Ultra-high Information-content Chemical Imaging with Broadband Coherent Anti-Stokes Raman and Two-photon Fluorescence Lifetime Microscopy

One challenge in biological studies is to accurately detect biological building materials in their native states and within the intact architecture of cells and tissues. Raman spectroscopy shows unprecedented potential for detecting intrinsic metabolites and chemical species within biological specimens and has been recently recognized as an emerging tool for metabolic profiling of biological specimens¹. By analyzing the peaks of various metabolic groups shown in Raman fingerprint spectra, increasing evidence shows that the fingerprint features can be used to detect the early stages of diseases such as cancer^2,3,4, cardiac disorders^5,6, and neurodegenerative diseases^7,8 and is particularly useful for addressing fundamental biological problems^9,10 including imaging intracellular structures without any labeling perturbation¹¹, differentiating various cell states^12,13, and cell type classification^14,15,16. Most of the chemical information ( > 90%) encoded in the Raman spectra is in the fingerprint region (500 to 1800 cm^-1), making this region by far the most useful for discriminating chemical compositions in biological samples¹⁷. However, the weak Raman signal in this frequency range makes the applications of Raman fingerprint spectroscopy to biological studies challenging, particularly for live, intact biological specimens¹.

Coherent Raman imaging (CRI) approaches significantly shorten the acquisition time of Raman signals, but most of these acquire relatively narrow spectral segments and are optimized for the strong CH stretch signals between 2800 - 3100 cm^-1. Narrowband CARS and stimulated Raman scattering (SRS) (Figure 1A,B) use two picosecond pulses, usually to acquire a narrow portion of the strong CH signal ^{18,19,20,21,22,23}. Multiplex CRI approaches, such as multiplex CARS (Figure 1C), use a combination of a picosecond and ~100 fs pulses to obtain a moderate spectral range (usually ≤ 400 cm^-1) usually in the CH region^{24,25,26,27,28,29,30}. Acquiring a whole biological-relevant Raman spectrum (from 600 - 3200 cm^-1) per pixel using these imaging approaches is usually prohibitive because most fingerprint signals are 10-100 times weaker than CH-stretch signals³¹. Unlike most CRI approaches, broadband coherent anti-Stokes Raman scattering (BCARS) is optimized for single-shot broadband acquisition that covers the fingerprint and CH-stretch regions^31,32,33. BCARS uses intra-pulse interactions of a ~10 fs pulse to efficiently excite low-frequency (fingerprint) vibrations (Figure 1D), and uses the same mechanism as multiplex CARS, the combination of the fs pulse with a ps pulse to generate vibrational coherence in the CH stretch region (Figure 1C)^32,33. Furthermore, the BCARS signals are intrinsically calibrated due to their interaction with the non-resonant background (NRB), resulting in a quantitative and instrument-invariant full Raman spectrum per laser pulse at each pixel³¹, with the capability of rapidly acquiring high-information-content data from cells, intact tissues, and live biological organisms^{32,34,35,36,37,38,39}.

The acquired raw BCARS spectra contain a strong so-called "non-resonant background" (NRB). The NRB is the contribution from the electronic response to the laser pulses that is not chemically specific. The NRB coherently interferes with the vibrationally resonant signal, acting as a heterodyne amplifier³². The NRB also distorts the underlying Raman spectrum (Figure 2A), but the Raman spectrum can be retrieved as long as significant contiguous spectral sections are acquired. Currently, there is no post-processing method available to separate the resonant CARS signal from the non-resonant component using a narrowband CARS approach. Raman signal retrieval from the broadband CARS signal can be achieved by utilizing the time-domain Kramers-Kronig relation (or Hilbert transform)^33,40 (Figure 2B), maximum entropy^41,42, or machine-learning^{43,44,45,46,47,48} approaches. Note that the first two methods analytically retrieve the phase (the resonant Raman signal) from the raw BCARS spectra, the machine-learning approaches can be highly dependent on the instrumentation and the training models.

While BCARS microscopy allows for label-free characterization of metabolic profiling of biological specimens, the same supercontinuum laser (950 - 1200 nm for the BCARS setup presented in this work) used for BCARS signal generation can also excite a wide range of dyes, such as Rhodamine-based, Bodipy-based, Alexa Fluor-based, and cyanine-based dyes, and various fluorescent proteins, including GFP and mCherry^49,50. Compared to single-photon excitation using visible continuous wave (CW) lasers, multiphoton fluorescence microscopy, such as two-photon fluorescence lifetime imaging microscopy (2p-FLIM) and two-photon excitation fluorescence microscopy (TPEF), using near infrared (NIR) light offers a deeper sample penetration depth, intrinsic 3D sectioning capability, high-contrast images, and good resolution comparable to state-of-the-art confocal microscopy, making it an attractive tool for biological research^51,52. Fluorescence lifetime imaging microscopy (FLIM) using a time-correlated single photon counting (TCSPC) approach detects the fluorescence decay signals repeatedly excited by a time-modulated laser source^53,54. The fluorescence spectrum of a given fluorophore may change with environmental factors such as solvent polarity, but these changes are usually subtle. On the other hand, fluorescence lifetimes often vary by factors of two or more with environmental changes such as pH values⁵⁵, viscosity⁵⁶, polarity⁵⁷, and concentration of ionic species^58,59, binding interactions, the presence of quenchers, or structural changes^53,54. This unique property of fluorescence lifetime provides an additional signal contrast in fluorescence imaging and has successfully addressed the issues of autofluorescence interference in cell and tissue imaging⁶⁰, unmixing multiple fluorophores using a spectrally resolved FLIM approach⁶¹, cancer diagnosis and treatment monitoring⁶², label-free detection of cellular heterogeneity^63, and metabolic changes^64,65.

In this work, we introduce a multimodal imaging platform combining BCARS, 2p-FLIM, and TPEF. To the best of our knowledge, this is the first time presenting an imaging platform allowing simultaneous BCARS and 2p-FLIM/TPEF imaging. We demonstrate the in vivo imaging capabilities of this instrument using a widely used genetically tractable model organism, Caenorhabditis elegans (C. elegans). We show that the BCARS spectra, the 2p-FLIM decays of Nile Red (a vital lipid dye), and the two-photon fluorescence GFP signals of a reported lipid marker (DHS-3::GFP) can be simultaneously acquired in vivo from Nile-Red-stained worms. A practical guide for the imaging approach presented here includes (i) the sample should be optically transparent and with a thickness typically less than 100 µm; (ii) the staining condition can be different depending on the dyes and samples. In the Nile-Red staining of C. elegans results shown here, the Nile Red concentration is at ~µM level, and staining time can range from a few hours to overnight⁶⁶; (iii) a dark sample environment is required to avoid visible light leakage into the 2p-FLIM/TPEF detectors. These imaging conditions are identical to the imaging conditions of most nonlinear optical and fluorescence microscopy. In addition, our system demonstrated here can be further applied to detect other dyes and fluorescent reporters in addition to Nile Red and GFP by introducing an additional Ti:Sapphire fs laser that covers 700 -950 nm range for the two-photon fluorescence excitation. Thus, while the representative results described in this protocol are the in vivo imaging of dye-stained C. elegans expressing GFP, the ultra-high information-content chemical bioimaging platform has applications that are broad and far-reaching.

1. C . elegans sample preparation and Nile Red Vital staining

NOTE: C. elegans has been heavily used for multiple imaging techniques due to its whole-body transparency. The internal organs and tissues can be easily imaged in the intact animal without dissection. In addition, ethics approval is not required for research involving C. elegans, as they are not considered vertebrates and are not subject to the same ethical regulations as higher animals like rodents. This section describes the synchronization of C. elegans worms and the preparation of Nile Red-stained samples for fluorescence imaging. Synchronized one-day adult worms are obtained via a two-generation egg-laying protocol and stained using freshly prepared Nile Red-OP50 plates.

1. Maintain the C. elegans strains (e.g., N2, LIU1 ldrIs1 [dhs-3p::dhs-3::GFP + unc-76(+)]) on standard Nematode Growth Medium (NGM) plates seeded with OP50 bacteria at 20 °C.
2. Synchronize the worms
   1. Pick 10-20 gravid adult worms onto a freshly made OP50-seeded NGM plate.
   2. Incubate the plate at 20 °C for 1-3 h to allow egg laying.
   3. Remove all adult worms, leaving only the freshly laid eggs on the plate.
   4. Incubate the plate at 20 °C until worms reach the one-day adult stage.
   5. Repeat the above steps for an additional generation to improve synchronization precision.
3. Prepare Nile Red stock solution
   1. Dissolve 15.92 mg of Nile Red in 10 mL of DMSO to a final concentration of 5 mM.
   2. Sonicate for 10-30 min if needed to aid dissolution.
4. Prepare Nile Red-OP50 staining plates.
   1. Mix OP50 bacterial slurry thoroughly with the Nile Red stock to achieve a final Nile Red concentration of 5 µM per NGM plate. For example, mix 10 µL of the Nile Red stock with 20 µL OP50 bacterial slurry, and then immediately apply the whole mixture to an unseeded 6 cm (with ~10 mL volume) NGM agar plate to make a final Nile Red concentration of 5 µM per NGM plate.
   2. Spread the mixture evenly onto unseeded 6 cm NGM plates.
   3. Store plates in the dark at room temperature overnight to allow bacterial growth and dye absorption.
5. Stain synchronized worms
   1. Transfer 30-50 synchronized one-day adult worms onto each prepared Nile Red-OP50 plate.
   2. Incubate the worms in the dark at 20 °C for 1-2 h before imaging.
6. Immobilize worms for imaging
   1. Apply a drop of 100 mM sodium azide on a 2% agarose pad on a microscope slide (approximately 10-15 µL for 20 worms)
   2. Transfer stained worms to the anesthetizing agent (sodium azide) drop and cover the worms with a coverslip prior to microscopy.
      NOTE: When picking 10-20 gravid adults onto a freshly made NGM OP50 plate in the first step of egg-laying synchronization, only pick the adults, but not any stray eggs or larvae. This is to make sure that all the remaining eggs on the plates are produced by the picked worms within the 1-3 h egg-laying time window. In addition, sodium azide is classified as a hazardous material, although it requires a very small amount for nematode anesthetization. The usage and disposal of sodium azide waste must follow the guidelines of the institutional Environmental Health and Safety department.

2. Probe beam generation

NOTE: This protocol describes the generation of the picosecond (ps) probe beam using a home-built optical parametric oscillator (OPO) followed by second harmonic generation (SHG) with a home-built signal feedback loop system to stabilize the output power of the probe beam.

1. Switch on the 1030 nm OPO pump laser (Figure 3A or the near 1ps, ~3 W OPO pump laser shown in Figure 4) and allow at least 2 h for thermal equilibration. Stable output is critical for OPO operation.
2. Guide the output to the OPO cavity
   1. Select the appropriate poling period on the OPO PPLN (Figure 3A) crystal to match the target signal wavelength.
   2. Direct the beam into the first curved mirror (CM) (Figure 3A), forming part of a linear cavity.
   3. Ensure the beam passes through the OPO periodically poled Lithium Niobate (PPLN) crystal (Figure 3A) to facilitate optical parametric down-conversion (Figure 3B).
   4. Ensure the beam reflects off a second curved mirror (CM) and then off the end mirror (EM) (Figure 3A) to form the cavity loop.
   5. Configure the output coupler (OC) (Figure 3A) to extract the generated signal beam from the cavity.
3. Adjust the temperature of the OPO PPLN crystal using the temperature controller (TC) (Figure 3A) to select the signal wavelength. A typical signal wavelength is 1530 nm, corresponding to an output power of approximately 0.8 W when pumped with 3 W at 1064 nm.
4. Optimize OPO alignment
   1. Adjust the position and tilt of two CMs and EM to maximize output power and beam quality.
   2. Adjust the cavity length to shift the signal spectrum to the desired frequency.
5. Route the OPO signal beam through a set of collimating and focusing optics and guide it into the SHG stage.
   1. Select the appropriate poling period on the SHG PPLN crystal (Figure 3A) to match the target signal wavelength.
   2. Precisely control the crystal temperature using a temperature controller to maximize SHG efficiency.
   3. Direct the beam into the SHG PPLN crystal for second harmonic generation, converting 1530 nm OPO signal to 765 nm (Figure 3C). The typical SHG conversion efficiency is approximately 30% under optimal alignment conditions.
   4. Insert a 1300 nm shortpass filter (SPF in Figure 3A, specifications in Table of Materials) after the SHG PPLN to remove residual 1530 nm fundamental light.
   5. Collimate the resulting 765 nm beam using an appropriate lens (as L in Figure 3A).
6. Implement power stabilization
   1. Use a wedge window (WW) to sample the SHG beam and direct it to a photodiode (PD).
   2. Ensure a feedback loop compares the photodiode signal (PD in Figure 3A) to a reference value and actuates the Picomotor on the end mirror (EM) to adjust the OPO cavity length in real-time, thereby stabilizing the SHG output power.
      NOTE: To determine the reference value, first scan the EM position using the Picomotor while recording the SHG output power. Identify the peak intensity in this profile, then set the reference value to 90% of the peak. This choice ensures a clear direction of feedback correction, enabling the system to compensate for both increasing and decreasing trends in power and effectively stabilizes the output power over time.

3. Supercontinuum (SC) generation for BCARS

NOTE: The simplified scheme of the BCARS-based multimodal system is shown in Figure 4. This protocol describes the generation of the SC beam.

1. Adjust the pre-chirp (Group Delay Dispersion, GDD) to the femtosecond (fs) pulse output (1030 nm, 150 fs, ~1 W) to obtain a wide enough SC bandwidth covering the fingerprint region signals (SC bandwidth ≥ 1800 cm⁻¹). Achieve this by adjusting the separation of a grating pair (see Figure 4), allowing for precise control over the dispersion.
   NOTE: The application of pre-compensated GDD is crucial for generating the desired pulse and optimizing SC generation. The required GDD value may vary depending on the specific characteristics of the laser system used.
2. Incorporate a beam telescope consisting of two lenses with the same focal lengths to adjust the divergence of the pump beam.
   NOTE: When the beam travels over meters before entering the broadening fiber, precise control of the beam divergence becomes essential. Maintaining a collimated beam during this distance is critical to ensure efficient SC generation.
3. Set up the auto-aligner (Figure 4) using kinematic mirror mounts equipped with piezo actuators for beam steering. Combine these with optical wedges and a Position Sensing Detector (PSD) to actively monitor and correct beam alignment.
4. Position the components such that the distance from the wedge to the first mirror is equal to the distance from the wedge to the first PSD. This symmetric configuration ensures beam-pointing stability into the broadening fiber.
   NOTE: Active alignment is essential for maximizing spectral broadening efficiency and minimizing the risk of fiber input damage.
5. Use a half-wave plate (HWP) and a polarizing beam splitter (PBS) to control the laser power before coupling into the fiber. Rotate the HWP to adjust the polarization angle relative to the PBS, thereby tuning the transmitted power. While these components are not shown in Figure 4, they are critical for safe alignment and power control.
6. Couple approximately 1 W of laser power into a 10 cm Large-Mode-Area Polarization Maintaining (LMA-PM-5) photonic crystal fiber (PCF) using a fiber coupler. The coupling procedure is as follows:
   1. First, reduce the laser power to ~20-50 mW during the initial alignment to avoid damage to the fiber.
   2. Direct the collimated beam into the fiber, ensuring some light emerges from the other side. Optimize the beam path by adjusting its position to maximize the output power.
   3. Insert the aspherical lens (Figure 4), fine-tune the beam again to maximize the output power.
   4. Adjust the z-position of the aspherical lens to ensure precise focus on the fiber facet.
   5. Repeat the alignment and focus adjustments iteratively to maximize coupling efficiency.
   6. Increase the laser power to 1 W for operation.
7. Collimate the resulting broadband supercontinuum (SC) output from the PCF using an off-axis parabolic (OAP) mirror with focal length f = 15 mm, as shown in Figure 4.
   NOTE: Reflective optics, such as the OAP mirror, are preferred for broadband beams to minimize chromatic aberration introduced by refractive elements.
8. Guide the collimated broadband supercontinuum (SC) output into a prism compressor (Figure 4) consisting of a pair of Schott Flint (SF10) prisms.
   1. Direct the SC beam into the apex of the first prism to initiate spectral dispersion.
   2. Position the second prism such that the dispersed beam enters through its apex.
   3. Mount both prisms on a rail system to enable coarse adjustment of the prism separation. Translate the second prism along the rail to adjust the optical path length and overall compression. The separation for this system is 44 cm.
   4. Mount the first prism on a precision linear translation stage. Adjust the insertion depth of the first prism to fine-tune the dispersion compensation.
   5. Iteratively adjust both the prism separation and the insertion length of the second prism to achieve optimal temporal compression of the SC pulse. Determine the optimal temporal compression by matching the experimentally observed degenerate four-wave mixing (DFWM) signal to simulated results that account for group delay dispersion (GDD) and third-order dispersion (TOD). This approach involves performing a simulation of the supercontinuum generation process, incorporating the effects of GDD and TOD, which provides a reference for the expected temporal and spectral characteristics of the optimally compressed SC pulses. Then, iteratively adjust the prism compressor parameters to compensate for the GDD and TOD present in the system until the experimental DFWM signal matches the simulation predictions. A typical result of the spectral profile of the SC is shown in Figure 5.
      NOTE: Proper dispersion compensation enhances the spectral coherence of the SC. Under optimal alignment, the SC output has a power of approximately 150 mW, a center wavelength around 1060 nm, and spans a spectral range from 950 to 1200 nm (corresponding to 1800 cm⁻¹). These values may vary slightly depending on alignment and fiber coupling efficiency.
9. Combine the probe and supercontinuum beams using a dichroic mirror (see DM in Figure 4). Align them with a pair of pinholes to ensure both beams are collinear. Focus the combined beams onto a high-frequency photodiode (1 GHz bandwidth). For optimal signal quality, both beams must be overlapped in both the spatial and temporal domains. The process is as follows:
   1. Direct the output from the photodiode to a high-frequency oscilloscope (500 MHz bandwidth). Use the laser trigger as the reference clock.
   2. Block the probe beam and measure the SC pulses. Record the temporal differences between the SC pulses and the laser reference.
   3. Block the SC beam and measure the probe pulses. Adjust the mechanical delay line (as shown in Figure 4) such that the probe pulses overlap with the SC pulses in time. Ensure that the two beams reach their maximum simultaneously.
   4. Remove the photodiode, and let the beam proceed into the microscope.
      NOTE: The above steps describe a coarse adjustment for beam overlap. Further optimization for more precise alignment is covered in steps 6.1 to 6.9.

4. Setting up the fluorescence detection for 2p-FLIM and TPEF and introducing an additional Ti:Sapphire fs laser

1. Turn on the Ti:Sapphire laser and allow it to reach thermal equilibrium. Wait approximately 2 h to ensure stable output power and beam pointing.
   NOTE: In this study, TPEF and FLIM imaging were performed using the SC source, which provides broad spectral coverage and sufficient power for fluorophores such as GFP and Nile Red. The Ti:Sapphire laser is included in the system as an optional excitation source for fluorophores with two-photon absorption peaks below 950 nm, which may not be efficiently excited by the SC.
2. Activate the mode-locking mechanism on the Ti:Sapphire laser. Confirm stable mode-locking via the built-in monitor.
3. Pre-chirp the Ti:Sapphire pulses for optimal temporal compression at the sample using a prism compressor. Arrange the prisms in a configuration similar to that described in step 3.8 and adjust the insertion length of the second prism to fine-tune the pulse duration.
   NOTE: Proper temporal compression ensures peak intensity and efficient two-photon excitation at the sample plane.
4. (Optional) Couple the Ti:Sapphire beam into a polarization-maintaining single-mode fiber if the laser is located on a separate optical table. Adjust the input coupling optics and monitor the fiber output to minimize coupling losses.
5. Combine the Ti:Sapphire beam with the supercontinuum (SC) and probe beams using a dichroic mirror (see the combining point at PBS shown in Figure 3). Align the beams carefully to ensure spatial overlap at the dichroic and co-propagation toward the microscope.
6. Guide the combined beams to the microscope and fine-tune their spatial overlap. Use a fluorescent sample (e.g., a slide with 1 µm fluorescent polystyrene beads) to verify the alignment of the Ti:Sapphire beam relative to the BCARS excitation path. Adjust the Ti:Sapphire alignment mirrors to ensure that the fluorescence emission is co-localized with the BCARS focal point within the same imaging field. Confirm the colocalization by visualizing the overlap between the BCARS and TPEF images that are acquired simultaneously, where the spatial offset should be less than one pixel.
   NOTE: Achieving spatial overlap between fluorescence and BCARS signals is essential for multimodal image registration.
7. Insert a dichroic mirror before the microscope input to reflect back-scattered fluorescence toward the detection path. Select a dichroic with an appropriate cutoff wavelength to efficiently separate fluorescence from the excitation beams.
8. Insert a second dichroic mirror downstream in the emission path to split the fluorescence signal into two detection branches:
   1. Transmit wavelengths ≥ 550 nm to the FLIM detector.
   2. Reflect wavelengths ≤ 550 nm toward a photomultiplier tube (PMT) for analog two-photon excited fluorescence (TPEF) detection.
      NOTE: Install bandpass filters in each detection branch to block residual excitation light and to isolate specific emission bands.
9. Connect the PMT output to a preamplifier and feed the amplified analog signal into a data acquisition (DAQ) device. Share the global synchronization trigger from the field programmable gate array (FPGA) with the DAQ to ensure temporal alignment across imaging modalities (Figure 7).
10. Connect the FLIM detector output to a time-correlated single-photon counting (TCSPC) module.
    1. Connect the FLIM detector signal to the start input of the TCSPC module.
    2. Connect the Ti:Sapphire laser's (or BCARS SC's) synchronization output to the reference input of the TCSPC module.
    3. Connect the global camera trigger (generated by the FPGA) to the frame sync input of the TCSPC module to synchronize with line-scanned image acquisition.

5. Setting up the BCARS detection path

NOTE: This protocol describes setting up a descanned detection path for line-scanning hyperspectral imaging. The 1D galvo mirror scans the excitation beam along the y-axis with 2-4 ms dwell time per pixel for sufficient signal integration. After each line scan, the sample stage advances one step along the x-axis. Descanning using a synchronized secondary galvo mirror is essential to maintain consistent spatial registration of the spectral signal on the camera during fast beam scanning, eliminating signal crosstalk between adjacent pixels and preserving spatial-spectral fidelity³⁹.

1. Introduce the supercontinuum (SC) and probe beams into a one-dimensional (1D) galvo scanner (as shown in Figure 4 and more details in reference³⁹). Position the galvo mirror conjugate to the back aperture of the focusing objective to enable accurate beam scanning across the field of view.
2. Expand each beam using a reflective beam expander (as shown in Figure 4) to match the back aperture of the excitation objective. This ensures optimal focusing and maximum spatial resolution at the sample plane.
   NOTE: Use reflective optics to minimize chromatic aberration and maintain the broadband spectral content of the SC beam.
3. Place an alignment target at the back aperture plane of the excitation objective. Adjust the SC and probe beams so that both pass through the center of the target, ensuring co-alignment along the optical axis.
   NOTE: Precise spatial overlap maximizes BCARS signal strength and ensures maximum spatial resolution at the focal point.
4. Focus the beams onto the sample using a high-NA excitation objective (e.g., 1.2 NA, 60x water-immersion objective) mounted on an inverted microscope.
5. Prepare an alignment sample by sandwiching a drop of water between two coverslips to create a sealed microscope slide. Place the prepared slide onto the microscope stage and adjust the focus to bring the sample into view.
6. Collect the generated anti-Stokes signal in transmission using a collection objective placed opposite the excitation path.
7. Implement a 4f descanning system (as shown in Figure 4) to preserve spatial mapping of the anti-Stokes signal onto fixed rows of the detector.
   1. Relay the back focal plane (BFP) of the collection objective onto the axis of the descanning galvo mirror using a 4f optical system. This consists of two precision tube lenses (see Table of Materials), arranged such that:
      1. The first lens is placed one focal length (200 mm) after the BFP of the collection objective.
      2. The second lens is placed one focal length before the galvo mirror.
   2. Synchronize the descanning galvo with the scanning galvo using an FPGA-controlled waveform generator.
   3. Integrate a 7-bit digital potentiometer into the FPGA control system to allow adjustment of the descanning galvo amplitude.
      NOTE: The descanning galvo mirror compensates for the angular deflections introduced by the primary scanning galvo. By dynamically adjusting its angle in synchrony, it counteracts beam movement, ensuring that the anti-Stokes signal remains aligned along the optical axis leading into the spectrometer. This stabilization is essential to maintain spatial registration on the detector and prevent lateral displacement of the spectral image during scanning.
8. Configure the FPGA system (see Table of Materials) to synchronize galvo scanning and camera triggering.
   1. Generate analog outputs for the scanning and descanning galvos.
   2. Accept an input stage trigger to mark the start of each line scan.
   3. Output frame capture trigger signals to the scientific complementary metal-oxide-semiconductor (sCMOS) camera at each scan position.
   4. Independently control the descanning amplitude to accommodate different collection objectives.
      NOTE: A typical FPGA pulse train includes a scanning galvo drive signal, a descanning galvo drive signal, a Stage trigger input, and a pixel clock for synchronized frame timing.
9. Pass the collimated anti-Stokes signal through two short-pass filters to block residual pump and SC excitation light.
10. Focus the filtered anti-Stokes signal through the spectrometer entrance slit and onto the sCMOS detector. Adjust the angles of both the scanning and descanning galvos to ensure that the beam is scanned parallel to the slit's long axis, aligning the optical image plane with the slit. This precise alignment maximizes spectral resolution and minimizes lateral signal crosstalk across the detector rows.
11. Configure the spectrometer to disperse the anti-Stokes spectrum vertically along the camera rows. Select an appropriate grating (e.g., 300 lines/mm, blazed at 700 nm) for optimal spectral coverage.

6. Signal optimization and spectral resolution calibration

1. Launch the home-built acquisition software and begin real-time signal monitoring from the sCMOS camera. The acquisition software is home-built in Python 3.8+ (Windows 10) with dependencies including Hamamatsu DCAM-API and Measurement Computing Universal Library drivers, outputting data in HDF5 format. Software is available upon reasonable request from the corresponding author.
2. Prepare a spectral calibration sample by placing a drop of benzonitrile between two coverslips to form a sealed slide. Place the slide onto the microscope stage and bring it into focus using the excitation objective.
3. Locate the BCARS signal on the sCMOS camera. Adjust the z-position of the collection objective to maximize anti-Stokes signal intensity.
4. Adjust the lateral position of the collection objective to ensure that the anti-Stokes light is correctly focused through the entrance slit of the spectrometer. This alignment ensures optimal coupling into the spectrometer and improved spectral fidelity.
5. Optimize the spatial overlap between the supercontinuum (SC) and probe beams at the sample plane. Use mirrors in the probe beam path to finely adjust its lateral position and incident angle.
6. Translate the optical delay line in the probe beam path to ensure temporal overlap with the SC pulses. Monitor the signal intensity and adjust until the anti-Stokes signal is maximized, confirming proper temporal synchronization.
7. Identify a strong, sharp Raman peak in benzonitrile (e.g., 1002 cm^-1) on the sCMOS image.
8. Record the probe wavelength and its corresponding pixel position on the camera. Use the Rayleigh line (0 cm^-1 at the probe wavelength) and known benzonitrile Raman peaks at 1002 cm^-1 and 3073 cm^-1 to establish a linear pixel-to-wavenumber mapping across the sCMOS detector. The Rayleigh line provides the zero-wavenumber reference, while the benzonitrile peaks determine the wavenumber increment per pixel.
   NOTE: Since the analysis requires only relative wavenumber information (mapping between pixel position and Raman shift), absolute wavelength calibration using emission lamps is not necessary.
9. Input the pixel-wavenumber mapping into the acquisition software. This calibration should remain valid unless changes are made to the optical path, grating setting in the spectrometer, or alignment.

7. Signal acquisition

1. Load the C. elegans sample prepared in steps 1.1 to 1.6 onto the microscope stage. Locate a region of interest and bring it into focus using the excitation objective.
2. Identify a region on the slide that lacks Raman-active signal (water background or a bare glass coverslip) to serve as the nonresonant background (NRB) reference.
3. Locate the BCARS signal on the sCMOS detector. Fine-tune the z-position of the collection objective to maximize the anti-Stokes signal intensity.
4. Adjust the lateral position of the collection objective to ensure that the anti-Stokes light is properly aligned through the spectrometer entrance slit.
5. Start the synchronized scanning and descanning galvo mirrors using the FPGA. Adjust the amplitude of the descanning galvo using the digital potentiometer until the angular deflection from the primary scan is fully compensated. The anti-Stokes signal should remain confined within ~12 rows on the camera across the entire scan range.
6. Define a temporal acquisition sequence
   1. Acquire a cross-correlation signal between the SC and probe pulses by delaying the probe pulse. To acquire this, plot the integrated three-color BCARS signal vs. the time delay of the probe pulse. The strongest signal will appear at time 0 when the SC and probe pulses are temporally overlapped.
      NOTE: Since the FLIM and TPEF detection systems share the same trigger as BCARS, their signals are simultaneously acquired during this time point.
   2. Record at a later delay of the probe beam (e.g., 1-2 ps) for BCARS and a later delay of the SC beam for nonresonant background (NRB spectra in Figure 6A). When a time delay is applied to the narrowband pulse (probe), the NRB in the three-color CARS signal can be effectively suppressed, and thus the signal-to-noise ratio can be improved⁶⁷.
   3. Acquire a dark image with no temporal overlap between the SC and the probe to capture the baseline detector noise (dark reference).
   4. Save all the BCARS, NRB, and dark images, including the metadata within a single h5 file.
7. BCARS Signal Processing Protocol
   1. Open the CRIkit2 software⁶⁸.
   2. Load all necessary datasets using Open HDF tool as shown in Figure 6B: Load BCARS hyperspectral data, NRB reference, and dark reference. Select frequency region within BCARS signal generation bandwidth using the Freq Window tool (from 400 cm^-1 to ≥ 3200 cm^-1 for the BCARS system demonstrated here).
   3. Perform preprocessing steps:
      1. Use the Anscombe tool (as shown in Figure 6B) to variance-stabilize the BCARS and NRB datasets.
      2. Use the De-noise tool to perform Singular Value Decomposition (SVD): Retain principal components based on visual inspection. Preserve components exhibiting spatial distributions matching the morphology of the sample and discard components dominated by unstructured or speckle like noise. This denoising step improves signal-to-noise ratio (SNR), particularly for low-intensity spectral features.
      3. Use the Inverse Anscombe tool to restore the data to its original domain.
         NOTE: These steps may be skipped if performing real-time processing on acquisition.
   4. Retrieve Raman-like spectra using the Kramers-Kronig (KK) transform:
      1. Open the KK tool (as shown in Figure 6B).
      2. Enable Norm to NRB to ensure appropriate spectral reference normalization.
      3. Set parameters including NRB bias, pad factor, and edge pixel averaging.
      4. Execute the KK transform to obtain retrieved Raman spectra.
   5. Perform post-KK error corrections
      1. Correct phase errors using Savitzky-Golay smoothing: Open the phase error correction tool (Figure 6B). Adjust the smoothing window size and polynomial order. Inspect the original vs. corrected spectra using the preview panel.
      2. Correct scale errors: Open the scale error correction tool (Figure 6B). Normalize spectra to correct relative intensity scaling across the dataset. Ensure consistency across spatial and spectral dimensions.
   6. Calibrate spectral axis (optional, if misaligned): Open the calibration tool (Figure 6B). Compare the measured wavenumbers to the known reference peaks. Apply the corrected slope/intercept to realign the spectral axis.
   7. Define and extract data from regions of interest (ROIs)
      1. Use the ROI Spect. or Pt Spect. tool (Figure 6B) to mark spatial regions of interest or points of interest in the BCARS image.
      2. Visualize and compare the spectral content of different ROIs.
      3. Export retrieved spectra for further quantitative or chemometric analysis.

Figure 5 shows the measured spectra of the narrowband probe and the broadband supercontinuum (SC) pulses used for BCARS excitation. The probe pulse, generated via second harmonic generation of a ~1530 nm signal from a home-built optical parametric oscillator (OPO, Figure 3), is centered at 764.2 nm with a full width at half maximum (FWHM) of 0.2 nm, providing the spectral resolution at ~6.8 cm-1 in the resulting Raman spectra. The SC pulse, produced by coupling a pre-chirped femtosecond 1030 nm laser pulse into a 10 cm PCF (see Table of Materials), spans from approximately 950 nm to 1200 nm (≥ 1800 cm-1), covering the whole biologically relevant Raman spectral region (400 to ≥ 3200 cm-1) from fingerprint (through intra-pulse 3-color excitation), silent region (the mix of the 3- and 2-color excitations), and the CH/OH stretch region (through inter-pulse 2-color excitation). Together, these beams enable effective coherent Raman excitation with high spectral specificity and broad vibrational coverage.

To achieve precise synchronization between scanning, descanning, and image acquisition, we implemented a custom control system using an FPGA. The FPGA generates analog waveforms that drive both the scanning galvo and the descanning galvo (Figure 7A). The FPGA receives a line clock to mark the beginning of each scan line and produces duration-adjustable pixel triggers synchronized with the camera exposure (2-4 ms per trigger, equal to pixel dwell time). These signals are distributed across all detection channels, including FLIM, TPEF, and BCARS, to ensure accurate temporal and spatial co-registration. To compensate for slight optical differences introduced by varying objective lenses, the descanning amplitude can be finely tuned via a 7-bit digital potentiometer. A timing diagram illustrating the relationship between galvo signals and trigger outputs is shown in Figure 7B.

Figure 8 shows example images of BCARS, 2p-FLIM, and TPEF images acquired simultaneously from a live, Nile-Red-stained adult worm expressing DHS-3::GFP, which has been accepted as a lipid droplet marker in C. elegans lipid research69. As a reference, we first acquired the BCARS and 2p-FLIM signals from the autofluorescent gut granules of adult wild-type worms without any staining. The autofluorescence gut granules show a low Raman lipid signal (2850 cm-1) and a short fluorescence lifetime ~1 ns (the black spectrum and FLIM decay curve shown in Figure 8E,F, respectively), where our 2p-FLIM results are consistent with recent FLIM findings70. To investigate whether those autofluorescent gut granules contained starch components that were possibly originated from digested bacteria consumed by the worms, we measured the decay curve of a second harmonic generation (SHG) signal. Starch is known to generate SHG signals with zero fluorescence lifetime and therefore can be an SHG reference to estimate the instrument response function (IRF)71. Using our 2p-FLIM system and a thin ZnSe crystal as an SHG reference, we recorded the SHG response. The SHG decay curve showed a very fast lifetime of ~0.12 ns (gray curve in Figure 8F), which is much shorter than the ~0.86 ns lifetime we observed in the autofluorescent granules. This comparison indicates that starch SHG is unlikely to be the main contributor to the measured lifetime of autofluorescent granules. Next, we compared the retrieved Raman spectra from several intestinal regions with DHS-3::GFP signal (ROI 1-4 shown in Figure 8A,C,E) in the Nile-Red stained adult worms expressing DHS-3::GFP. The spectra from ROI 1 and 3 show weak protein signatures but an obvious 2850 cm-1 lipid Raman peak and other lipid-related signatures including the vibration signals at 1740 cm-1 (C=O), 1640 cm-1 (C=C), 1450 cm-1(CH def.), 1080 cm-1 (C-C), and 870 cm-1(C-C). In contrast, although ROI 2 exhibits strong DHS-3::GFP and Nile-Red signals (Figure 8A,B), the absence of lipid signatures and the presence of strong protein peaks including 1003 cm-1 (Phe.), 1665 cm-1 (Amide I), and 2930 cm-1 (CH3) in the ROI2 indicates that this particle was falsely labeled with lipid marker DHS-3::GFP and Nile Red dye. We also found that the ROI 4 spatially contacts a larger lipid particle ROI 3 (Figure 8C) and contains phospholipid-related Raman signatures such as 760 cm−1 (ethanolamine) and 880 cm-1 (νasN+(CH3)3). The spectrum of ROI 4 shows strong lipid and protein peaks indicated in Figure 8E, suggesting that this particle is a yolk particle interacting with a lipid droplet (ROI 3). This can be further confirmed by the Nile Red 2p-FLIM signals that were obtained simultaneously with BCARS. Both ROI 1 and ROI 3 are triglyceride-rich lipid droplets that exhibit a relatively longer lifetime (~3.38 ns and ~3.21 ns, respectively), but ROI 4 shows a shorter lifetime at ~2.73 ns (Figure 8F), which is a typical lifetime value when Nile Red stains polar lipids, consistent with our previous 2p-FLIM finding66. While ROI 2 was stained by Nile Red, its lifetime is significantly shorter than that of lipid-binding Nile Red dye, suggesting that Nile Red falsely tagged ROI2, but this non-specific staining can be separated based on the distinct lifetime. Overall, these suggest that although ROI 1-4 were all tagged by a well-accepted lipid droplet marker DHS-3::GFP, two of them (ROI 2 and 4), which are not triglyceride-rich lipid droplets, were labeled inaccurately. We conclude that multimodal imaging combining BCARS, 2p-FLIM, and TPEF generates high information-content data and can provide rigorous results.

Figure 1: Jablonski diagrams of coherent Raman scattering. (A) Stimulated Raman Scattering (SRS), where Raman-active vibrational modes are coherently driven by two incident fields (pump and Stokes), and the resulting modulation is detected on the pump (SRL) or probe (SRG) beam. (B) Narrowband Coherent Anti-Stokes Raman Scattering (CARS), using narrowband pump and Stokes beams to excite a single Raman mode, with the anti-Stokes signal read out using a degenerate probe. (C) Two-color or multiplex CARS, where a narrowband probe is used in combination with a supercontinuum (SC) Stokes beam to access multiple Raman shifts. (D) Three-color CARS, where both the pump and Stokes pulses are derived from a broadband SC. Abbreviations: SRS - Stimulated Raman Scattering; CARS - Coherent Anti-Stokes Raman Scattering. Please click here to view a larger version of this figure.

Figure 2: Simulated BCARS, NRB, and retrieved Raman spectra. (A) Raw BCARS spectrum and the exact estimation of the NRB reference. (B) Retrieved Raman signal when the Kramers-Kronig (KK) transformation is applied correctly without errors, representing an ideal spectrum retrieval. Abbreviations: NRB: Non-resonant Background; KK: Kramers-Kronig. Please click here to view a larger version of this figure.

Figure 3: Dual output laser and the home-built OPO/SHG system for BCARS probe beam generation. (A) The femtosecond branch passes through a grating compressor, generating a broadband supercontinuum (SC) beam through photonic crystal fiber (PCF). The pump branch drives an optical parametric oscillator (OPO) with a focusing lens (L), two curved mirrors (CM), an end mirror (EM), an output coupler (OC), and a periodically poled lithium niobate (PPLN) crystal. The narrowband signal at ~1530 nm is frequency-doubled in a second PPLN crystal to produce a ~765 nm probe via second harmonic generation (SHG) and is combined with the SC beam using a dichroic mirror (DM). An additional Ti:Sapphire beam is collinearly combined with the probe beam through a polarizing beam splitter (PBS) before the DM. A photodiode (PD) monitors SHG output for cavity stability feedback by adjusting the OPO cavity length (the position of EM) in real-time. (B) Jablonski diagram of the OPO process showing parametric down-conversion from 1030 nm to produce a narrowband signal and idler photon pair. (C) Jablonski diagram of the SHG process for frequency doubling the signal beam to generate the probe pulse. (D) Simplified schematic of the full beam path, highlighting major components and nonlinear optical conversions used to generate the narrowband probe beam. Abbreviations: PPLN: Periodically Poled Lithium Niobate; CM: Curved Mirror; EM: End Mirror; OC: Output Coupler; PD: Photodiode; SPF: Shortpass Filter; PCF: Photonic Crystal Fiber; DM: Dichroic Mirror; L: Lens; TC: Temperature Controller; WW: Wedge Window; PBS: Polarizing Beam Splitter; OPO: Optical Parametric Oscillator; SHG: Second Harmonic Generation. Please click here to view a larger version of this figure.

Figure 4: The multi-modal imaging setup combining BCARS, 2p-FLIM, and TPEF. The supercontinuum (SC) is generated from a 10 cm photonic crystal fiber (PCF) by pumping a 1 W, 130 fs 1030 nm laser. The prechirp (grating compressor) and prism compressor are set to compensate for the dispersion caused by the PCF, aspherical lens (Asph. lens), glasses, and objective lens before the sample stage. The probe pulse is generated from a home-built OPO/SHG system (see the details in Figure 3). A mechanical delay line is set to adjust the length of the probe beam path to match that of SC. After collinearly combining SC and probe beams with a dichroic mirror (DM), the beams are guided into a 1D-galvo/reflective beam expansion system, resulting in a 1D laser scanning geometry with a beam size similar to the back aperture of the focusing objective lens. The forward BCARS signal is collected by another objective lens, through a 4f-descan system, short-pass filters that reject the SC and probe, and finally focused onto the slit of the spectrometer. The 2p-FLIM and TPEF signals are collected with a non-descan geometry in the epi-direction. The additional Ti:Sapphire laser (not shown in this scheme) is collinearly combined with the probe before the DM (see the details in Figure 3). Abbreviations: DM: Dichroic Mirror; Asph. Lens: Aspherical Lens; PCF: Photonic Crystal Fiber; OAP: Off-axis Parabolic mirror; SP filter: Shortpass filter. Please click here to view a larger version of this figure.

Figure 5: The spectra of SC and probe. Both probe and SC spectra are shown on the same plot for comparison. The probe spectrum is centered at 764.2 nm with a full width at half maximum (FWHM) of 0.2 nm. The supercontinuum (SC) spectrum spans from 950 nm to 1200 nm, covering ≥ 1800 cm-1. Please click here to view a larger version of this figure.

Figure 6: BCARS signal processing pipeline. (A) The pipeline of raw BCARS signal processing. (B) The user interface and the steps using customized CRIKit2 software68 to retrieve Raman signal by applying time-domain Kramers-Kronig (TDKK), phase error correction, scale error correction, and wavenumber correction. The detailed mathematical formula can be found in33. Please click here to view a larger version of this figure.

Figure 7: Descanning and synchronization architecture. (A) Schematic diagram of the optical and electronic pathways used for signal descanning and synchronization. Fluorescence emission is collected via dichroic mirrors (DM) and directed to photomultiplier tubes (PMTs) for time-correlated single-photon counting (TCSPC) and analog detection. The anti-Stokes signal for BCARS is descanned using a galvo mirror placed in a 4f relay system. The scanning and descanning galvo mirrors are driven by analog signals from a field-programmable gate array (FPGA), which also generates synchronized trigger signals for pixel clocking for camera, ADC (Analog-to-Digital Converter) and TCSPC. The data are collected by a computer. (B) Timing diagram showing the scanning and descanning galvo drive signals, along with the corresponding line and pixel clock outputs. The FPGA precisely controls both galvos to maintain the anti-Stokes beam within a fixed region on the camera during scanning. Pixel and line clocks are used to synchronize data acquisition across all detection channels. Abbreviations: FPGA: field programmable gate array; PMT: photomultiplier tube; TCSPC: time-correlated single-photon counting; PA: pre-amplifier; ADC: analog-to-digital converter. Please click here to view a larger version of this figure.

Figure 8: The representative results of simultaneous in vivo imaging of BCARS, 2p-FLIM, and TPEF obtained from a Nile-Red stained adult worm expressing DHS-3::GFP. The overlapping images (A) of BCARS (yellow) and GFP signals (cyan) and (B) 2p-FLIM Nile Red (magenta) and GFP (cyan) signals. The scale bar = 10 µm. (C)(D) The enlarged image showing the BCARS, 2p-FLIM, TPEF signal of two connecting particles. (E) Raman spectra retrieved from selected regions of interest (ROIs 1 - 4 shown in (A-D)) along with an autofluorescent reference region from adult wild-type worms without any Nile-Red staining. (F) The fluorescence decay curves of ROI 1 – 4, the autofluorescence reference (from autofluorescent gut granules in wild-type adult worms), and SHG decay curve (gray). The fluorescence lifetimes were exponentially fitted after removing the interference of the instrument response function (IRF). Please click here to view a larger version of this figure.

This work demonstrates the architecture of BCARS and 2p-FLIM/TPEF imaging platform in detail. While a dual-output 1030 nm laser is used in our current setup, most high-power fs 1030 nm fiber lasers can be used as a master laser source for building a BCARS microscope. By carefully choosing the length and period of OPO/SHG PPLN crystals, generating a few-ps probe pulse with a sub-ps OPO 1030 nm pump is achievable. Although a relatively high laser repetition rate (100 MHz) or a relatively lower pulse peak power is demonstrated here, the BCARS signal can still be effectively generated using our BCARS signal excitation scheme that combines intra- and inter-stimulation (3-color and 2-color signal generation). The in vivoC. elegans images shown in the representative results further confirmed the feasibility of performing bioimaging using home-built SC/probe BCARS laser sources. While we previously showed BCARS spectroscopic bioimaging can be achieved by the sample-stage-scanning method^32,34, the same quality of the spectral and spatial resolution can also be achieved by using the 1D-galvo scanning and descan geometry demonstrated here. Finally, we demonstrate the expansion of BCARS imaging platform by integrating 2p-FLIM and TPEF modality. Although the BCARS system is relatively complex, the multimodal BCARS/2p-FLIM/TPEF imaging platform can provide pixel-coregistered high-quality images with a combination of unbiased chemical and spatial information; changes in the fluorescence lifetime responding to the microenvironment, and good spatial distribution of both intrinsic metabolites and fluorescence labels of interest. More importantly, this can be done within live, intact specimens, providing invaluable insights into biological research and can be broadly applied to various biomedical research topics.

The BCARS data processing pipeline using customized CRIKit2 software^33,68 is described in detail (Figure 6). While these procedures can be done in CRIKit2 with a customized Python script, the denoise steps (the steps in the yellow and orange regions shown in Figure 6A) can be done in real-time through a carefully orchestrated combination of multithreaded CPU stages and sequential GPU processing. The CPU pipeline (yellow region shown in Figure 6A) is responsible for high-throughput data preparation and cleaning, operating on a stream of spatial pixels captured from the camera. It is implemented as a series of thread-isolated stages connected through queues, enabling parallel and efficient data flow. The multithreaded CPU stages are configurable based on the signal being processed. For instance, NRB signal processing bypasses stages like Anscombe transformation and quantization. This modular design allows the system to adapt to different acquisition modes with minimal changes to the pipeline structure. Once accumulated, the signal tensor is transferred to the GPU for more compute-intensive sequential processing. The GPU pipeline (orange region shown in Figure 6A) begins with a singular value decomposition (SVD) to identify dominant spectral components. First, all the SVD components will be saved for the case of user-defined postprocessing and reconstruction. After that, a randomized SVD is performed multiple times, allowing the system to determine the number of significant singular values to be retained statistically. The signal is then reconstructed using only the selected components, enabling denoising and dimensionality reduction, followed by inverse Anscombe transformation. This on-the-fly processing approach has been implemented in our acquisition software, greatly reducing the overall BCARS data processing time. We expect that all data processing will be automated in the future by applying parallel calculation for the remaining steps, including TDKK, phase error correction, and scale error correction.

The multimodal system described here can be further expanded by adding additional FLIM and TPEF detectors that enable many-channel detection through implementing a wavelength dispersion element to achieve spectral-resolved FLIM and TPEF imaging. While the 2p-FLIM signal shown in this work used the BCARS SC ranging from 950 nm to 1200 nm as the excitation laser source, an additional Ti:Sapphire fs laser can be guided into the BCARS system. As shown in Figure 3, a tunable Ti:Sapphire fs laser, which has been implemented in our multimodal imaging platform (the dashed cyan line), was collinearly combined with the probe pulse, significantly broadening our two-photon fluorescence imaging capability for a wide range of fluorophores covering from blue to deep red fluorescence emission that have been routinely used in biological studies. However, the additional Ti:Sapphire laser has to be carefully adjusted and calibrated to ensure the overlap between BCARS and its point spread functions. This can be done by using dyed polystyrene beads for fine-tuning the axial and lateral overlapping. Other potential issues with introducing a third laser could be laser damage and interference between the fluorescence excited by SC and Ti:Sapphire pulses if the fluorophore exhibits a wide range of two-photon absorption wavelengths. The former issue can be addressed by experimental determination of the maximum input laser power. The latter can be mitigated by choosing a different laser repetition rate and polarization direction. In our case, we introduced an fs laser with a repetition rate of 76 MHz, which is different from our BCARS laser's 100 MHz repetition. The polarizing beam splitter (The PBS shown in Figure 3A) was used to combine the Ti:Sapphire fs laser and BCARS probe beam, allowing an orthogonal polarization between BCARS SC/probe pulses and the 2p-FLIM excitation pulse to minimize the possible interference between them.

Both BCARS and 2p-FLIM results are system-invariant. The BCARS signal is intrinsically calibrated with an intrinsic reference NRB signal, which has a fixed phase and amplitude relationship to the BCARS signal³³. Consequently, BCARS yields Raman spectra with absolute amplitudes that do not vary day-to-day, or even between instruments. The lifetime data obtained from 2p-FLIM are also system-invariant. The results of FLIM phasor analysis are solely dependent on the intrinsic fluorescence decay property and the microenvironment of the fluorophore because its fluorescence intensity is normalized during the phasor component conversion^72,73. These unique properties enable BCARS and 2p-FLIM to be attractive tools for quantitative comparison among samples and across instruments.

In the representative results, the BCARS spectra and fluorescence decay curves of several particles labeled with DHS-3::GFP are shown. Consistent with our previous finding³⁴, we confirmed that in the adult worms, the DHS-3::GFP can report false-positive results, for example, although ROI 2 shown in Figure 8 was tagged by the DHS-3::GFP reporter, it did not show effective lipid signals in BCARS and 2p-FLIM detections. In our previous 2p-FLIM work⁶⁶, it required a staining time of 4-8 h to examine lipid particles in Nile-Red stained living worms. By contrast, we found that only 1-2 h of staining time is sufficient for the 2p-FLIM imaging using the multimodal system described here. This is because the two-photon fluorescence excitation efficiency using our BCARS SC beam is much higher. Since the SC wavelength is longer than the laser wavelength we used previously for Nile-Red 2p-FLIM imaging⁶⁶, less interference of autofluorescence from the gut granules was observed. The SHG-based signals can be possible separated due to their very short lifetimes. Although several fluorescence decay curves from selected particles are demonstrated in this protocol (Figure 8F), the 2p-FLIM signal of each pixel can be converted to phasor components^72,73 and analyzed with various machine-learning (ML) approaches, as increasing ML methods have been successfully applied to 2p-FLIM data, including ensemble machine-learning for identifying sub-populations of lipid particles⁶⁶, U-Net convolutional neural network (CNN) and Uniform Manifold Approximation and Projection (UMAP) for cell classification^74,75, and combinatorial analysis of FLIM and spatial transcriptomics⁷⁶. While the images of Nile-Red-stained living worms are shown as an example here, the introduction of an additional Ti:Sapphire fs laser described in this work can be broadly applied to almost all the fluorescence labeling routinely used in biological studies, facilitating the understanding of fluorescently labeled objects and their chemical compositions, surrounding microenvironments, and the possible metabolic activities and changes.

In summary, the ultra-high information-content chemical bioimaging combining BCARS, 2p-FLIM, and TPEF modalities can provide valuable, coupled chemical and spatial information within live, intact specimens at sub-cellular resolution. Most importantly, all the BCARS/2p-FLIM/TPEF and potentially second harmonic generation (SHG) / third harmonic generation (THG) images can be obtained simultaneously in a spatially co-registered manner, enabling more complex and comprehensive spatial-spectral and cross-modality data analysis. The platform setup and protocols described here offer interested readers the possibility to adapt this multimodal imaging platform to their research. The practical limitations of the presented system include (i) high equipment cost to build a new system, which can be close to1 M USD, largely depending on the costs of lasers (BCARS and the additional Ti:Sapphire lasers), microscope/objective lenses, spectrometer/camera, TPEF/FLIM detectors and TCSPC card, and galvo-scanning systems and other optics; (ii) high instrumental complexity as the whole system contains more than one hundred components and requires customized controlling, image acquisition, and data analysis programs and software; (iii) a nonlinear optical laser expert is required for Laser beam path and system maintenance to ensure the performance of the system; (iv) user expertise is required as the new users need to be properly trained by an experienced BCARS expert before they can acquire high-quality and reproducible data. Although the technical threshold of building such a system based on BCARS could be high compared to other CRI approaches, we believe that future improvements can be done by simplifying the SC and probe beam generation, integrating a 2D laser scanning and descanning geometry, or introducing a time-domain detection scheme.