Supporting Online Material for

Label-Free Biomedical Imaging with High Sensitivity by Stimulated
Raman Scattering Microscopy

Christian W. Freudiger, Wei Min, Brian G. Saar, Sijia Lu, Gary R. Holtom, Chengwei
He, Jason C. Tsai, Jing X. Kang, X. Sunney Xie*

*To whom correspondence should be addressed. E-mail: xie@chemistry.harvard.edu

Published 19 December 2008, Science 322, 1857 (2008)
DOI: 10.1126/science.1165758

This PDF file includes:

Materials and Methods
Figs. S1 to S6

Other Supporting Online Material for this manuscript includes the following:

(available at www.sciencemag.org/cgi/content/full/322/5909/1857/DC1)
Movies S1 to S3





Supporting online materials


Detailed SRL Apparatus

Synchronized, mode‐locked laser pulse‐trains are provided by an optical parametric oscillator (OPO) (Levante Emerald, APE‐Berlin) synchronously pumped by a frequency doubled Nd:YVO4 laser (picoTRAIN, High‐Q, 532nm, 7 ps, 76 MHz repetition rate). Additionally the pump‐laser (picoTRAIN) provides a separated output of the fundamental beam (1064nm). The OPO uses a temperature‐tuned noncritically phase matched LBO crystal. The coarse tuning of LBO temperature and the fine tuning of a stacked Lyot filter provide wavelength access to all Raman shifts from 500cm‐1‐3400cm‐1. We utilized the signal‐beam from the OPO as the pump‐beam and the fundamental 1064nm‐beam of the pump‐laser as the Stokes‐beam. The idler‐beam of the OPO is blocked with an interferometric filter (Chroma Technology, CARS 890/220m). The pump beam is spatially overlapped with the 1064nm Stokes‐beam with a dichroic beam‐combiner (Chroma Technology, 1064 DCRB). Temporal overlap between pump and Stokes pulse trains is ensured with a delay‐stage and measured with an autocorrelator (APE GmbH, PulseCheck).

The Stokes beam is modulated with a Pockel Cell (ConOptics, model 360‐80) before entering the microscope. The 1.7MHz clock is provided to the driving electronics and the lock‐in amplifier by a square‐wave function‐generator (Stanford Research Systems, DS345, Stanford Research Systems). Care is taken to provide proper RF‐shielding. A polarization‐analyzer is used to transform the polarizationmodulation of the Pockel‐Cell into amplitude‐modulation. The analyzer is positioned after the dichroic beam‐combiner to guarantee identical polarization of pump‐ and Stokes‐beams. Modulation amplitude and DC offset of the RF‐driver (ConOptics, model 25D) are adjusted to maximize the SRL signal from a pure compound tuned into vibrational resonance. Polarization‐modulation instead of intensity‐modulation is also possible in principle, as different tensor elements of the third‐order nonlinear polarizability have different magnitude and would therefore generate an amplitudemodulation of the SRL signal. However the modulation‐depth would be smaller for this approach, resulting in weaker signals. In addition, the average power in the focal spot is higher if polarization modulation is used.

Pump‐ and Stokes‐beams are coupled into a modified laser scanning upright microscope (BX61WI/FV300, Olympus) optimized for near‐IR throughput. For concentration studies and tissue imaging a 60x 1.2NA water (UPlanApo / IR, Olympus) and for cell imaging a 60x 1.1NA water dipping lens (LUMFI, Olympus) are used as excitation objectives. The beam‐size is matched to fill the back‐aperture of the objectives. Light is collected in transmission with a 60x 1.35NA oil objective (UPlanSAPO, Olympus) as condenser, which is aligned with white‐light transmission



from a lamp. A telescope is used to image the scanning mirrors onto the photo‐diode to avoid beam‐movement due to laser scanning.

For epi‐SRL imaging, a polarizing beam‐splitter is positioned ~2cm before the excitation objective with its transmissive polarization direction parallel to the excitation pump and Stokes beams. Between this polarizing beam‐splitter and the objective, an achromatic quarter‐waveplate is inserted into the optical path with its fast axis at 45° with respect to the polarization direction of the excitation beams. Neglecting de‐polarizing effects due to tissue scattering, after double passing the quarter‐wave plate, the back‐scattered light has a perpendicular polarization to the excitation light at the polarizing beam‐splitter and thus is reflected to a custom‐built side‐port. A high OD band‐pass filter (Chroma Technology, CARS 890/220m) is used to block the Stokes beam (at 1064nm) and transmit the pump‐beam only. This filter works for imaging of all Raman‐shifts from 500cm‐1‐3400cm‐1. No leak‐through of the modulated Stokes‐beam could be measured.

A large‐area photodiode (FDS1010, Thorlabs) with reversed bias of 64V is used for detection of the pump beam. The output‐current is low‐pass filtered (Mini‐Circuits, BLP‐1.9+) to suppress the strong signal due to the laser pulsing (76MHz), and then terminated with 50Ω. A high‐frequency lock‐in amplifier (Stanford Research Systems, SR844RF) is used to demodulate the pump‐intensity. The analog R (i.e. modulus) or x (i.e. in phase component)‐output of the lock‐in amplifier is fed into a modified input of the microscopy A/D‐converter.

For the SRL spectral acquisition, we developed an RS232 computer control interface with the OPO in collaboration with APE GmbH (manuscript in preparation). In brief, we tuned the wavelength with the Lyot‐filter within the phase‐matching bandwidth of LBO crystal at a given temperature, allowing up to 300cm‐1 tuning range. The microscope is set to point‐scan, the OPO is tuned, and the intensity for each given Δω is recorded. Intensity variations of the tuned pump‐beam are corrected by simultaneously recording the pump‐intensity with a photodiode. The compensation can be easily performed based on the linear power‐dependence of SRL signal on pump beam intensity (Figure S3).



Fig S1: A high power Nd:YVO4 oscillator (High Q Laser, PicoTrain) provides 22 W of output power at 1064 nm with a 76 MHz repetition rate and 7 ps pulses. The majority of the output power is frequency doubled to produce a beam at 532 nm (~11W) which is split by a 50:50 beamsplitter and used to synchronously pump two optical parametric oscillators (OPOs) (APE GmbH, Levante Emerald). The Stokes beam for SRL is a portion of the undepleted 1064 nm fundamental from the Nd:YVO4 oscillator, which is amplitude modulated by an electro‐optic modulator (ConOptics). An acoustooptic modulator can also be used. Each of the OPOs provides an independently tunable beam which is used as the pump beam for the SRL process. The three beams are combined using two dichroic mirrors. DM1 is an 880 nm longpass mirror (Chroma Technology, 880 DCXR), which transmits pump 2 and reflects pump 1. This filter choice and the tuning range of signal beam of the OPO (680nm‐1010nm) effectively limit the usable range of pump 2 to 890‐1010 nm, and the usable range of pump 1 to 680 nm‐870 nm. An optical delay stage allows the two pump beams to be overlapped in time. After the two pump beams are overlapped, they are combined with the Stokes beam (1064nm) on DM2 (Chroma Technology, 1064dcrb), which reflects 1064 nm and transmits below 1000 nm. An additional optical delay stage allows the Stokes beam to be overlapped in time with the two pump beams. The combined beams are coupled into a custom‐modified laser scanning microscope as described above. After the collection optics, the two pump beams are then separated by DM3 (identical to DM1) and each beam is detected by a photodiode coupled to a lock‐in amplifier as above. An additional narrowband optical filter may be supplied in front of each detector to reduce cross‐talk between the channels. The two lock‐in outputs are digitized on a PC to provide the simultaneously two‐color image.







SRG imaging

In SRG, the pump beam is modulated and the resulting amplitude modulation of the Stokes beam is detected (the opposite of SRL). In principle the SRL and SRG provide equivalent information, as shown in Fig. S4. To implement SRG, we placed an acousto‐optic modulator (Crystal Technology, model 3080‐122) in the pump beam and introduced an amplitude modulation at 10 MHz by modulating the RF power applied to the modulator. A set of 2 emission filters (Chroma Technology, D1125/150m) which blocked the ~800 nm pump laser beam and transmitted the 1064 nm Stokes beam were placed in front of the photodiode detector, and the signal was again processed by a lock‐in amplifier. Because the responsivitiy of the silicon photodiodes that we use is higher for detection of the pump laser beam than for the Stokes beam, and because the collection optics more efficiently transmit the pump beam, SRL is preferred in our instrument and is used for all of the data shown in this report with the exception of Fig. S4. A schematic of the SRG detection scheme is shown in Fig. S2.

Fig. S2: Schematic of SRG. In SRG, the pump laser beam is amplitude modulated and the amplitude modulation is transferred into the Stokes beam as a gain in intensity due to the SRG process. This amplitude modulation of the Stokes beam can be detected by a lock‐in amplifier. The SRG process only occurs when the difference between the pump and Stokes frequency is tuned into vibrational resonance with a molecular species in the sample volume.

Sample preparation concentration curves, spectra and powerdependence.

Methanol, ethanol and trans‐retinol are used as purchased (Sigma Aldrich). Aqueous solutions are prepared with deionized water. Trans‐retinol bottles are opened immediately before use. For spectral measurements and power dependence, we used sample chambers which were made from two cleaned No.1 coverslips separated by 120μm spacers (Grace Biolabs). For the concentration curve, we built a flow‐cell from No.1 coverslips and a spacer to allow quick sample exchange without changing the sample‐environment or focusing depth into the sample.






Sample preparation for live cell imaging

The human lung cancer cell line A549 was obtained from the American Type Culture Collection (ATCC, Rockville, MD). A549 cells are maintained in DMEM (Invitrogen, Carlsbad, CA, USA) supplemented with 5% fetal bovine serum (Invitrogen) at 37°C in a humidified 5% CO2 air incubator. Cells are cultured on uncoated glass bottom dishes (P35G‐1.0‐14‐C, MatTek Cooperation). First we image the control cells without adding EPA to the growth medium. For the uptake study, we add 25μM EPA into the growth medium, and cells are incubated for 24h in the 37°C incubator. After that, cells were taken out of the incubator and imaged under the microscope within 30min.

Sample preparation for tissue imaging & drug delivery

Tissue from wild‐type, white mice are obtained from Dr. S. Kesari and coworkers at Dana Farber Cancer Institute (Boston, USA). Directly after sacrificing, the brain is removed from skull, and ear is removed. Tissue is transported in iced phosphate buffer and imaged within one hour after sacrificing. Brain slices are cut with a razor blade and sandwiched between two coverslips. Thin mouse ear can be imaged whole. No further sample preparation is needed.

For drug delivery experiments with DMSO and RA, mouse ear skin from white, wild‐type mice was used. For DMSO delivery experiments, approximately 25 μl of a 20/80 v/v water/DMSO (Sigma Aldrich) mixture is pipetted onto the 5 x 5 mm piece of skin. The ear tissue is then incubated at 37° C and saturating humidity for one hour. The ~1‐2 mm thick ear tissue is placed between two #1 coverslips and imaged using SRL at 670 cm‐1 for DMSO and 2845 cm‐1 for the endogenous skin lipids.

For retinoic acid delivery experiments, we found that penetration could be optimized by use of a commercially available sonication device that is marketed for transdermal drug delivery in human patients (SonoPrep, Sontra Medical Corp., Franklin, MA). We followed the manufacturer instructions to sonicate the mouse ear tissue. The tissue was removed from the animal and placed in the disposable foam target ring. The ear was cleaned with the recommended skin preparation pad (containing glycerin, methyl‐paraben, propylparaben, benzyl alcohol, potassium sorbate, DMDM hydantoin and water) and the sonicator was placed over the sample. Coupling medium was applied to the sample, followed by a 30 s sonication period.
After sonication, the excess coupling medium was blotted from the sample and 25 μl of a 2% retinoic acid in myritol (Sigma Aldrich) solution was pipetted onto the skin. Five minutes after drug application, the skin was blotted dry, placed between two #1 coverslips and imaged using SRL at 1570 cm‐1 for retinoic acid.

Raman Spectroscopy

Spontaneous Raman spectra are acquired with a confocal Raman microspectrometer (LabRam HR800, Jobin Yvon) under 5mW 633nm He‐Ne laser illumination, with 100x 0.9 NA air objective (MPlan, Olympus) and 10 minutes integration time. The spectrometer/CCD was calibrated with a characteristic line of Si at 520cm‐1.






Image progressing

/wp-content/uploads/2013/09 are acquired with FluoView scanning software and processed using ImageJ (with UCSD plugins for microscopy), IgorPro and Photoshop. LUTs are applied in ImageJ. Graphs were made with IgorPro.





















Fig. S3: The power dependence of the SRL signal. The SRL signal is taken from 10% methanol/water solution tuned to the CH3 stretching mode at 2840 cm‐1. The linear dependence of the signal on both pump beam and Stokes beam power is clear.

Fig. S4: SRL and SRG spectra of pure methanol tuned across the CH3 stretching modes at 2840cm1 and 2950cm1. Within experimental uncertainty, these two spectra are identical as predicted by theory.




Fig. S5: Determination of methnol concentraton limit. For methanol, above, and Fig. 1E, data is sampled in the point‐scan mode of the Olympus FluoView software as a time‐trace, and the solutions with different concentrations are exchanged using a homemade flow cell and a syringe pump. Every concentration measurement is followed by a measurement of the zero‐level (water). No significant change for water signal is found with increasing sample concentration. The lock‐in signal is oversampled by the microscope A/D converter based on the time constants chosen. Time traces ~60 times longer than the time‐constant are recorded. With this time race, the average and standard deviation can be determined at each concentration level. In the concentration curve, the error‐bars are determined from the standard deviation of the recorded SRL signals. With a signal‐to‐noise‐ratio >1.6, the limit of detection is 5 mM and 50 μM for methanol/water and retinol/ethanol solutions, respectively, with laser intensities <40mW for each beam. This corresponds to approximately 3∙105 methanol and 3,000 retinol molecules within the excitation focal volume (about 0.1 femto liter) of the microscope objective.











Fig. S6: Control A549 Cells. A549 human lung cancer cells are grown without adding EPA in the growth medium. SRL /wp-content/uploads/2013/09 at 2920 cm‐1 (left) is much brighter than that at 3015 cm‐1 (right), suggesting that most of the fatty acids inside cells are saturated. Also, fewer lipid droplets can be seen in the cytoplasm of these control cells due to the restricted lipid supply.


SRLmicroscopyBrain.avi: Movie shows three‐dimensional image stack acquired with SRL microscopy in brain. CH2 contrast highlights myelin sheaths around neurons. Three dimensional sectioning capability of SRL microscopy is evident.

SRLmicroscopySkin.avi: Movie shows three‐dimensional image stack acquired with SRL microscopy in brain. CH2 contrast highlights lipid rich structures: inter‐cellular space of stratum corneum, hair follicles in the viable epidermis and sebaceous gland in the dermis (structures are mentioned is the chronological order of the movie).

SRLmicroscopyDMSO.avi: Movie shows three‐dimensional image stack acquired with simultaneous two color SRL microscopy in murine skin. DMSO is applied to the skin and contrast is obtained by tuning into the 670 cm‐1 peak (green channel). Endogenous lipid contrast is obtained from the 2845 cm‐1 peak (red channel). The movie covers a depth of approximately 80 μm from the surface and shows DMSO penetration via a hydrophilic pathway.