Supercontinuum fiber laser-based coherent anti-Stokes Raman scattering microscopy for label-free chemical imaging

1. Introduction

Coherent anti-Stokes Raman scattering (CARS) microscopy has revolutionized our ability to visualize the noninvasive imaging complex systems with high-spatial resolution, high sensitivity, and label-free chemical specificity.¹ CARS is a four-wave mixing process in which a pump-probe beam (${\omega }_p$) and a Stokes beam (${\omega }_s$) interact with molecules in the specimen. When the frequency difference (${\omega }_p-{\omega }_s$) coincides with a given vibrational mode ($\mathrm{\Omega }$), a strong anti-Stokes signal is generated at the frequency of $2{\omega }_p-{\omega }_s$. Since the wavelength of the CARS signal has blue-shifted from that of the incident radiation, we can easily eliminate the background caused by single-photon fluorescence.² It has tremendous applications in several disciplines including chemistry, medicine, and biology,^3,4 especially for visualizing the biomedical samples such as living cells⁵ and tissues.⁶ Current multifarious CARS microscopy techniques have been developed using two-synchronized laser sources,⁷ for example, Nd:YVO₄ laser and optical parametric oscillator,⁸ single laser oscillator combining with a spatial light modulator (SLM),⁹ Ti: sapphire amplifier system and optical parametric amplifier (OPA),¹⁰ and combination of a photonic crystal parametric oscillator.¹¹ However, these typical laser sources are generally complicated and expensive, and they use solid-state lasers to generate synchronized pulses as pump and Stokes beams, leading to increased prices and difficulty in popularizing the entire CARS microscope. In addition, the remaining challenge of lasers in bulky size and sensitive to the environment has hindered the clinical and in vivo applications of CARS microscopy.

Supercontinuum fiber (SF) laser shows promise in meeting these characteristics of CARS, namely, good beam quality, low cost, and compactness, which makes CARS microscopy promising for chemical analysis and clinical applications.¹² Several previous studies have demonstrated CARS based on SF lasers, by combining SF Stokes pulses with pump pulses from a single laser oscillator.¹³ Kano et al. used a Ti: Sapphire oscillator and a photonic crystal fiber as the laser pulses, which are divided into the pump and seed pulses for the supercontinuum generation, to detect Raman spectra of cyclohexane.¹⁴ However, they only focus on CARS spectral measurements of specimens. Okuno et al. described a multiplex microspectroscopic system that used a nanosecond supercontinuum generation from a photonic crystal fiber and a sub-nanosecond pulse laser, to measure the vibrational spectra and images of yeast cells in the ultrabroadband spectral range ($>2500$cm${}^{-1}$) and high spectral resolution ($<0.1$cm${}^{-1}$).¹⁵ Li et al. developed an SF source using a Yb-doped dissipative soliton fiber laser and used it to build a broadband CARS micro spectroscope, providing a resolution of less than 10cm${}^{-1}$ in the spectral range from 700to 1900cm${}^{-1}$.¹⁶ Camp et al. constructed a broadband CARS platform to probe the correlated Raman resonance signals of whole organisms at a high spectral resolution of less than 10cm${}^{-1}$.¹² However, these studies primarily focused on the pursuit of high spectral resolution, which has led to a relatively coarse spatial resolution. In addition, all of these techniques typically acquired two-dimensional images of the spatial distribution of chemicals by moving the sample through a translational stage, which greatly hinders the imaging speed and the imaging field of view to only a few tens of microns.

Upon the excellent features of the SF laser and the various limitation of current techniques, in this paper, we homebuilt a low-cost CARS microscope based on a SF laser module (scCARS microscope). This SF laser module is specially customized by adding a time-synchronized seed source channel to the SF laser to form a dual-channel output laser. The Stokes and pump pulses are generated from the SF laser module, and the corresponding pump pulses are purified by a fast-tunable filter. The performance of the scCARS microscope is evaluated with dimethyl sulfoxide, whose results confirm a spatial resolution of better than 500nm and a detection sensitivity of millimolar concentrations. The dual-color imaging capability is further demonstrated by imaging different species of mixed microspheres. We finally explore the potential of our scCARS microscope by mapping lipid droplets in different cancer cells and corneal stromal lenses.

2. Materials and Methods

The scheme of a low-cost CARS microscope based on a SF laser module (scCARS microscope) is shown in Fig. 1, where Fig. 1(a) is a schematic diagram of the setup, and Fig. 1(b) is the photo of the scCARS microscope. This SF laser module was specially customized on the basis of a SF laser (NTK Photonics, Denmark), providing a pulse train with a repetition rate of 40MHz. This pulse train is the supercontinuum source with an adjustable output between 400 and 2400nm and is served as the pump beam for CARS signal excitation. The output power and pulse duration are 4W and 58ps, respectively. With the help of a fast-tunable filter (NKT SuperK VARIA, Denmark), a single wavelength can be selected with an adjustable bandwidth of 1–100nm. In order to be able to stimulate the CARS signal, the other time-synchronized pulse train is needed as the excitation source. This pulse train is centered at 1064nm and is employed as the Stokes beam, with a pulse duration of 186fs and the pulse power of 1.1W. The two pulse trains are specially customized to a SF laser module of dual-channel time-synchronous outputs. The Stokes beam is first passed through a half-wave plate and a beam splitter to adjust the power in a specific polarization direction, thereby reducing the interference of noise to the signal, and then is combined with the delayed pump beam by a dichroic mirror (DMLP1000, Thorlabs). The spatially overlapped Stokes and pump beams are expanded and collimated by a 4-f system, and then delivered to a microscope (BX51WI, Olympus). Delay stages are equipped to adjust the coincidence of the two beams in time as they reach the sample. A sliding rail with a long adjustable range of 300mm was used in the pump arm for coarse tuning to determine the coincident of the pulse of two excitation beams, and the other high-precision translation stage (XR50P/M, Thorlabs) was employed in the Stokes arm to precisely correct the time delay in a small range. A 60X water immersion objective lens ($\mathrm{\text{NA}}=1.2$, UplanSApo, Olympus) is used to focus the temporally and spatially coincident beams onto the sample. An oil condenser ($\mathrm{\text{NA}}=1.4$, U-UCD, Olympus) is employed to collect the generated CARS signal. A dichroic mirror (FF735-Di02-$25\times 36$, Semrock) and a bandpass filter (FF01-661/20-25, Semrock) are housed in a lens tube and used for purification of the CARS signal that is finally collected by a photomultiplier (PMT, H7422-40, Hamamatsu). The CARS images are assembled by scanning the temporally and spatially coincident beams using two one-dimensional galvo mirrors (GVSM001/M, Thorlabs).

3. Results and Discussion

We first characterized the performance of the scCARS microscope. Using dimethyl sulfoxide (DMSO) as the test object, we investigated the dependence of the scCARS signal on the laser powers, as well as the spatial resolution and the sensitivity of the scCARS microscope. Figures 2(a) and 2(b) show that the scCARS signal increased linearly with the power of the Stokes beam ($R^2=0.9982$) and quadratically with the power of the pump beam ($R^2=0.9951$). These results were consistent with the theory of CARS signal generation.¹⁰ We then measured the spatial resolution by acquiring the scCARS signal of DMSO at 2931cm${}^{-1}$. The scCARS signal map of DMSO is shown in Fig. 2(c). We extracted a profile across the DMSO-air interface and calculated its shock response. After fitting the shock response with a Gaussian curve (Fig. 2(d)), the spatial resolution can be determined from the full width at half maxima of the fitted Gaussian curve, with a value of 445nm. We further evaluated the sensitivity of the scCARS microscope by detecting the C–H bonds from DMSO diluted in deuterium oxide (D₂O). The pump and Stokes powers were 5.3 and 70.9mW, respectively. The pixel dwell time was 0.3ms. Figure 2(e) shows the measured signal-to-noise ratio as a function of DMSO concentration. These results indicated that our system can detect 141mM DMSO very significantly. The scCARS signal map at such concentration is shown in Fig. 2(f), showing that the interface of DMSO and air can be clearly resolved. From Fig. 2(e), we find that when the DMSO concentration was reduced to 70.5mM, the signal-to-noise ratio of the scCARS image was still greater than 1. This result proved a millimolar sensitivity ($<70.5$mM) of the scCARS microscope in C–H bond.

Next, we demonstrated the dual-color imaging capability of the scCARS microscope by imaging different species of mixed microspheres, including the polystyrene (PS) beads and polymethyl methacrylate (PMMA) beads, who have strong Raman peaks around 3057 and 2871cm${}^{-1}$, respectively. We first imaged PS beads (10$\mu$m in diameter) and PMMA beads (20$\mu$m in diameter) fixed with agarose gel, respectively. The excitation power of the pump beam was 4.5mW, and that of the Stokes beam was 42 and 56mW, respectively. The pixel dwell time was 0.1ms, thus, it takes about 9s to obtain an image. Figure 3(a) presents the scCARS images of PS beads at 2871cm${}^{-1}$ (on resonance) and 3259cm${}^{-1}$ (off-resonance), respectively. The scCARS images of PMMA beads at 2931and 3259cm${}^{-1}$ are shown in Fig. 3(b). We then imaged the mixture of PS and PMMA beads having 10 and 20$\mu$m diameters, respectively. The excitation powers of the pump and Stokes beams were 42 and 49mW, respectively, with a pixel dwell time of 0.1ms. The corresponding dual-color imaging results are shown in Fig. 3(c). These results demonstrated the validity of our scCARS microscope in resolving different components of samples.

As an example to detect biomolecules, we finally imaged lipid droplets in different cancer cells and corneal stromal lenses. Lipid droplets are organelles that store neutral lipids and are essential for energy metabolism. Their function in energy storage is firmly established and increasingly well-characterized and plays an important role in human health.¹⁷ We first imaged lipids of esophageal squamous carcinoma cells (EC109 cell lines), breast cancer cells (4T1 cell lines), and melanoma cells (B16F10 cell lines), as shown in Figs. 4(a)–4(c). To image the lipids in EC109 and 4T1 cells, the power of pump beams was tuned to 4.6mW and 4.2mW, respectively. The power of the Stokes beams was set at 63.5mW and 34.3mW. The pixel dwell time was 1.5ms for both types of cells, including a total acquisition time of 135s for an image. We first tuned the Raman shift to the region of CH₂ asymmetric vibration at 2931cm${}^{-1}$ to collect the resonant scCARS signals (First row in Figs. 4(a) and 4(c)). The corresponding images of nonresonant signals are then shown in the second row of Figs. 4(a) and 4(c), displaying no obvious signals, thus demonstrating that indeed the CARS signals were detected by our scCARS microscope. Next, we also detected the CH₃ stretching vibrations in B16F10 cells by tuning the Raman shift to 2910cm${}^{-1}$, as presented in the first row of Fig. 4(c), where the powers of the pump and Stokes beams were 4.5mW and 56.1mW. The pixel dwell time was 0.8ms, corresponding to a total acquisition time of 72s for an image. Similarly, an image from the nonresonant signals was collected at 2577cm${}^{-1}$ (Second row in Fig. 4(c)). These results further validated that the scCARS microscope can well identify the biomolecules, such as lipid and protein in cancer cells. Further, we also measured the lipid information in the corneal stromal lens with the scCARS microscope. The corneal stromal lens was provided by Xi’an People’s Hospital and was taken from the myopic population. This study was approved by the Ethics Committee of Xi’an People’s Hospital (Xi’an Fourth Hospital) with a number of 20210010. The cornea plays a critical role in the eye, including protecting the contents of the interior of the eye and providing approximately two-thirds of the eye’s refractive power. The corneal stromal lens is rich in collagen protein and accounts for nearly 90% of the thickness of the cornea. It is well known that the state of corneal hydration plays a crucial role in maintaining optimal vision. The ratio between Raman intensity of the water OH mode at 3400cm${}^{-1}$ and the lipid/collagen protein CH stretching mode at 2940cm${}^{-1}$ is usually used to characterize the hydration state of cornea.¹⁸ We first measured the Raman spectra of the corneal stromal lens by using our homemade confocal Raman spectroscopic imaging system. We found a strong Raman peak at 2940cm${}^{-1}$ and a relatively strong Raman peak at 3400cm${}^{-1}$, reflecting the presence of C–H stretching and O–H vibration information in the corneal stromal lenses. We, therefore, chose to acquire the scCARS images at these two Raman shifts, as shown in Fig. 4(d) and calculated the ratio between them ($R_{\mathrm{\text{OH}}}/R_{\mathrm{\text{CH}}}$) that reflects the hydration state of the cornea. The ratio was approximately 1.5, indicating that the water content is higher than that of lipid/collagen protein, and suggesting that the water balance in the corneal may be slightly imbalanced. The laser powers here were the same as those used in imaging of B16F10 cells, and the pixel dwell time was 1.0ms for acquiring scCARS images of $300\times 300$ pixels. Collectively, these results further highlighted the use of the scCARS microscope to detect the biomolecules in a biologism.

There is some room for further improvement. First, the scCARS microscope requires a relatively long acquisition time to obtain high-quality scCARS images. Currently, the branch of the SF laser that serves as the pump beam outputs a light beam with continuously adjustable wavelengths in the 400–2400nm range. Although it has a total power of up to 4W, the power allocated at a single wavelength is low, resulting in a relatively low sensitivity of the microscope. Therefore, in order to obtain higher quality scCARS images, we used a single point dwell time of milliseconds in the experiments. This can also be further improved in two ways. First, with the help of a deep learning network,¹⁹ rapidly acquired poor quality scCARS images can be of quality enhancement by using high-quality images as training sets and thus accelerate the scCARS imaging speed. Second, the surfaced-enhanced Raman scattering (SERS) technique can provide many orders of magnitude enhancement of specific Raman signals of samples.^20,21,22 Thus, the scCARS microscopy can integrate with the SERS strategy to enhance the image quality and thus reduce the acquisition time.

Second, in our experiments, the high excitation power was obtained by increasing the linewidth of the pump beam in order to acquire a good quality image. The increase in the linewidth of the pump beam causes a decrease in the spectral resolution of the scCARS microscope. For example, in our experimental measurements, the current scCARS microscope has a poor spectral resolution of up to 35cm${}^{-1}$. The poor spectral resolution can also be improved by incorporating the SERS strategy. The introduction of SERS can increase the Raman signal intensity by orders of magnitude, which can reduce the linewidth of the pump beam and ultimately improve the spectral resolution. Similar to the use of deep learning networks to enhance image quality in Ref. 19, we can also combine deep learning techniques for spectral resolution enhancement, which is the next step we will focus on.

Conclusions

The current CARS microscope usually uses a bulky, high-performance solid-state laser as the excitation light source, which is expensive and very sensitive to environmental changes. The low-cost SF lasers with good beam quality and compact space occupation hold great promise in CARS microscopy for chemical analysis and clinical applications. In this study, we homebuilt a low-cost CARS microscope based on a customized SF laser module (scCARS microscope), which has a spatial resolution of 445nm and a detection sensitivity of millimolar concentrations ($<70.5$mM). The laser module is a dual-channel output laser by adding a time-synchronized seed source channel to the SF laser. To demonstrate the performance of our scCARS microscope, we obtained the CARS resonance images of polystyrene beads, polymethacrylate beads, and polymethyl the mixture of these two at different Raman shifts, respectively. We also explored the potential of our scCARS microscope by mapping lipid droplets in different cancer cells and corneal stromal lenses. These results suggest the capability of our system in detecting the biomolecules in biologisms. Our work provides an alternative solution for the implementation of low-cost and miniaturized CARS microscopes, which are essential for their application in nonlaboratory environments.

Conflicts of Interest

The authors declare that there are no conflicts of interest relevant to this article.

Acknowledgment

Nan Wang and Jiaojiao Zhang contributed equally to this work. This work was supported in part by the National Key R&D Program of China (2018YFC0910600), the National Natural Science Foundation of China (81871397), the National Young Talent Program, the Shaanxi Science Fund for Distinguished Young Scholars (2020JC-27), the Key Research and Development Program of Shaanxi (2021ZDLSF04-05), the Shaanxi Young Top-notch Talent Program, the Best Funded Projects for the Scientific and Technological Activities for Excellent Overseas Researchers in Shaanxi Province (2017017), and the Fundamental Research Funds for Central Universities (QTZX2105). Xueli Chen would like to thank Dr. Chi Zhang at Purdue University for his help in building the CARS microscope.