Stimulated Raman photothermal microscopy: theory and implementation

Abstract

Stimulated Raman scattering (SRS) microscopy has shown enormous potential in revealing molecular structures, dynamics, and couplings in complex systems. For most biomolecules, the detection sensitivity of SRS is fundamentally limited to the milli-molar level due to the shot noise and the small modulation depth. Additionally, the operation of SRS imaging is complicated by cross phase modulation. We recently revisited SRS from the perspective of energy deposition. Via intensity gain in the Stokes beam and loss in the pump beam, the SRS process pumps molecules to their vibrationally excited states. The thereafter relaxation heats up the surroundings and induces refractive index changes. By probing the refractive index changes with a laser beam, stimulated Raman photothermal (SRP) microscopy is developed, where a >500-fold boost of modulation depth is achieved. Moreover, SRP imaging can be operated with a noisy fiber laser for excitation and a long working distance air condenser for signal collection. Two implementations and broad biological applications are reviewed. In summary, SRP microscopy opens a new way to perform chemical imaging with ultrahigh sensitivity and long working distance optics toward clinical translation.

1. INTRODUCTION

Since the discovery of Raman scattering [1], the abundant spectroscopic information provided by such a process has been very beneficial in the molecular and/or material structure characterization [2,3], chemical analysis [4,5] and has been applied in biomedical settings [6,7]. However, the small cross sections of spontaneous Raman processes result in low photon budget in the measurement, making it difficult for high speed, high throughput measurement.

Coherent Raman scattering microscopy was invented to solve the signal intensity issue. Coherent anti-Stokes Raman scattering (CARS) microscopy [8–10], was the first microscopic implementation of coherent Raman measurement. Benefit from high peak power, high numerical aperture excitation that enhanced the nonlinear coherent Raman processes, CARS measurements provided 6-orders faster imaging speed comparing to spontaneous Raman scattering, enabled real-time imaging of Raman signatures [11]. However, the non-resonant background in CARS measurements poses limit on the image quality of CARS microscopy [12].

Stimulated Raman scattering (SRS) microscopy was developed to overcome the non-resonant background issue [13–16]. In stimulated Raman scattering (SRS) (Fig. 1(A)), two laser pulse trains, naming pump and Stokes, overlap both spatially and temporally. These pulses interact coherently with Raman-active molecules that resonate at the laser beating frequency, resulting in a stimulated Raman gain (SRG) in the Stokes field and a corresponding loss (SRL) in the pump field. This gain or loss is typically measured using a modulation transfer technique. In this method, one of the beams (either pump or Stokes) is modulated in intensity, delay, or spectrum at a high frequency, before being directed to the sample. Through SRS interaction with the sample, the other beam experiences modulation at the same frequency. This modulation can then be detected with a lock-in amplifier. By measuring the non-parametric interaction components, SRS largely avoided the non-resonant background, provided better detection sensitivity and contrast. After 20 years of development, SRS microscopy has gained great success in a wide spectrum of applications, with the scales of interest spanning from virus [17], bacteria [18–20], mammalian cells [21–23], to the histology of large area tissues [24–27].

Figure 1.

Scheme (A) and energy diagram (B) of stimulated Raman gain (SRG) and stimulated Raman loss (SRL). GS, ground state; VS, virtual state. Reprinted from [28] under a Creative Commons license.

In Raman and coherent Raman spectroscopies, the existing methods largely focus on the “light” aspect of the light–matter interaction, i.e., the intensity gain, loss, or emission at new wavelengths. On the other side of the coin, the stimulated Raman gain and loss processes collectively drive real vibrational excitations, depositing energy onto the sample (Fig. 1(B)). The vibrational excitation energies quickly dissipate into heat, causing a temperature rise at the focus. Such temperature rise leads to refractive index change through the thermo-optic effect, which effectively creates a thermal lens, causing a stimulated Raman photothermal (SRP) effect [28].

Optically detected photothermal microscopy has been extensively developed and has achieved single-molecule sensitivity [29,30]. Photothermal spectroscopy, first introduced in the 1970s [31], involves optical absorption that raises the local temperature and induces a local refractive index change, which is then measured with a probe beam. Early research in photothermal microscopy focused on electronic absorption, targeting non-fluorescent dye molecules or metal nanostructures [29,31–33]. Recently, mid-infrared photothermal (MIP) microscopy has been developed, providing universal infrared-active vibrational spectroscopic features [34,35]. The MIP technique offers micromolar-level detection sensitivity and achieves spatial resolution at the visible diffraction limit.

In contrast, the thermal effects induced by Raman processes have traditionally been considered minimal due to the small cross-sections of Raman scattering. Very recently, we challenged this conventional understanding by employing low duty cycle coherent Raman excitation, which benefits nonlinearly from high laser peak power. Based on the substantial thermal effect of the SRS process, our stimulated Raman photothermal microscopy [28] enabled bond-selective imaging with high sensitivity. The experimental implementation of SRP microscopy echoes with the recent theoretical development by Min and coworkers who formulated the stimulated Raman process as a two-photon vibrational absorption process [36,37]. The theory shows that for Raman active modes, SRS as a two-photon absorption process has a “apparent cross section,” defined as ${\sigma }_{\mathit{SRS,apparent}}={\sigma }_{\mathit{SRS}}\cdot {\mathbf{\Phi }}_{\mathit{Stokes}}$, close to that of mid-infrared absorption with the help of pre-resonance.

In this review, we introduce the theory and implementation of stimulated Raman photothermal (SRP) microscopy [28]. We first present the theoretical background of SRP microscopy, along with the numerical simulation. We then introduce the implementation of SRP microscopy. The key advantages and features of SRP microscopy will be discussed, along with a comparison with other vibrational imaging modalities. We conclude this review with an outlook of SRP microscopy.

2. SRP THEORY

2.1. SRS Induced Heating and Temperature Change

2.1.1. Calculations Based on Existing Interaction Intensity Constants

To estimate the energy deposition per pair of SRS excitation laser pulses, multiple approaches can be considered. The most straightforward method involves direct calculation using electric field strength and light–matter interaction susceptibility. Taking stimulated Raman loss (SRL) as an example:

$$E_{SRL}=N\cdot Im\left\{{\chi }^{(3)}\right\}{\left|E_S\right|}^2{E}_p,$$ (1)

where $E_{SRL},E_S$, and $E_p$ refer to the electric fields of SRL, Stokes field, and pump field, respectively. N denotes the number of molecules inside the excitation volume. ${\chi }^{(3)}$ is the third-order susceptibility of the sample molecules. The intensity of SRL is then:

$$I_{SRL}={\left(E_{SRL}+E_p\right)}^2.$$ (2)

Afterward, such an interaction intensity could be converted to energy deposition power:

$$P_{SRL}=A\cdot I_{SRL}\frac{{\omega }_{SRL}}{{\omega }_p},$$ (3)

where $P_{\mathit{SRL}}$ is the power of energy deposition induced by SRS, A is the interaction area, ${\omega }_{\mathit{SRL}}$ and ${\omega }_p$ are the wavenumber of the corresponding Raman transitions and the pump field (local oscillator for SRL measurement), respectively.

This approach has been successfully employed to estimate photodamage in CARS microscopy [38,39]. However, the lack of a comprehensive third-order nonlinear susceptibilities database poses a challenge for broader application of this method.

Another perspective to describe the light–matter interaction is cross-sections and photon fluxes of the excitation laser fields:

$$R_{SRS}=N\cdot {\sigma }_{SRS}\cdot {\phi }_p{\phi }_S,$$ (4)

where $R_{\mathit{SRS}}$ is the rate of SRS transitions, unit in $\text{photon}·{\mathrm{s}}^{-1}$. ${\phi }_p$ and ${\phi }_S$ denote the photon flux of pump and Stokes beam, respectively, unit in photon·cm⁻². Photon flux is linked to light intensity with relationship: $\phi =I/\hslash \omega$. The SRS excitation rate could be converted to SRS heating power with:

$$P_{SRS}=R_{SRS}\cdot \mathrm{\hslash }{\omega }_{SRS}.$$ (5)

Notably, the SRS cross sections here are not the spontaneous Raman cross sections, but the absolute SRS cross sections (unit in cm⁴·s·photon⁻¹) that are defined in Min’s work [36,37].

2.1.2. Calculation of Energy Deposition Based on Modulation Depth

In practice, one can always estimate the energy deposition based on empirical measurement. One of the perspectives is from the modulation depth, which is defined as the intensity change ratio of the measured local oscillator.

With known photon energy of a laser beam at a specific wavelength, the number of photons in a laser pulse can be determined. Subsequently, the change in photon number, derived from the product of modulation depth and total photon number, allows for the calculation of excitation events and, consequently, the energy deposition.

In stimulated Raman loss (SRL) measurements, the relationship between the pump laser intensity (I_p), the modulation depth of the pump beam (η), and the change in photon number per pulse (ΔN) can be described by the following equation:

$$\frac{I_p\eta }{2}=\mathrm{\Delta }N\cdot \mathrm{\hslash }{\omega }_p\cdot f_{rep},$$ (6)

where ℏ is the reduced Planck’s constant, ω_p is the angular frequency of the pump field, and f_rep is the laser repetition rate. The factor of 2 accounts for the 50% modulation duty cycle. Using this relationship, the energy deposition per pair of SRL pulses can be estimated as $\mathrm{\Delta }N\hslash {\omega }_R$, where ${\omega }_R$ is the target Raman shift.

Previous studies have demonstrated that with a Stokes beam power of 25 mW (modulated at a 50% duty cycle) and a pump beam power of 15 mW at an 80 MHz repetition rate, the SRL modulation depth for the 2913 cm⁻¹ mode of dimethyl sulfoxide (DMSO) reaches 0.04% [40]. By substituting these values into the equations, the energy deposition per pair of laser pulses is calculated to be 8.7 fJ, which corresponds to 0.7 μW with an 80 MHz pulse train.

2.1.3. Calculation of Energy Deposition Based on the Shot Noise Limit

The modulation depth induced by SRS is usually very small (down to 10⁻⁶) and cannot be accurately measured together with the DC background of the strong local oscillator. This is especially the case when estimating the heat deposition caused by relatively weak stimulated Raman processes, e.g., Raman mode with small cross sections, of low concentration molecules or small nanoparticles. In these cases, filtering and amplification are necessary for modulation depth measurement, which may introduce distortion to the measurement and lead to inaccurate estimations.

For high performance systems that operate at the shot noise limit, one could estimate the modulation depth, hence the SRS heating power, based on the signal-to-noise ratio (SNR) of a measured SRS signal. Shot noise as a Poisson noise equals the square root of the total photon number:

$$N_{\mathit{noise}}=\sqrt{N_{\mathit{photon}}},$$ (7)

where $N_{\mathit{photon}}$ could be obtained with laser power (P) and angular frequency (ω) of the local oscillator and integration time (t):

$$N_{\mathit{photon}}=\frac{Pt}{\hslash \omega },$$ (8)

and for a signal with SRL (or SRG) photon count $N_{\mathit{signal}}$, the SNR could be defined as:

$$\mathrm{S}\mathrm{N}\mathrm{R}=\frac{N_{\mathit{signal}}}{N_{\mathit{noise}}}.$$ (9)

Here we define noise $\left(I_{\mathit{noise}}\right)$ as the signal standard deviation of the signal when there is only the local oscillator but no modulation. Signal $\left(I_{\mathit{signal}}\right)$ could be obtained under the same condition, but with SRS modulation. And SNR hence can be defined and measured:

$$\mathrm{SNR}=\frac{I_{\mathit{signal}}}{I_{\mathit{noise}}}.$$ (10)

Combining Eqs. (9) and (10), $N_{\mathit{signal}}$ could be then calculated with a given set of excitation parameters and SNR, assuming shot-noise-limited measurement. And $N_{\mathit{signal}}$ equals to the number of SRS excitation events. From here, the energy deposition could be estimated with equations mentioned in Section 2.1.b.

2.1.4. SRP-Induced Temperature Rise

After obtaining the heating power from the above methods, it’s still necessary to convert it into temperature rise to comprehend the consequence of SRS heating. To quantitatively assess the SRP effect, a finite-element model (Fig. 2) has been developed to simulate heat deposition both spatially and temporally, based on Fourier’s law:

$$dQ=-kA(dT/dr),$$ (11)

where (dQ) represents the conducted heat energy within a given time window, (k) is the thermal conductivity, (A) is the surface area where conduction happens, and (dT) is the temperature difference over the heat conduction distance (dr). In the model, the space has been segmented into spherically symmetric layers to reduce complexity. The SRS on-off process induces a heating and cooling cycle at the focal point, with heat transfer occurring during both phases. During the heating phase, SRS pulses deposit heat into the sample instantaneously with a Gaussian spatial distribution, as the vibrational excited state relaxation time is much shorter than the simulation time grid. Finally, by combining the heating estimated from the modulation depth with the thermal conduction calculated using Fourier’s law, the temperature distribution over time and space is determined.

Figure 2.

Simulation of stimulated Raman induced thermal effect. (A) Thermal effect generation scheme. (B) Schematic of a heat diffusion model. The system is meshed into concentric spherical shells. The red arrow indicates the thermal propagation direction. Reprinted from [28] under a Creative Commons license.

The simulation results (Fig. 3) indicate that, with aforementioned excitation conditions and sample (DMSO), the temperature at the center of the laser focus can increase by up to 2.4 K after 12 μs of SRS excitation, which corresponds to 960 pairs of pump/Stokes pulses. At (t = 12 μs), the temperature increases by 0.54 K at 1 μm and 0.12 K at 2 μm from the focal center. The temperature distribution at various time points is also depicted in Fig. 3. The full width at half maximum (FWHM) of the temperature rise field at (t = 12 μs) is calculated to be 1.1 μm, indicating highly localized heating.

Figure 3.

Simulation of temperature rise induced by SRP effect in temporal (top) and spatial (bottom) domains. Spatial scale bar: 1.0 μm. Reprinted from [28] under a Creative Commons license.

2.2. Experimental Validation of Temperature Rise

We have used fluorescence thermometers to validate the simulation results, leveraging the well-documented temperature dependence of certain fluorophores’ emission intensities [41]. Specifically, the fluorescence intensity of Rhodamine B is known to decrease by approximately 2% per Kelvin at room temperature [42]. This characteristic has been previously utilized in fluorescence-detected mid-infrared photothermal microscopy [43,44]. For SRP, this method has been used to quantify the temperature rise at the SRS foci, with Rhodamine B serving as the fluorescence thermometer. Two synchronized femtosecond pulse trains are chirped to excite SRS processes in a spectral focusing mode [45,46]. When the chirped pump and Stokes lasers were focused into a DMSO solution containing Rhodamine B, the molecules at the SRS focus were electronically excited via multiphoton absorption, resulting in fluorescence emission. Concurrently, tuning the differential frequency between the pump and Stokes lasers to resonate with the C–H vibration in DMSO induced the SRP effect, which elevated the temperature and consequently reduced the two-photon fluorescence intensity of the Rhodamine B molecules (Fig. 4).

Figure 4.

Energy diagram for the fluorescence thermometer experiment. Reprinted from [28] under a Creative Commons license.

The same parameters as those in the simulation have been utilized in the experiments. The fluorescence intensity change curve measured at the off-resonance frequency was first subtracted from the on-resonance measurement to eliminate the photobleaching effect. The results showed that the fluorescence intensity decreased by approximately 2.3% after 12 μs of on-resonance SRS, which corresponds to a temperature rise of about 1.2 K (Fig. 5), closely matching the simulation result of 2.4 K. The discrepancy between these values is likely due to their different sampling regions. The experimental measurement captured the average temperature rise across the entire laser focus, whereas the simulation result represented the temperature rise at the center of the laser focus ~(20 nm). This relationship can be mathematically described as:

$$F(\stackrel{\rightharpoonup }{r},t)=\frac{1}{V}\int (1-T(\stackrel{\rightharpoonup }{r},t))\times \frac{dF}{dT}{d}^3\stackrel{\rightharpoonup }{r},$$ (12)

where $F(\stackrel{\rightharpoonup }{r},t)$ and $T(\stackrel{\rightharpoonup }{r},t)$ denotes the fluorescence intensity and temperature as a function of space and time, respectively; $\frac{dF}{dT}$ is the fluorescence change per degree Kelvin. By calculating the weighted average of temperature increases throughout the heating region and considering the two-photon excitation intensity of a Gaussian beam, the fluorescence change curve closely aligned with the simulation (Fig. 5(C), gray line). Control experiments have also been reported with temperature-insensitive Rhodamine 800 (Rh800) as the sample. In this case, the fluorescence intensity remained constant with SRS excitation on or off resonance, excluding other possibilities that may alter the fluorescence intensities, such as thermophoresis.

Figure 5.

Fluorescence intensity of rhodamine B in DMSO during an SRS process. (A) Fluorescence intensity of RhB with SRS excitation off-resonance. (B) Fluorescence intensity of RhB with SRS excitation on-resonance. (C) Cropped top part of (A) and (B). (D) Control experiments with Rh800 as the sample. The beating frequency (ω_p – ω_s) is tuned to 2913 cm⁻¹ for on resonance and 2850 cm⁻¹ for off resonance. BG, background. The on resonance (BG removed) curve is obtained by subtracting the off resonance (BG) from the on-resonance curve, to eliminate the non-photothermal contributions. Reprinted from [28] under a Creative Commons license.

2.3. Ball Lens Model

The conversion of temperature modulation into light contrast is a crucial aspect of photothermal microscopy. While existing models [29] that describe this process are comprehensive and accurate, they can often be too complex to facilitate an intuitive understanding of the topic. Consequently, there is a need for a simplified, yet easy-to-interpret analytical model that can help build intuition and clarify the physical processes involved. Such a model would provide valuable insights into the contrast mechanism and the parameter dependencies of this imaging technique, potentially guiding the development of more efficient photothermal measurements.

An analytical description of the actual excitation profile of SRS, along with the changes in light propagation due to the highly localized (near diffraction limit) refractive index change, presents significant challenges. To develop a more comprehensible model, several assumptions were made: (1) the excitation region is spherically symmetric and uniform; (2) a non-diffraction, ray-optics model is used; and (3) the paraxial approximation is applied. With these assumptions, the optical measurement of SRP can be simplified into a ball-lens model using lens maker equations (as detailed in the following section). This model only considers the SRP effect induced by instantaneous heating and does not account for thermal diffusion.

The model (as shown in Fig. 6. is defined in pure liquid sample under the SRS excitation of pump and Stokes laser pulses with peak photon flux ϕ (photon·cm⁻²·s⁻¹) defined as:

$$\varphi =\frac{P}{hv\cdot A}\cdot \frac{1}{f_{rep}{\tau }_{exc}},$$ (13)

where P is average laser power, h is the Planck constant, ν is the laser frequency, f_rep is the laser repetition rate, and τ_exc is the laser pulse width. The number of SRS events (N_SRS) induced by a single pair of pump and Stokes pulses can be calculated by:

$$N_{SRS}=N_{mol}\cdot {\sigma }_{SRS}\cdot {\varphi }_{pump}\cdot {\varphi }_{Stokes}\cdot {\tau }_{exc},$$ (14)

where $N_{\mathit{mol}}$ is the number of targeted chemical bonds in the laser foci, ${\sigma }_{\mathit{SRS}}$ is the stimulated Raman cross-section (unit in ${\mathrm{c}\mathrm{m}}^4\cdot \mathrm{s}\cdot {\mathrm{p}\mathrm{h}\mathrm{o}\mathrm{t}\mathrm{o}\mathrm{n}}^{-1}$), ${\varphi }_{\mathit{pump}}$ and ${\varphi }_{\mathit{Stokes}}$ are the photon flux of pump and Stokes lasers, respectively. The energy deposition $(Q)$ through the SRS process is then given by:

$$Q=\mathrm{\hslash }{\omega }_{SRS}\cdot N_{SRS},$$ (15)

where ℏ is the reduced Planck constant and ω_SRS is the Raman transition energy. Under the approximation that SRS heating region in a spherically symmetric, its radius (r) was defined as half of the lateral resolution of SRS excitation:

$$r=\frac{1}{4NA}\sqrt{1/\left(\frac{1}{{\lambda }_{\mathit{pump}}^2}+\frac{1}{{\lambda }_{\mathit{Stokes}}^2}\right)},$$ (16)

and excitation volume:

$$V_{SRS}=\frac{4}{3}\pi r^3.$$ (17)

Figure 6.

Thermal ball lens model. The solid line indicates the propagation in the absence of a ball lens. The dashed line indicates the propagation of the probe laser in the presence of a ball lens. n₀ is the refractive index of the medium. Δn is the change of refractive index induced by the SRP effect. Reprinted from [28] under a Creative Commons license.

Here, (C_p) represents the heat capacity at constant pressure of the medium, and (V_SRS) denotes the volume of the SRS excitation. Thermal diffusion is neglected because the energy deposition from a single pair of laser pulses is completed within the thermal confinement time (approximately 1 μs). The pressure is assumed to remain constant, as the stress confinement time (less than 1 ns) is too brief to be detected.

The elevated temperature then changes the refractive index of the medium (Δn):

$$\mathrm{\Delta }n=\mathrm{\Delta }T\cdot \frac{dn}{dT},$$ (18)

where dn/dT is the thermo-optic coefficient of the medium.

As shown in Fig. 6, the changed refractive index effectively forms a thermal ball lens, with its focal length f satisfying the relationship that:

$$\frac{1}{f}=\frac{2}{r}\left(\frac{n_0+\mathrm{\Delta }n}{n_0}-1\right)=\frac{2\mathrm{\Delta }n}{n_{0}r}.$$ (19)

Here, r is the radius of the ball lens, n₀ is the refractive index of the medium.

To measure the thermal lens effect, a focused Gaussian probe laser was introduced, aligned collinearly with the SRS laser beams. The presence of the ball lens alters the propagation of the probe light. In the accompanying diagram, the dashed lines represent the light propagation in the absence of a ball lens, while the solid lines depict the light propagation with the ball lens in place:

$$\frac{1}{f}=\frac{1}{q}-\frac{1}{z_f},$$ (20)

where z_f denotes the offset between the SRS and probe laser focus.

Under the paraxial approximation, for an annular ray of light with incident angle θ as shown in Fig. 6, Eq. (20) becomes:

$$\frac{1}{f}\approx \frac{\mathrm{\Delta }\mathrm{\Theta }(\theta )}{z_f\mathrm{s}\mathrm{i}\mathrm{n}\theta \mathrm{c}\mathrm{o}\mathrm{s}\theta },$$ (21)

where $\mathrm{\Delta }\mathrm{\Theta }(\theta )$ is the change of propagation angle of the probe light at angle Θ.

Combining Eqs. (19) and (21) yields the change of laser propagation angle ΔΘ induced by the ball lens as a function of incident angle θ:

$$\mathrm{\Delta }\mathrm{\Theta }(\theta )=\frac{2z_f\cdot \mathrm{\Delta }n\cdot \mathrm{s}\mathrm{i}\mathrm{n}\theta \mathrm{c}\mathrm{o}\mathrm{s}\theta }{n_{0}r}.$$ (22)

The far-field annular intensity distribution of the probe laser, I(θ), can be written as:

$$I(\theta )=I_0\pi Z_R^2\cdot \theta \cdot \mathrm{e}\mathrm{x}\mathrm{p}\left(-\frac{Z_R^2}{{\omega }_0^2}\cdot {\theta }^2\right).$$ (23)

Here, (Z_R) represents the Rayleigh length of the laser focus, and (ω₀) denotes the beam waist. After passing through the ball lens, the intensity of the probe laser is measured by a photodetector with an angular detection aperture size ${\theta }_{\mathit{Det}}$. Consequently, the SRP signal is determined by calculating the intensity difference with and without the ball lens:

$$I_{SRP}={\int }_0^{{\theta }_{Det}}I(\theta +\mathrm{\Delta }\mathrm{\Theta }(0))d\theta -{\int }_0^{{\theta }_{Det}}I(\theta )d\theta \approx I\left({\theta }_{Det}\right)\cdot \mathrm{\Delta }\mathrm{\Theta }\left({\theta }_{Det}\right).$$ (24)

Combining all the above equations gives:

$$\begin{gathered} I_{SRP}\approx I\left({\theta }_{Det}\right)\frac{4z_f\cdot sin{\theta }_{\mathit{Det}}\cdot \mathit{cos}{\theta }_{\mathit{Det}}}{n_{0}r}\cdot \frac{1}{C_p}\frac{dn}{dT}\cdot \frac{1}{V_{SRS}} \\ \cdot \left(\hslash {\omega }_{SRS}\right)\left(N_{mol}{\sigma }_{SRS}{\varphi }_{pump}{\varphi }_{Stokes}{\tau }_{exc}\right). \end{gathered}$$ (25)

In Eq. (25), $\frac{1}{C_p}\frac{dn}{dT}$ denotes the thermal and thermo-optic properties of the sample environment. $V_{\mathit{SRS}}$ denotes the volume of heating. $\hslash {\omega }_{\mathit{SRS}}$ represents the energy of vibrational transition. $N_{\mathit{mol}}{\sigma }_{\mathit{SRS}}{\varphi }_{\mathit{pump}}{\varphi }_{\mathit{Stokes}}{\tau }_{\mathit{exc}}$ represents the number of transitions caused by SRS.

Extracting the core terms, we have:

$$I_{SRP}\propto N_{mol}{\sigma }_{SRS}{\varphi }_{pump}{\varphi }_{Stokes}{t}_{exc}\cdot \hslash {\omega }_{SRS}\cdot \frac{1}{V_{SRS}}{C}_p{n}_0\frac{dn}{dT}.$$ (26)

In this model, SRS is treated as a two-photon vibrational excitation process. The first term, $\left(N_{\mathit{mol}}{\sigma }_{\mathit{SRS}}{\varphi }_{\mathit{pump}}{\varphi }_{\mathit{Stokes}}{\tau }_{\mathit{exc}}\right)$, represents the number of SRS events, or vibrational excitations. The second term, $\left(\hslash {\omega }_{\mathit{SRS}}\right)$, corresponds to the energy of the vibrational transition. The final term describes the properties of the measurement environment. The product of the first two terms quantifies the energy deposition, while the last term determines the rate of heat to refractive index conversion. This term is inversely proportional to the excitation volume and heat capacity, and directly proportional to the thermo-optic coefficient. Although the environmental properties term has the potential to enhance signal intensity, it is not applicable to SRS measurement.

In this model, only the instantaneous heating induced by SRS is considered, with thermal diffusion being ignored for simplicity. However, in practical scenarios, thermal diffusion inevitably occurs and must be accounted for in its impact on the size and temperature change of a thermal lens. Qualitatively, thermal diffusion reduces the peak temperature increase while expanding the size of the thermal lens, both of which contribute to a reduction in the photothermal signal. Notably, this model is based on ray-optics even though it’s discussing phenomenon very close to the diffraction-limited laser focus. As a result, although the qualitative picture of this model is relatively accurate, some of the predictions, such as a varying probe beam divergent angle with different refractive index changes, do not agree very well with the experiments.

2.4. Finite-Difference Time-Domain Model

To develop a comprehensive understanding of the SRP microscope, it is essential to model the conversion process from temperature modulation of the sample to light intensity modulation on the photodetector. Given a specific temperature field from a previously established layered finite-element model, this problem can be translated into the propagation of light through a medium with a defined refractive index field, as illustrated in Fig. 7. However, due to the complex nature of the temperature fields and corresponding refractive index fields, analytically describing the refractive index field and its scattering is impractical. Instead, this process can be numerically calculated using a finite-difference time-domain (FDTD) model. Subsequent parameter sweeps in the FDTD simulation will provide insights into the parameter dependencies of the model.

Figure 7.

Simulation of thermal lensing. Top, thermal lens induced by SRP effect. Bottom, illustration of FDTD simulation of thermal lensing effect. Reprinted from [28] under a Creative Commons license.

The numerical description of the refractive index field was constructed based on the previously simulated temperature field. The temperature field was first converted into a refractive index field using the given refractive index and thermo-optic coefficient of the simulated medium. This layered refractive index field was then remapped onto a standard x–z coordinate system, with a resolution of 200 × 200 pixels and a pixel spacing of 10 nm. The resulting 2D refractive index field, representing the thermal lens, was imported into Lumerical FDTD simulation software as a refractive index object.

A focused light source along the z-direction, with a specified NA and distance to the thermal lens, was used to illuminate the thermal lens. An electric field detector was positioned approximately 10 μm away from the laser focus to measure the resulting light field. The far-field angular distribution of light intensity was then calculated, as shown in Fig. 8. The modulation depth was determined by comparing the light intensity change within the collection NA to the scenario where the thermal lens was absent. A z-scan was performed by adjusting the z-position of the thermal lens in the FDTD simulation. To investigate parameter dependencies, a parameter sweep was conducted on illumination NA, axial offset, and collection angle.

Figure 8.

FDTD simulated far-field angular distribution of probe light intensities. The right panel shows the calculated change of probe light intensity induced by the thermal lens.

Figure 9 presents the simulation results as a function of z-offset and the illumination NA of the probe beam, with the thermal lens parameters held constant and the collection NA set to 0.2. For each illumination NA, two distinct maxima in absolute modulation depth are observed, corresponding to specific z-offset values. In cases of high NA illumination (NA = 1.49 or 1.1), these maxima occur at approximately ±2 μm z-offset. The highest signal intensity is achieved with an illumination NA of 1.49, which then gradually decreases as the NA is reduced. This trend suggests that higher illumination NA enhances signal intensity, independent of the thermal lens characteristics.

Figure 9.

FDTD simulated z-scan shows the relationship between signal modulation depth and different probe beam illumination NA. The excitation NA is set to 1.47, and the collection aperture is 0.2.

Further analysis was conducted by calculating the signal modulation depth as a function of the collection NA under various z-offsets, as illustrated in Fig. 10. The thermal lens parameters and the illumination NA were kept constant. Due to the anti-symmetric nature of the photothermal signal, only the negative half of the z-scan was performed for simplicity. As shown in Fig. 10, although z-offsets of −10 μm and −0.5 μm may yield similar modulation depths when the collection NA is approximately 0.2, their far-field modulation distributions differ significantly. The total signal that can be collected with a −0.5 μm z-offset is much greater than that of the −10 μm case, considering the higher total light intensity with a larger collection aperture. An interesting observation is that when the z-offset is zero, no modulation is detected. This can be explained by the thermal lens model, which suggests that when a light beam is focused precisely at the optical center of a lens, the lens does not alter the light’s propagation, resulting in no observed signal.

Figure 10.

FDTD simulated z-scan shows the relationship between the signal modulation depth, the pump–probe focal offset (inset), and the collection NA. The illumination NA is 1.47.

Overall, the simulation results align well with previous experiments and theoretical analysis that have been presented in visible photothermal microscopies. Sharing the same thermal lensing measurement component, similar z-scan profiles could be observed in both SRP and visible photothermal microscopies. However, the nonlinear excitation in SRP measurement brings extra z-sectioning property than visible photothermal measurement, resulting in much sharper profile in z-scan experiments. This may help SRP with better-defined optimal axial alignment conditions, as well as reducing the image artifacts caused by out-of-focus objects.

3. IMPLEMENTATION OF SRP MICROSCOPY

3.1. OPO-Based SRP Microscope

3.1.1. Justification

Rooted from the same coherent Raman process, SRP could be measured on any existing SRS microscope, with appropriate modification of the system. In this section, we demonstrate the first OPO-based SRP microscope that’s converted from an OPO-based SRS microscope.

There are two major differences between the instruments of SRS and SRP. First, different laser beams are measured in SRS or SRP. In SRS, one of the ultrafast lasers serves as a local oscillator, and the intensity of it is measured. Therefore, it must maintain a pseudo continuous wave mode, with minimal perturbation on its noise performance so that the final intensity measurement can be carried out in an efficient way. In practice, the local oscillator beam is usually directly sent to the microscope without any active optical components such as an acousto-optic modulator (AOM). However, in SRP, there is no local oscillator required, so the modulation could be performed in a more flexible way without harming the overall noise performance. Second, high NA collection is mandate in SRS while low NA collection is preferred in SRP. SRS is an absorption measurement in nature, where complete light collection is necessary to avoid cross-phase modulation background. SRP, on the other side, is a scattering measurement, which measures the re-distribution of the probe laser, therefore collecting part of the laser beam provides the optimal signal intensity.

3.1.2. Instrument

In OPO-based SRP imaging, we used an Insight X3 laser as the excitation source for SRS. The modulation of both pump and Stokes beams is achieved using two synchronized AOMs, which output first-order beams at a modulation frequency of 125 kHz with various duty cycles. Additionally, a 765 nm continuous probe laser (TLB6712-D; Spectral Physics) is introduced post-dichroic mirror via a polarized beam splitter, creating a three-beam copropagating collinear system (Fig. 11). These beams are directed to a two-dimensional galvo scanning unit (GVS002; Thorlabs), which is conjugated through a four-focal system to the back aperture of a 100× oil objective (NA = 1.49, UAPON100XOTIRF; Olympus). The condenser’s numerical aperture is set to 0.4 to facilitate the detection of the thermal lensing signal. Detection is performed using a broadband silicon photodiode (Hamamatsu) with 50-ohm resistance, a 22 kHz high-pass radio frequency filter (Mini-circuits), and a 46 dB low-noise amplifier (SA230-F5; Wayne). A tilted bandpass optical filter (FL780-10; Thorlabs) is positioned before the detector to block SRS beams, allowing exclusive detection of the probe beam. The output signal is digitized using a high-speed data acquisition card (Alazar card, ATS9462; Alazar Technologies). For imaging biological samples in aqueous environments (Fig. 19), viruses (Fig. 20(A)), and comparing SRS with SRP (Fig. 21), the 1045-nm femtosecond laser output is directed to an LBO crystal to generate a 522.5 nm SHG laser beam, which is then chirped to function as the SRP probe. This picosecond probe, with its reduced coherence length, minimizes interference from the medium/cover-glass interface.

Figure 11.

SRP microscope design (A) and characterization of SRP modulation (B). Experimental setup. AMP, amplifier; AOM, acousto-optic modulator; COND, condenser; DL, delay line; DM, dichroic mirror; HP, highpass filter; HWP, halfwave plate; OBJ, objective; PBS, polarizing beam splitter; PD, photodiode; SM, scanning mirror; SP, spectral filter. Reprinted from [28] under a Creative Commons license.

Figure 19.

SRP imaging of a live cell. (A) Live SJSA-1 cell in PBS imaging at 2930 cm⁻¹. (B) Time trace imaging of the live SJSA-1 cell with color coded temporal dynamics of the lipid droplets inside the cell. Scale bar: 10 μm. Reprinted from [28] under a Creative Commons license.

Figure 20.

SRP imaging of a fixed SJSA-1 cell in PBS at 2930 cm⁻¹ (a). Scale bar: 10 μm. (b) Phasor decomposition of the hyperspectral dataset yields organelle map. (c) The SRP spectrum at the C–H region for each phasor component. (d) Mapping of each component from phasor analysis. Reprinted from [28] under a Creative Commons license.

Figure 21.

SRP analysis of SJSA-1 cell in PBS with (A) and (B) or without (C) and (D) palmitic acid-d31 (PA-d31) treatment. Phasor analysis separates PA-d31 SRP signal (red) and cell background (cyan) (B) and (D) with spectral features shown in (A) and (C). Scale bar: 10 μm. (E) and (F) Hyperspectral SRP imaging of SJSA-1 cell in PBS. Data were shown with 20 cm⁻¹ spectral steps. The cells were incubated with (E) or without (F) deuterated palmitic acid (PA-d31) for 18 h prior to imaging. Scale bar: 10 μm. Reprinted from [28] under a Creative Commons license.

The SRP system achieves electronic synchronization via a NIDAQ (National Instrument, PCIe-6363) card, which manages the galvo scanning unit and generates both the transistor–transistor logic (TTL) trigger for sampling control of the Alazar card and the pixel trigger for the function generator. The function generator produces rectangular waves with varying duty cycles at 125 kHz in burst mode, which modulate the pump and Stokes pulse trains through two acousto-optic modulators (Fig. 11). The amplified signal from the detector is digitized by the Alazar card at a 20 MHz sampling frequency and subsequently transmitted to the host computer for further analysis.

Following the recording of signals from each pixel, a Whittaker smoother [47] is employed to eliminate the fluctuating baseline caused by transmission variations. Matched filtering is subsequently applied to enhance the signal-to-noise ratio (SNR) [48]. The baseline-corrected signal undergoes a Fourier transform to produce the signal spectrum. To mitigate out-of-frequency noise, a matched multi-band-pass filter is applied to the signal spectrum, targeting up to seven harmonic frequencies of the modulation frequency (Fig. 12). The width of each bandpass filter is set to 13.89 kHz, corresponding to a pixel dwell time of 72 μs. The matched-filtered spectrum is then inversely Fourier transformed back to the time domain. The SRP signal intensity for each pixel is determined by calculating the average peak-dip contrast of the processed time-domain data.

Figure 12.

Data processing of a thermal dynamic trace recorded by an SRP microscope. The signal is acquired from a single 100-nm PMMA nanoparticle at a Raman shift of 2950 cm⁻¹ immersed in glycerol-d8. FFT, fast Fourier transform; iFFT, inverse fast Fourier transform. Reprinted from [28] under a Creative Commons license.

The SRP ball lens model highlights that the properties of the medium significantly influence photothermal signal intensity. The relationship between photothermal signal intensity (S), thermo-optic coefficient $(dn/dT)$, and heat capacity $\left(C_p\right)$ is given by $S\propto \left(\frac{dn}{dT}/C_p\right)$, as corroborated by previous studies. However, water, the most prevalent medium in biological samples, has a low thermo-optic coefficient (−1.13 × 10⁻⁴ K⁻¹) and a high heat capacity (4181 J·kg⁻¹·K⁻¹). To enhance signal intensity, we explored various common liquid media (Table 1). By inserting parameters to the relationship, it has been found that glycerol could provide 4.27-fold signal improvement compared to aqueous environment. Further numerical simulation of the SRP heating and heat diffusion of a PMMA particle inside different medium has also been carried out as shown in Fig. 13. The result suggests that glycerol increases signal intensity by 3.21 times compared to water. The discrepancy between the quantitative predictions from medium property calculation ( 4.27-fold) and numerical simulation ( 3.21-fold) is mainly attributed to the contribution of PMMA material properties in the simulation setting. Glycerol is also highly bio-compatible and commonly used as a mounting medium or clearing agent in bio-imaging. Consequently, glycerol was selected to enhance the sensitivity of SRP imaging. Given the Raman-active vibrational characteristics of glycerol, deuterated glycerol (glycerol-d8) was used for SRP measurements in the C–H and fingerprint regions.

Table 1.

Thermal and Thermo-Optic Properties and Estimated Relative Signal Intensity of the Materials Related to This Study [28]

Thermal Property	Unit	DMSO	Hexane	Glycerol	Water
Heat capacity	J/(kg·K)	1966	2251	2400	4184
Thermal conductivity	W/(m·K)	0.200	0.124	0.283	0.598
Thermo-optic coefficient dn/dT (10−4)	K−1	−4.93	−5.20	−2.30	−1.13
Refractive index		1.479	1.375	1.473	1.333
Relative signal intensity	a.u.	10.3	8.83	4.27	1

Figure 13.

Simulation results for the shape of temperature gradient (A) and thermal lens (B) induced by the SRS absorption of a 100-nm PMMA particle in glycerol or water. The PMMA particle is placed at x = 0. Two local minimums of refractive index change appear at two PMMA-medium boundaries, due to the sudden change of thermos-optic coefficients (dn/dT) at the boundaries. Reprinted from [28] under a Creative Commons license.

3.1.3. Performance Characterization

The spectral fidelity of the SRP microscope was evaluated using well-defined samples. As illustrated in Fig. 14, the SRP spectra are consistent with the SRS spectra for both bulk liquid samples and PMMA nanoparticles. It is noteworthy that the SRP intensity is proportional to ${\omega }_{\mathit{SRS}}$, although its effect on the SRP spectrum is minimal within a narrow spectral range. This high spectral fidelity establishes a robust basis for comparing the detection sensitivities of the two techniques.

Figure 14.

SRP spectral fidelity. (a) Comparison between SRS and SRP spectra of pure DMSO. (b) Comparison between SRS and SRP spectra of a 500-nm PMMA particle. Reprinted from [28] under a Creative Commons license.

For a pure liquid, under a 5% duty cycle and 125 kHz modulation frequency, the induced modulation on the probe beam was sufficiently strong to allow direct measurement of the SRP signal in the direct current channel without amplification (Fig. 11(b)). Using appropriate laser powers to excite the C–H symmetric stretching mode (2913 cm⁻¹) of DMSO, the modulation depth achieved was 22.3%. This represents an increase of over 500-fold compared to the SRS modulation depth (0.04%) at the same average power. This significantly higher modulation depth provides a foundation for enhanced detection sensitivity.

The limit of detection (LOD) for DMSO was measured, focusing on the 2913 cm⁻¹ mode. To maintain consistent thermal and optical properties, deuterated DMSO (DMSO-d6) was used as the solvent for diluting DMSO. As depicted in Fig. 15(a), the SRP spectrum was clean and smooth at high DMSO concentrations, with detectable signals at concentrations as low as 5.1 mM. The LOD was calculated to be 2.3 mM using the formula LOD = 3σ/k, where σ represents the standard deviation of the baseline and k is the slope of the intensity-concentration linear calibration curve. In comparison, the LOD for SRS under identical average laser powers was 39 mM, indicating that SRP measurement provides approximately a 17-fold improvement. Additionally, LODs for C≡C and C–D bonds were measured in DMSO media using 1,7-Octadiyne and DMSO-d6. In both cases, SRP demonstrated superior sensitivity compared to SRS, with a 12-fold LOD increase for 1,7-Octadiyne and a 4-fold LOD increase for DMSO-d6. This discrepancy between these different extents of improvement majorly originate from the differential absorption properties of DMSO-d6 and DMSO in their respective C–H and C–D measurement windows. Higher linear and nonlinear absorption in these solvents leads to elevated photothermal background and, consequently, increased background fluctuations. In addition, the presence of hω term in the SRP signal intensity equation (Eq. (26)) leads to higher heat deposition when exciting higher wavenumber modes, which favors C–H (2913 cm⁻¹) measurement over C–D (2125 cm⁻¹). Notably, sensitivity is related to, but distinct from, modulation depth. Modulation depth refers to the relative signal intensities alone, whereas sensitivity is proportional to the signal-to-noise ratio. In SRP measurements, although modulation depth improved by more than 500-fold, the relatively higher noise floor resulted in a smaller improvement in LOD of approximately 17-fold.

Figure 15.

SRP spectroscopy and imaging performance characterization. SRP (A) or SRS (B) signal with gradient concentrations of DMSO dissolved in DMSO-d6. Concentration unit: mM. Insert shows the signal intensity as a function of concentration. (C) SRP and SRS image of 100 nm PMMA beads at 2950 cm⁻¹ with the same average power, at the same field of view. Scale bar: 500 nm. The beads were immersed in d8-glycerol. (d) The Gaussian fitting measured size of the bead is 218 nm (FWHM). Reprinted from [28] under a Creative Commons license.

The enhanced sensitivity of the SRP microscope facilitates high-quality imaging of nanoparticles. Using this system, a hyperspectral image of 100 nm PMMA beads was obtained, as depicted in Fig. 15(C). The SRP spectrum revealed a distinct Raman peak for PMMA at 2950 cm⁻¹, clearly separated from the background spectrum (Fig. 16), with a signal-to-noise ratio (SNR) of approximately 7.0 after BM4D [49] denoising. In contrast, the SRS measurement of the same sample, under identical average laser power, showed no discernible contrast for the 100 nm beads. Overall, SRP demonstrated superior sensitivity compared to SRS for both liquid samples and nanoparticles. Notably, the introduction of a third probe beam at a shorter wavelength enhanced spatial resolution (Fig. 15(D)). The intensity profile across a pair of 100 nm PMMA beads showed a Gaussian-fitted full width at half maximum (FWHM) of approximately 218 nm. After deconvolution with the bead size, the FWHM was approximately 194 nm, which is below the theoretical resolution limit of SRS under the same conditions (approximately 217 nm, FWHM of the Airy disk).

Figure 16.

SRP spectrum of a single 100 nm PMMA nanoparticle immersed in glycerol. The peak around 2950 cm⁻¹ is from CH₃ stretching vibration in PMMA. Reprinted from [28] under a Creative Commons license.

SRS microscopy relies on measuring either the gain in Stokes or the loss in the pump beam, necessitating a high-NA objective to maximize collection efficiency and minimize cross-phase modulation. In contrast, SRP microscopy detects light scattering due to thermal expansion or refraction from the thermal lens, favoring light collection with a low-NA objective or an air condenser. As demonstrated in Fig. 17, using a 0.5 NA air condenser, the SRP image is clear and sharp, whereas the SRS image is comparatively noisier. Additionally, SRP maintains high spectral fidelity (Fig. 17(C)), while the SRS spectrum is distorted by enhanced cross-phase modulation background. Quantitatively, the SRP image exhibits a 21.2-fold improvement in SNR and a 7.8-fold increase in signal-to-background ratio (SBR) compared to the SRS image (Fig. 17(D)).

Figure 17.

Comparison of SRP and SRS performance under different light collection conditions. SRP (A) and SRS (B) images of 500 nm PMMA nanoparticles recorded with 1.2 (NA) water objective for focusing and a 0.5 NA air condenser for signal collection. Raman shift: 2950 cm⁻¹. Scale bar: 1.5 μm. (c) Comparison between the SRP and SRS spectra of a single 500 nm PMMA nanoparticle. The broader line profile in SRS with low NA collection (blue line) is caused by a Raman-independent cross-phase modulation background. (d) Image quality comparison. Reprinted from [28] under a Creative Commons license.

The relaxation of the high NA oil condenser requirement brings convenience for SRP applications. Previously in SRS imaging, a pair of high NA focusing and collection optics is required to acquire high quality images. The working distance of high NA optics are usually very small, which allows only a few millimeters of space in between. Such a shallow distance limits the scope of sample formats that is compatible with the measurement. For instance, 96-well plate, which is widely applied in high-throughput culture of bacteria and cells, is too thick for SRS imaging. And extra training is necessary for non-expert end users. However, in SRP, low NA and long working distance collection optics can be used without affecting the SNR of the signal. As shown in Fig. 18, a simple and budget-friendly condenser lens could collect the SRP signal of liquid sample (DMSO) with almost no SNR loss. The 18-mm long working distance allows a more diverse scope of sample formats.

Figure 18.

SRP imaging performance comparison with different collection optics. Sample of interest: DMSO, 2913 cm⁻¹. (a) SRP signal collected with high NA oil condenser. (b) SRP signal collected with a single piece condenser lens.

3.1.4. Biological Applications

Label-free SRP imaging on live SJSA-1 osteosarcoma cancer cells has been conducted to investigate the potential of SRP in bio-imaging. As illustrated in Fig. 19, lipid droplets, the endoplasmic reticulum (ER), and the nucleolus exhibited strong signals at 2930 cm⁻¹. The nuclear membrane displayed significant contrast, suggesting locally enhanced photothermal intensity in an aqueous environment. At an imaging speed of approximately 2.8 seconds per frame, the dynamics of lipid droplets were captured, with all trajectories shown in Fig. 19(B).

Subsequently, hyperspectral SRP imaging of SJSA-1 cells in the C–H stretching vibration region has been performed. As shown in Fig. 20, subcellular structures such as lipid droplets and the nucleolus demonstrated clear contrast. Phasor analysis was then applied to segment the subcellular structures, identifying five major components (Fig. 20(B)). Detailed maps of each phasor components are displayed in Fig. 20(D).

To evaluate the applicability of SRP imaging within the silent window, the cellular uptake of deuterated palmitic acid (PA-d31) was investigated. The spectral profile obtained from phasor analysis (Fig. 21(B)) exhibited a peak at 2100 cm⁻¹, corresponding to the C–D stretch vibration. In the control sample without PA-d31 treatment, only the inherent cellular components were observed (Figs. 21(C) and 21(D)). Notably, the background signal remained independent of the Raman shift (Figs. 21(B) and 21(D)). This background likely arises from the overtone absorption of C–H vibrational transitions, which also deposit energy and generate photothermal contrast. The cell background in the C–D SRP images was more pronounced than in the C–H SRP images (Fig. 21(C)), likely due to the 855 nm pump laser used for C–D excitation being near the 920 nm overtone absorption band of the C–H stretching vibration. Details of the hyperspectral dataset are displayed in Figs. 21(E) and 21(F).

Using mouse brain slices, the performance of SRP imaging on tissue specimens was assessed. Three fields of view (FOVs) were selected to represent typical brain structures. As shown in Figs. 22(A)–22(C), the detailed structure of the myelin sheath is clearly resolved, displaying characteristic lipid-rich Raman spectral features. The myelin sheaths oriented differently exhibit varying signal intensities, indicating a strong dependence of SRP intensity on laser polarization. In Fig. 22(B), densely packed cytoplasmic organelles are observed. The SRP spectra (Fig. 22(D)) reveal a lipid peak at 2845 cm⁻¹ and a protein peak at 2930 cm⁻¹ for the myelin sheath, while the cytoplasmic signal is predominantly protein-based. The SRP signal from the medium is weak and independent of the Raman shift. A picture of the brain slice sample is shown in Fig. 22(E).

Figure 22.

SRP imaging of a mouse brain immersed in PBS at 2930 cm⁻¹ in different regions. Scale bar: 12 μm. (D) Spectrums of two representative spectrum in the mouse brain indicated with triangles in (A) (myelin sheath) and in (B) (cytoplasm). PBS signal is acquired at no sample region. (E) Image of the brain slice sample. Reprinted from [28] under a Creative Commons license.

Compared to water, glycerol’s high thermo-optic coefficient and low heat capacity present an opportunity to further enhance the sensitivity of SRP microscopy. Building on the successful imaging of 100-nm nanoparticles with high SRP contrast (Fig. 15(C)), we evaluated SRP imaging of viral particles. As illustrated in Fig. 23(A), individual varicella-zoster viral particles (approximately 180 nm in diameter) were distinctly resolved from the background, achieving an SNR of around 20. The SRP spectrum of a single virus (Fig. 23(B)) exhibited a peak at 2950 cm⁻¹, indicating a significant contribution from the nucleic acids at the virus’s core.

Figure 23.

SRP imaging of single virus. (A) SRP imaging of a single varicella-zoster virus immersed in glycerol-d8, 2950 cm⁻¹. Scale bar: 2 μm. (B) Single virus Raman spectrum at C–H region, acquired from a single virus in (A). Reprinted from [28] under a Creative Commons license.

To enhance the quality of SRP imaging of mammalian cells, glycerol-d8 was used as a mounting medium. The pancreatic cancer cell line MIA PaCa-2 served as the testbed (Fig. 24(A)). Glycerol-d8 replaced the PBS buffer to immerse the cells, thereby improving SRP contrast. Imaging in the high-wavenumber C–H vibration region revealed significant contrast from membranes and intracellular lipids. Phasor analysis was employed to segment the cellular compartments, successfully identifying up to six distinct components (Fig. 24(B)). Notably, the nuclear membrane was distinctly visible from the cytoplasm and nuclear matrix (Fig. 24(D)), underscoring SRP’s potential for studying fine membrane structures. This high contrast is likely due to the high thermo-optic coefficients and low heat capacities of membranes.

Figure 24.

(A) SRP imaging of fixed Mia-Paca2 cell immersed in glycerol-d8, at 2950 cm⁻¹. Scale bar: 5 μm. (B) Color coded chemical map through phasor analysis of (A). (C) SRP spectrum at C–H region for each component. (D) Mapping of each component from phasor analysis. Scale bar: 5 μm. Reprinted from [28] under a Creative Commons license.

The addition of glycerol enhances the sensitivity of SRP, enabling the detection of weak Raman bands in the fingerprint region. Figure 25(A) presents the SRP image of a 10 μm thick OVCAR-5 cancer tissue at 1650 cm⁻¹, targeting the Amide I band in proteins and the C=C vibration in lipids. High-quality spectra were obtained from the hyperspectral image stack. As shown in Fig. 25(B), lipid (region of interest 1, ROI 1) and protein (ROI 2) species are clearly distinguishable.

Figure 25.

(A) SRP spectroscopic imaging of OVCAR-5 tissue, cleared with glycerol-d8, at 1650 cm⁻¹. (B) Single pixel spectrum at the circled region of interests (ROI) in (A). Reprinted from [28] under a Creative Commons license.

Finally, a direct comparison between SRP and SRS at the same FOV on a SKOV3 cell in glycerol was conducted, with conserved average laser power and dwell time. The results are shown in Fig. 26. The representative cellular structures (lipid droplets, cell nuclei, and nucleolus) can be clearly resolved in both images. Comparison between profiles at the lines indicated by arrows on the 2980 cm⁻¹ frames show an SNR of 53 for SRP and 17 for SRS. Furthermore, the SRP image clearly shows sharper contrast for the intranuclear structures.

Figure 26.

Comparison between SRP and SRS images of SKOV3 cell immersed in glycerol-d8. (A) SRP and SRS images taken at three Raman shifts for lipid, protein and DNA. Scale bar: 10 μm. (B) Intensity profiles along the arrow-indicated lines in the SRP and SRS images at 2967 cm⁻¹. Reprinted from [28] under a Creative Commons license.

3.2. Fiber OPO-Based SRP Microscope

3.2.1. Justification

Benefiting from a robust, high-power, and low-noise solid-state OPO laser source, the SRP phenomenon has been successfully demonstrated and implemented in microscopy. However, to broaden the use of optical microscopy techniques, both portability and user-friendly design are essential. In the context of SRS, moving from solid-state to fiber lasers is desirable for translational applications. Despite their convenience, fiber lasers introduce significant drawbacks—most notably, elevated laser noise that prevents achieving an optimal signal-to-noise ratio (SNR) with heterodyne detection [50]. Although auto-balanced detection has been attempted [51], its complexity hinders widespread adoption. Meanwhile, strict light collection requirements and the need for high-NA optics in conventional SRS limit samples to thin, two-sided contact formats [52].

As discussed in previous sections, SRP microscopy probes the SRS-induced thermal lensing effect using a third beam [28]. While still relying on modulation transfer, the use of a separate probe beam circumvents intensity fluctuations in the local oscillator that can interfere with detection in SRS (Fig. 27(A)). In fiber laser–based SRS, even at high modulation frequencies, relative intensity noise (RIN) in the local oscillator (the pump in SRL or the Stokes beam in SRG detection) can easily overwhelm the SRS signal, as the laser intensity can fluctuate on the order of 10⁻⁶—comparable to the modulation depth from typical biological samples (Fig. 27(B)). By contrast, a low-noise continuous-wave (CW) laser exhibits much greater intensity stability at the modulation frequency, enabling shot noise-limited detection. The SRS-induced heat modulates the CW probe laser through thermal lensing, while the 10⁻⁶ level fluctuations in the pump or Stokes beam do not affect SRP detection via modulation transfer (Fig. 27(B)).

Figure 27.

SRP on high relative intensity noise fiber laser source. (A) Modulated fiber laser dual output Stokes and pump pulses induced SRP signal detected by the continuous-wave (CW) probe beam. The SRP signal shows a clear thermal heat accumulation and decay signature. without the influence from the SRS beam fluctuations. (B) The schematic of noise level comparison in SRS or SRP system. Reprinted from [53] under a Creative Commons license.

Another major advantage of SRP for translational applications is its relaxed requirement for light collection. As discussed earlier, thermal lensing detection relies on collecting the probe beam with a low-NA objective to isolate the modulated component in the far field. Unlike SRS, which necessitates a high-NA condenser, this low-NA approach allows a longer working distance and expands the sample space (Fig. 28(A)). An inverted microscope design can capitalize on this feature, making SRP compatible with most commercially available confocal epi-fluorescence microscopes.

Figure 28.

Schematic of a fiber laser simulated Raman photothermal (SRP) microscope. (A) Comparison of SRP and SRS design in the sample space. (B) Fiber laser-based SRP setup. AOM, acousto-optic modulator; BS, polarized beam splitter; COND, condenser; DM, dichroic mirror; L, lens; OBJ, objective. PD, photodiode; SU, scanning unit; TL, translational lens. The red dashed line box illustrates the SRP optical contrast by detailed light path alternation due to the presence of the SRP-induced thermal lens. Zoom-in shows the long working distance air condenser that allows large format samples such as multi-well plate. Reprinted from [53] under a Creative Commons license.

Combining these advancements, the fiber laser-SRP microscope was developed to address challenges associated with SRS in translational applications, including sensitivity limitations and stringent sample requirements [53].

3.2.2. Instrument

The setup employed a compact, dual-output picosecond tunable fiber laser (Picus Duo, Refined Laser Systems), capable of synchronized tuning over the Raman vibrational range of 700 cm⁻¹ to 3100 cm⁻¹ within 100 milliseconds. The pump and Stokes beams were modulated by acousto-optic modulators (AOMs) for heterodyne detection. A low-NA air condenser was used for light collection, enabling long working distances and accommodating diverse sample types, including multi-well plates and thick tissues (Fig. 28(B)).

Signal detection was achieved via a third CW probe laser, which interacted with the thermal lensing effect induced by Raman absorption. The system employed a lock-in amplifier to demodulate the SRP signal, enhancing sensitivity and robustness. The inverted microscope design facilitated the imaging of live and fixed samples in both aqueous and glycerol-based environments without requiring complex optical adjustments.

3.2.3. Results

The SRP system achieved a two-order-of-magnitude improvement in the signal-to-noise ratio (SNR) compared to fiber laser-based SRS without auto-balanced detection, and one-order-of-magnitude over with auto-balanced detection (Figs. 29(A)–29(C)). Sensitivity was demonstrated through detection limits of 13.77 mM in the carbon–deuterium (C–D) region (Fig. 29(D)) and 11.13 mM in the carbon–hydrogen (C–H) region (Fig. 29(E)).

Figure 29.

SNR comparison of SRP (A), SRS with auto-balanced detection (B), and SRS without balance detection (C) on DMSO. Pump power on sample: 28 mW, no modulation. Stokes power on sample: 90 mW, 50% duty cycle modulation. Probe power on sample: 23 mW. Dwell time: 20 μs/pixel. (D) Spectral fidelity (left) and intensity-concentration relation (right) of serial diluted dimethyl sulphoxide-d6 (DMSO-d6) dissolved in DMSO. Data represents 400-pixel averages from measurements with a 20 μs pixel dwell time. (E) Spectral fidelity (left) and intensity-concentration relation (right) of serial diluted DMSO dissolved in DMSO-d6. Data represents 400-pixel averages from measurements with a 20 μs pixel dwell time. Reprinted from [53] under a Creative Commons license.

Fiber laser SRP imaging was performed on fixed cells immersed in PBS. Hyper-spectral imaging (Fig. 30(A)) in the C–H region (2790–3030 cm⁻¹) enabled phasor segmentation of major cellular compartment, generating organelle maps (Fig. 30(B)) of mitochondria, membrane, cytoplasm, nucleus, and ERs. The inverted microscope configuration enabled contact-free imaging, minimizing contamination risks and streamlining sample changes.

Figure 30.

T24 cancer cell imaging in aqueous (PBS) and thermal enhancement medium (8 M Urea). (A) Fixed T24 cell imaging in PBS immersion condition. Scale bar: 8 μm. (B) Phasor segmentation results generated from hyperspectral dataset in (A). (C) Imaging of the newly synthesized lipid droplets of the OA-d34 treated T24 bladder cancer cell at 2105 cm⁻¹ and protein and lipids at 2930 cm⁻¹. Urea solution was used as the immersion medium. Scale bar: 10 μm. Reprinted from [53] under a Creative Commons license.

Glycerol or urea immersion enhanced SRP contrast by exploiting its high thermal-optic coefficient and low heat capacity. Lipid metabolism in OA-d34-treated cells was visualized (Fig. 30(C)), revealing cytoplasmic lipid distributions. Imaging in the silent C–D vibrational region (2105 cm⁻¹) complemented observations in the C–H region, providing insights into newly synthesized lipids and their spatial colocalization with proteins. Enhanced sensitivity facilitated the detection of subtle spectral differences, enabling detailed analysis of lipid droplet polarization and molecular orientation.

The fiber OPO-based SRP microscope demonstrated exceptional capability for volumetric imaging of rat brain tissue cleared with an 8 M urea solution. Two color imaging distinguished between cell bodies and extracellular matrix, offering histological insights into brain architecture (Fig. 31(A)). Hyperspectral imaging in the C–H region enabled the decomposition of biomolecular components, such as cholesterol, proteins, and fatty acids, with subcellular specificity (Fig. 31(B)). Depth-resolved imaging extended to over 200 μm (Figs. 31(C) and 31(D)), revealing intricate neural networks, including myelinated axons and cellular connections. The use of thermal-enhancing media further amplified signal strength, making SRP a powerful tool for non-invasive exploration of brain tissue in three dimensions. These capabilities highlight the system’s potential for advancing neuroscience research and understanding brain pathologies.

Figure 31.

Volumetric SRP imaging of tissue clearance rat brain. (A) Two-color SRP histology of neuron inside 300 thick rat brain slice. (B) Biomolecule decomposition by hyper-spectral SRP in C–H region. Scale bar: 5 μm. (c) Depth-resolved SRP imaging of urea-cleared rat brain at 2930 cm⁻¹. Depth: 200 μm. Scale bar: 10 μm. (d) Average intensity of each frame in the depth-resolved image in (C). Reprinted from [53] under a Creative Commons license.

3.3. OPA-Based Single Pulse SRP Microscope

Zhu et al. [54] recently advanced SRP microscopy by implementing a single-pulse configuration (spSRP) using optical parametric amplifier (OPA) lasers, which provide significantly higher peak powers (70-fold before chirping, 5-fold after chirping) than conventional OPO systems while operating at lower repetition rates (1 MHz). To maximize SRS excitation efficiency under these high peak power conditions while minimizing photodamage, the authors employed extensive pulse chirping to 30 picoseconds, guided by theoretical modeling of SRS saturation effects. Combined with radially segmented balanced detection for noise suppression, spSRP achieved a single-pixel limit of detection of 890 μM for DMSO—representing a 44-fold improvement over conventional SRS microscopy and 2.5-fold improvement over OPO-based SRP. The modulation-free single-pulse approach enabled high-speed hyper-spectral imaging at 10 frames per second with minimal photodamage, allowing tracking of lipid droplet dynamics in live cells. Notably, the enhanced sensitivity of spSRP enabled the first direct, label-free visualization of cholesterol-rich membrane domains in intact HeLa cell plasma membranes, with structures co-localizing with caveolin immunofluorescence and exhibiting characteristic cholesterol spectral signatures at 2875 cm⁻¹. This work demonstrates that OPA-based SRP microscopy can overcome the sensitivity limitations of conventional coherent Raman techniques to detect subtle chemical variations in nanoscale biological structures.

4. DISCUSSION

4.1. Advantages

Compared to SRS measurements, SRP offers 4 major advantages. First, it provides superior sensitivity. Due to the significantly higher modulation depth in SRP measurements, the detection sensitivity is approximately ten times greater than that of conventional SRS. Such sensitivity improvement translates to 10-fold LOD improvements, as demonstrated in both OPO-based SRP systems. It also enables the Raman imaging of smaller plastic nanoparticles, as demonstrated in OPO-based SRP. The SRP images of biological samples also show better image quality compared to SRS images in the same field of view.

Second, SRP allows for low NA forward or epi signal collection. As a scattering measurement, unlike the absorption measurement in SRS, SRP does not require complete light collection. Thus, a low NA lens is sufficient to gather the light signal and produce quality images. This flexibility accommodates various sample formats, such as 96-well plates and flow cytometers, and simplifies microscope operation. This advantage also relaxes the operational difficulty of coherent Raman measurement, making SRP potentially a good candidate for commercialized coherent Raman imaging product.

Third, SRP is less affected by ultrafast laser noise. Since the ultrafast lasers used for SRS excitation are not directly measured in SRP, their noise levels have minimal impact on the final image noise. This characteristic enables the use of high-performance, albeit noisy, and ultrafast laser sources in coherent Raman measurements. These new lasers may bring unique advantages, from the perspective of laser power, laser repetition rate, tuning range and speed, cost, etc.

Fourth, SRP offers an advantage over SRS in terms of spatial resolution. By introducing a probe laser, the ideal effective point spread function (PSF) becomes the product of the PSFs of all three lasers (pump, Stokes, and probe). Additionally, a probe laser with a shorter wavelength can result in a sharper SRP effective PSF. In the OPO-based SRP, using a 1.49 NA objective, the ideal resolution achieved was approximately 167 nm with a 765 nm probe laser and around 137 nm with a 522 nm probe laser. It is important to note that achieving such ideal resolution improvements require high modulation frequencies (>MHz), as thermal diffusion during the modulation period can enlarge the thermal lens, thereby compromising resolution. At very low modulation frequencies, the resolution degrades to that of SRS. As in the OPO-based SRP, with a modulation frequency of 125 kHz, the spatial resolution improvement has been significant, from approximately 217 nm with SRS to about 194 nm with SRP.

4.2. Features

SRP keeps high spectral fidelity. Based on the previous theoretical analysis, the spectral information provided by SRP is very faithful compared to the SRS spectrum, which has been confirmed with both bulk liquid and particle samples. It is important to note that the SRP intensity is also proportional to ω_SRS. This factor may affect the fidelity over a very broad range of spectrum, however, within a narrow spectral window, this factor will not impact the SRP spectrum significantly.

Background is another factor that may affect spectral and image quality. In SRP, only the non-parametric processes that produce heat are measured, so the commonly observed non-resonant background in CARS and SRS do not exist in SRP. However, other non-parametric processes, such as the electronic absorption of naturally occurring pigments [55] and the overtone or combinational modes of vibrational transitions [56,57]. There are several types of pigments that appear at different compartments of biological systems, NADH, heme, FAD, etc., of which the absorption spectrum usually lays at UV to visible region. In practice, one can usually photobleach the pigments before the imaging, to avoid spectral distortion caused by pigment absorption. In addition, the overtone transition of C–H stretching modes also resides in the NIR region, yielding a weak linear absorption background. Originating from chemical bonds, such overtone absorption could not be photobleached. However, such background shows linear power dependence, instead of quadratic power dependence as in SRS. Therefore, the signal to background ratio could be improved with elevated excitation peak power. On the other hand, some higher order nonlinear absorption processes may occur once the laser peak power is too high. This is especially pronounced when applying femtosecond excitations. Simple reduction of laser peak power can reduce such background and improve SBR. Both types of background could be spectrally resolved based on its spectral bandwidth. Single color absorption background is roughly constant when spectrally scanned within a small range (e.g., 200 cm⁻¹), so it could be easily removed by subtracting the offset of the spectrum. Multicolor multiphoton absorption shows different spectral features with different spectral scanning techniques. With OPO-tuned spectral scanning, multicolor multiphoton absorption background is roughly constant within a small range; and with spectral focusing technique, multicolor multiphoton absorption background usually resembles the cross-correlation function between pump and Stokes laser pulses.

A notable characteristic of SRP, in contrast to SRS, is its reliance on the properties of the sample environment. Consequently, lipid-rich regions in cellular images, such as lipid droplets and lipid bilayers in membranes, display enhanced contrast compared to SRS images. It enables signal augmentation by addition of bio-compatible media in biological samples, such as glycerol [58] and urea [25]. These media serve as both tissue/cell clearance agents that improve the image quality of scattering measurement, and contrast enhancement agents that improve the signal intensities. This attribute also presents opportunities to further amplify the SRP signal intensity by using a carefully engineered medium, such as critical Xenon [59], which has been shown to achieve approximately a 400-fold enhancement in visible photothermal microscopy. For aqueous samples, the theory predicts 4-fold signal reduction compared to glycerol or 10-fold comparing to DMSO, overall making the performance close to conventional SRS. However, a major contribution of the signal reduction is the high heat capacity of water which leads to inefficient heating. Therefore, introducing a higher peak power excitation laser may deliver enough energy to sufficiently heat up water and recover the signal back. However, it is crucial to recognize that the sample-dependent nature of SRP signals may complicate quantitative analysis in highly heterogeneous environments.

The choice of probe laser is crucial to the performance of the whole SRP system, with two main factors for consideration. First, the laser noise level at the range of hundreds of kHz to several MHz should be as low as possible. As the laser beam of which the intensity is finally measured, the noise level of the probe laser determines the noise level of the whole measurement. Note that, the root-mean-square (RMS) noise parameter that’s usually provided by the laser manufacturer is measured at a relatively low frequency (mostly below 100 Hz), which could not faithfully represent the laser noise level at the measurement frequencies. Second, the temporal coherence of the laser should not be very good. High temporal coherence and long coherent length usually lead to strong interference at any reflective or refractive index mismatching interfaces, which results in artifacts in the images that overwhelm any weak signals. This is a very severe issue with narrow linewidth continues wave lasers. Considering both factors, mode-lock lasers are good candidates for the probe laser, given their low noise and low temporal coherence. However, appropriate chirping must be applied to reduce the photodamage induced by high peak power mode-lock lasers.

5. COMPARISON WITH OTHER MODALITIES

Targeting the molecular vibrational information of the specimen, there have been a few types of spectroscopic and microscopic techniques developed. While sharing many common properties, each of these techniques also have their own advantages and applications. And many of their frontier advances could inspire further improvement of SRP technique. In the following section, we compare SRP microscopy with a collection of vibrational imaging techniques, including SRS microscopy, MIP microscopy, shortwave infrared photothermal microscopy (SWIP), stimulated Raman photoacoustic (SRPA) microscopy, atomic force Stimulated Raman microscopy (AFM-SRS), and impulsive stimulated Raman scattering microscopy (iSRS).

5.1. Comparison with Stimulated Raman Scattering Microscopy

SRP and SRS are rooted in the same photophysical processes and provide similar information in many cases. However, their different accesses to spectroscopic information render different features and properties. Therefore, it is important to discuss the respective advantages and disadvantages, so appropriate techniques are chosen at a given application scenario. The comparison between the key features of two techniques is summarized in Fig. 32.

Figure 32.

Comparison between SRP and SRS. N/A, not applicable. The light propagation of the probed beam is illustrated in geometry (i.e., pump beam in SRL, probe beam in SRP). Deeper color refers to weaker light field. Red circle refers to the SRS focus.

Since the first report of high-speed SRS imaging in 2008 by the Xie group [14], a series of technical innovations have been explored. On detection sensitivity, different mechanisms have been harnessed to either reduce the noise floor or increase the signal intensity, including quantum-enhanced SRS [60,61], electronic pre-resonance SRS [62–64], plasmon-enhanced SRS [65] and stimulated Raman excited fluorescence [66,67]. On imaging speed, video-rate SRS microscopy has been achieved [68] with the fastest scanning axis on the spatial domain; or high speed spectrum acquisition with the fastest scanning axis on the spectra domain [40,69] for better spectral quality. High speed SRS imaging with sparse sampling reconstruction has also been reported with good performance [70]. On spatial resolution, different methods have been proposed and demonstrated by advanced algorithms [71], coupling with fluorescence superresolution techniques [67], saturation excitation [72], expansion microscopy [73], or directly applying shorter excitation wavelengths [17,74]. On imaging depth, tissue clearing and Raman dye (Fig. 33) [25] helps achieve SRS imaging through 200 μm tissue. On data analysis methods, different algorithms have been published to improve certain performance of SRS, such as Furnace [17] for denoising and resolution improvement, A-PoD [71] for resolution improvement; or help translate the spectral result to chemical information, such as multivariate curve resolution (MCR) analysis [75,76], least absolute shrinkage and selection operator (LASSO) unmixing (Fig. 34) [77]. These innovations lay the foundation for broader applications of SRS microscopy in biology settings.

Figure 33.

Example of SRS images for tissues. Clearing-enhanced volumetric chemical imaging of glioblastoma in mouse brain. (A) Representative 2D images of the tumor in the whole coronal slice at increasing imaging depths. (B) Tumor margins outlined from (A). (C) and (D) Zoomed-in 2D images of the white-boxed region in (A) at increasing imaging depths. (E) Protein/lipid ratio-metric image at 200-μm depth. (F) Plots of protein/lipid signal ratios as a function of imaging depths in tumor and normal tissues. Reprinted with permission from Wei et al., Proc. Natl. Acad. Sci. USA 116, 6608–6617 (2019) [80], Copyright 2019 by the Proceedings of the National Academy of Science of the United States of America.

Figure 34.

LASSO unmixing helps interpret hyperspectral dataset of U87 cell at C–H region. (A) Raman spectra of BSA, glucose, TAG, cholesterol, and purified RNA from cells. (B) Normalized SRS spectra of BSA, glucose, TAG, cholesterol, and purified RNA from cells. (C) Schematic illustration of pixel-wise LASSO spectral unmixing for chemical mapping generation. N_x, N_y, and N_λ represent dimensions of the hyperspectral image, and D, C, and S stand for the data matrix, concentration matrix, and spectral reference matrix, respectively. The circled elements represent the dominant components in each row. (D) Representative hyperspectral SRS image (sum of all channels), mapped protein, carbohydrate, fatty acid, cholesterol, and nucleic acid images, as well as the merged image of metabolites mapping for U87 cells through LASSO, LS fitting, or MCR separation processing. Reprinted from [77] under a Creative Commons license.

SRP spectroscopy is an indirect measurement of scattering or deflection signal caused by thermal effect of SRS, while SRS spectroscopy is a direct measurement of light absorption. The key difference between scattering measurement and absorption measurement lays on the requirement of light collection. Scattering processes do not change the total photon count, instead redistribute the light energies to different directions. Therefore, measuring part of the total light field could already capture the scattering signal very well, and different strategies measuring different portions of the outcoming light fields yield different properties of the measurement, e.g., interferometric scattering (iScat) provides superior sensitivity to small particles on the surface [78], forward scattering provides better signal for large scale features [33]. To the contrary, absorption measurement changes the total photon counts but does not redistribute the light energies. Therefore, one must collect as much as possible of the probed laser beam to avoid any scattering-like artifacts (e.g., cross phase modulation (XPM) and photothermal) and faithfully measure the absorption-originated signals. When performing high resolution measurements, such a rigid requirement leads to the necessity to use a high NA condenser for the measurement to avoid background. As a result, the allowed space between two high NA optics is very confined.

The thermal processes measured in SRP evolve at microsecond scale, while SRS interactions are instantaneous. The slow evolution of thermal processes brings the benefits of SRP measurement that the temperature rise of the sample could build up upon the pseudo-continues excitations induced by a burst of laser pulses, which help improve the peak signal intensity hence sensitivity. Whereas in SRS, the instantaneous interaction could not build up, limiting the modulation depth of the measurement. On the other side of the coin, the slow dynamics of thermal processes also leads to long cooling time for the system to recover back to equilibrium, as a result high frequency modulation in SRP measurement reduces signal intensity, as the heat will accumulate rather than sufficiently dissipate. Whereas in SRS, the instantaneous nature of the interaction allows high modulation frequencies up to half of the repetition rate of the excitation laser sources [79,80]. High modulation frequency helps mitigate the laser intensity noise, making it possible to reach shot-noise limitation.

SRP signal intensity is affected by environmental factors, while SRS is always linearly proportional to the sample concentration. As shown in Eq. (26), the refractive index, heat capacity, and thermo-optic coefficient play their role in the formation of SRP signals. Such a feature brings the possibility of specific optimization of these terms could help improve the signal intensity and image quality of SRP imaging, which is not possible in SRS measurement. In native cellular environments, such a feature also helps enhance the signal intensity in some organelles such as lipid droplets. On the other side of the coin, the signal intensities of SRP in heterogeneous environment are also affected by the environmental heterogeneity, which brings difficulties comparing signal intensities across different environment, e.g., different organelles in cells. Whereas in SRS, the absorption signal is always proportional to the number of molecules inside the laser focus, which makes it perfect for quantitative measurements.

The two techniques have different origins of backgrounds. In SRP, a continuous wave (or pseudo-continues wave) probe laser is used to probe the real part of index changes. Such index changes are usually induced by thermal effect, meaning that any dissipative processes happen between pump and/or Stokes beams and the sample, other than SRP, leads to background. These usually include linear absorption of either excitation laser beams and/or some nonlinear absorption processes when the peak power is high. Details about the composition and corresponding method to suppress background in SRP have been discussed previously. In SRS, there are 3 major origins of background, i.e., cross phase modulation (XPM), transient absorption (TA), and photothermal (PT). XPM originates from the optical Kerr effect [81], which means the refractive index change induced by an optical electric field. In SRS measurements, the modulated laser field (e.g., Stokes in SRL measurements) could induce refractive index changes instantaneously, and such index changes could lead to propagation direction change of the probed laser beam (e.g., pump beam in SRL measurements), making a scattering-like background. This background is not present in SRP measurement, because the XPM processes are instantaneous, and the probe beam in SRP will only experience such index changes at extremely low duty cycle (typically <0.2%), thus the overall impact is much suppressed. PT in SRS background originates from the thermal effect caused by the modulated laser beam [82]. It is also a scattering-like background as in XPM. TA refers to the process where the first pump beam of laser alters the electronic states of the target molecules, and the consequent optical density changes of the molecules alters the absorption of the second probe beam of laser [83–86]. In SRS, the modulated beam may serve as the pump beam that imprints its modulation on the probed beam through TA process, showing as TA background. TA background is essentially an absorption background, so it cannot be avoided optically. All three types of backgrounds in SRS display could be spectrally resolved, and the specific spectral shape depends on the spectral scanning technique.

There have been many advancements, especially data analysis methods of SRS that improve its performance, many of which could be transferred to SRP with good potential. For detection sensitivity, 2 orders of magnitude improvement has been demonstrated with electronic pre-resonance effects [62–64] for SRS, and we expect SRP could also benefit from such an interaction enhancement, hence sensitivity improvement. For spatial resolution, many established algorithms that have been proven working for SRS, such as A-PoD [71] and Furnace [17], should also work for SRP, given the similarity between the two modalities. For spectral information interpretation, many of the established methods, such as phasor analysis [87], MCR unmixing [75], and LASSO unmixing [77], could be readily applied to SRP.

Overall, SRP is more sensitive and more compatible with a broader spectrum of devices such as 96-well plate, noisy laser sources. Therefore, we expect broader applications of SRP microscopy in the future, especially by non-experts. SRS is still a very powerful technique that has advantages in specific cases, such as where quantitative analysis is crucial.

5.2. Mid-Infrared Photothermal Microscopy

MIP microscopy, also called optical photothermal infrared (O-PTIR) microscopy, is an advanced chemical imaging technique that measures the photothermal effect induced by an infrared absorption of chemical bonds. Since the first report of MIP microscopy in 2016 [34], a broad spectrum of different photothermal detection methods have been adopted to MIP to improve the performance in different aspects, such as thermal lensing [34], dark field scattering [88], iScat microscopy [89,90], phase measurement [91,92], or couple with fluorescence detections [43,44] (Fig. 35). Thanks to the performance improvement brought by these innovations, MIP microscopy has been widely applied (Fig. 36) to bacteria [93] and virus [90] fingerprinting, live cell imaging [34], protein secondary structure characterization [94], etc.

Figure 35.

Representative measurement schemes of MIP microscopy. (A) Thermal lensing technique. Reprinted from [34] under a Creative Commons license. (B) Dark field scattering technique. Reprinted with permission from Zong et al., ACS Photonics 2021, 8, 11, 3323–3336 (2021) [91], Copyright © 2021 by the American Chemistry Society. (C) Fluorescence detection. Reprinted with permission from Zhang et al., J. Am. Chem. Soc. 143, 30, 11490–11499 (2021) [44], Copyright © 2021 by the American Chemistry Society.

Figure 36.

Representative applications of MIP microscopy. (A) MIP images of C. elegans at different wavenumbers. Reprinted from [34] under a Creative Commons license. (B) MIP images of E. coli. Reprinted with permission from Zong et al., ACS Photonics 2021, 8, 3323–3336 (2021) [88], Copyright © 2021 by the American Chemistry Society. (C) Fluorescence detected MIP image of Shigellaflexneri bacteria expressing GFP. Reprinted with permission from Zhang et al., J. Am. Chem. Soc. 143, 11490–11499 (2021) [44], Copyright © 2021 by the American Chemistry Society.

MIP microscopy provides high fidelity infrared absorption spectral information of the sample compared to conventional Fourier transform infrared (FT-IR) spectroscopy. Additionally, the spatial resolution in MIP reaches the diffraction limit of the visible probe laser, which is much sharper than the excitation IR beam. Combining the high IR absorption cross sections and highly sensitive photothermal detection, MIP could also provide micro-molar level sensitivity.

Both measuring scattering-like signals, MIP and SRP share many common features on their detection side, and it is possible to translate most of the technological advancements about photothermal detections between each other. A broad collection of detection schemes has been demonstrated in MIP, such as thermal lensing detection, fluorescence detection, and dark-field detection. Among these techniques, thermal lensing detection has been demonstrated in SRP. Fluorescence detection could be possibly implemented for SRP detection, for their unique advantages in measurement specificity. Dark-field detection is also a good option for SRP given its high sensitivity for weakly scattering signals.

Different excitation sources in two modalities bring most of the major differences between SRP and MIP. In SRP, the heat deposition originates from Raman scattering, while in MIP it is from infrared absorption. Therefore, these two modalities show different selection rules for their accessible vibrational transitions, hence different sensitivity for different molecules. Generally, polar bonds are stronger in MIP, while Raman processes favor non-polar bonds.

The cross sections of IR transitions are usually larger than their Raman counterparts. Therefore, linear absorption of mid-IR light is sufficient to generate enough heat for photothermal imaging, while high peak power, ultrafast lasers are necessary to coherently excite Raman process and generate measurable amount of heat in SRP. As a result, the sensitivity of MIP is usually better than SRP. It is also easier to achieve wide field imaging in MIP, while the spatial multiplex potential of SRP is limited due to the requirement for high peak power density to achieve efficient nonlinear excitations.

In SRP, the spectral region of the excitation lasers lay at visible to near-infrared (vis-NIR, 400–1100 nm), while in MIP, the excitation laser is at the mid-infrared (mid-IR, 3–20 μm) region. The SRP excitation lays at water-transparent region, while mid-infrared light could be strongly absorbed by water. Therefore, it is possible to measure SRP deep inside the sample, while the penetration depth of MIP imaging is very limited. Such water absorption also brings strong water absorption background in MIP.

The SRP excitation wavelengths are very close to the probe wavelength, which is usually within the vis-NIR region as well. Therefore, it is easy to collinearly align all three laser beams and achieve laser scan for SRP. In MIP, there is a huge wavelength mismatch between the excitation laser (mid-IR) and the probe laser (vis-NIR), thus one must use reflective optics to achieve collinear alignment, which limits the available NA and/or transmission. Laser scan could still be achieved with non-collinear, counter-propagate scheme [95] in MIP with high performance, but at the cost of system complexity.

As a nonlinear process induced by vis-NIR lasers, SRP only happens within a very confined space, which provides excellent spatial resolution and z-sectioning ability. On the other hand, mid-IR light in MIP could only be focused down to a few micrometers due to the long wavelength, and the linear absorption process lacks z-sectioning ability. As a result, the resolution of MIP is solely defined by its scattering detection part.

Overall, MIP is a very successful technique with many application demonstrations, that advantages especially in detection sensitivity. However, water absorption and background pose challenges for many application scenarios. SRP is a complementary technique that measures Raman active vibrational modes with minimal water absorption and background. Therefore, it is important to choose the appropriate technique based on the mode of interest, as well as the environment of the sample. The difference between SRS, SRP and MIP are summarized in Table 2.

Table 2.

Comparison between Different Vibrational Imaging Modalities

Feature	SRS	SRP	MIP (O-PTIR)
Mechanism	Stimulated gain/loss	Thermal lensing (Raman)	Thermal lensing (IR)
Excitation	Two NIR pulses	Two NIR pulses	Mid-IR + Visible Probe
Resolution	300 nm	200 nm	500 nm (Probe limited)
Sensitivity	mM	Sub-mM to mM	μM to mM
Water background	Minimal	Minimal	High
Quantitation	Excellent	Environment-dependent	Environment-dependent
Key advantage	Speed/linearity	Sensitivity/Low-NA collection	High sensitivity
Key applications	Quantitative measurement	Cleared tissue, isolated particles, and membranes	Isolated particles, single layer cells, and bacteria

5.3. Shortwave Infrared Photothermal Microscopy

Shortwave infrared (SWIR) usually refers to the spectral window roughly from 0.9 to 1.7 μm. Within this window reside the first- (1.6–1.8 μm) and second- (1.2–1.3 μm) overtones and some related combinational bands of C–H stretching modes. In addition, the second- (1–1.3 μm) and third- (1.5–1.9 μm) near-infrared window where the water absorption as well as tissue scattering are minimized, also provides access to a deeper penetration of ballistic photons.

Appreciating the special properties of this spectral window, short-wave infrared photothermal (SWIP, Fig. 37) has been demonstrated by the Cheng group to improve the chemical imaging depth [57]. In this work, 1730 nm nanosecond laser was used to excite the first overtone of C–H bond, and another 1310 nm continuous wave laser was then sent to the sample to probe the photothermal effect. With both long wavelength lasers, the absorption and scattering are much reduced, and 800 μm imaging depth has been demonstrated. Such a high imaging depth is still limited by the objective working distance, instead of the strong scattering of the sample.

Figure 37.

SWIP microscopy. (A) Scheme of SWIP microscopy. (B) SWIP images of mouse brain slice atup to~1 mm depth. Reprinted with permission from Ni et al., Nat. Photonics 18, 944–951 (2024) [57]. Copyright © 2024 by Springer Nature.

In this work, efforts have been made to differentiate lipid and background water based on their different thermal relaxation dynamics. However, single wavelength excitation could only provide very limited chemical information. A tunable excitation could certainly be beneficial to provide more specificity of this technique, but it is still limited by the intrinsically simple features of vibrational transitions within this spectral region, where most of the peaks are related to C–H stretching modes. In comparison, SRP provides access to Raman fingerprint information of the sample, which is crucial in many applications.

The principle of SWIP that achieves high imaging depth could also be applied for SRP. By switching the laser wavelengths of SRP from NIR to SWIR, one could reasonably anticipate a great improvement of imaging depth. However, special care should be taken of the chromatic aberrations between the pump and Stokes of SRS excitations, given the long travel distance through complex tissue samples.

Other than SWIP, the second overtone of C–H stretching modes at 1.2 μm that resides at the second NIR window could also be applied as the excitation source for photothermal chemical imaging. Le and coworkers reported overtone photothermal (OPT) microscopy [56], where they modulated a femtosecond optical parametric oscillator (OPO) laser to periodically excite the second overtone transitions and scanned the spectrum by directly tuning the output wavelength of the OPO laser. Based on their different compositions of CH₂ and CH₃ moieties, protein and lipid could be spectrally differentiated.

5.4. Stimulated Raman Photoacoustic Microscopy

SRPA microscopy is an interesting sibling technique of SRP microscopy. As in SRP, SRPA also measures the sample response of SRS excitations. Both techniques use ultrafast laser pulses trains to excite the SRS transitions and heat up the sample. Afterward, instead of measuring the thermo-optic effect with a probe laser beam, SRPA microscope measures the acoustic wave induced by thermal expansion with an ultrasonic transducer.

SRPA microscopy was reported in 2010 by Yakovlev and coworkers (Fig. 38) [96], where in this work, they spectrally shaped a very high energy (>20 μJ) laser pulse train and excited the SRS process of chloroform in a capillary tube at 1 kHz repetition rate. An ultrasonic transducer was coupled to the sample through distilled water to measure the outcoming acoustic waves. The maximum amplitudes of the acoustic waves were recorded with an oscilloscope to generate the contrast. By spectrally scanning the SRS excitation lasers, a spectrum of SRPA intensities as a function of excitation wavenumber was generated with close resemble to the spontaneous Raman spectrum, which confirmed the SRS origin of signals.

Figure 38.

SRPA microscope. (A) Scheme of SRPA microscopy. (B) SRPA images of chloroform in glass capillary. Reprinted with permission from Yakovlev et al., Proc. Natl. Acad. Sci. USA 107, 20335–20339, (2010) [96]. Copyright © 2010 by Proceedings of the National Academy of Science of the United States of America.

SRPA microscopy promises deeper penetration [97]. Due to the long wavelength of acoustic waves, the loss of acoustic wave propagation in tissues is much lower than its optical counterpart. Therefore, SRPA is less affected by the tissue scattering, which could potentially lead to deeper penetration. To realize such an advantage, one must focus SRS excitation lasers deep into the highly scattering medium as well, thus adaptive optics may be necessary. SRPA also has potential in high-sensitivity detection, thanks to the recent development of photoacoustic detection methods which could reach a minimal pressure of 800 Pa [98].

However, it has also been shown that during non-radiative decay of photo-excitations, only a small portion of the energy is converted into acoustic wave, while most of the energy is thermalized and dissipated through diffusion. PA process can be modeled as essentially a heat engine that transfers thermal energy to mechanical work [99], of which the efficiency is fundamentally limited by Carnot limit:

$${\eta }_{\mathit{max}}=1-\frac{T_c}{T_h},$$ (27)

where ${\eta }_{\mathit{max}}$ denotes the Carnot limit efficiency, $T_c$ and $T_h$ is the environmental temperature and maximal temperature after excitation, respectively. For SRS excitations, the temperature change $\left(T_h-T_c\right)$ in biological imaging setting is usually lower than 1 degree, therefore the efficiency is lower than even 1%. Such low efficiency will negatively impact the detection sensitivity of SRPA microscopy. This also echoes with the comparison between PA and PT measurement done by Ni and coworkers [57], where they concluded that PT is more sensitive than PA.

To date, few works have been published about SRPA microscopy, majorly due to its lack of sensitivity. More explorations are needed to realize the promised advantages of SRPA and quantitatively compare its performance with other coherent Raman measurements.

5.5. Atomic Force Stimulated Raman Microscopy

Atomic force Stimulated Raman microscopy (AFM-SRS) is another type of sample response measurement methods for coherent Raman spectroscopy. In AFM-SRS, SRS excitation lasers are first sent to the sample to heat up the sample through Raman resonant transitions. An AFM is then applied to measure the consequent thermal expansion of the sample.

AFM-SRS is first reported by the Kumar group (Fig. 39) [100]. In this work, two continues wave lasers with low laser power ( 40 μW) were modulated at different frequencies, such that the beating frequency between the two resonated with one of the intrinsic frequencies of the AFM cantilever. By measuring the SRS induced force at the beating frequency, coherent Raman imaging for the 1143 cm⁻¹ and 1183 cm⁻¹ modes of azobenzene thiol was achieved with unprecedented resolution of 10 nm. Later the same group reported AFM-SRS with plasmonic enhancement [101]. Very high sensitivity down to 42 molecules has been achieved.

Figure 39.

AFM-SRS microscope. (A) Scheme of AFM-SRS microscopy. (B) AFM-SRS images of azobenzene thiol. Reprinted from V. A. Tamma, et al., “Stimulated Raman spectroscopy and nanoscopy of molecules using near field photon induced forces without resonant electronic enhancement gain,” Appl. Phys. Lett. 108 (2016) [100], with the permission of AIP.

AFM-SRS is a very powerful technique in terms of its high sensitivity and high resolution, benefiting from the powerful AFM measurement technique. However, as a near field method, the mechanical tip also limits the imaging speed. And AFM only measures features at the surface of the sample, while incapable of measuring anything inside. These limitations confine the applications of AFM-SRS to thin and/or surface samples, such as studying surface characters of biological systems, for example, the SRS imaging of cellular surface proteins. In addition, the complex modulation and demodulation scheme as well as the delicate gold coated tip make AFM-SRS very difficult to perform.

5.6. Impulsive Stimulated Raman Scattering Microscopy

Impulsive stimulated Raman scattering (iSRS) [102] also measures the sample response to SRS excitations, at femto- to pico-second time scale. In iSRS, a spectrally broadband and temporally short pulse of laser is used to excite all vibrational modes of which the period of oscillation is longer than the pulse duration, or alternatively, of which the vibrational frequency is within the bandwidth of the excitation laser. Under this condition, any of the two spectral components within the excitation pulse could drive coherent Raman modes that resonate with their difference frequency. Collectively, such short excitation produces vibrational coherence of all accessible Raman modes, which leads to a periodic modulation of refractive index of the sample. A delayed ultrafast pulse could sense the index modulation with high temporal resolution, and the following Fourier transform of the time domain readout yields the coherent Raman spectrum.

Early spectroscopic work of iSRS dated back to 1985 by the Nelson group [104]. Different measurement scheme with potential to be applied in microscopy, including Kerr lensing [105], spectral shift [106], phase shift [107], etc. have been explored in more recent years, and eventually lead to iSRS imaging [103,107]. In 2019, Raanan and coworkers reported the first high speed iSRS microscopy (Fig. 40) [103]. In this work, they use a Dazzler to achieve high speed delay tuning between the pump and probe pulses, and both Kerr lensing and spectral shift contrast mechanisms are demonstrated for imaging. Carbamazepine dehydrate and anthracene crystals are used as the test bed. However, the applications of this technique are not well-explored, with most of the demonstrations focusing on crystal and/or pure liquid samples.

Figure 40.

iSRS microscope. (A) Scheme of iSRS microscopy. (B) iSRS spectroscopic imaging of carbamazepine dehydrate crystal. Reprinted with permission from [103] © Optical Society of America.

iSRS advantages in low-wavenumber coherent Raman measurement, where conventional SRS microscopy usually suffers from difficult spectral separation of pump and Stokes beams when measuring low-wavenumber Raman mode <200 cm⁻¹. High-wavenumber iSRS, such as Amide I regions (1600 cm⁻¹), C–D (2100 cm⁻¹), and C–H (2900 cm⁻¹), is more challenging due to its requirement of ultrafast laser sources (<10 fs) and sophisticated dispersion compensation over a large range of spectrum.

For SRP, on the other hand, it is not required to spectrally separate pump and Stokes beams. Instead, a probe laser beam with flexible wavelength is measured. In addition, the excitation used in SRP could be spectrally focused at a specific wavenumber, hence a single wavenumber image could be generated in a single scan. While in iSRS, the complete temporal coherence measurement is required to extract any spectral information. However, low-wavenumber Raman excitations only deposit small amount of energy, which may pose challenge for low-wavenumber SRP measurement. Collectively, low-wavenumber SRP is an interesting direction that is worth exploring.

6. CONCLUSION AND OUTLOOK

SRP microscopy is an emerging imaging modality that provides high fidelity spectroscopic information with much improved detection sensitivity, along with other advantages features such as compatibility with low-NA collection elements and noisy ultrafast laser sources. The special measurement scheme also brings possibilities, such as higher spatial resolution, epi-detection that’s friendly to bulk and in vivo measurement. These features promise SRP to be potentially the next generation of coherent Raman measurement scheme, and further exploration will almost certainly bring new and powerful features that may help translation of SRP into a product and enable a broader spectrum of applications.

Looking ahead, there are several areas where SRP imaging performance can be significantly improved. First, there is a mismatch between modulation depth improvement and sensitivity enhancement. In the OPO-based SRP microscope, a greater than 500-fold increase in modulation depth was achieved, yet the limit of detection improvement was only around 10-fold, which is much lower than anticipated. Even with the application of an SHG probe laser with significantly reduced low-frequency noise, this discrepancy persists, suggesting that laser noise is not the cause. This phenomenon may indicate that thermal dynamic fluctuations within the sample system may contribute to the elevated noise floor, warranting further investigation.

Second, other photothermal measurement schemes could be incorporated with SRP and bring their unique advantages to SRP measurement. Current SRP implementations all relay on the forward thermal lensing measurement, which is relatively simple and shows great performance in image quality. It’s also possible to measure SRP with epiphotothermal detection [33], which may enable in vivo measurement of SRP signals with high sensitivity and spectral fidelity; or iScat photothermal measurement [78], which may improve the detection sensitivity for small particles. Other measurement schemes, such as fluorescence detection [43,44] or nano-mechanical photothermal sensing [108], could also bring their advantages in detection selectivity or sensitivity, respectively.

In summary, SRP microscopy provides a versatile platform for sensitive chemical imaging across diverse biological and clinical contexts. SRP microscopy is uniquely positioned for single-virus metabolic profiling, offering the sensitivity required to resolve viral cores without the water-absorption background that limits MIP. It also holds great potential for imaging fine membrane structures, which exhibit enhanced contrast due to their high thermo-optic coefficients. Finally, SRP’s low-NA collection removes the rigid optical constraints of SRS, enabling label-free high-throughput screening in standard 96-well plates, which makes it a good candidate for anti-microbial susceptibility test.

Biographies

Yifan Zhu is currently a postdoctoral associate at Boston University, working with Dr. Ji-Xin Cheng. He received his Ph.D. degree in 2024 under the supervision of Dr. Ji-Xin Cheng. He joined the Cheng group in 2017 with a passion for novel spectroscopy. He has been focusing on the development of new spectroscopic imaging tools to see the unseen in materials and living systems. He demonstrated stimulated Raman photothermal microscopy for the first time, opening new access to Raman chemical information. He is also a part-time Chinese popular science writer, with >10k online subscribers.

Xiaowei Ge is currently an engineer at Apple. She graduated from Huazhong University of Science and Technology in 2017 and obtained her master’s degree at University of Dayton in 2019. From 2019 to 2024, she carried out her Ph.D. study at Boston University under the supervision of Dr. Ji-Xin Cheng. Her major research interest is the development of advanced coherent Raman imaging techniques and their biological applications. She also has a cute cat called “Stokes Raman.”

Hongli Ni is a postdoctoral associate at the Massachusetts Institute of Technology. Her current research focuses on large-scale volumetric microscopy for neural activity imaging. She received her Ph.D. from Boston University under the supervision of Prof. Ji-Xin Cheng. Her Ph.D. research focused on developing advanced label-free chemical microscopy, including stimulated Raman microscopy and photothermal microscopy. She demonstrated shortwave infrared photothermal microscopy for millimeter-deep, micron-resolution imaging for the first time, opening new possibilities for chemical imaging.

Ji-Xin Cheng attended University of Science and Technology of China (USTC) from 1989 to 1994. From 1994 to 1998, he carried out his Ph.D. study on bond-selective chemistry at USTC. As a graduate student, he worked as a research assistant at Universite Paris-sud (France) on vibrational spectroscopy and at the Hong Kong University of Science and Technology (HKUST) on quantum dynamics theory. After postdoctoral training on ultrafast spectroscopy at HKUST, he joined Sunney Xie’s group at Harvard University as a postdoc (2000–2003). Cheng joined Purdue University in 2003 as an Assistant Professor in Weldon School of Biomedical Engineering, promoted to Associate Professor in 2009 and Full Professor in 2013. He joined Boston University as the Inaugural Theodore Moustakas Chair Professor in Photonics and Optoelectronics in summer 2017.

DATA AVAILABILITY

Data newly presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.