Molecular spectroscopy in biodiagnostics (from Hippocrates to Herschel and beyond)

Molecular spectroscopy in biodiagnostics (from Hippocrates to Herschel and beyond)

Journal of MOLECULAR STRUCTURE ELSEVIER Journal of Molecular Structure 347 (1995) 187-206 Molecular Spectroscopy in Biodiagnostics (From Hippocrate...

1MB Sizes 0 Downloads 8 Views

Journal of MOLECULAR STRUCTURE

ELSEVIER

Journal of Molecular Structure 347 (1995) 187-206

Molecular Spectroscopy in Biodiagnostics (From Hippocrates to Herschel and Beyond) Henry Mantsch and Michael Jackson Institute for Biodiagnostics, National Research Council of Canada, 435 Ellice Avenue, Winnipeg, Manitoba, R3B 1Y6, Canada. After two decades of intense research on the spectroscopic properties of biological molecules in isolated systems, infrared spectroscopy is now being applied to the study of human tissues. Extending this approach, it is possible to use the sensitivity of infrared spectroscopy to probe the biochemical events underlying transformation from normal to a diseased state within tissues, and so develop novel diagnostic methods. We highlight some of the areas of research within our group aimed at developing clinically useful methodologies based upon infrared spectroscopy.

I. INTRODUCTION The development of the Fourier transform infrared spectrometer presented structural biologists with a powerful tool for characterising the complex molecules which constitute the building blocks of life. In the last two decades a great deal of information has been obtained by FTIR spectroscopic methods relating to lipid polymorphism (1, 2), protein structure-function relationships (3, 4) and the structural properties of DNA and RNA (5). Of course, as instrumentation improved and experimentalists became more adventurous, interactions in binary and ternary mixtures of these important molecules were studied with great success, generating data on assemblies closer to real-life systems. Yet, information on real-life systems only comes from the analysis of lipids, proteins or nucleic acids in their physiological environment, that is, in cells and tissues. That obtaining such information by infrared spectroscopic means is more challenging is amply demonstrated by the lack of infrared studies on cells and tissues reported in the literature. However, more recently the wealth of information obtained over the last two decades from studies of isolated biomolecules is at last being applied to the analysis of human tissues (6, 7). We believe that this approach may be taken one step further, and that infrared spectroscopy now can be applied to the study of disease states in humans. The rationale for this is straightforward. All disease states, without exception, are caused by fundamental changes in cellular and/or tissue biochemistry. Current diagnostic methods rely upon detection of these abnormalities either through clinical chemistry techniques or the appearance of some defined physiological response which, by virtue of the experience of the physician, is correlated with a particular disease state. The result is often a highly 87-206

0022-2860/95/$09.50 © 1995 Elsevier Science B.M SSDI 0022-2860(95)08545-9

All ,ignts reserved

188 subjective diagnosis. It therefore becomes imperative to find means to detect the biochemical changes by a non-subjective method, preferably an instrumental method of analysis. That detection of these changes by such methods is indeed possible is demonstrated by the success of PET, CAT and MRI scanning techniques. In principle, infrared spectroscopy has many advantages over these more established techniques, including low cost (tens of thousands of dollars as opposed to millions), no need for external perturbation and no dependence upon physical state. Furthermore, changes in tissue biochemistry must precede any morphological or symptomatic manifestations, thus allowing spectroscopic diagnosis at an earlier stage of the disease. Current work aimed at developing diagnostic techniques based upon infrared spectroscopy will be discussed.

2. INFRARED-ACTIVE CONSTITUENTS OF HUMAN TISSUES Whereas the assignment of absorptions in the infrared spectra of simple organic molecules can be achieved with a high degree of certainty based upon theoretical studies, in more complex molecules - such as those comprising biological tissues - theoretical studies are at best impracticable. In fact, in the poorly defined (in chemical and physical terms) environment of tissues such an approach becomes impossible. For this reason the discussion of infrared bands in human tissues must rely on extrapolation of infrared spectra of the major components of human tissue (lipids, proteins and polynucleotides) recorded in isolation and in complex assemblies. In principle, the IR spectrum of most tissues should be closely approximated by summation of these three spectral types weighted according to relative concentrations (with obvious exceptions such as calcified tissues). With this in mind, the major spectral features characteristic of each of these building blocks will be discussed. The mid-IR spectrum of a typical phospholipid (dimyristoyl phosphatydylcholine, or DMPC) is shown in Fig. 1A. The dominant absorptions in this spectrum - as in the spectra of all lipids - are found in the region 2800-3000 cm -1 and by analogy with IR spectra of alkanes, are assigned to the asymmetric and symmetric stretching vibrations of CH 3 (2956 and 2874 cm -1) and CH 2 (2922 and 2852 cm-1) groups of alkyl/acyl chains. As expected, given the greater number of methylene groups in phospholipids, the intensity of the CH 2 absorptions is some 10-20 times that of the corresponding CH 3 absorption. In model systems (i.e., liposomes) the position and width of the CH 2 and CH 3 stretching absorptions can provide important information regarding the packing characteristics of the acyl chains, which in turn may be related to the degree of fluidity of the membrane lipids (1, 2). In spectra of nucleic acids (Fig. 1B) and proteins (Fig. 1C) the intensity of absorptions in this spectral region is drastically reduced. Only weak absorptions due to the CH 2 stretching vibrations of the carbohydrate residues and the C-H stretching vibrations of the nucleotide bases are seen in the spectra of polynucleotides. The weak features in the spectrum of proteins in this region reflect the almost equal proportion of CH 2 and CH 3 groups in protein side chains.

189 The region between 1800-2800 cm -1 is generally devoid of absorptions in biological materials. Infrared bands in the region 1600-1800 cm -1 are typically attributed to C=O stretching vibrations. The band at 1736 cm -1 in the spectrum of DMPC arises from the stretching vibration of the ester C=O groups of this lipid. In simple lipid systems this vibration is indirectly sensitive to phase behaviour due to hydration-induced changes that accompany phase transitions (1, 2). In nucleic acids, typically two major absorptions are seen in this region (at 1717 and 1666 cm -1 in Fig. 1A), which are attributed to the C=O stretching vibrations of the purine and pyrimidine bases (5). The absorption at 1717 cm -1 is a sensitive probe of base pairing. In proteins, the major absorption in this region is the so-called amide I band which arises predominantly from the C=O stretching vibration of the amide C=O. It has been repeatedly demonstrated that this absorption is conformationally sensitive and can be used to predict protein secondary structure (3,4). Two other amide modes are also used to assess protein conformation, the amide II absorption (predominantly an N-H bending vibration coupled to C-N stretching, which is seen between 1500-1560 cm -1) and the amide III absorption (attributed to a complex vibration involving C-N stretching and N-H in-plane bending with significant contributions from CH 2 wagging vibrations, which occurs at 1250-1350 cm-l).

1547__~ ,

1656

A II

f

II

1088

.

1 6 6 6 A "~--

1717

2922

1230 '~

2852 1468

~

--~-

1736 2874.--~

~..,,~-.2 9 5 6

1000 12~50 1500 17~50 2600 2250 2500 2-750 3000 Wavenumber, cm --1

Figure 1. A: IR spectra of a representative phospholipid (DMPC), B' of a typical DNA fragment, and C: of: a typical globular protein (human haemoglobin). The region between 1250-1500 cm q is populated by a number of sharp but weak absorptions. A relatively strong absorption band, particularly in the spectra of lipids, is the CH 2 scissoring vibration (at 1468 cm'l). The major absorption bands in the 1000-1250 cm "1 region arise from vibrational modes of phosphate groups. In nucleic acids, the phosphodiester linkages of the

190 polynucleotide chain lead to two strong IR bands, the antisymmetric (1224 cm"1) and symmetric (1087 cm-1) PO 2- stretching vibrations. In RNA, an additional PO 2- stretching absorption appears at 1240 cm-1 (5). In phospholipids, the exact frequency of the PO 2bands provides important information concerning headgroup hydration. In proteins (Fig. 1C), this region contains no strong IR bands. In addition to lipids, proteins and nucleic acids, carbohydrates are present in all cells. The major absorptions from carbohydrates are found in the 1000-1200 cm -1 region of the spectrum and are attributed to C-O stretching vibrations. Significant carbohydrate absorptions are only expected in those tissues with large carbohydrate stores (e.g., the liver), or in tissues which contain high proportions of connective tissue.

3. GENERAL HISTOLOGICAL CONSIDERATIONS It is evident from the brief discussion presented above that the infrared spectrum of a sample of human tissue will be complex, with considerable overlap of absorptions from the major species present. Correct assignment of these absorptions requires an understanding of the composition of the tissue. While the cellular and chemical composition will vary from tissue to tissue, certain constituents are common to all human tissues. Thus, lipids, protein and nucleic acids will be found in all tissue, but their relative proportions will vary. The effect of such variation can be illustrated by discussion of the spectra of human central nervous system tissue. To the untrained eye, the human brain is remarkably homogeneous. Visual inspection reveals two distinct types of tissue, referred to as grey matter and white matter. The IR spectra of brain tissue also fall into two broad categories, reflecting differences in the histology of grey and white matter. Grey matter consists predominantly of neuronal cell bodies. White matter, on the other hand, consists predominantly of the conducting protrusions (axons) of the neurons, surrounded by a lipid-rich insulating layer (the myelin sheath) which gives white matter its characteristic colour. These histological differences should give rise to significant differences between their IR spectra, the spectrum of white matter exhibiting more intense lipid absorptions than that of grey matter. This is in fact what is observed (7). The intensity of the CH 2 and CH 3 asymmetric and symmetric stretching absorptions at 2800-3000 cm"1 are significantly more pronounced in spectra of white matter. The major lipid absorption between 10001800 cm -1 is seen at 1467 cm -1 and is assigned to the CH 2 scissoring vibration of the acyl chains. As expected, given the differences in lipid content, the intensity of this absorption is much greater in white matter than in grey matter. In addition to these pronounced differences, other, more subtle differences are apparent. For instance the amide I:amide II ratio is greater in grey matter than in white matter. It may be speculated that this reflects some difference in the protein content or the environment of the proteins in the two tissues. However, a more acceptable explanation is easily found if one considers the chemical composition of the tissues. Of particular importance in this respect is water, as water has a significant absorption in the region of the amide I mode. Variations in the water content of tissues, and so in the intensity of the

191 O-H bending absorption, may be expected to result in variations in the apparent intensity of the amide I mode. The greater the water content of the tissue, the greater the apparent intensity of the amide I band and so the greater the amide I:amide II ratio. In line with this reasoning, the greater amideI/amide II ratio in grey matter reflects the greater water content in grey matter than in white matter (80% v s 72%). The examples discussed above highlight the usefulness of a knowledge of tissues on a chemical and cellular level. However, one component that is found in all tissues (except blood and most areas of the central nervous system), namely the connective tissue matrix which holds all tissue together, is often overlooked by biological spectroscopists. The connective tissue matrix is composed of a cellular element (fibroblasts) and the proteins which these cells secrete. The major connective tissue protein (indeed the major protein in the body accounting for 60% of body protein and 15% of total body mass) is the glycoprotein collagen, a unique triple stranded protein with great mechanical strength. In view of the high collagen content of tissues it is likely that significant absorptions arising from collagen will be present in the IR spectra of tissues, and an understanding of the spectroscopic properties of collagen is desirable. ~o eo

o0

~

I

e,4

~

I

B

A

lObO 1160 1200 1300 14-b0 1500 1600 1700 1800 wavenumber, cm --1

Figure 2. Infrared spectra of human haemoglobin (A) and type I collagen (B). The spectrum of type I human collagen (Fig. 2) shows a number of relatively intense absorptions (marked with an asterisk), not apparent in spectra of other proteins such as haemoglobin. We have assigned these absorptions (8) to the amide III vibration with significant mixing with CH 2 wagging vibrations (1204, 1236, 1282 and 1337 cm -1) and to the C-O stretching vibrations of the carbohydrate residues (1031 and 1082 cm "1) of collagen. These absorptions are not present in spectra of white or grey matter but are observed in the choroid plexus and arachnoid vessel, specialised regions of the central nervous system responsible for the secretion and reabsorption of cerebrospinal fluid. This observation is consistent with the histology of these areas of the central nervous system.

192

While most areas of the central nervous system are devoid of connective tissue, and so collagen, the choroid plexus and the arachnoid vessel are among the limited central nervous system regions which do have a significant connective tissue matrix. Thus, the choroid plexus and arachnoid vessel show significantly different spectra compared to other regions of the brain, due principally to the presence of collagen (8). The presence of connective tissue in almost all human tissues can therefore be seen to have potentially important spectroscopic consequences, particularly when one considers that major contributions from collagen occur in the spectral regions in which characteristic absorptions from DNA, RNA and phospholipids are seen. The implications of the presence of collagen absorptions for the diagnosis of disease states from IR spectra will be discussed in more detail later.

A

B I

I000

I

I

1100

I

I

1200

wevenumber,

I

I

I

1300

I

I

I

14-00

cm

--I

Figure 3. IR spectra of two spatially distinct samples of a xenografled human breast tumour (A) and two spatially distinct regions of the same sample of the tumour recorded with an infrared microscope (B). As discussed, variations in collagen content between tissue types exist and must be taken into account in a spectroscopic assessment of human tissues. However, significant variations in the distribution of collagen within the same piece &tissue also occur. Fig. 3A illustrates the variation in collagen content which may be observed between two small samples taken from spatially distinct regions of the same xenograffed human breast tumour (produced by the inoculation of a nude mouse with human breast tumour cells). Absorptions marked C are attributed to collagen, and it is apparent that variations in collagen content are apparent, suggesting heterogeneity of the tissue on a macroscopic level. In addition to variations in collagen content on a macroscopic level, significant variations are also seen on a microscopic level. Microscopic variation is best assessed with an infrared microscope, which effectively functions as a beam condenser to allow spectra

193

to be taken of very small samples (of the order of 20 ~tm2). We have used an infrared microscope to assess the variation of collagen content on the microscopic level. Fig. 3B shows two spectra obtained from a thin microtomed section of a xenogratted breast turnout. Whereas the spatial separation of the samples giving the spectra in Fig. 3A was of the order of millimeters, the spectra shown in Fig. 3B were recorded from regions of tissue separated by only 20 ~tm. It can be seen that similar variations in collagen content are seen on the macroscopic and microscopic level. The variation in sample histology on the macro- and microscopic level has important consequences. Variation on a microscopic level means that if the area of tissue sampled is too small, an incorrect picture of the true histological composition of the sample will be obtained. If micro-sampling is to be used, then the whole tissue section should be sampled. With the availability of computer controlled stages for IR microscopes this presents no problem other than the longer time required for data collection. However, it does allow a complete spectroscopic picture of the tissue to emerge. Similarly, macroscopic variation suggests that spectra of a number of samples of a tissue specimen are required to provide a true picture of the tissue biochemistry.

J

lOOO

i 120o 1400 wclverq~lmb,~

1 4 6 5 ~ ' ~ ~ i 1600 -1 r-- , cm

\B

L

D

1800

Figure 4. Infrared spectra of a representative breast tumour (A), a breast tumour with significant adipose tissue content (B), the material remaining on the transmission windows after recording spectrum B and of a tissue oil droplet (C). Although collagen appears to be the major source of interfering absorptions in human tissues, it is not the only source of spectral "contamination". The spectrum of a typical human breast tumour is shown in Fig. 4A. The spectrum of a second tumour is shown in Fig. 4B and significant differences are apparent between the spectra of the two tumours. In particular, a strong ester C--O absorption is seen at 1748 cm -1, together with a prominent CH 2 bending vibration at 1465 cm "1. After dismantling the cell used to record the spectrum in Fig. 4B, an oily liquid was retained on one of the transmission windows,

194 which gave rise to the spectrum shown in Fig. 4C. From the spectral features exhibited by this oily material it appears to be a mixture of triglycerides and phospholipids and presumably arises from the presence of a significant proportion of adipose material stored in the second breast tumour sample. This is confirmed by a microscopic analysis of the tissue. A considerable number of small microscopic oil droplets were present in the tissue. The spectrum of one of these droplets obtained with an infrared microscope is presented in Fig. 4D, and is almost identical to that shown in Fig. 4C These spectra strongly suggest that the spectroscopic features present in Fig. 4B which are not observed in other breast tumours are related to the presence of a significant adipose tissue content. This brief discussion of the consequences of variations in histology of tissue samples highlights the care with which spectroscopic studies of human tissues should be undertaken. In order to fully appreciate the significance of variations in spectra, an understanding of the composition of the sample is essential. Although the composition of many tissues can be determined intuitively, at least to some extent, based upon a spectroscopic analysis, we recommend close collaboration with a histologist or pathologist.

4. DIAGNOSIS OF DISEASE STATES With an understanding of the composition of the tissues or biofluids to be studied, it now becomes feasible to begin to use IR spectroscopic methods for the identification of pathological changes in tissues and/or biofluids. For convenience we have identified three distinct methodological approaches. (i) A first approach, which could be referred to as IR clinical chemistry, involves the spectroscopic analysis of common biological fluids such as urine or blood, but also of less common biofluids such as synovial fluid, saliva, or sweat. (ii) A second, more complex approach, which could be referred to as IR pathology, involves the e x vivo analysis of human tissues, generally obtained from biopsied material. (iii) Finally, the most complex approach involves the in vivo diagnosis of disease states. Examples for each approach will be discussed. 4.1 Infrared clinical chemistry: Arthritis Arthritic disorders are among the most common disabling diseases in the developed world, affecting 4 million people in Canada alone (out of a total population of about 25 million). The economic and social costs of the disease are enormous, arthritis accounted for 44.5 million lost working days in Canada in 1992 and is the major cause of premature retirement. Yet, arthritis remains a difficult disease to diagnose and particularly difficult to stage. Current diagnostic methods rely upon X-ray investigations, coupled to the experience of a rheumatologist. In addition, clinical laboratory tests are often required. Diagnosis by these methods involves (i) considerable subjectivity (being dependent upon the experience of the physician), (ii) is usually only possible when considerable irreversible damage to joint has already occurred, and (iii) is costly and time consuming. It is apparent that a non-subjective method for the rapid diagnosis of arthritis would be of great benefit. The challenge then, is to find infrared spectroscopic methods for the diagnosis of arthritis.

195

The first and most logical step in the search for a spectroscopic diagnostic method is to analyse the fluid which is found inside the joint. This fluid (synovial fluid) is essentially a plasma filtrate, formed by filtration of plasma across the synovial membrane. In addition to the normal plasma analytes, synovial fluid also contains hyaluronic acid, a complex polysaccharide secreted by the synovial membrane, which endows the synovial fluid with a high viscosity and allows it to function as an effective lubricant. Significant changes in the composition of the synovial fluid accompany the transition from a healthy to an arthritic joint, including a reduction in the viscosity and an increase in the volume of the fluid. As synovial fluid is routinely sampled by physicians (for diagnostic purposes and to alleviate discomfort due to the swelling associated with the increased volume of fluid within the joint), the fluid is readily available for a spectroscopic analysis. The question remains as to whether spectroscopic analysis will prove diagnostic. Our hypothesis was that the changes in the composition of the synovial fluid could be detected by infrared spectroscopy, and that therefore infrared spectra of synovial fluid could be used in the non-subjective diagnosis of arthritic disorders. To this end, we have undertaken an extensive study into the spectroscopic characteristics of human synovial fluid from a variety of arthritic conditions.

A5 I

I

1000

1500

2000

2500

WGvenumber,

3000 cm

3500

4000

-1

Figure 5. Infrared spectra of water (A), normal synovial fluid (B) and synovial fluid from an osteoarthritic knee (C). The infrared spectra of normal (control) and osteoarthritic synovial fluid, together with the spectrum of water, are shown in Fig. 5. It is immediately apparent that the major absorptions in all three spectra are due to water itself, with only weak absorptions in the 1000-1600 cm -1 range arising from the non-aqueous synovial fluid components. In spectra dominated by absorptions arising from water it is common practice to interactively subtract the spectrum of water from the sample spectrum to generate a spectrum of the analytes present. Subtraction of water absorptions from aqueous samples requires the

196

assumption that the water in the sample closely approximates bulk water in spectroscopic terms. Experimentally, spectra of water and the fluid must be measured under identical conditions (pathlength, temperature etc.) Only under these conditions will subtraction of water absorptions result in artifact free spectra. Subtraction of water from aqueous samples is best achieved by monitoring the water absorption at around 2100 cm -1. Correct subtraction of water results in a flat baseline between 1900-2500 cm "1. The presence of positive or negative features in the resulting difference spectrum around 2100 cm-1 suggests incomplete or over-compensation of water, while the appearance of derivative-type artifacts implies that the spectroscopic characteristics of tissue water are not the same as those of bulk water. For synovial fluid spectra, it was found to be impossible to obtain an artifact free subtraction applying these criteria, due to disruption of the solvent water structure by the macromolecules present in the fluid. Thus the strong water absorptions of synovial fluid present a significant problem. This problem may be alleviated in two ways. The first method involves removing the water by formation of a thin film. Although this may seem an unusual method to employ for the analysis of biological fluids, most of the chemical information is retained. In addition, spectral reproducibility is greatly enhanced (problems associated with detector linearity at high absorbance values and reproducibility of pathlength are avoided). Valuable information concerning solute hydration and perturbation of water structure is of course lost, but additional information concerning the solutes present within the fluid is gained. In most instances, the advantages of such a procedure will outweigh the disadvantages.

pro;in NH

proteinC=O --I~ 1 proteinNH

~ I

4-000

id

I

I

I

I

I

I

5500

3000

2500

2000

1500

1000

wGvenumber,

crn

I

-I

Figure 6. Infrared spectrum of a film formed by drying of synovial fluid. The infrared spectrum of a film formed by drying of synovial fluid is presented in Fig. 6. The attributions of the major absorptions in the spectrum are noted on the figure. An unusual feature in this spectrum, in a region typically devoid of absorptions in biological/biomedical samples, is a weak absorption at 2337 cm-1 (9). A similar absorption

197

has been reported for C O 2 trapped within cages of ice (10). Of more physiological relevance, CO 2 found within germinating seeds has been demonstrated to give rise to absorptions in this region of the spectrum (11). Therefore this feature must originate from CO 2 trapped within the film. In terms of diagnostic utility, the level of CO 2 appears elevated in spectra of fluids from arthritic joints. Three questions arise at this point: 1: What is the source of this CO 2 2: What is the location of the CO 2 and 3: Why do CO 2 levels appear to have some correlation with arthritic diseases? Of course, the answers to all three questions are likely to be related. In terms of the source of the CO 2, the most likely explanation seems to be simply metabolic activity, as metabolic activity is the primary source of CO 2 within the body. The location of the CO 2 is less intuitive. The unusual composition of the synovial fluid may be related to the presence of this unique feature, the complex protein-carbohydrate matrix potentially providing numerous sites for enclathration of small molecules. However, dilution of synovial fluid with distilled water and freeze-thaw cycles lead to the disappearance of the band, a phenomenon which would not be expected if the CO 2 was enclathrated within the protein-carbohydrate matrix. There is an alternative explanation if one considers the effects of freeze-thaw cycles and dilution. Both of these procedures will result in lysis of cells present in the fluid, causing release of intracelllular contents. This implies that the CO 2 may be contained within cellular elements of the fluid, a suggestion which appears quite likely if the source of the CO 2 is indeed metabolic activity. The final question now remains to be answered: why do levels of CO 2 in synovial fluids show a tendency to be elevated in arthritis? In many instances, arthritis is accompanied by inflammation of the affected joint. This is particularly true of rheumatoid arthritis. Inflammation in the joint will be accompanied by infiltration of cells into the joint as part of the immune response, and the metabolic activity within the joint will be increased. As a result, the level of metabolically produced CO 2 in an inflamed joint will be elevated. It should be stressed that this explanation does not provide a diagnostic marker for arthritis, rather the level of spectroscopically determined CO 2 is a marker for inflammation. A diagnosis of the specific nature of the arthritic disorder from synovial fluid films requires a more sophisticated analysis of the spectra. Such a sophisticated analysis may take the form of a spectroscopic interpretation of the data, involving assignment of the major spectral features and interpretation of any changes accompanying the disease process. While this is an approach we are pursuing, for clinical usage this is unacceptable as it requires a skilled spectroscopist. A non-subjective methodology would be most useful. This can be achieved by the application of multivariate statistical analyses to the spectra, for example by using linear discriminant analysis. The results of a linear discriminant analysis based upon principle component analysis of spectra of synovial fluid films for control and three arthritis conditions is shown in Table I. The best results for this analysis were obtained using the CH stretching region of the spectrum, (2800-3050 cm-l). Utilising this region alone, spectra of synovial fluid may readily be classified as arising from joints clinically classified as control, osteoarthritic, rheumatoid arthritic or suffering from spondyloarthropathy. For a total of 62 patients (with 168 aspirates), only one misclassification occurred, which gives an overall success rate of greater than 98%. We are currently expanding our analysis to include a number of other arthritic

198 disorders and have met with considerable success to date. Thus, the combined application of infrared spectroscopy and multivariate analytical methods provides a rapid (of the order of minutes), non-subjective and accurate method for the diagnosis of the major forms of arthritic disorders. Combined with the low cost, minimal operational training and low service requirements of FTIR instrumentation these results suggest that FTIR spectroscopy may find a valuable place in the diagnosis of arthritic disorders in the general practitioners office, as well as in clinical chemistry laboratories.

C RA OA SPA

C 6 0 0 0

RA 0 25 1 0

OA 0 0 18 0

SPA 0 0 0 12

%Accuracy 100 100 96.2 100

Table I. Results of linear discriminant analysis on synovial fluid film spectra. C=control, RA=rheumatoid arthritis, OA=osteoarthritis, SPA=spondyloarthropathy. Rows represent multivariate classification, columns represent clinical diagnosis. Off-diagonal elements are misclassifications. A second way to alleviate the problems associated with the strong absorptions of water is to move into the near infrared region of the spectrum. This approach provides a more reproducible and reliable method for the diagnosis of arthritic disorders based upon an analysis of the structure of the solvent water. In addition, there will be diagnostic information contained in the overtone and combination bands arising from CH, CH2, CH 3 and NH groups. Our preliminary studies suggest that multivariate analyses for the diagnosis of arthritis are as successful in this spectral region as in the mid infrared region.

4.2

Infrared pathology

4.2.1 Multiple sclerosis (MS) Multiple sclerosis ("multiple scars") is the most common demyelinating disease in the developed world and one of the most common central nervous system (CNS) disorders affecting young adults. Characterised by recurring bouts of paralysis, often accompanied by blindness, limb incoordination, muscle spasms and numbness, the causes of MS remain unclear. Most of the available evidence suggests an auto-immune disease, in which for unknown reasons the body begins to attack the insulating layer around nerve cells in the CNS, forming regions of scarring (termed plaques). This insulating layer, the myelin sheath, is a unique membrane which wraps around axons and facilitates impulse conduction. Degradation of this membrane leads to "short circuits" within the brain and spinal cord resulting in the symptoms of MS. Multiple sclerosis is a particularly difficult disease to diagnose, due to the fact that the symptoms are not localised in either time or space. In addition, many of the symptoms

199 may be attributed to other disease states. Finally, the clinical course of MS depends upon age. All of these features make diagnosis of MS problematic. Current diagnostic methods include MRI and neurological assessment. However, the only definitive diagnosis comes from autopsy. Autopsy examination requires the post mortem removal, dissection, fixation, staining and mounting of tissue. While many steps of this traditional approach have been automated, it is still time consuming and expensive. In addition, the final analysis of the tissue requires an experienced pathologist and always contains some degree of subjectivity.

14(]0 14561467 1361

~

j.,

I I I I 1350 14001450 1

1200

/ ~

I

I

1400

I

Plaque

//'~~l ~

I

I

White

I

I

I

1600 1800 2800 2 9 0 0 wqvenumber, cm

Grey 3000 I

Figure 7. IR spectra of control grey and white matter and an MS plaque. Demyelination would be expected to result in significant changes in the IR spectrum of affected areas, reflecting degradation of the myelin sheath and accumulation of breakdown products. Spectra of control white and grey matter and of an MS plaque are shown in Fig. 7. As discussed, the major differences between the spectra of grey and white matter are related to differences in lipid and water content. The difference in lipid content of grey matter is most obvious in the reduced absolute intensity of the CH 2 stretching vibrations and the CH 2 bending vibration (2800-3000 cm -1 and 1467 cm -1 respectively). In addition to an absolute reduction in the intensity of the CH 2 bending and stretching vibrations, the ratio of the CH2:CH 3 stretching vibrations is reduced, reflecting a relative increase in the number of CH 3 groups in grey matter. This in turn reflects the relative increase in protein content of grey matter. As proteins contain almost equal proportions of CH 2 and CH 3 groups in their side chains, the ratio of the intensities of the CH2:CH 3 stretching vibrations will be almost one in tissues with a high protein content. On the other hand, in lipid rich tissues the CH 2 groups from the acyl chains will dominate the spectrum. A similar reduction in the ratio of the intensity of the CH 2 bending and CH 3

200 asymmetric bending (1456 cm -1) vibrations is also seen (see inset of Fig. 7), again due to differences in the relative concentrations of lipid and protein in white and grey matter. The differences between white and grey matter are distinctive. It may be expected that demyelination of white matter may produce tissue which has spectral characteristics similar to that of grey matter, due to the decreased lipid content. To a first approximation, this is indeed the case (7). As can be seen from Fig. 7, MS plaques share many of the spectral features of grey matter, including a marked decrease in the absolute intensity of absorptions arising from CH 2 vibrations. In addition to these major differences, there are also subtle differences between white matter and MS plaques and grey matter. Of particular note is the absence in grey matter and MS plaques of a feature at 1484 cm -1, which is attributed to vibrations of the CH 3 groups of choline. Thus, it is possible to infer not only that grey matter and plaques contain less lipid than white matter, but to be more specific and characterise grey matter and MS plaques as having a low phosphatidylcholine content. While MS plaques share many of the spectroscopic features of grey matter, the differences between MS plaques and white matter are more pronounced than those seen between grey and white matter (7). The ratio of the intensities of the CH2: CH 3 stretching vibrations is even more reduced in MS plaques as compared to grey matter, implying a lower lipid content in the region of demyelination and/or an elevated protein content. Similarly, the ratio of the intensities of the CH 2 bending:CH 3 asymmetric bend is reversed in MS plaques, the CH 3 asymmetric bending vibration giving rise to the stronger absorption in plaques. The reason for this is unclear at the present time. However, MS is characterised by events other than demyelination. For instance, the spectra presented here were obtained from chronic plaques, which are relatively old plaques with considerable scarring. Formation of scar tissue is associated with deposition of a collagen matrix. As we have seen, collagen is associated with a number of characteristic absorptions. Examination of the spectrum of collagen presented in Fig. 2. shows a currently unassigned feature at 1456 cm "1 . It may be speculated that the increased intensity at 1456 cm -1 in chronic MS plaques represents the formation of scar tissue due to collagen deposition, The other features characteristic of collagen, between 1200-1350 cm -1 are not observed in spectra of plaques. However, the increase in the intensity at 1456 cm -1 in the plaque material is relative small, and may possibly arise from the presence of only small amounts of collagen, which are insufficient to cause interfering absorptions in other spectral regions. The formation of scar tissue in a chronic plaque has other spectroscopic consequences. In particular, it leads to a decrease in the amide I/amide II intensity ratio. As discussed above, variations in this ratio in tissues can arise from variations in water content. In this case, the reduced intensity in the amide I region of the spectrum reflects the decreased water content found in scar tissue. It can be seen that it is relatively trivial, with some knowledge of the disease process and an understanding of the spectroscopic properties of biological molecules and tissues, to distinguish between central nervous system tissue which has been affected by multiple sclerosis and control tissue, based upon a spectroscopic analysis of IR spectra of brain tissue. In an attempt to remove the necessity for trained spectroscopists in the IR pathological classification of tissue we again turned to multivariate analysis. Preliminary analyses using

201 linear discriminant analysis and hierarchical clustering have shown that it is possible to couple infrared spectroscopy and multivariate analysis to obtain a non-subjective classification of MS tissue (in fact, with an understanding of the spectral features characteristic of the disease under investigation, we have found that classification is possible using selected spectral regions containing only 30-50 data points).

4.2.2 Alzheimer's disease

Alzheimer's disease (AD) is the commonest dementing disorder of adult life, causing memory loss and disorientation in person, time and place. Pathologically, AD is characterised by gross atrophy of cortical grey matter as a result of the deposition of aggregates of a cytotoxic peptide to form neuritic plaques. Other major biochemical changes include hyperphosphorylation of microtubule associated proteins and alterations in the lipid content of white matter. Our preliminary spectroscopic studies of AD tissue have demonstrated the presence of spectral features which appear to be associated with the disease process, including an amide I feature characteristic of aggregated proteins and peptides which we have attributed to the formation of neuritic plaques. Unfortunately the appearance of these features is somewhat variable (due to difficulties in sampling and the gross degeneration of brain tissue in AD), precluding diagnosis of AD purely from a spectroscopic analysis. We therefore subjected our spectroscopic data to a variety of multivariate analyses, including hierarchical clustering, linear discriminant analysis and artificial neural network analysis. The result of the artificial neural network analysis is shown in Table II.

CG ADG CW ADW

CG

ADG

CW

ADW

16

0

0

0

0 0 0

22 0 1

0 6 0

2 0 7

% Accuracy 100 91.7 100 87.5

Table II. Results of artificial neural network analysis of IR spectra of control white and grey matter (CW, CG) and AD white and grey matter (ADW, ADG). Rows represent multivariate classification, columns represent pathological classification. Off-diagonal elements are misclassifications. These results show that neural networks can separate human CNS tissue into four classes, identified as control white and grey matter and AD white and grey matter with a high degree of success (94.5% overall). All misclassifications which occurred were conservative in nature, AD white matter being classified as AD grey matter and vice versa. Such misclassification most likely results from contamination of white matter with small amounts of grey matter and highlights the necessity of good sample preparation. If the analysis is restricted to a t w o class problem, namely classifying tissue as normal or AD, the analysis is 100% successful. Thus, as with arthritic disorders, the combination of multivariate analytical

202 methods and infrared spectroscopy can provide a rapid, non-subjective diagnosis of AD in a pathological setting.

4.2.3 Cancer Despite decades of research into the aetiology, diagnosis and treatment of cancer, it remains one of the leading causes of premature death in the developed world. Survival depends critically upon early diagnosis, but unfortunately reliable methods for early diagnosis for the most part remain elusive. The enormous literature which is available concerning the molecular events underlying transformation from a normal to a cancer cell has shown that significant differences exist between the biochemistry of normal and turnout cells, which should be detectable by a sensitive spectroscopic technique. Recent work has demonstrated that IR spectroscopy may have the potential to discriminate between healthy and malignant tissue. Most IR spectroscopic studies to date have concentrated on carcinoma of the cervix and colon (6), for reasons which will become clear. In both tissues, considerable differences have been observed between IR spectra of normal and malignant tissue. Of particular interest are changes in absorptions attributed to the asymmetric stretching vibration of DNA phosphate groups (6). In normal tissue, an absorption at 1240 cm -1 has previously been attributed to the PO 2- asymmetric stretching vibration of phosphate groups in DNA which are not involved in hydrogen bonds. In malignant tissue a shift to lower wavenumbers is seen, which upon derivation or deconvolution is seen to arise from an increased absorption at around 1225 cm "1. Absorptions at such frequencies have been attributed to the PO 2asymmetric stretch of phosphate groups of DNA involved in hydrogen bonds. Thus it has been suggested that this spectral changes may indicate changes in the hydrogen bonding pattern of DNA (6), which in turn indicates structural alterations in the DNA of malignant cells. However, such a significant redistribution of intensity in the IR spectrum suggests that a significant proportion of the cellular DNA must have undergone structural changes. It is difficult to envisage a cell surviving such gross modifications to its nuclear material. The spectral changes reported in malignant tissue may have an alternative explanation. As we and others have demonstrated, significant absorptions from collagen arise in infrared spectra of most human tissues between 1200-1350 cm -1. In isolated DNA and nuclei the PO 2- asymmetric stretch may be shown to absorb at around 1225 cm -1 (see Fig. 1). Thus the feature centered at 1240 cm -1 in human tissue is probably a composite absorption dominated by absorptions arising from collagen, with contributions from DNA phosphate groups. Unfortunately for the IR spectroscopist, the connective tissue matrix is not a static structure, but may be considerably altered in a number of diseases. Furthermore, the nature of the alteration may depend upon the stage of the disease. For example, in the human breast benign epithelial tumours are composed of both an epithelial component and a substantial connective tissue matrix with variable collagen deposition. In situ carcinomas may stimulate collagen deposition. Progression to an invasive carcinoma is associated with changes in the basement membrane, which is normally maintained and modulated by a balance of secreted products from both epithelial and adjacent stromal cells. Alteration of this balance and degradation of the basement membrane (for example due to secretion of collagenase) is a major factor in the

203 process of invasion by malignant tumour cells. Thus the overall and regional collagenous composition of an organ can vary significantly with pathological states. Variations in the collagen content of malignant tissue will have pronounced effects upon the IR spectrum of malignant tissues. Decreases in the amount of collagen present in samples, consistent with a highly invasive carcinoma, will lead to a decrease in the intensity of the collagen absorptions seen between 1200-1350 cm -1. In addition to decreases in collagen content, malignant tissues are characterised by the presence of cells showing genetic changes (reflected in nuclear changes such as altered chromosomal structure or DNA content) and functional changes (abnormal control of cell division and growth). Increased DNA content will increase the intensity of the PO 2- asymmetric stretching absorption of DNA. Together, these two changes (decreased intensity at 1240 cm-1 due to collagen degradation and increased intensity at 1225 cm-1 due to elevated DNA levels) will result in an apparent shiR of the composite collagen/DNA absorption to lower wavenumber and the appearance in resolution enhanced spectra of a feature at 1225 cm-1. Thus, rather than denoting a primary biochemical event in cell transformation (i.e. alterations in DNA structure), the spectroscopic changes may actually reflect a secondary event, the alteration of the connective tissue matrix in which the cells sit. As mentioned above, the IR spectroscopic studies to date have concentrated to a large extent upon carcinomas of the cervix and colon. The structure of these tissues facilitates studies in which samples of the epithelium (the tissue component usually transformed in cancer) may in principle may be obtained without contamination from the underlying connective tissue. The general trends appear to be the same in isolated malignant epithelium as in intact malignant tissue, although the spectral differences are much less pronounced. It remains to be seen to what extent the small quantities of connective tissue which will contaminate the epithelial samples will contribute to the spectra and confound the analysis.

l B A

1000 1160

12b0 13b0 1400 15b0 16b0 17b0 18b0 Wovenumber,

cm

-1

Figure 8. IR spectra of a representative human breast tumour (A) and a suspension of cultured human breast tumour cells (B). Collagen absorptions are marked *

204

This question may be answered by a rigorous biochemical/histological analysis of samples used for IR spectroscopy, so that the collagen content can be correlated with spectra. An alternative approach is to make use of the widespread availability of human epithelial cells, both normal and cancerous, in cultured form. The usefulness of such an approach is illustrated in Fig. 8, which presents spectra of a human breast tumour and human tumour cells from culture. As can be seen, human breast tumours contain large quantities of collagen (unlike the cervix and colon, it is impossible to isolate the breast epithelium from the connective tissue matrix), which precludes any analysis of changes in DNA absorptions in breast cancer. The spectrum of human breast tumour cells grown in culture shows none of the spectral features characteristic of collagen. Comparison of normal and transformed cultured cells should provide a useful model system with which to evaluate the spectroscopic features characteristic of cancer, and hopefully point the way to a method of detecting tumour cells in tissue samples with more certainty. Currently both avenues of exploration (correlation of tissue spectra with detailed histological information and the analysis of tumor cell lines) are under investigation in our laboratory.

4.3.

In vivo diagnosis : the future?

The above discussions have centered upon diagnosis of disease states where tissue or biological fluids have been sampled, and represent an extension of existing clinical techniques. However, tissue or fluids are not always available. In other cases, tissues only become available after death. While the spectroscopic characterisation of this tissue in a pathology lab is of benefit to the pathologist, there is no benefit to the deceased patient. Obviously, noninvasive methods which can be applied to living subjects are preferable for a more clinically relevant diagnosis. The applications of the mid infrared region of the spectrum to non-invasive diagnosis are restricted by the strong absorptions shown by biological molecules, particularly water. A technique which has a pathlength limitation of 6-10 ~tm is obviously ill suited for in vivo diagnosis. However, there are areas in which mid infrared spectroscopy may yet prove to be useful. Short pathlengths will not necessarily prove to be a limitation when we are interested in an exposed surface, for example the skin. If all we require is surface information then mid infrared reflectance spectroscopy can be used to probe the molecular events happening at the skin surface. This raises the intriguing possibility of a technique for diagnosis of skin disorders (basal cell carcinoma, psoriasis, UV damage etc.) based upon existing mid infrared technologies, perhaps utilising flexible fibre optic/ATR systems. Of particular interest, given the pronounced spectral features characteristic of collagen, may be monitoring of the integrity of the skin barrier by analysis of collagen content (degradation of collagen in the skin is associated with alterations in the physical and mechanical properties of the skin and accompanies aging and UV radiation damage). Other exposed surfaces, such as the tooth or fingernail, may also prove to be amenable to study. In our laboratory we have already demonstrated using photoacoustic infrared spectroscopy that high quality spectra of intact human teeth can be obtained ex vivo, and that depth profiling of the tooth surface can provide

205 useful information concerning the composition of the tooth (12). It seems likely that this study can be extended, using fibre optic systems, to the in vivo analysis (including depth profiling) of human teeth. Such a technique would have a significant impact on dental care, including the non-subjective detection of dental caries before they are advanced enough to be visible to the naked eye and preventative dentistry based upon analysis of carbonate content and the crystallinity of the phosphate complexes of dental enamel. The near infrared region of the spectrum is more suited to in vivo diagnostic use, due to the dominance of scattering effects rather than absorption. To date, the non-invasive clinical usage of NIR spectroscopy has been limited to the measurement of blood oxygenation levels, particularly to the assessment of neonatal cerebral blood oxygenation. However, such oxygenation measurements utilise a few low lying electronic transitions close to the visible region of the spectrum, and ignore the much richer region of the spectrum containing the overtones and combinations of C-H, N-H and O-H vibrations. In recent years this has begun to change. Many groups, ourselves included, are beginning to realise the benefits which would be afforded by using these overtones and combinations. For laboratory methods, effort is being directed at the use of NIR spectroscopy for the simultaneous determination of a number of analytes (e.g. haemoglobins, urea, glucose, lactate) in biological fluids, particularly blood (13). Of course mid infrared methods may also be applied to this problem (14). Ultimately it would be preferable to have an analytical method which does not require withdrawal of blood. Mid infrared methods are unsuitable for this type of analysis due to the inherent problems associated with the strength of the fundamental absorptions in biological systems and associated short pathlengths. However, the enhanced penetration of NIR radiation into tissues should allow the refined NIR laboratory method to be adapted, using fibre optics, to develop real-time, non invasive methods for the analysis of blood. Such a methodology would prove invaluable to clinicians in critical care units where continuous monitoring of many variables is desirable. In addition to in vivo clinical chemistry, the near infrared spectroscopic approach may possibly be extended to m vivo studies of neurological disorders. It has already been demonstrated that NIR spectroscopy can detect significant changes in the biochemistry of stroke affected brains of gerbils in a non-invasive fashion (15). We are currently assessing the use of such techniques in the analysis of other neurological disorders in animals, such as MS. Studies of this type are enhanced by the availability of powerful tunable laser diodes at relatively low cost. The increased power of laser diodes compared to standard NIR sources significantly increases the number of photons reaching the detector, partially alleviating the scattering problems common to NIR experiments on tissues, and increasing signal to noise ratios. In the future, with continued advances in technology, perhaps we may see spectrometers based around such tunable laser diodes, with no moving parts and no dispersive elements, for use in the overtone and combination region of the spectrum. Such instrumentation would greatly facilitate biomedical research.

206 4.4

References

1. H. L. Casal and H. H. Mantsch, Biochimica Biophysica Acta 779 (1984) 381. 2. M. Jackson and H. H. Mantsch, Spectrochimica Acta Review 15 (1993) 53. 3. H. Susi and D. M. Byler, Biochem. Biophys. Research Commun. 115 (1985) 391. 4. W. K. Surewicz and H. H. Mantsch " Infrared Absoprtion Methods for Examining Protein Secondary Structure" in Determination of Protein Structure in Solution by Spectroscopic Methods, H.H. Havel, Ed., VCH Publishers, New York, 1994, Chapter 8. 5. E. Taillandier, J. Liquier and J. A. Taboury, "Infrared Spectral Studies of DNA Conformations" in Advances in lnfrared and Raman Spectroscopy, R .J.H. Clark and R. E. Hester, Eds., Wiley, New York, 1985. Chapter 8. 6. B. Rigas, S. Morgello, I. S. Goldman and P .T .T. Wong, Proc. Natl. Acad. Sci. USA, 87 (1990) 8140.

7. L.-P. Choo, M. Jackson, W. C. Halliday and H. H. Mantsch, Biochimica Biophysica Acta 1182 (1993) 333. 8. M. Jackson, L.-P. Choo, P. H. Watson, W. C. Halliday and H. H. Mantsch. Biochimica Biophysica Acta (1994) in press. 9. H. H. Eysel, M. Jackson, G. T .D. Thomson and H. H. Mantsch, Applied Spectroscopy 47 (1993) 1519.

10. F. Fleyfel and J.P. Devlin, J. Physical Chemistry 95 (1991) 3811. 11. S. Sowa and L. E. Towill, Plant Physiology 95 (1991) 610. 12. M. G. Sowa and H. H. Mantsch, Calcified Tissue International 54 (1994) 481. 13. J. W. Hall and A. Pollard, Clinical Chemistry 38 (1992) 1623. 14. H. M. Heise, R. Marbach, T. Koschinsky and F. A. Gries, Applied Spectroscopy 48 (1994) 85.

15. J. M. Carney, W. Landrum, L. Mayes, Y. Zou and R. A. Lodder, Analytical Chemistry 65 (1993) 1305.