2.1.

Experimental Setup

Near-infrared Raman spectroscopy has been described previously.¹⁵ In brief, a wavelength-stabilized diode laser (HL-7851G, Hitachi, Japan) produced a laser beam (at a wavelength of 780 nm and a power of 20 mW), which was circularized by a pair of anamorphic prisms, spatially filtered, and then directed into an inverted differential interference contrast microscope (Nikon TE2000-U). This microscope was equipped with an objective ($100\times$, numerical aperture of 1.30) to form an excitation source and a single-beam optical trap. The beam was focused through a quartz sample chamber with a thickness of 0.1 mm. For these experiments, we equipped a Focht Chamber System 2 (FCS2, Bioptechs), which is a closed-chamber system for live-cell observation, with a Raman system to ensure a constant temperature as well as perfusion of the culture medium and carbon dioxide which are beneficial for cell survival. Cells were grown in the FCS2 system and were probed by the laser. Raman-scattered light from the cells was collected by the objective, focused onto the entrance slit of an imaging spectrograph (Acton Research Co., SpectraPro 2300i, $600\text{lines}/\mathrm{mm}$ blazed at 650 nm) and recorded using a charge-coupled device (CCD, Roper Scientific Spec-10, Princeton Instruments). The resolution of the spectrograph was $6{\mathrm{cm}}^{-1}$. A suspension of polystyrene beads ($2\mu \mathrm{m}$, Bands Laboratories) was used to calibrate the relative Raman shifts prior to each measurement.

2.2.

Pancreatic Islet Isolation and Single-Cell Acquisition

Sprague–Dawley rats (8 to 10 weeks old, body weight 250 to 300 g, provided by the Guangxi Medical University Laboratory Animal Center) were sacrificed, and collagenase (Sigma) was injected into the ductal system for tissue digestion.¹⁶ The tissue was dispersed under shaking after collagenase digestion. The islets were isolated through density-gradient centrifugation using Histopaque-1077 (Sigma). The islets were dispersed into single cells via trypsin (Sigma) digestion. The cells were plated and grown in culture dishes containing 40-mm-diameter coverslips overnight in Roswell Park Memorial Institute 1640 (RPMI 1640) culture medium (Invitrogen) supplemented with 10% ($\mathrm{vol}/\mathrm{vol}$) fetal calf serum, $100\mathrm{IU}/\mathrm{ml}$ penicillin, $0.1\mathrm{mg}/\mathrm{ml}$ streptomycin, and $5.6\mathrm{mmol}/\mathrm{l}$ glucose, under environmental conditions of 37°C and 5% ${\mathrm{CO}}_2$.

INS-1 insulinoma cells (from the Conservation Genetics Chinese Academy of Sciences Kunming Cell Bank, China, Kunming) were cultured in RPMI 1640 with $2.05\mathrm{mmol}/\mathrm{l}$ l-glutamine (Invitrogen) supplemented with 10% fetal calf serum, $100\mathrm{IU}/\mathrm{ml}$ penicillin, $0.1\mathrm{mg}/\mathrm{ml}$ streptomycin, $2\mathrm{mmol}/\mathrm{l}$ l-glutamine, $1\mathrm{mmol}/\mathrm{l}$ sodium pyruvate, and $50\mu \mathrm{mol}/\mathrm{l}$ b-mercaptoethanol (Sigma). Sterile conditions were maintained for all operations involving islet isolation and cell culture. All cells were attachment cultured in the FCS2 system.

2.3.

Collection of Raman Spectra

Each circular coverslip (40 mm in diameter, Bioptechs) on which adherent cells were grown was transferred to the FCS2 system, which was mounted on a Nikon microscope equipped with a CCD camera and the Raman spectroscopy setup. After an image of the cells was acquired, the laser spot was repositioned onto three to four different regions within a cell to acquire a Raman signal for each of these cell regions to obtain the approximate overall information of the cell. The background spectra of the culture medium were acquired over the same scan area but in the absence of cells. The integration time (measurement time) for each cell was set to 30 s.

2.4.

Identification of Primary Pancreatic $\beta$ Cells

Primary pancreatic islets have been classically described as containing four different types of cells: insulin-producing $\beta$ cells, glucagon-producing $\alpha$ cells, somatostatin-producing $\delta$ cells, and pancreatic-polypeptide-producing PP cells. To acquire the spectra of pure primary rat pancreatic $\beta$ cells as contrast samples, we first detected the Raman spectra of the primary islet cells, and a series of photographs of each cell was acquired using a microscope. The relative positions of all detected cells were recorded. After Raman detection was complete, we fixed the cells with 4% ($\mathrm{w}/\mathrm{v}$) paraformaldehyde. Through staining via immunohistochemistry techniques (using an insulin antibody, as described below), the insulin-positive cells were determined to be $\beta$ cells. Based on the relative positions of the positive cells, the $\beta$ cells, whose Raman data had been previously obtained, were confirmed. The immunohistochemistry-negative cells were thus considered to be non-$\beta$ cells, and their data were discarded.

2.5.

Immunohistochemistry Staining

Slices with adherent cells were dried and fixed using paraformaldehyde and then washed three times with phosphate-buffered saline (PBS). The cell sections were then placed in 3% hydrogen peroxide (${\mathrm{H}}_2{\mathrm{O}}_2$) for 10 min and washed three times with PBS. After being soaked in a permeabilization buffer (containing 0.4% Triton X-100) for 10 min and rinsed in PBS, all sections were incubated in 10% normal goat serum for 10 min and then incubated overnight in rabbit anti-insulin (H-86, diluted 1:200 in PBS, pH 7.4, 1% BSA, Santa Cruz Biotechnology, Inc.) in humidified chambers at 4°C. These sections were rinsed in PBS and incubated in biotinylated goat anti-rabbit IgG (Beijing Zhongshan Biotechnology Co., Ltd., China) diluted in PBS (pH 7.4, 1% BSA) in humidified chambers for 10 min at 37°C. The sections were rinsed again and incubated in enzyme tag chain mildew avidin horseradish (Beijing Zhongshan Biotechnology Co., Ltd., China) for 10 min in humidified chambers at 37°C. The sections were then rinsed again in PBS. The immunoreactivity was visualized by incubating the sections for 5 min in the chromogen 3,3-diaminobenzidine hydrochloride in the presence of 0.01% hydrogen peroxide. The sections were rinsed several times in tap water, and the dyed sections were then rinsed with hematoxylin for ~3 or 4 s, dehydrated through a graded series of ethanol solutions, and soaked in xylene. The slices were then dried at room temperature and sealed. Finally, the slices were observed under an optical microscope (Olympus BX 40, Japan); in the resulting images, positive particles appeared brown and negative particles appeared blue.

2.6.

Statistical Analysis

Raman spectra were collected within the spectral region from 510 to $1800{\mathrm{cm}}^{-1}$. This region corresponds to a molecular fingerprint region that contains most relevant biological fingerprints, including those of disulfide bonds which are present in abundance in insulin. The cellular spectra of the different cell regions were averaged; other processes, including the subtraction of background spectra, intensity calibration, five-point smoothing, normalization and baseline calibration, were performed using Micro Origin 8.1 software. The Raman peak intensities (relative counts) were used as the measurement data and were processed using the SPSS 13.0 software package for Windows. Multivariate analysis of variance (MANOVA) was used to analyze the difference between the two cell types. Student’s $t$-test was used to test the difference in each peak.⁴ PCA was applied to all spectra to extract the major spectral features for individual cells. Subsequently, Fisher LDA, which is a standard linear technique for discriminant analysis, was applied to establish the separation between the rat pancreatic $\beta$ cells and the INS-1 insulinoma cells.^17,18 $P$ values of $<0.05$ were considered to be significant.

3.1.

Raman Spectra of Pancreatic $\beta$ Cells and Insulinoma Cells

The Raman spectra of samples containing pancreatic $\beta$ cells and insulinoma cells were acquired. The mean Raman spectrum of an insulinoma cell and that of a single pancreatic $\beta$ cell are shown in Fig. 1(a). Figure 1(b) shows a single insulinoma cell under Raman spectroscopy, whereas Figs. 1(c) and 1(d) show three islet cells and their immunohistochemical staining, respectively, in the same optical section. A series of Raman peaks can be observed, including the vibration of the DNA backbone ($785{\mathrm{cm}}^{-1}$); phenylalanine ($1002{\mathrm{cm}}^{-1}$), $\mathrm{C}─\mathrm{C}$ and $\mathrm{C}─\mathrm{H}$ vibration modes associated with proteins and lipids ($1445{\mathrm{cm}}^{-1}$); and protein amide I-band vibrations ($1655{\mathrm{cm}}^{-1}$). The Raman spectrum of an individual cell includes contributions from all of its cellular components (nucleic acids, proteins, and lipids).

Fig. 1

(a) Comparison of the mean spectra for a pancreatic $\beta$ cell (black line, $n=60$) and an insulinoma cell (red line, $n=77$); (b) insulinoma cell with the effect of differential interference contrast under the microscope; (c) three islet cells under the microscope; (d) immunohistochemical staining of the same three islet cells, only the middle of which appears immunohistochemistry-positive brown, indicating that it is a pancreatic $\beta$ cell. The scale bars represent $10\mu \mathrm{m}$.

The MANOVA results indicate an overall significant difference between the two cell types ($P<0.05$). With the spectra represented in the same coordinate system, as shown in Fig. 1(a), each main distinct Raman peak was employed for statistical testing using Student’s $t$-test. The peaks at 562, 640, 725, 785, 852, 893, 937, 1003, 1053, 1175, 1320, 1337, 1605, 1615, and $1660{\mathrm{cm}}^{-1}$ exhibited significantly different mean intensities, as determined via $t$-test analysis ($P<0.05$). The normalized intensities of the Raman peaks at 562, 640, 725, 852, 893, 937, 1002, 1175, 1605, 1615, and $1660{\mathrm{cm}}^{-1}$ were more intense for the pancreatic $\beta$ cells than for the insulinoma cells, whereas the Raman bands at 785, 1053, 1320, and $1337{\mathrm{cm}}^{-1}$ exhibited higher intensities in the insulinoma cell samples. The specific assignments of individual peaks can be found in Table 1, and Fig. 2 shows the means and standard deviations of the significantly different peaks.

Table 1

Peak assignments for the Raman shifts of a pancreatic β cell.

Fig. 2

The mean intensity values of significant peaks and the associated standard deviations for pancreatic $\beta$ cells (gray) and insulinoma cells (white).

3.2.

PC and LDAs of Raman Spectra

Individual Raman spectra were analyzed via PCA to compress redundant spectral information and differentiate insulinomas from pancreatic $\beta$ cells. Each spectral data set was imported into the SPSS software package for PCA and LDA. The PCA technique involves calculating the principal components (PCs) that describe the greatest proportion of the variance in the spectral data from its mean. Based on the PCA eigenvalues and the results of the PCA scree tests, 43 PCs were retained and calculated for LDA.

A three-dimensional (3-D) plot of the first principal component (PC1) versus the second principal component (PC2) and the third principal component (PC3) of the Raman spectra for the pancreatic $\beta$ and insulinoma cells is shown in Fig. 3. PC1, PC2, and PC3 are responsible for 32.1%, 16.9%, and 9.6% of the variance, respectively. From Fig. 3, we can see that the two cell groups can be readily separated, thereby indicating inherent molecular differences between the cells, as a natural separation of the data is visible without supervised statistical manipulation.

Fig. 3

A three-dimensional mapping of the PCA results for pancreatic $\beta$ cells (white hollow balls) and insulinoma cells (black solid balls). PC1: the first principal component; PC2: the second principal component; and PC3: the third principal component.

The 43 PCs obtained from the PCA calculation were used as multivariate data for classification. Using the SPSS software package, the discriminant function was established via LDA and back-substitution tests of all cell spectra. The results are presented in Tables 2 and 3. Table 2 presents the classification results for the original cell data for which the discriminant algorithm was established based on all spectra; using this algorithm, each cell can be classified as a certain cell type. In the cross-validation, the discriminant algorithm was established using “leave one spectrum out” cross-validation. This involved removing one spectrum from the data set and constructing a diagnostic algorithm from the remaining spectra. The algorithm then predicted the cell type of the “left out” spectrum and stored the result. This process was repeated, with each spectrum left out in turn, until the cell type had been predicted for each of the spectra. The overall predictive accuracy of the algorithm was calculated and expressed in terms of sensitivity and specificity for each cell type. As shown in Table 2, there were 60 actual pancreatic $\beta$ cells in total, of which 55 cells were correctly classified, whereas of the 77 total insulinoma cells, 76 cells were correctly classified; thus, the total accuracy was 95.6%. Similarly, Table 3 shows that 92.7% of the cross-validated grouped cases were correctly classified.

Table 2

Classification results for the original cell data.

Table 3

Classification results for the cross-validation test.