2.1.

Experimental Setup

A schematic of a CARS experimental setup is shown in Fig. 1. The prototype module of a laser light source, which can generate both pump and broadband Stokes lights by using a photonic crystal fiber (PCF), was used to perform multiplex CARS.¹⁶ Pulse duration, center wavelength, average power, and repetition rate of the microchip laser (Horus laser HLX-I-F040) used in the module were 1.1 ns, 1064 nm, 495 mW, and 27 kHz. The laser light was branched to PCF (NKT Photonics SC-5.0-1040-PM) and single-mode fiber (Nufern PM980-XP) by using a half-wavelength plate and polarizing beam splitter. The pump and Stokes lights were combined by a dichroic mirror and focused into a sample by an objective lens (Nikon CFI Apo LWD 40xWI, NA 1.15). When pump and Stokes lights are focused at the same position in a sample, CARS light is generated. Average powers of pump and Stokes lights measured after passing through the objective were approximately 50 and 10 mW, respectively. CARS spectrum was obtained by a spectroscope (Princeton PIXIS400, Acton SP-2358), and Raman spectrum was reconstructed by the maximum entropy method (MEM).¹⁷ After the reconstruction, the quality of spectra was improved by singular value decomposition (SVD).¹⁸ It is one of the advantages of multiplex CARS to be able to use SVD comparing with monochromatic CARS. To adjust power ratio between pump and Stokes lights, a half-wavelength plate was rotated. Divergence of Stokes light was adjusted by changing position of a collimator lens for PCF to adjust multiplexing condition of pump and Stokes lights.

Fig. 1

Schematic of experimental setup for multiplex CARS.

CARS images were obtained by changing position of a sample by using a piezo stage. The typical size of CARS image and acquisition time to obtain the image were $50\times 50\mu {\mathrm{m}}^2$ and 5 min, respectively. Figure 2 shows a CARS image indicating distribution of ${\mathrm{CH}}_2$ stretch ($2900{\mathrm{cm}}^{-1}$) in $1\text{-}\mu \mathrm{m}\varphi$ polystyrene beads. The lateral resolution was estimated at approximately $1\mu \mathrm{m}$ from measured width of the beads.

Fig. 2

(a) Bright-field image and (b) CARS image of $1\text{-}\mu \mathrm{m}\varphi$ polystyrene beads.

2.2.

Sample

Assuming application to regenerative medicine, cartilage cells differentiated by the following method were measured as a sample. Approximately $4\times {10}^4$ C3H10T1/2 cells were seeded onto a $35\text{-}\mathrm{mm}\varphi$ glass-bottom dish and incubated in 2-mL D-MEM (Dulbecco's Modified Eagle Medium) D6429 under 5% ${\mathrm{CO}}_2$ at 37°C for 2 days. To induce differentiation, the medium was replaced with 2 mL of the D-MEM containing $100\mathrm{ng}/\mathrm{mL}$ bone morphogenetic protein 2 (BMP-2) in another 2 days. For comparison, the medium was just changed to a new D-MEM in a control sample. After culturing for 3 days, the cell density was adjusted to $1\times {10}^4$ to obtain CARS images as a single cell.

Figures 3(a) and 3(b) show typical bright-field images of the cells cultured with and without BMP-2, respectively. It is difficult to identify differentiation by cell morphology, so cell composition needs to be analyzed. Because one of the major components of cartilage is collagen, we focused on distribution of proline, which generally accounts for approximately 20% of collagen. Spectra of gelatin and adipose cell were compared with confirm signal of proline because gelatin is extracted from collagen and was expected to show greater signal than adipose. Adipose cells were also used to measure the effect of optical adjustment because adipose shows several spectrum peaks in fingerprint region and is suitable to investigate signal-to-noise ratio (SNR) of spectrum. To validate a CARS image of proline, the stained image by using sirius red F3B of the same cell was obtained after CARS measurement.

Fig. 3

Typical bright-field images of the cell cultured (a) with and (b) without BMP-2.

3.1.

Optical Adjustment to Obtain CARS Images of Fingerprint Region

To obtain CARS images of weak spectrum peak in a fingerprint region such as proline, an optical adjustment method was investigated. Because CARS is a third-order nonlinear optical effect, adjustment of the power ratio between pump and Stokes lights is essential, especially under the condition with limited laser power. Figure 4(a) shows the spectra of broadband Stokes light generated by PCF. The spectrum changed depending on incident laser power, and it is needed to exceed a threshold power to generate Stokes light with a specific wavelength. For example, to generate Stokes light with a wavelength of 1550 nm, which contributes to the generation of CARS light with approximately $3000{\mathrm{cm}}^{-1}$, the incident laser power higher than 50 mW is necessary. On the other hand, the threshold is 20 mW to generate Stokes light with 1200 nm, and the rest of the laser power could be used as pump light. CARS light power is proportional to Stokes light power and the square of pump light power, so the signal intensity can be calculated by using the generation characteristic of broadband Stokes light. Figure 4(b) shows the relationship between power branching ratio and CARS signal intensity calculated by the following equation:

Fig. 4

(a) Stokes light spectra depending on PCF incident laser power. (b) Relationship between CARS signal intensity and power branching ratio of incident light on PCF to that on single-mode fiber to propagate pump light.

Furthermore, intensity of CARS signal is highly sensitive to the multiplexing condition of pump and Stokes lights. When Stokes light is focused in a measurement sample by an objective lens, chromatic aberration occurs due to broadband characteristics and focal point differs slightly depending on the wavelength. Because CARS light is generated at only region where both pump and Stokes lights are focused, the shape of the CARS spectrum changes depending on multiplexing condition of pump and Stokes lights.

Figure 5(a) shows the CARS spectra at the same position of adipose by changing optical adjustments described previously. A strong peak of $\mathrm{C}─\mathrm{H}$ stretch at $2900{\mathrm{cm}}^{-1}$ was obtained; however, intensities of peaks in the fingerprint region were extremely weak before the adjustment. On the other hand, although the peak of $\mathrm{C}─\mathrm{H}$ stretch disappeared, peaks in the fingerprint region improved after the adjustment. The SNR of peaks in fingerprint region is compared in Fig. 5(b). We defined signal as the difference between top and bottom levels of the peak. Noise was defined as the average intensity of high-frequency component, which corresponds to fluctuation at less than approximately $5{\mathrm{cm}}^{-1}$, after fast Fourier transformation of the spectrum. The SNR was confirmed to improve by more than 5 dB. The SNR of low wavenumber peaks such as amide III was improved greater than high wavenumber signal by the adjustment targeting at lower wavenumbers.

Fig. 5

(a) CARS spectra measured at the same position of adipose by changing optical adjustment. (b) SNR comparison before and after optical adjustment.

3.2.

CARS Hyperspectral Imaging of Cartilage

To confirm signal of proline, a spectrum of gelatin was investigated prior to measurement of a cartilage cell. Figure 6 compares the spectra of gelatin and an adipose cell. The spectrum peak of proline¹ was obtained at approximately $860{\mathrm{cm}}^{-1}$. Figures 7(a) and 7(b) show a bright-field image and CARS spectra at points A and B of a cartilage cell, respectively. Several spectrum peaks containing proline were obtained at point A (region of cell). CARS images of amide I ($\sim 1650{\mathrm{cm}}^{-1}$), phenylalanine ($\sim 1000{\mathrm{cm}}^{-1}$), and proline ($\sim 860{\mathrm{cm}}^{-1}$) are shown in Figs. 8(a), 8(b), and 8(c), respectively. CARS hyperspectral images were successfully obtained by the optical adjustment described above. Images of amide I and phenylalanine indicate that those molecules were distributed throughout the cell. On the other hand, periphery of the cell is highlighted in the CARS image of proline. Figure 9 shows the bright-field images and CARS images of proline of other cells. Periphery of the cell was confirmed to be highlighted in each CARS image.

Fig. 6

Spectrum comparison between gelatin and adipose cell.

Fig. 7

(a) Bright-field image of cartilage cells. (b) CARS spectra at points A and B.

Fig. 8

CARS images of (a) amide I, (b) phenylalanine, and (c) proline.

Fig. 9

Correspondences of bright-field image and CARS image of proline.

Figure 10 shows the comparison of CARS image of proline and the stained image. Periphery of the cell was highlighted in Fig. 10(b) as discussed previously. Figure 10(c) is the stained image of the same cell by using sirius red F3B. Collagen and other protein were stained green and purple, respectively. The extraction image of the region containing purple by image processing of Fig. 10(c) is shown in Fig. 10(d). Comparing Figs. 10(b) and 10(d), periphery of the cell is emphasized in both images. These results suggest that the CARS image of proline shows a distribution of generated collagen, an extracellular matrix. Because cells were seeded as a single cell, collagen would be localized at a portion of the cell periphery.

Fig. 10

(a) Bright-field image, (b) CARS image, (c) stained image, and (d) extraction image of region containing purple of cartilage cell.