Coherent Anti-Stokes Raman Scattering Microscopy and Its Applications

Coherent anti-Stokes Raman scattering (CARS) microscopy can provide high resolution, high speed, high sensitivity, and non-invasive imaging of specific biomolecules without labeling. In this review, we first introduce the principle of CARS microscopy, and then discuss its configuration, including that of the laser source and the multiplex CARS system. Finally, we introduce the applications of CARS in biomedicine and materials, and its future prospects.


INTRODUCTION
Raman scattering was named after its discoverer C. V. Raman, the celebrated Indian scientist, in 1928 [1]. The discovery of Raman scattering not only proved that the energy level of light is discrete but also opened up a new research area for the study of the chemical characteristics of molecules. First, Raman scattered photons in the mid-infrared range carry information about molecular vibration. The switch from infrared to visible light circumvents the strong optical absorption of water in the infrared region, which is of great significance in the field of biomedical science. Second, Raman spectroscopy obviates the need for chemical labeling when obtaining chemical information, so that the biological functional characteristics of the sample can be studied without chemical interference.
Although it has many advantages, spontaneous Raman scattering is a very weak second-order radiative process. The cross-section of Raman scattering is approximately 10-30 cm 2 per molecule, which is 10 11 times smaller than the corresponding cross-section of infrared absorption. The small cross-section of the scattering leads to a low spontaneous Raman scattering efficiency and consequently necessitates a long integration time. This is acceptable for Raman spectroscopy, but for imaging, it is not feasible to collect real-time data with such a long integration time. The development of the femtosecond laser with high peak power has enabled the observation of many nonlinear processes, and promoted the study of coherent Raman scattering.
As a new nonlinear optical microscopy technique, coherent Raman scattering microscopy overcomes the limitation of the imaging speed of spontaneous Raman micro-imaging [2][3][4][5][6][7][8]. In coherent Raman scattering microscopy, typically, two beams of light are used to simultaneously excite the sample, one beam is referred to as the pump beam and the other is the Stokes beam. When the frequency difference between the two beams of light corresponds to a chemical bond vibrational frequency of the target sample, four simultaneous coherent Raman processes will occur. These processes are coherent anti-Stokes Raman scattering at the frequency (ω p − ω s ) + ω p , coherent Stokes Raman scattering at the frequency ω p − (ω p − ω s ), stimulated Raman gain at ω s , and stimulated Raman loss at ω p . The CARS signal is generated at a frequency that is different from the excitation light. Stimulated Raman gain and stimulated Raman loss are both employed in another coherent Raman microscopy technique referred to as Stimulated Raman Scattering (SRS) Microscopy.
The first observation of CARS was reported by Terhune and Maker [9], as a third-order nonlinear effect in 1965 at Ford Motor Company, the car manufacturer. However, the official name of CARS was not coined until 10 years later by Begley [10]. Begley and his co-workers used CARS to study the chemical reaction kinetics of combustion. CARS microscopy was first implemented by Duncan [11] in 1982 using a non-collinear geometry to match the phase and obtain two-dimensional images. In 1999, Zumbusch [12] changed the excitation light configuration to overlap collinearly with the Stokes beam, and used a femtosecond laser for close focusing to achieve fast threedimensional imaging. This work greatly promoted the rapid development of modern CARS microscopy technologies. Over the next few years, a systematic study of CARS was performed by Cheng and co-workers [2,[13][14][15][16]. In 2001, Cheng exploited polarization-CARS by using the difference between the polarizations of resonant and non-resonant CARS signals to eliminate the non-resonant background. Laser scanning CARS microscopy based on the picosecond pulsed laser was developed for high-speed imaging of biomedical samples [16], and the imaging speed of CARS reached the level of video-rate (20 frames/s) in 2005 [17]. Many strategies, including frequency modulation [18], nonlinear interferometric vibrational imaging [19], and spectral mixing [20] were used to suppress the nonresonant background or extract the pure Raman responses from the CARS spectrum.
In this review, the history and early development of CARS are introduced, as well as the implementation of both single frequency and multiplex CARS. Subsequently, recent applications in biomedical and material science are presented. Finally, the future directions of CARS are discussed.

CARS PRINCIPLE
CARS is a third-order nonlinear optical process. Under the interaction of E(p), E(s) of the Stokes light field and E′(p) of the probe light field, the sample molecules will produce an anti-Stokes field E (as) with a frequency of ω as 2ω p − ω s . Figure 1 shows the energy diagram of the CARS process. Figures 1A,B show the generation process of the resonant CARS and nonresonant CARS signals, respectively, under the action of the pump light. In most cases, both the pump and probe light originate from the same laser beam; thus, in the above process, I CARS describes the intensity of the CARS signal and is given by Where k i 2π/λ i is the wavevector, z is the thickness of the sample, and the wavevector mismatch Δk is defined as Δk k as − (2K p − k s ). When Δk z is close to zero, the sinc function is maximized. This condition is known as the phasematched condition. Figures 1C exhibits the phase-matched conditions for forward-and epi-detected CARS signals. Here χ (3) is the third-order polarizability which is usually expressed in the following form: where Δ is the Raman shift, which can be described as Δ ω p − ω s − Ω R , and Ω R is the center frequency of a Raman line with bandwidth Γ. Hence, the CARS signal can be written as As can be seen from Eq. 3, the CARS intensity is determined by three terms: the first term is known as the non-resonant  background and is independent of the Raman shift. The second term is the dominant contribution to probing strong resonant scatterers. The third term includes a mixture of resonant and non-resonant contributions, which contains the real part of the vibrational response.

CARS MICROSCOPY
Two pulsed laser trains at different frequencies, i.e., ω p and ω s , are necessary to build a CARS microscope ( Figure 2A). One beam of the lasers is tunable to adjust the desired Raman shift, ω p − ω s . A pair of mirrors on a motorized linear stage is used to adjust the time delay between the pump and Stokes beams. Expanded beams are combined in a collinear geometry, and then directed to an inverted or upright microscope. Finally, the two temporally and spatially overlapped beams are focused onto the sample using a water/oil immersion objective lens. The strong CARS signals are generated both in the forward and backward direction, and are referred to as F-CARS and E-CARS, respectively ( Figure 2B). Each signal is collected by a photomultiplier tube (PMT) or an avalanche photodiode after filtering the useless fluorescence with a bandpass filter, and focused by a lens. CARS images are obtained by a laser scanning system similar to that of a laser scanning confocal fluorescence microscope.
The cost and complexity of coherent Raman microscopy are mainly due to the laser light source. The two excitation beams must accurately control the absolute wavelength and linewidth. Although a pulsed dye laser in the visible band was used in the first coherent Raman microscopy system in history [11], the current mainstream light source for CARS is the Ti: sapphire laser or optical parametric oscillator (OPO) [17,21]. Although the newly developed fiber laser-based coherent Raman microscopy can greatly reduce cost, it still cannot surpass solid-state lasers in performance [22,23].
At least two excitation light sources are required for coherent Raman microscopy, one of which must be tunable to match the Raman frequency difference of different molecular vibration frequencies. For laser pulse width, a balance must be established; femtosecond has a shorter pulse width, higher peak power, and stronger nonlinear effect, while picosecond laser has narrower linewidth (l-10 cm −1 for a similar molecular vibration width) and can excite coherent Raman scattering more efficiently. Experiments show that pulses of 1-10 picoseconds can ideally achieve this balance [24]. For high-speed imaging, a repetition rate of at least 10 MHz is required, and ideally a higher rate. In addition, the laser excitation of the spectrum in the near infrared region can minimize the generation of the non-resonant background in CARS [12], providing less light-induced damage than visible light, and deeper penetration into the tissue. Further, because near-infrared light has a low transmission efficiency in a coherent Raman microscopy system (typically 10%-20% from excitation light to the sample), the power of the excitation source is approximately at the milliwatt level. Several groups have reported that they employed 50-200 fs pulse widths rather than a 2-6 picosecond pulse width [22,[25][26][27]. In spite of the fact that coherent Raman scattering processes could be excited by femtosecond pulses, this resulted in decreased signal levels, broader spectral resolution, and higher intensity non-resonant background signals.
CARS microscopy offers many unique advantages, including 1) nondestructive molecular imaging without labeling, 2) ease of spectral separation from the single-photon fluorescence background, and 3) 3D sectioning capability. Despite all of its advantages, however, the non-resonant background that limits the sensitivity of CARS is a major disadvantage.

MULTICOLOR CARS MICROSCOPY
Multicolor CARS microscopy refers to a method that can detect all or part of the vibration spectrum information on a sample at the same time. Multicolor CARS microscopy can be understood as a technique that can acquire a certain vibration spectrum such as a spontaneous Raman spectrum, but which has good imaging speed and spatial resolution. Although the imaging speed of multicolor CARS microscopy is still not as fast as that of single-color CARS, it has two essential advantages. The first is that multi-color CARS can obtain a full range of spectral information, while single-color can only obtain information at a certain frequency. The second is that multicolor CARS can better achieve quantitative analysis. Most microscopy techniques endeavor to achieve qualitative analysis at a certain point of the sample; however, in the fields of biology and material science, quantitative analysis of the composition of the sample is often required.
There are several common methods of implementing multicolor CARS microscopy, as shown in Figure 3. Figures  3A shows multicolor CARS microscopy using picosecond pump and femtosecond Stokes pulses [28,29], in which a 100-fs Stokes pulse is used to provide a frequency domain window of approximately 400 cm −1 , which is sufficient to obtain the spectrum of the entire CH stretching range. In addition, by adjusting the center wavelength of the Stokes pulse, other regions of the vibration spectrum can also be studied, although not simultaneously. Using a 10-ps pump pulse can achieve a high spectral resolution of 2-3 cm −1 . Figures 3B shows multicolor CARS microscopy using picosecond pump and supercontinuum Stokes pulse [30][31][32][33][34][35][36][37]. Although a femtosecond Stokes pulse can include 400 cm −1 of spectral information in one-shot, this is still not sufficient for a vibrational frequency domain of 4,000 cm −1 . With the rapid development of photonic crystal fiber-based supercontinuum light sources, it has become feasible to use supercontinuum light sources as Stokes sources for CARS microscopy and to achieve a window of 4,000 cm −1 . Femtosecond and nanosecond sources are used to generate the supercontinuum spectrum. Figures 3C shows multicolor CARS using a femtosecond pump and Stokes pulse with spectral focusing [30,[38][39][40][41][42][43][44], and Figures 3D shows the principle of the spectral focusing method. In general, by using broadband pump and broadband Stokes pulses, the spectral resolution of the CARS signal is reduced. Hellerer [38] proposed the method of spectral focusing in 2004, Frontiers in Physics | www.frontiersin.org December 2020 | Volume 8 | Article 598420 which can obtain a high spectral resolution when excited by a broadband light source. In this method, a dispersive medium is used to linearly chirp the light source. Although the linewidth of the chirped source remains the same, the instantaneous linewidth of the interaction between the two beams becomes very narrow. This instantaneous linewidth determines the spectral resolution of the system. The desired CARS spectrum can be obtained easily and quickly by adjusting the time delay between the pump and Stokes beams. In addition to being ingenious, this method has the advantage of being simple and fast. The disadvantages are the same as that of the ps-pump-fs-Stokes pulse method, i.e., it is not feasible to obtain a wide range of spectral signals simultaneously.
Over the past few years, it has become possible to study complex samples obtaining more detailed information for both biological and material science samples by multiplex CARS microscopy. Multiplex CARS microscopy confers the two key advantages of the full spectrum being available, and of being quantitative. The full spectrum is essential when studying complex samples with many overlapping peaks and the quantitative nature of the spectrum makes multiplex CARS microscopy a particularly useful technique for probing such complex samples. The extremely rich information obtained from imaging and spectral information constitutes a hyperspectral dataset. With the advent of phase retrieval techniques such as the maximum entropy method (MEM) and time-domain Kramers-Kronig (TD-KK) transformation, such spectra become amenable to analysis by multivariate techniques such as principal component analysis and related approaches. Therefore, the different components in the sample can be distinguished based on their vibrational information in an unlabeled manner, and the sensitivity of each component can be increased to the mM-μM level by using electronic resonance enhancement. Coupled with rapid spectral acquisition, these sophisticated algorithms should make it possible for researchers to investigate phenomena in the biological and material science fields that manifest as only subtle changes in the vibrational spectrum and occur on sub-micrometer length scales at sub-ms timescales.

APPLICATIONS OF CARS MICROSCOPY
Cell Imaging CARS microscopy has been widely used for cell imaging [36,[45][46][47][48][49][50][51][52][53]. The strong CARS signal in the C-H stretching region can provide rich and unexpected chemical information, such as that related to lipids and proteins. For example, Nan et al. [54] used the CH 2 stretching vibration to map neutral lipid droplets in living fibroblasts. There was a significant process of disappearance of the lipid droplets in the cells in the early stage of differentiation, after which differentiation and accumulation began again. Paar et al. [55] used CARS microscopy to study the metabolism of intracellular lipid droplets during lipid decomposition. Experiments have shown that lipid droplet growth is achieved through the transfer of lipids between one organelle and another. This lipid transfer is not a rapid spontaneous process but completed within a few hours. Experimental data show that the growth of lipid droplets is a highly regulated process, and the decomposition and production of lipids occur in parallel in cells to prevent cell acid overflow. In another work [56], cancer cells in vivo and in vitro were monitored by CARS microscopy to assess the effect of diet on the metastasis of cancer cells. The results show that the physical disturbance of free fatty acids to the cancer cell membrane in lipid rich tumors is related to tumor metastasis. CARS has also been used to study prostate circulating carcinoma cells, and the results show that prostate circulatory carcinoma cells have a CARS signal intensity seven times stronger than leukocytes [57]. The absorption and decomposition of surfactants in mammalian cells were studied by CARS spectroscopy [58]. The results show that living cells retain surfactants not only on the cell membrane but also throughout the intracellular membrane system. Label-free chemical mapping of cells by hyperspectral CARS microscopy has led to promising results. In another work, Di Napoli [59] fed different ratios of Palmitic acid and α-linolenic acid to human adipose derived stem cells and the results showed good contrast between cellular components and the corresponding CARS spectra.

Tissue Imaging
CARS has also been posited as a unique tool for label free tissue imaging. Visualization and characterization of the myelin sheath in the peripheral nervous system [61,62] and central nervous system [63][64][65][66][67] is an important application. The sciatic nerve is a perfect candidate for in vivo CARS imaging since it is the largest single nerve in mammals and located in the peripheral nervous system isolated from the movement induced by breathing and the heartbeat. By 3D imaging of fat cells that surround the nerve, the underlying contrast mechanisms of in vivo CARS are found to arise from both interfaces and back reflection of the forward CARS signal [61]. Longitudinal in vivo CARS imaging of the spinal cord has been demonstrated through a careful experimental design to overcome complexities induced by laminectomy surgery [64]. To reduce the amount of tissue to be removed and improve the survivability and recovery time of the animals, the spine was exposed by making an incision through the skin and muscle tissue at T10, where the natural curvature of the spinal cord makes this region more superficial to the skin than at other locations [64]. The myelin sheath [63], myelinated axons [61], and the node of Ranvier [64] were distinguished clearly by CARS microscopy, as shown in Figures 4B left, middle, and right, respectively. The lipid metabolism in the atherosclerotic tissue of mouse and human subjects has been studied by CARS microscopy. Employing picosecond lasers, Wang et al. explored the imaging capabilities of CARS-based multimodal microscopy imaging of atherosclerosis. The study first demonstrated the capability to identify different atherosclerotic types [68] based on the scheme of atherosclerosis classification suggested by the American Heart Association (AHA) by imaging. Furthermore, Wang et al. [68] showed that a multimodal approach employing CARS and sum-frequency generation (SFG) signals allows quantitation of collagen and lipid content in lesions from early to advanced stages. Additional studies have shown increasing interest in using CARS microscopy for the studies of atherosclerotic lesions in animal models. Lim et al. utilized multimodal NLO microscopy to quantitatively measure the impact of a high-fat, high-cholesterol Western diet on the composition of atherosclerotic plaques in ApoE-deficient mice [69]. Using photonic crystal fiber-based CARS microscopy, the capability of multimodal imaging was demonstrated by an alternative femtosecond system to study lesion development in myocardial infarction-prone, hyperlipidemic rabbits [70]. Moreover, CARS has been used to skin biopsy [17,[71][72][73][74] and for stain-free histopathology. An imaging platform based on broadband coherent anti-Stokes Raman scattering (BCARS)

Material Science
CARS microscopy lends itself well to the study of polymer films, thanks to the intrinsic sectioning capability of CARS and the intensity of the polymer peaks. Von Vacano et al. [35] were able to discriminate between the different polymers in tertiary blends of polystyrene (PS), polyethylene terephthalate (PET) and polymethylmethacrylate (PMMA), as well as to construct a virtual section of an adhesive and its carrier film, as shown in Figure 5. Jeoung group has also used the sectioning capability of multiplex CARS microscopy to quantitatively determine the thickness of sub-micron polymeric films [76,77]. CARS microscopy has also been used to study reactions in zeolite crystals. CARS is an ideal tool to study the interplay between catalyst architecture and reactivity and hence obtain previously inaccessible information about fundamental processes of catalysts. Specifically, the precursor states of thiophene conversion over individual H-ZSM-5 zeolite crystals have been studied [78]. Maps of the catalyst loaded with thiophene revealed a heterogeneous, diffusion-limited distribution of thiophene throughout the zeolite, with the analytes accumulating at the center of the crystal and along defect sites.
More recently, Kotaro [79] developed a high-throughput flow cytometry system based on a Fourier-transform CARS spectrometer, enabling a high throughput of 2000 events/s. High chemical specificity and classification accuracy without labeling is demonstrated by the differentiation of polymer beads and Euglena gracilis. Super-resolution vibration imaging by higher-order CARS microscopy that detects higher-order antistokes Raman signals, such as six-wave mixing (SWM) at ω SWM 3ω p − 2ω S and eight-wave mixing (EWM) at ω EWM 4ω p − 3ω S , increasing the respective lateral resolutions to 230 and 196 nm is reported by Li [80]. To increase the signal-to-noize radio and imaging speed, the wide-field CARS implementation was improved. Duarte [81] developed wide-field Fourier-transform coherent anti-Stokes Raman scattering microscopy with highcontrast, and outline detail of the wide-field FT-CARS signal. Heuke [82] implemented wide-field CARS by using a rotating disk with an imprinted modulation pattern and a PMT detector, referred to as spatial frequency modulated imaging (SPIFI). The results demonstrate that SPIFI-CARS could detect a much larger number of photons than that of laser-scanning microscopy (LSM)-CARS, and further, could be operated in ambient light.

OUTLOOK
Although the technological development of CARS is close to maturity, its application has just begun; however, the results are encouraging. Although CARS has broad application prospects, it will not replace other imaging methods in all fields. Because the cost of the spontaneous Raman system is relatively low and the system is relatively simple, it will likely be chosen when imaging speed is not a priority or there are technical difficulties (such as sample environment) to apply the coherent method. However, if the sample does not have second harmonic generation (SHG) characteristics, and fast acquisition is important, CARS or coherent Raman microscopy is the best label-free optical imaging method.
The clinical translation of CARS microscopy has long been hindered by traditional solid-state lasers that are sensitive to the environment. Ultrafast fiber lasers can potentially overcome these shortcomings but have not yet been fully exploited for CARS microscopy, as previous implementations have suffered from high intensity noise, a narrow tuning range, and low power, resulting in low image quality and slow imaging speed [23,[83][84][85][86][87][88][89][90]. Therefore, high-performance fiber lasers are highly required.
Another recent trend has been focusing on label-free multimodal nonlinear microscopy based on CARS, twophoton excited fluorescence (TPEF), and SHG [91][92][93][94]. The coupling of two-photon fluorescence, second-harmonic generation, and coherent anti-Stokes Raman scattering has allowed for investigation of a broad range of biological phenomena concerning lipid metabolism, cancer development, cardiovascular disease, and skin biology. By integrating the strengths of each non-linear optical (NLO) imaging modality, different structures and their interactions in a complex biological system can be simultaneously visualized. Overall, as a label-free optical imaging method, CARS has a bright future with the development of other technologies such as laser light source and detection devices.

AUTHOR CONTRIBUTIONS
SL, LL and JQ wrote and edited the manuscript. SL, YL, RY, LL and JQ conceived the work and organized the manuscript.