3360 OPTICS LETTERS / Vol. 37, No. 16 / August 15, 2012

Stokes shift highlights differences of cancerous and normal human tissues

Yang Pu, Wuabo Wang, Yuanlong Yang, and R. R. Alfano* Institute for Ultrafast Spectroscopy and Lasers, and Department of Physics, City College of the City University of New York, Convent Avenue at 138th Street, New York, New York 10031, USA *Corresponding author: [email protected] Received May 15, 2012; accepted June 25, 2012; posted June 28, 2012 (Doc. ID 168649); published August 6, 2012 The Stokes shift spectroscopy (S3) offers a simpler and better way to recognize spectral fingerprints of fluorophores in complex mixtures. The efficiency of S3 for cancer detection in human tissue was investigated systematically. The alterations of Stokes shift spectra (S3) between cancerous and normal tissues are due to the changes of key fluor- ophores, e.g., tryptophan and collagen, and can be highlighted using optimized wavelength shift interval. To our knowledge, this is the first time to explicitly disclose how and why S3 is superior in comparison with other conventional spectroscopic techniques. © 2012 Optical Society of America OCIS codes: 170.0170, 170.6510, 300.0300, 300.6170.

Fluorescence spectroscopy has been widely investigated These differences may reflect tissue fluorophores’ for diagnosing cancer since the study by Alfano et al. in change during the evolution of cancer. In order to under- the 1980s [1]. The difference between the emission and stand which components mainly contribute to these absorption peaks is known as the Stokes shift interval, changes, S3 of the main fluorophores in breast tissue, Δλss. A spectroscopic method was proposed to acquire e.g., tryptophan, collagen, NADH, and flavin, need to be the signal by a fixed wavelength shift inter- measured. S3 val (Δλi) between the excitation λexc and emission λem, The of a mixture solution of tryptophan, NADH, and which is termed as Stokes shift spectroscopy (S3) by flavin with Δλi ˆ 40 nm is displayed as solid line in 3 Alfano and Yang [2]. Although this spectral approach was Fig. 2(a).TheS of collagen for same value of Δλi used in multiple fluorophore analysis in tissue, no other was superposed as a dashed line. Since collagen is not groups in the United States are applying it in cancer de- soluble in water, it is hard to obtain the S3 of the mixture tection. This is because the previous studies did not in- including collagen. The aqueous collagen suspension was vestigate the reason why S3 is superior over absorption, shaken evenly before the measurements. Compared with fluorescence, and excitation-emission matrix (EEM) Fig. 1, Fig. 2(a) shows that the main peak at ∼290 nm measurements. Therefore, few researchers are aware for the S3 of the breast tissues is from tryptophan. The of the excellence of S3 technique in cancer diagnostic secondary main peak at ∼340 nm corresponds to col- application in tissues. lagen, and the very tiny peak at ∼380 nm stands for This Letter will demonstrate the efficiency of S3 to re- NADH. No obvious peak of flavin was observed. There- 3 cognize spectral fingerprints of fluorophores in complex fore, the S profiles of breast tissues acquired with Δλi ˆ mixtures and its application to highlight the difference 40 nm are mainly contributed from tryptophan, collagen, between cancerous and normal tissues. We will also and NADH. show how to select an optimal wavelength shift interval To study the relative content changes of fluorophores, to obtain the best Stokes shift spectra (S3) for the pur- an analytical method, namely the nonnegative least pose of cancer detection, and discuss why S3 is superior square (NNLS) method, was applied to the S3 of the over other conventional spectroscopic techniques. As an example, the averaging S3 of 15 pairs of cancer- ous (solid line) and normal (dash line) breast tissues 0.024 were recorded by setting Δλi ˆ 40 nm in the synchro- 0.16 nized scan mode of a spectrometer (Perkin-Elmer LS 50), 0.020 385nm and shown in Fig. 1. The scan speed was 300 nm per 0.016 minute. Each spectral profile was normalized to a unit 0.12 value of 1 (i.e., the sum of squares of the intensity ele- ~294 nm 0.012 ments in each was set as 1). The sali- 0.08 0.008 S3 370 380 390 400 410 ent difference of between cancerous and normal wavelength (nm)

breast tissues can be observed as two reverses of the ~340 nm 0.04 peak intensities at ∼I294 and ∼I340: (1) Ic >In at Intensity (arb. units) cancer normal ∼294 nm, while Ic I340 for cancer, while I294

(a) 1000 cancerous and normal breast tissues, shown in Fig. 1 to 1000 (b) 800 extract relative contents of the fluorophores, e.g., trypto- 800 3 × 10 phan, collagen, and NADH, using the measured S 600 600 20nm signal acquired with Δλi ˆ 40 nm shown in Fig. 2(a). 60nm 80nm 400 400 Figure 2(b) shows the scatter plot of the relative content 100nm 120nm 200 200 Intensity (arb. units) for tryptophan versus collagen of cancerous (square) and Intensity (arb. units) 140nm normal (circle) breast tissues. A separating line on the 0 0 300400 500 600 300 400 500 600 scatter plots was loaded by the Linear discriminant Wavelength (nm) Wavelength (nm) analysis (LDA) model for the diagnostic significance of Fig. 3. (Color online) S3 of mixed solution of tryptophan, tryptophan versus collagen. The sensitivity and speci- NADH, and flavin obtained by (a) Δλc ˆ 20 nm (solid), 60 nm ficity were calculated as 80% and 86.7%, respectively. (dash), and 80 nm (dot); and (b) 100 nm (solid), 120 nm (dash), The most salient feature of Fig. 2(b) is that all data points and 140 nm (dot). for the normal tissue locate in the upper-left side in com- parison with the data for the cancerous tissue, indicating 3 Δλ that the relative contribution of collagen to the S signal When i reaches 120 nm, the NADH signal almost takes in the normal tissue is higher that of the cancerous over the flavin signal. S3 Δλ tissues, while the relative contribution of tryptophan in To quantitatively study the changes of with i, the S3 normal tissue is lower than that in the cancerous tissue. FWHM and the peak intensities of profiles as a func- Δλ Reproducible results were observed for other kinds of tion of i for three fluorophores in the solution are human tissue studies, such as prostate and lung cancers. shown as Figs. 4(a) and 4(b), respectively. Figure 4 is Δλi Similarly, two inverse spectral properties of (1) Ic >In at very useful for researchers to choose an optimal . For signal processing, two most important properties de- ∼294 nm, while Ic I340 for cancer, while I294

400 listed. (a) tryptophan 0.8 (b)

0.7 300 collagen 160 40nm (a) 1000 (b) collagen 0.6 200 NADH 0.5 120 800

flavin ) collagen cancer tryptophan 0.4 nm tryptophan 100 normal NADH 600 NADH 80 flavin flavin Intensity (arb. units) 0.3 400 FWHM ( 0 0.2 300 400 500 600 0.5 0.6 0.7 0.8 0.9 40 200 Wavelength (nm) tryptophan Peak intensity (arb. units) 3 Fig. 2. (Color online) (a) S of mixed solution of tryptophan, 2040 60 80 100 120 140 20 40 60 80 100 120 140 Wavelength shift ∆λ∆λ (nm) Wavelength shift ∆λ∆λ (nm) NADH, and flavin obtained by Δλc ˆ 40 nm (dash) and aqueous suspension collagen; (b) The relative content of tryptophan ver- Fig. 4. (Color online) (a) FWHM and (b) S3 peak intensities of sus collagen in cancerous and normal breast tissues by analyz- tryptophan (square–solid), NADH (circle–dash) and flavin 3 ing their S obtained with the selective Δλc ˆ 40 nm. The (hexagon–dot) in the mixed solution changed as a function separating lines were calculated using the LDA method. of Δλc. 3362 OPTICS LETTERS / Vol. 37, No. 16 / August 15, 2012

3 Table 1. S3-Related Parameters of Key Fluorophores Therefore, the S with selective Δλi ˆ 40 nm highlights tryptophan collagen NADH flavin the difference between cancerous and normal tissues and causes the two inverse spectral properties exhibited λ abs (nm) 280 340 340 450 by Fig. 1. λem (nm) 340–350 380–390 440–460 520 λ – – – The conventional spectral methods applied in tissue ss (nm) 70 80 40 50 100 120 70 optics studies are absorption (O.D.), emission, excita- Δλ (nm) 80 — 120 60 drop tion, and EEM measurements. In the absorption mea- surements, only a few chromophores can match the detectable level in tissue. The fluorescence spectro- Table 1 can be used to explain the changes of the S3 scopy, either emission or excitation, can detect low level peak intensities with Δλi for different fluorophores. 3 fluorophores, but because of the fixed pump (detecting) When S was acquired by Δλi ˆ 20 nm, the excitation wavelength, the strongest emission (excitation) signal of all three fluorophores is far from the Stokes shift in- can be acquired only for one or two fluorophores. In ad- terval, Δλ , which results the smallest S3 signal intensity. ss dition, the emission signals of most fluorophores have Since the Stokes shift interval, Δλ of flavin and trypto- ss very wide FWHM. So emission and excitation measure- phan is ∼70 nm and ∼70–80 nm, respectively, the mag- ments provide poorer resolution and much less informa- nitude of S3 signals from these two fluorophores is tion than S3. Although EEM can be used to ensure the boosted with the increase of Δλi within the range of coverage of all endogenous fluorophores, the acquisition Δλ at first, but the intensity drops at Δλi ˆ 60 nm ss is extremely time consuming, and thus not suitable for for flavin and 80 nm for tryptophan after it exceeds their clinical application. Furthermore, redundant information Δλ . The same reason causes the peak intensity of ss of EEM conceals alteration of the spectral fingerprints of NADH to regress back at Δλi ˆ 120 nm after it increases. cancerous and benign breast tissues. In contrast, S3 mea- The observed Δλ is in good agreement with Δλ . The drop ss surements can be used to acquire enough information of drop point for collagen is expected to be ∼40–50 nm. multiple key fluorophores in a much lower concentration Tryptophan, collagen, NADH, and flavin are key mole- and relatively higher resolution by employing a single cules in cancer diagnosis using spectroscopy [3]. For scan (compared with EEM). This approach thus can breast cancer, the most common grading system used dramatically reduce data acquisition time and keep the in the U.S. is the Scarff–Bloom–Richardson (SBR) system classification accuracy reasonably high. [4], which is a breast cancer staging system to determine In summary, the S3 method offers an efficient way how aggressive and invasive the cancer is [4]. According to rapidly measure spectral fingerprints of complex to the features described by the SBR system, the higher mixtures such as tissue and highlights the differences be- cell density is the hallmark of breast cancer; therefore the tween cancerous and normal tissues. To our knowledge, increase of fluorescence from the main fluorophores in- it is the first time to study why and how the wavelength side cells—e.g., tryptophan, NADH, and flavin—should 3 interval Δλ should be optimized to obtain better S for be expected. The primary fluorophore in the breast tissue i cancer detection in human tissue. This study demon- extracellular matrix is collagen [5]. For invasion and sub- strates that the S3 measurements can be used to acquire sequent metastasis, tumor cells degrade the surrounding information for different key fluorophores in one scan extracellular matrix (ECM), which is composed mainly and used to investigate the changes of the relative con- by collagen [4]. Understanding these changes during tents of the key fluorophores in breast tissues due to the breast cancer evolution is critical to reveal the contribu- development of cancer. tions of the fluorophores in tissues using spectroscopic techniques. This research is supported in part by the U.S. Army Based on Figs. 2 and 3, one should choose the optimal Medical Research and Material Command (USAMRMC) Δλi as small as possible if resolution is the only consid- under grants W81XWH-08-1-0717 (CUNY RF # 47170- eration. However, in order to enhance the signal-to-noise 00-01) and W81XWH-11-1-0335 (CUNY RF # 47204-00-1). ratio (SNR), the optimal Δλi should be chosen as close as Yang Pu acknowledges useful discussion with Prof. to Δλss. In particular application in breast cancer detec- Y. Sun at the Electrical Engineering Department of City tion, the alterations of fluorophores due to the cancer de- College of New York. The authors acknowledge the help velopment direct an optimal Δλi. In the spectral analysis of Cooperation Human Tissue Network and National Dis- on cancer detection, it is very difficult to calculate the ease Research Interchange for providing normal and can- absolute concentration of the fluorophores. One needs cerous tissue samples for the measurements. to find an unchanged component to be a reference to ob- References serve the changes of the fluorophores of interest. It is better to use our approach because of evidence of an in- 1. R. R. Alfano, D. Tata, J. Cordero, P. Tomashefsky, F. Longo, crease of tryptophan and a decrease of collagen in and M. Alfano, IEEE J. Quantum Electron 20, 1507 (1984). 9 cancerous tissue [4,5]. Since Δλ ˆ 40–50 nm for col- 2. R. R. Alfano and Y. Yang, IEEE J. Quantum Electron , 148 ss (2003). lagen and Δλ ˆ 70–80 nm for tryptophan, Δλi ˆ 40 nm ss 3. Y. Pu, W. B. Wang, G. C. Tang, and R. R. Alfano, should be chosen as an optimal scan wavelength interval, J. Biomed. Opt. 15, 047008 (2010). which gives balance between the resolution and the SNR. 11 3 4. H. J. G. Bloom and W. W. Richardson, Br. J. Cancer , 359 Furthermore, the S with Δλi ˆ 40 nm investigates the (1957). signal arisen from tryptophan and collagen in tissue, 5. G. Fenhalls, D. M. Dent, and M. I. Parker, Br. J. Cancer 81, which change inversely in cancerous and normal tissues. 1142 (1999).