Advances in Magnetic Resonance in Food Science (Special Publications)

Advances in Magnetic Resonance in Food Science Advances in Magnetic Resonance in Food Science Edited by P. S. Belton...

Author: P. BELTON | B. HILLS | G. WEBB

79 downloads 1352 Views 13MB 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

Advances in Magnetic Resonance in Food Science


Edited by

P. S. Belton Institute of Food Science, Norwich, UK B. P. Hills Institute of Food Science, Norwich, UK G. A. Webb University of Surrey, Guildford, UK

RSmC ROYALSOCIETY OF CHEMISTRY

The proceedings of the Fourth International Conference on Applications of Magnetic Resonance in Food Science, held on 7-9 September 1998 in Norwich, UK

Special Publication No. 23 1 ISBN 0-85404-724-7 A catalogue record for this book is available from the British Library 0 The Royal Society of Chemistry 1999

All rights reserved Apart from any fair dealing for the purpose of research or private study, or criticism or review as permitted under the terms of the UK Copyright, Designs and Patents Act, 1988, this publication may not be reproduced, stored or transmitted, in any form or by any means, without the prior permission in writing of The Royal Society of Chemistry, or in the case of reprographic reproduction only in accordance with the terms of the licences issued by the Copyright Licensing Agency in the UK, or in accordance with the terms of the licences issued by the appropriate Reproduction Rights Organization outside the UK. Enquiries concerning reproduction outside the terms stated here should be sent to The Royal Society of Chemistry at the address printed on this page.

Published by The Royal Society of Chemistry, Thomas Graham House, Science Park, Milton Road, Cambridge CB4 OWF, UK For further information see our web site at www.rsc.org Printed and bound by MPG Books Ltd, Bodmin, Cornwall, UK.

Preface

The fourth International Conference on Applications of Magnetic Resonance in Food Science was held in Norwich, UK, between 7th and 9th September 1998. The meeting attracted 120 scientists from 20 countries and was thud a truly international occasion. The success of the Conference is reflected in the high quality of the oral and poster presentations which it attracted. This volume contains the material given in the oral presentations; the science covered in the fifty posters is additional. The Conference comprised major and minor oral contributions divided among five symposia which, taken together, ably demonstrate the protean nature of magnetic resonance techniques in dealing with problems arising in many areas of food science. The order of the chapters in this volume shows a parallelism to that in which the lectures were given at the Conference. Symposium A covered Magnetic Resonance in Food: The Developing Scene; the first four chapters relate to this Symposium. Symposium B dealt with Water, Ions and Small Molecules in Food; Chapters 5 to 10 relate to this material. The following eight chapters belong to the largest of the Symposia, Symposium C, which was devoted to Functional Constituents of Food. Chapters 19 to 21 are from Symposium D, which dealt with Signal Treatment and Analysis in Magnetic Resonance. The topics presented in Symposium E, relating to Applications of Magnetic Resonance to Food Processing and Engineering, are covered in the final five chapters of this volume. The Editors wish to express their gratitude to the authors for the prompt submission of their camera-ready copy manuscripts and to the production staff at the Royal Society of Chemistry for their kind co-operation in the genesis of this volume.

Contents

Magnetic Resonance in Food: The Developing Scene From Solid-Liquid Ratios to Real Time Tomography - The Development of NMR in Food Applications P.J. Lillford and S. Ablett

3

Time Domain NMR Studies under Controlled Shear Conditions S. Ablett, A. Darke and D. Martin

16

Internal Structure Characterization of Soft Cheeses by MRI F. Mariette, G. Collowet, P. Marchal and J.M. Franconi

24

Protein Aggregation Studies Using PFG NMR Diffusion Measurements W.S. Price, F. Tsuchiya and Y. Arata

35

Water, Ions and Small Molecules in Food A Multistate Theory of Water Relations in Biopolymer Systems B.P. Hills, C.E. Manning and J.Godward

45

Molecular Mobility of a System: Waxy Maize, Glycerol and Water Studied by NMR D.C.P. Jardim, J.R. Mitchell, W. Derbyshire, J.M.V. Blanshard and J.A.G Ar2as

63

Water Dynamics in Gelatine. Relaxation and Diffusion Analysis L. Foucat, A. Traore' and J.P. Renou

73

Probing the Physical and Sensory Properties of Food Systems Using NMR Spectroscopy S.J. Schmidt

79

'H Relaxation of Hydrated Carbohydrate Systems J.M. V. Blanshard, W. Derbyshire, W. MacNaughtan, S. Ablett, D. Martin and M.J. h a r d

95

Thermodynamics of Relaxation Phenomena in Freeze-dried Wheat Starch Gel S. Poliszko, D.M. Napierala, R. Rezlar and G. Hojjkann

105

Functional Constituents of Food NMR of Food Biopolymers P.S. Belton

115

...


Vlll

Solid State I3C NMR Studies of Wheat High Molecular Weight Subunits A.M. Gil, E. Alberti, A. Nait6, K. Okuda, H. Sait6, A S . Tatham and S.Gilbert

126

The Application of Electron Spin Resonance Spectroscopy to the Detection and Transfer of Free Radicals in Protein-Lipid Systems N.K. Howell and S. Saeed

135

Editing the Information in Solid State Carbon-13 NMR Spectra of Food R.H. Newman

144

Cross-polarisation Kinetics and the Determination of Proton Mobility in Hydrated Plant Cell Walls M.C. Jarvis. M A . Ha and R.J. Vietor

158

Proton Relaxation in Plant Cell Walls and Model Systems H. Tang and P.S. Belton

166

Probing Molecular Motions of Low Moisture Starch Gels by Carbon-1 3 NMR Y. Vodovotz and P. Chinachoti

185

Applications of ESR Imaging in Food Science D.G. Gillies

193

Signal Treatment and Analysis in Magnetic Resonance Analysis of Time Domain NMR and Other Signals D.N. Rutledge, A S . Barros, M.C. Vackier, S. Baumberger and C. Lapierre

203

Comparative Chemometric Analysis of Transverse Low-field 'H NMR Relaxation Data I.E. Bechmann, H.T. Pedersen, L. N@rgaardand S.B. Engelsen

217

Quality Evaluation of Atlantic Halibut (Hippoglossus hippoglossus L) during Ice Storage Using 'H NMR Spectroscopy B. Sitter, J. Krane, I.S. Gribbestad, L. J@rgensenand M. Aursand

226

Applications of Magnetic Resonance to Food Processing and Engineering Magnetic Resonance Temperature Mapping A.G. Webb and J. B. Litchfeld

24 1

Study and Modelisation of Starch Gelatinisation in Potatoes with Magnetic Resonance Imaging C.A. Toussaint, F. Lungevin, J.-P. Pain and A. Goullieux

256

Contents

ix

Online Magnetic Resonance Imaging for Detection of Spoilage in Finished Packages T.W. Schenz, B. Dauber, C. Nicholls, C. Gardner, V.A. Scott, S.P.Roberts and M.J. Hennesy

264

Magnetic Resonance Mapping of Solid Fat Content of Adipose Tissues in Meat

272

A. Davenel, P . Marchal, A. Riaublanc and G. Gandemer

Time Domain 'HNMR: Its Relevance to the Processing and Storage of Starch Systems 280 I.A. Farhat, J.M. V. Blanshard and J.R. Mitchell Subject Index

289

Magnetic Resonance in Food: The Developing Scene

From Solid-Liquid Ratios to Real Time Tomography The Development of NMR in Food Applications P. J. Lillford and S. Ablett UNILEVER RESEARCH COLWORTH, COLWORTH HOUSE, SHARNBROOK, BEDFORD MK44 lLQ, UK

1 INTRODUCTION Most food materials are of complex chemical composition, heterogeneous structure, and reactive. i.e. just like the biological materials from which most of them arise. Food technology is the developing skill which allows all these variables to be understood, controlled and manipulated to produce nutritious, attractive, entertaining and, most of all, safe food. To understand and control anyhng requires that firstly measurements must be made, and it is easy to write down the information that is required to build a predictive model of the behaviour of any food product or process, (and processes not only involve the fabrication and assembly of products, but also their degradation during storage, use and consumption). We need to know: What are the molecules present, and how many of each? Where are they within the product? Are they stationary or moving, and if so at what speed? What are they reacting with? - and though it is reasonable to construct models to simplify the study of each of these questions, it would be ideal if we could have a single instrument, to measure all of them, in real time and on the immediate subject of interest. Fortunately, the food technologist is not alone in asking these questions. It is also what all medical researchers would like to know, and they have the added problem that their subjects of interest are not cheap or easily and frequently manufactured and dissected. So a non-invasive route to all of these measurements is also advantageous to all of us. Fortunately, Physical Science gave us the route to the solution of these problems over 50 years ago by the discover of nuclear magnetism, and the demonstration of nuclear magnetic resonance (NMR). The solution to all our measurement problems are encapsulated in two basic equations. Viz.

4


M = ( N y 2h21(I+ l)B) l3kT

(1)

and

-T2= c

(

32+

52

+

1+W2Z2

1

1 + 422 02r2

The first tells us that if we can measure the net magnetisation of nuclei, they will be labelled by their specific magnetic moment (y), and we can count how many (N). Also, by controlling the external field (B), we can expose internal field shifts giving molecular information, or even gross position in space; and if all these were constant we could even measure temperature (T). The second tells us that the decay of magnetisation gives direct information on the molecular correlation times (T) and therefore movement of the nucleus and the molecule in which it is contained. All our measurement problems are over. It’s just a question of doing it! The real rate of progress in solving our measurement problems has been in the hands of electronic engineers,physicist and latterly computer scientists, who have identified how to extract the parameters and latterly how to reassemble complex data sets to provide the molecular and structural information we need. We still don’t have machines that can answer all our questions but before we complain we should now review what we have done with what they have already provided. I will approach this on an approximately chronological basis, but like most scientific developments, progress is rarely linear but transfers developmentsfrom one field to another on an entrepreneurial basis. 2 THE CONTINUOUS WAVE ERA

The early instrumentsused relatively low magnetic fields corresponding to I 60 MH, nuclear fiequencies for protons and used continuous wave radio fiequencies for excitation. There were two parallel developments, firstly in search of information on molecular structure - “high resolution” which operated well on solution spectra of mobile molecules. Liquid foods (and drinks) were open to study, and many of us, as undergraduates, first learnt the principles of chemical shifts and spin coupling from the spectra of alcohol and sugar solutions. The first protein spectra was published in 1957’. Certainly, by the 1960s Unilever was examining high resolution spectra of tea, only to find that molecular complexity rather than intrinsic resolution limited the interpretationof the spectra. Even 30 years later, with advanced in high field and pulsed Fourier Transform instruments, it is still not easy (see Figure 1).


,

~

8.0

'

7.8 7.6

~

7.4

,

~

7.2 7.0 6.8 6.6

5

~

,

l

6.4 6.2 6.0

~

l

~

~

~

ppm

Figure 1 Proton NMR Spectrum of a Black Tea Extract: Expansion showingpolyphenol Signals Broad band instruments were capable of identifying relaxation time differences of solids and liquids by analysis of lineshapes, with immediate impact on food formulation in the areas of fats and oils, and water. Most edible fats exist as mixtures of liquid and crystalline forms at body temperature so that the proportions of each can be measured directly from N M R lineshape (Figure 2).

t

Signal Intensity

I

Field-

Figure 2 Continuous Wave NMR Signalfiom a Sample Containing a Mixture of Solid and Liquid Phases.

.

~

6


Table 1 A Comparison of the analytical D20contents of crisps with the concentration of

D20detected by NMR (-1969) Total D 2 0 content

D,O concentration detected i.e. bound water concentration

Solid-like water concentration

19.8 10.1 6.5 5.8

16.1 0.9 0.6 0.5

3.7 9.2 5.9 5.3

A simple method, allowing the quantification of amounts of solid and liquid fats as a function of blending source and temperature has been of enormous commercial value. It probably pays my salary. Observations of water in dried foods showed not dissimilar spectra. Some of the water appeared to be solid at temperatures well above its freezing point. Table 1 shows early results of deuteron studies in potato crisps. The ‘liquid’ water appears somewhere between 5 and 10% w:w, which is the same point where crispness is lost. It appeared that N M R could now measure textural properties and the search for ‘bound water’ was afoot. But continuous wave experiments were slow, signal to noise was awful and we needed to speed things up.

3 PULSED N M R STUDIES It is amazing to think that as late as 1968, Emst was still doubtful that pulsed nuclear excitation followed by Fourier Transform of the resulting decay signal would become a regular way of conducting NMR. His doubts related to the difficulty in collecting sufficient detailed data to produce high resolution spectra. But for ‘broad line’ practitioners there was no such problem. The collection of decay rates provided direct access to relaxation times (T, and T,) and was much faster and could be averaged. The CW instruments were superseded by pulse machines with simple averagers, and as many pulse sequences as one could afford or build oneself. Solid-liquid ratios became cheap and almost on-line, and the study of water in foods and model systems began in earnest. 3.1 “Water binding”

As early as 1954’, it was reported that the line width of water was increased by the presence of biological material (deoxyribonucleic acid). Broad line studies also showed that food gels and biological tissues had anomalously broad water peaks, and the simplistic explanation that all the water was more solid-like, bound or less structurally mobile was briefly advanced. This did not last long after the realisation that rapid exchange according to the Zimmermann-Britten model was the probable reason. i.e.

Magnetic Resonunce in Food: The Developing Scene

1

1 1 =x, -+(l-xB)-

T20bs

T2B

(3)

TZF

where XBrelates to the proportion of water “bound “ to substrates and TZobs, T,, and T,, relate to observed, bound and free water relaxation time. Furthermore, an estimate of XB could be envisaged by connecting with the observation that N M R saw a ftaction of water with reduced mobility, but not frozen, below the fleezing point. Solutes clearly affect the mobility of some water, if not all of it3. Real foods showed the same behaviour and firtherrnore the processes of rigor mortis and cooking were reflected in the water proton relaxation behaviour4.As data collection improved, so did the number of apparent water relaxation times, from 2 to 3 and up to 5, where the number of fittable parameters exceeded the reasonable number of domains where water could be thought to reside. An alternative approach, transforming the decay curve to a relaxation time spectrum was proposed’, where the origin of the complex decay can be related to the spatial heterogeneity of the sample over scales of lo’s of microns (Figure 3). Transverse proton relaxation in a 4.8%agarose gel:

!h

b. Freezethaw damaged gel

a. Homogeneous gel

i s”

40

Time (msec)

200

40

120

Time ( m e )

Schematic representation of the structure of agarose gel: a. Homogeneous gel

b. Freezethaw damaged gel

Figure 3 Spatial Heterogeneity and Dzfision Lengths

8


The NMR machine can now be used as structure measurement tool, capable of measuring the effect of processes relevant to food processing. The approach has been investigated and developed extensively by Brian Hills et a16 and will be referred to in a later lecture. The relaxation time spectrum has also been correlated with the sensory impression of juiciness in the mouth. Not surprisingly, the water least influenced by the architecture of the food is released the most quickly. 3.2 Water droplets Many foods e.g. margarine, low fat spreads, dressing, are oil continuous emulsions, containing disperse droplets of aqueous solution. The stability, both microbiological and physical, and even the mouthfeel are dependent on the droplet size distribution. The insertion of controlled field gradients pulses within a normal echo train causes the refocussed signal to be dominated by the diffusion of water rather than its intrinsic relaxation time (Figure 4). If diffusions is restricted (by the boundaries of a dro let) then P anomalous diffusion is observed fkom which the droplet sizes can be estimated .

6

c)

z

b = exp{-y

2 2

6

AG

2

G D

(A-6/3)}

Y = Magnetogyric Ratio G= Magnetic Field Gradient D= Diffusion Coefficient

Figure 4 Pulse sequencefor determination of self diffusion 3.3 Diffusion in Foods

The interest in diffusion rates is not limited to emulsions. Essentially, the self diffusion coefficient of any small molecule in a matrix of any other is vital to the reaction rates of deteriorative processes, flavour retention, and migration between components. All


9

of these can be measured by the pulsed field gradient method. Figure (5) shows a recent and surprising result, that the water within a “glassy” polysaccharide still exhibits considerable diffusive motion, and is quite independent of the glassing of the polymer itsel?.

D (m2/sec) 10-10 L

* * --

lo-”

0

10

20

30

40

50

60

70

80

90

I

100

Temperature (C)

Figure 5 SelfDiffusion Coeficient of the Water in 81 % Pullulan 4 FOURIER TRANSFORMED HIGH RESOLUTION SPECTRA

Emst’s early worries were unfounded. As computer power developed, and NMR spectroscopists found clever ways of enhancing signal to noise, FT-NMR became the dominant experiment and allowed amazing results to be delivered. As early as 1970, Stahl and McNaught showed that with a little hydrolysis to reduce viscosity, NMR could be the most effective method of analysing starches and chemically modified food ingredientsg. Access to natural abundance CI3 spectra made even further possibilities available, such as this detailed assignments of hyaluronate polymer resonance. Chemical shift and line width changes can be quantitatively measured as association or conformational changes take place. Wiithrich reported the first N M R structural determination of a protein in solution in 1985”. All of these advances now allow structural studies of macromolecule unfolding to be measured under conditions relevant to commercial processing. Developments in the medical field showed that 31Pspectra could be obtained in whole cells and organs”. This was seized upon by the food industry to monitor pre to post rigor changes in most muscle tissue, and the location and turnover of phosphates added to enhance water retention in processing and freezing. All these high resolution phenomena rely on sufficient molecular mobility to allow resonances to be observed. But many food systems contain real solids, crystalline

10


materials or polymers associated to form effectively solid state motional restriction - but help was at hand. Since 1958, Andrew and co-workers had been “cheating” by high speed spinning of solid samples at the magic angleI2. As soon as commercial spectrometers became available, the food industry was on board, producing high resolution spectra of crystalline polymorphs of triglycerides, which had previously been lost in the broad line signalsI3 (Figure 6)

170.0

140.0

110 0

80.0

50.0

20.0

0.0

PPM

Figure 6 CP/MAS Spectra from Tripalmitin

But whilst this cadre of workers were scheming to increase resolution, others were playing about with the magnetic field.

5 NMRIMAGING In 1973, Paul Lauterbur published an image of 2 glass tubes by sweeping the external magnetic fieldI4. For those of us doing NMR, it was such an annoyingly clever experiment that we wished we had thought of it first, and several of us repeated it, just to prove that we could! I was lucky enough to be working with Nottingham at the time of the enormous advances made by Mansfield, Andrew et a1 and seeing the first image cross section of Waldo Hinshaw’s finger, and then his wrist, and then other bits of people as the magnets got bigger. Foods found their way into the magnets as fast as we could borrow time on the machines, and the first image of food was shown in 1980 at the Royal S~ciety’~. The publication was not ours, but the apple pie was (Figure 7).


Figure 7 NMR image of Apple Pie

Early work simply showed that images were possible, and that changes such as water migration within composite structures could be measured as a function of time or temperature (Figure 8).

Figure 8 NMR images of Egg Custard - aft.. baking

11

12


The images rapidly improved in quality but what did they mean and how could they be used? The relative intensity of signals was derived from a combination of proton density and relaxation time. Provided one could examine the specimen by visual means, the results were interpretable, but this largely defeats the objective of non invasive testing. Fortunately, the problem was recognised and shared with medical imaging, and food science gave something back to medicine via construction of calibration samples of varying proton density and relaxation time’6. Nowadays, image sequences weighted for density, relaxation time and even diffusion rates are reasonably well understood. After the heady days of image collection, questions began concerning the real value of imaging in foods. Taking photographs of structures is not enough. The technique must provide information not otherwise obtainable, or it would be seen as an expensive luxury. We must answer the basic questions posed in the Introduction to this paper. The ability of NMR to measure dynamic processes in 3-dimensions provides one valuable and novel piece of information. Early experiments used proton density change as the indicator of water flow. Such experiments have been extended to the macroscopic visualisation of whole products (Figure 9), and at the microscopic level of wheat grains (Figure lo), allowing mathematical modellin of heat and mass transfer processes to be matched to the actual flows mapped by NMR The capability of spatial labelling by imaging has allowed the development of “velocimetry” by NMR18. This is particularly important for the direct measurement of flow in Non-Newtonian liquids frequently encountered in food materials, which is important in the measurement of constitutive behaviour (rheology) (Figure 11). And the direct imaging of complex flows in complex geometries (Figure 12)

5’.

Figure 9 Rehydration of Dried Pasta


13

Figure 10 Corrected NMR images of moisture contentfor grains boiled in distilled water.

10

,a,

..

:.***

.'....I

I

.

I

.'.".,

.

..-.,.I

-

0.1 7

H

5

I

*%m

ai u!

a

.-

I111.1))

0

' 0

h

&. .

I

*.

1-

v

""'.I

0.01 7

0.001

111

0

a

I

in Water Flow cutve from MRI Measuredon rheorneter (wne&plateflow)

I I11.11

Figure 11 NMR Rheometry.

. . '., I

'3.

I

..

I .111(

I

.

I ..mum,

*: 7

1

1 I11.11

14


Figure 12 Axial Velocity Maps This means that the developing capabilities of Computational Fluid Dynamics can be challenged by experimental measurement of actual flows in the same complex environments. This should provide some healthy competition between modellers and measurers for a few years. 6 THE WAY FORWARD

So where do we go from here. Firstly, in “conventional” spectroscopy, we need further developments to obtain molecular information of both the complex mobile components but also the solid-like or real solid elements in heterogeneous foods. Fortunately, the developments in magic angle spinning (MAS) allows reduction of both the motional and susceptibility induced broadening, so that high resolution spectra are becoming available either by the simple addition of MAS, or the combined techniques of spinning and cross polarisation (CPMAS). In imaging, there is still the requirement for even higher spatial resolution and shorter timescales. We still cannot achieve the quality of answers to questions originally set in this paper. Even with the current capabilities there are new opportunities only recently

15


identified. Who would have believed that N M R could measure the brain’s response to stimuli likely to be induced by the appearance, smell and taste of food? The developments in imaging food manufacturing processes suggests that the eating and digestion of food is equally amenable to real time study, with real foods inside whole people. We will continue to depend upon the skills of instrument designers and the financial input from medical research. But we must continue to “steal with honour” from all available sources. NMR still has a lot more to offer. ACKNOWLEDGEMENTS The chronology of N M R developments has been described in much greater details elsewhere. Interested readers are recommended to study the excellent chapters in Progress in Nuclear Magnetic Spectroscopy, 28,(1995). Thanks to all the N M R specialists who have taken the brave step of examining foods. Not all have been referred to here, but they know who they are.

References 1. 2. 3. 4.

5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18.

z,

M. Saunders, A. Wishnia and J.C. Kirkwood, J. Amer. Chem. SOC.,1957, 3289. B. Jacobsen, W.A. Anderson and J.T. Amold, Nature, 1954,173,772. I.D. Duff, PhD. Thesis, University of Nottingham, Oct. 1973. for example: P.S. Belton, R.R. Jackson and K.J. Packer, Biochim. Biophys. Acta.,1972,286, 16. C.F. Hazlewood et al. Biophys. J., 1974,14,583. P.J. Lillford et al. ACS Symposium Series, 1980, 127. A.H. Clark and P.J. Lillford, J. Mag. Res., 1980,3,42. B.P. Hills, S.F. Takacs and P.S. Belton, Food Chem., 1990,37,2,95. K.J. Packer and C. Rees, J. Colloid Int, Science, 1972,4,2,206. and J.C. van den Enden et al, 1.Colloidfnt. Science, 1990,140. 1, 105. S. Ablett, A.H. Darke, M.J. Izzard and P.J. Lillford, in “The Glassy State in Foods” ed. J.M. Blanshard, P.J. Lillford, Nottingham Press, 1993, 189. H. Stahl and R.P. McNaught, Cereal Chem., 1970, flz, 345. M.P. Williamson, T. Have1 and K. Wuthrich, J. Mol. Biol., 1985,182, 195. D.I. Hoult et al., Nature, 1974,252,285. E.R. Andrew et al., Nature, 1958,182, 1659. S.M. Bociek et al., J.A.O.C.S., 1985,62,8, 1261. P.C. Lauterbur, Nature, 1973,242, 190. S.F.J. Cox et al., Phil. Trans. Roy. SOC.,1980,289. 1037. P.M. Walker et al, Phys. Med. Biol., 1989,2, 1,s. A.G.F. Stapley et al., Int. J. ofFood Sci. and Technol. 1997,2,355-375 For example: I.T. Callaghan and Y. Xia, J. Mag. Res., 1 9 9 1 , a , 326. B. Newling et al., Chem. Eng. Sciences, 1997,2, 13,2059.

Time Domain NMR Studies under Controlled Shear Conditions Steve Ablett, Arthur Darke and Dave Martin UNILEVER RESEARCH COLWORTH, COLWORTH HOUSE, SHARNBROOK, BEDFORD MK44 ILQ, UK

1 INTRODUCTION

Time domain NMR is now a well established techtuque within many Food Science laboratories for the characterisation of food microstructure. These range from simple routine measurements of the solids content of fat blends’, through to the more complex analysis of water relaxation decays to probe the microstructure present within foods293. Many food processes involve the application of shear to generate the required final product structure, and time domain NMR has proven to be a powerhl technique for characterising the changes that have taken place. However, these N M R measurements have to be performed off-line, which means it is only possible to characterise the initial and final structure produced. The ability to make these types of NMR measurements directly under shear conditions would be beneficial because it would allow the dynamics of the microstructural changes to be directly monitored. Shear cell devices have previously been constructed for N M R spectrometers, but these have primarily been used to study the flow patterns of rather than for bulk NMR measurements to characterise any shear induced microstructural changes. In addition, these measurements have tended to have been undertaken on specialised MRI equipment which most food scientists would not have ready access to. In this paper we describe the design of a shear cell, a simple addition to a standard commercial benchtop spectrometer, which allows the dynamics of shear induced structural changes in fluid samples to be directly investigated. The application of this device is illustrated by demonstrating how it can be used to study the effect of shear on the crystallisation kinetics of edible fats, and on the gelation properties of what are referred to as biopolymer ‘fluid gels”.

2. DESCRIPTION OF THE SHEAR CELL

A concentric cylindrical couette type device has been designed and constructed for use with a Resonance Instruments MARAN benchtop spectrometer. It has been designed to operate inside a standard NMR tube which required no major modifications to the spectrometer, allowing the cell to be easily installed without compromising the performance of the spectrometer when it is required for other applications. The cell has

17


been constructed using a standard 9 inch long 18 mm O.D. precision glass NMR tube (Wilmad code no.18-PP-9), together a 13.55 mm O.D. central cylinder made of a plastic material (PEEK), giving a sample gap of 1.5 mm. The bottom of the N M R tube is fitted with a glass and ceramic bearing to keep the inner cylinder central. The design of the shear cell is shown schematically in figure 1, and a photograph of the actual cell components is shown in figure 2. The inner cylinder is driven by a standard laboratory motor (Jake & Kunkel type RW 20 DZM), which together with a 1 O :l step down gear box permits rotation speeds of 5 rpm to 400 rpm to be readily achieved, allowing shear rates in the range -3 sec-' to 230 sec-' to be applied to the fluids. The shear cell is clamped to the top of the magnet box, and a frame has been constructed around the magnet box to hold the motor firmly in place relative to the shear cell (figure 3). Temperature control is within the range -20°C to 1OO"C, with a typical sample volume of 5 ml.

u

R

MotorDrive

Rotating Inner Cylinder

Nh4R Tube (18mm OD) Sample (1.5mm gap)

C- Magnet Pole Piece

Bottom Bearing

Figure 1

Schematic of the design of the shear cell.


Figure 2 Photograph of the components of the shear cell showing the standard MMR tube Jitted with the bottom bearing, the inner plastic cylinder, and the top bush to attach the cell to the magnet.

Figure 3 Photograph showing the attachment of the motor and the shear cell to the spectrometer magnet.


19

3. RESULTS AND DISCUSSION 3.1 Characteristics of the Shear Cell

The characteristics of this device were evaluated using proton relaxation studies of both pure water (T2-2sec) and water doped with gadolinium chloride (T2-5Omsec) to establish if the cell itself, or the shear induced flow of the sample, caused any artefact to the recorded NMR signal. These two test samples were chosen since they were considered to represent the typical range of relaxation times to be studied by this method. 3.1.1 Effect of the Shear Cell on the NMR Signal. The central cylinder is machined fiom PEEK which gives a strong proton signal with a very short relaxation time which completely decays within -70 ps. The close proximity of the motor causes small fluctuations in the magnetic field homogeneity which also affects the FID. This means the device is not suitable for recording FID's, i.e. for studying samples with very short relaxation times. However, no background signal from the shear cell was detected with the Carr Purcell Meiboom Gill (CPMG) pulse sequence', and the operation of the motor did not have any detectable effect on this signal. 3.1.2. Effect of Applied Shear on the CPMG signal. CPMG pulse sequences are routinely used in time domain NMR studies to eliminate the effect of molecular diffusion on the NMR signal. The amplitude of an echo in the CPMG pulse sequence is given by-

A(echo at time t) oc exp[-(t/Tz) - 0.33y2G2Dz3] where G is the magnetic field gradient due to magnet inhomogeneity D is the diffusion coefficient z is the 90"- 180' pulse spacing (i.e. echo spacing =22). The value of z is chosen to be sufficiently short such that the second term in equation 1 is negligible, aflowing the value of T2 to be determined directly fiom the CPMG decay. However with this shear cell, in addition to normal diffusion processes there is also forced fluid flow taking place. This could increase the magnitude of the second term in equation 1 such that it is no longer negligible. The effect of changing the 5 spacing was evaluated for both the pure water and the doped water over the range of z values typically used (50 KS < r < 250 ps). It was found that the application of shear had no significant effect on either the measured amplitude or the relaxation time for both samples studied over the range of z spacings investigated. The effect of the rotation speed of the central cylinder was also investigated. No significant effect on the recorded CPMG signal was observed for both the pure water and the doped water samples over the range of rotation speeds (0 to 100 rpm) investigated. It was concluded that the second term in equation 1 remains negligible despite the increased motion caused by the application of shear, and therefore the application of shear introduces no measurable artefacts to the recorded CPMG decay.


LU

3.2 SolidILiquid Fat Measurements One of the widest applications of bench top NMR spectrometers is the determination of the solid/liquid ratio of fats'. The N M R signal of a fat typically contains two components, a very rapid decay from the crystalline component and a much slower decay from the liquid component. Two different approaches have been developed for the routine analysis of the solidhquid ratio of fats'. The so called 'direct method' calculates the solids content directly from the FID signal. The amplitude of the signal relating to both the solid and liquid components is measured close to the start of the FID, and the amplitude of just the liquid component is recorded some 60 ps later when it is assumed the signal from the solid component will have completely relaxed. The solids content is then calculated directly from these two amplitude values. In contrast to this approach, the 'indirect method' only measures the amplitude of the liquid component of the NMR signal at the temperature of interest. This amplitude is then compared to the increased signal amplitude recorded at a higher temperature where it is known that the fat will be totally liquid. A reference sample of oil which remains totally liquid over the whole temperature range of interest also needs to be measured to compensate for the temperature dependence of the NMR signal amplitude. The solids content of the fat is then calculated using the following equation %Solids = 100[ ~-(R~&I)(S,I/S,O)]

(2)

where t 1 is the measurement temperature tO is the temperature at which the sample is 100% liquid R is the signal amplitude of the reference oil S is the signal amplitude of the sample. Both of the two above methods are widely used throughout the Food Industry to evaluate the solids content of fats. They are primarily used to construct phase diagrams for the fats which are subsequently used to help with product formulations. However, it is well known that fats can readily supercool, in which case, supplementary information on the rate of crystallisation is also likely to be required. Fortunately, the speed of the NMR measurement is sufficient short (i.e. a few seconds) that in most cases, this non equilibrium solids content can be determined. However, many food products are processed under shear conditions which could potentially moditjl the crystallisation rate of fats. For this situation, the measurement of the solids content while shear is directly applied to the fat is ideally required. It has already been shown that this device is not suitable for studying FID's so the direct method cannot be used, However the indirect method can be used to study the crystallisation behaviour of a fat, by measuring the initial amplitude of the liquid component using the CPMG pulse sequence. A fat sample (hardened palm kernel oil) was initially heated to 50°C in the shear cell to ensure it was completely melted, and then slowly cooled from 30°C to 15°C at a rate of l"C/minute. The CPMG signal was recorded every minute (i.e. at each 1°C step), initially for the sample under static conditions (i.e. 0 rpm), and again with the sample under shear conditions (i.e. 400 rpm). This measurement protocol was then repeated with pure sunflower oil to allow the corresponding reference oil signal amplitudes to be acquired. The NMR signals were fitted to a single exponential function to allow the initial amplitude of the liquid component present in the sample to be determined. The solids contents were

21


calculated using equation 2, and the results are shown in figure 4. It can be seen that the

30

25

20

15

Temperature IC

Figure 4 Solid fat content of hardened palm kernel oil as a firnction of decreasing temperature (I Qminute) with (crosses) and without (open circles) applied shear. 3.3 Biopolymer Fluid Gels

Biopolymers have been used for many years by the Food Industry for their gelling properties which impart textural improvements to food products. More recently it has been shown that novel properties can be achieved with these materials when shear is applied during the gelation process’. This produces what are referred to as ‘fluid gels’. The action of shear during gelation causes very small gel particles to be formed whose dimensions are typically on the micron scale. This results in the final material having no long range network structure, and so produces a pourable colloidal type liquid instead of a solid like gel which would have normally been produced under static conditions. Agarose has been extensively studied as a model for biopolymer gelation, and it has been established that the gelation mechanism is via the formation of a three dimensional network of aggregated double helices. Previous NMR studiesg on agarose have shown the water relaxation time (Tz) to be particularly sensitive to the degree of aggregation present and is a powefil tool for probing the dynamics of the network formation during gelation. The application of NMR to the study of biopolymer fluid gel formation is severely limited because it is not possible to study this process directly within a conventional NMR probe. Previously measurements have had to be restricted to the initial and final structure formed, but the development of this shear cell provides the potential for directly studying the complete dynamic process of fluid gel formation. The NMR signal from 3% agarose was recorded using the CPMG pulse sequence as a function of temperature from 70°C to 28°C at a cooling rate of 1°C/minute and then held at 28°C for a further 30 minutes. Both static and shear conditions (i.e. 100rpm) were

22


investigated, and the results are shown in figure 5 . This shows the kinetics of structure formation within the fluid gel to be the same as those of the normal quiescently formed gel. However, the final Tzvalue of the water within the fluid gel is significantly longer than the corresponding value Erom the quiescent gel. This suggests some microstructural modification within the fluid gel particles compared that that of the quiescent gel, although the origin of these structural difference is unclear at present. It should be noted that similar results to these have also been observed using another cylindrical couette type shear cell inside a MRI spectrometer"

loooo 1000

I------

. v)

E cu

I-

100

10

0

20

40

60

80

Time lmins

Figure 5 Tz relaxation time of the water in 3% agarose as a function of temperature with (crosses) and without (open circles) applied shear. The solution was cooled (I Wminute) from 70 97 to 28 T and held at this temperaturefor afirther 30 minutes. 4. CONCLUSIONS

A shear cell has been designed and constructed for use with a low cost commercial bench top Nh4R spectrometer. It operates inside a standard NMR tube, allowing easy installation without compromising the overall performance of the spectrometer when it is required for other applications. It has been shown that no detectable artefacts are introduced into the NMR signal using the CPMG pulse sequence, either from the shear device itself or from the flow of the sample which results from the applied shear. The potential of this type of shear device to the Food Industry has been demonstrated by showing how it can be used to follow shear induced effects to both the crystallisation behaviour of fats and the dynamics of biopolymer fluid gel formation.

5 , ACKNOWLEDGEMENTS

The authors wish to thank Mr C. Marriott of the Engineering Support Group (Unilever Research Colworth) for the design and construction of the shear device.


23

References

K. Van Putte and J. Van den Enden, J. Amer. Oil Chem. Soc., 1974,51,316. P. J. Lillford, A. H. Clark and D. V. Jones, ‘Water in Polymers’, Washington DC, 1980, ACS Symp. Series No. 127, p177. 3. P. S. Belton and R. G. Ratcliffe, Prog. NMR Spectrosc., 1985, 17,24. 4. P.T. Callaghan and Y. Xia, J. Mug. Rex, 1991, 91, 326. 5. R. L. Powell, J. E. Maneval, J. D. Seymour, K. L. McCarthyandM. J. McCarthy, J. Rheol., 1994, 38, 1465. 6. S. J. Gibbs, D. Xing, T. A. Carpenter, L. D. Hall, S . Ablett, I. D. Evans, W. Frith and D. E. Haycock, J. Rheol., 1994,38, 1757. 7. I. T. Norton, T. Foster and C. R. T. Brown, ‘9th International Conference on Gums and Stabilisers for the Food Industry’, Wrexham, 1997. 8. S. Meiboom and D. Gill, Rev. Sci. Inst., 1958, 29, 688. 9. S. Ablett, P. J. Lillford, S . M. A. Baghdadi and W. D. Derbyshire, J. Colloid Interface Sci., 1978, 67, 355. 10. A. L. Hanlon, PhD Thesis, University of Cambridge, 1998. 1. 2.

Internal Structure Characterization of Soft Cheeses by MRI F. Mariette,' G. Collewet,' P. Marchal' and J. M. Franconi2

'

CEMAGREF, TECHNOLOGY DIVISION, 17 AVENUE DE CUCILLE, CS 64427,35044 RENNES, FRANCE * SIEMENS SA, 3 9 4 7 BOULEVARD ORNANO, 93527 SAINT DENIS, FRANCE

1 INTRODUCTION Structure is a relevant characteristic of food products. It is often related to texture and flavour but also stability during storage. Consequently the structure is very important for those involved in the chain of production, fiom farmer to consumer'. The structure of soft cheeses can range from hard to soft, almost semi-liquid as the Camembert cheese. Moreover the structure can be highly heterogeneous inside a single product. All these aspects increase the difficulty to develop instrumental techniques for structure characterisation. If numerous works have been published for micro structure description, relatively few works have been done for macro structure description. To achieve this goal, Magnetic Resonance imaging (MRI) technique presents numerous advantages. The object can be studied in a non-invasive and nondestructive way. Structural information can be obtained such as air or liquid pockets, and cracks. Moreover information can be obtained on chemical composition and molecular dynamic. Since the last decade the applications of the MRI technique in the field of food science have been increasing. Numerous review papers have been written2933.Applied on dairy food MRI has been evaluated to study hard cheese openings formation and di~crimination~.~.~, dairy gels shrinkage' and internal fat and moisture measurements*. The objective of our work was to discriminate soft cheeses texture from their structure measured by MRI. Numerous MRI sequences were defined and the MR images obtained were analysed.

2 MATERIAL AND METHODS 2.1 Cheese Collection The soft cheese collection was provided by Bongrain company. The cheeses were selected in order to investigate a wide structure and texture range variation. Six different cheese technologies were studied : a) one ultrafiltrated (UF) technology b) one thennophilic (THERMO) and c) four mesophilic (MESO I, MESO 11, MESO 111, MESO IV). For each technology 6 cheeses from different batches were analysed. Two mesophilic technologies (MESO I and MESO 11) were characterized by an heterogeneous structure formation. To take this effect into account, a kinetic procedure was proposed. This kinetic procedure consists in making two


25

measurements, one when the cheeses were young and one at the end of life. Consequently the cheese collection was composed of 48 cheeses.

2.2 MRI Data Sets MR images were acquired using a 0.2 T MR scanner (Open System, Siemens) with a head coil. All the images were acquired in the plane parallel to the longest cheese surface using a two-dimensional (2D) Fourier transform technique. Three different pulse sequences were used : a) a standard spin-echo (SE) sequence to produce proton-density (PD) , T2-weighted (T2) and TI-weighted (TI) images, b) a gradient echo (FLASH) sequence to produce magnetic susceptibility-weighted images and c) a multi-echo (CPMG) sequence to calculate T2 parametric images. Three sets of images were obtained from the SE pulse sequence , proton density weighted images (TE =17 ms and TR = 1500 ms), Tz weighted images (TE = 80 ms and TR = 1500 ms) and TI weighted images (TE = 15 ms and TR = 3000 ms). The parameters of the FLASH sequences were TE = 23 ms, TR = I500 ms. Three slices were acquired per cheese, with a field of view (FOV) = 180 * 180 mm, slice thickness = 4 mm and a 256 x 256 pixel matrix. The multi-echo sequence parameters were TE=17 ms, 24 equidistant echoes, TR= 1500 ms, slice thickness = 4 mm and a 128 x 128 pixel matrix. One T2 parametric images was acquired per cheese. For each sequence the gain and FFT scale parameter were kept constant. The T2 relaxation times were calculated according to the Marquartd algorithm using a monoexponential fitting model. All the MRI measurement were performed at 16°C. 2.3 Image Analysis

Three different kinds of information were extracted from the MR images: T2 value distribution, opening characteristics and information on image heterogeneity 2.3.1 Tz Distribution. The T2 map analysis was achieved from the T2 histogram representing the T2 value from 1 to 100 ms and the amount of pixel at a specific TZvalues. The histogram frequency is normalised by the total pixel number to prevent any distortion of the histogram from the cheese MR image size. 2.3.2 Opening Characteristics. The openings were segmented in each proton density image using an adaptive thresholding (see an example in Figure 1). For each opening the following characteristics were computed : the surface S, the perimeter P (computed with the Crofton formulae), the compacity C = P2/4*ll*Swhich is equal to 1 for a perfect circle and decreases with the complexity of the shape, the distance D to the nearest opening neighbour. Seven characteristics were extracted from the distribution histograms of S, P, C and D (minimum, maximum, mean, standard deviation, variance, skewness and kurtosis values). Finally, for each cheese the ratio R of the total openings surface to the cheese surface and the number N of openings per surface unit were computed. The three slices of each cheese were collected.

26


Figure 1 From lefr to right :original proton density and segmented openings 2.3.3 Textural Analysis. In order to quantify the image heterogeneity, textural features were computed. The image texture is the visual information characterized by pixels grey levels properties and their spatial relationships to each other. Texture classification for image segmentation or products characterisation is used in many applications fields such as medical imagery, remote sensing and products inspection or classification8~93'0~''~'2. Within the several approaches to compute textural feature^'^"^, the one proposed by Haralick" was found to be the most powerful to describe texture in generalI6. This method gives 13 parameters extracted from a statistical study of grey level variations between one pixel and its neighbour at distance d in direction 6 (Table 1). Among these parameters some are easily linked with images features such as the angular second moment (or energy) which measures the homogeneity in the image, the contrast and the inverse different moment are related to the local variations in the image, the entropy measures the randomness of the grey levels, the correlation corresponds to the grey level linear dependencies in the image. We computed the 13 parameters with d=l in all directions. These parameters were computed on proton density, TZ weighted and magnetic susce tibility weighted images. The images were first filtered with a Nagao type . filter to improve signal to noise ratio. More, 7 parameters (minimum, maximum, mean, standard deviation, variance, skewness and kurtosis values) were extracted from the grey-level histograms.

I7

Sum Variance Correlation Variance

Entro Difference Variance Difference Entro

no9 nol 1

Sum average

Table 1 Textural parameters

Principal component analysis (PCA) and discriminant 2.3.4 Data Analysis. analysis (DA) included in the Stalab software (SPL infoware, Paris) were used. The data were subjected to PCA to identify data structure and to achieve dimensionality reduction. The PCA method expresses the variable vector as a linear combination of a set of orthogonal (uncorrelated) vectors called principal components (PC). The first principal component is chosen to account for the largest possible fraction of the total variance. Each successive principal component is then chosen to account for the largest possible fraction of the remaining variance. The correlation of the variables, or of the samples and the principal components were assessed and are helpful to

Magnetic Resonance in Food: lhe Developing Scene

21

identify data structure. Then a discriminant analysis (DC) was performed on principal component to separate the groups in multidimensional feature space. The misclassification rate was estimated with the cross validation leave-one-out method.

3 RESULTS

Figure 2

Proton density weighted images. From top to bottom, left to right : UF, THERMO,young MESO I, old MESO I, young MESO I1 and old MESO 11, MESO Ill, MESO N.

Figure 2 shows examples of proton density weighted images of the 8 types of cheese.These cheeses present significant differences in many points. young MESO I and young MESO I1 have different grey level in the centre and near the rind while the other cheeses are more uniform. The number of openings is quite different from UF with no openings at all to MESO IV with a lots of openings and THERMO with few openings. MESO 111 and MESO IV present very large openings contrary to old MESO I or old MESO I1 for example. Figure 3 shows the effect of MR parameters on the young MESO I MR images contrast. Note the great effect of TE on the contrast between the non mature and more mature area while the FLASH sequence exalted the openings because of the magnetic susceptibility effects between curd and openings.

Figure 3 From left to right :MR images from proton density, T2 weighted and FLASH sequence on a young MESO I cheese.

28


3.1 T2 Histogram Analysis An example of T2 histograms is given in Figures 4 and 5. According to the cheese technology the T2 distribution varied from a narrow distribution for homogeneous cheese such as the ultrafiltrate one to a bi-modal distribution for the heterogeneous cheese such as young MESO I.

1

14

27

40 53

66

T2 (ms)

Figure 4

79

92

1

14

27

40

53

66

79

92

T2 (ms)

T2 histogram of UF cheese Figure 5 T2 histogram ofyoung MESO MR image I MR image

The mean T2 value variation was between 30 to 65 ms. In order to identify data structure between T2 histogram a PCA was performed. The results showed that the total data variation was distributed among many principal components. For example the first two principal components PC 1 and PC 2 explained only 45.44 % of the total variation. These resuIts could be explained by two concomitant factors. The first factor was related to the non linearity of the data. A change of Tz mean value or a change of the histogram dispersion was converted through the PCA method into a plane. The second factor resulted from the intra-variation of a T2 histogram from a single cheese compared to the total data variation. Despite these effects main tendency could be pointed out. To identify relationships between the principal components and the T2 histogram data, both the correlation coefficient and the eigenvectors were calculated and plotted as a function of the T2 values (Figure 6).

Figure 6 Correlation coeficient (-) and eigenvectors (a) of the three first principal components The most important variables in PC 1 were the small T2 values around 20 ms and the high T2 values around 70 ms. The small TZ value presented a negative eigenvector, while the high T2 values showed a positive eigenvector. So PC 1 opposed the cheese characterised by small T2 to those with high T2 value. The second principal component, PC 2 which represented 11 % of the total variation was mainly correlated (r2 = 0.45) to the T2 value centred at 50 ms. In the same way the third principal component, PC 3 explained 9 % of the total variation and was positively correlated to the T2 value centred at 40 ms and negatively correlated to the T2 value


29

centred at 55 ms. Compare to PC 1 which explained large T2 variation between cheese T2 histogram, PC 2 and PC 3 explained small T2 changes. The representation of each cheese T2 histogram on the factorial 3D map is given in Figure 7.

pcl

I

6 FCl

Figure 7 Principal components analysis Figure 8 Discriminant analysis of T2 of T2 histograms histogram after variable number reduction by PCA From the three first PC it should be possible to distinguish some cheese groups associated to the T2 distribution. For example the both MESO I ,young and old, the young MESO 11, the UF and the T H E M 0 cheeses. While MESO I11 and old MESO I1 cheese T2 histograms seemed more confhsed with the others. Note also the large individual dispersion which should be related to the difficulty to keep a constant cheese chemical composition for all the batches. However the PCA subspace representation showed that a good discrimination between cheese could be achieved. A discriminant analysis was carried out. Because of the large number of T2 variable, 100 T2 variables, compare to the 48 cheese MR images, the discriminate analysis was performed from the reduced variables obtained by the PCA method. This dimension reduction increases the discriminant analyse reliability . So the first ten principal components were selected. Learning sets and test sets composed of one cheese T2 histogram per cheese were built and the discriminant analysis was applied several times. From this method the misclassification rate estimated from the learning sets was very low, around 10 %. The Figure 8 represents the factorial map composed by the two first discriminant components, DC 1 and DC 2. Some cheeses were well separated such as UF, MESO 11, MESO IV and THERMO. While, MESO I and MESO I11 were disperse and were not well discriminated from the previous ones. The misclassification rate estimated from the test sets increased to 33 %. This effect was consistent with the PCA results obtained. The T2 values of the cheeses were very sensitive to the batch chosen. The difficulty to obtain a homogeneous cheese population resulted from chemical composition and ripening effects. These two effects will induce large internal structure modifications which have a great effect on the relaxometric behaviour of the cheese protons dynamic. Therefore the discrimination is very efficient on the learning sets but become less efficient on test sets.

30


3.2 Opening Data Analysis Compared to hard cheeses such as Emmental which are characterised by large and regular openings, soft cheese openings could offer a wide distribution of size and shape. Figure 9 shows the frequency distribution of the size and compacity of openings for one old MESO I1 cheese and for one MESO 111 cheese (the compactity is given only for the openings with a surface greater than 4 pixels i.e 1.96 mm', value under which this measure is not reliable). The old MESO I1 cheese has many small openings with regular shape (high value for compacity) while the surfaces for the MESO 111are more widely spread as well as the compacities. MESO 111 Compacitiesdistribution 0.3 7

MESO I1 old Compacitiesdistribution

1

1 2 3 4 5 6 7 8 91011 Compacity * 10

MESO 111 surfaces distribution 0.4 I

MESO II old surfaces distribution

I

1 3 5 7 9 11 13 15 17 1921 surfacen (mm'mm)

1 3 5 7 9 11 13 15 17 1921 surfacel2 (mmlmm)

Figure 9 Compacities and surfaces distributionfor old MESO II and MESO III cheese openings The opening data analysis was performed on the calculated parameters (mean, max, min, standard deviation, variance, kurtosis and skewness values) from the different histograms (opening surface, opening perimeter, opening shape and opening distance). In order to prevent any distortion of the PCA all the UF cheeses were taken off from the data base because no openings were observed into these cheeses. The two first principal components PC 1 and PC 2 explained 54 % of the total variation. These two components were correlated with the perimeter standard deviation ? =0.94,the distance standard deviation =0.78 and the surface standard deviation r2 =0.73 variables. This implies that 54 % of the total variation of the total data base are attributed to the distribution of the opening characteristics. The representation of the cheeses openings parameters on the 2D factorial map built from PC 1 and PC 2 were very disperse (Figure 10). No data organisation as a function of cheese structure was observed.


31

Figure 10 Principal components analysis of opening parameters However some cheese samples were well correlated to this 2D factorial map. For example, some young MESO I and 11, MESO IV were correlated to PC 1 (?>0.6) and were characterized by large openinf size irregularities. Old MESO and mainly THERMO cheeses were correlated (r >0.7) with PC 2 which explained the distance standard deviation. The specific distance standard deviation of the THERMO openings could be explained by the low opening number observed into these cheeses. The lower the number of openings, the more the distance between them becomes sensitive to the opening position and so the standard deviation increases. On the other hand a cheese characterised by a large opening density should have distance parameters insensitive to the opening position. For the other variables no clear features have been observed. A discriminant analysis was performed from the first six principal components which explained 91 % of the total data variation. The misclassification rate obtained from the learning sets was 24 %. This value increased to 36 YOfor the test sets. These results indicate that the opening characteristic could not be considered as discriminating variable for the cheese structure. 3.3 Texture Data Analysis

A PCA was performed and the correlation and the eigenvectors for the first principal components as a function of the texture variables are given in Figure 11. PC 1 which explained 54 % of the total variability, was highly correlated with the textural features. The other components PC 2 and PC 3 were correlated with the grey level histogram parameters such as skewness (?=0.8 l), Kurtosis (?=0.62), minimum ( ?=0.75) of the T2 weighted grey values and mean proton density grey values (?=O .63).


32

1 2

3 4 5 6 7 8 9 1011 1213 Textural features

1 oPD weighted

HT2 weighted .mag

susc weighted

Figure 11 Correlation between texture variables and PC 1 for each MR image contrast. The numbers refer to Table I

image

1 Figure 12 Correlation parameter extracted @om the Co-occurence matrice of the PD and magnetic susceptibility image

Among all the textural features, some of them had correlation coefficient independent of the MR contrast, for example the angular moment or energy (l), the contrast (2) and the entropy (9). These textural features were strongly correlated with each other suggesting that the information provided by these parameters was independent of the MR contrast. On the other hand correlation coefficient of the textural correlation (3), variance (4) and sum average ( 6 ) textural parameters were dependent on the MR contrast. The major effect was observed between the correlation textural parameter calculated from the proton density and the magnetic susceptibility for which we observed the larger difference. The effect was explained by the UF cheeses (Figure 12). For this technology the linear grey level dependencies were higher for the magnetic susceptibility contrast images compare to proton density images. This observation was related to the signal to noise ratio which was higher for magnetic susceptibility contrast images. Indeed the UF images were highly homogeneous because they were free of openings. Consequently the grey level dependencies were strongly dependent to the signal to noise ratio. More the signal to noise ratio increase more the correlation textural parameters increase.

Figure 13 Principal components analysis of textural features

Figure 14 Discriminant analysis of texturalfeatures after variable number reduction by PCA


33

The study of the correlation between the cheese textural features and the principal components showed that PC 1 ordered the cheese according to texture regularity. PC 1 set the energy and the inverse different moment against the sum variance and the entropy. More the texture is regular ( high value of energy) more the cheese will be negatively positioned on the this principal component. On the map composed by PC 1 and PC 2 (Figure 13), the UF cheeses were highly distinguished from the other cheeses. In opposite the MESO IV had the most irregular texture. The other cheese technology were distributed along the first axes. The young MESO I were correlated with PC 2 because of a high T2 kurtosis value which is sensitive to the flattening of the grey level histogram. The third principal component separated The old MESO I and the THERMO cheeses from the young MESO I1 according to their mean grey level values obtained on the proton density images. In the same way more minor components, such as the fifth one which opposed the old MESO I from the others had to be take into account. To perform the discriminant analysis, the ten first principal components were retained (Figure 14). They explained 96 % of the total data variability. From the learning and test set the rate of misclassification were nearly null. 100% of the cheeses were well classified from the learning set and a rate of 92 % was obtained from the test set. Those results compared to the discriminant analysis from T2 histogram or opening parameters were very good, knowing that the discriminant analysis could also separate the cheese as a function of their ripening age. 4 DISCUSSION AND CONCLUSION The MRI data presented in this paper could be divided into three groups : micro scale data, macro scale and localised data, macro scale and global data. The micro-scale data group is constituted by relaxometric data. Indeed MRI T2 values which are of course sensitive to water and fat amount, are also sensitive to protein structure through protein-water interactions. During ripening the proteolyse modifies the gel network structure and the microbiological activities change the pH and ionic strength gradient. These two phenomena act on the water relaxation through diffusive and proton exchange evolutions. Consequently the T2 cheese distribution are function of the cheese composition, the cheese structure and the ripening stage. Thus the difficulty to control those effects from batch to batch explained the misclassification rate of the discriminant analysis. This misclassification rate should be reduced by increasing the sample number. The micro scale and localised data are constituted by the opening information. The open texture in cheese is the result of gas production from microbiological activities on the one hand and from mechanical curd handling practices during the cheese manufacturing process on the other. The opening number and shape are out of control from batch to batch. Except for UF cheese which were free of openings the other cheese presented too much intra group variation of the opening parameters compared to the inter group variation. The sensitivity of the computed parameters and the small sample number contributed to increase the misclassification rate of discrimination. Textural features could be considered as macro scale and global data because of the statistical approach retained. The co-occurrence matrix method appeared to be


34

very powerful. In fact this method integrates much information provided by the MR images. The opening and all the curd heterogeneity were took into account. But the effect of small image variation changes were ignored compared to the opening analysis. Moreover because we used a multi contrast strategy, the T2 effect were also analysed from the Tz weighted images. Thus the co-occurrence method properties provide the lowest rate of misclassification. This work was financed by a share cost contract no 96-1056 DG12 within the FAIR program TIROS of the EC. We thank J.M. Soulie and P. Fortier from Bongrain company and P. Marty Mahe from Cemagref for helpful discussions and advices.

References 1. P. Walstra and P. Peleg, 'Rheological and Fracture Properties of Cheese', Bull. Int. Dairy Fed. 1991,268,3 2. S.J. Schmidt and H. M. Lai, 'Water Relationships in Food', E. H. Levine and L. Slade, Plenum press, New York, 1991,405 3. M.J. Mc Carthy, 'Magnetic Resonance Imaging in Foods' Chapman & Hall, New York, 1994 4. M. Rosemberg, M.J. Mc Carthy, R. Kauten, Food Struct. , 1991, 10, 185 5. M. Rosemberg, M.J. Mc Carthy, R. Kauten, J. Dairy Sci. , 1992,75,2081. 6. M. Ozilgen and R. Kauten, Process Biochem. , 1994,29, 373. 7. R. Ruan, K. Chang, P. L. Chen, R. G. Fulcher, E. D. Bastian, J. Dairy Sci., 1998, 80, 9 8. S Kim, M.J. McCarthy and P.Chen, J. Magnetic Resonance Analysis, 1996, 2 (4), 281. 9. K.J Khiani, S.M. Yamany and A.A. Farag, Proceedings from the ANNIE96, St.Louis, November 1996. 10. B.B.B. Khoo, M.P.C. McQueen and W.J. Sandle, J. Biorned. Eng. 1991,13, November, 489 1 1. J.A. Throop, D.J. Aneshansley and B.L. Upchurch, Proceedings of the ASAE International Winter Meeting, Atlanta, December 1994. ASAE Paper 946580. 12. S.A. Shearer and R.G. Holmes, Transactions of the ASAE, 1990,33 (6), 2037 13. J.S. Weszka, C.R. Dyer and A. Rosenfeld, IEEE Transactions on Sytems, Man and Cybernetics, 1976, SMC-6 ( 4 ), 269. 14. P.P. Ohanian and R.C. Dubes, Pattern Recognition, 1992,25 (8), 819 15. R.M. Haralick, K. Shanmugam and I. Dinstein, IEEE Transactions on sytems, Man and Cybernetics, 1973, SMC-3 ( 6), 610. 16. R.W. Conners and C.A. Harlow, IEEE Transactions on Pattern Analysis and Intelligence, 1980, 2 (3), 204 Machine 17. M. Nagao and T.Matsuyama, Computer Graphics Image Processing, 1979, 9, 394

Protein Aggregation Studies Using PFG NMR Diffusion Measurements William S. Price, Fumihiko Tsuchiya and Yoji Arata WATER RESEARCH INSTITUTE, SENGEN 2-1-6, TSUKUBA, IBARAKI 305, JAPAN

1 INTRODUCTION The propensity of lysozyme to aggregate is well-known. The aggregation and crystallisation behaviours of lysozyme are closely linked as it has been reported that the critical nucleus most likely consists of four monomers and the growth unit is probably not the monomer but is more likely to be the octamer. The aggregation process has a complex dependence on pH, temperature and the protein and salt concentrations. This complex behaviour results from intermolecular forces since proteins are both colloids and polymers.’ Lysozyme has an isoelectric point of pH 11 and thus at most pHs has a net positive charge. In low ionic strength solutions, lysozyme interacts mainly through a combination of electrostatic repulsion and attractive dispersion forces.2 Addition of salt lowers the electrostatic barriers thereby achieving supersaturation of the protein at lower protein concentrations. In the present study the solution and aggregation properties of lysozyme at different pH, temperature, protein and salt (i.e., ionic strength) concentrations were studied using the self-diffusion coefficient, D, obtained from pulsed field gradient (PFG) NMR measurement^.^" D has a very direct correlation with molecular weight. However, care must be taken in relating the measured diffusion coefficients with protein aggregation and in this article we discuss the data interpretation in some detail. Some of the experimental results are compared to those obtained from crystal growth rate data by Li et al?’

2 PFG MEASUREMENTS OF AGGREGATION The basic scheme for studying protein aggregation using PFG NMR diffusion measurements is depicted in Figure 1 and the individual steps are outlined in the following subsections. 2.1 Protein Shape and Diffusion

To a first approximation we can take monomeric lysozyme to be spherical, in which case, D can be related to the Stokes radius, Ro, and the solvent viscosity, 7,via the Stokes-Einstein relation,

36


Ensemble Averaging , ~

Figure 1 Schematic description of the steps involved in studying protein aggregation using PFG NMR diffusion measurements. A cogent model for the aggregate distribution must be used in calculating the apparent diffusion coefjcient.

It is important to realise that Eq. (1) only applies to a dilute solution such that the aggregating species diffuse independently of each other. The Stokes radius, which is the effective hydrodynamic radius, is generally larger than the radius derived from the molecular structure itself. The concept of a hydration shell is often invoked to account for this discrepancy. Although the difference may also be related to the rugosity of the protein surface. In reality lysozyme is not exactly spherical, nevertheless taking it to be spherical for the point of view of modelling the diffusion data is a reasonable approximation. The next problem is how to model the hydrodynamic characteristics of the different oligomeric states. The simplest way is to approximate the monomer-monomer contact as hard-sphere contact,8x9in which case the ratio of the diffusion coefficient of an i-mer, D,, to that of the monomer, D,=Ican be modelled by


37

and the values of Fifor various geometries are given by Teller et a1.' Whilst there is only one possible geometry for dimer formation, many possibilities exist for higher oligomers. Consequently, we have simplistically taken all oligomers to be hydrodynamically spherical, thus the friction coefficient (i.e., the denominator of Eq. (1)) increases according to the inverse cube root of the molecular weight. In fact the friction coefficients calculated from Eq. ( 2 ) for reasonable geometrical possibilities for the oligomeric shapes are all quite close to that obtained for a sphere of equivalent volume. 2.2 Crowding Effects on Diffusion

The previous section considered the hydrodynamics of lysozyme at infinite dilution (i.e., no interaction between the diffusing aggregating species). In reality, aggregating lysozyme samples are crowded systems and consideration must be given to the effect of such crowding on the diffusion coefficient, irrespective of any aggregation process since both processes lead to a reduction in the measured diffusion coefficient. Crowding is a complicated many-body problem and at present only approximate means of estimating the effects of crowding on the diffusion coefficient exist. A simple model to account for the effects of crowding based on scaled particle theory is given by"

D= ~~exp(-%c)

(3)

where

In Eq. (3) DO is the true (uncrowded) difhsion coefficient, v p is the volume fraction (mllg) of the protein and Ar is the step size and R is the radius of the diffusing particle. From the Smoluchowski equation ArlR = 213. This model is based on all of the protein present existing in the monomeric state. Consequently as the degree of aggregation increases, this simple model overestimates the reduction in diffusion."

2.3 PFG NMR Measurements of Diffusion A detailed description of the PFG NMR diffusion experiment can be found el~ewhere.~,'For a single diffusing species the equation relating the echo signal attenuation to the experimental variables is given by

In( E ) = -y 2 g 2D J 2( A - 6/3)

(5)

where y is the gyromagnetic ratio, g is the magnitude of the gradient, 6 is the width of the gradient pulses and A is the separation between the leading edges of the gradient pulses. Importantly, A defines the timescale over which the diffusion is measured. In a polydisperse system the attenuation curve is more complicated than that described by Eq. (5). If the species are in slow exchange, E should be multi-exponential. If the exchange rate is intermediate the attenuation curve is further complicated. While in the fast exchange rate the attenuation will again be single exponential (i.e., similar to Eq.

38


(5)) but with D representing a mass averaged diffusion coefficient (see Eq. (7)). However, all of the experimental data was well described by a single apparent difhsion coefficient even under the conditions most likely to produce aggregation. Thus, it is probably not unreasonable to assume that the exchange is slow on the time-scale of the PFG difhsion experiment. Since an aggregating protein solution is crowded it is likely that some ensemble averaging process occurs leading to a narrower distribution of diffusion coefficients. This averaging probably results from collisions and crowding since the mean free displacement is greater than the average spacing between the lysozyme molecules. We allow for this by taking the cumulant expansion (to 2"d order), ln(E) = -b(D)w + -b2 ((D): -(D2)w)7 2 where b = -y2g262(A- 6/3) and the mass averaged diffusion coefficient is defined by

where M, is the molar mass and ni is the number of the i-th aggregating species. In fact for the values of b used in the present work we can safely omit the quadratic term in Eq. (6) and thereby recover the single exponential form of Eq. (5). Equation (7) can be used to simulate the observed diffusion coefficient given a model for the oligomer distribution and an estimate of the Di.

3 RESULTS AND DISCUSSION We used 'H PFG NMR measurements to obtain the difhsion coefficients of three Iysozyme concentrations, 1S,2.8 and 10 mM at various temperatures. The experiments were performed using a Bruker DMX 500 NMR spectrometer equipped with a triple axis gradient probe (only the z-axis gradient was used). The 1.5 mM lysozyme samples were studied at pH 3, 5 and 8 and in the presence of 0, 0.15 and 0.5 M NaC1. The 1.5 mM lysozyme samples were used to study the onset of aggregation and only at the highest pH and salt concentrations were the solutions saturated at this low protein concentration. Measurements were also conducted on a lysozyme sample containing 2.8 mM lysozyme at pH 4.6 in the presence of 0.15 M salt which results in a saturated sample. These sample conditions were chosen so as to correspond with those used by Li et aL6.' who used crystal growth measurements to study the aggregation process. Finally a sample containing 10 mM lysozyme in the presence of 0.15 M salt at various pH values was used to study lysozyme in a supersaturated solution. Theoretical calculations of the lysozyme monomer diffusion coefficient were performed using the three dimensional structure of lysozyme" assuming that the backbone atoms were of equal size (0= 5.0 A) with the program DIFFCI2 which is based on the bead model appro~imation.~


39

3.1 The Onset of Aggregation (1.5 mM Lysozyme) The results of the diffusion measurements for the 1.5 mM lysozyme samples are shown in Figure 2. In the absence of salt there is no obvious pH-dependence consistent with there being no aggregation. The theoretical predictions overestimate the monomer diffusion coefficient. However, if the effect of crowding on the theoretical diffusion coefficient is considered, the agreement with experiment becomes better. In the presence of 0.15 M NaCl the situation is rather different and there is a clear pH-dependence of the measured diffusion coefficients. The lower pH samples have higher apparent diffusion coefficients consistent with monomeric lysozyme being the dominant form. At low pH in the absence of salt, the lysozyme molecules are repulsive towards one another and thus they effectively exclude other molecules from diffusing in their neighbourhood (i.e., the self-obstruction effect is increased due to electrostatic interactions), as the ionic strength increases this effect is decreased and thus the difhsion coefficient increases (i.e., compare the results for the 0 and 0.15 M NaCl samples). In 0.5 M NaCl aggregation is possible and D decreases. Interestingly at the highest salt concentration and lowest temperature (i.e., 0.5 M and 283 K) the diffusion becomes less pH sensitive. A likely reason is that at this temperature the conditions are such that aggregation is now relatively similar at all three of the measured pH values.

... ... ........ ..+' ...

a:. . ..

i-....'-. . ..

E?.A

"+:I.

... ..

_ .. . ..' _ . .

....'.. .

'+ :....._...

3.20

3.25

3.30

3.35

3.40

3.45

3.50

0.15

M

3.55

1 O O O R (K-')

Figure 2 Diffusion of 1.5 mM lysozyme at pH 3.0 (square), 5.0 (circle) and 8.0 (triangle) centre) and 0.5 (solid centre) M versus temperature containing.0 (open centre), 0.15 NaC1. The 0.15 and 0.5 Mdata have been offset in the temperature axis (i.e., away from the vertical lines). The theoretically derived monomer diffusion coefficients are also shown (dotted line) and corrected for the effects of crowding (dotted line with plus symbols). ('-I

40


3.2 Diffusion in Saturated Solution (2.8 mM Lysozyme)

The results of the diffusion measurements for the 2.8 mM lysozyme sample are shown in Figure 3. Due to the higher protein concentration the crowding effects are more significant. After accounting for the effects of crowding the experimental difhsion coefficients for 2.8 mM lysozyme at pH 4.6 and 0.15 M NaCl indicate that there is some degree of aggregation although the average molecular weight is clearly less than that of a dimer. The crystal growth data of Li et a16,’ appear to underestimate the degree of association at high temperatures but overestimate it at low temperatures.

...

Figure 3 Diffusion of lysozyme at p H 4.6for a sample containing 2.8 mM lysozyme and 0.15 M NaCl (closed squares). Since this is very similar to the conditions as used by Li et al.,62 the apparent diffusion coefficients (closed circles) were calculated from the theoretical monomer diffusion coefficient using the association parameters of Li et a1 and assuming that the higher oligomers were spherical. The theoretical values for the monomer diffusion coefficient at each temperature are denoted by the dotted line and those corrected f o r crowding are denoted by the dotted line with plus symbols


41

3.3 Diffusion in Supersaturated Solution (10.0 m M Lysozyme)

To observe aggregation in supersaturated protein solutions, samples were prepared containing 10 mM lysozyme in 0.15 M NaCl at various pH values and the results of the diffusion measurements are shown in Figure 4. The effects of crowding at this protein concentration are very pronounced and, after allowing for the effects of crowding; the diffusion coefficient of a monomer is reduced to that of a (uncrowded) dimer. At this protein concentration it is found that only at the lowest pH and highest temperature does the observed diffusion coefficient approach that of a (crowded) monomer. The .diffusion coefficient decreases significantly with increasing pH, consistent with high degree of aggregation. Large changes are also observed in the corresponding 'H NMR spectra (not shown). This is consistent with large reductions in the rotational correlation time as would be expected with the formation of higher oligomers.

0.2 0.0 0.4

-

n

"E

-0.2

-

-0.4

-

W

2-

2 Q

-c

W

-0.8 -0.6

-1.0

1

3.20

~

3.25

1

3.30

~

3.35

1

3.40

I

3.45

~

I

3.50

,

I ]

3.55

1000/T (K-') Figure 4 Change in the diffusion coeficient of lysozyme with p H @H 3.2: up triangle; pH 4.0: down triangle; p H 5.0: circle and p H 6.2: diamond) for a sample containing I 0 mM lysozyme and 0.15 M NaCl versus temperature. The calculated diffusion coeficients for the lysozyme monomer (dotted line), dimer (dashed line) and octamer (solid line) uncorrected for crowding are also shown. The monomer diffusion coeficient corrected for crowding is denoted by the plus symbols connected by the dotted line (NB this is almost perfectly coincident with the dimer dflusion coeficient which was derived from the uncorrected monomer diffusion coeficient).

~

l

42


4 CONCLUSIONS The results of our diffusion measurements on lysozyme show that the aggregation state and the interactions between the protein molecules are very sensitive to the experimental conditions (i.e., protein and salt concentrations, pH, and temperature). Even in the most supersaturated lysozyme solution studied (i.e., 10 mM) it can be inferred from the observed difision coefficient that the average oligomer is less than that of an octamer. PFG NMR diffusion measurements are a powerful technique for studying the solution behaviour of proteins since the method can be used at high protein concentrations in contradistinction to many traditional techniques. Further, compared to NMR relaxation measurements, the interpretation of the diffusion data is more straightforward. Nevertheless, to properly interpret the diffusion data factors leading to non-ideal solution behaviour such as crowding and electrostatic effects need to be considered. A major problem in this respect is that there are as yet only very approximate models for accounting for the effects of crowding.

5 ACKNOWLEDGMENTS Dr. Bertil Halle (University of Lund, Sweden) and Dr. V. V. Krishnan (Lawrence Livermore National Laboratory, USA) are thanked for useful discussions and performing the theoretical calculations of lysozyme diffusion, respectively.

References

L. R. De Young, A. L. Fink and K. A. Dill, Acc. Chem. Rex, 1993,26,614. D. E. Kuehner, H. W. Blanch and J. M. Prausnitz, Fluid Phase Equilibria, 1996,116, 140. 3 W. S. Price, in Annual Reports on NMR Spectroscopy; ed. G.A. Webb, Academic Press, London, 1996, p. 5 1 . 4 W. S. Price, Concepts Magn. Reson., 1997,9,299. 5 W. S. Price, Concepts Magn. Reson., 1998,10, 197. 6 M. Li, A. Nadarajah and M. L. Pusey, J. Cryst. Growth, 1995,156,121. 7 A. Nadarajah, M. Li and M. L. Pusey, Acta Cryst., 1997, D53, 524. 8 D. C. Teller, E. Swanson and C. De Haen, Methods Enzymol., 1979,61, 103. 9 J. Garcia de la Torre and V. A. Bloomfield, Q. Rev. Biophys., 1981, 14, 81. 10 J. Han and J. Herzfeld, Biophys. J., 1993,65, 1155. 1 1 L. J. Smith, M. J. Sutcliffe, C. Redfield and C. M. Dobson, J. Mol. Biol., 1993, 229, 930. 12 V. Yu. Orekhov, D. E. Nolde, A. P. Golovanov, D. M. Korzhnev and A. S. Arseniev, Appl. Magn. Reson., 1995,9,581.

1 2

Water, Ions and Small Molecules in Food

A Multistate Theory of Water Relations in Biopolymer Systems B. P. Hills, C. E. Manning and J. Godward INSTITUTE OF FOOD RESEARCH, NORWICH RESEARCH PARK, COLNEY, NORWICH NR4 7UA, UK

1. INTRODUCTION

N M R and MRI techniques now exist for monitoring water mobility over distance scales ranging fiom the molecular to the macroscopic. However, there is still much that needs to be learnt before dynamic information on the molecular distance scale can be used to predict the transport behaviour of water on the microscopic and macroscopic distance scales in complex, heterogeneous food systems. In this paper an attempt is made to show how, beginning with a new multistate theory of water relationships on the molecular distance scale, transport behaviour on larger distance scales can be understood. Such understanding is essential if we are to predict the effects of biopolymer engineering and novel processing operations on the quality of manufacured foods. 2. THE STATE OF WATER IN SINGLE COMPONENT BIOPOLYMER SYSTEMS 2.1 The multistate formalism

Recent multinuclear N M R experiments and molecular dynamics calculations have clearly demonstrated that it is usually sufficient to consider three stqtes of water in biopolymer systems and biological tissue. The first could be called “~tructural” or strongly interacting water hydrogen bonded inside the cavities and gropves of globular proteins and polysaccharides and which plays an important role in determining the structure and dynamics of the biopolymer chains. There is a spectrum of lifetimes of this “structural” water ranging from nanoseconds to microseconds depending on the nature of the water-biopolymer interaction. In addition there is “surface” or “m$tilayer” water, which is water at the biopolymer surface having a dynamic state pgrturbed by the presence of the interface. This surface water extends for several molecular layers from the surface and is extremely mobile, having exchange lifetimes on the order of subnanoseconds. Bulk water comprises the third state. In this section we take this simple molecular model and try to elucidate its implications for the dependence of readily measurable quantities such as water activity, G, NMR water relaxation rates, y, and water diffusivity, D, on water content and composition in single and multicomponent biopolymer mixtures,

46


The starting point for the model is the set of equations

yav =

ci xiyi

These equations give the observed “average” value as the weighted average of the value over all states of water, where xi is the mol fraction of water in the state i. The first equation is a consequence of the Ergodic theorem of equilibrium thermodynamics which states that the time averaged property of an individual water molecule as it diffuses between the various states in the system equals the time-independent ensemble average over the system, so that a,

=

limt,,

I/t I,”“ dt a(t)

= Xi Xiai

[41

Equation [2] is the well-known fast exchange limit equation in Nh4R relaxometry which was originally derived by Zimerman and Britten It remains valid provided there is fast exchange of all water proton pools on the N M R measurement timescale (typically a few 100’s of microseconds), so that the water proton relaxation is single exponential. It also neglects the dephasing effects of chemical shift offsets between the pools, so that proton exchange between biopolymer and water protons is, for the moment, neglected.

’.

2.2 The sorption isotherm for a single component biopolymer system

We first show how the well-known “sigmoidal” shape of a sorption isotherm for a simple one-component biopolymer-water system emerges from equation [ 11. Acknowledging the three states of water explicitly, equation [11 becomes,

structural

multilayer

bulk

where we have assumed there are n states of the structural water in the biopolymer. If mi is the mass of the ith state of water per gram of dry biopolymer, then xi = m,/W where W is the total mass of water per gram of dry biopolymer. To proceed to a sorption isotherm, we assume, for simplicity, that the order of ‘Yilling” of the various states as water molecules are added to a dry biopolymer is first the structural states, then the multilayer state and finally bulk water. Of course, a more rigorous theoretical approach would be to incorporate a Boltman distribution over the energy levels characterizing each state of the water, but for present purposes this complexity is left for a later generalization.


41

2.2.I The structural states

These are occupied at very low water contents and, because of the limited number of such sites, also over a narrow range of water content, W, typically less than 15-20%. Because there are many different structural sites in any particular biopolymer, such as a globular protein it is probably unwise to try to elucidate the detailed form of the isotherm in this region, but merely leave the general form,

where a, is the stuctural water activity and m;(max) is the mass of water when the structural state i is saturated. The shape of the sorption isotherm in this very low water content region will clearly depend on the biochemical details of the number and binding energies of the hydration sites in the biopolymer. 2.2.2 The multilayer state

When all the structural sites are occupied, further addition of water creates the “multilayer” state and in the water content range for which there is only structural water and multilayer water, equation [11 becomes

which is equivalent to the sorption isotherm,

where the constant C, is

2.2.3 The bulk water state

When all structural and multilayer sites are occupied, it is assumed that further addition of water causes formation of the bulk state, for which,

which can be rewritten in the form, aav=

+ Cdw

for m,(max) I WI m

where the constant cb is defined as

48


Unlike equation [6] describing the structured water region, equations [8] and [12] have the same dependence on W’, which no doubt explains the frequently observed sigmoidal shape of sorption isotherms. The equations also suggest that to identifjr the states and their water content ranges, the data should be plotted as water activity verses IN, or alternatively, against the solid content, (I-W,)N,, where W, is now the mass of water per total mass of water plus biopolyrner. In this way the constants C ,, Cb can be determined and the populations of the various states can be identified from breaks in the slope. Figure 1 shows an example for the adsorption and desorption isotherms of pregelled potato starch. The desorption isotherm shows breaks at water activities of about 0.9 and 0.3, indicating that structural water is characterized by water activities less than 0.3 while multilayer water exists in the activity range between 0.3 and ca. 0.9. The small additional break in the adsorption plot at water activities close to unity indicates that adsorption leads to greater amounts of essentially bulk water at low solid contents. Figure 2 shows the more conventional sorption and desorption isotherm plots , together with the theoretical fits. Because of our ignorance about the number and states of structural water in the pregelled starch no attempt has been made to fit the very low water content regime where only structural water exists and equation [6] should apply.

Figure I . i%e multistate theory for the desorption and adsorption isotherms of pregelledpotato starch. The straight lines are best fits of equations [8] and (111 expressed in terms of solid content.

49


Pm-gelled potato starch

0

0.4

~

adsorpaon desorphontheoly

-adsorption

02

theory I

0

20

60

40

80

100

WY Figure 2. The adsorption and &sorption isothermsfor pregelled potato starch based on the linearfits infigure 1. . 2.3 Water relaxation in a single component biopolymer system

The theory for the dependence of the single exponential water relaxation rate on water content can be developed in complete analogy with equations [5] to [12]. It follows that,

where the constant B, is

where the constant Bb is defined as

These equations neglect the effects of proton exchange and secular dipolar cross relaxation and focus exclusively on the states of water. Figure 3 shows the water proton transverse relaxation rate for the adsorption and desorption isotherms of pre-gelled potato starch corresponding to figures 1 and 2. Note how the desorption plot shows a break at the same water content as the water activity plot in figure 1.


50

Pre gelled potato starch 3000

2500

i

2000

-

In \

1

1500

N

K

1000

0 0 0

abosrbR2 desorbR2max

-des reg R2 max

500

-des

0

5

10

15

reg R2max

20

(1-WY)MfY

Figure 3. The dependence of the CPMG water proton transverse relaxation rate (measured at a short p l s e spacing of 200.11.9agaimt solid content.

The proton exchange contribution can vary with water content so it is undoubtedly better to test the theory with the proton dewupled water oxygen-17 relaxation rate or in situations where proton exchange is known to be negligible such as all glassy states of the biopolymer and also in the low water content regimes lacking bulk water. This has been done on model Sephadex and silica systems4but not yet in biopolymer systems.

2.4 The relationship between water activity and NMR relaxation in the dilute regime Equations [ 111 and [ 151 imply that, over the same range of water contents there is a simple linear relationship between the water activity and the water relaxation rate. In dimensionlessform, this relationship takes the form,

Figure 4 shows the linear dependence predicted by equation [16] for the adsorption isotherm of pregelled potato starch. The break at a water activity of ca. 0.9 corresponds

51


pre gelled potato starch adsorption 1800 T

/

1600

1200

2 m :.Iloo0

"1

/

600

/

"d reg wet __t_l

0

02

0.4

0.6

0.8

1

1-aw

Figure 4. The linear relationship between the water relaxation rate and (I-aw) for pregelledpotarostarchpredicted by equation [I61 to the bulk water break in the adsorption plot in figure 1. The oxygen-17 data in reference (5) also lends support to the simple form of this equation. It should, however, be remembered that factors such as proton exchange, that can also contribute to water relaxation and that these may show changes at water contents independently of water activity. For this reason, linear activity-relaxation relationships should only be expected over limited ranges of water content, temperature and pH. Moreover, there is no implied fimdamentaJ physical relationship between water activity, which is an equilibrium thermodynamic quantity, and NMR relaxation, which is in essence a non-equilibrium, kinetic phenomena. AU that is implied is that the changing states of water in a biopolymer system will affect both activity and relaxation in parallel ways. 2.5 FID amplitudes for a single component biopolymer system

At low water contents, where structural water is of special importance, relaxation times become too short for measurement by the CPMG method and relaxation times need to be extracted from the FID itself One complication in doing this is the appearance of fast decaying Gaussian components from the biopolymer itself. While these are of interest in their own right, especially for studies of the mechanics of plasticization, they complicate the analysis of the various states of water. One way to circumvent this complication is to focus on a single time point, say t., in the FID, chosen so that all solid biopolyrner signal has decayed to zero, leaving only the mobile water component. The FID amplitude at t, is then,


52

amplitude 1.25

1 .oo

0 0000 00000 000 000 0 00 0

0.75 MO 0.50 0.25

xxxxxxxxxxxxxxxxxx

......

0 c

0 0.0

0.00 0

0.

o.oooo..

I

I

I

I

0.02

0.04

0.06

0.08

I

0.1 0.12 time (msec)

Fig. 5 : Normalized amplitude of the FID curvesfor gelatines observed at 20 MHz and 20°C at dflerent water contents ( 0 :4.93 ; :15.86 ;x :20.1 ; 9 :57.73%, wet basis). The extrapolated signal intensityfor the slow relaxing component, MO,is indicated as well as the total signal intensity at 11 p,FIDf1. Takenfrom reference 9.

-

where yav is given by the equations above. For example, taking equation [ 151, we deduce that In M(ta) = In A - Em

[181

where the constants A and E are given as A=M(0)eXp{-ybda} and E is mo{ ys-ybullr)fa. Figure 5, taken from reference 9, shows the proton FIDs for gelatine gels at various water contents and figure 6 shows that the FID amplitude does indeed have a very similar “sigmoidal” dependence on moisture content as the sorption isotherm itself In figure 6, the straight line segments on the FID amplitude plot as the water content increases correspond to addition of structural, multilayer and bulk water respectively. The derivation of equation [18] assumes that the mobile component comprises only water. It should, however be remembered that in many biopolymer systems there may also be contributions from mobile side chains that increase with increasing water content


53

1.8 1.6 1.4 1.2 1.0 0.8

0.6 0.4 0.2

Fig.6 : Initial 'H A&&? signal amplitude of the slow relaxing protons divided by the amplitude of the FID signal at I 1 psec ( 8, :MOsrJFID~J and sodium relaxation times ( : 23iVal / R 2 ) as a finction of the water content. A sudden change in slope is visible for both parameters in the 10-20% water content region. Takenfrom reference 9.

as they become increasingly plasticized. The use of water oxygen-17 relaxation would alleviate this complication. 2.6 Water diffusivity in a single component biopolymer system

Because the biopolymer self diffusion coefficient is negligible compared to that of bulk water it is usually safe to assume that the water self diffusion coefficients in the structural and multilayer states are also negligible, so we can write,

54


or

xi - x,,,. Of course, a fill calculation of the water self diffusion where mJW = coefficient must also take account of the well-known obstruction effects of the biopolymer on the bulk water translation. 3. THE STATE OF WATER IN MULTICOMPONENT BIOPOLYMER SYSTEMS 3.1 Sorption isotherms for multicomponent biopolymer/solute/water systems

Let the moisture content fraction of water associated with the jth biopolymer or solute component be x(j), then clearly X,x(i) = 1. Including the various water states, i, associated with each component, we can write, Cijxi(j)= 1. Except for the bulk phase, the activities will depend on both the state of the water (j) and the component, i, so

This assumes an “ideal” mixing, whereby the water is not preferentially associated with one or other component. In other words, if the system is made by mixing component (1) with water fraction x(1) with component (2) of water fraction x(2), then the corresponding water fractions in the mixture remain x( 1) and x(2). If this is not the case, we can define preference coefficients c(j) for each component and generalize equation [21] by writing

such that Zij c(i)xi(i) = 1. Using these equations it is now possible to derive sorption isotherms for both ideal and non-ideal multicomponent biopolymer/solute/watersystems. For notational simplicity we combine the structural and multilayer states into a single state “s” and specialize to the situation where some bulk water exists. The derivation for other cases follows along obvious lines. For the two-component case, equation [21] becomes

But -(I) becomes

=

%&(2) = 1 and using the normalization condition, the sorption isotherm

But for the sorption isotherms measured on the separate components at the same water contents,

Water. Ions and Small Molecules in Food

55

Equations [23] to [26] show that, for the case of ideal mixtures, for which the preference coefficients, Es(j), are all unity, a,(1,2) = a d 1 ) a 4 2 )

+ second order terms in &(1)&(2)

~ 7 1

which, to first order is the Ross equation for multicomponent sorption isotherms. However, in general, this is only a starting approximation and the fill equation [23] should be used in the non-ideal mixture. Clearly, an outstanding task is to acquire tables of the preference coefficients in analogy with activity coefficients. 3.2 Water relaxation and diffusion in multicomponent biopolymer/solute/water systems An analogous derivation shows that the water relaxation rate in a two component mixture, yav(1,2), is given as

where yb& is the relaxation rate in the bulk water state, usually pure water. For ideal mixing, equation [28] implies the validity of a Ross-type equation for relaxation,

The limits of validity of this relationship have yet to explored experimentally. In like manner, the corresponding relationships for water difisivity are

which, for ideal mixing, leads to the corresponding Ross relation: 1,2)/ Dbud = (DS(1)/ h u l k )(Ds(2)/ h u l k )

+ higher order terms

[3 11

3.3 The origin of hysteresis in sorption isotherms and relaxation state diagrams

Figure 2 shows a typical example of hysteresis during the adsorbtion and desorption of water in a single component biopolymer system. Hysteresis in sorption isotherms is often assumed to be caused by differing microscopic condensation patterns of water condensed in capillaries during adsorption and desorption. The idea being that surface tension effects cause a depression of vapour pressure and hence of water activity'. However simple order of magnitude calculations* serve to show that, in most systems, this is unlikely to be the whole explanantion. Indeed, the large differences in water activity observed in figure 5 at the same water content would require capillary pores of the order of a few water molecular diameters, when the very concept of surface tension breaks

56


down. The multistate theory provides a simple alternative explanation of hysteresis since it merely requires that the number and nature of the structural (and, possibly, multilayer) states of water differ during the adsorption and desorption processes. Any change in biopolymer conformation or state of aggregation induced by the adsorption and desorption cycles would therefore serve to alter the water activity. Indeed, in principle, the multistate theory provides the opportunity for quantitatively predicting the effect of engineering a biopolymer on its sorption isotherm.

A further consequence of the multistate theory is that sortion-desortion isotherms should also exist in measurements of the water relaxation rates and FID amplitudes. Figure 6 gives an example of this behaviour and confirms that the isotherm and relaxation loops have similar shapes.

I

I

Pre-gelled potato starch: relaxation hysteresis

0.8

0.7

20

I

40

60

wy %

80

100

~

I

Figure 7. Hysteresis in both the water proton transverse relaxation rate and the water activity during adsorption and desorption isotherms of pregelled potato starch

3.4 Ion solvation and the multistate theory

Most biopolymers, including proteins and important food polysaccharides, carry ionic groups and their properties therefore depend, often sensitively, on the nature of the counter ions. The question therefore arises as to the relationship between the various states of water and the dynamic properties of the counter ion. A priori one would expect that ionic mobility would be minimal in the dry biopolymer system when only structural


57

water exists, because there would be little or no water available for ionic solvation. Conversely, ionic mobility would be essentially that of the hlly hydrated aqueous ion when a bulk water state exists. It follows that there should be a transition zone in ionic mobility as the multilayer state of water is formed. To test this hypothesis, =Na NMR relaxation was used to monitor sodium ion mobility in a gelatine gel system over a wide range of water contents. The data, along with the normalized amplitudes of the mobile fraction in the proton FID at a t, of 170ps, is shown in figure 6 and supports the hypothesis In the structural state, corresponding to moisture contents of ca. 15% the sodium relaxation time is too short to measurement. As the multilayer states are occupied between moisture contents of ca. 15-20%, there is a rapid increase in the sodium transverse relaxation time and, as the bulk water state is formed at water contents above ca. 20%, there is a dramatic change of slope and a slower increase in sodium relaxation time corresponding to dilution of the now, fully hydrated, ions. Consistent with the multistate model, these sodium relaxation transition zones correspond very nicely with the changes predicted in the amplitude of the mobile fraction of the proton FID.

’.

4. GENERALIZATION TO MULTICOMPARTMENT BIOPOLYMER SYSTEMS

4.1 The dynamic state of water in multicompartment, multicomponent systems Up to this point it has been assumed that the biopolymer system is spatially homogeneous on all distance scales greater than the macromolecular. This assumption breaks down in phase-separated mixed biopolymer systems and in all biological tissue so it is necessary to consider how the multistate theory applies to such systems. Figure 8 shows a schematic of a two compartment system in which water can exchange, by diffusion, between compartments 1 and 2. Suppose that the inner compartment consists of a dense biopolymer network such as a water-saturated starch granule and the outer compartment is bulk water. The equilibrium water activity measured fiom the ratio of vapour pressures will, of course, be unity, because there is a bulk water phase. This does not, however, imply that the equilibrium water activity inside the biopolymer compartment is also unity, as is sometimes supposed. Indeed, it is manifestly less than unity because it comprises a rather concentrated biopolymer gel network. The resolution of this paradox must be the presence of additional terms in the expression for the water chemical potential inside the biopolymer network, most probably an osmotic pressure term, xV1, because the bulk water outside will exert a swelling pressure on the biopolymer compartment. We can therefore write,

where V, is the partial molar volume of water. This example serves to demonstrate that the water activity in a microscopically heterogeneous system can vary spatially. It should therefore be possible to once again invoke the Ergodic theorem and define an average water activity as the volume integral over all microphases in the ensemble:

a,

= fdv a(r) = Cka(k)V(k)

Advances in Mugnetic Resonance in Food Science

58

where the sum is taken over all compartments k of volume V(k). While this average value can be defined, it is clearly not the same as the overall water activity measured from the equilibrium vapour pressure and it is questionable whether this Ergodicallydefined water activity has any measureable significance. In contrast the analogous equation for water relaxation has obvious measureable significance as it is the average relaxation rate, yN , observed when diffusive exchange of water between all the compartments is fast on the NMR measurement timescale, so that the observed water proton relaxation is single exponential, and

Of course, if the diffusive exchange is very slow on the NMR measurement timescale then a multiple exponential relaxation is observed, each exponential component arising from one compartment. More complicated multiple exponential behaviour is observed when the exchange and relaxation timescales are comparable, in which case the BlochTorrey equations must be solved. Analogous expressions exist for water diffusivity and these hold provided diffusion is unrestricted by permeability barriers at the compartment interfaces. Then D,

=

Jdv D(r) =

ck D(k)V(k)

1351

If there are diffusive barriers, then the q-space formalism can be used to relate the NMR d i s i o n measurements to

0

mr

001

I

Figure 8, The dependence of the water proton transverse relacation time distribution (in seconds) for packed be& of native corn starch g r m l e s on water content. The dotted line shows the water-saturated bed Reducing the water content progressively increases the relative amplitude (%) of thefmter relaxing peak.


59

4.2 Application to native, granular starch and cellular tissue

Figure 8 shows the distribution of water proton transverse relaxation times for a bed of native (i.e.ungelatinized) corn starch granules at various water contents. Assigning the two major peaks is far from straightforward. Based on comparison with electrical conductivity data the author originally assigned the long relaxation time peak to water outside the starch granules and assumed its shift to lower relaxation times was a result of exchange with water inside the granules on an intermediate timescale’2. However this model fails to explain the q-space diffusion data and suggests that the peaks in figure 8 actually arise from water in internal domains within the starch granules. For example, the longer relaxation time component could be water inside the amorphous growth rings inside the granule; whereas the shorter component to water inside the semi-crystalline stacks. The observation of a third “shoulder“ in the case of potato (see figure 9) is consistent with the larger and more heady pitted nature of the potato starch granules. When water is replaced by DzO the starch proton relaxation time distribution also shows two peaks (figure lo), consistent with the more mobile nature of the amorphous region compared to the semi-crystalhe domains. Furthermore, the deuterium transverse relaxation time distribution is also multiple exponential, consistent with the proton data.

Figure 9. The dependence of the water proton transverse relaxation time distribution (in second) for packzd beds of native potato starch granules on water content. The dotted line shows the water-saturated bed Reducing the water content progressively increases the relative amplitude (%) of the fmter relaxingpeak-s.

60


Figure 10. The distribution of starch proton transverse relaxation times for a packed bed of corn starch granules saturated in D20.

The q-space diffusion data, acquired with the stimulated echo pulse sequence, is unusual in that it is dependent on the product q2A and yet can be fitted only with a twodimensional diffusion model (see reference 12 and figure 12). It would appear that, on the NMR disfhion timescale of milliseconds the water is confined to a s i o n inside the granules, most probably in the channels created by the amorphous growth rings.

Figure 11. The dependence of the relative amplitude of the stimulated echo on $A for a packzd bed of corn starch granules containing 35% water. Note the f i t of the 2dimensional d f i s i o n model.


61

In contrast to carrot, apple tissue displays three distinct water proton transverse relaxation times peaks whose position depends on diffusion of water between the vacuolar, cell wall and cytoplasmic compartments. In this case the effects of drying and freezing on this relaxation time distribution require solution of the Bloch-Torrey equations in a three-compartment system14. The details of the analysis can be found in reference (14). 4.3 Combining distance scales

Developing theoretical models which consistently incorporate transport and mobility data over all the distance scales accessible by NMR and MRI " remains an outstanding Cture challenge. Consider, for example, the drying of raw potato tissue. MRI has already been used to monitor moisture gradients set up in a piece of raw potato during drying and these were interpreted using a Fickian diffusion model involving an effect water dfision coefficient and ~hrinkage'~. However, at the microscopic distance scale it is the arrangement of cells and air-gaps in the tissue as well as membrane permeability barriers that determine the observed macroscopic diffusivity and shrinkage. As we have seen, NMR relaxometry and q-space microscopy can be used to probe water compartmentation and diffusion on this distance scale. What is lacking is the theoretical formalism for relating the macroscopic parameters to microstructure. One possibility is to treat the cellular structure of the tissue with some lattice or tessellation algorithm and include subcellular compartmentation in the model by incorporating a two- or threecompartment cell model 14,17. In principle tissue drying or freezing could then be analysed by the removal of subcelluar compartments and, at the tissue level, by the random removal of cells, which also results in the shrinkage. However, these possibilities have yet to be systematically explored. At the molecular level, the values of the intrinsic diffisivities and relaxation rates characterizing the microscopic compartments could, in principle, be calculated as a hnction of concentration during drying or freezing, by use of the proton exchange formalism andor the multicomponent theory discussed above. Clearly much remains to be done before this integration of distance and timescales can be achieved in any given multicompartment and multicomponent food system.

Acknowledgments The author wishes to thank the Biological and Biotechnology Science Research Council (BBSRC) for financial support.

References 1. R.M.Brunne, E.Liepinsch, G.Otting, G.Wuthrich, and W.F. van Gunsteren, JMoIec. Biol., 1993, 231, 1040. 2. V.P.Denisov, K.Venu, J.Peters, H.D.Horlein and B.Halle, JPhys. Chem B., 997, 101,9380. 3. J.R.Zimmerman and W.E.Britten, JPhys. Chem., 1957,61, 1328. 4. B.P.Hills and C.E.Manning, JMolec. Liq., 1998, 75, 61. 5. B.P.Hills, C.E.Manning, YRidge, and T.Brocklehurst, JSci. Food and Agrzc. 996, 71, 185.

62


6. M.C.Vackier and D.N.Rutledge, Food Chem. 1996,57,287. 7. R.Zsigmondy, Z.AnorgChem., 1911, 71, 356. 8. P.S.Belton, in Food Freezing: today and tomorrow, ed. W.B.Bald, Springer Verlag, Berlin, 1989 ch. 1 . 9. M.-C.Vackier, B.P.Hills and D.N.Rutledge,JMagn.Reson., in press. 10. P.T.Callaghan,Principles of hMR microscopy, Oxford Science Publications, Oxford, 1991. 1 1 . B.P.Hills,Magnetic Resonance Imaging in Food Science, John Wiley & Sons, New York, 1998. 12. B.P.Hills, J.Godward, C.E.Manning, J.L.Biechlin and K.M.Wright, Magn. Reson. Imaging, 1998, 16, no. 516 inpress. 13. B.P.Hills, Molec. Phys., 1992, 76,489. 14. B.P.Hills, and B.Remigereau, Int. J. Food Sci. and Tech., in press. 15. R.Ruan, S.J. Schmidt, A.R.Schmidt, and J.B.Litchfield, J Food Process Eng., 1991, 14,297. 16. A. Szafer, J.Zhong, and J.C.Gore, Magn. Reson, inMed., 1995,33,697. 17. B.P.Hills and J.E.M.Snaar, Molec. Phys., 1992,76,979.

Molecular Mobility of a System: Waxy Maize, Glycerol and Water, Studied by NMR D. C. P. Jardim,' J. R. Mitchell? W. Derbyshire,2 J. M. V. Blanshard2 and J. A. G. Areas3*

' INSTITUTO DE TECNOLOGIA DE ALIMENTOS, CP 139, CAMPINAS, S. P., BRAZIL FACULTY OF AGRICULTURAL SCIENCES, UNIVERSITY OF NOlTINGHAM, SUlTON BONINGTON, UK DEP. DE NUTRICAO, FACULDADE DE SAI~DE PUBLICA DA USP, AV. DR. ARNALDO, 715, CEP 01246-904, SAO PAULO, SP, BRAZIL

ABSTRACT Molecular mobility and physical state of biopolymers are important and informative aspects related to food stability. The glass transition temperature (Tg) better describes key constituents of foods, and it is calculated either for each component or for the whole system The present work studied the molecular mobility of a system constituted by starch and the plasticizers glycerol and water, by means of several techniques, namely, DMTA, water sorption isotherms, X-ray diffraction, texture measurements and NMR relaxation methods. Waxy maize was used as such or extruded to produce regular unexpanded semi-transparent ribbons of gelatinised starch. After drying to 22% moisture these ribbons presented glassy state characteristics. The sample was then placed in contact with glycerol water solutions of several Aws for 21 days both through ambient atmosphere in sealed chambers (system I ) and immersed in the solutions (systemII). System I samples lost water, even those placed in environments with relative humidity higher than the samples. On the other hand, system I1 samples gained water after this period. Crystal structure of starch was lost after extrusion and did not rewver on the samples of system I. However, a gradual increase on crystalline order was observed on system I1 samples as their moisture increased. Samples of system I showed a slight increase on log E (E=Young's modulus, calculated from three point bend test) with decrease on moisture content of the samples, whereas system I1 presented a marked dependence of Young's modulus and moisture wntent of the samples. There was also a correlation between log E and glass transition temperature determined by DMTA. Relaxation times of the samples, determined by CPMG, was practically unaffected in system I, whereas presented in system I1 a marked increase in TZas the Aw of glycerol solutions increased. Tg and TZwere also correlated, indicating the relationship between proton mobility and glass transition temperature. Food stability is related to biopolymers low mobility that is achieved by keeping them in environments of low water activity, preferably below their Tg. Relaxation parameters obtained by NMR may contribute to describe the idealised situation for low biopolymer activity. This technique has proved valuable to assess molecular mobility in the systems under study.

64


1 INTRODUCTION Molecular interactions, molecular mobility and physical state of biopolymers are important and informative aspects related to food stability. Several phenomena as retrogradation, crystallisation and gelatinization have been more efficiently explained when one uses molecular mobility or state transition concepts. The glass transition temperature (Tg) better describes the physical state of key constituents of foods, and it is calculated either for each component or for the whole system. Starch is an example of a partially crystalline biopolymer that presents physical changes governed by nonequilibrium phenomena that affect deterioration and shelf-life of starch products (Biliaderis et al, 1986; Slade and Levine, 1991; Noel et al, 1996; Tsoubeli et al, 1995; Chinachoti, 1993; Karel, 1985; Kalichewski et al, 1992; Jardim, 1998). The glass transition temperature concept has been used more recently preferentially to water activity for forecasting food deterioration. Whereas this approach has proved successful to predict deterioration due to chemical transformation that is determined mainly by molecular mobility of food components, it has been useless to predict microbial growth in the same materials. Molecular motion around this transition state has proved to be an important piece of information amenable to be addressed by non destructive, non invasive, low cost, low resolution proton NMR, which can add more information to food systems. The present work studied the molecular mobility of a system constituted by starch and the plasticizers glycerol and water, by means of several techniques, namely, DMTA, Aw, X-ray diffraction, texture measurements and NMR relaxation methods.

2 EXPERIMENTAL

Samples: Waxy maize starch was provided by National Starch and Chemical Co (Manchester, UK). It was used as such or extruded in a Clextral BC-21 extruder (Clextral, France) with successive extrusion temperatures of 48, 100, 120 and 99°C in consecutive sections of the equipment, producing a continuous ribbon (rectangular die of lmm x 3 cm) of homogeneous gelatinised starch in the rubbery state without air bubbles. These ribbons were dried over P205 under vacuum to a final moisture of 22% (d.s.b.) and cut in regular pieces for further experiments. After drying these ribbons presented characteristic glassy state behaviour. Glass transition temperature: Glass transition temperature (Tg) was determined in a dynamic mechanic thermo-analyser (DMTA) from Polymers Laboratories (UK) at a 5”C.min” heating rate, 1 Hz frequency and deformation xl in a single cantilever bending mode. Tg was obtained in the peak of tan 6 in the resulting thermogram. Glass transition was also calculated using Couchman-Karasz equation (Couchman and Karasz, 1978):

Tg = w,ACP, Tg, + W2ACP2TiT2 + W,ACP, Tg, W,ACp, + W2ACp, + W,ACp,

(1)


65

where: W1.2,~ - mass fraction of each component, ACpl,z,3- specific heat difference between the liquid and glass state of each component, Tg,,~,3 - glass transition temperature of the pure components.

Glass transition temperatures utilised were: 139K for water, 500K for the wax maize and 180K for the glycerol with respective specific heat differences of 1.94Jg-1K-7, 0.41Jg-'K-' and 0.88Jg-'K1. (Davies and Jones, 1953; Kalichewsky et al, 1992) Texture: Texture parameters were obtained using TA-XT2 (Stable Micro Systems, Ltd., UK) using the three point bend test. Young's modulus was calculated as: F

L3 E = -D 4bd3 -

where: E = elastic modulus, FA3 = gradient of the first section of the deformation line, b and d = width and thickness of the sample, respectively, L = distance between the two supporting points of the sample (9.3 mm).

M R experiments: They were performed in a spectrometer (Bruker Minispec PC 120) at 4OoC. The signal amplitudes and the decay time (T2)for the solid and liquid portions of the samples were obtained from the Free Induction Decay @ID)recorded after a single pulse. TZ of the whole sample was also calculated by CPMG (Carr and Purcell, 1954; Meibom and Gill, 1958) experiments, with T spacing of 1500ps between xi2 and x pulses. Aw measurements: Water activities were determined in a Decagon hygrometer mod. CX-1 (Decagon, USA) at 25°C. Before each experiment calibration curves were obtained through known salt saturated solutions and corrections of the measured values were made according to Greenspan (I 977).

X-ray diflaction: Crystalline patterns were obtained in X-ray diffractometer Phillips, model APD-15 (Phillips, Netherlands), fitted with a copper tube X-ray generator operating at 40kV and 50mA, producing a aCuK radiation of 1.54A wavelength. Data were acquired in the range 4 to 38", in 0.005" interval (28). 3 RESULTS AND DISCUSSION For simplicity, experiments will be herewith referred to as: System I: extruded waxy maize, equilibrated m chambers containing water glycerol solutions of several Aws without direct contact with the solution. Classical experiment for obtaining water sorption isotherms.

66


System 11: extruded waxy maize, glycerol and water, with direct contact of the starch ribbons with the glycerol water solutions of several Aws. Experiments for detecting diffusion of both the water and the glycerol. System 111: Native starch pre-equilibrated 21 days in an ambient of controlled RH. After this period the determined Aw of this native starch was 0.52. These samples were then placed in direct contact with a glycerol solution of same Aw. AAer 21 days equilibrium, extruded starch fiom both systems I and I1 had their Aws determined and the results are presented in Figure 1. Water content of these same samples are presented in Figure 2. All samples in System I1 presented an increase in

' I 0.8

'

EXprpeded

Original

.

Aws

Sample

sflm I1 0.6

.

sflm I 0.4

'

'

0.2 0.2

I

I

I

0.4

0.6

0.8

1

Aw ofthe glycerol solutions

Figure 1 - Water activity of gelatinised starch after 21 days of storage either with environments of known R H s provided by glycerol solutions of known Aws (system I); or direct contact with glycerol solutions of known Aws (system 11)

water content after the diffusion period, except the ones in contact with glycerol solution 0.34 Aw. However, their water activities experimentally determined were below the expected Aws after equilibrium. Samples of the System I lost water; even those placed in environments with relative humidities higher than the original sample (RHs = 68.7, 72.4 and 82.4%). Their observed Aws were also below the expected. The crystallinity of starch was monitored in all conditions studied and the results are presented in Figures 3 and 4, for System I and 11, respectively. The results clearly indicate that starch samples equilibrated in chambers without direct contact with glycerol solution had their crystalline pattern lost and it was not recovered even at high relative humidities (Figure 3). On the other hand, when direct contact occurred, the samples that were initially amorphous presented a gradual increase in crystalline order of starch as water activity of the glycerol solutions (and water content of the samples) increased.

61


Glass transition temperatures determined by the peak of tan 6 obtained in DMTA thermograms were also calculated for the system according to equation (1). The plasticizer effect of water was clearly observed when plotting Tg against moisture (not shown). The plot of the calculated Tgs pointed to the same pattern as for the experimental ones, but the absolute values were discrepant, being the calculated values

' I 0.8 0*9

1

A

system1

0'41i

0.2 0.3

20

10

0

40

30

70

60

SO

g H,O/lOO g dry sample

Figure 2 - Moisture of gelatinised starch, in indirect (system I) and direct (system 11) contact with glycerol solutions of known Aws

Aw

I

I

4

8

.

12

=

0.34

1

I

I

I

1

I

16

20

24

28

32

36

40

ze

Figure 3 - X-ray dieaction spectra for System I samples after 21 days of storage in chamber of various RHs produced by glycerol solutions of known Aw.

68


h 0.63

h 01.

4

8

1

2

1

~

2

0

2

4

2

~

3

2

3

0

4

0

20

Figure 4 - X-ray diffraction spectra for System I1 samples in direct contact with glycerol solutions of known Aw.

well below the experimental ones. Young moduli calculated from equation (2) by three point bend test of both systems are presented against the experimentally determined Tg in Figure 5 , where

Figure 5 - Log E (Young’s modulus) as a function of Tg after 21 days of storage in indirect (system I) and direct (system 11) contact with glycerol solutions of known Aws. Appearance of the samples are indicated as: R - rubbery; G - glassy.

observations about the visual aspect of both systems are also indicated. These results showed that system I was practically unaffected by the different Aw of the glycerol


69

solutions. For system 11, however, a clear dependence was noted and, after an initial decreasing, Young's modulus increased steadily, as the water content of the samples increased. Solid-liquid relationship of the systems was assessed by NMR based on signal amplitude at two distinct times of data acquisition. Values of Tz, which are related to the mobility of the whole system, obtained by the CMPG sequence, are displayed in Figure 6. The initial amplitude (T = 11ps, detector dead time) was taken as an indication of the total protons present, whereas the value obtained at 70 ps was considered signal from solids. Solid and liquid FID amplitudes are presented in Figures 6 and 7 for both systems studied Molecular mobility and solid and liquid components, assessed in this way, were usefd to describe the system.

. t

01 0.9

system11

System I

*

_.-

".

A

:

0.4

0.5

A.

0.6

0.7

0.8

0.9

1

glycerol solution Aw

Figure 6 - Tz relaxation (CPMG)of starch samples after 21 days of storage indirectly (system I) and directly (system 11) with glycerol solutions.

Based on signal amplitude the relative composition of solid and liquid were estimated. For systems I and I1 the results presented an expected behaviour, with more liquid component on system 11, which was the one with more water uptake. For the system I11 (native starch), however, it was observed (Figure 9) that there was more solid than expected due to liquid, probably glycerol, behaving like solid. This was checked by determining the density of glycerol solution after equilibrium, which presented lower values than expected. This difision of glycerol to starch granules altered mobility behaviour only detected by using NMR. The observed change in Tz as Tg varied indicated the relationship between samples in these two states.

Advunces in Magnetic Resonance in Food Science

70 7 -

System I

6 -

0

a

a

a^

E

5 -

Original

4 -

+

B

2

B j 3

O n

0

swan I1

2 -

+*

1 0.2

I

1

I

0.4

0.6

0.8

1

glycaol solution Aw

Figure 7 - Solid amplitudes after 21 days of equilibrium with indirect (system I) and direct (system 11) contact of the extruded starch samples with glycerol solutions of known Aw.

8

Original Sample

system I

0

60

a

0

I

0.2

0.4

0,s

0.8

1

glyceol solulicm Aw

Figure 8 - Liquid amplitude (FID) of extruded starch samples after 21 days in indirect (system I) and direct (system 11) contact with glycerol solutions of known Aw.

71


15

I

/ 0

-

did 51-

0

1

2

3

4

5

6

7

8

9

10

11

@ymd sdlllm:anylopedin

Figure 9 - Calculated and experimental solid and liquid amplitude estimated from FID of native starch immersed in glycerol solution of Aw = 0.52 (system III).

Food stability is related to biopolymers mobility that is achieved by keeping them in environments preferably below their Tg. Relaxation parameters obtained by NMR also contribute to describe the idealised situation for low biopolymer mobility and represents the best tool to describe this property. This technique has proved valuable to assess this characteristic in the systems under study and complemented information derived from other techniques, providing thus a better understanding of the physical aspects of food components related to deterioration. 4 REFERENCES

Biliaderis, C.G.; Page, C.M and Maurice, T.J. (1986) Carbohydrate Pol. 6: 269-288. Cam, H.Y. and Purcell, E.M. (1954) Pys. Rev. 94: 630.

Chinachoti, P. (1993) Food Technol., Jan. 134-140. Couchman, P.R. and Karasz, F.E. (1978)MacromoZecuZes 11: 117-1 19. Davies, R.O. and Jones, G.O. (1953) A h . Phys. 8: 370-410. Greenspan, L. (1977) J. Res. Nut. Bur. Stand A. Phys. Chem. 1: 89-96. Jardim, D.C.P. (1998) MobiZidade Molecular de um Sistema: Amido Glicerol e Agua, PhD Thesis, Universidade de Siio Paulo, Brazil, 140 pp. Kalichevsky, M. T.; Jaroszkiewicz, E.M.; Ablett, S.; Blanshard, J.M.V. and Lillford, P.J. (1992) Carbohydr. Polym. 18: 77-88.

12


Karel, M. (1985) In Simatos, D. and Multon, J.L. Properties of water in Foods in Relation to Quality and Stability, Martinus Nijhof Publishers, N.York, pp 153169. Meibom, A. and Gill, D. (1958) Rev. Sci. Instrum. 29: 688. Noel, T.R; Park, R. and Ring, S.G. (1996) Carbohydrate Res. 282: 193-206. Slade, L. and Levine, H. (1991) Crit. Rev. FoodSci. Nutr. 30: 115-360. Tsoubeli, M.N.; Davis, E.A. and Gordon, J. (1995) Cereal Chem. 72: 64-69.

Water Dynamics in Gelatine. Relaxation and Diffusion Analysis L. Foucat, A. Traor6 and J. P. Renou STRUCTURES TISSULAIRES ET INTERACTIONSMOLECULAIRES, SRV, INRA-THEE, 63122 ST GENkS-CHAMPANNELLE, FRANCE

1 INTRODUCTION

Gelatines are water soluble products of thermal and chemical degradation of collageneous tissues. A reversible sol-gel phase transition is one of the most characteristics of watergelatine systems. The gelling of gelatine results in the formation of the three-dimensional network of cross-links.' The specific interactions between water and gelatine chains play an important role in the stabilisationof gelatine gel. NMR relaxation rates (R, = l/T1 and R2 = l/TZ) and diffusion coefficient (D) provided an insight into the dynamics of water molecules and their local environments. The effects of various experimental variables such as gelatine concentration, gel strength, pH, temperature and measurement frequency were taken into account to characterise water-protein interactions.

2 MATERlALsANDMJXHODS Gelatine powders (Sigma) of two gel strengths (60 and 300 Bloom) were dissolved in deionised, distilled water at 60°C. NaN3 (400 ppm) was added to prevent microbiological growth, and pH adjusted with either NaOH (1M) for basic gelatines or HCI (1 M) for acid gelatines. The final concentrations were in the range of 5 to 20% (weight of dry gelatine per total weight). NMR measurements were canied out at 400 MHz on a Bruker AMX400 spectrometer (equipped with a microimaging accessory) and at 20 MHz on Minispec Bruker spectrometer PC20. Field frequency lock was not required. The temperature was controlled to f 0.1"C. The longitudinal relaxation rate (RI=l/TI) was measured using Inversion Recovery. The Cam-Purcell-Meiboom-Gill (CPMG) pulse sequence was used to determine transverse relaxation rates. The 90-180" pulse spacing (z) was varied between 50 ps and 2 ms. At 20 MHz, the acquisition time was maintained constant (800 ms) irrespective of the interpulse delay by varying the echo number between 40 and 160. Diffusion experiments were performed at 400 MHz with pulse field gradient multi-spinecho (PFGMSE) sequence.2


74

3 RESULTS AND DISCUSSION For each sample studied, a single R, value was observed.' This result is explained by an exchange process between water and gelatine protons. According to the fast exchange model in the limit of diluted systems R1 can be expressed by the relation ?

where Pb is the exchangeable proton population with the longitudinal relaxation rate Rlb and (I-pb) is the fraction of bulk water protons with the relaxation rate R,. The linear increase of RI with gelatine concentration (Figure 1) is consistent with an increase of P b and in good agreement with equation 1.

0.2

I

1 ' 0

5 10 gelatine concentration (%)

15

Figure 1 Variation of the relaxationparameter RI as afinction of gelatine (300 Bloom; pH 4.85) concentration at 400 MHzfor 2 temperatures (0) : I0"C; (0) :40°C This exchange phenomenon is confirmed by the transverse relaxation measurements. Rz values are frequency dependent and vary (at 400 MHz) with the inverse of the CPMG interpulse delay (g=l/z) (Figure 2).

100

1000

g=lh (s-1)

10000

100000

Figure 2 Variation of Rz as a &netion of the inverse of interpulse delay g (=I/z) for gelatine (300 Bloom, I5%, pH 4.85, 40°C) at two frequencies 20 and 400 MHz


75

In the range of g values used on Minispec (at 20 MHz), a much smaller variation in R2 was observed. Indeed, the contribution of the exchange process to the transverse relaxation mechanisms is 400 (Square of frequency ratio) times less important at 20 MHz than at 400 MHz. As a result, the R2 at 20 MHz depends almost exclusively on dipolar interactions. At 400 MHz, the behaviour of R2 as a function of g can be described (see Figure 2) by the Luz-Meiboom relation5as : R, =(R,),+k;'x[l-g

k;'xth(k,g-1)]xpb62

where k, is the rate of proton exchange, Pb the exchangeable proton population and 6 the chemical shift difference between water and exchangeable protons, (1-pb) is the fraction of bulk water protons with the transverse relaxation rate R2w.R2bis the weighted sum of spin-spin relaxation rates of water molecules at the protein interface that have lifetimes dependent on the number of binding hydrogens.6 From our experimental results, the different parameters k,(R2)o and pb62 were determined and are given in Table 1. Table 1 Exchange parameters determined >om equation 2 as a firnction of experimental parameters (gel strength,pH, temperature and gelatine concentration) Bloom pH "C concentration (R2)o (s-') k, (x lo3 s-l) ph6' ( x I O ~ S - ~ ) 300

4.85 40

300

6 40 7.15 8

5% 10% 15% 20% 15 %

0.49 f 0.03 0.6 f 0.1 0.8 f 0.1 0.8 f 0.3

7.3 f 0.3 8.3 f 0.6 8.9 f 0.4 8.8 f 0.6

13.9 f 0.6 35 f 3 60f3 96f8

0.71 f 0.08 1.OfO.1 1.5 f 0.2

8.6f 0.1 10.4fO.l 9.5 f 0.1

92*9 163 f 12 229 f 16

300

6

10

15%

2.4 f 0.2

10.3 f 0.1

60

6

40 10

15%

1.7 f 0.2 4.3 f 0.4

9.1 f 0.1 10.2 f 0.1

25

*2

111 f 10 37*4

The exchange rate k, is slightly influenced by gelatine concentration and by pH, and its values are very similar in "sol" and "gel" states, as already reported.' The (R2)o values depended on concentration and pH. Assuming a number of exchangeable-gelati-protons (hydroxyl and amino protons) equal to 0.33 per 100 g of gelatine at pH 4.85,' the P b was calculated for each concentration. From this assumption, a linear relationship was found between (R2)o and Pb (R2 = 0.993; F = 297). The intercept of 0.39 s-' corresponds to the RzWand is in good agreement with the transverse relaxation rate of pure water. From the slope of 64 s-' corresponding to R2b, the population associated with water molecules held

76


by four hydrogen bonds was less than 1% of the hydration layer, consistent with previous results obtained on bovine serum albumin.6 The pbZ2values correlated closely with Pb (R2 = 0.991; F = 228). Hence 6 did not vary within the gelatine concentration range and from the slope 6 was about 1.5 ppm. This value agrees with previous results of Hills.' This 6 low value forecasts a small contribution to the exchange process at 20 MHz. (R2)0 and pb6* decrease with the increase in gel strength. The increase in Bloom number is derived from the formation of covalent intermolecular cross links. These protein-protein interactions increased at the expense of the water-protein interactions and induced the decrease in Pb. In the gel state, the diffusion through the high internal magnetic field gradient G may affect the transverse magnetisation, according to the echo amplitude detected at time t with the CPMG sequence :

[ (

M(t) = M, x exp -t x R, + D$2)]

(4)

where D is the molecular self diffusion constant. Figure 3 displays R2 versus g-2.With the first four experimental values, a linear relationship between R2 and g" was found in the gel state, whereas in the sol state no linearity was observed. Assuming a water D value of about 1 in gelatine gel, the G value will be 600 Gausdcm, according to the CarrPurcell t h e ~ r yAt . ~ higher values of g-2, R2 becomes independent of g-2.[o,'*According to the restricted diffusion theory," the microheterogeneity length of 1.5 pm was determined. This microheterogeneity distance, which may characterise some biological systemsi2 is greater than the 0.02-0.1 l m of the nucleated junction zones in gelatine gel reported by Djabourod3 from the electronic microscopy study. This inner magnetic field gradient may be attributed to the presence of air-bubbles trapped in gelatine during the gelling process.

lE09

1E-07 g-'

(s')

Figure 3 R2 as a function of g" for gelatine 15%, 300 bloom, pH 6

17


Diffusion coefficient values determined at 400 MHz are reduced by a factor 1.5 to 2 compared with pure water at the same temperature, and decrease linearly with the increase in gelatine concentration. Hydration and obstruction models predict this behaviour. The study of D in function of the diffusion time (A) shows a decrease of D when A increase (Figure 4). This non-Brownian behaviour may be due either to the inner magnetic field gradient G, or to the restricted diffu~ion.’~ Approximating the distribution of the internal gradients Go by a Gaussian distribution, the relation DaPp= D[1- 0.5yZdDA(z- A/2)’]

(5)

is derived,I4 where D is the diffusion coefficient fiee from the magnetic susceptibility effects and o2stands for the variance of the internal gradients. The D,,-vs-A plot can be used to derive $. The observed dependence of D,, (Figure 4) is not well simulated by equation 5. At all events, the d values thus determined for different gelatines are very small (lo0 Hz. This was attributed to a state of intermediate rigidity. PSRE N M R was used to distinguish two categories of galacturonans, i.e., relatively rigid polymers with a low degree of methyl esterification and more mobile polymers with a higher degree of methyl esterification. PSRE Nh4R has also been used to distinguish two categories of galacturonan signals in I3C NMR spectra of strawberry cell walls.' Values of T 2 0 for the relatively highly methyl-esterified galacturonans became lengthened, and the response to cross-polarization weakened, as the h i t ripened.* These observations indicated a transition to a more liquidlike state of molecular mobility, helping to explain the softer texture of the ripe h i t .

4.CONCLUSIONS The objective approach to PSRE N M R was successll in distinguishing two or three categories of molecular rigidity, in all of the cases studied in this assessment. Cellulose and other crystalline constituents showed the best signal strength and sharpest peaks. Noncrystalline components showed signals that were weakened by inefficient crosspolarization processes. Values of TI@), T 2 0 , linewidths and signal strengths all provided information about molecular rigidity in the noncrystalline constituents. Published studies based on PSRE NMR have shown that the method can provide insights into the


156

molecular origins of textural variations in fruit, vegetables, and cereal products. The new objective approach might be helpfd in extending applications to other foods. 5. EXPERMENTAL

Samples were packed in 7 mm diameter cylindrical silicon nitride or sapphire rotors and retained with end caps machined from Vespel or Kel-F. Poly(chlorotrifluoroethy1ene) grease (Halocarbon 25-58) was used to ensure a water-tight seal without contributing to I3C NMR spectra obtained with cross-polarization excitation. Samples were spun at about 4 kHz in a magic-angle spinning probe @oty Scientific Inc.) for N M R at 50.3 M H z in Varian XL-200 or Inova-200 spectrometers. Typical operating parameters were: proton preparation pulse 6 ps, cross-polarization contact time 1 ms, data acquisition time 30 ms, signal recovery delay between 1 and 2 s. The proton transmitter output was increased to yB~l(2n)> 60 kHz for proton spin decoupling during data acquisition. Experiments were left running for periods between several hours and 2 days in order to achieve the high signal-to-noise ratios required for PSRE NMR. Acknowledgement

The author thanks Dr J. A. Hemmingson for preparing the cellulose-galactomannan complex. References 1. T. J. Foster, S. Ablett, M. C. McCann and M. J. Gidley, Biopolymers, 1996,39, 51. 2. M.-A. Ha, B. W. Evans, M. C. Jarvis, D. C. Apperley and A. M. Kenwright, Carbohydr. Rex, 1996, 288, 15. 3. D. L. VanderHart and E. Perez, Macromolecules, 1986,19, 1902. 4. R. H. Newman and J. A. Hemmingson, Holzforschung, 1990,44,351. 5. C. M. Preston and R. H. Newman, Can. J. Soil Sci., 1992, 72, 13. 6. R. H. Newman, M.-A. Ha and L. D. Melton, J. Agric. FoodSci., 1994,42, 1402. 7. R. H. Newman, L. M. Davies and P. J. Harris, Plant Physiology, 1996,111,475. 8 . T. H. Koh, L. D. Melton and R. H. Newman, Can. J. Botany, 1997,75, 1957. 9. K. R. Morgan, R. H. Furneaux and R. A. Stanley, Curbohydr.Res., 1992,235, 15. 10. K. R. Morgan, R. H. Furneaux and N. G. Larsen, Carbohydr. Res., 1995,276,387. 11. T. T. P. Cheung and B. C. Gerstein, J. Appl. Phys., 1981,52, 5517. 12. R. H. Newman and L. M. Condron, Solid State M R , 1995, 4, 259. 13. R. H. Atalla and D. L. VanderHart, Science, 1984,223,283. 14. R. H. Newman and J. A. Hemmingson, Cellulose, 1995, 2, 95. 15. R. H. Newman, Holzforschung, 1998, 52, 157. 16. J. F. Revol, A. Dietrich and D. A. I. Goring, Can. J. Chem., 1987,65, 1724. 17. B. G. Smith, P. J. Harris, L. D. Melton and R.H. Newman, Plant Cell Physiol., 1998 (in press). 18. C. Sterling and F. Shimazu, J. Food. Sci., 1961, 28,479. 19. C. A. Willis and A. A. Teixeira, J. FoodSci., 1988, 53, 111. 20. E. Garcia, T. M. C. C. Filisetti, J. E. M. Udaeta and F. M. Lajolo, J. Agric. Food Chem., 1998,46,2110. 21. M. C. McCann, B. Wells and K. Roberts, J. Cell Sci., 1990, 96, 323.

Functional Constituents of Food

157

22. S. E. C. Whitney, J. E. Brigham, A. H. Darke, J. S. G. Reid and M. J. Gidley, Plant J., 1995, 8, 491. 23. S. E. C. Whitney, J. E. Brigham, A. H. Darke, J. S . G. Reid and M. J. Gidley, Carbohydr.Hex, 1998,307, 299. 24. J. F. Ang and W. B. Miller, CerealFoods World, 1991, 36, 558. 25. R. H. Newman and J. A. Hemmingson, Carbohydr.Polym., 1998, in press. 26. R. Ilker and A. S. Szczesniak, . I Texture Studies, 1990, 21, 1. 27. K. M. Fenwick, M. C. Jarvis, D. C. Apperley, G. B. Seymour and C. R. Bird, Phytochem., 1996,42,301. 28. W. P. Rothwell and J. S. Waugh, J. Chem. Phys., 1981, 74,2721. 29. D. L. VanderHart, W. L. Earl and A. N. Garroway, J. M a p . Heson., 1981,44,361. 30. W. L. Earl and F. W. Parrish, Carbohydr.Res., 1983,115,23. 3 1. R. K. Harris, ‘Nuclear Magnetic Resonance Spectroscopy’, Pitman, London, 1983, p 87. 32. B. G. Smith, P. J. Harris, L. D. Melton and R. H. Newman, Physiologia Plantarum, 1998, in press.

Cross-polarisation Kinetics and the Determination of Proton Mobility in Hydrated Plant Cell Walls M. C. Jarvis, M. A. Ha and R. J. Vietor CHEMISTRY DEPARTMENT, GLASGOW UNIVERSITY, GLASGOW G12 SQQ, SCOTLAND, UK

1. INTRODUCTION Plant cell walls are complex nanostructures which, in their natural hydrated form, have remarkable properties of strength, resilience and controlled flexibility'. Their mechanical properties, when they are under tension from the internal turgor pressure of the cell, define the textural properties of fresh fruit and salad crops. When distended by starch swelling pressure instead of turgor the cell walls control the textural quality of cooked and processed starch-containing vegetables like potatoes, peas and beans. In cereal products their influence on texture is less central but they modulate the availability and movement of water during processing. The constituent polymers of hydrated cell walls exhibit an extraordinary range of physical properties, despite the small scale of their internal structuring (nanometers or tens of nanometers)'. Cellulose microfibrils are solids of moderate crystallinity, although less than 10 nm in diameter. In contrast the p( 1,4')-linked D-galactan and a(1,5')-linked Larabinan side-chains of pectin behave like tethered liquids with respect to their chain mobility. These properties are reflected in 'H T2 values of ca. 10 p for cellulose and up to 1 ms for the pectic galactan?. A wide range of motional and relaxation behaviour is represented in other polymers between these two extremes. It follows that NMR relaxation experiments on hydrated plant cell walls are making a valuable contribution to our understanding of their influence on food quality. We are only now beginning to see how the essential viscoelastic properties of the cell wall emerge from the properties of rigid and flexible polymer chains, but NMR methods have been responsible for much of the progress that has been made. Such methods probably are the key to future developments in this area. In a typical cross-polarisation, magic-angle spinning (CP-MAS) NMR experiment it is not only the 'H and I3C relaxation processes that are sensitive to thermal motion. The CP process itself is capable of being disrupted by motional effects, and it is well known that

Functional Constituents of Food

159

between the ranges of polymer mobility accessible in CP-MAS and conventional solutionstate 13C NMR experiments their lies a region in which polymers are 'invisible' because they cross-polarise too slowly to give a detectable 13C signal before the 'H magnetisation is lost by the T I , process3. This paper deals with the use of CP kinetics as an alternative probe of molecular mobility, and specifically with the potential of slow-CP measurements as a way to explore the mobility of hydrated polymers in the 'invisible' range of mobility.

2. THEORY OF HARTMA"-HAHN CROSS-POLARISATION TheoreticaI descriptions of Hartmann-Hahn CP are available in the l i t e r a t ~ r e ~Here - ~ . we are concerned only with the kinetics of the process from a practical, descriptive viewpoint. A typical time-course for Hartmann-Hahn CP in a dry polysaccharide material is shown in Figure 1. Polarisation in 13C builds up in two phases before decaying through the Tlp process in the proton spin reservoir with which it is, by that time, in equilibrium. The initial CP phase is a flip-flop process between the I3C nucleus and the proton(s) with which it is most closely associated - normally by covalent bonding, although this is not essential and only spatial proximity and a heteronuclear dipolar interaction are required. The second, slower CP phase involves spin diffusion from more distant protons.

0

I

,

I

1

2

I

3

T,ms

Figure 1 Evolution, with increasing Hartmann-Hahn contact time, of I3C signal intensity at 105 ppm (cellulose C - I )from dry citrus cell walls

160


For static single crystals the kinetics of the initial phase can be described by a damped oscillating function':

S = Sd2 (1 - exp(-3/2Rz) cos b2/2)..........................................................

(1)

Where So is the theoretical maximum signal intensity, b is the dipolar coupling between the I3C nucleus and the covalently bonded proton, and R is the rate of spin diffusion within the surrounding shell of nonbonded protons. Hediger' has re-examined this relationship and found that an exponent of (-3Rz) instead of (-3/2R.t) gave better fits to the experimental data. In the case of powder samples, integrating over all orientations allows (1) to be approximated by an exponential function', which with the addition of terms for the second, slower CP phase and for proton spin-lattice relaxation in the rotating frame gives:

where TCHRand TCHDare time constants related inversely to R and b respectively". For a I3C nucleus with a single covalently bonded proton s = 0.5 since, in the approximation of an isolated two-spin system, the equilibrium polarisation is shared equally by the 'H and 13C nuclei. Thus the fast and slow components of cross-polarisation are predicted to be equal in magnitude. For CH2 and CH3 groups the directly bonded protons can, in principle, make a larger contribution, i.e. s < 0.5. Figure 1 shows that the progress of CP can be fitted by a biexponential curve as predicted by equation (2). Substituting experimentally estimated, rather than fitted, values for TCHR gives CP rates that are not as close to the observed values but are of the correct order of magnitude (data not shown). It should be emphasised that equation (2) refers to the static case. The effect of MAS, with a centreband Hartmann-Hahn matching condition, is to reduce the efficiency of the initial, rapid CP phase which depends on the C-H dipolar interaction. Under high-speed MAS conditions the contribution of this phase is expected' to diminish to zero i.e. s + 1, but at MAS rates of lpm. For larger samples up to 10 mm, using gradients an order of magnitude smaller reduces the attainable resolution to = 10 pm. Dielectric losses at X-band limit aqueous samples to < 1.5mm. Decreasing the frequency reduces the losses and allows larger samples. At L-band ( x 1.1 GHz) lossy samples up to 30 mm may be accommodated with a resolution of = 1 mm. For larger samples, we have used radiofrequencies in the range 200 to 400 MHz." In our laboratory we have obtained 2-D spatial images of spin density of nitroxyl spin probes by using two orthogonal gradient coils in order to rotate the gradient in the plane so that spectra can be obtained from typically 16 or 32 projections.20 The gradient-broadened spectra are Fourier deconvolved using the zero-gradient spectrum. 2-D images are constructed using the filtered back-projection reconstruction method25with Ram-LakZ6prefiltering. Extension to three spatial dimensions was demonstrated for a sample of irradiated

196


quartz.z7 At L-band, Zweier and co-workersz8 reported 3-D images of the rat heart with a resolution of 1-2 mm using a nitroxyl radical. For situations where the spectrum varies with spatial position (other than just in amplitude) spectral-spatial imaging was developed.z9330Spectra of a pseudo object of length AJ3 in the spectral dimension and L in the spatial dimension are obtained at a series of magnetic field gradients corresponding to projections at angle a to the spectral axis. The technique was demonstrated at X-band for one spectral and one spatial dimen~ion.~' More recently Zweier and co-workers have extended the technique to four dimensions in their studies of rat hearts at L - b a ~ ~ d . ~ ' 5 INSTRUMENTAL CONSIDERATIONS AND TECHNIQUES 5.1 Resonators

For imaging purposes it is important that the radiofrequency field is as uniform as possible. The loop-gap resonator, originally developed for X-band by Hyde and F r o n ~ i s z ~ ~ has been used for larger samples at both L-band and radiofrequencies. The radiofrequency field is perpendicular to the axis of the cylinder. In our laboratory we have used birdcage resonators at radiofreq~encies.~~ These are used for MRI and have good homogeneity of the radiofrequency field which conveniently is directed along the axis of the cylinder which is co-axial with the axis of the static field. For variable-temperature operation on food systems we have built an 8 cm 0.d. birdcage on a quartz d e ~ a r Recently, .~~ a birdcage resonator operating at L-band has been reported.35 5.2 Magnets

At radiofrequencies where the magnetic field is in the order of 10 mT, it is convenient to use air-core magnets. In our earlier work we built a simple Helmholtz electromagnet similar to that of Halpern et al.36 Recently, we have built a solenoid magnet around a 40 cm diameter shim set from a horizontal-bore superconducting NMR magnet previously used for MRI at 2.3 T.34 This allows operation at up to 400 MHz on samples up to 10 cm in diameter. Conveniently the BO shim is used to produce a uniform modulation field at 10 m. 5.3 Phase noise

Larger samples require higher levels of excitation. This causes problems with phase noise from the frequency source since the signal is detected in the presence of the excitation. Random fluctuations in frequency are discriminated by the bridge and are manifest as amplitude fluctuations in the spectrum. Since our original workz4 we have reduced the level to the practical limit by using an improved signal generator37 and by increasing the modulation frequency to 10 kHz.29 Further increase in power causes thermal effects which make the tuning and matching too unstable.

Functional Consiituenls of Food

197

5.4 Longitudinally Detected Electron Spin Resonance

This technique, LODESR, circumvents the phase noise problem and has been demonstrated at radiofrequency by Lurie and c o - w o r k e r ~ .The ~ ~ radiofrequency excitation is modulated at a frequency which is less than the linewidth, typically about 200 kHz. Applied at a level of several watts, this takes the electron resonance into and out of saturation at the modulation frequency. A coil tuned at the modulation frequency detects the changes in magnetisation along the z axis. Not only is the phase noise problem eliminated but also there is no need for field modulation. This is a promising technique for food applications, although in the presence of different radicals or radicals in different motional environments, the response may be significantly different on account of differences in saturation characteristics. 5.5 Dynamic Nuclear Polarisation

Enhancement of NMR signals by simultaneous saturation of the electron resonance was the original Overhauser effect.3924oIn particular, a 2 mM solution of '%TMIOD in triethylene glycol dimethyl ether (triglyme) is used in magnetometers in the earth's field and provides enhancement factors of between 1000 and 2000 over a temperature range of 2 5 125 OC.'* Proton-electron double resonance imaging, PEDRI,41combines proton NMR imaging with the sensitivity enhancement arising from irradiation of the electron resonance. In a field of 10 mT conventional proton images would be obtained at about 425 kHz with the electrons being irradiated at radiofrequency. Further sensitivity improvement can be achieved by the use of a field-cycling magnet4' where the proton magnetisation is established at high field, the electron enhancement is carried out at low field followed by proton imaging at high field. 5.6 Multiple-technique Spectrometerhager

A system which combines conventional radiofrequency ESR and EMRI, LODESR and PEDRI has been described recently.43 In our laboratory, LODESR mode would be achieved by installing a solenoid coil tuned to say 425 kHz along the field axis. For PEDRI the proton signals at a similar frequency would be detected by another birdcage coil. The proton resolution would be optimised using the shim set and the gradients would be applied through the X, Y and Z shims or the supplementary gradient coils which provide up to 1 mT cm-'. 5.7 Pulse Radiofrequency EMRI Despite the problem of short electron relaxation times, Sotgiu and c o - w o r k e r ~have ~~~~ reduced the dead time associated with recovery from the pulse excitation to enable a 2-D image to be obtained from lithium phthalocyanine in a 40 cm3 phantom at 220 MHz under physiological conditions. There is a prospect of a dramatic improvement in signal to noise but at the moment the dead time is still too long for practical food samples. However, the reduction is still insufficient for practical food samples.

198


6 IMAGING IN FOOD SYSTEMS

1-D imaging of diffusing mobile spin probes or spin-labelled species, performed either as discussed in Section 3 or by DID-ESR (Dynamic Imaging of Diffusion by ESR)46,47or by 2D spectral imaging provides information on transport of materials. Simultaneously the spatial dependence spectrum indicates changes in the local microviscosity and, in suitable cases, changes in the local pH and oxygen concentration. The experiments on gels in capillaries at X-band” exemplify this application. The principle of studying radicals diffusing into plant stems was established many years ago by 1-D and 2-D L-band experimentson celery stems.48 In principle hydration phenomena in large food samples could be followed by the spread of a water-soluble nitroxyl radical. Dehydration would presumably leave at least some radical behind in an immobile state with a broad spectrum which would cause problems for EMRI. The PEDRI technique would require the protons to be in a mobile state and would indicate the spatial distribution of the spin probe. Normal TI-weighted MRI could also indicate this through enhanced relaxation. As discussed in Section 2, radicals tailored to partition into hydrophilic or hydrophobic phases give detailed information on changes in molecular motion as the temperature is varied and we have built a resonator to study large food samples at radiofrequenciesover a temperature range from -100 to +150 OC. In principle the changes taking place during cooking or freezing or thawing can be monitored. Oxygen plays a key role in food systems. The technique of oximetry is well-developed for in vzvo applications and utilises the line broadening effects on nitroxyl radicals in soluti0n,4~or implanted solid particles such as lithium phthalocyanine” or carbohydrate chars.51 The use of spectral-spatial imaging was demonstrated for nitroxyl radicals and techniques for following oxygenation and deoxygenationusing nitrogen were reported.49 In conclusion, there is a range of applicationsfor EMRI in food systems, particularly in association with ESR spectroscopy. At present we are working on dough, ice cream and emulsions. The range of mobilities encountered makes this a challenging area for EMRI.

References B. Hills, ‘MagneticResonance Imaging in Food Science’, Wiley, USA, 1998. ‘EPR Imaging and In Vivo EPR’, eds. G.R. Eaton, S.S. Eaton and K. Ohno, CRC Press, USA, 1991. 3. S.A. Fairhurst, D.G. Gillies and L.H. Sutcliffe, Spectros. World, 1990, 2, 14. 4. Res. Chem. Intermed, 1996,22, No. 6. 5. Phys. Med Biol., 1998,43, No. 7. 6. L.H. Sutcliffe,Phys. Med Biol., 1998,43, 1987. 7. ‘Spin Labeling: Theory and Applications’, ed. L.J. Berliner, Academic Press, New York, 1976. 8. R. Bolton, D.G. Gillies, L.H. Sutcliffe and X. Wu, J. Chem. SOC. Perhn Trans 2, 1993,2049. 9. R. Bolton, L.H. Sutcliffe and X. Wu., J. Labelled Compounds and Radiopharmaceuticals, 1994,34, 663, 1. 2.

Funcfional Constituenfsof Food

199

10. D.G. Gillies, L.H. Sutcliffe and X. Wu, J. Chem. Soc. Farachy Trans., 1994, 90, 2345. 1 1 . D.G. Gillies, L.H. Sutcliffe and X. Wu, Food Chemistry, 1996,55,349. 12. D.G. Gillies, L.H. Sutcliffe and M.R. Symms, J. Chem. SOC.Faraahy Trans., 1994, 90,267 1. 13. D.G. Gillies, L.H. Sutcliffe and X. Wu, unpublished results. 14. P.S. Belton, A. Grant, D.G. Gillies, A.I. Smirnov, L.H. Sutcliffe and X. Wu, 1998, 4th International Conference on Applications of Magnetic Resonance to Food Science, Norwich, UK, abstract D13. 15. Scientific Software Services, Bloomington, 305 E Locust, I1 61701, USA. 16. C.A. Beadle, D.G. Gillies, L.H. Sutcliffe and X. Wu, J. Chem. SOC.Farachy Trans., 1995,91, 887. 17. S.A. Fawthrop, D.G. Gillies, L.H. Sutcliffe and M.R. Symms, Mugn. Reson. Chem., 1995,33, S107. 18. E. Belorizky, D.G. Gillies, W. Goreki, K. Lang, F. Noack, C. Roux, J. Struppe, L.H. Sutcliffe, J.P. Travers and X. Wu, J. Phys. Chem. A, 1998,102,3674. 19. R. Damadian, M. Goldsmith and L. Minkoff, Physiol. Chem. Phys., 1977,9,97. 20. P.J. McDonald, Prog. hMR Specfrosc., 1997,30,69. 21. P. Mansfield and P.G. Moms, “MR Imaging in Medicine’, Academic Press, New York, 1982. 22. D.G. Gillies, L.H. Sutcliffe and M.R. Symms, J. Chem. Soc. Farachy Trans., 1994, 90,2671. 23. M. Jkeya and T. Miki, J.Appl. Phys., 1987,26, L929. 24. N.M. Bolas, D.G. Gillies, L.H. Sutcliffe and M.R. Symms, Res. Chem. Intermed, 1996,22, 525. 25. R.A. Brooks and G. Di Chiro, Radiology, 1975,117,561. 26. G.N. Ramachandran and A.V. Lakshminarayanan,Proc. Nut. Acad Sci. USA, 1971, 68,2236. 27. R.K. Woods, G.C. Bacic, P.C. Lauterbur and H.M. Swartz, J. M a p . Reson., 1989, 84,247. 28. P. Kuppusamy, P. Wang and J. Zweier,Magn. Reson. M e d , 1995,34,99. 29. M.M. Maltempo, J. M a p . Reson., 1986,69,82. 30. M.M. Maltempo, S.S. Eaton and G.R. Eaton, J. Magn. Reson., 1986, 72,77. 3 1 . M.M. Maltempo, S.S. Eaton and G.R. Eaton, . I M a p . Reson., 1988, 77, 75 32. J.L. Zweier, M. Chzhan, P. Wang and P. Kuppusamy, Res, Chem. Intermed, 1996, 22,615. 33. W. Froncisz and J.S. Hyde, J. Magn. Reson., 1982,62,79 34. D.G. Gillies, to be published. 35. J.A.B. Lohman, M.A. Allan, W.A. Miller, A.J. Illsley and R. Ladbury, 1997, 39th Rocky Mountain Conference on Analytical Chemisriy, Denver, USA, abstract 120. 36. H.J. Halpern, D.P. Spencer, J. van Polen, M.K. Bowman, A,C. Nelson, E.M. Dowey, and B.A. Teicher, Rev. Sci. Instrum., 1989,60, 1040. 37. Marconi Type 2041, MarcoN Instruments Ltd., Stevenage, UK. 38. I. Nicholson, F.J.L. Robb and D.J. Lurie, J. Map. Reson., 1994, B104,284 39. A.W. Overhauser, Phys. Rev., 1953,92,411. 40. A. Abragam, Phys. Rev., 1955,98, 1729. 41. D.J. Lurie, D.M. Bussell, L.H. Bell and J.R. Mallard, J.Magn. Reson., 1988, 76, 366.

200


42. D.J. Lurie, M.A. Foster, D. Yeung and J.H.S. Hutchison, Phys. Med. Biol., 1998, 43, 1877. 43. S.J. McCallum, I. Nicholson and D.J. Lurie, Phys. Med. Biol., 1998, 43, 1857. 44. G. Placidi, J.A. Brivati, M. Alecci and A. Sotgiu, Phys. Med Biol., 1998,43, 1845 45. M. Alecci, J.A. Brivati, G. Placidi and A. Sotgiu, J. M a p . Reson., 1998, 130,272. 46. J.K. Moscicki, Y-K. Shin and J.H. Freed, J. M a p . Reson., 1989,84,554. 47. J.K. Moscicki, Y-K. Shin and J.H. Freed, ‘EPR Imaging and In Vzvo EPR’, eds. G.R. Eaton, S.S.Eaton and K. Ohno, CRC Press, USA, 1991, Chapter 19. 48. L.J. Berliner and H. Fujii, Science, 1985, 227, 1985. 49. H.M. Swartz and J.F. Glockner, ‘EPR Imaging and In Vzvo EPR, eds. G.R. Eaton, S.S.Eaton and K. Ohno, CRC Press, USA, 1991, Chapter 24. 50. A.I. Smirnov, S-W. Norby, T. Walczak, K.J. Liu and H.M. Swartz, J. M a p . Reson., 1994, B103,95 51. R.B. Clarkson, B.M. Odintsov, P.J. Ceroke, J.H. Ardenkjaer-Larsen, M. Fruianu and R.L. Belford, Phys. Med. Biol., 1998,43, 1907.

Signal Treatment and Analysis in Magnetic Resonance

Analysis of Time Domain NMR and Other Signals D. N. Rutledge, A. S. Barros, M. C. Vackier, S. Baumberger and C. Lapierre INSTITUT NATIONAL AGRONOMIQUE, 16, RUE CLAUDE BERNARD, 75005 PARIS,FRANCE

1 INTRODUCTION

has been Until recently, Time Domain - Nuclear hhgnetic Resonance (TD-NMR) used almost exclusively to quantify major constituents in agro-food and petrochemical products or to monitor their evolution during processing. In this context the technique is often referred to as “Low Resolution NMR”. This situation has changed with the advent of more sophisticated instruments that can be used to perform NMR experiments previously only possible on much more expensive, high-field spectrometers. Time Domain - N M R is therefore now being used both for quality control in industry and for research purposes. In TD-NMR, unlike other instrumental techniques such as I&md spectroscopy, it is possible to generate a wide range of responses by using different sequences of radio fkquency pulses to excite the protons in the sample. The resulting relaxation curves may vary as a fimction of the physicochemical properties of the product. This apparently unlimited number of possible signals is both an advantage and disadvantage for TDN M R : on the one hand, it increases the range of potential applications of the technique, on the other, it complicates the development of new analyticalprocedures. Chemometric techniques, such as Analysis of Variance (ANOVA) and Partial Least Squares Regression (PLS), will be shown to be very u s e l l way of getting around this problem inprder to determine whether a particular TD-NMRor other signal contains any relevant information and to then extract and use that information. Techniques for the simultaneousanalysis of several signals will also be presented. Chemomtrics has already been applied to TD-NMR signals by Davenel et ul. who used Principal Components Analysis to study the relaxation curves of doughs during cooking, by Gerbanowski et al. applying PLS and Multiple Linear Regression (MLR) to relaxation curves and calculated relaxation parameters (TI, TZ and initial signal amplitudes) and Airmu et ul. using Principal Components Analysis and Evolving Factor Analysis to study the influence of a complexation reaction on relaxation curves and calculated relaxation parameters. Vackier el ul. 4, applied ANOVA, MLR and PLS to both relaxation curves and calculated relaxation parameters of gelatines while Clayden et ul. applied Factor Analysis to Time Domain-NMR FID signals of PTFE samples with different crystalliuities.

’

1.1 Chemometric Techniques Applied

As Analysis of Variance (ANOVA) is a univariate statistical technique it can be very rapid. It is therefore prekrable in order to have a quick indication of the information

Advances in Magneric Resonance in Food Science

204

content of the signals. If the signal is shown to be interesting, multivariate techniques such as PCA or PLS regression may then be applied. I.1.1. ANOVA. If the samples can be classed into groups, it is possible to calculate the part of the variability of a measurement due to differences between these groups and compare it with the variability within the groups. For each measurement one calculates the Between-Group or Group Variance (VG) and Within-Group or Residual Variance (VR):

’

J=I

with : g = number of groups n = number of samples in group j = value for sample i in group j 7 - = mean value for group j 7 = grand mean value N = total number of samples J

J

When the measurements are in fact points in a signal, such as a TD-NMR relaxation curve, where the information content of successive points is strongly correlated, it is interesting to plot these variance values as a function of the position in the signal. Regions that vary systematically fiom one group to another will give high VG values. If there are no other important differencesbetween the samples, then VR will be low and will not have a structured distribution as a function of position in the signal. In this way one can not only determine whether a signal is interesting and whether all the significant sources of variability have been taken into account, but also highlight those parts of the signal which are most important. 1.1.2. Partial Least Squares Regression. PLS regression may be used to generate predictive regression models. PLS is a multivariate, least squares regression procedure where a reduced set of non-correlated, linear combinations, T, of the original independent variables, X, are regressed onto the dependent variable, Y. PLS differs fiom Principal Components Regression in that the T are not simply the Principal Components, but are calculated iteratively, maximising their covariance with Y. This predictive regression model obtained is ofthe form : Y=X*B+& (3)

*

where B, the vector of B-coefficients, is calculated fiom the loadings of the X variables on the T vectors. 1.1.3. The Durbin-Watson Statistic. The Durbin-Watson D statistic is classically used as a measure of the randomness of residuals after a regression. We propose to use it

205


as a measure of the structure, non-randomness or information content of the loadings vectors and B-coefficients vectors produced by the PLS regressions. In this way, the Durbin-Watson statistic can be used to characterise the “signal f noise” ratio and thus indicate to determine the optimal number of Factors to use in the regression model. This statistic is given by:

D=

c” c

(6x,

- %I

l2

1=2

(6Xi *6 X i )

i=2

(4) . .

where “ i and “ i - 1 are the residuals for successive points in a series. For n>lOO, the distribution is random with a 95% confidence interval for D between 1.7 and 2.3. 1.1.4. Outer Product Analysis. It is often interesting to compare the simultaneous variations in two types of signals, such as spectra, relaxation curves or chromatograms, as a function of the evolution of some particular property of a set of samples. Several procedures have been proposed to highlight the covariations in two sets of si nals. Barton er al. lo used Ordinary Least Squaresto correlate variations in signals. Noda developed a means to detect correlated and anti-correlated vibrations in Infrared spectra as a fbnction of an imposed perturbation using 2D-Correlation Spectra. Devaux et al. l2 applied Canonical Correlation Analysis to highlight similar evolutions in Near and Mid-Inhred spectra of edible oils as a h c t i o n of their degree of unsaturation. Barros et al. l 3 demonstrated the utility of applying ANOVA to the Outer Product matrices of these same sets of I n h e d spectral vectors to detect correlated variations in the spectra and thereby attribute particular combinations of Mid-I&ared vibrations to given features in the Near I n h e d spectra. This idea has since been generalised as Outer Product Analysis (OPA) by applying other statistical techniques such as Principal Components Analysis, PLS Regression and Factorial CorrespondenceAnalysis. l4

’k

1.2 Samples Studied

1.2.1. “Light”and “Traditional”Butters and Margarines. The moisture contents of 12 “light” and “traditionalyy butters and margarines were determined in triplicate by the Karl-Fischer method. The values ranged fkom 12 to 60%. lOmm outer-diameter Nh4R tubes were filled up to a height of 1Omm. The tubes were thermostated at 20°C before transfer to the N M R apparatus. The measurements were performed at 20°C on a 20 M H z TD-NMR apparatus (QP20+, OXFORD INSTRUMENTS) with phase quadrature detection. A set of relaxation curves was acquired by inserting a Carr-Purcell-MeiboomGill (CPMG) sequence into an Inversion-Recovery(I-R) sequence :

I-R = [180° -- 2 *(1.55) i-l -- CPMG -- RD} N CPMG = 90” -- T -- [{ 180”, -- 2r -- } 3 -- 18OoY-- T -- measure] M withN = 20, i=l to N, T = lms, M = 100, RD = 3s, Scans= 4 The resulting signals may be represented either folded as 2-Dimensional relaxation surfaces (Figure 1) or unfolded as a series of I-R weighted CPMG curves (Figure 2). As can be clearly seen in Figure 2, there are two Ti and T2 components, and the less mobile component at the beginning of the CPMG curves (short T2) having a shorter TI, as it relaxes more quickly in the I-R curve.

206


-2500 5000

0

0

Figure 1. A rypical2D-Relaxation surface for a traditional butter.

moo 6000

a r

4000

5

2000

P eE

o

e a

-2000 4000 4000

NMR Data Point

Figure 2. A typical series of I-R weighted CPMG curvesfor a traditional butter, Figures 3 and 4 present the 2D VGand VR surfaces of these signals calculated with the grouping “Light’T’Traditional”, reflecting the moisture contents of the samples. Figure 3 shows that the greatest differences are in the middle of the I-R axis and at the beginning of the CPMG axis. This indicates that the TI of the samples changes with moisture content, shifting the null-point of the I-R curve and that the T2 or the quantity of the fast relaxing component is modified. On the other hand, Figure 4 shows that there is a significant pmportion of structured variability in the signal intensities which is not explained by the “Light”/“Traditional” grouping, especially towards the end of the CPMG signals. A detailed analysis of the CPMG relaxation curves using CONTIN and

207


MARQT Is showed that this difference is due to the longer relaxation times of the aqueous phase in the margarine samples.

0

0

Figure 3. 2 0 VGplotfiom ANOVA on I-R weighted CPMG curves, based on groups “Light’’/ “Traditional”,

0 0

Figure 4. 2 0 VRplot from ANOVA on I-R weighted CPMG curves based on groups “Light”/ “Traditional”,

While it is possible with the QP20+ to program complex pulse sequences in order to acquire 2D relaxation surfaces, many commonly used TD-NMRinstnUnents do not have this capability. In such cases it is possible to create artificial 2D relaxation surfaces by calculating the Outer Product of independently acquired I-R and CPMG curves, as shown in Figure 5.

Figure 5. Unfolded Outer Product matrix for a traditional butter. Prior to calculation, the CPMG curves were range scaledfrom 0 to 1

It is clear that these OP matrices do not contain all the information to be found in the true 2D signals. However they can be used to detect some of the simultaneous variations in the 2 signals. Figures 6 and 7 present the VC and VR surfaces of these OP matrices based on the grouping “Light”PITraditiona1”. The similarity between these s u r h e s &om the OP ANOVA and those obtained using the true 2D relaxation surfaces is evident, in particular at the beginning along the CPMG axis.


208

0 0

Figure 6. 2 0 VGplot of matricesfrom Outer Product of I-R and CPMG curves.

Figure 7. 2 0 VRplot of matricesfrom Outer Product of I-R and CPMG curves.

Having demonstrated that these two types of signal (2D relaxation surfaces and OP surfaces) contain significant information, PLS regressions were performed on them both to create predictive models for the moisture content of the samples. The evolution, as a function of the number of Factors included in the models, of the Y-Variances and the Durbin-Watson values of the X-Loadings and B-Coefficients vectors (see below) was used to determine that 3 was the optimal number of Factors in both cases to limit over-fitting. Figures 8 and 9 show the B-coefficients surfaces for the two models The similarity between these surfaces is again evident. Figures 10 and 11 are the regression lines obtained using the two models. It is clear that there is no significant difference between them.

0 0

0

Figure 8. 2 0 B-Coeflcients plotfiom PLS on 2 0 I-R / CPMG surfaces.

.. *.

..

0

Figure 9. 2 0 B-Coeflcients plot from PLS on OP matrix of I-R and CPMG curves.

- .. I

e;

o

Figure 10. Predicted vs. observed moisture from PLS on 2 0 I-R / CPMG surfaces. RZ=O.996, MSEC=8.22%

10

20

ao

40

LO

00

70

Figure 11.Predicted vs. observed moisture from PLS on OP matrices of I-R and CPMG curves. R2=0.974,RMSEC=7.20%

1.2.2. Plant and Vegetable Oils. The Iodine Number (IN) of 20 plant and animal oils of pharmaceutical quality (SociktC Industrielle des Oleagineux, Coopkration

209


Yharmaceutique Franqaise) were either taken directly ffom the accompanying Analytical Bulletins or calculated fiom the given fatty acid compositions. The IN values, which reflect the degree of unsaturation of the oils, ranged fiom 70 to 150. For each oil, 3 NMR tubes (10od) were filled to 1Omm. The samples were thermostated and measured at 20°C in the QP2W instrument. Several different types of pulse sequences were tested but the Analysis of Variance on the relaxation curves showed that only the TI curves contained information on the Iodine Number. This result confirms the correlation between IN and TI observed by Brosio et al. l6 and El Khaloui et al. " The following InversionRecovery sequence was used :

'-'

I-R = [180" -- 2 *(1.36) -- measure -- RD} N withN=30,i=l toN,RD=5s, Scans=4 Fourier Transform-Mared spectra were acquired fiom 345Ocm-I to 550cm-'for the same oils on an FTS60 spectrometer (BioRad) using a ZnSe single-reflectionHATR cell at room temperature, with 128 scans and a resolution of 16cm-'. The Analysis of Variance on the spectra gave a clearly structured VG plot with, as expected, particular vibrations strongly influenced by the Iodine Number (results not shown). For each sample, the Outer Product matrix was calculated between the I-R relaxation curve and FT-IR spectrum. PLS regressions with fiom 1 to 10 Factors were performed between the X-matrix of unfolded OP matrices and the Y -matrix of IN values. The evolution of the X- and Y-Variances, and the Durbin-Watson @) values of the XLoadings and B-Coefficientsare presented in Figure 13. 1.5

I

100

L

.

*

1 -

*

. 0

.

E

x

+

0

x

- 1

4

a a n

0 0

O

0.5 -

x

X

oi

T '0

'

+

'

.+.

0

+

+ '

'

'

'

'

'

'

0

+ '

'

'

3 2

-25 m (1A

0

+

4

a,

- 0.1 0

'

.

0

x

.+. '

.

7 0.01 '

'

+'

'

0.001

Number of Factors in PLS regression model

Figure 13. X-Variances (*), Y-Variances (0)and the D values of the X-Loadings (+) and B-Coeficients M, as afunction of number of Factors used in the PLS regression between the I-R /FT-IR Outer Product matrices and the Iodine Numbers. The sudden increase in the D values between 3 and 4 Factors indicates a significant increase in the randomness in the X-Loadings and B-Coefficients vectors due to the inclusion of noise to improve the adjustment of the PLS regression model. This corresponds to the point where the X-Variance abruptly drops below 10%. If the signals only contained noise and it were equally distributed among the 10 Factors, their X-


210

Variances would all be equal to 10%. Therefore any Factors with X-Variances greater than this value may be assumed to contain information. The D-values and Variances both indicate that a 3 Factor PLS model is optimal. Figure 14 shows the B-Coefficients surhce for the 3 Factor PLS regression between the I-R/FT-IR OP matrices and the Iodine Number while Figure 15 shows 2 profiles through the B-Coefficients surface, parallel to the FT-IRaxis.

Figure 14. B-Coeflcients sur$ace for the 3 Factor PLS regression between the I-WFT-IR Outer Product matrices and the Iodine Numbers.

0.6 -

Od -

f -1d " 3600

'

'CH2

" * * " _ .

3000

'

2600

~ ~ " ' " * " " " ' ' '

2000

1600

1000

SO0

Figure 15. Profiles through the B-Coeflcients surface in Figure 14, parallel to the FT-IR mcis at 109 ms (-- 4 and 14.9 s (A on the I-R axis.

Although in this case the use of the OP matrix did not signiftcantly improve the predictive ability of the PLS regression model compared to that of the FT-IR spectra


21 1

alone, it does have the advantage of introducing another dimension along which the signals can be resolved. As both Brosio and El Khaloui have shown, TI increases with the Iodine Number. Therefore oils with low IN values (fewer unsaturations) contribute more to the earlier part of the I-R curve while those with high IN values (more unsaturations) contribute more to the later part of the I-R curve. The vibrations in the Inhued spectra corresponding to different degrees of unsaturations are then resolved along this axis in the B-Coefficients surfice. 1.1.3. Starch-Lignin Mixtures. Starch was pre-extruded in presence of water before incorporation of lignin at 5 concentrations : 0, 5, 10, 15 and 30%. The starch and lignin powder were mixed and then extruded at 12OOC in presence of water to produce canes. Films were also produced by compressing the canes at 140°C and 250 bar for 10 minutes. Although it has not yet been verified by microscopy, this compression most probably results in a modification in the crystallinity of the samples. In order to have information about the behaviour of water in the samples, they were equilibrated for 10 days at 2loC, at 2 different Relative Humidities (33%, above a saturated solution of MgClz and 75%, above a saturated solution of NaCl). An adjustment to the G.A.B. sorption curves gave average moisture contents of 3.5% (MgC12)and10% (NaCl) for the mixtures. TD-NMR measurements were done, in triplicate and at 2OoC, on the 20 starch lignin mixtures using the QP2O-t. It was not known which, if any, of the almost unlimited number of possible pulse sequences would produce an informative signal concerning the moisture content, lignin content and form (cane or film)of these samples. Therefore, several different sequences were applied and the resulting signals analysed statistically. Because of the very low moisture content, the CPMG sequence could not be used and so a simple FID was used to characterise the proportions of "solid" and "liquid" protons and their apparent transverse relaxations (Tz*).Normal and Inverse Goldman-Shen (GS) sequences were used to see if there was any cross-relaxation (CR) between these two phases and, if so, whether it varied with the composition of the samples. An Inversion-Recovery sequence was used to observe the longitudinal (TI) relaxation while a Multiple Pulse - Spin Locking sequence was used to observe the rotating h e longitudinal (TI& relaxation. The instrumental parameters of these sequences are given below :

-

Free Induction Decay (Tz *): 90°, [r - measure1 - RD solid + liquid components : 100 points; fiom 11ps to 3 1ps liquid component : 100 points; fiom 3 1ps to 23 1ps signal averaged to 2*20 points RD=2s Normal and Inverse Goldman-Shen (CR): {9OoX- t, - 90°,, - rvar- 90°,, - t , - measure1 - t, - measure2 - RD), N=30 war = 0.001*(1.74) ms; for i = 1 to N t l = 30ps; t2 = 12 ps (solid + liquid); t3 = 69 ps (liquid) each point acquired is the average of 5 with dwell of 0.2 ps RD = 2s

212


Inversion-Recovery (TI): (180°, - rvar- 90°, - t2 - measure - RD}, N=20 war = 0.5*(1.55) i-' ms; for i = 1 to N t2 = 11 ps (solid + liquid) RD = 2s Multiple Pulse Spin Locking (TI&: 90°, - T - {(9OoY- 211, - 9OoY - 2.5 - 9ODy- T - measure - T) - RD N

N = 80; M = 2 T=

10 ps (solid + liquid)

RD = 2s averaged to 40 points The GS liquid signals were subtracted fiom the corresponding GS solid+liquid signals to clearly isolate the effects of any cross-relaxation and longitudinal relaxation on the two compartments. The 8 signals for each sample were then concatenated, range scaled fiom 0 to 1 and centred. The shape of the normal and inverse GS solid curves in the typical vector of concatenated TD-NMR signals shown in Figure 16 indicate that this sample does not present any cross-relaxation- the evolution of the curves would simply be the result of longitudinal relaxation. Figure 17 shows the complete matrix of concatenated signals for 58 samples (2 signals eliminated as outliers). Here it is clear to the eye that the signals in the data matrix contain information on the moisture level, the form of the samples and the lignin level (not shown).

-

-

7 0

0.4

m E

u)

2

0

t s

s -

0.4

u)

(3 -LyLL

0.6

0

20

40

60

80

100 120 140 160 180 200 220

2s+L* CR+T 1s -CR+T 1~ T IS+L T2 ~ * CR-T IS -CR-T IL T 1 ~ S + L Figure 16. A lypical vector of range-scaled, centred concatenated TD-NMR signals.

213


so

0

100

150

200

Figure 17. Data matrix for starch-lignin mixtures. Top curve is a typical signal vector, leji curve indicates moisture levels and right curve shows alternating cane andfirmforms.

This somewhat subjective appreciation is confirmed by ANOVA. In Figure 18, the VG and VR plots for the 3 ANOVAS, based on moisture, lignin and form show clearly that :1) the moisture level influences the liquid part of the FID, the liquid parts of the inverse GS, the end of the I-R, and the start and end of the MP-SL; 2) the lignin level influences the middle of the I-R, and the middle the GS solid signals; 3) the form has a very strong effect on the ends of the GS solid signals and the inverse GS liquid signal. The complementarity of the VRand VGplots makes it clear that these three factors are the only significant sources of variability in the signals.

1

.e

0 08

0.004

0 08

0 0035

0.07

0 003

000

0 0025

001

i 1:

0 001

00015 0 001

0 02

o.oms

0 01 0 1

I1

101

151

~Dat.PoMl

201

0 5t

101

151

201

mRmhpoh(s

Figure 18. VG (lefl) and VR(righr)plotsfor ANOVAS based on moisture (o),lignin ( A ) andform (*).

PLS regressions with from 1 to 10 Factors were performed between the X-matrix of concatenated TD-NMR signals and the Y-matrices of moisture and lignin contents, and form as binary values. The X- and Y-Variances and the D values of the X-Loadings and B-Coefficients are presented in Figure 19. The evolution of these curves shows that the optimal number of Factors to include in the three models, in order to limit the overfitting, are respectively 3, 3 and 2. Figure 20 presents the B-Coefficients for the three models. These curves are very similar to the VG plots in Figure 17, but include the sign of the contribution of the signals to the moisture, lignin or form values.

214


0

2

4

6

8

1

0

0

Number of PCs

2

4

6

0

1

0

Number of PCs

Number of PCs

Figure 19. X- Variances (4, Y-Variances (0)and the D values of the X-Loadings (0) and B-Coeflcients (4 as a function of number of Factors used in the PLS regression between the concatenated signals and moisture (lefl), lignin (centre) andform (right). For moisture, there is a positive contribution of the liquid part of the FID, a negative contributionof the normal and inverse solid GS and a positive contribution of the inverse liquid GS. The positive contribution of the middle of the I-R curve indicates a decrease in the TI of the samples with increase in moisture content. This decrease in TI has been confirmed by multiexponential decomposition of the I-R curves. For lignin, there is no contribution fiom any of the liquid signals. Adding lignin does not modify the state of the water in the samples. There is however an increase in the solid FID signal and a positive contributions in the middle of the normal and inverse solid GS and fiom the middle of the I-R curve. The addition of lignin to the samples only slightly increases the solid FID signal. The positive contribution of the middle of the solid GS and I-R curves indicates a decrease in the TIof the samples without a signifkant modification in the proportion of solid content. This decrease in TI has been also confirmed by multiexponential decomposition of the I-R curves. 0.04

I

1

51

101

161

201

Figure 20. B-Coeficients plots for the PLS regression between the concatenated signals and moisture (top), lignin (centre) and form (bottom).

215


The situation for the form of the samples is more complex. The change fiom cane to film is associated with a negative contribution fiom both the normal and inverse solid GS and the normal and inverse liquid GS. Similarly, there is a slight increase in both the solid and liquid FID intensities. This evolution is not easy to explain but may be due to a modification in the relaxation properties of both the solid and liquid components of the sampIes. This could explain the modification in the shape of the solid FID signal from lorentzian to more gaussian. Multiexponential decomposition of the 1-R curves shows a slight decrease in TI but only for the highest lignin contents. The observed and predicted values for the 3 PLS regression models are presented in Figure 21, along with the associated statistics. ln 0

-a 0.14

g4 E

0.12 0.1 0.08

$

2

40

1.s

30

1

20 0.08 0.04 0.02

0

0.5

10

;

s

50

0 1

11 21 31 41 51

Object Number

-10

0 1

11 21 31 41 51

Object Number

0.5

1

11 21 31 41 51

Object Number

Figure 21. Observed andpredicted values for moisture (lefi), lignin (centre) andform (righr). The statistics of the PLS regression models are moisture :RMSEC = 7.66%, R2 = 0.970 (3Factors); lignin :RMSEC = 13.93%, R2 = 0.954 (3Factors);form : RMSEC = 33.63%, R2 = 0.766 (2Factors). 1.3 Conclusions These results confim the interest of applying chemometric techniques directly to

TD-NMRsignals as a means of detecting information and to have an indication of the changes taking place. However, a complete understanding of the relaxation phenomena occurring in the samples still requires the calculation of relaxation parameters such as TI, ‘Tz and the cross relaxation rate. The statistical analysis of Outer Product matrices (Outer Product Analysis - OPA) can in many cases be useful to facilitate the comparison of variations occurring simultaneously in two sets of signals and to artificially increase the resolution of the signals. References 1. A. Davenel, P. Marchal, J.P. Guillement, “Rapid coolung control of cakes by low resolution NMR” in : Magnetic resonance infood science, P.S. Belton, I. Delgadillo, A.M. Gil, G.A Webb (eds), Royal Society of Chemistry, Cambridge, 1995, p.146.

2. A. Gerbanowski,D.N. Rutledge, M. Feinberg, C. Ducauze, Science des Aliments, 1997, 17,309. 3. C Airiau, F. Gaudard, A.S. Barros, D.N.Rutledge, Analusis, 1998,26,66.

216


4. M.C. Vackier and D.N. Rutledge, J. of Mag. Res. Anal., 1996,2(4) 321

5. M.C. Vackier and D.N. Rutledge, J of Mag. Res. Anal., 1996,2(4) 31 1. 6. N.J. Clayden, R.J. Lehnert, S. Turnock, Analytica Chimica Acta, 1997,344,261. 7. J. Czerminski, A. Iwasiewicz, Z. Paszk and A. Sikorski, Statistical Methods in Applied Chemisw, Elsevier, Amsterdam 1990, p. 186.

8. A. Hoskuldson, J. Chemometrics, 1988,211. 9. J. Durbin and .G.S. Watson, Biometriku, 1950,37,409. 10. F.E. Barton, D.S. Himmerlsbach, J.H. Duckworth, M.J. Smith, Appl. Spectrosc., 1992, 46,420.

1 1 . I. Noda, Appl. Spectrosc., 1993,47,1329 12. M.F.Devaux, P.Robert, A.Qannari, M.Safar, E.Vigneau, Appl. Spectrosc., 1993,47 , 1024. 13. A S . Barros, M. Safar,M.F. Devaux, D. Bertrand, D.N. Rutledge, Appl. Spectrosc., 1997,51(9), 1384. 14. N. Gouti, M-F Devaux, B. Novales, D.N. Rutledge, M. Feinberg, Analusis, 1998, (in press) 15. D.N.Rutledge and A.S. Barros, The Analyst, 1998,123,55 1. 16. E. Brosio, F. Conti, A. Dinola, S. Sykora, J. Fd. Technol., 1981, 16,67. 17. M. El Khaloui D.N.Rutledge, C.J.Ducauze, J. Sci. FoodAgri., 1990,53,389.

Comparative Chemometric Analysis of Transverse Low-field 'H NMR Relaxation Data Iben Ellegaard Bechmann, Henrik Toft Pedersen, Lars Nflrgaard and Sflren Balling Engelsen* FOOD TECHNOLOGY, THE ROYAL VETERINARY AND AGRICULTURAL UNIVERSITY, ROLIGHEDSVEJ 30, DK-1958 FREDERIKSBERG C, DENMARK

Corresponding author 1 ABSTRACT

Transverse relaxation data from low-field (23.2 MHz) 'H N M R was investigated by different data analytical approaches. The quantitative information in the relaxation data of 200 intact salmon samples with respect to overall water and fat content was evaluated by four different data analytical methods: (a) bi-exponential curve fitting followed by linear regression, (b) forward selection of variables followed by multiple linear regression (FSMLR), (c) partial least squares regression (PLS) and (d) non-negative alternating least squares regression (NN-ALSR). The investigation demonstrates that the quantitative prediction performance is significantly enhanced (reduction of the prediction error from 14% to 34%) by the use of multivariate chemometric procedures such as PLS. While PLS is an extraordinarily robust and efficient algorithm its strictly orthogonal latent variables suffer from a difficult qualitative interpretation. NN-ALSR and FS-MLR which exhibit a quantitative performance comparable to PLS do not suffer from this problem, but are more unstable and ineffective data analytical techniques. 2 INTRODUCTION

Fat and water are very important quality parameters for fish flesh, not only because of the nutritional importance of fish fat as a source of unsaturated fatty acids, but also because they influence most of the functional properties of the product. Increasing demands for quality, assurance and product quality documentation in the seafood industry have led to the need for rapid, simple, inexpensive and objective analytical methods for assessing seafood quality. The use of near infrared spectroscopy' has been proposed as a rapid method for assessing seafood quality but low-field pulsed 'H NMR provides a costeffective alternative method which is rapid, direct, volume-based, non-invasive and may be non-destructive. It has been shown that transverse water proton relaxation from lowfield NMR can be used to detect changes in fish muscles during frozen storage or processing*. Recently, we have demonstrated the great potential of using low-field NMR in combination with chemometrics as a rapid analytical technique for the determination of water, fat and water-holding capacity in intact fish flesh3. The application of chemometric data analysis in LF-NMR is only sparsely treated in the literature. Data structures produced by N M R and analytical chemical systems in

218


general can be divided into classes reflecting the complexity of the data, ranging from scalars, vectors, and matrices to higher order data structures. More than 90% of the papers published in analytical chemical journals are based on zero-order data structures4 (one datum per sample). The use of zero-order data requires that the measured signal is a known function of the property of interest and e.g. univariate linear regression is based on zero-order data. An improvement of the model can often be achieved by using first-order data structures4 which gives the possibility for detecting outlying samples. The relaxation data presented here are typical first-order data where each sample gives rise to a vector of intensities recorded at the same time-points and a set of calibration samples yields a matrix. LF-NMR papers published on first-order data using two-way data analytical methods, for instance PLS regression, have only recently begun to emerge5x6. The objective of this work was to investigate and compare the quantitative performance of four different data analytical strategies to low-field relaxation data. The reference point of the comparison will be the quantitative performance of onedimensional analysis assuming exponential decay functions. Of the other three methods two will use latent variables and one will use variables in original variable space. In two respects low-field NMR relaxation data is extreme: the co-linearity of the data is very high and the exponential type of decay functions are extremely non-linear. While the latter has impact on the convergence properties of curve fitting algorithms, the high colinearity can be considered as a challenge for the multivariate chemometric algorithms which are usually quite robust in handling highly co-linear data. Table 1 lists a comparison of the co-linearity of low-field NMR data ( 1 00 samples and 5 12 echoes) with typical NIR data spectra7 (98 pectin samples and 525 wavelengths) usually considered being highly co-linear. Table 1 Co-linearity of Low-field Relaxation Data Given in Percent Method of NTC t R

water content % (wlw d.s.b)

Figure 2 Storage conditions of processed starch systems in relation to the glassrubber transition line. 2 NMR MEASUREMENTS

All 'H NMR experiments were performed using a Bruker bench top PC120 (20 MHz) Minispec operating at 4 0 . 1 " C with recycle delays typically of 1 to 2s. The FID was fitted to 2 gaussians: a solid-like component with a T; of a few tens of ps and a liquid-like component with a Tz of a few hundreds of ps. The spin-echo decay acquired with a 9Oo-18O0 pulse spacing of 262 ps was fitted to a single exponential. The spin-latticerelaxation times (TI) were measured using the inversion-recovery pulse sequence with twenty 180"-90" pulse spacing values (2 to 4096 ms). The amplitudes of the FID recorded at 11 and 70 ps were used to describe the TI of the rigid (yll - y70) and the mobile (y70) components respectively. The inversion recovery signal was best described as a single exponential. In the majority of cases reported in this paper, 0 the y70 components suggesting a comparable TI values were recorded for the ~ 1 1 - ~ 7and cross-relaxation process.

283

Applications of Magnetic Resonance to Food Processing and Engineering

3 STARCH CONVERSION

3.1 Sample Preparation

Waxy maize starch, w m s (National Starchesand Chemicals Co.,Manchester, UK) was extruded using a Clextral BC-21 co-rotating, intermeshing twin screw extruder at 120°C. The water level was adjusted to obtain non-expanded extrudates containing 43% water (w/w dry solid basis). The samples were sealed and stored at room temperaturefor 4 weeks allowing “fiJll” retrogradation as measured by XRD6. This was performed in order to deconvolute the effect of crystallinity fiom that of the processing history. The sample were then dried (vacuum oven 7OoC,15 hours), ground and rehydrated over saturated salt solutions. The water vapour sorption isotherms (Figure 4) showed, for RH values greater than 45%, a higher water uptake for the processedretrograded samples compared with native non-extruded waxy maize starch. 40-

retrograded

A

B

z

A

0

20

40

60

80

100

Relatlve Humidity %

Figure 3

Water vapour sorption isotherms (29’C) of native andprocessed wms

3.2 Results and Discussion

The plot of TI versus water content showed a pattern very similar to the familiar TI versus temperature plot with a clear minimum at -15% water (w/w d.s.b). At this water content, Tgvalues of 60°C and 70°C were reported for partially crystalline starch by DSC and spin-spin NMR relaxometry respectively’. Therefore, the TI minimum is occurring at approximately 30°C below the Tg measured by DSC. This phenomenon has been observed for other biopolymer systems* and is attributed to the fact that water retains a relatively high of mobility in biopolymer-waterglasses. The effect of processing on starch was clearly detectable in both spin-lattice and spin-relaxation results (Figure 4). The extruded then retrograded system showed TI values consistently smaller than the native control. Another interesting feature of the spin lattice relaxation results is the broader minimum of the plot of TI versus water

284


content for the processed sample which indicate a broader distribution of correlation times around the value T ~ -5. lo-' s calculated using the BPP equation. The CPMG spin-spin relaxation times of the extruded sample were slightly higher than the control values particularly for water contents >15%. The decrease in the TZ at high water content could be related to the onset of fast exchange.

b 2

0'

60

0

5

10

15

20

25

30

35

40

45

Water Content % (w/w dsb)

0

"

5

10

"

IS

20

"

25

30

" 3S

40

WaterCoateat %(w/w dsb)

Figure 4 Eflect of the processing on (a) the TIand (b) the Tzfor a range of water contents. The empty triangle in the TI plot represents the value measured on the wet sample before drying, The lines in both charts were added to ease viewing. 4 STARCH RETROGRADATION

4.1 Sample preparation

Non-expanded wms-water systems were extruded as described earlier. The samples were stored in sealed 8 mm NMR tubes at 4OoC +O. 1 in the spectrometer probehead. 4.2 Results and discussion

Both the spin-spin relaxation parameters recorded from both FID and CPMG decays and the spin-lattice relaxation times showed a strong dependence on the duration of storage. The implication is that the NMR properties were affected by the retrogradation process and by the extent of the reordering of the gelatinised starch'. The recrystallisation of amylopectin was accompanied by a decrease of the TZ of the solidlike component of the FID (Figure 5a) and an increase of the contribution of this same component to the total signal (Figure 6b) indicating that the increased crystallinity leads to a reduced molecular mobility. The increase of the Tz of the liquid-like component of the FID (Figure 5a) with the progress of retrogradation could be explained by the fact that an increasing population of starch protons that had a liquid-like behaviour in the freshly gelatinised starch becomes more rigid in the recrystallised system and contributes therefore to the rapidly decaying component6. The total NMR signal showed no dependence on the storage time indicating that, within the experimental error limits, there was no noticeable moisture change (Figure 5b). The spin-echo TZmirrored that of the solid component of FID. This behaviour is in agreement with the theoretical

Applications of Magnetic Resonance to Food Processing and Engineering

285

calculation relating the spin-spin relaxation rates of the water to that of the rigid polymer matrix lo. 425

a

A

9 -4QJ

-

375

s

z15, P

25l 0

"

20

"

40

Storage time Q

"m 60

0.15 0

20

40

60

Storage time ( I )

FID Tin-pin relaxationparameters as afinction of storage time (400C) Figure 5 for a 100:60 wmshvater extruhtes. The T1 results recorded on a comparable system (wmdwater 100:65) increased with retrogradation (Figure 6b). This behaviour is somewhat surprising, particularly that these systems, in the experimental conditions of this study are on the right of the TI minimum (TIversus temperature or water content), e.g. were T1 shows a positive dependence on molecular mobility. No significant difference was found between the relaxation times of the component recorded at llps and that at 70ps suggesting that water provides a relaxation path for the starch protons'.

Figure 6 Changes in the relaxation times (a) T2for a 100:60, and (b) TI for a I00:65 wmshvater extruhtes during retrograahtion.


286

The changes in all NMR relaxation parameters due retrogradation reached plateau values after approximately 20 h. This kinetic is in agreement with wide angle xray results obtained on the same samples (Figure 7).

o

5

10

m

15

25

30

20

XRD spectra recorded during the ageing-retrogradtion (25+2 T)of a Figure 7 100.60 wmshvater extrudale.

5 HYDRATION OF MIXED SYSTEMS

5.1 Sample preparation

Maize grits (Maizecor Foods Ltd.) were used for this study. The specifications from the supplier indicated 8-9.5% protein and a maximum 1% lipids. The starch was composed of approximately 25% amylose and 75% amylopectin. Maize and maize-sucrose 100:10 were extruded at 22 YOwater (wlw d.s.b) and a temperature of 170°C. After the expanded extrudates had been cooled, they were ground to fine powders, dried and rehydrated as described earlier lo. 5.2 Results and discussion

For water contents 47% for the maize and

Advances in Magnetic Resonance in Food Science (Special Publications)

Magnetic Resonance Imaging in Food Science

Magnetic Resonance in Food Science: From Molecules to Man (Special Publications)

Recent Advances in Actinide Science (Special Publications)

Magnetic Resonance in Food Science Challenges in a Changing World

Advances in Magnetic and Optical Resonance, Volume 19 (Advances in Magnetic and Optical Resonance)

Advances in Magnetic and Optical Resonance, Volume 19 (Advances in Magnetic and Optical Resonance)

Magnetic resonance in colloid and interface science

Magnetism and Magnetic Resonance in Solids

Magnetism and magnetic resonance in solids

Magnetism and Magnetic Resonance in Solids

Advance image processing in magnetic resonance imaging

Nuclear Magnetic Resonance

Principles of Magnetic Resonance

Magnetic resonance imaging

Magnetic Resonance Imaging in Ischemic Stroke

Magnetic Resonance Imaging in Orthopedic Sports Medicine

Nuclear magnetic resonance in solid polymers

Nuclear Magnetic Resonance Spectroscopy in Environmental Chemistry

Coronary Magnetic Resonance Angiography

Coronary Magnetic Resonance Angiography

Modern Magnetic Resonance

Magnetic resonance imaging

Nuclear Magnetic Resonance

Cardiovascular Magnetic Resonance Imaging

Cardiovascular Magnetic Resonance

Magnetic Resonance Tomography

Modern Magnetic Resonance

AGRI-FOOD QUALITY II (Special Publications)

Food Colloids:: Self-Assembly and Material Science (Special Publications)

Advances in Magnetic Resonance in Food Science (Special Publications)

Magnetic Resonance Imaging in Food Science

Magnetic Resonance in Food Science: From Molecules to Man (Special Publications)

Recent Advances in Actinide Science (Special Publications)

Magnetic Resonance in Food Science Challenges in a Changing World

Advances in Magnetic and Optical Resonance, Volume 19 (Advances in Magnetic and Optical Resonance)

Advances in Magnetic and Optical Resonance, Volume 19 (Advances in Magnetic and Optical Resonance)

Magnetic resonance in colloid and interface science

Magnetism and Magnetic Resonance in Solids

Magnetism and magnetic resonance in solids

Magnetism and Magnetic Resonance in Solids

Advance image processing in magnetic resonance imaging

Nuclear Magnetic Resonance

Principles of Magnetic Resonance

Magnetic resonance imaging

Magnetic Resonance Imaging in Ischemic Stroke

Magnetic Resonance Imaging in Orthopedic Sports Medicine

Nuclear magnetic resonance in solid polymers

Nuclear Magnetic Resonance Spectroscopy in Environmental Chemistry

Coronary Magnetic Resonance Angiography

Coronary Magnetic Resonance Angiography

Modern Magnetic Resonance

Magnetic resonance imaging

Nuclear Magnetic Resonance

Cardiovascular Magnetic Resonance Imaging

Cardiovascular Magnetic Resonance

Magnetic Resonance Tomography

Modern Magnetic Resonance

AGRI-FOOD QUALITY II (Special Publications)

Food Colloids:: Self-Assembly and Material Science (Special Publications)

Recommend Documents