. Visualization of GFP-expressing tumors and metastasis in vivo. Biotechniques 30:1016-1026, 2001. JE Bugaj, S Achilefu, RB Dorshow. R Rajagopalan. Novel fluorescent contrast agents for optical imaging of in vivo tumors based on a receptor-targeted dyepeptide conjugate platform. J Biomed Opt 6:122-133, 2001. S Achilefu. RB Dorshow, JE Bugaj, R Rajogopalan. Novel receptor-targeted fluorescent contrast agents for in vivo tumor imaging. Invest Radiol 35:479485, 2000. S Achilefu, JE Bugaj, RB Dorshow, HN Jimenez, R Rajagopalan. New approach to optical imaging of tumors. Proc SPIE 4259:110-114, 2001. S Achilefu, JE Bugaj, RB Dorshow, HN Jimenez, R Rajagopalan, RR Wilhelm, EG Webb, JL Erion. Site-specific tumor targeted fluorescent contrast agents. Proc SPIE 4156:69-78, 2001. A Becker, C Hessenius, K Licha, B Ebert, U Sukowski, W Semmler, B Wiedenmann C Grotzinger. Receptor-targeted optical imaging of tumors with near-infrared fluorescent ligands. Nat Biotechnol 19:327-331, 2001. C Bremen C Tung, R Weissleder. In vivo molecular target assessment of matrix metalloproteinase inhibition. Nat Med 7:743-748. 2001. B Ebert, U Sukowski, D Grosenick, H Wabnitz, KT Moesta, K Licha, A Becker, W Semmler, PM Schlag, H Rinneberg. Near-infrared fluorescent dyes for enhanced contrast in optical mammography: phantom experiments. J Biomed Opt 6(2): 134-140, 2001.

33.

34.

35.

36.

37.

38. 39.

40.

41. 42.

43.

44. 45.

Near-Infrared Imaging 46.

523

JC Finlay, DL Conover, EL Hull, TH Foster. Porphyrin bleaching and PDTinduced spectral changes are irradiance dependent in ALA-sensitized normal rat skin in vivo. Photochem Photobiol 73:54-63, 2001. 47. BW Rice, MD Cable, MB Nelson. In vivo imaging of light-emitting probes. J Biomed Opt 6(4):432-440, 2001. 48. NS Soukos, MR Hamblin, S Keel, RL Fabian, TF Deutsch, T Hasan. Epidermal growth factor receptor-targeted immunophotodiagnosis and photoimmunotherapy of oral precancer in vivo. Cancer Res 61:4490-4496, 2001. 49. Y Yang, N Iftimia, Y Xu, H Jiang. Frequency-domain fluorescent diffusion tomography of turbid media and in vivo tissues. Proc SPIE 4250:537-545, 2001. 50. CM Leevy, F Smith, J Longueville. Indocyanine green clearance as a test for hepatic function. Evaluation by dichromatic ear densitometry. JAMA 200: 236-240, 1967. 51. K Kogure, NJ David, U Yamanouchi, E Choromokos. Infrared absorption angiography of the fundus circulation. Arch Ophthalmol 83(2):209-214, 1970. 52. EM Sevick-Muraca, G Lopez, TL Troy, JS Reynolds, CL Hutchinson. Fluorescence and absorption contrast mechanisms for biomedical optical imaging using frequency-domain techniques. Photochem Photobiol 66:55-64, 1997. 53. ML Landsman, G Kwant, G Mook, WG Zijlstra. Light-absorbing properties, stability, and spectral stabilization of Indocyanine green. J Appl Physiol 40: 575-583, 1976. 54. S Mordon, JM Devoisselle, S Soulie-Begu, T Desmettre. Indocyanine green: physiochemical factors affecting its fluorescence in vivo. Microvas Res 55: 146-152, 1998. 55. KA Schafer, G Kelly, R Schrader, WC Griffith, BA Muggenburg, LA Tierney, JF Lechner, EB Janovitz, FF Hahn. A canine model of familial mammary gland neoplasia. Vet Pathol 35:168-177, 1998. 56. V Ntziachristos, X Ma, B Chance. Time-correlated single photon counting imager for simultaneous magnetic resonance and near-infrared mammography. Rev Sci Instrum 69:4221-4233, 1998. 57. KM McMaster, AE Giuliano, MI Ross, DS Reintgen, KK Hunt, VS Klimberg, PW Whitworth, LC Tafra, MJ Edwards. Sentinel lymph node biopsy for breast cancer—not yet the standard of care. N Engl J Med 339:990-995, 1998. 58. D Krag, D Weaver, T Ashikaga, F Moffat, S Klimberg, C Shriver, S Feldman, R Kusminsky, M Gadd, J Kuhn, S Harlow, P Beitsch, P Whitworth, R Foster Jr, K Dowlatshahi. The sentinel node in breast cancer—a multicenter validation study. N Engl J Med 339:941-946, 1998. 59. K Licha, A Becker, F Kratz, W Semmler. New contrast agents for optical imaging: acid-cleavable conjugates of cyanine dyes with biomolecules. Proc SPIE 3600:29-35, 1999. 60. MM Huber, AB Staubili, K Kustedjo, MHB Gray, J Shih, S Fraser, RE Jacobs, TJ Meade. Fluorescently detectable magnetic resonance imaging agents. Bioconj Chem 9:242-249, 1998. 61. X Li, B Chance, AG Yodh. Fluorescence heterogeneities in turbid media,

524

62.

63.

64.

65.

66.

67. 68. 69.

70.

71.

72.

73.

74. 75.

76.

77.

Sevick-Muraca et al. limits for detection, characterization, and comparison with absorption. Appl Opt 37:6833-6843, 1998. XD Zhu, S-P Wei, XW Guo. Imaging objects in tissue-like media with optical tagging and the diffuse photon differential transmittance. J Opt Soc Am 14: 300-305, 1997. RC Haskell, LO Svassand, T-T Tsay, T-C Feng, MS McAdams, BJ Tromberg. Boundary conditions for the diffusion equation in radiative transfer. J Opt Soc Am A 11:2727-2741, 1994. MS Patterson, B Chance, B Wilson. Time resolved reflectance and transmittance for the non-invasive measurement of tissue optical properties. Appl Opt 28:2331-2336, 1989. AH Hielscher, SL Jacques, L Wang, FK Tittel. The influence of boundary conditions on the accuracy of diffusion theory in time-resolved reflectance spectroscopy of biological tissues. Phys Med Biol 40:1957-1975, 1995. TJ Farrell, MS Patterson, B Wilson. A diffusion theory model of spatially resolved, steady-state diffuse reflectance for the noninvasive determination of tissue optical properties in vivo. Med Phys 9:879-888, 1992. EM Sevick-Muraca, CL Burch. Origin of phosphorescence re-emitted from tissues. Opt Lett 19:1928-1930, 1994. R Richards-Kortum, EM Sevick-Muraca. Quantitative optical spectroscopy for tissue diagnostics. Ann Rev Phys Chem 47:555-606, 1996. E Kuwana, EM Sevick-Muraca. Fluorescence lifetime spectroscopy in multiply scattering media with dyes exhibiting multi-exponential decays kinetics. Biophys J 83:1165-1176, 2002. V Chenomordik, D Hattery, I Gannot, AH Gandjbakhche. Inverse method 3-D reconstruction of localized in vivo fluorescence—application to Sjogren syndrome. IEEE J Sel Top Quant Electron 54:930-935, 1999. D Hattery, V Chernomordik, I Gannot, M Loew, AH Gandjbakhche. Fluorescence measurement of localized, deeply embedded physiological processes. Proc SPIE 3978:377-382, 2000. MS Patterson, BW Pogue. Mathematical model for time-resolved and frequency-domain fluorescence spectroscopy in biological tissue. Appl Opt 33: 1963-1964, 1994. CL Hutchinson, JR Lakov.'icz, EM Sevick-Muraca. Fluorescence lifetimebased sensing in tissues: a computational study. Biophys J 68:1574-1582, 1995. S Fantini, E Gratton. Fluorescence photon density waves in optically diffuse media. Optic Comm 173:73-79, 2000. RH Mayer, JS Reynolds, EM Sevick-Muraca. Measurement of fluorescence lifetime in scattering media using frequency-domain photon migration. Appl Opt 38:4930-4938, 1999. AE Cerussi, JS Maier, S Fantini, MA Franceschini, WW Mantulin, E Gratton. Experimental verification of a theory for time-resolved fluorescence spectroscopy of thick tissues. Appl Opt 36:116-124, 1997. A Chen. Effects of fluorescence and phosphorescence lifetime on frequency


78.

79.

80. 81. 82.

83.

84.

85. 86.

87. 88.

89.

90. 91.

92.

93.

525

domain optical contrast for biomedical optical imaging. MS thesis, Purdue University, West Lafayette, IN, 1997. MA O'Leary, DA Boas, B Chance, AG Yodh. Reradiation and imaging of diffuse photon density waves using fluorescent inhomogeneities. J Lumin 60, 61:281-286, 1994. J Wu, Y Wang, L Perleman, 1 Itzkan, RR Dasari, MS Feld. Time-resolved multichannel imaging of fluorescent objects embedded in turbid media. Opt Lett 20:489-491, 1995. J Chang, RL Barbour, H Graber, R Aronson. Fluorescence Optical Tomography. Proc SPIE 2570:59-72, 1995. MA O'Leary, DA Boas, XD Li, B Chance, AG Yodh. Fluorescence lifetime imaging in turbid media. Opt Lett 21(2):158-160, 1996. DY Paithankar, AU Chen, BW Pogue, MS Patterson, EM Sevick-Muraca. Imaging of fluorescent yield and lifetime from multiply scattered light reemitted from random media. Appl Opt 36:2260-2272, 1997. J Wu, L Perelman, RR Dasari, MS Feld. Fluorescence tomographic imaging in turbid media using early-arriving photons and Laplace transforms. Proc Natl Acad Sci 94:8783-8788, 1997. J Chang, HL Graber, RL Barbour. Improved reconstruction algorithm for luminescence when background luminophore is present. Appl Opt 37:35473552, 1998. H Jiang. Frequency-domain fluorescent diffusion tomography: a finite-element-based algorithm and simulations. Appl Opt 37(22):5337-5343, 1998. EL Hull, MG Nichols, TH Foster. Localization of luminescent inhomogeneities in turbid media with spatially resolved measurements of cw diffuse luminescence emittance. Appl Opt 37:2755-2765, 1998. MJ Eppstein, DE Dougherty, DJ Hawrysz, EM Sevick-Muraca. Three-dimensional optical tomography. Proc SPIE 3497:97-105, 1999. MJ Eppstein, DE Dougherty, TL Troy, EM Sevick-Muraca. Biomedical optical tomography using dynamic parameterization and Bayesian conditioning on photon migration measurements. Appl Opt 38:2138-2150, 1999. R Roy, EM Sevick-Muraca. Truncated Newton's optimization scheme for absorption and fluorescence optical tomography: Part II Reconstruction from synthetic measurements. Opt Exp 4:372-382, 1999. R Roy, EM Sevick-Muraca. Active constrained truncated Newton method for simple-bound optical tomography. J Opt Soc Am A 17(9): 1627-1641, 2000. MJ Eppstein, DJ Hawrysz, A Godavarty, EM Sevick-Muraca. Three-dimensional, near-infrared fluorescence tomography with Bayesian methodologies for image reconstruction from sparse and noisy data sets. Proc Natl Acad Sci 99:9619-9624, 2002. DJ Hawrysz, MJ Eppstein, J Lee, EM Sevick-Muraca. Error consideration in contrast-enhanced three-dimensional optical tomography. Opt Lett 26(10): 704-706, 2001 R Roy, EM Sevick-Muraca. Three-dimensional unconstrained and constrained image-reconstruction techniques applied to fluorescence, frequency-domain photon migration. Appl Opt 40 (13):2206-2215, 2001.

526 94.


R Roy, EM Sevick-Muraca. A numerical study of gradient-based nonlinear optimization methods for contrast-enhanced optical tomography. Opt Exp 9(l):49-65, 2001. 95. V Ntziachristos, R Weissleder. Experimental three-dimensional fluorescence reconstruction of diffuse media by use of a normalized Born approximation. Opt Lett 26(12):893-895, 2001. 96. J Lee, EM Sevick-Muraca. 3-D fluorescence enhanced optical tomography using references frequency-domain photon migration measurements at emission and excitation measurements. J Opt Soc Am A 19:759-771, 2002. 97. EM Sevick-Muraca, DY Paithankar. Fluorescence imaging system and measurement. U.S. patent #5.865,754, issued 2/2/99. 98. JM Schmitt, A Knuttel, JR Knutson. Interference of diffusive light waves. J Opt Soc Amer A 9:1832-1843, 1992. 99. DY Paithankar, A Chen, EM Sevick-Muraca. Fluorescence yield and lifetime imaging in tissues and other scattering media. Proc Soc Photo-Opt lustrum Eng 2679:162-175. 1996. 100. B Pogue, T McBride, J Prewitt. U Osterberg, K Paulsen. Spatially varying regulari/ation improves diffuse optical tomography. Appl Opt 38:2950-2961. 1999. 101. MJ Holboke, AG Yodh. Parallel three-dimensional diffuse optical tomography. Biomedical Topical Meetings, OS A, Miami Beach, FL, 2000, pp 177-179. 102. M Xu, M Lax, RR Alfano. Time-resolved fourier diffuse optical tomography. Biomedical Topical Meetings, OSA, Miami Beach, FL, 2000, pp 345-347. 103. SR Arridge, JC Hebden. M Schweiger, FEW Schmidt, ME Fry, EMC Hillman, H Dehghani, DT Delpy. A method for three-dimensional time-resolved optical tomography. Int J Imaging Sys Technol 11:2-11, 2000. 104. MJ Eppstein, DE Dougherty. Optimal 3-D traveltime tomography. Geophysics 63:1053-1061. 1998. 105. MJ Eppstein, DE Dougherty. Efficient three-dimensional data inversion: soil characterization and moisture monitoring from cross-well ground penetrating radar at a Vermont test site. Wat Resources Res 34:1889-1900. 1998. 106. MJ Eppstein. DE Dougherty. Three-dimensional stochastic tomography with upscaling. U.S. patent application 09/110,506, (9 July 1998). 107. TJ Yorkey, JG Webster, WJ Tompkins. Comparing reconstruction algorithms for electrical impedance tomography. IEEE Trans Biomed Eng BME-34, 843— 852, 1987. 108. MJ Eppstein, DE Dougherty, DJ Hawrysz, EM Sevick-Muraca. Three-dimensional Bayesian optical image reconstruction with domain decomposition. IEEE Trans Med Imaging 20(3): 147-163, 2000. R Roy. A Godavarty, EM Sevick-Muraca. Fluorescence-enhanced, optical tomography using referenced measurements of heterogeneous media. IEEE Trans Med Imaging (in press). R Roy. EM Sevick-Muraca. Truncated Newton's optimization scheme for absorption and fluorescence optical tomography: part I theory and formulation. Opt Exp 4:353-371, 1999. J Lee. EM Sevick-Muraca. Fluorescence-enhanced absorption imaging using


112. 113. 114.

115.

116.

117.

118.

527

frequency-domain photon migration: tolerance to measurement error. J Biomed Opt 6(l):58-67, 2000. SR Arridge. Optical tomography in medical imaging. Inverse Problems 15: R41-R93, 1999. MJ Eppstein, DE Dougherty. Simultaneous estimation of transmittivity values and zonation. Wat Resour Res 32:3321-3336, 1996. Y Pei, F-B Lin, RL Barbour. Modeling of sensitivity and resolution to an included object in homogeneous scattering media and in MRI-derived breast maps. J Biomed Opt 5:302-219, 1999. BW Pogue, KD Paulsen. High-resolution near-infrared tomographic imaging simulations of the rat cranium by use of a priori magnetic resonance imaging structural information. Opt Lett 23:1716-1718, 1998. MJ Holboke, BJ Tromberg, X Li, N Shah, J Fishkin, D Kidney, J Butler, B Chance, AG Yodh. Three-dimensional diffuse optical mammography with ultrasound localization in a human subject. J Biomed Opt 5:237-247, 2000. SW Provencher, RH Vogel. Regularization techniques for inverse problems in molecular biology. Numerical Treatment of Inverse Problems in Differential and Integral Equations. Boston: Birkhauser Press, 1983. AN Tikhonov, VY Arsenin. Solution of Ill-Posed Problems. Washington, DC: V. H. Winston & Sons, 1977.

15 Fluorescence in Photodynamic Therapy Dosimetry Brian C. Wilson Ontario Cancer Institute, University of Toronto, and Photonics Research Ontario, Toronto, Ontario, Canada Robert A. Weersink Photonics Research Ontario, Toronto, Ontario, Canada Lothar Lilge University of Toronto and Photonics Research Ontario, Toronto, Ontario, Canada

1.

INTRODUCTION

Photodynamic therapy (PDT) is a technique for treating a variety of malignant and nonmalignant conditions based on the use of light-activated drugs (photosensitizers). Typically, the photosensitizer is administered either systemically (intravenously or orally) or topically to the tissue to be treated. After allowing time for uptake of the photosensitizer to the target tissues or tissue structures, light of an appropriate wavelength to activate the drug is applied. This results in the photoproduction of one or more cytotoxic agents, leading to the intended cellular and tissue effects. For most photosensitizers used or under investigation clinically, it is likely that the main photophysical pathway is production of singlet oxygen, 'O2. Singlet oxygen is an excited form of oxygen that is highly reactive with biomolecules, leading typically 529

530

Wilson et al.

to oxidative damage. The Jablonski energy diagram for this so-called type II process is shown in Fig. 1. The absorption of a photon by the groundstate photosensitizer activates the molecule to the excited singlet state. This short-lived state (typically nanoseconds) may de-excite to the ground state either nonradiatively or by fluorescence emission, or may undergo a change of spin to the triplet state. This is relatively long lived, since decay to the ground state is a quantum mechanically forbidden transition. Energy exchange with ground-state oxygen is, however, an allowed transition, since 3 O2 is also a triplet state, and this excites the oxygen to 'O2. (Note that type I processes are those in which the reactive species are generated directly from the photosensitizer-excited singlet or triplet state and may or may not be oxygen dependent. They also lead to photosensitizer fluorescence from the singlet state.) Most of the current photosensitizers have some fluorescence emission. (The in vivo measurement of nonfluorescent drugs is discussed briefly later.) For the PDT efficiency to be as high as possible, a high triplet state quantum yield is desirable, which competes with the fluorescence quantum yield. However, using the fluorescence emission to monitor the photosensitizer does not require a high yield (e.g., a few percent), so that in practice a high triplet-state yield can be selected. Figure 2 shows the absorption and fluorescence emission spectra of some common PDT drugs. Note that the fluorescence excitation spectrum is generally very similar to the absorption spectrum, but that these may both change in biological media compared to the spectra in simple solutions, due to substrate (e.g., protein) binding. Shifts in the absorption peaks by several nanometers are common.

SINGLET STATES

"

BIOLOGICAL »-"-• ^

EFFECT

GROUND STATE

OXYGEN

FIGURE 1 Jablonski diagram for type II photodynamic reactions. The vertical axis indicates the energy for the distinct electronic states of the photosensitizer and oxygen molecules.

Photodynamic Therapy Dosimetry

531

O CD

•j=> Q) X O LU Q

400

600

500 600 Wavelength (nm)

700

800

(a)

FIGURE 2 (a) Absorption and fluorescence emission spectra of some common PDT photosensitizers. (b) Corresponding excitation-emission matrices.

Most PDT photosensitizers have spectra across the visible range, often with a strong Soret band in the UV-A/blue region. In order to obtain maximal light penetration in tissue for treating larger lesions (such as solid tumors), many newer PDT drugs have strong absorption bands in the far-red (650700 nm) and near-infrared (700-850 nm), and correspondingly their fluorescence is also at these long wavelengths. Figure 3 summarizes the classes of disease for which PDT is being investigated as a possible therapeutic method. Several of these treatments have regulatory approvals and are entering routine clinical practice. The potential applications are wide, reflecting the facts that (1) PDT can be applied to many body sites through the use of fiberoptic light delivery, as summarized schematically in Fig. 4, and (2) with different photosensitizers the mechanisms of action are varied, and include direct cell kill (by necrosis

Wilson et al.

532

Photofrin E •S c g ro

•3x ill

o in. cs

300

400 500 600 Emission (nm)

700

AIPcS4

o 10 c ,2 o

5 8

'Q x Hi

i 300

400 500 600 Emission (nm)

700

.m

(b) FIGURE 2

Continued

and/or apoptosis), vascular destruction, and immune stimulation or suppression. The photosensitizer fluorescence may be utilized in a number of ways, including: 1. 2. 3.

Lesion detection in vivo Lesion localization in vivo for therapy guidance, both PDT and surgical Quantification of the concentration of the photosensitizer in the target and other tissues in vivo for the purposes of a. Determining the drug pharmacokinetics in order to select the optimum time for light activation, or

Photodynamic Therapy Dosimetry In Vivo

photosensitizer

533

treatment light • solid tumors • dyslasias • papillomas • rheumatoid arthritis • age-related macular degeneration • cosmesis • psoriasis • endometrial ablation • localized infection • prophylaxis of arterial restenosis

Ex Vivo

• extracorporeal photophoresis • blood purging: HIV, hepatitis B, protozoa • bone marrow purging photosensitizer

I

FIGURE 3

treatment

Potential clinical applications of PDT.

b.

4.

5. 6.

Individualizing the light delivery to produce the required biological response ("dosimetry") Measurement of the photosensitizer photobleaching during PDT light activation, in order to estimate the effective PDT "dose" delivered Photosensitizer quantification in tissue samples (e.g., biopsies) ex vivo Determining the microdistribution of photosensitizer in tissues (ex vivo) or cells (in vitro) by fluorescence microscopy

Tissue autofluorescence may be used in addition or as an alternative to photosensitizer fluorescence for purposes 1, 2, and 4, or as a monitor of tissue response to treatment. Fluorescence imaging of other "contrast agents" may be of value for PDT in particular sites, such as the use of fluorescein angiography in PDT of age-related macular degeneration, a dis-

Wilson et al.

534

JLJJ

Superficial

Interstitial

Endoscopic

Whole Surface Intracavitory

FIGURE 4 Light irradiation methods for PDT, showing the wide variety of target geometries.

ease that leads to blindness due to growth of abnormal blood vessels in the choroid layer of the eye near the macula, which is the area of central acute vision. In this case, as illustrated in Fig. 5, fluorescein angiography is used both before treatment to show the area of choroidal neovasculature so as to target the treatment light and after treatment to evaluate the area and completeness of vascular closure. In this chapter, the emphasis will be on topics 2-6, since lesion detection and localization for treatment guidance are dealt with elsewhere, in the general context of fluorescence "contrast agents." It is worth noting, however, that many of the in vivo fluorescence-based methods and instruments that have been developed over the past 20 years have been strongly


535

FIGURE 5 Fluorescein angiograms before and after PDT for age-related macular degeneration, showing closure of the area of abnormal neovasculature.

associated with the corresponding development of PDT as a treatment modality, since they provide complementary "search and destroy" capabilities. Thus, for example, autofluorescence bronchoscopy (Chapter 11) grew from the original endoscopic spectroscopy and imaging work with the photosensitizer hematoporphyrin derivative (HPD) and was originally aimed at localizing tumors in the lung for PDT treatment. (Interestingly, HPD itself was developed from hematoporphyrin in an attempt to improve its fluorescence properties and was only subsequently discovered to be an effective PDT drug.) Studies were performed with successively decreasing doses of HPD in order to reduce the associated skin photosensitization, and in the case of early bronchial cancer/dysplasia it was found that the detection accuracy was greatest with zero dose, i.e., using only the autofluorescence. However, fluorescence of HPD continues for the other objectives listed above, and many developments in fluorescence spectroscopy and imaging in vivo continue to be tightly linked to improving and monitoring PDT treatments, although they are also developing independently. In addition to PDT applications exploiting the photosensitizer fluorescence, fluorescence-based optical fiber probes are being developed for light fluence monitoring in tissue. In these, a small point-like volume of specific fluorophore is incorporated into an optical fiber and placed on or within the tissue. The PDT treatment light then excites the fluorescence, a fixed fraction of which is transmitted along the fiber and detected. Probes incorporating several different fluorophores at different positions along the fiber allow

536

Wilson et al.

multipoint measurements to be made simultaneously by spectral deconvolution. The concept can also be extended to tissue oxygen measurements using phosphorescence lifetime measurements. This technology and its applications will be discussed at the end of the chapter. 2. 2.1

PHOTOSENSITIZER QUANTIFICATION IN VIVO Relative Measurements

Figure 6 shows a simplified schematic diagram of noninvasive measurement of photosensitzer fluorescence in vivo. The excitation light from a light source (laser, or filtered lamp or light-emitting diode) at wavelength Aex is delivered to the tissue surface by an optical fiber (or fiber bundle). Then one or more collection fibers pick up some fraction of the light remitted through the tissue surface; this comprises elastically scattered light (diffuse reflectance), photosensitizer fluorescence (at wavelength A cm ), tissue autofluorescence, and inelastic (Raman) scattered light. Typically, the order of magnitude of total photons remitted for each of these components, relative to the incident light, is around 10 ', 10~ 4 , 10 \ and 10 7, respectively, a small fraction of which is collected by the detector fibers. The elastic scattering is removed by optical filtering, typically, placing a long-pass filter (>Aex) between the output end of the collection fibers and the photodetector. The Raman signal is usually negligible, whereas the photosensitizer fluorescence is observed superimposed on the tissue autofluorescence background. Figure 7 shows such fluorescence spectra, for different administered doses of a specific agent, in this case the prodrug aminolevulinic acid (ALA), m

source

f photodetector i

xpPH;!H tissue

FIGURE 6 Schematic of fiberoptic detection of fluorescence at a tissue surface.


537

0.005

0.000

?„ (nm)

0.05

0-00

tQmg/kg@3hf

'

'

02

0.0 500 —-Barrett's

"-""I

(nm)

30mg/kg @ 6 Nr

600

700 Squarnous

FIGURE 7 In vivo fluorescence spectra in Barrett's and normal human esophagus, at different times post ALA administration and for different ALA doses. Individual spectra represent measurements at different points on the tissue and are normalized to the autofluorescence background at 500 nm. (See color insert.)

which induces increased biosynthesis of heme, the penultimate step of which is the fluorescent photosenisitizer protoporphyrin IX (PpIX). These data illustrate a number of points. First, the fluorescence intensity is highly variable from patient to patient and even point to point in the same patient for a given drug level and time point. This type of variability has been reported for other photosensitizers and is, of course, a major reason why quantitative measurements are required. Second, the drug fluorescence spectrum is known and can be easily identified on the autofluorescence background, at least for higher levels of drug; however, at low levels, the added drug spectrum may not be easily seen, especially if the autofluorescence itself is highly variable (which is certainly the case for diseased tissue, and indeed is the basis of autofluorescence-based diagnostics). Third, the increase in the PpIX fluorescence with ALA dose and the dependence on time after administration can be observed, at least in the average spectra of each tissue type. In order to obtain a quantitative estimate of the PpIX concentration, a fit may be made to the autofluorescence background: in the simplest case, by a linear interpolation between points just outside the main PpIX fluores-

Wilson et al.

538

cence peak, as illustrated in Fig. 8a, or by a polynomial fit, taking the (local) shape of the background into account (Fig. 8b). The residual photosensitizer signal, PSF*, is then proportional to the (average) tissue concentration, C ps . If all factors other than C—namely, the excitation light, excitation and detection geometry, fluorescence quantum yield, and tissue optical absorption and scattering coefficients at Aex and A ox —remain constant, then the FSF* values yield the relative drug concentrations. The assumption of these factors remaining constant usually holds best in following the photosensitizer pharmacokinetics in a given location in an individual patient. Even in this idealized case, however, complications arise. For example, photosensitizer aggregation/disaggregation depends in many cases on the local microenvironment (pH, polarity, etc.) and this alters the fluorescence yield. The yield can also change when the photosensitizer binds

550

600

650

700

750

Wavelength (run)

b

> CO

O

o C 300 c/? 0

c 200J CD O

b 100 2

(c) FIGURE 9

4

Concentration by Tissue Extraction (mg/kg) Continued

solution to that in an optically turbid medium. PLS is one of a number of multivariant statistical techniques that are used widely in spectroscopic analysis, and here is considered as a "black box" calibration technique. It requires simulated or measured training data sets (here, fluorescence spectra) for a range of known C values. This set must encompass the range of possible variations in the spectra and expected photosensitizer concentrations. These two methods were compared in phantoms, where the prediction error was —5% for PLS and — 1 8 % for the KM model. To achieve this low prediction error in PLS, the input data set consisted of the combination of fluorescence spectra collected at two excitation wavelengths. The limitation of PLS is that building of an adequate training set requires an extensive set of measurements that encompasses all possible C values across the range of tissue types. This must be repeated for each photosensitizer. Furthermore, a wide spectral measurement range must be used to provide an adequate spectral "signature" in the tissues. Applying it to layered tissues or inhomoge-


543

neous drug distributions may also present problems. Finally, although it can be very accurate, it provides no physical insight, so that it is difficult to assess what would be the effect of altering any of the measurement conditions. 2.4

Model-Dependent Techniques

To avoid the need for tissue-dependent calibration, assumptions can be made about the fluorescence escape function and excitation fluence distribution, based on a priori knowledge of the tissue optical properties. One approach is to collapse the escape function into 1-D by employing a broad-beam excitation with point collection and assume a single exponential dependence on depth. Another simplification is to assume that the optical properties at Aex and Aem are the same. Thus, Potter and Mang [3] postulated that the fluorescence signal was proportional to C*§2, where 8 is the penetration depth of light in the tissue at Aex = 630 nm. Separate measurements of 8 using diffuse reflectance were required using a second optical fiber probe. This was tested in vivo in an animal model with Photofrin. However, correcting the fluorescence signal by 82 still resulted in errors in the prediction of C. Profio et al. [4] used similar approximations but derived a relationship between the ratio of the fluorescence signal and the quantity (1 — y)/(l + y), where y is the diffuse reflectance. At Acx = 633 nm in tissue-simulating phantoms containing Photofrin, the prediction error for C was —20%, for a range of background absorption coefficients tissue/Ji

>

c

CO

2

CD T3 O

D

CD T3 O

4-1

a?

w

^

CD

~ •«=

CD

4-J

X

CD

Q) C

4-J

O SI Q.

O

SI

CL T3 C CO

CD O C CD O

> a

w

CO

0)

L.

111!

o

CD

.c

_D

I-

LL

o E

O

>-> C/) "'

—

CO

c Q_ if

"a o •M

LL

5

.Q D CO CD > "4-"

O CO

O O

4-1

S^ t! C

Illli

^

1 8 CD

£

co

2

II Hill!

si

Q_

H—

O V) CD

t CD Q. O CO

it

o

"co CD T3 — i— CD

CD

'55

o

CO

t a5

° °

Ô < sI- O

2 Ql

x

LLI

CO

co

Q-

O 4-
T3 °

CL c/5

33 +-< 72. '*j '72 •*-> a aa 000 a aa

CA

0 0 CO 13 Ô Q. '^_, '^_i • • Q. Q. CA

co Q. CL >

O

o

f

c

0

0

0

CA .^

C35 O) O) Q. CO CO CO CO •—O i_ i_ i_ CT c

+

1

1 +

1

1

+

+

4-

CO

c in Lf) CO

« ~O if,

co

|g

CM

U 0

<M

CO 00 ^" ^t CQ LU

'-P ^

' c n n 3

ccÊ^ÎJE^cE^cEÊ^îeE <EScc^^2E«:EDCQ:;

**""~

•—

w

C D C D C D C D C D C D ( D C D C D C D 0 C D OOC/3C/3C/3C/5C/)C/)C/)C/)C/5C/)C/) 3 3 3 3 3 3 3 3 3 3 3 3

^

Ê

LO

LO LT)

LO LO LO LO LO LO CO

LO

r--ococoLOcocr>oooococorô

OLOCMLOOOOOO

O O ^ O O O O O O O O O O L O 00

L O O O O L O L O r - ^ - ^

c

o

Q.

E o

CJ CD .>

E ® GO

4-J

^

O 03

C/) ^

o

o

S>

^~ 6)

CO Q_

O5 LU

O Q

o ^

Q-D fl^

c

.^

0 Q

\U

o

CD CD

JJ

^CJ OCJ CJ CJ OCJ °r CJ 0 O 0 U ^Q_ D_ Q_ Q_ Q_ Q- Q_ TO Q- CD Q0 n D C ^ Q - Q C D - Q - C C Q - i r Q - D C O Q - O)Q- $ D - D C ^ CC Q- DC L U Q F Q O F C J Q U CJQQCJQQCJQOQCJQ-Q

(A

M— O

1

ti _5 —

?

3 o

Q.

H

c

.2

o

.2 DO

CD >

"D

'4-1

Tf

uj

m H

CJ CD

2

o Q-

X

00

o

0 < r

~O "O i o o io 3 3 CQOQC

X

< < 2

i i i 1 1 1 1 1 1 r" r~ (^ QJ [^ o i x x i r o i o oi o oi o oi o io ;rn : 3 3 3 3 3 3 3— QOQCQOQDQCQm 2

^goooooooooo |_

i

O

O

O

O

O

O

O

O

O

^»

c —

c £5

C3 '^

c ~

O(/3(/)cn-C^

^ ^

O

^ D _ C D C D C D C D C D C D C D C D C D 0 Q _

ooCDCUOOCDOCDCDCDOCDl

> O O O 0 0 ? C C

i^^^^^cococo

o

CO 3

CO E

0) CD

CD

CD

CO D

CO 3

CO D

CO D

O O 0 O

^ ^ 20 ro Is ^ s s ^^s

"o 'c >5 •£,o CD 0 c o ••• -—i CJ

Si >-D

s1

"CD — > O -C >" ^ "co

Is CO

•| g.

ss ss> CM CM CN LO r r LO r- r- CM t- CM 00 00 O O 0 0 O O o o

^. CD

%.

Handbook of Biomedical Fluorescence

Biomedical Applications of Hydrogels Handbook

Biomedical Applications of Hydrogels Handbook

Handbook of Single Molecule Fluorescence Spectroscopy

Handbook of Single Molecule Fluorescence Spectroscopy

Handbook of Practical X-ray Fluorescence Analysis

Handbook of Single Molecule Fluorescence Spectroscopy

Biomedical Photonics Handbook

Biomedical Photonics Handbook

Biomedical Engineering Handbook

Standardization and Quality Assurance in Fluorescence Measurements II: Bioanalytical and Biomedical Applications (Springer Series on Fluorescence)

Principles of Fluorescence Spectroscopy

Handbook of biomedical image analysis. Registration Models

The Biomedical Quality Auditor Handbook

Clinical Engineering Handbook (Biomedical Engineering)

Standard Handbook of Biomedical Engineering and Design

Handbook of Biomedical Nonlinear Optical Microscopy

Handbook of biomedical image analysis. Segmentation models

Handbook of biomedical image analysis. Segmentation models

Handbook of Biomedical Image Analysis: Registration Models

Handbook of Neuroprosthetic Methods (Biomedical Engineering)

Handbook of Photonics for Biomedical Science

Clinical Engineering Handbook (Biomedical Engineering)

Principles of Fluorescence Spectroscopy

Molecular Fluorescence

Atlas of cell organelles fluorescence

Topics in Fluorescence Spectroscopy, Protein Fluorescence

Series on Fluorescence. Fluorescence Spectroscopy in Biology

Fluorescence of Supermolecules, Polymers, and Nanosystems (Springer Series on Fluorescence) (Springer Series on Fluorescence)

On Fluorescence Radiation of Nitrogen

Reviews in Fluorescence 2005

Handbook of Biomedical Fluorescence

Biomedical Applications of Hydrogels Handbook

Biomedical Applications of Hydrogels Handbook

Handbook of Single Molecule Fluorescence Spectroscopy

Handbook of Single Molecule Fluorescence Spectroscopy

Handbook of Practical X-ray Fluorescence Analysis

Handbook of Single Molecule Fluorescence Spectroscopy

Biomedical Photonics Handbook

Biomedical Photonics Handbook

Biomedical Engineering Handbook

Standardization and Quality Assurance in Fluorescence Measurements II: Bioanalytical and Biomedical Applications (Springer Series on Fluorescence)

Principles of Fluorescence Spectroscopy

Handbook of biomedical image analysis. Registration Models

The Biomedical Quality Auditor Handbook

Clinical Engineering Handbook (Biomedical Engineering)

Standard Handbook of Biomedical Engineering and Design

Handbook of Biomedical Nonlinear Optical Microscopy

Handbook of biomedical image analysis. Segmentation models

Handbook of biomedical image analysis. Segmentation models

Handbook of Biomedical Image Analysis: Registration Models

Handbook of Neuroprosthetic Methods (Biomedical Engineering)

Handbook of Photonics for Biomedical Science

Clinical Engineering Handbook (Biomedical Engineering)

Principles of Fluorescence Spectroscopy

Molecular Fluorescence

Atlas of cell organelles fluorescence

Topics in Fluorescence Spectroscopy, Protein Fluorescence

Series on Fluorescence. Fluorescence Spectroscopy in Biology

Fluorescence of Supermolecules, Polymers, and Nanosystems (Springer Series on Fluorescence) (Springer Series on Fluorescence)

On Fluorescence Radiation of Nitrogen

Reviews in Fluorescence 2005

Recommend Documents