Real-Time, Two-Color Stimulated Raman Scattering Imaging of Mouse Brain for Tissue Diagnosis

Summary

Stimulated Raman scattering (SRS) microscopy is a powerful, nondestructive, and label-free imaging technique. One emerging application is stimulated Raman histology, where two-color SRS imaging at the protein and lipid Raman transitions are used to generate pseudo-hematoxylin and eosin images. Here, we demonstrate a protocol for real-time, two-color SRS imaging for tissue diagnosis.

Abstract

Stimulated Raman scattering (SRS) microscopy has emerged as a powerful optical imaging technique for tissue diagnosis. In recent years, two-color SRS has been shown to be able to provide hematoxylin and eosin (H&E)-equivalent images that allow fast and reliable diagnosis of brain cancer. Such capability has enabled exciting intraoperative cancer diagnosis applications. Two-color SRS imaging of tissue can be done with either a picosecond or femtosecond laser source. Femtosecond lasers have the advantage of enabling flexible imaging modes, including fast hyperspectral imaging and real-time, two-color SRS imaging. A spectral-focusing approach with chirped laser pulses is typically used with femtosecond lasers to achieve high spectral resolution.

Two-color SRS acquisition can be realized with orthogonal modulation and lock-in detection. The complexity of pulse chirping, modulation, and characterization is a bottleneck for the widespread adoption of this method. This article provides a detailed protocol to demonstrate the implementation and optimization of spectral-focusing SRS and real-time, two-color imaging of mouse brain tissue in the epi-mode. This protocol can be used for a broad range of SRS imaging applications that leverage the high speed and spectroscopic imaging capability of SRS.

Introduction

Traditional tissue diagnostics rely on staining protocols followed by examination under an optical microscope. One common staining method used by pathologists is H&E staining: hematoxylin stains cell nuclei a purplish blue, and eosin stains the extracellular matrix and cytoplasm pink. This simple staining remains the gold standard in pathology for many tissue diagnoses tasks, particularly cancer diagnosis. However, H&E histopathology, particularly the frozen sectioning technique used in an intraoperative setting, still has limitations. The staining procedure is a laborious process involving tissue embedding, sectioning, fixation, and staining¹. The typical turnaround time is 20 min or longer. Performing H&E during frozen sectioning can sometimes become more challenging when multiple sections are processed at once due to the need to evaluate cellular features or growth patterns in 3D for margin assessment. Moreover, intraoperative histological techniques require skilled technicians and clinicians. Limitation in the number of board-certified pathologists in many hospitals is a constraint for intraoperative consultation in many cases. Such limitations may be alleviated with the fast development interests in digital pathology and artificial intelligence-based diagnosis². However, the H&E staining results are variable, depending on the experience of the technician, which presents additional challenges for computer-based diagnosis².

These challenges can potentially be addressed with label-free optical imaging techniques. One such technique is SRS microscopy. SRS uses synchronized pulsed lasers—pump and Stokes—to excite molecular vibrations with high efficiency³. Recent reports have demonstrated that SRS imaging of proteins and lipids can generate H&E-equivalent images (also known as stimulated Raman histology or SRH) with intact fresh tissue, which bypasses the need for any tissue processing, significantly shortens the time needed for diagnosis, and has been adapted intraoperatively⁴. Moreover, SRS imaging can provide 3D images, which offers additional information for diagnosis when 2D images are insufficient⁵. SRH is unbiased and generates digital images that are readily available for computer-based diagnosis. It quickly emerges as a possible solution for intraoperative cancer diagnosis and tumor margin analysis, especially in brain cancer^6,7,8. More recently, SRS imaging of chemical changes of tissue has also been suggested to provide useful diagnostic information that can further help clinicians stratify different cancer types or stages⁹.

Despite its tremendous potential in tissue diagnosis applications, SRS imaging is mostly limited to academic laboratories specialized in optics due to the complexity associated with the imaging platform, which includes ultrafast lasers, the laser scanning microscope, and sophisticated detection electronics. This protocol provides a detailed workflow to demonstrate the use of a common femtosecond laser source for real-time, two-color SRS imaging and the generation of pseudo-H&E images from mouse brain tissue. The protocol will cover the following procedures:

Alignment and chirp optimization
Most SRS imaging schemes use either picosecond or femtosecond lasers as the excitation source. With femtosecond lasers, the bandwidth of the laser is much larger than the Raman linewidth. To overcome this limitation, a spectral focusing approach is used to chirp the femtosecond lasers to a picosecond timescale to achieve narrow spectral resolution¹⁰. Optimal spectral resolution is only achieved when the temporal chirp (also known as the group delay dispersion or just dispersion) is properly matched for the pump and the Stokes lasers. The alignment procedure and the steps needed to optimize the dispersion of the laser beams using highly dispersive glass rods are demonstrated here.

Frequency calibration
An advantage of spectral focusing SRS is that the Raman excitation can be quickly tuned by changing the time delay between the pump and the Stokes lasers. Such tuning affords fast imaging and reliable spectral acquisition compared to tuning laser wavelengths. However, the linear relationship between excitation frequency and time delay requires external calibration. Organic solvents with known Raman peaks are used to calibrate the Raman frequency for spectral focusing SRS.

Real-time, two-color imaging
It is important to increase the imaging speed in tissue diagnosis applications to shorten the time needed for analyzing large tissue specimens. Simultaneous two-color SRS imaging of lipids and proteins obviates the need to tune the laser or time delay, which increases the imaging speed by more than two-fold. This is achieved by using a novel orthogonal modulation technique and dual-channel demodulation with a lock-in amplifier¹¹. This paper describes the protocol for orthogonal modulation and dual-channel image acquisition.

Epi-mode SRS imaging
The majority of SRS imaging shown to date is performed in transmission mode. Epi-mode imaging detects backscattered photons from tissue¹². For pathology applications, surgical specimens can be quite large. For transmission mode imaging, tissue sectioning is often necessary, which undesirably requires extra time. In contrast, epi-mode imaging can work with intact surgical specimens. Because the same objective is used to collect backscattered light, there is also no need for aligning a high numerical-aperture condenser required for transmission imaging. Epi-mode is also the only option when tissue sectioning is difficult, such as with bone. Previously we have demonstrated that for brain tissue, epi-mode imaging offers superior imaging quality for tissue thickness > 2 mm¹³. This protocol uses a polarizing beam splitter (PBS) to collect scattered photons depolarized by tissue. It is possible to collect more photons with an annular detector at the expense of the complexity of customized detector assembly¹². The PBS approach is simpler to implement (similar to fluorescence), with the standard photodiode already being used for transmission mode detection.

Pseudo-H&E image generation
Once two-color SRS images are collected, they can be recolored to simulate H&E staining. This paper demonstrates the procedure for converting lipid and protein SRS images to pseudo-H&E SRS images for pathology applications. The experimental protocol details critical steps needed to generate high-quality SRS images. The procedure shown here is not only applicable to tissue diagnosis but also can be adapted for many other hyperspectral SRS imaging applications such as drug imaging and metabolic imaging^14,15.

General system requirements
The laser system for this protocol must be able to output 2 synchronized femtosecond laser beams. Systems ideally feature an Optical Parametric Oscillator (OPO) for broad wavelength tuning of one of the laser beams. The setup in this protocol uses a commercial laser system Insight DS+ that outputs two lasers (one fixed beam at 1,040 nm and one OPO-based tunable beam, ranging from 680 to 1,300 nm) with a repetition rate of 80 MHz. Laser scanning microscopes, either from major microscope manufacturers or home-built, can be used for SRS imaging. The utilized microscope is an upright laser scanning microscope built on top of a commercial upright microscope frame. A pair of 5 mm galvo mirrors are used to scan the laser beam. For users choosing to adopt a homebuilt laser scanning microscope, refer to a previously published protocol for the construction of a laser scanning microscope¹⁶.

Protocol

All experimental animal procedures were conducted with 200 µm, fixed, sectioned mouse brains, in accordance with the protocol (# 4395-01) approved by the Institute of Animal Care and Use Committee (IACUC) of the University of Washington. Wild-type mice (C57BL/6J strain) are euthanized with CO₂. Then, a craniotomy is performed to extract their brains for fixation in 4% paraformaldehyde in phosphate-buffered saline. The brains are embedded in a 3% agarose and 0.3% gelatin mixture and sectioned into 200 μm-thick slices by a vibratome.

1. Initial alignment

NOTE: Ensure that the beam size and divergence of both arms are matched for best sensitivity and resolution. Collimate the pump and the Stokes beams and adjust their sizes before they enter the laser scanning microscope. To do this, use a pair of achromatic lenses for each beam before combining them on the dichroic mirror. Always wear proper laser goggles for beam alignment.

1. Beam collimation
   1. Install a pair of achromatic lenses for the pump beam. As a starting point, use a 100 mm lens and 200 mm lens to magnify the laser beam size by 2-fold. Ensure that the distance of the two lenses is roughly 300 mm. Align the pump beam through the center of both lenses.
   2. Place a mirror after the second lens to send the beam toward a wall (>1 m away). Take care when sending the beam across long distances. Trace the beam from the mirror to the wall with an IR card and check if the beam changes in size. Collimate the beam if the beam changes in size as a function of distance.
      1. If the beam is converging (decreasing size with propagation), move the two lenses closer.
      2. If the beam is diverging (increasing size with propagation), move the two lenses further apart. Adjust the distance until the beam is collimated.
   3. Repeat steps 1.1.1 and 1.1.2 for the Stokes beam to collimate the beam.
2. Beam size adjustment
   1. If a beam profiler is available, measure the collimated beam size for each beam. Alternatively, estimate the beam size using the IR card and a ruler to obtain a beam diameter of 4-5 mm.
   2. If the beam size is too small or too large, change the lens pair used in step 1.1. Adjust the lens pair until both beams have a diameter of 4-5 mm.
      NOTE: The magnification of the beam is the ratio between the focal length of the second lens to the first lens (f₂/f₁).
3. Spatial overlap
   NOTE: SRS imaging requires both laser beams to be combined in space and time to excite molecular vibrations. A schematic of the spectral-focusing SRS imaging is shown in Figure 1.
   1. Combine the two laser beams by installing a dichroic mirror with several steering mirrors for adjustment. Optimize spatial overlap of the pump and Stokes by monitoring beams after the dichroic mirror at two different positions far apart (~1 m). Iteratively adjust the steering mirror before the dichroic and the dichroic mirror to align the Stokes beam with the pump beam.
      NOTE: If the two beams are spatially overlapping at both positions, they are sufficiently overlapped.
   2. Ensure that the combined beams are sent to the center of the scan mirrors of the laser scanning microscope by adjusting a pair of steering mirrors when the scan mirrors are in the parked position. Ensure both beams travel through the center of the microscope objective and the condenser.
   3. After the condenser, use another pair of lenses with focal lengths of 100 mm and 30 mm, respectively, to relay the transmitted beam onto the photodiode. Ensure both beams are contained within the photodiode and install two low-pass filters to block out the modulated Stokes beam.

2. SRS signal detection

1. Electrooptical modulation (EOM)
   NOTE: EOM of 20 MHz is used to modulate the Stokes amplitude. As discussed later, the EOM is derived from the 80 MHz laser pulse train, which is required for orthogonal modulation. Other modulation frequencies can be used if only single-color or hyperspectral SRS is performed. In that case, synchronization of modulation frequency to laser frequency is unnecessary. A frequency generator with an RF power amplifier can be used to drive the EOM. As a result, steps 2.1.1-2.1.4 can be skipped.
   1. Place a beam sampler in the Stokes beam path to pick up 10% of the beam and send it to a fast photodiode to detect the 80 MHz pulse train.
      NOTE: The photodiode signal is sent to a frequency divider to generate a 20 MHz TTL output. This output is further sent into a fanout buffer to replicate the output into four identical 20 MHz outputs. One of the outputs is used to trigger the oscilloscope.
   2. Take one of the outputs of the fanout buffer and filter it with a bandpass filter to get a 20 MHz sinusoidal wave. Use an RF attenuator to adjust the output peak-to-peak voltage to ~500 mV.
   3. Send the resulting output to a phase shifter, which allows fine adjustment of the RF phase with a voltage source. Send this output to an RF power amplifier and connect the output of the amplifier to the EOM.
   4. Unblock the Stokes beam and optimize the modulation depth of EOM1 by placing a photodiode in the beam path. Adjust the EOM voltage and quarter-wave plate until the modulation depth (valley to peak ratio) appears satisfactory.
      NOTE: At 20 MHz modulation (1/4 of the laser repetition rate), two pulses are expected every 50 ns.
2. Temporal overlap
   NOTE: Temporal overlap of the pump and Stokes is achieved by delaying one of the two laser pulse trains with a retro-reflector mounted onto a delay stage (Figure 1 shows the Stokes being delayed). Coarse overlap is monitored with the oscilloscope, and fine overlap is monitored by the SRS signal. Fine temporal overlap can also be achieved with an autocorrelator if available.
   1. Place a photodiode after the dichroic mirror to detect the laser beam. Block the Stokes beam first. Zoom in on one of the pump pulse peaks on the oscilloscope. Place a vertical cursor to mark the temporal position of this peak with the oscilloscope.
   2. Block the pump beam and unblock the Stokes beam. Translate the delay stage to temporally match the peak position on the oscilloscope to the marked position in the previous step. See Figure 2 for a display of the temporal overlap of two beams.
      1. (OPTIONAL) If the translation of the delay stage is insufficient to temporally match the two beams, then move the delay stage to the middle of its movement range.
      2. Calculate the delay distance required to match the two beams by taking the temporal difference between the two beams and multiplying the difference by the speed of light to find the amount of distance needed to match the two beams temporally.
      3. Elongate the beam path of the faster beam or shorten the beam path of the slower beam to roughly match the temporal delay accordingly.
   3. Prepare a microscope slide sample with DMSO and double-sided tape as a spacer to hold the sample between the slide and a coverslip.
   4. Place the sample on the microscope with the coverslip side facing the microscope objective. Change the microscope to brightfield illumination and observe the sample from the eyepiece. Find the focus of the sample by first finding the focus at both the top and bottom layer of air bubbles at the glass-tape interface, and then move the focus to be in between the two layers of tape.
      NOTE: Ensure the laser beams are blocked before looking into the eyepiece.
   5. Set the tunable beam output to 798 nm. Based on the optical throughput of the condenser, adjust the optical power to be ~40 mW each for the pump and Stokes beams at the objective focus.
   6. Open ScanImage in MATLAB (or other scanning software that controls the microscope) and click on the button labeled FOCUS to start scanning.
      NOTE: The laser beams will be raster-scanned through the sample to generate an image. The low-frequency signal output from the photodiode (<100 kHz) is directly sent into channel 1 of the data acquisition card (referred to as the DC channel). The high-frequency output (>100 kHz) from the photodiode is sent into the lock-in amplifier, and the X-output of the lock-in amplifier signal is sent into channel 2 of the data acquisition card (referred to as the AC channel).
   7. Adjust the steering mirror before the galvo scanner to center the DC signal on the channel 1 display. Move the motorized delay stage and closely observe the lock-in output shown on the channel 2 (i.e., AC channel) display.
      NOTE: When the pump and Stokes coincide in time, a signal will show up on the AC channel. It is helpful to adjust the color scale of the AC channel to display the small intensity change.
   8. Maximize the AC signal intensity by finely adjusting the time delay. Adjust the dichroic mirror to center the SRS signal on the AC channel (while keeping the DC channel centered). Adjust the phase of the lock-in amplifier to maximize the signal. See Figure 3 for a satisfactory signal.

3. Spectral resolution optimization

NOTE: The pump and Stokes beams reaching the sample should have the same amount of group delay dispersion (GDD) to maximize spectral resolution. The dispersion depends heavily on the experimental setup. The experimental setup described here utilizes femtosecond pulses at 1,040 nm and 800 nm as Stokes and pump, respectively. Dense flint glass rods (H-ZF52A) are used as the pulse-stretching medium.

1. Insert 48 cm of a highly dispersive glass rod (H-ZF52A or equivalent dense flint glass) into the 800 nm beam path. Estimate the GDD using Eq (1):
   (1)
   NOTE: GVD of various glass materials at different wavelengths can be found from the refractive index database resource. For example, H-ZF52A has a GVD of 220.40 fs²/mm at 800 nm. The total GDD is 105792 fs².
2. Calculate how many cm of the dispersive glass rod is required to add to the 1,040 nm beam path to match the GDD of the pump. Insert the appropriate length of dispersive glass rods to the 1,040 nm beam path to roughly match the GDD of the 800 nm beam. Note that the addition of glass rods will change the temporal overlap of the two beams, and adjustment of delay may be necessary.
3. Calibration of spectral resolution
   1. Make a microscope slide sample with DMSO. Place the slide onto the microscope and check the power of the beams coming out of the microscope condenser. Adjust the power accordingly to have ~40 mW each at sample focus.
   2. Open ScanImage from MATLAB. Find the maximal SRS signal by scanning through the delay stage, which corresponds to the 2,913 cm^-1 Raman peak of DMSO. Estimate the stage position based on the previous stage position with the increased optical path length due to the insertion of rods. Realign the beam spatial overlap because of the small deviation of the beam when glass rods are added.
   3. Save a hyperspectral SRS scan by sequentially taking a series of SRS images while moving the motorized stage.
      NOTE: The delay scan range covers two Raman peaks, corresponding to the 2,913 cm^-1 and 2,994 cm^-1 Raman peaks of DMSO, respectively. These two transitions are observed when utilizing an 800 nm pump and 1,040 nm Stokes laser.
   4. Plot out the SRS spectra of the DMSO solution using either ImageJ or MATLAB. Fit the large DMSO 2,913 cm^-1 peak to a Gaussian or Lorentzian function in MATLAB to calculate the Full Width at Half Maximum (FWHM) of the peak.
      NOTE: Representative results are shown in Figure 4. If only one broad peak is present, that means either the spectral resolution is too poor to distinguish the two peaks and more glass rods are required, or the scanned range was too small to detect the second peak. Typically, an acceptable spectral resolution DMSO is ~20-25 cm^-1 when glass rods length of >60 cm are used. A lower resolution is often used for tissue imaging to trade for higher signals with shorter pulses¹⁷.
4. (OPTIONAL) Use an autocorrelator or a FROG (Frequency-Resolved Optical Gating) to determine the pulse duration of each arm to calculate exactly the amount of GDD and the length of rods needed to match the GDD between the pump and the Stokes.
5. Repeat steps 3.3.2-3.3.4 for different rods lengths on the Stokes beam to find the optimal spectral resolution, which means the best GDD match has been found experimentally. Use multiple sets of glass rods differing in length to achieve optimal spectral resolution.

4. Signal to noise (SNR) characterization

1. Ensure that step 4.2 is performed after complete spatial and temporal alignment.
2. Acquire an SRS image corresponding to the 2,913 cm^-1 Raman peak of DMSO. Open the image in ImageJ and select a small area in the center of the frame. Use the measure function to calculate the mean and standard deviation of values in the selected area.
3. Divide the mean value of the selected area by the standard deviation to find the SNR value, as in Eq (2).
   (2)
   NOTE: A good SNR for the system (with a lock-in time constant of 4 μs) using DMSO at 40 mw/40 mw at focus for both arms is >800. Lower concentrations of DMSO or lower power can be used for a more accurate estimation of the SNR if the data acquisition card has a limited bit depth.
4. If the SNR is too low, realign the laser pulses to optimize spatial overlap, temporal overlap, beam size/collimation matching, and/or dispersion matching. For an objective with an aberration correction collar, optimize the signal by adjusting the correction collar.

5. Frequency axis calibration

NOTE: This step is performed to relate the delay stage position to the scanned Raman transition. Careful selection of solvents is required to generate an appropriate "Raman Ruler." DMSO is an effective solvent for CH bonds as it has two sharp Raman peaks at 2,913 cm^-1 and 2,994 cm^-1.

1. Save a hyperspectral scan with the delay-stage range covering the 2,913 cm^-1 and 2,994 cm^-1 Raman peaks of DMSO. Save the stage positions corresponding to the hyperspectral dataset.
   NOTE: The global maximum peak of the spectrum corresponds to the DMSO 2,913 cm^-1 Raman shift and the second maximum peak corresponds to the DMSO 2,994 cm^-1 Raman shift.
2. Perform linear regression for the stage positions and Raman shifts at 2,913 cm^-1 and 2,994 cm^-1. Using the linear regression equation relating the stage position to the Raman shift, convert the delay positions to the corresponding Raman frequencies.

6. Orthogonal modulation and two-color imaging

NOTE: The orthogonal modulation step is only necessary when real-time two-color imaging is needed. A schematic of this scheme is shown in Figure 5. The orthogonal modulation uses a pair of EOMs driven at a quarter of the laser frequency (20 MHz for 80 MHz laser) with a 90° phase shift between the two. This orthogonal modulation step can be skipped for single-color SRS imaging or hyperspectral SRS imaging.

1. EOM1 modulation
   1. Install a PBS (PBS2), a quarter-wave plate (QWP2), and a second EOM (EOM2) into the Stokes beam path after the first EOM. Unplug the signal input to EOM2. Plug in the signal input to EOM1 and turn it on.
   2. Modulate the Stokes beam (fixed at 1,040 nm) at 20 MHz (f₀/4) by sending the beam through the first EOM. Adjust the tilt and position of EOM1 to ensure that the beam is hitting straight and centered through the EOM crystal.
   3. Monitor the modulation depth by observing both polarizations coming out of PBS1 with two photodiodes and displaying the modulation on an oscilloscope.
   4. Adjust the QWP1, EOM1 voltage, and phase of the 20 MHz input (using a phase shifter) to optimize the modulation depth of the transmitted beam to be close to 100%. See Figure 2B for an illustration of good modulation depth.
2. EOM2 modulation
   1. Unplug EOM1 and plug in EOM2.
   2. Send in the high voltage output of the second amplifier at 20 MHz to EOM2. Adjust the tilt and position of EOM2 to ensure that the beam is hitting straight and centered through the EOM crystal.
   3. Once again, monitor the modulation depth by looking at both polarizations coming out of PBS2 with an oscilloscope. Adjust the QWP2, EOM2 voltage, and the phase shifter as needed to achieve close to 100% modulation for both polarizations.
   4. Ensure that the pulse train modulation has a 90° phase shift from the first modulation.
      NOTE: If the two pump pulse trains are not 90° orthogonal, crosstalk between the two channels will be a problem.
   5. Test the orthogonality of the modulation by turning on and plugging in both EOM1 and EOM2. Monitor both polarizations being split by PBS2 with an oscilloscope. Reoptimize EOM1 and EOM2 individually if the pulse train after the second PBS does not resemble Figure 2C.
   6. Install a 20 mm birefringent quartz crystal (BRC) and HWP downstream of EOM2. Plug in both EOMs at once and monitor the pulse train such that it resembles Figure 2D.
      NOTE: For the chirp used in this experiment, 20 mm BRC induces a time delay that corresponds to an 80 cm^-1 Raman shift. A different BRC length may be needed if a different chirp is used.
3. Calibration
   1. Calibrate the system using DMSO by detecting signals from the lock-in amplifier X and Y channel output (sent to channels 2 and 3 of the data acquisition card).
   2. Check whether the signals generated by the 2,913 cm^-1 peak from the faster polarization and that from the slower polarization are close to 90° out-of-phase on the lock-in amplifier. If this is not the case, adjust the EOM alignment until the two signals are close to 90° out-of-phase.
   3. Once calibration is complete, find the delay position that probes the protein transition at 2,930 cm^-1 for one of the orthogonal beams. Ensure that the other polarization probes the lipid transition of 2,850 cm^-1.

7. Epi-mode SRS imaging

NOTE: In the transmission mode imaging scheme, the objective focuses the laser into the sample, and then a condenser lens directs the transmitted beam to a photodiode for lock-in detection. In the epi-mode imaging scheme, light that is backscattered and depolarized by the sample is recollected by the focusing objective and isolated using a polarizing beam splitter. The isolated and backscattered photons are sent to a photodiode through a pair of relay lenses for lock-in detection. Figure 6 depicts the epi-mode imaging scheme.

1. Install an HWP before the beam enters the microscope to change the polarization of the beam going into the microscope. Place a PBS above the objective to allow the depolarized back-reflected beam to reach the detector.
2. Use a pair of lenses consisting of a 75 mm achromat lens and a 30 mm aspheric lens to relay the backscattered photons from the back aperture of the objective to the photodetector. Mount the detector to collect the backscattered light directed by the PBS. Install a filter to block out the modulated beam from entering the detector.
3. Place the tissue sample under the objective. As the condenser is unnecessary for epi-mode imaging, remove it if more space is required.
4. Imaging
   1. Block off the beam with a shutter; shine a white light source onto the sample from the side; and use brightfield to find objective focus.
   2. Unblock both beams and use the precalibrated delay positions to acquire lipid and protein SRS images from tissue from the two outputs of the lock-in amplifier.
   3. Adjust the lock-in gain and pixel bin factor to acquire good-quality images.

8. False-color staining

1. Open the image stack with ImageJ.
2. Pull out the two images that correspond to the lipid (2,850 cm^-1) and protein (2,930 cm^-1) species by right-clicking on the image and clicking on Duplicate.
3. Rename the lipid image to lipids and the protein image to proteins.
4. Go to Process | Image Calculator and perform proteins subtract lipids.
5. Combine the images by going to Image | Color | Merge Channels, setting lipids to green and proteins to blue. Open the image channels tool (Image | Color | Channels Tool) and adjust the brightness and contrast (Image | Adjust | Brightness/Contrast).
6. Adjust the brightness and contrast for each channel using the channels tool. For the lipid channel, adjust the contrast until the cellular features appear dark. For the protein channel, adjust the contrast until the cellular features appear blue. Convert the merged channel green/blue image to an RGB image by going to Image | Type | RGB Color. Export this image by File | Save As | Tiff.
   NOTE: For false H&E staining, the color scheme shows pink cytoplasm, while the nuclei are dark blue-purple.
7. Download the HE.m MATLAB script from the false H&E staining script resource in the Table of Materials.
8. Run the HE.m script in MATLAB. Select the exported RGB image from the previous step to generate an artificially H&E stained image.
9. (OPTIONAL) Normalize the image intensity for large field-of-view imaging because the image appears darker in the periphery than in the center.
   1. To perform field normalization of the images, average as many images as possible. Then, remove the intensity features with ImageJ (Process | Filters | Gaussian Blur | Radius=50).
   2. Measure the maximum intensity of the blurred image (Ctrl+M). Divide the blurred image by the maximum intensity (Process | Math | Divide). Divide the raw SRS image by the blurred image (Process | Image | Calculator).

Representative Results

Optimizing spectral resolution:
Dispersion through a material is affected by the dispersive medium (length and material) and wavelength. Changing the dispersion rod length affects the spectral resolution and the signal size. It is a give-and-take relationship that can be weighed differently depending on the application. The rods stretch out the beam pulse from being wide in frequency and narrow in time to being narrow in frequency and broad in time. Figure 7 shows the effect of varying rod lengths on the spectral resolution. Figure 7A demonstrates very poor spectral resolution; this setup has no glass chirping rods, and the two Raman peaks from DMSO are not resolved at all. In Figure 7B,C, an increase in the number of glass chirping rods starts to resolve the two peaks. Lastly, Figure 7D shows how matched chirping resolves both peaks and can be used to calibrate stage positions to frequency.

Calibration with DMSO:
DMSO has two sharp Raman peaks at 2,913 and 2,994 cm^-1 that are convenient for calibration in the C-H region. Once a DMSO spectrum is obtained from a hyperspectral SRS scan, a simple linear regression converts the stage position to the Raman shift. If the spectral resolution is poor (as shown in Figure 7A) and the two peaks are not separable, then calibration with linear regression is impossible. In this case, better dispersion matching is necessary by either adding or removing glass rods. The most common difficulties in finding a DMSO signal for calibration are due to errors in either temporal overlap or spatial overlap of the two beams. Before attempting DMSO calibration, repeat the spatial and temporal alignment steps to optimize the SRS signal.

Two-color DMSO:
In two-color SRS, two pump pulse trains are generated with orthogonal polarization with a fixed time delay (due to a birefringent crystal). Evaluation of modulation depth and 90° phase shift is done with a photodiode followed by a DMSO SRS signal. Figure 8 demonstrates acceptable modulation depth and temporal separation. Although the spectral resolution of DMSO in Figure 8 is not ideal, it is often sacrificed in tissue imaging experiments to achieve a higher signal with shorter pulses. Figure 9 demonstrates poor phase shifts that give rise to inverted or negative peaks.

Two-color SRS of mouse brain tissue in epi-mode:
Epi-mode (detecting backscattered photons) SRS is used for imaging thick tissue (>1 mm). Figure 10A,B demonstrates real-time, two-color SRS imaging at 2,850 cm^-1 and 2,930 cm^-1 of ex vivo mouse brain tissue. The raw images from the lipid and protein channels (Figure 10A,B) were color-coded to produce a single image depicting lipid and protein contributions (Figure 10C). False H&E staining was performed (Figure 10D) on Figure 10C to mimic H&E staining. Poor imaging quality can be a result of large imaging depth or poor calibration (frequency axis or modulation depth). The typical imaging depth in brain tissue is 100-200 μm¹³.

Figure 1: Schematic of one-color SRS imaging setup. Construction of a spectral-focusing SRS microscope in transmission mode. X and Y represent the orthogonal outputs. Abbreviations: SRS = stimulated Raman scattering; DL = retroreflector-based delay line; Div = divider; FB = fanout buffer; AT = attenuator; PS = phase shifter; PA = power amplifier; DCM = dichroic mirror; GM = galvo mirrors; EOM = electrooptic modulator; POM = pick-off mirror; PBS = polarizing beam splitter, BRC = birefringent crystal; QWP = quarter-wave plate; HWP: half-wave plate; PD = photodiode; GR = glass rod; BB = Beam Block; SPF = shortpass filter; CL = collimating lens; BS = beam size changing lens. Please click here to view a larger version of this figure.

Figure 2: Representative temporal overlap. (A) The pump and Stokes beams are seen to be temporally overlapped on the oscilloscope. Oscilloscope cursors are used to mark the temporal positions of the pump and Stokes beams. This overlap is satisfactory as a starting point to further tweak temporal overlap with a delay stage. (B) Satisfactory representative modulation depth of one EOM at 20 MHz. (C) Satisfactory pulse modulation while two EOMs are in use. (D) Satisfactory pulse train modulation after birefringent quartz crystal and half-wave plate installation on the doubly modulated Stokes arm. Abbreviation: EOM = electrooptic modulator. Please click here to view a larger version of this figure.

Figure 3: Representative SRS and pump signals. (A) Misaligned pump signal detected in DC channel. (B) Misaligned SRS signal detected by photodiode. (C) Satisfactory, centered pump signal in the DC channel. (D) Satisfactory SRS signal centered on the AC channel. Abbreviation: SRS = stimulated Raman scattering. Please click here to view a larger version of this figure.

Figure 4: Spectral resolution characterization. A gaussian function was fit to the 2,913 cm^-1 Raman DMSO peak. The FWHM calculated gave a resolution of 15 cm^-1 for the system. Abbreviations: DMSO = dimethyl sulfoxide; FWHM = Full Width at Half Maximum. Please click here to view a larger version of this figure.

Figure 5: Schematic of two-color SRS imaging setup. Construction of a two-color SRS microscope in the transmission mode. X and Y represent the orthogonal outputs. Abbreviations: DL = retroreflector based delay line; Div = divider; FB = fanout buffer; AT = attenuator; PS = phase shifter; PA = power amplifier; DCM = dichroic mirror; GM = galvo mirrors; EOM = electrooptic modulator; PBS = polarizing beam splitter; BRC = birefringent crystal; QWP = quarter-wave plate; HWP = half-wave plate; PD = photodiode; GR = glass rod; BB = Beam Block, SPF = shortpass filter; CL = collimating lens; POM = pick-off mirror; BS = beam size changing lens. Please click here to view a larger version of this figure.

Figure 6: Schematic of two-color SRS imaging Epi-mode set-up. Construction of a two-color SRS microscope in the epi-mode. X and Y represent the orthogonal outputs. Abbreviations: DL = retroreflector based delay line; Div = divider; FB = fanout buffer; AT = attenuator; PS = phase shifter; PA = power amplifier; DCM = dichroic mirror; GM = galvo mirrors; EOM = electrooptic modulator; PBS = polarizing beam splitter; BRC = birefringent crystal; QWP = quarter-wave plate; HWP = half-wave plate; PD = photodiode; GR = glass rod; BB = Beam Block, BPF = bandpass filter; CL = collimating lens; POM = pick-off mirror; BS = beam size changing lens. Please click here to view a larger version of this figure.

Figure 7: Optimizing spectral resolution with 25% DMSO. (A) Zero glass chirping rods were used to obtain the DMSO spectrum. The two peaks are not resolved. (B) Glass chirping rods were used, 20 cm on the pump arm and 24 cm on the Stokes arm. Two peaks are starting to be resolved to a satisfactory point. (C) Chirping rods were used, 40 cm long pump and 24 cm long Stokes, to obtain a better resolved DMSO spectrum. (D) Chirping rods were used, 64 cm long pump and 60 cm long Stokes, to obtain higher spectral resolution. Abbreviation: DMSO = dimethyl sulfoxide. Please click here to view a larger version of this figure.

Figure 8: Two-color SRS, varying polarization and time delay. Two time-delayed pulse trains of orthogonal polarization (s&p) are used to image DMSO spectra to show the time delay (and Raman frequency difference) between the SRS excitations. Abbreviations: DMSO = dimethyl sulfoxide; SRS = stimulated Raman scattering. Please click here to view a larger version of this figure.

Figure 9: SRS spectrum resulting from a poor two-color SRS phase shift. A negative 2,994 cm^-1 peak near stage position 48 is indicative of poor phase difference. Abbreviation: SRS = stimulated Raman scattering. Please click here to view a larger version of this figure.

Figure 10: Generation of false two-color H&E staining of SRS images. (A) Raw protein SRS image from a mouse brain tissue at the 2,930 cm^-1 transition. (B) Raw lipid SRS image from mouse brain tissue at the 2,850 cm^-1 transition. (C) Merged and color-coded channels from A and B with lipid contribution in green and protein contribution in blue. (D) False H&E recoloring was performed on C to mimic H&E staining for pathological applications. Scale bar = 50 µm. Abbreviations: SRS = stimulated Raman scattering; H&E = hematoxylin and eosin. Please click here to view a larger version of this figure.

Discussion

The two-color SRS imaging scheme presented in this protocol hinges on the proper implementation of one-color SRS imaging. In one-color SRS imaging, the critical steps are spatial alignment, temporal alignment, modulation depth, and phase shift. Spatially combining the two beams is accomplished by a dichroic mirror. Several steering mirrors are used for fine adjustment when sending the beams to the dichroic mirror. Once the beams are combined with the dichroic mirror, spatial alignment can be confirmed by picking off the combined beam path with a mirror to send toward a wall >1 m away while checking the combined beam to see if it overlaps with an IR card. Following spatial alignment through the microscope, the temporal overlap is performed by adjusting the path length with a retroreflector-based delay stage, which has the benefit of preserving spatial alignment. The 1,040 nm arm is the delayed beam in this protocol. Temporal overlap is impossible if the delay stage does not have the range to cause the two beams to overlap temporally. In that case, it may be necessary to move the entire delay stage to a different location to ensure that temporal overlap is close to the middle-of-the-delay-scan range. For example, if the temporal difference between the two beams is measured as 6 ns, then 1.8 m of adjustment is required. The adjustment required denotes the length of elongation of the beam path for the faster beam or the shortening of the beam path for the slower beam.

Besides alignment, beam size matching and modulation of the Stokes beam are also important to maximize the SRS signal. Ideally, both beams should be collimated at the objective plane and match the size of the objective back aperture. It is useful to check collimation and beam size if the objective plane is exposed to the user. If the beam is converging or diverging, it is necessary to adjust the second lens of the collimation lens pair to achieve collimation (protocol step 1.1). Similarly, if the beam size is too small at the objective back aperture, a lens pair with high magnification needs to be used (protocol step 1.2). SRS scales linearly with the modulation depth of the Stokes beam. It is important to ensure that the pulse height at the valley is <10% of the pulse height at the peak on the oscilloscope. Poor modulation of the Stokes beam directly translates to poor SNR.

When using femtosecond lasers for SRS imaging, spectral focusing is necessary to convert the time delay between pump and Stokes to Raman frequency shift. The conversion factor depends on the amount of chirp applied. It is critical to match the GDD of the pump and Stokes to achieve optimal spectral resolution. The GDD can be estimated based on the length of the dispersion material used and its GVD. For example, the SF11 dense flint glass rod has a GVD of 123.270 fs²/mm at 1,040 nm and 187.486 fs²/mm at 800 nm. For 240 mm of SF11 in the 800 nm beam path, GDD is 240 mm × 187.486 fs²/mm = 45,000 fs². For 240 mm of SF11 in the 1,040 nm beam path, the GDD is 240 mm × 123.270 fs²/mm = 30,000 fs². This example calculation means that additional glass rods should be added to the 1,040 nm arm, or glass rods should be removed from the 800 nm arm to match GDD. To achieve higher spectral resolution, a larger GDD is required, which means longer rods. However, in calculating GDD, the contributions from the EOM and objective have been neglected. The actual matching of GDD is determined experimentally by finding the best spectral resolution for the given rod lengths on the pump or the Stokes. The calculations serve as a good beginning step. Experimental optimization through iterative addition and removal of glass rods is still necessary to optimize spectral resolution.

It is important to realize that a large chirp provides higher spectral resolution but at the expense of signal intensity. Smaller chirps and shorter pulses are beneficial for tissue imaging in the C-H region because protein and lipid Raman peaks are broad. Lower spectral resolution can be traded for a higher SRS signal (or faster imaging speed) without sacrificing two-color tissue contrast¹⁷. In other applications where high spectral resolution is needed, especially in the fingerprint region, it is necessary to apply a larger chirp using long glass rods. Alternatively, it might be easier to use grating stretchers to obtain longer pulses (for pulse duration longer than 3 ps). However, one main limitation of the utilized commercial laser is its spectral bandwidth. The spectral coverage of spectral-focusing SRS depends on the bandwidth of the excitation lasers. The utilized laser system has a spectral coverage typically around 200-250 cm^-1. This is barely sufficient to cover the C-H region. A larger spectral coverage is typically needed to resolve chemical species for imaging in the fingerprint region. This problem can be addressed with a fiber laser add-on that broadens the bandwidth of the Stokes laser from 6 nm to 60 nm¹⁸. Another important limitation to the two-color SRS imaging technique is that only two transitions are monitored. That approach would be unsuitable for complex samples with many overlapping Raman peaks or multiple species.

The real-time two-color SRS imaging method affords high-speed tissue imaging by removing the need for laser or delay tuning. However, it is difficult to set up due to the challenges in achieving near-perfect modulation depth for both EOMs. It is best to optimize EOM1 and EOM2 independently. If EOM1 is on, then EOM2 must be unplugged and vice versa. Once both modulations are optimized to near-perfection (>95% modulation depth), both EOMs are connected to enable orthogonal modulation. The length of the time delay between the two orthogonal pulse trains is dependent on the length of the BRC and the chirp. This modulation method is not immediately feasible for clinical applications as complex electronics are needed to provide two RF pulse trains with a tunable phase shift to drive the two EOMs. The alignment of the EOM also needs to be near-perfect to ensure high transmission, good modulation, and orthogonality between the two channels. This technique is generally applicable to other applications that require fast, simultaneous imaging of two Raman peaks due to motion or sample changes. Examples include water temperature measurements or tracking of moving objects such as lipid droplets or organelles^11,19.

A robust fiber laser is required for future clinical applications⁴. The approach described by the protocol could also be extended to improve acquisition speed up to video rate, which is important to scan large tissue specimens within a reasonable time. If imaging time or motion artifacts are not a concern, then protein and lipid SRS images can be acquired sequentially by moving the motorized delay stage frame-by-frame. Another real-time, two-color SRS imaging method is a dual-phase scheme that recycles the Stokes beam to provide the same two orthogonal channels²⁰. However, implementation of the dual-phase scheme requires extra alignment that involves three beams. It also requires matching of the beam size and divergence of three laser beams. A potential avenue to improve both techniques to overcome the simultaneous two-color limit is the incorporation of a rapidly tunable fiber laser to probe specific spectral regions²¹. The processing of images is the same as outlined in the protocol to generate simulated H&E images.

Finally, this protocol demonstrates epi-mode SRS imaging. It typically generates lower-quality images compared to transmission mode imaging for thin tissue sections because of the lower pump power reaching the detector¹³. For thick tissue imaging (>1-2 mm) or samples that are highly scattering (e.g., bone tissue), epi-mode imaging can perform better than transmission mode imaging. When imaging fresh tissue such as brain tissue, the SRS imaging depth is typically limited to 200-300 µm. For fixed tissue, scattering is stronger, and the imaging depth is 100-200 µm. Deeper imaging can be achieved with higher power, aberration correction, or optical clearing^22,23. Nevertheless, epi-mode imaging is a preferable approach for tissue diagnosis because it does not require any tissue sectioning, and the alignment is simpler without the high NA condenser. Future tissue diagnosis applications will benefit from rapid, large area imaging on intact surgical specimens followed by machine learning/deep learning-based diagnosis. The protocol presented here is also suitable for in vivo applications, such as brain imaging or skin imaging, where epi-mode imaging is the only option, and high-speed imaging is important to avoid motion artifacts.

Materials

Name	Company	Catalog Number	Comments
100 mm Achromatic Lens	THORLABS	AC254-100-B	Broadband, 650 - 1,050 nm, achromatic lens focal length, 100 mm
20 MHz bandpass filter	Minicircuits	BBP-21.4+	Lumped LC Band Pass Filter, 19.2 - 23.6 MHz, 50 Ω
200 mm Achromatic Lens	THORLABS	AC254-200-B	Broadband, 650 - 1,050 nm, achromatic lens focal length, 200 mm
Achromatic Half Waveplate	Union Optic	WPA2210-650-1100-M25.4	Broadband half waveplate
Achromatic Quarter Waveplate	Union Optic	WPA4210-650-1100-M25.4	Broadband quarter waveplate
Beam Sampler	THORLABS	BSN11	10:90 Plate Beamsplitter
Dichroic Mirror	THORLABS	DMSP1000	Other dichroics with a center wavelength around 1,000 nm can be used.
DMSO (Dimethyl sulfoxide)	Sigma Aldrich	472301	Solvent for calibration of Raman shift. Other solvents with known Raman peaks can be used.
Electrooptic Amplitude Modulator	THORLABS	EO-AM-NR-C1	Two EOMs are needed for orthogonal modulation and dual-channel imaging. Resonant version is recommended so lower driving voltage can be used.
False H&E Staining Script	Matlab		https://github.com/TheFuGroup/HE_Staining
Fanout Buffer	PRL-414B	Pulse Research Lab	1:4 TTL/CMOS Fanout Buffer and Line Driver, for generating the EOM driving frequency and the reference to the lock-in
Fast Photodiode	THORLABS	DET10A2	Si Detector, 1 ns Rise Time
Frequency Divider	PRL-220A	Pulse Research Lab	TTL Freq. Divider (f/2, f/4, f/8, f/16), for generating 20MHz from the laser output.
Highly Dispersive Glass Rods	Union Optic	CYLROD01	High dispersion H-ZF52A Rod lens 120 mm, SF11 Rod lens 100 mm
Insight DS+	Newport		Laser system capable of outputting two synchronzied pulsed lasers (one fixed beam at 1, 040 nm and one tunable beam, ranging from 680-1,300 nm) with a repetition rate of 80 MHz.
Lock-in Amplifier	Liquid Instruments	Moku Lab	Lock-in amplifier to extract SRS signal from the photodiode. A Zurich Instrument HF2LI or similar instrument can be used as well.
Mirrors	THORLABS	BB05-E03-10	Broadband Dielectric Mirror, 750 - 1,100 nm. Silver mirrors can also be used.
Motorized Delay Stage	Zaber	X-DMQ12P-DE52	Delay stage for fine control of the temporal overlap of the pump and the Stokes lasers. Any other motorized stage should work.
Oil Immersion Condensor	Nikon	CSC1003	1.4 NA. Other condensers with NA>1.2 can be used.
Oscilloscope	Tektronix	TDS7054	Any other oscilloscope with 400 MHz bandwdith or higher should work.
Phase Shifter	SigaTek	SF50A2	For shifting the phase of the modulation frequency
Photodiode	Hamamatsu Corp	S3994-01	Silicon PIN diode with large area (10 x 10 cm2). Other diodes with large area and low capacitance can be used.
Polarizing Beam Splitter	Union Optic	PBS9025-620-1000	Broadband polarizing beamsplitter
Refactive Index Database			refractiveindex.info
Retro-reflector	Edmund Optics	34-408	BBAR Right Angle Prism. Other prisms or retroreflector can be used.
RF Power Amplifier	Minicircuits	ZHL-1-2W+	Gain Block, 5 - 500 MHz, 50 Ω
Scan Mirrors	Cambridge Technologies	6215H	We used a 5mm mirror set with silver coating
ScanImage	Vidrio	ScanImage Basic	Laser scanning microscope control software
Shortpass Filter	THORLABS	FESH1000	25.0 mm Premium Shortpass Filter, Cut-Off Wavelength: 1,000 nm. For efficient suppression of the Stokes, two filters may be necessary.
Upright Microscope	Nikon	Eclipse FN1	Any other microscope frame can be used. If a laser scanning microscope is available, it can be used directly. Otherwise, a galvo scanner and scan lens needed to be added to the microscope.
Water Immersion Objective	Olympus	XLPLN25XWMP2	The multiphoton 25X Objective has a NA of 1.05. Other similar objectives can be used.