2.1.

Differentiation of Mesenchymal Stem Cells into Islet-Like Cells

Umbilical cords were provided by healthy newborns (approved by Guangdong provincial maternity and child care center), and MSCs were isolated as described previously.²⁵ After removing the cord vessels, umbilical cords were minced into $1\text{-}{\mathrm{mm}}^3$ fragments with sharp scissors under aseptic conditions. The fragments were washed with phosphate buffer saline (PBS) to remove contamination, and then treated with 0.2% collagenase II (Sigma) at 37°C for 2.5 h in a table concentrator. The digested mixture was first passed through a $100\text{-}\mu \mathrm{m}$ filter and then a $200\text{-}\mu \mathrm{m}$ filter to obtain cell suspensions. The dissociated cells were then washed with PBS, planted on uncoated T-25 culture flasks, and cultured in Dulbecco’s modified Eagle’s medium/Nutrient F12 Ham’s 1:1 mixture (DMEM/F12 Gibco) and 10% fetal bovine serum (FBS, Gibco Australia). Finally, after 3 days of culture, the medium was replaced and nonadherent cells were removed. While the cells were at 90% confluence, they were passaged with 0.25% trypsin.

As shown in Fig. 1(a), many MSCs with a long spindle shape were observed in the proliferation medium, which was composed of DMEM/F12 supplemented with 10% FBS and 1% penicillin and streptomycin. Next, MSCs were cultured at a density of $3\times {10}^5\text{cells}/\text{well}$ in a 6-well plate for 5 days with pancreatic differentiation medium,⁵ composed of DMEM/F12, 10 mM nicotinamide, 10 nM exendin-4, 10 nM pentagastrin, 100 pM hepatocyte growth factor, 2% B-27, and 2% N-2. As shown in Fig. 1(b), some floating cells with roughly spherical shapes were observed. While MSCs were cultured for 5 days, many floating cells with spherical shapes were observed [Fig. 1(c)]. Moreover, using a dithizone staining approach [Fig. 1(d)], these floating cells were shown to be islet-like cells, a type of pancreatic $\beta$ cell.⁵ MSCs were cultured in proliferation medium as the control group.

Fig.1

Bright-field images of mesenchymal stem cells (MSCs) differentiating into islet-like cells. (a) Many MSCs (white circle) with long spindle shapes were observed before differentiation. (b) Some floating cells appeared on the fourth day of differentiation (white circle). (c) Some islet-like cells appeared on the fifth day of differentiation (white circle). (d) Fluorescence image of islet-like cells stained by dithizone (red circle).

2.2.

Raman Spectroscopy and Data Preprocessing

A Renishaw inVia Raman spectrometer attached to a Leica upright microscope equipped with a $50\times$ objective, a 785 nm laser with a laser power of 50 mw, and a laser spot size of $2\mu \mathrm{m}$, and a CCD detector with a spectral resolution of $0.99{\mathrm{cm}}^{-1}$ was utilized to collect Raman spectra of individual cells for all cell types, in which five different areas were randomly probed within a cell and the integration time for acquiring a Raman spectrum of a cell was 10 s in all experiments. A Raman spectrum of each cell type was expressed by an average of 10 cells in the scan range of 600 to $1800{\mathrm{cm}}^{-1}$. Raman spectroscopic data processing was performed with MATLAB and Origin software, and the preprocessing procedure of the Raman spectrum was described in Fig. 2.

Fig. 2

The preprocessing procedure of Raman spectrum.

In this study, all spectra were normalized to the peak at $1655{\mathrm{cm}}^{-1}$, which is predominantly related to protein and lipid, to perform a direct comparison of the data. In the reported BCA fitting,²⁷ seven basic components—actin, albumin, triolein, phosphatidylcholine, DNA, RNA, and glycogen—were employed as the major organic molecules of biological cells, in which actin and albumin were attributed to proteins; triolein and phosphatidylcholine were attributed to lipids; glycogen was attributed to polysaccharides; and DNA and RNA were attributed to nucleic acids. The procedure of the reported BCA fitting was described in Fig. 3.

Fig. 3

The procedure of the reported biochemical component analysis (BCA) fitting.

As shown in Fig. 4, from the raw Raman spectrum of MSCs or islet-like cells, it was found that the signal-to-noise ratio (SNR) for several main peaks at 1004, 1448, $1655{\mathrm{cm}}^{-1}$ was more than 50:1, and even in these low-intensity peaks, such as $665{\mathrm{cm}}^{-1}$, the SNR was more than 10:1, indicating our Raman spectroscopic data are suitable for the quantitative analysis. In this study, we also chose the above seven biochemical components to perform the BCA operation, and the corresponding fitting results of MSCs and islet-like cells as in Fig. 4(a). It was shown that the correlation coefficient between the fitting spectrum and the raw spectrum was less than 90%, showing that the BCA fitting accuracy with the reported direct background subtraction approach is relatively low. One main reason is that the fluorescence background of biological samples is varied and unstable. To solve this problem, we employed three types of approximate fluorescence backgrounds, which were expressed either by the fifth-order polynomial fitting, the fourth-order polynomial fitting, or the penalized least-squares fitting.²⁸ As shown in Fig. 4(b), the correlation coefficient between the fitting Raman spectrum and the raw Raman spectrum was more than 97%, indicating a greatly improved fitting accuracy with the proposed BCA operation. The ratio of the fitting coefficient of each biochemical component to the sum of all fitting coefficients (excluding the coefficients from the three fluorescence backgrounds) was also obtained. Then the relative contribution of each basic biochemical component to the cell spectrum was determined (Fig. 4). Similarly, the final fitting fluorescence background could be obtained by the summation of each fluorescence background multiplied by its fitting coefficient. Taking the MSCs spectrum as an example, the fitting fluorescence background is shown in Fig. 5 (green line). After removing the fluorescence background, we obtained the final fitting Raman spectrum of MSCs (Fig. 5, red line). For the sake of comparison, the final fitting Raman spectra of MSCs and islet-like cells are presented in Fig. 6.

Fig.4

BCA fitting results with two different approaches. (a) The poorness of the BCA fitting result in which the background was directly subtracted from the raw Raman spectrum. (b) The goodness of the BCA fitting result in which three approximate fluorescence backgrounds were used as BCA fitting factors. (c) Spectrum contribution of each biochemical component with the improved BCA approach. Asterisk indicates a significance level of $P<0.05$.

Fig. 5

Fitting fluorescence background obtained by the improved BCA operation (green line) and the final Raman spectrum of MSCs (red line).

Fig. 6

Final Raman spectra of MSCs (red line) and islet-like cells (black line) obtained by the improved BCA approach.

3.1.

Raman Data Analysis with Biochemical Component Analysis and Multiple Peaks Fitting

Using the improved BCA approach, we first obtained the quantitative biochemical changes before and after MSC differentiation [Fig. 4(c)], in which the error bar represented the standard deviation (SD) of the fitting coefficients measured in 10 cells. A Wilcoxon signed-rank test showed differences in all components between these two differentiated modes ($P<0.05$). In addition, to verify the reliability of the above BCA results, we also measured the relative content changes of biochemical constituents between MSCs and islet-like cells by using five-kind conventional biochemical methods, as shown in Table 1. As is known, it was very difficult to obtain the percentages of each biochemical constituent or the relative content changes of all biochemical constituents by a one-kind biochemical examination method. Therefore, the measured value, representing the content of each biochemical constituent, was given as the relative value of control. Data were expressed as $\text{mean}\pm \mathrm{SD}$ from three independent experiments, with significant differences from control designated as *$P<0.05$ and **$P<0.01$. Cellular actin content was obtained by the Western blot analysis. Total protein was extracted and electrophoresed using sodium dodecyl sulfate polyacrylamide gel electropheresis (SDS-PAGE) and polyvinylidene fluoride (PVDF) membranes (Merck-Millipore). The chemiluminescence signal was imaged using ChemiDoc XRS (Bio-Rad) and quantified using Image J. Cellular albumin content was measured by the bromocresol green assay kit (Sigma–Aldrich, Germany) according to the manufacturer’s instructions. Cellular phosphatidylcholine and glycogen contents were, respectively, assessed using the commercially available kits (BioVision) according to manufacturer’s instructions. The total RNA of MSCs or islet-like cells was extracted for cDNA synthesis; the generated cDNAs were analyzed by a quantitative real-time polymerase chain reaction (TRIzol reagent, Invitrogen). RNA purity and concentration were determined using a spectrophotometer (Ultrospec 2100 pro UV/Visible, Amersham Bioscience, Germany). Cellular triolein content was detected by the above spectrophotometer, in which the purified triolein was obtained from Perimed Crop. (Beijing, China). Cellular DNA content was measured by DNA-image cytometry, in which MSCs or islet-like cells and the controls were stained using the Feulgen reaction and assessed for integrated optical density by image analysis, and the ploidy status was assessed.

Table 1

Relative content changes of biochemical constituents between mesenchymal stem cells (MSCs) and islet-like cells.

From Fig. 4 and Table 1, we can see that the relative content changes of biochemical constituents between MSCs and islet-like cells with the proposed BCA method were closely consistent with the conventional biochemical methods, indicating that the employed seven basic biochemical components were suitable for the BCA operation of MSCs’ differentiation.

Moreover, we also introduced an MPF approach to quantify the intensity changes of Raman peaks, in which the profile of each Raman peak was considered a Gaussian curve.²⁵ As shown in Fig. 7, the correlation coefficient between MPF and the raw Raman spectrum was greater than 99%. As a result, we can easily determine the accurate intensity changes of each Raman peak. This good MPF result shows that the fluorescence background embedded in the raw Raman spectrum has been effectively eliminated by using the accuracy-improved BCA approach. Like the differentiation of other cells,^10,22,29 Raman spectral changes during the differentiation of MSCs into islet-like cells also reflect their active cell cycles and the extent of protein synthesis. For example, the decreased RNA content in islet-like cells relative to MSCs suggests a lower translational rate of mRNA in these undifferentiated cells; accordingly, the protein content in islet-like cells was slightly higher than that in MSCs. The lower DNA content in islet-like cells may be indicative of the lower proliferation of these differentiated cells due to their development into a more mature phenotype. Moreover, the increase of glycogen content in islet-like cells relative to MSCs shows that differentiation is an active process that requires energy. Specially, note that two peaks at $1047{\mathrm{cm}}^{-1}$ (unassigned) and $1322{\mathrm{cm}}^{-1}$ (assigned to protein) were not observed in the BCA fitting result [Fig. 4(b)], suggesting that a new biochemical component not attributed to the seven basic biochemical components might exist in MSCs and their differentiated cells.

Fig. 7

Multiple peaks fitting (MPF) results of (a) MSCs; (b) islet-like cells, in which the black line and red dashed line (the upper row) represent the actual Raman spectra and the spectra obtained with the MPF fitting approach, respectively, and the other colored lines (the lower row) represent the different Gaussian fitting peaks. The upper left insets represent the zoomed image of the black square area of MSCs or islet-like cells spectra around the peak at $665{\mathrm{cm}}^{-1}$.

Next, combining the accuracy-improved BCA fitting and MPF results, we performed a quantitative comparison of the biochemical component and intensity changes of Raman peaks before and after MSC differentiation. Interestingly, as shown in Fig. 7 and Table 2, we can see that five peaks at 623, 644, 665, 784, and $1090{\mathrm{cm}}^{-1}$ were assigned to DNA, but four peaks at 623, 644, 784, and $1090{\mathrm{cm}}^{-1}$ were lower in intensity in islet-like cells than in MSCs. One peak at $665{\mathrm{cm}}^{-1}$ in islet-like cells had nearly twice the intensity as that in MSCs. It is impossible to determine the quantitative changes in DNA content by the intensity changes of Raman peaks using the MPF approach. However, using BCA fitting (Fig. 4), we can determine the decrease in DNA content after differentiation. It has been reported that a lower DNA content might be indicative of a lower proliferation of differentiated cells,^22,29 suggesting that our islet-like cells have developed into a more mature phenotype.

Table 2

Raman peak frequencies and their assignments.

Subsequently, we also observed that the peak at $811{\mathrm{cm}}^{-1}$ assigned to RNA was higher in intensity in MSCs than in islet-like cells (Fig. 7). This is because once the functional mature cell is generated after differentiation, its RNA content will be lowered following mRNA transcription and protein synthesis.²² MSCs are human progenitor cells and, before they become functional mature cells, their synthetic protein content is less than that in islet-like cells. Thus, the lower transcription rate of mRNA in MSCs leads to a higher RNA content in MSCs relative to islet-like cells. Moreover, various synthetic proteins are required to induce the generation of islet-like cells. Therefore, the RNA content in islet-like cells is lower than that in MSCs. Our BCA fitting result also demonstrates that the RNA content decreased after differentiation.^22,29,30 Furthermore, with the exception of the collagen peaks at 1322 and $1347{\mathrm{cm}}^{-1}$, the other peaks at 758, 830, 1030, 1157, 1174, and $1208{\mathrm{cm}}^{-1}$ (assigned to proteins) were more prominent in the differentiated cells. Our BCA operation also showed that the protein content was slightly increased after differentiation. Similar to other reports of cell differentiation,^22,29 our BCA fitting showed that an increase in protein content and a decrease in DNA content were observed during the differentiation of MSCs into islet-like cells, so the peak intensity ratio of protein to nucleic acid (such as $758{\mathrm{cm}}^{-1}/784{\mathrm{cm}}^{-1}$) was increased after differentiation.

Note that the intensity increase of the three peaks at 1004, 1030, and $1208{\mathrm{cm}}^{-1}$ assigned to phenylalanine (Phe) was also observed after differentiation, in which Phe is an important cell factor for inducing insulin generation.^31–33 In particular, an intensity increase of the peak at $1267{\mathrm{cm}}^{-1}$, which was assigned to amide III and related to changes in protein secondary structure,³⁴ was observed after differentiation, indicating that the type of synthetic protein in the amide III region might be changed after differentiation. Importantly, the peak at $665{\mathrm{cm}}^{-1}$ assigned to DNA, as described in Refs. 35 and 36, should be decreased or unchanged in intensity after differentiation. However, as shown in Fig. 7, our MPF result revealed that the intensity of the peak at $665{\mathrm{cm}}^{-1}$ was about twice as much in islet-like cells as that in MSCs, while a decrease in the total DNA content was observed after differentiation. A possible explanation is that the peak at $665{\mathrm{cm}}^{-1}$, which has been demonstrated as a characteristic peak of insulin,^37–39 is produced by islet-like cells. Because the amount of insulin produced by MSC-derived islet-like cells is relatively low,^8,11,40 the peak at $665{\mathrm{cm}}^{-1}$ was lower in intensity than those Raman peaks between 1000 and $1100{\mathrm{cm}}^{-1}$ or around $1300{\mathrm{cm}}^{-1}$. Subsequently, we also observed that other DNA peaks at 623, 644, 784, and $1090{\mathrm{cm}}^{-1}$ were lower in intensity after differentiation. Collectively, along with the decrease in DNA content, we conclude that the increased intensity of the peak at $665{\mathrm{cm}}^{-1}$ after differentiation is induced by insulin generation, further indicating that the directional differentiation of MSCs into islet-like cells was successful. It should be reasonable to utilize the intensity increase of the peak at $665{\mathrm{cm}}^{-1}$ as a marker for the identification, physical separation, and enrichment of islet-like cells. Of course, several other peaks at 830, 1030, and $1174{\mathrm{cm}}^{-1}$, which are observed in the spectrum of insulin,³⁷ were also more prominent in islet-like cells than in MSCs. Similarly, we also can also use these peaks as characteristic parameters for the generation of islet-like cells.

It is well known that glucose serves as a rapid energy source. Our BCA fitting showed that glycogen content was increased after differentiation, indicating that glucose served as the energy source for the differentiation of MSCs into Islet-like cells. By combining the improved BCA and MPF approaches, we determined the quantitative biochemical changes and the quantitative intensity changes of Raman peaks for MSC differentiation into islet-like cells.

3.2.

Raman Data Analysis with Principal Component Analysis

To assess the discrimination of MSCs and their differentiated cells, we employed PCA to compare the background-subtracted and normalized Raman spectra of different groups in this study. In PCA, if one type of cells is separated into the same cluster, then the scatter plot generated from this group data represents the same type of cells.⁴¹ Because the first few principal components (PCs) and the corresponding loading values for each PC contain most of the spectral information of the raw Raman data, each spectrum can be expressed as a linear combination of the first few PCs. As shown in Fig. 8, we first constructed two-dimensional plots with different combinations of scores for the first three PCs [Figs. 8(a)–8(c)] and three-dimensional plots with three sets of scores [Fig. 8(d)]. Interestingly, it was observed that the first three PCs accounted for approximately 85% of the total spectral information. Because protein is the main component of biological cells and PC1 accounted for 55% of the total spectral information, the spectral profile of PC1 is very similar to the actual cell spectrum [Figs. 8(a) and 9]. Moreover, it was seen that cells before and after differentiation were separated into two distinct clusters in PC1, indicating that the discrimination of proteins before and after differentiation can be determined by PC1. However, using BCA fitting (Fig. 4), no obvious changes in protein content were observed, indicating that PC1 is not just composed of protein. As shown in Fig. 7 and Table 2, some of these peaks assigned to proteins increased in intensity while others decreased in intensity, and one possible explanation is that conformational changes in some proteins occurred.

Fig. 8

(a)–(c) Two-dimensional principal component analysis (PCA) scatter plots of the first two PCs, (d) three-dimensional PCA scatter plot, in which the contributions of PC1, PC2, and PC3 were 55%, 19%, and 11%, respectively.

Fig. 9

Spectra of the first three principal components in PCA: (a) PC1, (b) PC2, and (c) PC3.

As described in the above analysis, we can see that the data classification with BCA is different from that with PCA. The former is performed by the content changes of biochemical components, while the latter is performed by the spectral changes of Raman peaks. Even if the biochemical component content is not changed, when the structure of the biochemical component is changed, its corresponding Raman peaks will be changed. Thus, BCA is suitable for revealing the biochemical changes of cells, while PCA can supply the discrimination information of different type cells.

PC2 and PC3 accounted for only 19% and 11%, respectively, of the total spectral information [Figs. 8(a) and 8(b)], so it was difficult to assess the discrimination between MSCs and the differentiated cells with PC2 or PC3. As a result, we can see that there was a large discrimination between PC2 or PC3 and the actual cell spectrum (Fig. 9). Although the PC2 spectrum captured a peak $1064{\mathrm{cm}}^{-1}$, changes in the lipid content before and after differentiation were observed by BCA fitting. However, PC2 accounted for 19% of the total spectral information, so the discrimination of different cell groups in PC2 was less than that in PC1. Similarly, the discrimination of different cell types in PC3 was much less.