1 INTRODUCTION

The differentiation of benign and malignant skin tumors has been studied for several years both in vitro and in vivo [1–7]. Cancer causes changes in the structure of tissue cells that can be detected using microspectroscopy. The most commonly used spectral post-analysis methods include standard principal component analysis (PCA) [8, 9], multivariate curve resolution analysis (MCR) [3, 5], partial least squares discriminant analysis (PLS/DA) [8–13] and others that allow the detection of basal cell carcinoma (BCC) and squamous cell carcinoma (SCC). Processes in tumors lead to an increase in the nuclear-cytoplasmic ratio (the ratio of the area of the nucleus and the cytoplasm), disruption of the chromatin structure, and changes in the content of lipids and proteins [14]. Using confocal Raman microspectroscopy, the authors of one of the studies [15] demonstrated that the contribution of nucleic acids, histones and proteins to the Raman spectra of tumor cell nuclei differs from that in normal epidermal cells. It has been established that the dermis near the tumor borders is deficient in collagen and has signs of structural changes.

In this work, we propose a differentiation method based on the analysis of the second derivative of Raman spectra obtained upon excitation at wavelengths of 532 and 785 nm [9, 16, 17]. The search for hidden peaks by calculating the second derivative is widely used in the analysis of infrared (IR)-Fourier transmission/reflection spectra, for example, to study processes occurring under the influence of certain antibacterial agents [18, 19]. However, it is used less frequently in Raman spectroscopy [20–26], and for the analysis of skin tumors—for the first time in our article.

Analysis of the second derivative of Raman spectra makes it possible to detect differences in the parameters of spectral bands, such as wavenumber, FWHM of the band, and area, which contain important information about the conformational order/disorder of the corresponding molecular bond and help to understand the fundamental changes that occur during tumor formation. Thus, the broadening of amide bands is often associated with the resulting disruption of the secondary structure of proteins, which can be caused by cell damage or mutation. This kind of analysis of spectral bands is most effectively carried out using the Lorentz approximation of spectra, where the difference between control (healthy skin) and pathology is more clearly visible in the spectra of the second derivative.

In this work, we provide a comparative analysis of BCC, PCC, papilloma and normal skin by calculating the second derivative spectra and determining the peak parameters (spectral position, bandwidth, area under the curve).

2 EXPERIMANTAL DETAILS

Samples of healthy skin and tumors were obtained by surgical removal. A preliminary diagnosis was made by a dermatologist-oncologist, and an accurate diagnosis after removal of the tumor was obtained using histological studies (the “gold standard” of diagnosis). Each tumor along with intact tissue was placed in saline. Raman spectra were recorded for seven samples of BCC, five SCC, three papillomas and five healthy skin samples, from 10 to 20 spectra at different points in each sample. A total of 21 samples of skin tumors from male and female patients, aged from 37 to 73 years, were studied. The samples corresponded to diagnosed cases of papilloma, carcinoma, skin cancer and healthy skin. All samples were stored under the same conditions (0.9% NaCl solution). More detailed information is given in Table S1 in the supplementary materials.

Raman spectra were obtained using a Renishaw inVia Basis spectrometer (inVia InSpect, Renishaw, London, UK) with excitation at 785 nm, energy up to 45 mW, acquisition time 10 s using a 50× objective (NPlan 50/0.50 Leica, Wetzlar, Germany), as well as Confotec MR520 spectrometer (SOL Instruments, Minsk, Belarus) at a laser wavelength of 532 nm, laser power 15 mW, accumulation time 2 s, using a 40× objective (MPlanFL, Nikon, Tokyo, Japan) with a numerical aperture of 0.75 and a working distance of 0.66 mm.

Background fluorescence signal was removed from the recorded spectra using the Vancouver algorithm, which is an improved multi-polymodal baseline removal method optimized for fluorescence of biomedical samples (degree 5 polynomial was used) [27]. The resulting spectra were smoothed and normalized to the most intense Raman bands in the range 1200–1800 cm–1 [16].

The averaged processed spectra for each sample were approximated using the Lorentzian fit using OriginPro software (OriginPro 2019b 9.6.5.169) according to the established method [18, 19]. The fitting procedure consists in approximating the shape of the peaks in the processed Raman spectra to the shape of the Lorentz contour:

$$y = y0 + \frac{{2A}}{\pi }\frac{w}{{4(x - {{x}_{c}}{{)}^{2}} + {{w}^{2}}}},$$
(1)

where \(H = 2A{\text{/}}(\pi w)\) is the peak height, \(A\) is the area under the curve, w is the bandwidth (FWHM), \({{x}_{c}}\) is the maximum position (central wavenumber in inverse centimeters), \({{y}_{c}}\) and \({{y}_{0}}\) are the maximal and background signals, correspondingly (Fig. 1).

Fig. 1.
figure 1

(Color online) Schematic representation of the Lorentz contour, according to which the Raman spectra are approximated.

Full size image

Detection of Raman bands (hidden peaks) was carried out by determining local maxima of the second derivative with Savitsky–Golay smoothing. The found values (positions of the maxima of the second derivative) were fixed, after which all peaks were approximated by the Lorentzian fit (1), as a result of which the values of the bandwidth and area under the curve were obtained. The distribution of these values (standard deviation) was estimated by the program in accordance with the noise level estimate and the magnitude of the residual error.

3 RESULTS AND DISCUSSION

3.1 Second Derivatives of Raman Spectra

Averaged Raman spectra for normal skin, BCC, SCC, and papillomas obtained upon excitation at wavelengths of 532 and 785 nm (hereinafter mentioned in the text as \({{\lambda }_{{532}}}\) and \({{\lambda }_{{785}}}\)), and their second derivatives are presented in Fig. 2.

Fig. 2.
figure 2

(Color online) Averaged Raman spectra of healthy skin (normal skin, green line) and neoplasms (basal cell carcinoma (BCC), blue line); squamous cell carcinoma (SCC), purple line; papilloma (orange line), obtained at two excitation wavelengths: 532 nm (a) and 785 nm (b). The gray penumbra indicates the standard deviation. The corresponding spectra of the second de-rivative, obtained at two excitation wavelengths: 532 nm (c, e) and 785 nm (d, f), are presented in two regions: (c, d) from 900 to 1330 cm–1 and (e, f) from 1330 to 1800 cm–1.

Full size image

The parameters of the spectral bands were obtained by the Lorentz approximation method described in Section 2. These values are displayed in Tables S2, S3, where each cell includes up to three rows corresponding to the position of the maximum, the bandwidth (FWHM) and the area under the curves with standard errors (Table S2, interpretation of spectral bands is given according to data from [9, 28–35]; Table S1 in supplementary materials).

3.1.1. DNA/RNA. The band at 1085 cm–1, assigned to PO2 vibrations in nucleic acids [28, 34, 35], shifts towards higher wavenumbers in the spectra of BCC and papilloma at \({{\lambda }_{{785}}}\), and for all tumors—to lower wavenumbers at \({{\lambda }_{{532}}}\) (Fig. 2 and Table S2). At \({{\lambda }_{{532}}}\) the area under the curve decreases significantly in the spectra of SCC, and the bandwidth increases in the spectra of neoplasms, decreasing for all tumors at \({{\lambda }_{{785}}}\). A drop in intensity and a shift towards lower wavenumbers has been reported previously [36], and such deviation can be caused by single- and double-strand DNA breaks [37]. Another PO2-related band appears in all spectra of tumors at both laser excitation wavelengths in the range  ~1100–1104 cm–1 [38]. This band results from excessive DNA replication during tumor growth [31]. Band at 1175 cm–1, related to cytosine (C) and guanine (G) [39], shifts towards higher wavenumbers for BCC and papilloma and vice versa for SCC (Fig. 2 and Table S3). For \({{\lambda }_{{785}}}\) it shifts to higher wavenumbers in all tumor spectra.

The area for the DNA/RNA band centered at 1344 cm–1 [40] increases in the spectra of SCC and papilloma at both excitation wavelengths (Fig. 2 and Table S2). The broadening of the bands at \({{\lambda }_{{785}}}\) may result from the destruction of nucleic acids [28, 31, 37]. In the spectra of papilloma, the band area increases significantly, which can be explained by more intensive DNA replication in malignant neoplasms [28, 37].

Ring symmetric (breathing) modes in adenine (A), G are displayed in the spectra as a band at 1490 cm–1 [41–43]. This band was detected only in the spectra of SCC and papilloma at \({{\lambda }_{{532}}}\) and in all samples except papilloma at \({{\lambda }_{{785}}}\) (Fig. 1, Table S2).

3.1.2. Cell membrane damage. The band centered at 1031 cm–1, corresponding to molecular vibrations of phospholipids, narrows in SCC and broadens in papilloma at both excitation wavelengths; its intensity increases in the spectra of papilloma at \({{\lambda }_{{532}}}\) (Fig. 2 and Table S2) and in the spectra of SCC at \({{\lambda }_{{785}}}\). Stretching vibrations of the C-C chain of cell wall lipids lead to the appearance of a peak at 1063 cm–1 [28, 44], which broadens in the BCC spectra and its area increases. The peak is absent in SCC and papilloma. At \({{\lambda }_{{785}}}\) the peak increases, and its area grows in all tumor’s spectra. The 1127 cm–1 band, which can be attributed to lipids, in the spectra of neoplasms narrows at \({{\lambda }_{{532}}}\), although the corresponding area increases. At \({{\lambda }_{{785}}}\) there is no significant change in bandwidth or area.

3.1.3. Lipids. The band at 1301 cm–1, attributed to CH2 in saturated lipids, is sensitive to disruption of conformational order in cells [28–30, 45]. Its position shifts towards higher frequencies in the spectra of SCC and papilloma at \({{\lambda }_{{532}}}\) and in all tumors’ spectra—at \({{\lambda }_{{785}}}\) (Fig. 2, Table S2). The band broadens in BCC and SCC, its area increases at both laser excitation wavelengths. A shift towards higher wavenumbers indicates deformation of the molecular structure of lipids (higher content of gauche conformers) [28, 39, 45]. The band at 1398 cm–1, attributed to CH3 oscillations, in the spectra of skin tumors shifts towards lower wave numbers (Fig. 2, Table S2), and its intensity increases significantly in spectra of SCC and papilloma (Figs. 2a, 2b) at \({{\lambda }_{{532}}}\), while at \({{\lambda }_{{785}}}\) this band is present only for SCC and papilloma (Figs. 2c, 2d). CH2 oscillations are illustrated by peak at 1438 cm–1 [46], which broadens in all spectra of tumors at both laser excitation wavelengths. Its area decreases for BCC and SCC and increases for papilloma at \({{\lambda }_{{532}}}\), increases in BCC and stays unchanged in SCC and papilloma at \({{\lambda }_{{785}}}\). The similar decrease of the area under the curve was reported in earlier work [36]. The peak demonstrates a significant shift to the higher wavenumbers in the malignant skin spectra for both laser excitation wavelengths, with the variation being more pronounced for SCC and papilloma, which may be caused by stronger damage of the lipids’ molecular structure.

Another band at 1416 cm–1 was attributed to molecular oscillations in ceramide III (Fig. 2, Table S2). Ceramides are also present in the cell membrane, so the observed band shift towards higher wavenumbers for all skin neoplasms at both excitation wavelengths may be due to disruption of the molecular structure of lipids in skin malignancies [28, 31]. Its intensity increases in BCC and SCC at \({{\lambda }_{{532}}}\) and decreases for all tumors at \({{\lambda }_{{785}}}\) according to the second derivative spectra (Fig. 2).

The band at 1459 cm–1 is attributed to proteins and lipids both [28, 31], and its area decreases in SCC and papilloma (more pronounced in SCC), and increases in BCC for both laser excitation wavelengths (Fig. 2, Table S2).

Interestingly, initially single band at 1744 cm–1 in spectra of normal skin, attributed to ester oscillations in lipids [47], was split into two weak bands at \({{\lambda }_{{532}}}\) (Fig. 2, Table S2). Such trend was not observed for \({{\lambda }_{{785}}}\), where the band at 1745 cm–1 shifted to higher wavenumbers in SCC and to lower wavenumbers in papilloma, broadening in all spectra of tumors. This might be caused by the high noise level, making the changes insignificant.

3.1.4. Proteins. The band at 937 cm–1 was attributed to collagen [29, 39, 46], and in the malignant skin spectra it demonstrates broadening and area increase at \({{\lambda }_{{532}}}\), narrowing and area decrease—at \({{\lambda }_{{785}}}\) (Fig. 2, Table S2). The band at 971 cm–1, also attributed to proteins [39], in tumors’ spectra splits into two bands, almost disappearing in papillomas at \({{\lambda }_{{532}}}\). At \({{\lambda }_{{785}}}\) in malignant skin spectra only one of two bands is seen. Bands at 961 and 983 cm–1 illustrate molecular vibrations of \(\alpha \)-helices and \(\beta \)-sheets in proteins, correspondingly. Such splitting may be caused by the possible changes in keratinocytes, arising from the atypical nuclei (enlarged, asymmetrical, hyperchromic) in SCC [48]. The secondary structure of histones mainly consists of α-helices, while fibrin mainly consists of \(\beta \)-sheets, therefore the observed increase in the 983 cm–1 band intensity is caused by fibrin, surrounding the cancer cells and forming a protective layer from the immune system cells. Such trend was observed earlier in a number of works [49–52].

The band at 1004 cm–1 corresponds to phenylalanine (Phe) vibrations [28, 34], and it narrows in the tumors’ spectra at \({{\lambda }_{{532}}}\) and broadens at \({{\lambda }_{{785}}}\) (Fig. 2, Table S2). Supposedly, Phe is also present in spectra as bands at 1582 and 1605 cm–1 [53]. The 1582 cm–1 band width and area increase significantly SCC and papilloma (Fig. 2, Table S2), decreasing in BCC at \({{\lambda }_{{532}}}\), and vice versa—at \({{\lambda }_{{785}}}\). The band at 1605 cm–1 [54] narrows significantly in SCC and papilloma at \({{\lambda }_{{532}}}\), and broadens at \({{\lambda }_{{785}}}\). The corresponding area decreases in SCC spectra at \({{\lambda }_{{532}}}\) and stays unchanged at \({{\lambda }_{{785}}}\); still, one must take into account the contribution of tyrosine and cytosine in the band intensity. The strong area increase of the band at 1605 cm–1 in tumors was reported earlier [55]. The band area in BCC and papilloma increases for both laser excitation wavelengths (Fig. 2, Table S2). The band at 1609–1615 cm–1 is related to tyrosine; in SCC it shifts to the lower wavenumbers at \({{\lambda }_{{532}}}\), and to higher wavenumbers in all malignant skin at \({{\lambda }_{{785}}}\). It becomes more pronounced in tumors at both laser excitation wavelengths, which may result to its narrowing. According to the second derivative spectra, its intensity increases in malignant skin spectra at \({{\lambda }_{{532}}}\), area decreasing at \({{\lambda }_{{785}}}\). The band at 1204 cm–1, related to proteins, in the tumors’ spectra shifts to higher wavenumbers at \({{\lambda }_{{532}}}\); its corresponding area increases for both laser excitation wavelengths (Fig. 2).

The band at 1537 cm–1 corresponds to amide II [56]. Its intensity at \({{\lambda }_{{532}}}\) demonstrates a slight decrease in SCC, and it disappears completely in papilloma (Fig. 2); the band is not detected in spectra, acquired at \({{\lambda }_{{785}}}\).

Amide III (\(\beta \)-sheets) -related band in normal skin is located at 1264 cm–1. According to literature, its transition to disordered state provides another band at 1245 cm–1 [10, 28], and such trend was observed for all malignant skin samples (Fig. 2, Table S2). The band at 1264 cm–1 in the spectra of tumors has a lower intensity and width, than in normal skin (Table S2). The band shift to higher wavenumbers was reported earlier [57] and was attributed to the arising destruction of the proteins’ secondary structure. Another peak at 1252 cm–1 appears BCC for both laser excitation wavelengths, whereas in papilloma this peak is only seen at \({{\lambda }_{{785}}}\). This peak may also illustrate the disordered state of the proteins’ secondary structure. The band at 1264 cm–1 was earlier observed only in healthy cells and was absent in malignant skin tumors [55]. In our work, the pronounced peak at 1264 cm–1 becomes weaker in BCC, SCC and papilloma, and the 1245 cm–1 peak area increases. Such variation in the proteins’ secondary structure indicates the arising changes in the skin cells’ structure in the tumor, such as an increase in fibrin content [55, 57].

Amide I band at 1654 cm–1 shifts to lower wavenumbers in BCC and SCC and to higher wavenumbers—in papilloms; its bandwidth in tumors increases for both laser excitation wavelengths. Such trend was reported earlier in a number of [28, 36]. The peak area decreases in BCC and increases in SCC and papilloma. The band intensity decrease in BCC was reported earlier [58]. Such changes may be attributed to the fact, that histones’ secondary structure in normal skin mainly consists of \(\alpha \)-helices, and their content in cancer cells is lowered, or the nucleic acids’ structure is damaged [55]. Another amide I-related band at 1690 cm–1 is detected only at \({{\lambda }_{{785}}}\), and in SCC and papilloma its bandwidth and area increase.

C=C, N–H and C–N vibrations of amide II correspond to the band 1553 cm–1 [59], which demonstrates an increase in bandwidth and area in SCC and papilloma at \({{\lambda }_{{532}}}\) (Fig. 2, Table S2). The band narrows in SCC and papilloma at \({{\lambda }_{{785}}}\). This peak may be also attributed to tryptophan, and its increase was reported earlier in BCC and SCC [60].

The band at 1513 cm–1 was attributed to carotenoids [32, 33]. Its intensity increases in spectra of papilloma and decreases in SCC, staying unchanged in BCC. Despite the fact, that some works report the higher content of carotenoids in BCC than in normal skin [39], there was no such trend in our work. Also, the carotenoids’ content may vary depending on the individual patients’ characteristics and their lifestyle (smoking, chronic fatigue, diseases etc.) [42].

Formation of BCC is often accompanied by the nucleus shape change to oval or elongated with condensed chromatin [58, 61]. The densification of the chromatin structure, which consists of DNA and proteins, leads to the corresponding Raman bands’ narrowing. According to the second derivative spectra, the narrowing is observed in the bands at 1030, 1615, and 1655 cm–1, which correspond to proteins and DNA (true for both used laser excitation wavelengths).

In BCC and SCC, mutations of protein p53 mainly consist of C \( \to \) T and CC \( \to \) TT transitions, and are triggered by ultraviolet (UV) radiation [62]. Such transition may lead to decrease in C concentration. Area of the band at 1260 cm–1, which illustrates the С content, decreases in all spectra of tumors: BCC, SCC and papilloma. According to the literature [63], Raman bands, attributed to Т, are located at 936 cm–1 (this band increases in BCC, SCC); 1063 cm–1 (increases in BCC and SCC); 1104, 1174, 1489, 1550 cm–1 (increase in SCC); 1232–1239, 1340 cm–1 (increase in all tumors).

4 CONCLUSIONS

In this work, we differentiated skin tumors (basal cell carcinoma, squamous cell carcinoma, papilloma) and healthy skin by analyzing the Raman bands’ parameters based on the second derivative and the Lorentzian approximation of the spectra. It has been established that skin tumors can induce deformation of the secondary structure of proteins, resulting in a band at 972 cm–1 splitting into two peaks or shifting to lower wavenumbers. Another band, attributed to \(\alpha \)-helices in proteins (1655 cm–1), shows an increase in width and area in SCC and papilloma and a decrease in BCC. In addition, a band corresponding to the disordered state of amide III appears in tumors. Similar change in content of \(\alpha \)-helices can be explained by disruption of their molecular structure in tumors. At the same time, in tumors the bands which illustrate the lipid content (1061, 1127, 1297, 1439, 1745 cm–1), demonstrate broadening and/or area increase. Another feature associated with lipids is the splitting of the band at 1744 cm–1 into two peaks upon excitation at a wavelength of 532 nm and its significant broadening at 785 nm. The disruption of lipid structure, also indicated in several bands as a “red” shift in wavenumbers, along with an increase in their content, is possibly associated with increased fluidity of the cell membrane.

The study of the spectra of papillomas is preliminary and may not be entirely objective due to the small sample size. The main goal of the work was to test the method of analyzing Raman spectra using the Lorentzian approximation. The results of using this method to analyze the spectra of skin tumors showed its promise for identifying characteristic biomarkers of malignant tumors and ongoing changes in cell structure, which can be the basis for differentiation in early diagnosis.