211

J. Photochem. Photobiol. B: Biol., 16 (1992) 211-233

Quantitative histochemical using Raman spectroscopy

analysis of human artery

Ramasamy Manoharan, Joseph J. Baraga, Michael S. Feldt and Richard P. Rava George R. Harrison Spectroscopy Laboratory, Massachusetts Institute of Technology, Cambridge, MA 02139 (USA)

(Received April 10, 1992; accepted July 25, 1992)

Abstract

We have developed a method for using near infrared Raman spectroscopy to quantitatively analyze the histochemical composition of human artery. The main contributors to bands observed in the Raman spectra of normal and atherosclerotic aorta are the proteins collagen and elastin, cholesterol lipids, and calcium hydroxyapatite. The Raman scattering crosssections of different bands for these components have been determined in order to understand their relative contributions to the Raman spectra of biological tissue. The Raman signal is observed to behave linearly with the concentration of the components, even in a highly scattering medium such as a powder. Using these data, we have developed a linear model that can be used to extract the quantitative contribution of an individual component to the spectrum of a mixture. The model has been applied to several mixtures of known composition of tissue constituents in order to evaluate its precision and accuracy. The calculated fit coefficients from the spectra are in agreement with the measured values within experimental uncertainties. The spectra of different types of atherosclerotic aorta have also been modeled, and we have extracted quantitative information regarding the relative concentration of biological constituents in atherosclerotic aorta.

Keywords:

Raman

spectra,

artery,

histochemical

analysis.

1. Introduction

Among the physical methods which are capable of probing macromolecular structures of biological systems in situ, Raman spectroscopy has become an important tool in the arsenal of the biophysical chemist [l, 21. The technique has been utilized to study diverse biological systems such as protein and DNA conformations, drug interactions with proteins and DNA, and enzyme mechanisms. The evolution of lasers of diverse excitation wavelength, sensitive CCD detectors, and cheap persona1 computers in the past few years has helped many of the early expectations of the method. Medical applications of Raman spectroscopy as a diagnostic or analytical tool have also appeared. Pioneering work in the mid-seventies showed that Raman spectra could be obtained from intact viruses [3] and eye lenses [4]. The laser Raman microprobe has played a role in the histologic examination of various pathologies [S]. The major limitation of the use of Raman spectroscopy as a niedical tool, however, has been +Author to whom correspondence

should be addressed.

212

the endogenous fluorescence present in most human tissues, which obscures the Raman signal. In addition, tissue can easily be altered when visible or near ultraviolet radiation is utilized. Since the fluorescence decreases very rapidly at longer excitation wavelengths, most materials, including tissue, exhibit virtually no fluorescence emission with near infrared (NIR) irradiation. This has led several groups, including our own, to begin to examine the prospects of extracting histochemical information from human tissue using NIR Raman spectroscopy. Yu and co-workers have obtained NIR Fourier transform (FT) Raman spectra of eye lens and several hard tissues [6]. Alfano and his group have indicated that NIR FT Raman spectroscopy of normal and cancerous human breast provides several distinguishing features which may allow discrimination of these tissue types [7]. Ozaki et al [8] have utilized NIR Raman to examine the state of hemoglobin in blood. In our preliminary work on human aorta, we demonstrated that detailed information about atherosclerotic tissue can be obtained without sample preparation, specimen degradation or fluorescence interference [9, lo] when NIR excitation wavelengths are utilized. Also, utilizing a spectrograph/CCD system, we have recently demonstrated that NIR Raman spectra can be collected rapidly with high S/N from human artery [ll], which will allow data to be collected in vim. The results indicated that Raman spectroscopy can be utilized to determine collagen, elastin, cholesterol, cholesterol esters, triglycerides, calcium hydroxyapatite, carbonated apatites and carotenoids in normal and atherosclerotic artery In order to become a useful in situ histochemical method, NIR Raman spectroscopy must be able to extract quantitative information about the important biochemical species. Clinicians will then be able to utilize the information for guiding or evaluating treatment. In this manuscript, we analyze the ability of NIR Raman spectroscopy to extract quantitative information from human artery. After a brief review of the experimental methods in the following section, NIR Raman spectra of human aorta are compared with those found for pure bio-components in Section 3.1. We then model the Raman spectra of known mixtures in Section 3.2 in order to evaluate the quantitative limits of the current technology. The results are compared with several specimens of aorta in Section 3.3. In Section 4, the applications and limitations of these methods to artery and other tissues are discussed, and conclusions are summarized in the last section.

2. Materials

and methods

A Perkin-Elmer Fourier transform infrared spectrometer was utilized for the study of NIR FI Raman spectroscopy in a manner identical to our previous work [9, la], and thus will not be discussed further here. Briefly, the Raman accessory employs a 180” back-scattering geometry and a cooled (77 K) InGaAs detector. A 1064 nm CW Nd:YAG laser was used for exciting samples, with 400 mW laser power in a 1 mm diameter spot on the sample. Spectra of components are the sum of 256 scans recorded 18 min collection time), and those of tissues are at 8 cm-’ resolution (approximately the sum of 512 scans recorded at 8 cm-’ resolution (35 min collection time). In order to determine the ability of NIR Raman spectroscopy to extract quantitative histochemical information, relative Raman cross-sections were measured by using BaS04 as a Raman scattering internal intensity standard, and the behavior of the Raman signals of individual biomolecules with concentration was explored. Mixtures of a known weight

213

percent of the powder of the compound of interest and BaSO, were finely ground using a mortar and pestle until they visually appeared to be homogenized, and then placed in a fused silica cuvette. For each sample, at least three measurements were made by irradiating different spots on the sample; the variation in the cross-section values was within f 15%. Since, no polarization analyzer was employed, the weight cross-sections derived here represent the sum of the scattering contributions from both perpendicular and parallel polarizations. Mixtures of tissue components themselves, without BaS04, were also examined both as powders and as saline slurries. Biomolecules were purchased from either Sigma Biochemical or Calbiochem and used as supplied. Human aorta was chosen for initial study as an instance of atherosclerotic artery tissue. Samples were obtained at the time of postmortem examination, rinsed with isotonic saline solution (buffered at pH 7.4), snap-frozen in liquid nitrogen, and stored at -85 “C until use. Prior to spectroscopic study, samples were passively warmed to room temperature while being kept moist with the isotonic saline. Normal and atherosclerotic areas of tissue were identified by gross inspection, separated, and sliced into roughly 8X8 nun’ pieces. The tissue samples were placed in a suprasil quartz cuvette with a small amount of isotonic saline to keep the tissue moist, and one surface in contact with the window was irradiated by the laser. After spectroscopic examination, all specimens were histologically analyzed to verify the gross identifications.

3. Results 3.1. NIR Raman spectra of the biological molecules in artery In attempting to quantify the observed spectral signals from human artery, the first question which must be addressed is the choice of the biological substituents which should be examined. Normal human artery is composed of three distinct layers: intima, media and adventitia. The intima, normally 50-300 pm thick depending on the artery, is the innermost layer. It is mainly composed of collagen fibers and ground substance, primarily formed from proteoglycans [12]. A single layer of endothelial cells in the vessel lumen protects the intima from injury. Normal intima is composed of up to 30% dry weight collagen (types I and III) and 20% elastin [13]. The proteoglycans account for up to 3% of the dry weight. The media, several hundred microns thick, can be quite elastic or muscular depending on the artery. The structural protein elastin is the major component of aortic media, while smooth muscle cells make up the majority of the media in coronary artery. The outermost adventitial layer serves as a connective tissue network which loosely anchors the vessel in place, and is mainly made up of lipids, glycoproteins and collagen. During the atherosclerotic process, the intima thickens due to collagen accumulation and smooth muscle cell proliferation, lipid and necrotic deposits accumulate under and within the collagenous intima, and eventually calcium builds up, leading to calcium apatite deposits in the artery wall [12]. Collagen can account for up to 60% of the dry weight of the atherosclerotic intima, and lipids can account for up to 70% [13, 141 depending on the lesion type. Elastin is generally less than 10% and the ground substance is equivalent to that found in normal intima. The lipids in the atherosclerotic lesion are primarily composed of cholesterol and cholesterol esters, with cholesteryl palmitate, cholesteryl oleate and cholesteryl linoleate accounting for up to 75% of the cholesterol esters. These considerations suggest that the primary species which should be studied are collagen, elastin, cholesterol, the cholesterol esters of palmitic acid, oleic acid and

214

linoleic acid, and calcium hydroxyapatite. The proteoglycans should also be examined. This assumes that none of the species in artery discussed above, or other minor biological components, show resonance enhancement at the excitation wavelength being used in this study. As discussed in our previous work [lo], a known exception is carotenoids, which show particularly strong signals, with 1064 nm excitation, despite their low concentration in artery. The spectral regions containing the carotenoid bands will be ignored in this paper. Figure 1 shows the NIR Raman spectra obtained from typical specimens of normal, atheromatous and calcified human aorta. As demonstrated by comparing Fig. l(a) with the spectra of elastin (bovine neck ligament) and collagen type I (bovine achilles tendon) (Fig. 2), the spectrum of normal aorta is dominated by bands due to the

? 1600

1600

1400

1200

1000

600

600

Frequency (cm")

Fig. 1. NIR Raman spectra of (a) normal aorta (X8), (b) atberomatous calcified plaque.

1800

1600

1400

1200

1000

800

plaque (x4),

and (c)

I 600

Frequency (cm-')

Fig. 2. NIR Raman spectra of the structural proteins (b) collagen (bovine achilles tendon, type I).

(a) elastin (bovine neck ligament),

and

215

1800

1600

1400

1200

1000

800

6iO

Frequency (cm-')

Fig. 3. NIR Raman spectra of proteoglycans (a) chondroitin sulfate A, sodium salt (bovine tracheae), and (b) hyaluronic acid, sodium salt (bovine tracheae).

proteins. In particular, the bands observed at 1658 and 1252 cm-’ can be assigned to amide backbone vibrations, while the peak at 1452 cm-’ is due to C-H bending of the protein [l]. Note that bands due to proteoglycans, such as chondroitin sulfate A and hyaluronic acid (Fig. 3), which are known to make up the ground substance in artery wall, do not appear to contribute significantly to the spectra, as might be expected from their low concentrations. This is a qualitative indication of the detection limits of the current methods. The spectrum of the atheromatous plaque (Fig. l(b)) is distinctly different from that of normal aorta (Fig. l(a)). In particular, there are many more bands in the atheromatous plaque spectrum below 1000 cn- ‘. Consideration of the physiology of these plaques, as discussed above, and comparison of the spectra with several of the predominant cholesterol esters shown in Fig. 4, indicates that many of the bands in these spectra are due to cholesterol and its esters. In fact, the band at 700 cm-‘, due to the sterol ring, appears to serve as a marker for the existence of cholesterols in atherosclerotic lesions, while the other bands can be used to separate the various contributions of the esters to the spectrum. Some of the bands in the spectra of the cholesterol esters can be directly attributed to the spectra of the fatty acid side chains. This is demonstrated in Fig. 5(c), where the spectrum of cholesterol is subtracted from cholesteryl oleate. The result is a spectrum nearly identical to that found in Fig. 5(a) of oleic acid, with the exception of the ester vibrational band at 1737 cm-‘. This also points out the ability of the Raman method to distinguish between triglycerides (glycerol tri-esters), which have an ester frequency around 1745 cm-’ (Fig. 5(b)), and the cholesterol esters which have ester vibrational frequencies around 1737 cm-‘. The NIR Raman spectra of calcified plaques (Fig. l(c)) have additional bands at 960 and 1070 cm-‘. Comparison of calcified plaque spectra with the NIR Raman spectrum of calcium hydroxyapatite (Fig. 6) indicates that this salt is the primary contributor to the 960 cm-’ band. However, the 1070 cm-’ band seen in calcified plaque may have a large contribution from carbonate apatites (see below).

a

j

;

m f 1800

1600

1400

1200

1000

600

800

Frequency (cm-')

Fig. 4. NIR Raman spectra of cholesterol and cholesterol esters atherosclerotic lesions. (a) Cholesterol; (b) cholesteryl palmitate; cholesteryl linoleate.

800

1600

known to be significant in (c) cholesteryl oleate; (d)

600

Frequency (cm-')

Fig. 5. NIR Raman spectra of (a) oleic acid, (b) triolein, and (c) subtraction of the spectrum of cholesterol from cholesteryl oleate, (c) demonstrates that the major bands in the Raman spectrum of cholesteryl oleate is simply the sum of cholesterol plus oleic acid and the ester vibration at 1737 cm-‘. 3.2. Relative Raman scattering cross-section Having established the identity of the major contributors spectra of artery, we now wish to utilize the Raman spectra

to the NIR Raman to extract quantitative

217

1600

1600

1400

1200

1000

600

600

Frequency (cm-‘)

Fig. 6. NIR Raman

spectrum

of calcium

hydroxyapatite.

biochemical information. In order to do so, two pieces of information must be known. First, the Raman scattering cross-section for each of the species must be measured relative to a standard, so that meaningful comparison between bands of different molecules can be carried out. Secondly, the behavior of the Raman signals with respect to concentration in a highly scattering medium such as tissue must be measured. In order to address the first issue, we measured the integrated Raman intensities from the bands of many compounds known to be important in atherosclerotic tissue. As discussed in Section 2, the band intensities were studied in BaS04 powder mixtures in order to utilize the strong S04’- band at 987 cm-’ as an internal reference standard. For a given intensity, I0 (W cm-‘) and collection time, t (s), the integrated Raman signal in W for a band at a frequency vi, S(y) measured at the detector is given by

where n is the detector ,quantum efficiency (electrons/photon) and .$ is the efficiency of the optical system. The instrument throughput, 0 (cm’ sr), is given by the product of the collection area, A (cm*), and the solid angle of collection, 0 (sr), and the sampling length, 1 (mm), is primarily determined by the collection optics. p is the concentration in either g cmP3 or molecules cm -3; for the former concentration units (%/HI), is a weight Raman cross-section (cm’ (g.sr)-‘) while for the latter it is a molecular cross-section (cm* (molecule.sr)-‘). q, I,,, t, 5, 0 and 1 can be eliminated from consideration when using an internal standard. Comparing the BaS04 signal with the material of interest,

-au aa

(1 (an1 “i

au

“ease,

=

S(Y)P13aSO*

S(%SOJPi

(2)

218

We have ignored local field corrections for the local refractive indices in the condensed phase [15]. In Table 1, we report the relative Raman weight cross-sections compared with 1 g BaS04 for several bands in collagen, elastin, cholesterol, the primary cholesterol esters (cholesteryl palmitate, cholesteryl oleate and cholesteryl linoleate), the triglyceride tripalmitin and its fatty acid side-chain palmitic acid. We have chosen to report the relative Raman weight cross-sections because for many biological components (e.g. elastin) the precise molecular weights are unknown. As an example, Fig. 7 shows the NIR FT Raman spectrum of a cholesterol: BaS04 powder mixture (SO wt.% cholesterol). In this experiment, the CH2 bending mode of cholesterol at 1440 cm-’ is compared with that of the symmetric SO,,‘stretching vibration of BaS04 at 987 cm- ‘. The areas under each of the bands were determined and compared, yielding a relative Raman weight cross-section of 3.19. In order to test the linearity of the Raman signal in a highly scattering medium, the weight percentages of cholesterol and BaSO_, were varied, and the integrated intensity ratio of the Cl& bending mode of cholesterol at 1440 cm-’ to that of the BaS04 peak at 987 cm-’ was measured. The plot of integrated intensity ratio versus weight percentage of cholesterol is shown in Fig. 8 and is found to be linear. The linearity of this plot is an indication of both the homogeneity of the powder mixture and the absence of any chemical interaction between the components of the mixture that could alter the spectral features. The implication of this result is that apparently the tissue Raman spectra can be described in terms of a linear superposition of individual biochemical constituents as long as the specific scattering properties of tissue do not significantly distort the signal. Having established the linear and chemical behavior of the powder mixtures with BaS04, the molecular Raman scattering cross-section of each given band for various lipids was estimated using BaS04 as a standard (Table 2). In doing this, we utilize the relative weight cross-sections listed in Table 1, the known molecular weights of these compounds, and the value of the Raman cross-section of BaS04 reported in the literature [16]. For a given cholesterol lipid, the scattering cross-section for -CH2 bending vibration is higher than other modes. The molecular Raman cross-section (Table 2) of the CH2 bending modes of cholesterol esters are higher than the corresponding modes of cholesterol, consistent with the additional fatty acid sidechains in the case of esters. The increase in this value for cholesteryl oleate (C18:l) and cholesteryl linoleate (C18:2) relative to cholesteryl palmitate (C16:O) is likely due to the increase in the number of -CH2 groups in the side-chain. The degree of unsaturation, or number of double bonds in the fatty acid side-chain, of the lipids is manifested in the molecular Raman cross-section values of the band around 1670 cm-‘. For example, cholesteryl palmitate, which like cholesterol has only one double bond in the ring, shows a molecular scattering cross-section of 0.77 relative to cholesterol. The molecular scattering cross-section of this same band in cholesteryl oleate, which has one ring and one side-chain double bond, is 2.58 times larger than that of cholesterol; in cholesteryl linoleate, with a total of three double bonds, this cross-section is 3.13 times larger than in cholesterol. Both cholesterol and the cholestetyl lipids exhibit a unique Raman peak at 700 cm-’ as a result of the steroid nucleus. Defining the molecular scattering cross-section for this mode in cholesterol to be 1.00, the relative molecular scattering cross-section value for this mode is decreased to nearly 0.55 in the cholesterol esters. This might be attributed to the substitution-induced effect on the ring skeletal mode. The ester band molecular scattering cross-section of tripalmitin is nearly four times higher than that of cholesterol esters, primarily because tripalmitin has three ester groups compared

Amide I Amide I Amide Amide 1738 1738 1740 1737 1745 0.12 0.12 0.11 0.52 0.41

1.00 1.23 0.18 0.58

Crosssection

assignment

of different

1671 1667 1665 1665 -

Freq. (cm-‘)

-C&J-

contains

1450 1450 - 1400” - 1400a 1440 1440 1440 1440 1442 1440

Freq. (cm’)

CH2 bend

contributions

0.72 0.79 0.58 0.79 3.19 2.70 3.70 3.02 4.66 4.32

Crosssection

0.35 0.17 0.17 0.76 0.66

-

Crosssection

700 700 700 700 -

-

0.38 0.13 0.12 0.12

-

Crosssection Freq. (cm-‘)

Sterol ring stretch

aorta relative to that of 1

from other modes as well.

1130 1140 1146 1130 1130

Freq. (cm’)

C-C stretch

and lipids typically found in atherosclerotic

and probably

0.77 0.36 1.14 1.40

Crosssection

bands from proteins

“Calculated for the entire band in the region 1300-1500 cm-’

Collagen Elastin Chondroitin sulfate A Hyaluronic acid Cholesterol Cholesterol palmitate Cholesteryl oleate Cholesteryl linoleate Palmitic acid Tripalmitin

Freq. (cm’)

Ester, C=O

Vibrational

weight cross-sections

Biological component

Raman scattering g BaSO.,

TABLE 1

1800

1600

1400

1200

1000

800

600

Frequency (cm-')

Fig. 7. NIR Raman spectra of (a) cholesterol,

and (h) 5050 by weight cholesterokBaS0,

mixture.

-I 0

0.1

0.2

0.3

0.4

0.5

0.6

(weightpercentage) PCho,eSterO,

Fig. 8. Plot of integrated intensity ratio of the 1440 cm-’ band of cholesterol to 987 cn-’ peak of BaSO, vs. weight percentage of cholesterol in a cholesterol:BaSO, mixture (the symbols in the axes labels are as defined in eqn. (2) in the text). The slope of the line is 2.72; the regression coefficient is 0.997. with the one in the cholesterol esters. Similarly, the relative molecular scattering crosssections of all the modes of tripalmitin are nearly three times higher than those of palmitic acid. This is consistent with the molecular structure of tripalmitin, which is the triglyceride of palmitic acid. For calcium hydroxyapatite, the weight scattering cross-section of the symmetric phosphate stretching mode, 0.36, is ten times greater than that of the anti-symmetric mode. In tissue, additional bands appear around the phosphate anti-symmetric stretching frequency, and thus the relative intensity of this band is larger. The most likely candidate for these bands is carbonated apatites [173, as discussed below.

0.17 0.18 0.17 0.76

1 1.06 1.00 4.49 -

0.67 0.52 1.73 2.1

Absolute crosssection

Absolute crosssection

Comparativeb

-C=C-

assignment

Ester, C=O

Vibrational

1 0.77 2.58 3.13 -

ComparativeC

of different

2.85 3.91 5.58 4.53 2.77 8.07

1 1.37 1.96 1.59 0.97 2.83

0.50 0.26 0.26 0.45 1.23

Absolute crosssection

C-C stretch

for the wavelength

Comparative’

corrected

Absolute crosssection

CH, bend

dependence

1 0.52 0.52 0.9 2.46

Comparativeb

aorta=. Units for

[16].

0.34 0.19 0.18 0.18 -

Absolute crosssection

1 0.55 0.53 0.53 -

Comparativec

Sterol ring stretch

bands from lipids typically found in atherosclerotic

“The Raman cross-section value for SO 42- is 0.54~ 10m3’-’cm’ (molecule.sr)-‘, bMolecular cross-sections compared with given band of cholesteryl palmitate. ‘Molecular cross-sections compared with given band of cholesterol.

Cholesterol Cholesteryl pahnitate Cholesteryl oleate Cholestetyl linoleate Palmitic acid Tripahnitin

Biological component

Estimated absolute Raman scattering molecular cross-sections the absolute cross-section values are 10TM cm’ (molecule.sr)-’

TABLE 2

222 For equal weight percentage, the relative Raman cross-sections of lipid bands near 1440 cm-’ are higher than those of protein and glycosaminoglycan modes. This suggests that if equal amounts (by weight) of lipids and proteins are present in a mixture, lipids are expected to contribute to the integrated area of -CH2 bands nearly four times as much as proteins. 3.3. Modeling of artery Raman spectra to obtain quantitative information 3.3.1. Linear model for artery NIR Raman spectra In the previous section we demonstrated that the NIR FT Raman spectra of different biological components can qualitatively account for the observed features of the spectra of aorta. In addition, we showed that the signals behave in a linear fashion, even in the presence of a highly scattering medium such as BaSO,. Now we wish to determine whether the relative concentrations of the biological constituents can be determined quantitatively from the NIR Raman spectra, using the basis set of biomolecular spectra established above. Our model for the NIR Raman spectra is a simple linear superposition of the spectra of the biological substituents given by R(V) = zxiri( v) + poly3( V)

(3)

where R(V) is the observed Raman spectrum of tissue, ri(V) is the Raman spectrum of the ith component normalized to a particular band, and Xi is the fit coefficient describing the spectral contribution of the ith component. Poly3(u) is a third-order polynomial utilized to account for broad, featureless signals from tissue not accounted for by the basis set. In our procedure, the basis set of spectral lineshapes, ri(v), are given by the pure substance spectra (shown in Figs. 2, 4 and 6), with the integrated intensity of the CHH,bending band normalized to unity. The parametersx, are determined using a linear least-squares fitting procedure [18]. Using the relative Raman weight cross-sections of the CHz band for the individual components determined above, the weight percentage Wi of each component can then be computed as follows: wi=K

=

where K is determined by normalizing Alternatively, this can be written as

(4)

the sum of the weight percentages

to unity.

Writing eqn. (4) in this form makes it clear that the Raman cross-section for the standard, BaSO,, is not required to compute the weight percentages of individual components, as the weight percentages are measured relatively. In order to initially test the capabilities of this approach, we measured FT Raman spectra of mixtures of the biological constituents with varying weight percentages. Each mixture spectrum was then fit to eqn. (3), and the weight percentages calculated from eqn. (4) were compared with the known weight percentages of the mixtures. These

223

results are discussed in the following section. The model was then applied to several specimens of normal and atherosclerotic aorta to examine the applicability of the basis set and to establish typical limits of sensitivity of this approach. 3.3.2. Application of model to mixtures of biological constituents To evaluate the linearity of the Raman signals, the limits of detection of important tissue constituents, and the accuracy of the model, we prepared a series of mixtures of the pure biological constituents with weight percentages that span the known compositions of normal and atherosclerotic artery. In these experiments, the primary components of interest were those that play dominant roles in normal and atherosclerotic plaques: the proteins collagen and elastin, and cholesterol and cholesterol ester lipids. Ten separate mixtures of protein and lipid were prepared, with vatying protein/ lipid weight percents ranging from 100% protein/O% lipid to 0% protein/loo% lipid. The protein portion consisted of collagen type I (bovine achilles tendon) and elastin (bovine neck ligament) in equal weight percentages (collagen:elastin= l:l), and the lipid portion consisted of equal weight percentages of cholesterol and cholesterol ester (cholesterol:cholesteryl oleate:cholesteryl linoleate = 1:OS:OS). This range allowed evaluations of the accuracy of the linear model for all five components and of detection limits for total protein and total lipid, as well as for the individual proteins and cholesterol lipids. Two consecutive Raman spectra were recorded from the same spot for each mixture to check the reproducibility in measurement, and Raman spectra from two separate spots were recorded for two of the mixtures to check the homogeneity of the mixtures. Each Raman spectrum was then modeled using eqn. (3) with the Raman lineshapes recorded from the five individual components. Each resultant fit coefficient xi was then used along with the measured CH2 band Raman weight crosssection of that component (listed in Table 1) to compute the weight percentage, wi, for that component according to eqn. (4). The Raman spectrum of the 50% protein (collagen 25%, elastin 25%) 50% lipid (cholesterol 25%, cholesteryl oleate 12.5%, cholesteryl linoleate 12.5%) mixture is compared with the model calculation in Fig. 9. The residual of the fit (also shown in Fig. 9) falls within the noise level of the spectrum, indicating a reasonable fit to the spectrum. The weight percentages calculated from the fit coefficients for this spectrum are protein 64% (collagen 26%, elastin 38%) and lipid 36% (cholesterol 20%, cholesteryl oleate 5%, cholesteryl linoleate 11%). Given the + 15% uncertainties in the measured Raman cross-sections and the inhomogeneities in the mixture, the calculated protein and lipid weight percentages agree with the measured percentages to within the experimental error. The differences among the individual protein and lipid component weight percentages calculated from the model and the measured weight percentages is primarily attributable to uncertainties in the cross-sections, along with uncertainties in the fit coefficients due to spectral noise (see below). The weight percentages of total protein and total lipid calculated from the model are compared with the measured weight percentages in Fig. 10 for all the Raman spectra collected from the mixtures. These plots illustrate three important features regarding the calculated total protein and total lipid weight percentages. First, the calculated weight percentages are very linear over the entire range of mixture concentrations, supporting the validity of the linear model. Second, a linear correlation between calculated and measured lipid weight percentages yields a slope of 0.94, which is essentially consistent with the expected value of 1. Any small discrepancy between this value and an exact match (slope = 1) is attributable to systematic uncertainties from two sources. One source is the difficulty in achieving completely homogeneous

224

1800

1600

1400

1200

1000

600

t 10

Frequency (cm-')

Fig. 9. Measured Raman spectrum of 50% protein (25% collagen, 25% elastin) 50% lipid (25% cholesterol, 12.5% cholesteryl oleate, 12.5% cholesteryl linoleate) mixture, along with model calculated fit and residual. mixtures due to differences in the physical properties of the components. For example, collagen, elastin and cholesterol are powdery and cholesterol oleate and linoleate are pasty. The other source of systematic uncertainty derives from measurement errors in Raman cross-section values, which propagate in the calculation of weight percentages. Third, uncertainties in the calculated weight percentages due to spectral noise, which are illustrated by the scatter of the data points about the linear correlations in Fig. 10, are relatively small. These uncertainties determine the detection limits for lipid and protein; the data in Fig. 10 indicate that these limits are 5% or less for total lipid and lO-15% for protein. The difference in detection limits between protein and lipid are in large part due to the three-fold smaller CH2 band Raman weight crosssections for proteins (see Table 1). At a finer level of detail, the lipids can be divided into cholesterol and cholesterol esters. Cholesterol and total cholesterol ester (oleate + linoleate) weight percentages calculated from the Raman spectra are compared with the directly measured weight percentages in Fig. 11. The individual cholesterol ester (oleate, linoleate) weight percentages are plotted in Fig. 12. In all cases, the calculated and measured weight percentages appear to be linearly correlated to within the parameter uncertainties. However, the uncertainties in the calculation of weight percentages of individual components increase due to either or both of two factors: (i) the individual components occur over lower concentration ranges in the mixtures; (ii) spectral differentiation depends on distinguishing small spectral features above the given noise level. This differentiation is more difficult in components with similar Raman spectra such as collagen and elastin. For cholesterol and cholesteryl linoleate, the slopes of the linear correlations between calculated and measured weight percentages, 1.08 and 0.98, respectively, agree with the exact value of 1 to within the uncertainties in the measured Raman weight cross-sections. In the case of cholesteryl oleate, the slope of 0.64 is smaller than the expected value of 1, resulting in a slightly smaller than expected value of 0.81 for total cholesterol ester. The plots also demonstrate that the detection limits for cholesterol, total cholesterol ester, and the individual cholesterol esters are

225

-0.2 I -0.2

0

0.2 measured

(4

0.4 weight

0.6

0.6

1

1.2

percentage

0.6. 0.60.40.2O-0.2 (b)

-I

I -0.2

0

0.2 measured

0.4 weight

0.6

0.6

1

1.2

percentage

Fig.

10. Plot of component weight percentages calculated from model vs. measured weight percentages. (a) Total protein (collagen+elastin). The slope of the line is 0.94; the regression coefficient is 0.98. (b) Total lipid (cholesterol + cholesteryl slope of the line is 0.94; the regression coefficient is 0.98.

oleate + cholesteryl

linoleate).

The

roughly 5% each, which is similar to the total lipid detection limit. This is a consequence of the similar values for the CH2 band Raman weight cross-sections among cholesterol, cholesteryl oleate and cholesteryl linoleate. The protein fraction can also be further subdivided into collagen and elastin weight percentages. The calculated weight percentages for collagen and elastin are compared with measured weight percentages in Fig. 13. In these cases, the parameter uncertainties are significantly greater than in the case of the individual lipid components because of the relatively high degree of similarity between the collagen and elastin Raman spectra. These uncertainties obscure the linear correlations between the calculated and measured weight percentages, although a linear trend is consistent with the data. The detection limits for collagen and elastin individually are U-20%, or more than 3 times the 5% detection limits of cholesterol and cholesterol esters.

226

-0.1

0

(4

-0.1 I -0.1 @I

0.1

0.2

0.3

0.4

0.5

0.6

0.5

0.6

measured weight percentage

-I 0

0.1 measured

0.2

0.3

0.4

weight percentage

Fig. 11. Plot of component weight percentages calculated from model vs. measured weight percentages. (a) Cholesterol. The slope of the line is 1.08; the regression coefficient is 0.98. (b) Total cholesterol ester (cholestetyl oleate+cholestetyl linoleate). The slope of the line is 0.81; the regression coefficient is 0.97. 3.3.3. Application of model to aorta With the limits of validity of the model established and lipid mixtures, we applied the model to Raman human aorta. Six biological components were chosen

over a wide range of protein spectra collected from intact

for the initial basis set, ri(v): collagen (bovine achilles tendon) (Fig. 2(b)), elastin (bovine neck ligament) (Fig. 2(a)), cholesterol (Fig. 4(a)), cholesteryl oleate (Fig. 4(c)), cholesteryl linoleate (Fig. 4(d)) and calcium hydroxyapatite (Fig. 6). The carbonated apatite region between 1100 and 1025 cm-’ was excluded in fitting the model to the data, because no sample of this compound is available. Again, the CH2 bending band area of each protein and lipid basis spectrum was normalized to unity, as was the symmetric phosphate stretching band in the calcium hydroxyapatite basis spectrum. In addition, the Raman spectrum of the buffered saline was included, as it improved the quality of the fits in the 1650 cm-’ region, where the weak O-H bending vibration of water makes a small contribution to the signal. Addition of cholesteryl palmitate as a basis spectrum did not significantly improve the fits of the data.

227 1

-0.05 I -0.05

-I 0

0.1

0.05

(4

measured

0.15

0.2

0.25

0.3

weight percentage

0.35 0

0.33 =

E zl 0. g

0.25. 0.2. 0.15-

(::,

n

2 2

2

0

O 0.1.

8O

0.05. o-

0 0

-0.05

0

0

0

0

,

0

-0.05 ,

0’)

0

0

s

0

0.05

0.1

0.15

0.2

0.25

0.3

measured weight percentage

Fig. 12. Plot of component weight percentages calculated from model vs measured weight percentages. (a) Cholesteryl oleate. The slope of the line is 0.64; the regression coefficient is 0.93; (b) Cholesteryl linoleate. The slope of the line is 0.98; the regression coefficient is 0.93. Measured and calculated m Raman spectra of typical specimens of normal aorta, atheromatous plaque, and exposed calcified atheromatous plaques are shown in Figs. 14, 15 and 16, respectively. Residuals of the fits are also plotted in these figures. Weight percentages for each component were computed from the fit coefficients using eqn. (4) and are listed in Table 3. Here, we have adopted the normalization condition that the weight percentages of the organic components (collagen, elastin, cholesterol, cholesteryl oleate, cholesteryl linoleate) for each spectrum sum to 1. In tissue, the weight percentages of these constituents will not in general sum to one due to the presence of other components in the tissue not detected in the Raman spectrum. The calculated spectra for both normal aorta (Fig. 14) and atheromatous plaque (Fig. 15) agree quite well with the measured spectra, with only minor deviations from the noise level in the residuals. This suggests that not only does the linear model hold for tissue, but also that the chosen basis spectra are a reasonable and nearly complete representation of the Raman spectra of the tissue biomolecules to within the spectral signal-to-noise levels. In addition, the calculated weight percentages of all the components are well within the reported ranges of these components for these

-I

-0.1

0

0.1 0.2 0.3 0.4 0.5 measured weight percentage

0.7

0.6

”

0.6

i:

$/@+zf+

O.lo-0.1, -0.1

0

i 0

0.1 0.2 0.3 0.4 0.5 measured weight percentage

0.6

Fig. 13. Plot of component weight percentages calculated from model vs. measured weight percentages. (a) Collagen. The slope of the line is 1.21; the regression coefficient is 0.89. (b) Elastin. The slope of the line is 0.68; the regression coefficient is 0.73.

tissue types. For example, the calculated collagen:elastin content of the normal aorta spectrum is 31%:62%, while that of the atheromatous plaque is 36%:17%. Also, the normal aorta spectrum yields 6% total cholesterol, the majority being cholesterol ester (oleate), which is consistent with biochemically measured levels [19]. This calculated level is near the detection limit for lipid and is likely significant. In contrast, the computed total cholesterol (cholesterol + cholesterol esters) content for the atheromatous plaque is 47%, with 14% cholesterol, 21% cholesteryl oleate and 12% cholesteryl linoleate. Again, these values are consistent with the known composition of this lesion type.

The agreement between the measured and calculated spectra of exposed calcification (Fig. 16) is not as good as with the non-calcified specimens. The two primary bands associated with the deposited calcium salts, 1070 and 960 cm-‘, are inadequately modeled with the spectrum of calcium hydroxyapatite alone. In both cases, the discrepancies are attributed to the omission of carbonated apatites from the model

229 0.25

0.2

3

0.15

.z 3 $

0.1

a9 .z 5 0.05 I 0

-0.05-1 . 1800

..I. 1600

?.,...,...,...,~..I 1400

1200

1000

800

600

Frequency (cm.')

Fig. 14. Measured Raman spectrum of normal aorta, along with model calculated fit and residual. (The negative spike at 1500 cm-’ is due to spurious noise.)

3

.E

: 0.18 m 21 .z 5 =

0.06

-0.054. 1800

.I 1600

., 1400

1200

., 1000

I 800

600

Frequency (cm-')

Fig. 15. Measured Raman spectrum of atheromatous plaque, along with model calculated fit and residual.

Carbonated apatites exhibit a band at 1070 cm-’ attributed to the symmetric CO3 stretching mode [17]. In addition, the width of the 960 cm-’ phosphate stretching

band, which in tissue is slightly larger than in pure hydroxyapatite, is known to increase with increasing carbonate substitution in hydroxyapatite [17]. Of the soft tissue components, the model calculates 68% collagen, 0% elastin, 9% cholesterol, 4% cholesteryl oleate and 20% cholesteryl linoleate.

230

1800

1600

1400

1200

1000

800

600

Frequency (cm-')

Fig. 16. Measured Raman spectrum of calcified atheromatous plaque (exposed calcification), along with model calculated fit and residual. The residual has been offset from zero for clarity. TABLE 3 Weight percentages

for human aorta calculated

from the Raman spectra

Biological component

Normal

Atheromatous

Exposed calcification

Collagen Elastin Total protein Cholesterol Cholesteryl oleate Cholesteryl linoleate Total lipid” Total cholesteryl esterb

0.31 0.61 0.93 0.003 0.064 0.002 0.068 0.066

0.35 0.18 0.53 0.14 0.21 0.12 0.47 0.32

0.68 -0.006 0.67 0.088 0.036 0.20 0.33 0.24

“Cholesterol + cholestexyl oleate + cholesteryl %holesteryl oleate +cholesteryl linoleate.

linoleate.

4. Discussion In order for Raman spectroscopy of human tissue to become a viable clinical histochemical method, one must be able to extract quantitative biochemical information from the Raman spectra. In this paper, we have shown that NIR m Raman spectra of human aorta can be understood in terms of the spectra of the individual biomolecules which are most prevalent in the tissue, that the signals behave in a linear manner even in a highly scattering environment, and that the signals can be modeled to extract relative quantitative information about the biological composition of atherosclerotic lesions. The current linear model for extracting the biochemical information may be limited in several ways. First, the basis set contains three shortcomings which will be refined in future models. The first is that the basis spectra themselves are currently limited

231

in signal-to-noise ratio. This reflects itself in the ability of the model to obtain parameters accurate to 515% when fitting mixtures of the biological components, and is particularly true for the basis spectra of collagen and elastin. The basis spectra can be collected for longer times to increase the signal-to-noise ratio and overcome this difficulty. The second problem is that the basis spectra may not be an exact representation of the data from human tissue. The spectra of collagen and elastin, which come from nonhuman sources, are particularly subject to this criticism. In the near future, we will collect these spectra from individual morphological structures using a Raman microprobe. The basis spectra will then more faithfully represent the bulk tissue data. Third, it has been assumed that the Raman cross-sections of purified tissue components examined as powder are the same as those in tissues. Fourth, there are additional species in arterial tissue which may contribute to the Raman spectra but which are not accounted for in the model. For example, in the spectra of calcified plaques, the residuals indicate an additional band at 1070 cn-‘, likely due to carbonated apatites. Finally, the simple linear model makes no attempt to take into account the scattering and inhomogeneities in the tissue. This is likely to be an important issue for solid structures in the tissue such as calcium hydroxyapatite or cholesterol crystals. We are currently studying the question of accounting for the tissue optics and the effects of inhomogeneous scattering on the Raman spectra. The ability of the model to analyze the mixtures of biological molecules indicates that the model was able to quantitatively determine the character of even complex mixtures with S-15% accuracy. When applied to the spectra from real specimens of tissue, the model yielded biochemical information which appeared to match well with what might be expected from particular lesions. However, we did not measure the concentrations of the species in tissue in any other way. Experiments are currently underway in our laboratory to compare the measurements of the lipids and other biomaterials found in tissue with NIR Raman spectroscopy with standard biochemical methodology. To understand the potential diagnostic utility of NIR Raman spectroscopy, it should be compared with other methods currently utilized in the vascular system for obtaining diagnostic information. Angiography provides information about the length and diameter of a lesion, but cannot supply any biochemical information. Angioscopy allows visualization of a lesion which may permit diagnosis of a thrombus or other clearly distinct features, but is limited in the type of data available. Ultrasound can yield information about the density of the material, and thus circumstantially diagnose calcified lesions, but is also very limited in the type of information that can be extracted. Finally, magnetic resonance imaging provides information about the blood flow within the vasculature, but currently has been limited in yielding other chemical information. Thus, NIR Raman appears to be unique in the detail and quantitative nature of the biochemical information it provides. If the information demonstrated here could be obtained in the clinical setting, it could be used to guide treatment. For example, before deciding on a particular therapy, the physician would measure the histochemical information of a lesion such as the percent of cholesterol and cholesterol esters, using NIR Raman spectroscopy. If the lesion did contain a large amount of cholesterol, cholesterol lowering drugs might be indicated before proceeding with a more destructive procedure such as balloon angioplasty. Eventually, the information provided by the Raman data could be correlated with observations such as the incidence of restenosis after balloon angioplasty, which may allow for a better determination of the correct treatment modality. Finally, with the NIR Raman technique, the availability of biochemical data regarding the composition

232

of atherosclerotic lesions in vivo could contribute significantly to understanding the etiology and pathogenesis of atherosclerosis. Furthermore, the techniques described here are applicable to other tissues and pathologies as well. For instance, histological detection of malignancies and premalignancies depends in part on determining increases and/or alterations in nuclear material. Since Raman spectroscopy is capable of probing nucleic acids [20], this technique may potentially be used to monitor relative nucleic acid concentrations in vivo. NIR Raman spectral differences among normal, benign and malignant tissues have already been observed [7]. NIR Raman may also offer a method for real-time monitoring of blood components [8]. Again, clinical utility requires quantitative analysis of the spectra similar to that described above.

5. Conclusions

We have shown that NIR Raman spectroscopy of human tissue can be utilized to extract quantitative information about the histochemical composition of the tissue. Since the Raman signal appears to be linear with respect to concentration even in a highly scattering medium, we utilized a simple linear superposition of the spectra of individual tissue components expected to contribute significantly to the observed signal. These included collagen, elastin, cholesterol, cholesteryl oleate, cholesteryl linoleate and calcium hydroxyapatite. The model showed itself to be accurate to within 15% when tested with known powder mixtures. Currently, we are comparing the results obtained from this model with biochemical extractions carried out on human artery. In this fashion, we will determine the limits of applicability for real samples. Ih the long term, we wish to increase the sensitivity of the method to be able to extract biomolecules of lower concentration by increasing the signal-to-noise ratio of the spectra. These studies are being carried out using a spectrograph/CCD based system. In addition, we are exploring the application of these methods to other pathologies.

Acknowledgments

The authors would like to thank Vinny Moylan, Mark Knowlton and Connie Austin of the Brigham and Women’s Hospital in Boston for providing arterial samples. This work was completed at the NIH-supported M.I.T. Laser Biomedical Research Center (NIH Grant No. RR02594).

References

1 P. R. Carey, Biochemical Applications of Raman and Resonance Raman Spectroscopy, Academic Press, New York, 1982. 2 T. G. Spiro (ed.), Biological Applications of Raman Spectroscopy, Wiley, New York, 1987. 3 K. A. Hartman, N. W. Clayton and G. J. Thomas, Jr., Biochem. Biophys. Res. Commun., 50 (1973) 4

942.

N.-T. Yu, B. H. Jo, R. C. C. Chang and J. D. Huber, Arch. Biochem.

614. 5 G. J. Rosasco, in R. Spectroscopy, Vol. 7,

J. H. Clark and R. E. Hester (eds.), Advances Wiley-Heyden, London, 1980.

Biophys.,

I60 (1974)

in Infrared

and Raman

233

6 S. Nie, K. L. Bergbauer, J. J. Ho, J. F. R. Kuck and N.-T. Yu, Spectroscopy, 5 (1990) 24. 7 R. R. Alfano, C. H. Liu, W. L. Sha, H. R. Zhu, D. L. Akins, J. Cleary, R. Prudente and E. Cellmer, Lasers Life Sci., 4 (1991) 23. 8 Y. Ozaki, A. Mixuno, H. Sato, K. Kawauchi and S. Muraishi, Appl. S’ctrosc., 46 (1992) 533. 9 R. P. Rava, J. J. Baraga

10 J. J. Baraga, 11 J. J. Baraga, 12 R. S. Cortran, PA, 1989. 13 E. B. Smith,

and M. S. Feld, M. S. Feld and R. P. Rava, M. S. Feld and R. P. Rava, V. Kumar and S. L. Robbins,

Spectrochim. Actu, 47A (1991) 509.

Proc. Natl. Acad. Sci. USA, 89 (1992) 3473. Appl. Spectrosc., 46 (1992) 187. PathologicBasis of Disease, Saunders,

Philadelphia,

in W. D. Wagner and T. B. Clarkson (eds.), Advances in ExperimentalMedicine and Biology, Vol. 43, Plenum Press, New York, 1974, pp. 125-139. 14 D. M. Small, Arteriosclerosk, 8 (1988) 103. 15 N. Abe, M. Wakayama and M. Ito, J. Ruman S’ctrosc., 6 (1977) 38. 16 J. M. Dudik, J. R. Johnson and S. A. Asher, J. Chem. Phys., 82 (1985) 1732. 17 D. G. A. Nelson and B. E. Williamson, Aust. J. Chem., 35 (1982) 715. 18 W. H. Press, B. P. Flannery, S. A. Teukolsky and W. T. Vetterling, Numerical Rec&s in C, Cambridge University Press, New York, 1988, Chap. 14. 19 S. S. Katz, J. Biol. Chem., 256 (1981) 12275. 20 See, for example, T. J. Thamann, R. C. Lord, A. H. J. Wang and A. Rich, Nucleic Acids Res., 9 (1981) 5443.