Electronically preresonant stimulated Raman scattering microscopy in the visible

Vibrational spectra of molecules are typically composed of many bands with widths of a few tens of cm⁻¹ distributed over a range of 3000 cm⁻¹. This wealth of information allows the identification of molecules, which motivates the application of vibrational spectroscopy to the generation of molecule specific contrast in microscopy. Traditionally, this is done using infrared absorption in FT-IR microscopy¹ or spontaneous Raman scattering in confocal Raman microscopy.² Both methods, however, have drawbacks that are causes for their shadowy existence compared to fluorescence microscopy: Applications of FT-IR microscopy are hampered by low spatial resolution and the strong absorption of water, whereas spontaneous Raman microscopy suffers from sample autofluorescence and low sensitivity. During the last two decades, some of these limitations have been overcome by the development of non-linear Raman microscopy, i.e., coherent anti-Stokes Raman scattering (CARS) and stimulated Raman scattering (SRS) microscopy.^3,4 Sample autofluorescence is not a problem when using either of these methods, and image acquisition times are short since usually only one vibrational frequency is imaged at a time. Yet, the low sensitivity remains a problem also for these techniques.

With epr-SRS microscopy, detection sensitivities in the low μM range can be achieved, which corresponds roughly to one thousand molecules in the focal volume.^5,6,10,11 An important aspect of these experiments is that the excitation frequencies, ω_p and ω_S, have to be close to ω₀. A problem in this respect is that the laser systems used for SRS microscopy have to fulfill a number of other requirements: (i) high average power at high repetition rates; (ii) excitation pulse durations of approximately 1 ps; (iii) synchronization of pump and Stokes pulses; (iv) Stokes wavelength tunability independent from the fixed pump wavelength; and (v) low laser intensity noise. For standard SRS microscopy, most of these aims are met by using a ps near-IR pumped optical parametric oscillator (OPO). The fixed wavelength near-IR pump then serves as Stokes laser, and the signal output of the OPO is used as the tunable SRS pump laser.¹² For the detection of the SRS signal with a lock-in amplifier, one of the beams is preferentially modulated at half the repetition rate with an electro-optical modulator (EOM).¹³ The disadvantage of using similar setups in epr-SRS microscopy is that only a very limited range of compounds with electronic resonances at the red edge of the visible spectrum can be investigated.⁶ To date, two SRS experiments with excitation in the visible spectral range have been published.^11,14 In both cases, the spectra were only recorded over a range of approximately 200 cm⁻¹. A further complication arises from the fact that in these experiments, the pump beam wavelength was tuned to record SRS spectra. Since the pump beam provides the dominant contribution to the preresonant enhancement, changing its wavelength has a strong influence on the vibrational spectra.

Here, we report a method for broadband epr-SRS imaging with excitation in the visible region, using a combination of spectral focusing and wavelength tuning. Spectral focusing^15,16 allows us to record SRS spectra at high tuning speeds over short vibrational ranges [Fig. 1(b)], whereas the wavelength tuning of the central wavelength from the OPO extends the measurements over the whole vibrational region with little influence on the preresonance enhancement. Since wavelength tuning of OPOs is slow, recording of an SRS spectrum in the finger print region (⁠ $≃$ 1000 cm⁻¹) with acceptable spectral resolution (10 cm⁻¹) can take more than 30 min. In this work, we recorded SRS spectra up to 20 times faster by combining the wavelength tuning and the spectral focusing method. Figure 1(c) shows such an SRS spectrum of benzonitrile recorded by tuning the wavelength of the Stokes beam with step sizes of 100 cm⁻¹, from 600 to 3500 cm⁻¹.

The core of our setup [Fig. 1(d)] is a commercial laser system consisting of two synchronously pumped OPOs (CRONUS-2P, Light Conversion). The laser has three outputs: one at a fixed wavelength of 1030 nm, and two with wavelengths independently tunable between 670 and 960 and 960 and 1300 nm. The pulse durations are less than 160 fs for all outputs, and the repetition rate is 77.1 MHz. As is shown in Fig. 1(d), the SRS pump beam is obtained by frequency doubling of the longer wavelength tunable beam with a periodically poled lithium niobate crystal (PPLN, HC Photonics). The obtained pump beam was tunable between 625 and 650 nm. For epr-SRS experiments with molecules absorbing between 520 and 580 nm, we set the pump beam wavelength to 640 nm. The pump beam was sent over a delay stage before combination with the Stokes beam using a dichroic mirror. A water immersion objective (63×, NA 1.2, Leica Microsystems) served to focus the beams, and an oil immersion condenser (NA 1.4, Leica Microsystems) collected the light in the forward direction.

SRS signals were recorded from the in-phase component of the demodulated pump beam (Raman loss, SRL) using lock-in amplification (UHFLI, Zurich Instruments) with the 1030 nm output of the laser serving as a reference. The short wavelength laser output was used as tunable Stokes beam modulated at 19.27 MHz with a broadband EOM. The broadband EOM is based upon two birefringent thermally compensated rubidium titanyl phosphate (RTP) crystals in a double pass configuration and a broadband Faraday isolator, which facilitates wavelength tuning. The isolator, in conjunction with significant and uneven acoustic heating of the RTP crystals at the 19.27 MHz pulse picking rate, limited the overall pulse-picker extinction ratio to less than 1:100. Broadband optical operation of the EOM over the whole laser output wavelength tuning range was achieved by adjusting the voltage applied to the crystals together with the changing wavelength. Unlike resonant EOMs, the EOM being used can pick single or multiple pulses from the 77.1 MHz pulse train at any rate below half the source rate.

The described laser system delivers nearly transform limited fs pulses. Due to the large spectral bandwidth of the pulses, their direct usage for SRS would result in a poor spectral resolution. An approach to circumvent this problem is spectral focusing, i.e., equal chirping of both excitation pulses.^15,16 The frequency difference between the chirped pump and Stokes beams can be tuned by changing the temporal delay between them. To match the typical width of vibrational lines in condensed phase, we aimed at a spectral resolution of less than 30 cm⁻¹ for the SRS spectra. Pulses were chirped by using slabs of highly dispersive glass (H-ZF62, Gestione SILO).¹⁶ SRS spectra of the sample molecules are then obtained in two different ways. To rapidly scan spectral regions of approximately 200 cm⁻¹, we simply changed the delay between pump and Stokes pulses. The spectral axis in this mode is also plotted as a temporal delay with the respective central frequency [Figs. 2(a) and 2(d) and Fig. 4(e)]. For scanning larger spectral ranges [Fig. 1(c)], we use a combination of spectral focusing and wavelength tuning, consisting in changes of the time delay (⁠ $Δ t$⁠) and the central wavelength of the Stokes (ω_S). To obtain regular Raman spectra from the spectral focusing data, the varying strength of the excitation field changes during the temporal scan has to be accounted for. For this purpose, we record the sum frequency (SF) between a pump and a Stokes photon generated in a thin layer GaSe crystal as a pump loss in the SRL channel. The SF signal is used for both normalizing the intensity of SRS spectra and finding the time zero at every central wavelength point as shown in Fig. 2(a). Furthermore, a spontaneous Raman calibration sample is used to find the conversion factor between temporal and spectral resolution. The normalized spectra recorded at different central frequencies are merged together using a moving average with a step size of 8 cm⁻¹, Fig. 2(b). Combining this method with the fast tuning mode of the laser (⁠ $≃$ 15 s per wavelength) allows us to record a spectrum over a range of 1000 cm⁻¹ in 160 s.

Figure 2(c) shows a comparison between the experimental and theoretical results from Eq. (1) for the 1650 cm⁻¹ band in the SRS spectra of six different rhodamine based dyes absorbing in the preresonant region and the 1600 cm⁻¹ peak of benzonitrile, respectively. The two bands shown in Fig. 2(d) correspond to the totally symmetric stretching of the xanthene ring in the dyes and of the benzene ring in benzonitrile.^17,18 To give better insight into the enhancement factors, Fig. 2(d) depicts the SRS spectra for 1 mM solutions of the dyes compared to pure benzonitrile (this corresponds to 9.7 M). As expected, the dyes with electronic absorption closest to the pump wavelength exhibit the highest enhancement. Their SRS signal strength resembles that of the band of the 10⁴ times higher concentrated benzonitrile. Figure 2(e) shows that the preresonance enhancement of the signals increases the SRS sensitivities to 1 μM. It is, however, important to note that these measurements are accompanied by mainly two types of background. The first is mixed two photon absorption (TPA) that occurs when one photon from the Stokes and one from the pump beam are absorbed simultaneously by the sample molecules. This results in a pump loss signal that is detected as vibrational frequency-independent background.¹⁰ The second background source is due to cross phase modulation (XPM) caused by the optical Kerr effect and has previously been reported in other SRS experiments.^19,20 In this case, the pump loss signal arises due to transient scattering of the pump beam caused by a Stokes beam induced change in refractive index. Both of these backgrounds are much broader than the vibrational bands and can be removed by background subtraction with a Gaussian fit to the background to obtain the pure SRS signal that scales linearly with concentration [Fig. 2(e-I)].

We evaluated the imaging performance of our experimental setup by microscopy studies of polymer test samples and of fixed biological cells. First, a sample of three different types of 2 μm polystyrene (PS) beads was measured at different vibrational resonances with the aim of distinguishing them using epr-SRS. The beads were commercial PS beads labeled with an unknown dye with an electronic absorption at 580 nm (FluoroSpheres F-8801, ThermoFisher Scientific), commercial PS beads (Polysciences) dyed by encapsulating SolventRed197, a weakly fluorescing dye with an absorption maximum at 550 nm, following the procedure reported by Behnke et al.,²¹ and undyed sulfate-modified polystyrene microspheres (Polysciences). SRS and spontaneous Raman spectra were recorded for each type of beads and are depicted in Fig. 3(a). The SRS images of the beads in water were recorded with excitation powers of 3.2 mW at the pump wavelength of 640 nm and of 12.8 mW at varying Stokes wavelengths, Fig. 3(b). The band at 1000 cm⁻¹ originates from the PS matrix present in all beads. At 1500 cm⁻¹, only the FluoroSpheres are in resonance, whereas at 1580 cm⁻¹, SolventRed197 has the strongest resonance that, however, appears low in intensity due to the lower preresonance enhancement and possibly also a lower dye concentration than in the FluoroSpheres. Due to the onset of the strong resonance at 1600 cm⁻¹, the latter are still visible at 1580 cm⁻¹ but with less intensity than when excited at 1500 cm⁻¹. The image taken off resonance at a Raman shift of 1760 cm⁻¹ was used to subtract the SRS background from the other images. By comparison, the fluorescence signal recorded simultaneously with the SRS signal and allows the detection of the FluoroSpheres, but not of the undyed nor the weakly fluorescent SolventRed197 beads.

Finally, we used the epr-SRS setup to image labeled biological cells. To this end, we stained the nucleus of fixed HeLa cells with AZ568 functionalized with dibenzocyclooctyne (AZDye 568 DBCO, Fluoroprobes). The dye has an absorption maximum at 578 nm and is suited for bioorthogonal labeling through strain-promoted azide–alkyne cycloaddition [Fig. 4(a)].²² The SRS and spontaneous Raman spectrum of the compound in Fig. 4(b) indicate a strong resonance at 1660 cm⁻¹. We recorded epr-SRS images at this resonance and off resonance at 1710 cm⁻¹ simultaneously with the fluorescence images. While the fluorescence images remain unaffected by the wavelength shift, the vibrational resonance completely disappears. Since electronic preresonance enhancement does not require fluorescent sample molecules,²³ extending epr-SRS to visible excitation wavelengths is also interesting for the investigation of naturally occurring compounds in unlabeled biological samples. As an example, we recorded spontaneous Raman and epr-SRS spectra of (non-fluorescent) β-carotene [Figs. 4(c) and 4(d)]. Its main vibrational resonances are located at 1160 and 1520 cm⁻¹. While the absorption maximum of β-carotene is more than 5000 cm⁻¹ blue-shifted from the pump beam used, epr-SRS measurements of solutions of β-carotene in DMSO-d₆ with concentrations as low as 18 μM clearly show two vibrational resonances expected in the probed vibrational region, Fig. 4(e).

In conclusion, we have demonstrated an experimental scheme for recording spectra over the whole range of molecular vibrational resonances using epr-SRS with visible excitation. With this approach, sensitivities comparable to fluorescence imaging are realized. While moving the excitation to shorter wavelengths might lead to a higher risk of photodamage,²⁴ the sensitivity enhancement is high enough that excitation intensities can be reduced to an acceptable level.²⁵ In addition, stretching the pulses with the employed spectral focusing gives another handle on reducing potential phototoxicity. We, thus, expect that epr-SRS microscopy with visible excitation will, in the future, find numerous applications for the investigation of labeled and unlabeled biological samples.