Visible enhanced stimulated Raman scattering microscopy via frequency-doubling of picosecond pulses

Abstract

We report visible-enhanced stimulated Raman scattering (VIS-SRS) microscopy by direct frequency-doubling of picosecond pulsed lasers. Compared to previous reports from femtosecond sources, our setup is greatly simplified by removing the need to perform spectral focusing and allows for quick wavelength tuning for sparse multi-channel imaging across a large spectral range. We report a signal enhancement of 3.2-fold for the C–H bond from decane, consistent with theoretical analysis, and up to 483-fold with resonance enhancement. We showcase a multi-channel imaging application in HeLa cells and demonstrate a spatial resolution down to 47 nm from 5xFAD mouse brain tissues coupled with Raman-tailored sample-expansion. Finally, we discuss the future applications and limitations of our picosecond-doubled VIS-SRS.

1. Introduction

Raman spectroscopy, probing the vibrational modes of target molecules, allows for investigations of chemical structures in native states and has offered chemical-rich information for materials and biological investigations. However, for biological imaging, spontaneous Raman suffers from extremely slow imaging speed due to the weak signals and high auto-fluorescence background caused by visible excitation. To address this challenge, stimulated Raman scattering (SRS) significantly enhances Raman scattering efficiency by utilizing a secondary laser to drive stimulated emission and has been extensively employed to image live biological samples with high sensitivity, subcellular spatial resolution, and unparalleled chemical specificity [1–5].

Despite the improvement, SRS still faces a sensitivity bottleneck especially for detecting low concentration (∼ nM - µM) molecules [3–6]. Consequently, extensive efforts have been directed towards the theory of SRS and methods for improving SRS detection to increase the signal strength. Traditionally, SRS signal is detected by measuring either pump laser intensity loss (stimulated Raman loss, SRL) or Stokes laser intensity gain (stimulated Raman gain, SRG) resulting from stimulated emission, with SRL being preferred due to its higher photosensitivity with available photodiodes in the near-infrared (NIR) region. The SRL process is described as follows [1]:

$$SRL\propto -N·{\sigma }_{\text{Raman}}·{\Phi }_p·{\Phi }_S$$ (1)

where N denotes the number of bonds in the focal volume, ${\sigma }_{\text{Raman}}$ is the Raman cross-section, and ${\Phi }_p(or{\Phi }_S)$ are the photon flux of the pump (or Stokes) beams. The Raman cross-section can be further described by Eq. (2).

(2)

Here, K is a sample-dependent constant, the red-boxed terms are the molecule-independent non-resonant (NR) terms, and the blue-boxed term is the molecule-dependent Albrecht-A electronic pre-resonance (EPR) approximation for totally symmetric transitions under a strong dominating electronic transition [7]. Based on these terms, two strategies harnessing the enhancement in the non-resonance, or the pre-resonance condition can be developed separately or in combination to boost the SRL process.

SRS microscopy is conventionally performed with NIR laser excitation to offer fine biocompatibility with less photodamage in live biological samples [8]. The pump laser energy is set far from electronic excitation to avoid electronic background (Fig. 1(A), (left)). One significant strategy to enhance the imaging performance is through pre-resonance SRS imaging [9,10]. As the pump laser approaches the electronic excitation of the target molecules, the Albrecht-A term in Eq. (2) predicts a quadratic increase in the bonds’ Raman cross-section. Coupling the excitation lasers with NIR absorbing dyes to drive this pre-resonance condition (Fig. 1(A), (right)), Min et al., pushed the detection limit down to 250 nM [9]. Another direction for increasing sensitivity is through frequency doubling of the lasers that converts the NIR laser into the visible laser SRS (VIS-SRS) excitation, which is expected to enhance both the NR terms and the Albrecht-A term for molecules absorbing in the visible or even into the UV regime, such as retinoids, endogenous proteins, and DNA [11,12]. Such a strategy also offers much-improved spatial resolution with the shortening of the excitation wavelength. Powered by Fourier reweighting after denoising, resolution down to 86 nm was recently achieved for cellular imaging [13].

Fig. 1.

VIS-SRS microscopy with picosecond pulsed-lasers. (A) Energy diagram for non-resonance and pre-resonance SRS conditions. (B) Schematic of VIS-SRS microscope through frequency doubling on picosecond lasers. FM: flip mirror. $\lambda /2$: half waveplate. PD: photodiode. Ref: reference. BIBO: bismuth triborate. PPLN: periodically poled lithium niobate. SP: short-pass filter. BP: bandpass filter. DM: dichroic mirror. CL: cylindrical lens (C) Beam profile after SHG for pump and Stokes beams. Left: pump beam directly after BIBO, Middle: pump beam after correction with a pair of cylindrical lenses, Right: Stokes beam directly after PPLN.

So far, all VIS-SRS instruments are based on the frequency doubling of femtosecond (∼ 100 - 200 fs) pulses followed by spectral focusing of temporally-stretched picosecond (ps) laser pulses to retain high spectral resolution, lowered photodamage, and fast spectral tuning within the designated frequency range [14,15]. Spectral-focusing SRS requires highly dispersive media (e.g., rods or gratings) for pulse stretching, and exhibits challenges for multi-frequency imaging across a wide spectral regime due to the wavelength dependence of group delay dispersion, which can cause chirp mismatch and degrade spectral resolution unless compensated appropriately [15,16]. Managing these dispersion-related effects may in practice require adjustments to the optical path and system alignment when tuning across distant spectral regions. In this work, we explore the extension of picosecond-based SRS into the visible spectral regime, given its widespread use in the SRS community for its ease of use, with the goal of establishing a foundation for future VIS-SRS development based on picosecond platforms and providing practical insight into associated challenges.

Here, we report the VIS-SRS modality based on direct frequency-doubling of narrow-band picosecond lasers, bypassing spectral focusing, and greatly accelerating the wavelength tuning spanning across drastically different bond regions. We implement our system with the high-efficiency picosecond laser compatible conversion strategy and characterize our system by directly comparing the same samples between VIS-SRS and NIR-SRS. Using n-decane solution, we report a 3.2-fold non-resonance signal enhancement. Under our experimental design, we achieved a spatial resolution of 148 nm, and the same spectral resolution offered by NIR-SRS in both the cell-silent and the high-frequency C–H region. We next screened the (pre-) resonant enhancement of commercial dyes, reaching signal enhancements up to 480-fold. Finally, we discuss the nature and severity of photodamage associated with VIS-SRS on biological samples and establish a sample-expansion-based effective solution with VISTA-VIS-SRS [17]. We report an effective 47-nm spatial resolution on expanded 5xFAD mouse brain tissue.

2. Experimental setup

2.1. VIS-SRS microscope with picosecond lasers

A tunable pump (770-990 nm, 2 ps, 80 MHz) and Stokes (1031.2 nm, 2 ps, 80 MHz) beams from a picoEmerald laser system (Applied Physics & Electronics) are split through a long-pass dichroic mirror (Thorlabs, DMLP1000) to perform frequency doubling (Fig. 1(B)). In spectral-focusing VIS-SRS, thin (<1 mm) Beta-Barium Borate (BBO) and Lithium Borate (LBO) crystals were demonstrated with high conversion efficiency for the short fs pulses. However, these crystals provide very low SHG efficiency for ps pulses. As an alternative, we explored bismuth triborate (Bi₃BO₆, BIBO, Newlight Photonics, BITC5300-SHG800(I)-AR) for our tunable pump laser [18]. With a 3 mm BIBO crystal, which balances between the conversion efficiency and spatial walk-off effects, we obtained a conversion efficiency of 23% for the pump laser, yielding a total output power of 167 mW (Fig. 1(B)). To correct the astigmatism with a beam divergence ratio of about 1:5 (Fig. 1(C), (left)), we employed a pair of cylindrical lenses with focal lengths of 50 mm and 250 mm (Fig. 1(C), (middle)). In addition, a half-wave plate is used to correct for the type I SHG polarization change (Thorlabs, AHWP10N-580) (Fig. 1(B)). For our Stokes beam, a periodically poled lithium niobate (PPLN, Covesion, MSHG1030-0.5-3) crystal was chosen for its high conversion efficiency (50%), ease of phase matching criteria, and desirable type 0 SHG process that directly yields a round beam mode (Fig. 1(C), (right)) [19]. The remaining fundamental pump & Stokes laser pulses after frequency-doubling are cleaned up with optical filters (FF01-745/SP-25 & FF01-520/15-25, respectively). Due to the slight wavelength dependence of the spatial walk-off effect in the BIBO crystal, a small diagnostic path was added to realign the SHG pump beam and maintain spatial overlap with the Stokes beam. The frequency-doubled beams are then recombined with an additional dichroic mirror (Thorlabs, DMLP490R) before being routed into an Olympus FV3000 microscope (Fig. 1(B)). Light is focused on the sample with a 60x water immersion objective (UPLSAPO60XW, 1.20 NA, Olympus) and transmitted light is collected with an oil immersion condenser lens (1.40 NA, Olympus). The Stokes beam is filtered out (Semrock, FF01-498/SP-25) and the remaining pump beam is collected with a photodiode (S3590-19, Hamamatsu). The SRL signal from the pump beam is demodulated via lock-in amplification (SR844, Stanford Research Instruments) at the reference modulation frequency (20 MHz). Images were captured in a 16-bit grayscale using Olympus FluoView 3000 software. Images and graphs are visualized using FIJI and GraphPad Prism, respectively. Deconvolution to extract the system PSF was performed in MATLAB by fitting a convolution model to the measured bead profiles using nonlinear least-squares minimization (lsqnonlin, Optimization Toolbox). Imaging parameters for all measurements are tabulated in Supplement 1 ^{(1.1MB, pdf)} tables S2-S4.

3. Experimental results

3.1. System characterization

We first characterized the spatial resolution of our system by imaging 100-nm polystyrene (PS) beads (Fig. 2(A)), targeting the 2904 cm^-1 band and obtained a full width at half maximum (FWHM) of 162 nm (Fig. 2(B)). After deconvolving with the size of the beads, we calculate our effective point spread function to be 148 nm, which is in good agreement with the theoretical value of 141 nm, calculated from the Abbe diffraction limit, $D=\frac{0.50\lambda }{\text{N.A.}}$, where λ is the convolved wavelength of the excitation lasers (448.4 nm & 515.6 nm) and numerical aperture (N.A.) is 1.2.

Fig. 2.

Characterizations of VIS-SRS microscope. (A) A representative VIS-SRS image of 100-nm PS beads. (B) Sectional profile obtained from the bead indicated by the white arrow in (A). A full-width-at-half-maximum (FWHM) of 162 nm is obtained through Gaussian fitting. (C) Raman spectra of the C–H stretching of n-decane measured by spontaneous Raman (blue), VIS-SRS (red), and NIR-SRS (black). (D) SRS spectra of EdU measured using VIS-SRS (squares) and NIR-SRS (triangles). Corresponding Lorentzian fittings are overlaid. (E-F) SRS spectra of Coumarin 102 (E) and 3-Ethynylperylene (F) measured from VIS-SRS (red) and NIR-SRS (black).

We next calculated our non-resonance signal enhancement using n-decane, a small molecule with minimal electronic absorption far into the UV region [20], by directly comparing signals of the same samples obtained from VIS-SRS and NIR-SRS. Under the same NIR and VIS power conditions, we obtained an optimal 3.2-fold signal enhancement, lower than our theoretically calculated 8.4-fold enhancement ( Supplement 1 ^{(1.1MB, pdf)} Section 1). We reason that the enhancement discrepancy can be potentially attributed to the use of different objectives between measurements. For NIR-SRS, a 25x water immersion objective (XLPLN25XWMP, 1.05 NA, Olympus) is used, which however leads to severe chromatic aberration in the visible region, requiring us to instead use a visible-wavelength optimized objective (UPLSAPO60XW, 1.20 NA, Olympus). In our theoretical calculations, we only accounted for the NA difference, but other practical parameters (e.g. chromatic aberration correction) may differ. We also note that here we focused on analyzing the pure signal enhancement for quantitative understanding, instead of the signal-to-noise ratio.

With n-decane, we further confirmed that VIS-SRS achieves the same spectral resolution quality in the high-frequency C–H region through wavelength sweeping of the pump laser across the designated region (∼2825-3000 cm^-1, Fig. 2(C)), compared to that obtained by NIR-SRS and spontaneous Raman, indicating the well-maintained bond-selectivity and the pump laser tuning in VIS-SRS. As the cell-silent regime has been proven essential for performing specific dynamic metabolic imaging, we next moved to characterize 5-ethynyl-2'-deoxyuridine (EdU) solution with a sharp alkyne peak centered at 2125 cm^-1 [21]. Since EdU has a characteristic absorption peak at 288 nm (Fig. S1), we expect a slight pre-resonance enhancement in VIS-SRS. Indeed, we obtained a 7-fold total signal enhancement between VIS-SRS and NIR-SRS, confirming the presence of weak pre-resonance effects. Further spectral acquisition also confirms that VIS-SRS has comparable spectral resolution in the cell-silent window (Fig. 2(D), FWHM of 21.4 cm^-1 (VIS-SRS) vs 20.5 cm^-1 (NIR-SRS), respectively).

Finally, we screened candidates for pre-resonance and resonance imaging experiments by characterizing a series of commercially available dyes with varying absorption maxima. We identified two promising candidates in Coumarin 102 (Fig. S1, λ_e = 396 nm, λ_Pump = 476.3 nm & 477.4 nm, Fig. 2(E)), in the pre-resonance region and 3-Ethynylperylene (3-Perylene Fig. S1, λ_e = 459 nm, λ_Pump = 477.1 nm, Fig. 2(F)) in the rigorous resonance region. Despite Coumarin 102 suffering from a large transient absorption (TA) background (Fig. S2), we obtained a 30-fold signal enhancement compared to NIR-SRS when probing the C = C double bond. Notably, the C = C double-bond from 3-Perylene yielded a 483-fold signal enhancement, due to more significant resonance enhancements, with only a minor and flat background (Fig. 2(F)). However, the alkyne peak of 3-Perylene could not be resolved due to the high electronic background (Fig. S2, λ_Pump = 465.2 nm), likely due to the increased proximity of the pump laser to the absorption maximum. For the importance of alkynes in Raman imaging in the cell silent-region, we further investigated the pre-resonance enhancement for alkynes in VIS-SRS systematically. We synthesized and measured a series of polyyne [22,23] molecules that absorb in the UV-vis region with varying degrees of resonance enhancement (Table S1). We note that lowered signals for 3- and 4- yne were due to induced higher background.

3.2. Bio-imaging using VIS-SRS

We next characterized the biocompatibility of our VIS-SRS system. We first imaged fixed HeLa cells in the high-frequency C–H region targeted at the protein channel (2940 cm^-1) (Fig. 3(A)) and achieved similar performance with enhanced spatial resolution to visualize the cellular network to a previous report [11–13]. Through direct comparison to NIR-SRS, we obtained a 7-fold signal enhancement in the cytoplasm of HeLa cells. The slightly larger enhancement in HeLa cells compared to that of n-decane solution also indicates the presence of pre-resonant effects likely arising from the absorption band around 280 nm from proteins in the cytoplasm [11]. We also demonstrated that more clinically relevant samples, such as neuronal cells, benefited from this same resonant enhancement and could be imaged with strong image fidelity and signal strength (Fig. 3(B)). We further confirmed that our VIS-SRS system is live-cell compatible by imaging HeLa cells (Fig. 3(C)) targeted at the –CH₃ channel (2940 cm^-1) with similar image qualities to NIR-SRS (Fig. 3(D)). We note, however, live cells showed much-reduced photon tolerance for picosecond visible photons which originates from the higher order absorption induced photo-damage, preventing long-term imaging applications (Table S3).

Fig. 3.

VIS-SRS imaging of biological samples. (A-B) Protein distribution of a fixed HeLa cell (A) and neurons (B) with fine structural details targeting 2940 cm^-1. (C-D) Protein distribution in live HeLa cells by NIR-SRS (C) and VIS-SRS (D). (E-F) Two-channel imaging of HeLa cells under the same field-of-view showing protein distributions (yellow, E, 2940 cm^-1) and lipid droplets formed containing d₈-AA in C-D channel (magenta, F, 2250 cm^-1).

As rationalized above, we seek to demonstrate the flexibility and wide tunability of our system by performing two-channel imaging of the same cells in the cell-silent and high-frequency C–H regions. We imaged the protein channel (CH₃, 2940 cm^-1, Fig. 3(E)) mapping the endogenous proteins in HeLa cells and the cell silent region (C–D, 2250 cm^-1, Fig. 3(F)) where incubation with deuterated arachidonic acid (d₈-AA) results in the accumulation of the polyunsaturated fatty acid (PUFA) within the lipid droplets [24]. This set of images, including an off-resonance image at 2700 cm^-1 for background subtraction, was captured within five minutes, a time frame compatible with many live-cell imaging tasks. We note that while spatial realignment is needed to ensure optimal beam overlap after tuning, our imaging speed is ultimately limited by laser tuning (10-15 seconds for spatial realignment, and 60-90 seconds for laser tuning). Nonetheless, this imaging speed paves the way for metabolic imaging of UV-VIS absorbing target molecules, such as drugs, that are below the detection threshold in NIR-SRS.

3.3. Mitigating photodamage, and tissue imaging using VIS-SRS

The bluer wavelengths after frequency doubling increase photodamage and must be carefully considered in VIS-SRS for biological samples [13]. Symptoms of photodamage we observed during cell imaging include nuclei darkening, nucleoli shrinking, and cytoplasm bubbling and burning. One way to mitigate the damage is to use longer pulses. Fujita et al. demonstrated that stretching 2 ps pulses to 7 ps and 14 ps greatly reduced photodamage [25]. However, this strategy may suffer from insufficient laser power for high-quality VIS-SRS cell imaging applications. An alternative strategy to circumvent this limitation is to explore sample preparation techniques, such as pre-emptive cooling [26], sample dilution, or homogenization [17].

Therefore, we coupled sample expansion with VIS-SRS as a strategy to mitigate photodamage. We theorized that expanding samples isotropically would decrease sample density and homogenize sample refractive index to minimize local heat generation and facilitate heat dissipation. At the same time, the expanded sample coupled with VIS-SRS should render unprecedented super-resolution imaging capabilities for chemical imaging and allow deep-tissue imaging with optically cleared tissues. Consequently, we employ our previously developed VISTA framework [17] to expand 5xFAD brain tissues. Figure 4(A) shows a fixed 5xFAD tissue brain image before the expansion by VIS-SRS, with a protein and a lipid rich amyloid-beta aggregate region. After a 3.4-fold expansion, the reduced protein density with the removal of lipids indeed significantly raised the photodamage threshold, allowing us to fully capture an aggregate region (Fig. 4(B)) in 3D under dramatically higher power conditions compared to cells and unmodified tissues (Table S4). We measured the cross section of a tissue structure (Fig. 4(C)) and fitted it to the gaussian profile with a FWHM of 160 nm (Fig. 4(D)), resulting in the effective spatial resolution of 47 nm after accounting for 3.4-fold sample expansion.

Fig. 4.

VIS-SRS imaging on isotropically expanded brain tissues. (A) VIS-SRS image of a 5xFAD mouse brain tissue before sample expansion under 2940 cm^-1. (B) 3D VIS-SRS image (2940 cm^-1) of 5xFAD mouse brain tissue after 3.4-fold isotropic expansion. (C) A XY-slice of 3.4-fold expanded single frame of a tissue structure in 5xFAD brain tissue. (D) Spatial analysis of the yellow-line highlighted structure from (C).

4. Conclusion

In this work, we developed a widely tunable VIS-SRS system based on the frequency doubling directly from a picosecond output laser, which is capable of rapid multi-channel imaging across different bond regions, enabling various biological imaging applications. Using the picosecond VIS-SRS system, we achieved significant resonance enhancements in Coumarin 102 and 3-Ethynylperylene, highlighting potential targets for pre-resonance enhanced SRS imaging in future applications. We imaged HeLa cells across the high-frequency C–H stretching region and cell-silent region in a timely manner compatible with live-cell investigations. We discussed the destructive nature of photodamage and provided a solution based on the sample expansion, enabling label-free imaging of brain tissues in 3D that achieves an effective spatial resolution of 47 nm.

Compared to spectral-focusing VIS-SRS with fs laser sources, we note that our VIS-SRS has pros and cons, depending on the specific applications. While our VIS-SRS provides excellent multi-channel imaging across a wide range for live-cell metabolic studies, its tuning is slower if the bonds of interest are limited to the bandwidth of fs pulses (∼ 200 cm^-1). Due to the slight wavelength dependence of spatial walk-off in the BIBO crystal, quick optimization of the laser spatial overlap is still necessary after tuning shifts of 7–10 cm⁻¹. However, with the aid of the diagnostic light path described earlier (Fig. 1(B), dashed-lined box), realignment can be rapidly performed. Using this approach, two-channel imaging including an off-resonance image were acquired within 5 minutes, with laser tuning as the primary time bottleneck. Implementing a fast feedback loop involving motorized or piezo stage that automatically corrects spatial walk-offs should further speed up the spectral acquisition and make the system more user friendly.

Despite its increased photodamage compared to NIR-SRS, VIS-SRS still offers unique opportunities worth exploring. Most commercial drugs at their physiological concentrations cannot be detected by NIR-SRS due to the small Raman cross sections without pre-resonance enhancement. However, many drugs absorb in the UV-VIS region and frequently contain biorthogonal alkyne or nitrile moieties, making them promising imaging targets in VIS-SRS. With continuous improvements to this technique and advancements in statistical models predicting resonance dynamics [27], we envision VIS-SRS should find utility in the field of drug development and imaging.

Funding

National Science Foundation 10.13039/100000001 ( 2240092, 2139433, 1745301); Heritage Medical Research Institute 10.13039/100014221.

Disclosures

The authors declare no conflicts of interest.

Data Availability

Data underlying results in the paper is not publicly available but is obtainable upon request.

Supplemental document

See Supplement 1 ^{(1.1MB, pdf)} for supporting content.