SEARCH

SEARCH BY CITATION

Keywords:

  • Raman spectroscopy;
  • breast cancer;
  • optical diagnosis;
  • curve deconvolution

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS AND DISCUSSION
  6. CONCLUSIONS
  7. Acknowledgements
  8. REFERENCES

The aim of this study was to understand and correlate spectral features and biochemical changes in normal, fibroadenoma and infiltrating ductal carcinoma of breast tissues using Raman spectra that were part of the spectroscopic models developed and evaluated by us earlier. Spectra were subjected to curve fitting and intensities plots of resultant curve resolved bands were computed. This study has revealed that fat (1301 and 1440 cm−1), collagen (1246, 1271, and 1671 cm−1) and DNA (1340 and 1480 cm−1) bands have strong presence in normal, benign and malignant breast tissues, respectively. Intensity plots of various combinations of curved resolved bands were also explored to classify tissue types. Combinations of fat (1301 cm−1) and collagen (1246, 1271, and 1671 cm−1)/amide I; DNA (1340 cm−1) and fat (1301 cm−1); collagen (1271 cm−1) and DNA (1480 cm−1) are found to be good discriminating parameters. These results are in tune with findings of earlier studies carried out on western population as well as our molecular biological understanding of normal tissues and neoplastic processes. Thus the finding of this study further demonstrates the efficacy Raman spectroscopic approaches in diagnostic applications as well as in understanding molecular phenomenon in breast cancers. © 2009 Wiley Periodicals, Inc. Biopolymers 91: 539–546, 2009.

This article was originally published online as an accepted preprint. The “Published Online”date corresponds to the preprint version. You can request a copy of the preprint by emailing the Biopolymers editorial office at biopolymers@wiley.com


INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS AND DISCUSSION
  6. CONCLUSIONS
  7. Acknowledgements
  8. REFERENCES

Breast cancer is one of the major heath hazards amongst women in both developing and developed countries. It stands second in terms of incidence rates when both genders are considered together and fifth in mortality rate.1 The significant increase in the number of cases since 1970s can be partly attributed to urbanized lifestyles.2 The imaging technique, mammography is the gold standard for screening. False positives and risk of repeated exposure to harmful ionizing radiations are identified as major concerns in this method. The histopathology, gold standard of diagnosis, involves thorough microscopic examination of samples obtained via well-established invasive/minimally invasive tissue sampling procedures (i.e., excisional biopsy, axillary node dissection, sentinel node dissection, or fine needle aspiration), depending on the location, size, palpability, and characteristics of the abnormality.3 However, histopathological diagnosis has been shown to be subjective and depends on the experience of skilled pathologist. Thus, the complete conventional diagnostic regimen involves various examinations of triple tests (clinical breast examination, imaging and tissue sampling) alongwith multiple biopsies mounting financial burden and patient's anxiety. Hence, there exists a need to develop alternate/adjunct, user-friendly, cost effective, rapid, objective and unambiguous methods for early detection and diagnosis of breast cancers.

Potential applications of optical spectroscopic techniques in biology and medicine, including cancer diagnosis have been well investigated and documented.4–11 Though, Raman scattering is inherently weak, it is a more suitable tool for biomedical applications. This is due to following attributes of the technique: use of harmless near infrared radiation as excitation source; minimal or no sample preparation; and adaptability to biomedical applications especially for in vivo/in situ applications.

Pertaining to Raman spectroscopic applications in breast cancer diagnosis, several studies have clearly demonstrated the efficacy of Raman spectroscopy.12–20 Among Raman spectroscopy methods, Raman microspectroscopy, which facilitates high spatial resolution up to 1 μm, is often used for in depth analysis of selected regions. On the other hand conventional or macro Raman spectroscopy is better suited for diagnostic applications, especially for in situ/in vivo conditions. This is because, representative spectrum can be acquired by conventional Raman spectroscopy as probing area (20–100 μm) larger in this mode. Moreover, findings of conventional spectroscopy of ex vivo samples, to a reasonable extent, can be extrapolated to in vivo/in situ conditions. However, rigorous validation of the approach by several independent groups, preferably on different ethnic populations, is a prerequisite before contemplating to routine use. In view of these considerations, we had taken up conventional Raman spectroscopy investigations to classify ex vivo specimen of normal, fibroadenoma and infiltrating ductal carcinoma — the most prevalent benign and malignant conditions,6 respectively.21, 22 In these studies training sets were developed, verified and evaluated by larger data sets consisting of certified as well as blinded ex vivo specimen.6, 21, 22 As a logical and subsequent step, we have attempted to understand and correlate spectral features and possible biochemical changes in normal, benign and malignant breast tissues by curve fitting. The findings of curve deconvolution analysis of Raman spectra that are used in the established normal, benign and malignant models are discussed in the present paper.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS AND DISCUSSION
  6. CONCLUSIONS
  7. Acknowledgements
  8. REFERENCES

Sample Collection and Processing

Breast tissue specimens, from routine surgical resections, were procured in ice-cold saline from Department of Surgical Oncology, Shiridi Sai Baba Cancer Hospital and Research Center and Department of General Surgery, Kasturba Hospital, Manipal University, Manipal. The size of the tissue samples were ∼2–5 mm in thickness and 20–25 mm2 surface area. Around six or more sites were recorded on both sides of each specimen depending on the size. Each spectrum is treated as an individual sample.8, 23 In our earlier studies, a total of 258 sites/spectra (105 normal, 101 malignant, 52 benign) were recorded on histopathologically certified 29 normal, 24 infiltrating ductal carcinoma and seven fibroadenoma breast tissues.21 Among them, randomly selected 36 normal, 35 malignant, and 21 benign breast tissue spectra were used for development of spectroscopic models. These models were then verified and evaluated by 69 blinded samples.22 In the present study spectra that were part of above mentioned established models were analyzed by curve deconvolution methods.

Laser Raman Spectroscopy

Raman spectra were recorded with the set up assembled in our laboratory.6, 8, 23, 24 In brief, the set up comprises of diode laser; SDL-8530 (785 nm, 100 mW) and HR 320 spectrograph (600 g/mm blazed at 900 nm) coupled to Spectrum One liquid N2 cooled CCD (Jobin Yvon-Spex, Instruments S.A.) as excitation source and detection system, respectively. A holographic filter (HLBF-785.0, Kaiser Optics) was used to filter the excitation source. Rayleigh lines were removed by a notch filter (HSPF-5812, Kaiser Optics). In the present setup the theoretically achievable laser spot size is ∼17 μm. However, under the present experimental condition spot size could be larger than this.25 An integration time of 30 s and 20 accumulations were kept constant for all the spectral measurements. Tissues were kept moist with saline during spectral acquisition.

Data Analysis

Spectral Preprocessing

All the spectra were baseline corrected, smoothened and post calibrated with a cubic orders fit to known frequencies of Tylenol (4-acetamidophenol) using diode adjust algorithms by GRAMS 32 (Galactic Industries Corporation). Then the spectra were normalized to δCH2 band.

Curve Deconvolution

The preprocessed spectra were resolved into individual component bands using standard curve deconvolution technique based on original algorithm of nonlinear peak fitting as described by Marquardt which is also known as Lavenberg-Marquardt method.26 In these fits, the sum of the squared differences between observed and computed spectra are minimized to get the best fit. The spectral region used for the curve analysis ranges from 1180 to 1800 cm−1 region. Normal, malignant and benign breast tissue spectra were resolved into 17, 17, and 18 bands, respectively. Each band is characterized by four parameters: shape factor, peak position, peak intensity and full width at half maximum. The shape factor decides the band shape. In our curve analysis, Gaussian band has lead to good fit.

In order to understand biochemical level information, we have compiled scatter intensities plots of curve resolved bands. Area under the curve (integrated intensity) is used to compute intensity plots.

RESULTS AND DISCUSSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS AND DISCUSSION
  6. CONCLUSIONS
  7. Acknowledgements
  8. REFERENCES

As described earlier, in our previous studies we had developed spectroscopic models for normal, malignant and benign breast tissues from randomly selected spectra of certified samples. These training sets were challenged by large testing data from certified6, 21 as well as 69 (427 spectra) blinded specimen.6, 22 Typical spectra of established Raman models of normal, malignant and benign breast tissues are shown in Figure 1. The spectra, as can be seen from the figure, exhibit significant discriminant features which can be easily exploited for classification as already been demonstrated by us.6, 21, 22 Further, Raman spectra can also give significant clues to understand biochemical nature of specimen. Even cursory/qualitatively look of spectra suggests the predominance of lipids in normal tissues and proteins in pathological breast tissues.6, 21, 22 However, one should note that, as already been mentioned earlier, spectra obtained by conventional spectroscopy represent gross information acquired over larger probing regions consisting of different regions. Still, by proper data mining it would be possible to correlate biochemical variations through spectral profiles of pathological conditions. There are several approaches to achieve this. The most common method, as described above, is to assign the various vibrational modes of mean spectra. This approach could provide primary level understanding. Difference spectra which are computed by subtracting one spectrum from the other is slightly better approach.21, 27, 28 Factor loadings of PCA or analysis using models are often used for data mining. Although, multivariate statistical tools such as PCA are often used to develop objective diagnostic tool, this analysis may not be very useful for biochemical correlation. This is because, at times PCA is carried out using selected regions of either derivative or nonderivative spectra in order to bring out classification of tissues. In our studies of breast cancers PCA of 1400–1750 cm−1 spectral region gave best classification.6, 21, 22 Though, factor loadings are also used for biochemical correlation,29 most often PCA is employed for the reduction of dimensionality of the data and the eigenvectors may not always provide realistic shape for biochemical correlation.30

thumbnail image

Figure 1. Mean Raman spectra of normal, malignant and benign breast tissue. Solid line—normal; broken line—malignant; and dotted line—benign breast tissue.

Download figure to PowerPoint

Another approach is to analyze the data using spectroscopic models. In the case of breast cancers, a recent study has employed a model based on linear combination of basic spectra.19 This approach was implemented not only to classify tissue types but also to analyze biochemical composition of the specimen. We have explored classical spectral deconvolution analysis to understand biochemical differences due to radiation induced effects in mouse models31 and oral tissues.25, 32 We have also explored this approach for autofluorescence spectra of breast33 and oral tissues.34 To make curve fitting more consistent, analysis was carried out on larger data set and scatter plots depicting contribution (band intensity) of curve deconvoluted bands were computed.33, 34 Despite being subjective and tedious process to some extent from the classification/diagnosis point of view, this approach provides wealth of information about the biomolecules and their distribution.

In the present study, we have carried out spectral deconvolution of the spectra involved in the Raman spectroscopic models. As already been mentioned in “Materials and Methods”, normal, benign, and malignant spectra were deconvoluted by fitting 17, 17, and 18 peaks which are denoted by indo-arabic numerals (Figures 2–4). Typical curve deconvoluted spectra of normal, malignant and benign breast tissues are shown in Figures 2–4, respectively. Tentative assignments of the curve resolved bands which are shown in Table I were made based on the available literature data.29, 35–39

thumbnail image

Figure 2. Typical spectral deconvolution of Raman spectra of normal breast tissue. (A) Amide III, (B) δCH2, and (C) Amide I regions. (2D – 2nd Derivative; FT – Fitted Trace; R – Residual).

Download figure to PowerPoint

thumbnail image

Figure 3. Typical spectral deconvolution of Raman spectra of malignant breast tissue. (A) Amide III, (B) δCH2, and (C) Amide I regions. (2D – 2nd Derivative; FT – Fitted Trace; R – Residual).

Download figure to PowerPoint

thumbnail image

Figure 4. Typical spectral deconvolution Raman spectra of benign breast tissue. (A) Amide III, (B) δCH2, and (C) Amide I regions. (2D – 2nd Derivative; FT – Fitted Trace; R – Residual).

Download figure to PowerPoint

Table I. Tentative Assignments of Major Vibrational Modes of Raman Spectra Identified in Breast Tissues
Peak Positions/cm−1Major Assignments
1230PO2 ant. symmetric stretch
1245Amide III, disorder structures of proteins, collagen
1268δ([DOUBLE BOND]CH) phospholipids
1271Amide III; collagen
1301τ(CH2), lipids
1315Amide III (α–helix)
1340DNA (Adenine)
1364CH3 symmetric deformation
1376CH3 symmetric deformation
1422υs(COO) of proteins
1440δ(CH2), lipids
1456δas(CH3) and δ(CH2) of proteins
1471δas(CH3) and δ(CH2) of proteins
1480DNA (adenine, guanine)
1531Tryptophan (Trp)
1554Tryptophan (Trp)
1586Phe
1621Trp (IgG), Phe, Tyr
1639Amide I (both α–helix and β–structures)
1648Amide I (random coils of proteins)
1652Amide I; υ(C[DOUBLE BOND]O stretch of proteins)
1655Amide I (C[DOUBLE BOND]O)/υ (C[DOUBLE BOND]C stretch)cis phospholipids
1671Amide I (collagen)
1743υ(C[DOUBLE BOND]O) lipids, esteryl, carbonyl, phospholipids

The curve resolved bands of normal tissue spectra indicate strong lipid bands (1230, 1268, 1301, 1440, and 1743 cm−1) and weak protein (1246, 1271, 1315, 1364, 1422, 1458, 1620, 1639, 1671 cm−1) as well as nucleic acid bands (1340 and 1480 cm−1). In case of vibrational modes around 1655 cm−1 can be attributed υ(C[DOUBLE BOND]C)cis stretch of lipids and amide I (C[DOUBLE BOND]O stretch) of proteins. In the case of spectra of pathological conditions i.e., malignant and benign, lipid bands are quite weak (1268, 1301, 1440 cm−1) compared to protein bands (1246, 1271, 1315, 1424, 1456, 1531, 1554, 1586, 1621, 1650, 1671 cm−1) as well as nucleic acid bands (1340 and 1480 cm−1). These features suggest the dominance of collagen, noncollagenous proteins, nucleic acid in pathological tissues compared to normal tissues. Among the pathological conditions, intensities of curve resolved components attributed to proteins especially collagenous proteins are relatively stronger in benign tissue spectra, Figures 3 and 4. Therefore, stromal components, especially collagen seem to be abundant in fibroadenoma when compared to infiltrating ductal carcinoma.

In order to understand the relative contributions of these curve resolved bands, we have computed scatter intensities plots as shown in Figures 5–8. The scattered intensities plots of lipid bands: 1301 and 1440 cm−1 (Figures 5A and 5B) clearly indicate the dominance of lipid features in normal breast tissue. The contributions of these bands are found to be lowest in benign conditions and the clusters of malignant spectra occupy intermediate position between normal and benign clusters. Strong presence of collagen bands (1246, 1271, 1315, 1455, 1656, and 1671 cm−1) seen in fibroadenoma tissues. Contributions of collagen are the lowest for normal tissues and malignant tissues occupied intermediate position, Figure 7. Strong presence of nucleic acids (1340 and 1480 cm−1) are seen in malignant conditions, Figure 8. However, contributions of protein bands are not very conclusive. In this case, relatively higher presence of one of the amide III bands, at 1315 cm−1, attributed to helical structures seen in benign breast tissue, as shown in Figure 6A. normal and malignant tissues exibit similar levels of these bands. When we consider intensities of 1455 and 1655 cm−1 bands, presence of this peaks are higher for pathological tissues compared to normal tissues.

thumbnail image

Figure 5. Scatter plots of intensities of curve resolved lipid bands (A) 1301 cm−1 and (B) 1440 cm−1 in Raman spectra of normal (○), malignant (□) and benign (▵) breast tissue.

Download figure to PowerPoint

thumbnail image

Figure 6. Scatter plots of intensities of curve resolved protein bands (A) 1315 cm−1, (B) 1455 cm−1, and (C) 1656 cm−1 in Raman spectra of normal (○), malignant (□) and benign (▵) breast tissue.

Download figure to PowerPoint

thumbnail image

Figure 7. Scatter plots of intensities of curve resolved collagen bands (A) 1246 cm−1, (B) 1271 cm−1, and (C) 1671 cm−1 in Raman spectra of normal (○), malignant (□) and benign (▵) breast tissue.

Download figure to PowerPoint

thumbnail image

Figure 8. Scattered plot intensities of curve resolved DNA bands (A) 1340 cm−1 and (B) 1480 cm−1 in Raman spectra of normal (○), malignant (□) and benign (▵) breast tissue.

Download figure to PowerPoint

The above findings, i.e. abundance of lipids in normal, helical proteins and collagen in fibroadenoma, and nucleic acids and noncollagenous proteins in infiltrating ductal carcinoma are in good agreement with a recent study.19 In this study spectra were analyzed by model which were built by linear combination of basic spectra. Strong lipid features in normal tissue spectra could be due to the fact that majority of glandular breast is composed of adipose tissue (lipids) and lipids have a larger Raman scattering cross section relative to epithelial component. Therefore contributions from epithelial component might have been swamped by adipose tissue features as opined by the earlier workers.19, 22 As is well known, fibroadenoma is an expansile lesion characterized by increased levels of stromal components due to proliferating fibroblasts and accumulation of collagen. Our own autofluorescence spectroscopy studies which were carried out on sections of same normal, benign and malignant breast tissues also corroborate the abundance of collagen in benign conditions.32 Further, collagen and NAD(P)H are also well established intrinsic fluorescent spectral markers in malignancy.40 Neoplastic process is basically a cellular process and hence biochemically observed to have higher levels of noncollagenous protein. And enlargement of cell nuclei is a hallmark of cancer and is one of the basis of histopathological diagnosis of cancers. Thus, our observation of increased contributions from DNA bands at 1340 and 1480 cm−1 in malignant spectra (see Figure 8) is also in corroboration with basics of neoplastic process. However, as mentioned earlier, intensity plots of protein bands like 1315, 1455, and 1655 cm−1 are not conclusive.

Based on the above analysis, fat, collagen, and nucleic acids can be used for characterization of normal, fibroadenoma and infiltrating ductal carcinoma tissues. Therefore, we have explored various combinations intensities of these spectral markers in tissue type discrimination. Among the combinations explored of lipid band (1301 cm−1) against collagen lines (1246, 1271, and 1671 cm−1) has provided very good classification of three tissue types (Figure 9A, 9B, and 9D). We also observed that intensities of lipid (1301 cm−1) against amide I (1655 cm−1); DNA (1340 cm−1) against lipid (1301 cm−1); and collagen (1271 cm−1) against DNA (1480 cm−1) can bring out clear classification among three tissue types as shown in Figure 9C, 9E, and 9F.

thumbnail image

Figure 9. Discrimination of breast tissues: Scattered plot intensities of curve resolved bands. (A) Protein (1246 cm−1) vs Lipid (1301 cm−1), (B) Protein (1271 cm−1) vs Lipid (1301 cm−1), (C) Protein (1655 cm−1) vs Lipid (1301 cm−1), (D) Protein (1671 cm−1) vs Lipid (1301 cm−1), (E) Lipid (1301 cm−1) vs DNA (1340 cm−1), and (F) Protein (1271 cm−1) vs DNA (1480 cm−1). (○) Normal; (□) Malignant; and (▵) Benign.

Download figure to PowerPoint

CONCLUSIONS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS AND DISCUSSION
  6. CONCLUSIONS
  7. Acknowledgements
  8. REFERENCES

As evident from literature, several reports have indicated the efficacy of Raman spectroscopic approach in classification types. Raman spectral features are shown to be very useful tools for probing biomolecular composition of normal and pathological breast tissues. However, evaluation of these models by several independent groups and preferably over different ethnic groups is necessary before considering for routine clinical use.

The present study has demonstrated strong presence of lipids in normal, collagen in fibroadenoma and DNA in infiltrating ductal carcinoma tissues. Combination of fat and collagen features shown to provide clear classification of all three tissue types. These results corroborate findings of previous studies carried out on American population using models developed by linear combination of basic spectra. Thus the present study further supports the efficacy of Raman spectroscopic approach in diagnosis as well as in understanding biochemical composition of breast tissues.

Acknowledgements

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS AND DISCUSSION
  6. CONCLUSIONS
  7. Acknowledgements
  8. REFERENCES

This work was carried out under the project entitled: “Development of laser spectroscopy techniques for early detection and follow up of therapy in breast malignancy” (No. 5/13/23/2003-NCD-III), Indian Council for Medical Research, Government of India. The authors (K.K.K. and M.V.P.C.) are thankful to ICMR for senior research fellowships. Ms. Keerthi and Mr. Chethan N. Anand are acknowledged for their technical assistance in the sample handling/preservation and data acquisition.

REFERENCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS AND DISCUSSION
  6. CONCLUSIONS
  7. Acknowledgements
  8. REFERENCES