Lipid droplets (LDs) are cellular organelles that are highly specialized; they contain a hydrophobic core of neutral lipids, triacylglycerol and/or sterol ester, enclosed by a single-layer membrane composed of phospholipids (Thiam and Ikonen, 2021) (Fig. 1A). LDs are vital hubs of cellular metabolism, as they are not only important for lipid storage, but also take part in the regulation of lipid synthesis, incorporation and breakdown, as well as lipid signaling (Bustos and Partridge, 2017; Jarc and Petan, 2020). Owing to the hydrophobic nature of the lipid core, LD-associated proteins predominantly anchor to the phospholipid membrane (Thiam and Dugail, 2019; Zhang and Liu, 2019). These proteins are crucial for the maintenance of the morphology and functionality of LDs, and they regulate lipid incorporation and lipolysis at LDs; they also mediate the interaction of LDs with other organelles and contribute to lipid signaling (Cui and Liu, 2020; Herker et al., 2021; Zhang and Liu, 2019). The current model of LD formation suggests that free fatty acids are firstly esterified and accumulate within the intermonolayer leaflets of endoplasmic reticulum (ER) to form lipid lenses, which further grow and then bud off from the ER to become droplets in the cytosol (Olzmann and Carvalho, 2019). These newly released small droplets can expand to large droplets through incorporation of more neutral lipids and coalescence with each other (Fig. 1A). The turnover of LDs is closely correlated with metabolic status. In lipolysis, lipids are hydrolyzed to release free fatty acids that can be utilized in mitochondrial and peroxisomal β-oxidation to generate energy, in phospholipid biogenesis as building blocks and in signal transduction as lipid messengers (Lass et al., 2011; Walther and Farese, 2012). With the mobilization of the lipid core, LDs reduce their size, decrease contacts with other organelles and the excess proteins on the surface are degraded (Lass et al., 2011; Olzmann and Carvalho, 2019; Thiam and Dugail, 2019). In recent years, there has been recognition that the roles of LDs greatly extend beyond them just being inert lipid reservoirs, and we now know they are an active platform for organelle crosstalk and cellular signaling (Gao and Goodman, 2015) (Fig. 1A). LD homeostasis is vital for organism health, and its disruption leads to a variety of metabolic diseases, such as obesity, hepatic steatosis and lipoatrophy (Walther and Farese, 2012). Importantly, the functionality of LDs is tightly linked with their size, composition and spatiotemporal distribution within the cell and organism. Methods to quantitatively analyze these different LD parameters at high spatial resolution in a physiologically relevant condition are crucial for understanding the regulation of LDs and their contribution to health and diseases.

Fig. 1.

Coherent Raman scattering microscopic imaging of LDs. (A) Lipid droplets (LDs) store cholesterol ester (CE) and triacylglycerol (TAG) in the core, which is enclosed by a single layer of phospholipid membrane into which various proteins are anchored. The formation of LDs starts with fatty acid esterification and accumulation in the endoplasmic reticulum (ER). Upon budding off from the ER, small LDs form larger ones through lipid incorporation and protein recruitment and/or coalescing with each other. LDs undergo dynamic interactions with other organelles. (B) Principle of spontaneous Raman and coherent Raman scattering. In spontaneous Raman scattering (SRS), a small fraction of pump photons will be converted to photons with increased energy (anti-Stokes photons) or decreased energy (Stokes photons). In coherent Raman scattering (CRS), the synchronized pump and Stokes photons can generate enhanced anti-Stokes fluorescence and transfer energy between each other mediated by the molecules. CARS, coherent anti-Stokes Raman scattering; SRL, stimulated Raman loss; SRG, stimulated Raman gain, CSRS, coherent Stokes Raman scattering. (C) An example of instrumentation for coherent Raman scattering microscopes. Synchronized and overlapped pump and Stokes lasers are directed into a laser scanning microscope. The intensity of the Stokes laser is modulated by an electro-opto modulator (EOM). The combined beam is focused on the sample, and the reflected anti-Stokes signal is collected and detected by a photo-mulitplier tube (PMT). The transmitted lasers are filtered; only the pump beam will transmit and reach the photodiode. The photo-electric current is sent to a lock-in amplifier (LIA) for demodulation and signal amplification. DL, delay line; FG, function generator. (D) Representative SRS images of LDs in various samples. From the left to the right are: cultured HEK293 cells (red for SRS and cyan for YFP labeled LSDP5) (Wang et al., 2011), fat body of D. melanogaster (Yu et al., 2014), C. elegans, and mice liver tissue section (Yu et al., 2014). Scale bar: 10 µm for all images. Images are adopted with permission from references.

Fig. 1.

Coherent Raman scattering microscopic imaging of LDs. (A) Lipid droplets (LDs) store cholesterol ester (CE) and triacylglycerol (TAG) in the core, which is enclosed by a single layer of phospholipid membrane into which various proteins are anchored. The formation of LDs starts with fatty acid esterification and accumulation in the endoplasmic reticulum (ER). Upon budding off from the ER, small LDs form larger ones through lipid incorporation and protein recruitment and/or coalescing with each other. LDs undergo dynamic interactions with other organelles. (B) Principle of spontaneous Raman and coherent Raman scattering. In spontaneous Raman scattering (SRS), a small fraction of pump photons will be converted to photons with increased energy (anti-Stokes photons) or decreased energy (Stokes photons). In coherent Raman scattering (CRS), the synchronized pump and Stokes photons can generate enhanced anti-Stokes fluorescence and transfer energy between each other mediated by the molecules. CARS, coherent anti-Stokes Raman scattering; SRL, stimulated Raman loss; SRG, stimulated Raman gain, CSRS, coherent Stokes Raman scattering. (C) An example of instrumentation for coherent Raman scattering microscopes. Synchronized and overlapped pump and Stokes lasers are directed into a laser scanning microscope. The intensity of the Stokes laser is modulated by an electro-opto modulator (EOM). The combined beam is focused on the sample, and the reflected anti-Stokes signal is collected and detected by a photo-mulitplier tube (PMT). The transmitted lasers are filtered; only the pump beam will transmit and reach the photodiode. The photo-electric current is sent to a lock-in amplifier (LIA) for demodulation and signal amplification. DL, delay line; FG, function generator. (D) Representative SRS images of LDs in various samples. From the left to the right are: cultured HEK293 cells (red for SRS and cyan for YFP labeled LSDP5) (Wang et al., 2011), fat body of D. melanogaster (Yu et al., 2014), C. elegans, and mice liver tissue section (Yu et al., 2014). Scale bar: 10 µm for all images. Images are adopted with permission from references.

Traditional biochemical approaches, including chromatography and mass spectrometry, have been applied to LDs extracted from cells, tissues or whole organisms. These quantitative analyses have been employed to profile the composition of neutral lipids in the core and phospholipids in the membrane of the extracted LDs (Ding et al., 2013). However, these approaches lose spatial information regarding LD distribution and heterogeneity. Furthermore, transmission electron microscopy has been exploited to image LDs in situ, and its ultrahigh resolution has successfully revealed nascent LDs as small as 50 nm (Ohsaki et al., 2014; Wang et al., 2016). Despite the ultrahigh resolution, identification of LDs smaller than 50 nm is not trivial, because they can be easily confused with other small cellular structures, such as endocytic vesicles. Imaging mass spectrometry can map the spatial distribution of a large number of lipid species, but the resolution is far less than what is needed for LDs, and it cannot probe samples under physiological conditions (Djambazova et al., 2020). Fluorescence microscopy is powerful for LD visualization in vivo through coupling with lipophilic fluorescent dyes and/or fluorescent fusions of LD-associated proteins (Brasaemle et al., 1997; Fam et al., 2018; Kuerschner et al., 2005, 2008; Listenberger and Brown, 2007). In addition, label-free imaging approaches, which require no exogenous molecules for labeling, further advance LD imaging at the most physiological conditions. Third-harmonic generation utilizes the third-order susceptibility inhomogeneity within the excitation volume to probe LDs (Débarre et al., 2006). Quantitative phase imaging (QPI) makes use of the refractive index to reconstruct cellular components contours. Optical diffraction tomography, a technology based on the principle of QPI, provides fast non-invasive imaging of cellular LDs in three dimensions (Kim et al., 2016). However, despite their advantage in tracking LDs in vivo with high spatial resolution, these imaging methods are often non- or semi-quantitative and lack chemical specificity to reveal LDs with different lipid composition.

Vibrational spectroscopic approaches, including infrared (IR) and Raman scattering, acquire contrast by probing the energy exchange between incident photons and the vibration of the chemical bonds. Through targeting LDs by the spectral signature of the abundant aliphatic hydrocarbon in fatty acids, vibrational spectroscopy and microscopy offer imaging methods to identify LDs with chemical specificity. Compared with IR, which is based on absorption, Raman scattering is intrinsically free from large positive background signals, and it has less requirement on excitation wavelength, permitting higher spatial resolution with shorter excitation wavelengths in mapping (Jones et al., 2019). However, Raman scattering suffers from low signal intensity, therefore requiring long acquisition time, which limits its application in imaging live samples. Through coherent excitation and nonlinear processing, coherent Raman scattering (CRS) microscopy is able to enhance the Raman-scattering signal by several orders of magnitudes, enabling label-free fast imaging of LDs with high spatiotemporal resolution and chemical specificity (Cheng and Xie, 2015; Yu et al., 2014). In this Review, we will focus on CRS microscopy and introduce the principle behind its use to image LDs. We will also provide an overview of its recent biological applications in the discovery of new regulatory mechanisms of LDs and medical applications to understand disease pathology.

CRS microscopy utilizes the same signal as spontaneous Raman scattering, which originates from the interaction between photons and chemical bond vibrations. In spontaneous Raman scattering, the vibrational energy of molecules, the energy acquired from relative motion between atomic nuclei, can be exchanged with incident photons. Consequently, a small fraction of photons undergoes frequency shifts after the energy exchange with a chemical bond. The frequency that is shifted towards the lower-energy end of the spectra is called the Stokes Raman scattering, and the frequency shifted towards the higher-energy end of the spectra is called the anti-Stokes Raman scattering (Camp and Cicerone, 2015; Zumbusch et al., 2013) (Fig. 1B). Correspondingly, the vibrational energy level of the chemical bond transits from the ground state to the excited state in the Stokes Raman scattering or vice versa in the anti-Stokes Raman scattering. The energy gain or loss of the photon equals the energy difference between the two vibrational states of the chemical bond, which is called the Raman shift. Since the Raman shift is a characteristic feature of a chemical bond, the detection of these frequency-shifted photons will reveal the presence of a specific chemical bond in a molecule. A spectrum of these frequency-shifted photons from Raman scattering, which is called the Raman spectrum, provides information on the composition and structure of chemical groups in a sample without any labeling. However, microscopic detection based on spontaneous Raman scattering experiences a natural drawback, in that the chance of a scattering event to happen in a Raman process (Nagli et al., 2008; Tian et al., 2007) is more than ten orders of magnitude lower than that of a fluorescence event (Lincoln et al., 2012), therefore, limiting the practical use of Raman scattering for live imaging.

Coherent excitation overcomes the limitation of spontaneous Raman scattering and enhances the signal up to eight orders of magnitude in a focused excitation volume, therefore making it sufficient for real-time imaging (Evans and Xie, 2008; Min et al., 2011). Coherent excitation requires incident lights of two different wavelengths, λp (the pump light) and λS (the Stokes light). When the frequency difference between the pump light (λp) and the Stokes light (λS) equals the vibrational energy of a chemical bond, the beating frequency will excite the chemical bonds to vibrate in a coordinated manner. The coherently excited molecules will carry out two types of processes amid nonlinear interaction with the light. The first type is the parametric process, in which lights of new wavelengths are generated but no photonic energy is transferred to the molecules. It includes the generations of anti-Stokes photons in coherent anti-Stokes Raman scattering (CARS) and Stokes photons in coherent Stokes Raman scattering (CSRS). They are the results of scattering of pump or Stokes photons by the coherently excited molecules. CARS is the dominant mechanism of the parametric process adopted in CRS microscopy. The second type is the dissipative process, in which no new wavelengths of lights are generated but a share of photonic energy is transferred to the molecules. In the dissipative process, the intensity of the pump light (λp) and Stokes lights (λS) will change in stimulated Raman scattering (SRS) (Woodbury and Ng, 1962) due to the energy conversion from the pump photon to the Stokes photon coupled with the coherent excitation of chemical bonds. In SRS, the pump and Stoke light experiences a power loss and gain, called stimulated Raman loss (SRL) and stimulated Raman gain (SRG), respectively. SRL is mostly adopted in the home-built SRS microscope. All of the above processes happen simultaneously upon resonance when the frequency difference between the pump and Stokes photons meets the vibrational energy of specific chemical bonds. Thus, both CARS and SRS signals can be used to probe molecules based on their chemical specificity, but there are important differences between them. First, CARS and SRS suffer from different backgrounds. In CARS, a non-resonant background is the major contamination to the true signal. The non-resonant background originates from nonlinear electronic responses to excitation lights, therefore presenting regardless of the frequency difference between the pump and Stokes lights. The non-resonant background can distort the Raman spectrum of the target. Various methods have been developed to eliminate the non-resonant background (Ganikhanov et al., 2006; Pope et al., 2013; Bradley et al., 2016; Masia et al., 2013). SRS is frequently hindered by a background originating from other nonlinear optical processes, which can be removed using polarization encoding (Hill et al., 2019). In general, SRS obtains a spectrum of the sample that is identical to its spontaneous Raman spectrum, offering a simple straightforward way of quantification, whereas in CARS, the non-resonant background can distort the spectrum and complicate the quantification process. However, with the effective removal of the non-resonant background using well-designed algorithms, the recovered Raman spectra offer a quantification capability comparable to SRS (Bradley et al., 2016; Masia et al., 2013). Secondly, CARS and SRS signals show different dependence on local molecule concentration. Ideally, CARS has a quadratic dependence on molecule concentration (Evans and Xie, 2008); however, the actual quantitative relationship can be shifted to in between linear and quadratic due to the non-resonant background. SRS, in contrast, presents a linear relationship between signal intensity and molecule concentration (Freudiger et al., 2008) (Fig. 2B). As two most commonly used approaches in CRS microscopy, the basis of CARS and SRS microscopes will be explained in the next section.

Fig. 2.

SRS imaging of different LD parameters. (A) Representative SRS spectrum of dipalmitoylphosphatidylcholine (DPPC), a compound used to represent saturated lipid in research, and oleic acid (OA), showing the aliphatic hydrocarbon peak at 2845 cm−1. (B) The standard curve of palmitic acid concentrations in DMSO solution demonstrates a linear concentration dependence (Yu et al., 2014). (C) Quantification of fat content levels using SRS, colorimetric biochemical assay (Biochem) and gas chromatography–mass spectrometry (GC-MS) analysis, yielding comparable results. Data from Mutlu et al. (2020) and Wang et al. (2011), and are shown as mean±s.d. (n=5–20 worms for SRS, n=15,000 worms for biochemistry and GC-MS). **P<0.01, ***P<0.001 (one-way ANOVA). (D) Application of SRS hyperspectral imaging to distinguish LDs containing cholesterol ester versus triacylglycerol. Left, SRS hyperspectra of individual synthetic LDs containing cholesterol ester or triacylglycerol showing visible differences at 3015 cm−1 and 2965 cm−1 (indicated by arrows) and how they are separated into two groups. Middle, the histogram of the signal ratio for each pixel between images at 3015 cm−1 and 2965 cm−1 (R3015/2965) showing two peaks that correspond to either 100% of cholesterol ester or triacylglycerol. Right, image showing individual synthetic LDs containing either cholesterol ester (green arrows) or triacylglycerol (blue arrowheads). Scale bar: 20 µm. (E) Use of deuterated fatty acids to assess lipid dynamics. Left, SRS hyperspectra of deuterated palmitic acid (PA-D31) and oleic acid (OA-D34) showing the signal of D–C–D bonds at 2110 cm−1. Middle, upon 24-h labeling in C. elegans, PA-D31 signals are much lower than OA-D34 signals in LDs, which are visualized by SRS at 2110 cm−1. Scale bar: 20 µm. Right, the ratio between deuterium-labeled lipids and unlabeled total lipids (H-C-H 2850 cm−1) in LDs is quantified and compared between PA and OA at both 12 h and 24 h, revealing a difference in incorporation rate between different fatty acids. Date are mean±s.d. (n=400–500 LDs). ***P<0.001 (one-way ANOVA). Graphs and images in panels D and E adapted with permission from Fu et al. (2014) (https://pubs.acs.org/doi/10.1021/ja504199s; further permission to reuse this figure should be directed to ACS). a.u., arbitrary units

Fig. 2.

SRS imaging of different LD parameters. (A) Representative SRS spectrum of dipalmitoylphosphatidylcholine (DPPC), a compound used to represent saturated lipid in research, and oleic acid (OA), showing the aliphatic hydrocarbon peak at 2845 cm−1. (B) The standard curve of palmitic acid concentrations in DMSO solution demonstrates a linear concentration dependence (Yu et al., 2014). (C) Quantification of fat content levels using SRS, colorimetric biochemical assay (Biochem) and gas chromatography–mass spectrometry (GC-MS) analysis, yielding comparable results. Data from Mutlu et al. (2020) and Wang et al. (2011), and are shown as mean±s.d. (n=5–20 worms for SRS, n=15,000 worms for biochemistry and GC-MS). **P<0.01, ***P<0.001 (one-way ANOVA). (D) Application of SRS hyperspectral imaging to distinguish LDs containing cholesterol ester versus triacylglycerol. Left, SRS hyperspectra of individual synthetic LDs containing cholesterol ester or triacylglycerol showing visible differences at 3015 cm−1 and 2965 cm−1 (indicated by arrows) and how they are separated into two groups. Middle, the histogram of the signal ratio for each pixel between images at 3015 cm−1 and 2965 cm−1 (R3015/2965) showing two peaks that correspond to either 100% of cholesterol ester or triacylglycerol. Right, image showing individual synthetic LDs containing either cholesterol ester (green arrows) or triacylglycerol (blue arrowheads). Scale bar: 20 µm. (E) Use of deuterated fatty acids to assess lipid dynamics. Left, SRS hyperspectra of deuterated palmitic acid (PA-D31) and oleic acid (OA-D34) showing the signal of D–C–D bonds at 2110 cm−1. Middle, upon 24-h labeling in C. elegans, PA-D31 signals are much lower than OA-D34 signals in LDs, which are visualized by SRS at 2110 cm−1. Scale bar: 20 µm. Right, the ratio between deuterium-labeled lipids and unlabeled total lipids (H-C-H 2850 cm−1) in LDs is quantified and compared between PA and OA at both 12 h and 24 h, revealing a difference in incorporation rate between different fatty acids. Date are mean±s.d. (n=400–500 LDs). ***P<0.001 (one-way ANOVA). Graphs and images in panels D and E adapted with permission from Fu et al. (2014) (https://pubs.acs.org/doi/10.1021/ja504199s; further permission to reuse this figure should be directed to ACS). a.u., arbitrary units

With decades of development, numerous configurations of CRS microscopes have been developed. A couple of excellent reviews (Zhang and Cheng, 2018; Zumbusch et al., 2013) have covered these different schemes. Here, we give an example and discuss several major considerations related to the example, including the use of an ultrafast laser system for excitation, a scanning microscope for image acquisition and a series of electronic devices for signal retrieval (Fig. 1C). An ultrafast laser system that can provide two excitation wavelengths is usually utilized, specifically with one tunable beam. As a critical step to obtain an optimal CRS signal, the two beams need to be focused to the same focal point and have their pulses arrive at the same time. Careful alignment and the use of an optical delay line can ensure both requirements are met. The choice of pulse width is critical for effective coherent excitation. Once the spectral span of excitation pulse is close to the targeted Raman resonance, the efficiency will be optimal (Zumbusch et al., 2013). In biological samples, most Raman peaks obtain a spectral span of ∼10 cm−1, therefore constraining the width of the pulses to be used. Such a pulse width can be achieved by using either picosecond lasers or femtosecond lasers that are shaped to meet the requirement of the spectral resolution (discussed below).

The choice of a laser-scanning microscope depends on whether its optimal transmission band covers the near infrared region. The detector of a CARS microscope captures the emission of anti-Stokes fluorescence. Given the blue-shifted wavelength of anti-Stokes photons, this configuration follows the example of a multiphoton fluorescence microscope, in which the emission fluorescence is separated from excitation light using a short-pass dichroic mirror and selected by a fluorescence filter before detection. In the case of CARS, a dichroic mirror and a filter are specifically chosen based on the wavelength of emitted anti-Stokes fluorescence (Zumbusch et al., 1999). By contrast, the detection of SRS signals follows a completely different and unique strategy since no photons of a new wavelength are generated. In the SRS process, the pump and the Stokes lasers experience an intensity loss and gain, respectively. To probe this pump or Stokes laser intensity change, a high frequency modulation transfer scheme is employed (Min et al., 2011). The intensity of one beam of excitation lasers (usually Stokes) is modulated at a high frequency f. Such a periodic presence of modulated beam results in a periodic switch-on of the SRS process. Therefore, the energy loss or gain in the other beam is modulated at the same frequency f. This small periodic intensity variation can then be demodulated and amplified by a lock-in amplifier that is referenced at the frequency f. Overall, CARS and SRS can utilize the same laser and microscope systems and be integrated to work simultaneously (Freudiger et al., 2008).

For a complex biological system, the existence of diverse chemical groups and their complex combination require CRS imaging at many different Raman peaks. To address this challenge, recent advances have been made, including multiplexing in CARS (Camp et al., 2014; Ideguchi et al., 2013; Pope et al., 2013) and SRS (Freudiger et al., 2011, 2014; Fu et al., 2012, 2013; Ozeki et al., 2012; Zhang et al., 2017a), which allow efficient image acquisition covering a wide range of Raman peaks. To build a CRS microscope with spectrum-scanning capability, the key technical requisite is a rapid switch of the resonant beating frequency of the pump and Stokes lasers. Conventionally, a change of the targeted Raman peak is achieved by tuning the wavelength of either the pump or Stokes beam. Such tuning mechanism involves a temperature change of nonlinear crystals, which limits the tuning speed and therefore its application in following fast in vivo dynamics of biological systems (Kong et al., 2013). In newly developed approaches, femtosecond lasers with broad spectral bandwidth are exploited to enable rapid wavelength changing. Different methods have been developed to achieve this goal (Zhang and Cheng, 2018; Zumbusch et al., 2013). The first method, and also the most straightforward one, is combining femtosecond and picosecond pulses directly and generating different beating frequencies in the femtosecond pulse (Camp et al., 2014; Zhang et al., 2017a). Once there are chemical bonds in resonance, both CARS and SRS processes will be activated at the corresponding frequency components in the femtosecond pulse. The spectral SRS signal can be detected by an array of photodiodes after different frequency components are separated by dispersion. Meanwhile, the generated broadband CARS signal can be detected by a spectrometer. The second method performs selective filtering on the femtosecond laser such that only the desired Raman peaks are excited (Freudiger et al., 2011; Fu et al., 2012). The third method utilizes two femtosecond lasers that are chirped, and the excitation of a specific Raman peak is determined by the temporal overlap between the two pulse trains, which can be achieved with an optical delay line. The spectrum of a sample is obtained by sequential image acquisition at different delays (Fu et al., 2013; Pope et al., 2013). The final approach is to sweep the laser source wavelength for fast switching of Raman resonance (Ozeki et al., 2012). A comparison between these different technology approaches is provided in Table 1.

Table 1.

Comparison between different coherent Raman scattering microscopy approaches

The crucial impact of lipids on metabolic health is associated with their quantity, composition, distribution and dynamics. Emerging advances of CRS microscopy bring tremendous opportunity for the imaging, quantification and tracking of lipids. Various methods exploiting the advantages of CRS microscopy have been developed to facilitate different purposes in lipid research. Here, we will present four major applications, including in vivo quantification of the levels of total lipid content, visualization of lipid distribution and differentiation of lipid classes, as well as tracking of LD mobility. We will also discuss the potential of multimodal imaging systems that harness the power of both CRS and fluorescence microscopy.

Lipid quantity is a major parameter that correlates with different metabolic status, and can be influenced by both genetic and environmental factors. A systemic increase in lipid quantity can lead to obesity, while a systemic decrease results in lipodystrophy, which both are major risk factors of diverse metabolic and chronic morbidity, such as type II diabetes, cardiovascular diseases and cancer (Butler et al., 2020; Nordestgaard and Varbo, 2014; Taskinen, 2005). The ability to quantify lipid content levels in vivo at high resolution is critical for lipid research. Neutral lipids, including triacylglycerols and cholesterol esters stored in LDs, contain chains of fatty acids, which carry the symmetric stretching of aliphatic hydrocarbons with a Raman shift at 2845 cm−1 (Freudiger et al., 2008) (Fig. 2A). As noted above, the signal obtained by CRS microscopy shows an intensity dependence on the concentration of targeted molecules over a large range, and can be used for quantitative imaging (Freudiger et al., 2008; Masia et al., 2013) (Fig. 2B). Thus, by targeting the characteristic stretching energy from the fatty acid chain, CRS can directly quantify the level of total lipids without the need for any labeling in living cells, tissues and organisms (Bradley et al., 2016; Di Napoli et al., 2016; Freudiger et al., 2008; Wang et al., 2011) (Fig. 1D). In particular, the result of SRS-based quantification has been shown to be comparable to biochemical methods, such as triacylglycerol colorimetric measurement or chromatography coupled with mass spectrometry analyses (Lin and Wang, 2017; Mutlu et al., 2020; Wang et al., 2011) (Fig. 2C). The spatial resolution of CRS microscopy is optical diffraction limited in three dimensions, which makes it feasible to quantitatively measure the morphological parameters of a single LD and their changes in response to different environmental inputs (Cao et al., 2016; Di Napoli et al., 2016). Additionally, owing to utilization of near-infrared band excitation lasers (typically 805 nm for pump and 1045 nm for Stokes), CRS is suitable for long term in vivo imaging owing to its non-photo bleaching and low phototoxicity (Cao et al., 2016; Di Napoli et al., 2016; Freudiger et al., 2008), and offers deep-tissue imaging capacity up to 200–400 μm (Hill et al., 2020). The imaging depth can be further improved by using tissue clearing methods (Li et al., 2019b; Wei et al., 2019). Furthermore, for a transparent organism, such as Caenorhabditis elegans, CRS can easily reveal lipid content levels of different tissues at a single-LD resolution in live animals quantitatively, whereas for non-transparent organisms, such as Drosophila melanogaster and Mus musculus, lipid distribution can be visualized and quantified in dissected tissues in a label-free and fixation-free manner (Fig. 1D) (Dou et al., 2012; Ji et al., 2013; Wang et al., 2011; Yu et al., 2014).

Lipid classes are diverse, with different chemical structures and biological functions (Tvrzicka et al., 2011). The ability to image different lipid classes, revealing their distribution and quantifying their composition is critically important for the understanding of LD-associated biology. CRS microscopy offers such an ability in vivo, as it is able to differentiate lipid classes based on their characteristic Raman signals at the single LD level. Fatty acids, the fundamental building blocks of lipids, can be categorized into saturated and unsaturated fatty acids, depending on the presence of C=C double bonds in their hydro