Visualizing cell state transition using Raman spectroscopy

Abstract

System level understanding of the cell requires detailed description of the cell state, which is often characterized by the expression levels of proteins. However, understanding the cell state requires comprehensive information of the cell, which is usually obtained from a large number of cells and their disruption. In this study, we used Raman spectroscopy, which can report changes in the cell state without introducing any label, as a non-invasive method with single cell capability. Significant differences in Raman spectra were observed at the levels of both the cytosol and nucleus in different cell-lines from mouse, indicating that Raman spectra reflect differences in the cell state. Difference in cell state was observed before and after the induction of differentiation in neuroblastoma and adipocytes, showing that Raman spectra can detect subtle changes in the cell state. Cell state transitions during embryonic stem cell (ESC) differentiation were visualized when Raman spectroscopy was coupled with principal component analysis (PCA), which showed gradual transition in the cell states during differentiation. Detailed analysis showed that the diversity between cells are large in undifferentiated ESC and in mesenchymal stem cells compared with terminally differentiated cells, implying that the cell state in stem cells stochastically fluctuates during the self-renewal process. The present study strongly indicates that Raman spectral morphology, in combination with PCA, can be used to establish cells' fingerprints, which can be useful for distinguishing and identifying different cellular states.

Figure 1. Raman images of three cell-lines.

RGB reconstituted Raman images of NIH3T3 (A), EPH4 (B) and Hepa1–6 (C) cells. Raman peaks at 753 cm⁻¹ (cytochrome C), 1686 cm⁻¹ (proteins), and 2852 cm⁻¹ (lipids) are mapped in blue, green, and red, respectively. Scale bar, 10 µm. (D) Raman spectra of the nuclei of NIH3T3 (blue), EPH4 (purple) and Hepa1-6 (orange) cells. (E) Raman spectra of the cytosol of NIH3T3 (blue), EPH4 (purple) and Hepa1-6 (orange). The representative Raman spectra shown are the average of spectra at 49 pixel positions in the black circled region in A–C. Peak assignments are, RP; rebose-phosphate, BK; backbone OPO of nucleic acid, str; stretching, def; deformation . Peaks characteristic to cytochrome C are indicated with asterisks.

Figure 2. Difference in Raman spectra between cell-lines.

(A) Averaged Raman spectra of NIH3T3 (blue), EPH4 (purple) and Hepa1-6 (orange) cells in the fingerprint region (700–1800 cm⁻¹). Raman spectra are average of 10–34 cells for each cell-line. The lower envelope, which was estimated by a 4^th-order polynomial fitting, was subtracted from all spectra in order to make the spectral differences clearer for comparison . (B) Score plots calculated by PCA for three cell-lines. For PCA analysis, raw spectra without averaging was used. Each symbol represents a single cell. NIH3T3 (blue), EPH4 (purple) and Hepa1-6 (orange).

Figure 3. Raman images of cell-lines with differentiation capability.

Raman images of Neuro2a (A) and 3T3L1 (D) cells before (left panel) and after (right panel) the induction of differentiation (inset; bright-field image). (B, E) Averaged Raman spectra of N2a (B) and 3T3L1 (E) cells before (blue) and after (red) induction of differentiation. Spectra are average of 15–27 cells. Spectra from the fibroblast cell-line NIH3T3 are also plotted (black). Peaks characteristic to cytochrome C are indicated with asterisks. (C, F) Score plots of Neuro2a (C) and 3T3L1 (F) cells before (blue) and after (red) the induction of differentiation calculated by PCA. For PCA analysis, raw spectra without averaging was used. Data from the fibroblast cell-line NIH3T3 are also plotted (black). Each marker shows averaged score values of the spectra obtained from single nuclei. Error bar shows SD of the score values from the same nuclei.

Figure 4. Raman images of ESCs before and after induction of differentiation.

Bright-field (left panel) and Raman images (right panel) of undifferentiated (A) and differentiated (B) ESCs. (C) Averaged Raman spectra of undifferentiated ESCs (blue), differentiated ESCs (red), and MSCs (green) in the fingerprint region (700–1800 cm⁻¹). For PCA analysis, raw spectra without averaging was used. Spectra shown are average of 18–44 cells.

Figure 5. PCA analysis of Raman spectra obtained from ESCs.

(A) Calculated loading vectors of PC1–PC5 of the Raman spectra of the nuclei in the fingerprint region. For PCA analysis, raw spectra without averaging was used. (B) Plot of PC1 and PC2 scores. Each marker shows the average value of the PC1 and PC2 scores of the Raman spectra obtained from single nuclei. Error bars show SD of the score values from the same nuclei.

Figure 6. Transition of ESC state during differentiation.

(A–E) Score plots of Raman spectra at various stages of differentiation spanning a period of 2 weeks. Each marker shows averaged score values of the spectra obtained from single nuclei. Error bar shows SD of the score values from the same nuclei. (F) Histograms of the population distribution of the SD relative to PC1 and PC2 before (blue), 1 week (red) and 2 weeks (green) after LIF removal. Solid lines show the results of the fitting achieved with two Gaussian functions.