Combining the best of two worlds: Stimulated Raman excited fluorescence

The exquisite chemical specificity of Raman spectroscopy provides rich structure and dynamics information.¹ However, Raman scattering is well known to be extremely weak with a cross section of 10⁻³⁰ cm⁻² for typical chemical bonds.² Therefore, how to improve the sensitivity of Raman spectroscopy is one of the central themes in Raman research from the very beginning. In the community of far-field Raman spectroscopy, coupling Raman scattering with electronic resonance has long been harnessed as resonance Raman spectroscopy to enhance the Raman cross section, and stimulated scattering (a process analogous to stimulated emission) is widely used in coherent Raman spectroscopy, such as stimulated Raman scattering (SRS), to boost the Raman transition rate.^3–11 The synergistic combination of these two effects in the form of electronic (pre) resonance SRS under microscopy setting can enhance the effective Raman cross sections of electronically coupled vibrational modes by 10¹³ folds, which results in the overall Raman cross sections (∼10⁻¹⁷ cm⁻²) to be within one order of magnitude of the absorption cross section (∼10⁻¹⁶ cm⁻²) of a single light-absorbing molecule.^12,13 However, the SRS signal manifests as a small gain or loss on top of a strong laser field because of its heterodyne detection nature.³ As a result, the laser shot noise ultimately limits the detection sensitivity of electronic (pre-) resonance SRS to about hundreds of dye molecules.¹² To further improve the sensitivity, researchers have considered to couple Raman transitions to other optical observables that can potentially avoid the laser shot noise. In fact, the initial ideas and the relevant pioneering work along this line can be traced back four decades to the 1980s.

Laser induced fluorescence has been long known to the fields of physical chemistry and analytical chemistry as an extremely sensitive optical technique.¹⁴ Back to 1975, the pioneering work of Laubereau et al. employed fluorescence detection with time-resolved infrared spectroscopy to study vibrational relaxation dynamics.^15–17 In their method, a strong picosecond IR pulse was applied to pump the fluorophore (Coumarin 6) to a well-defined vibrational mode, followed by another picosecond pulse whose photon energy was merely enough to probe the energy gap between the vibrational excited state and the first electronic excited state [Fig. 1(a)]. The delay time between the two pulses was well controlled so that the detected fluorescence intensity represents the momentary degree of vibrational excitation. The fluorescence coupling to the vibrational excited state elegantly bypasses the giant laser noise and environmental interferences encountered in absorption measurements, which dramatically improved the sensitivity. Along this line, in 2019, Whaley-Mayda et al. have pushed the sensitivity of such fluorescence-encoded infrared (FEIR) spectroscopy to the impressive 10 nM–100 nM region with the help of high repetition rate laser and modern confocal microscopy.¹⁸ This level of sensitivity opens up the possibility of performing fluctuation correlation vibrational spectroscopy or even single-molecule measurements with further improvement.

Realizing that this detection scheme may be generalized to other absorption-based vibrational spectroscopy, Wright proposed the SRS counterpart [Fig. 1(b)] of this technique in early 1980s and summarized all the related techniques as the double resonance excitation.¹⁹ Later in 1983, the Wright group performed the first experimental attempt of double resonance fluorescence mediated by SRS [i.e., the stimulated Raman excited fluorescence (SREF) process named by us later]. However, this early attempt was hindered by the overwhelming two-photon fluorescence background from the perylene dye, and no signal was detected as the proposed SRS mediated double resonance fluorescence.²⁰ Long after this failed experiment, spectroscopists were still optimistic about the potential of this approach and further proposed in 2011 (likely inspired by the success of room-temperature single-molecule fluorescence spectroscopy) that this SRS mediated double resonance fluorescence method, when performed under modern optical microscope, might offer the ultimate single-molecule sensitivity of vibrational spectroscopy at optical far-field.²¹ However, no further experimental studies had been reported, suggesting the difficulty of implementing such a proposal.

After nearly four decades of twists and turns since its initial proposal, we reported in 2019 the first successful realization of double-resonance fluorescence mediated by stimulated Raman scattering [Figs. 2(a) and 2(b)], which we called stimulated Raman excited fluorescence (SREF).²² The initial demonstration was performed on a near infrared absorbing dye rhodamine 800 (Rh800). Figure 2(b) illustrates our microscope setup. Briefly, temporally and spatially overlapped pump and Stokes pulse trains are focused by a high numerical aperture microscope objective to perform SRS excitation of vibrational modes. In our demonstration, the pump pulse can also play the role of probe pulse for excitation to the electronic state. Fluorescence emission is then detected confocally. Backgrounds (such as laser beams or coherent anti-Stokes Raman scattering signal) are completely blocked by proper filters.

We then obtained its SREF excitation spectrum by scanning the pump wavelength. Remarkably, a pronounced Raman-like peak emerged at 2236 cm⁻¹ when detecting fluorescence emission [see the solid line in Fig. 2(c)]. Both its position and linewidth are highly consistent with the corresponding Raman peak of the nitrile stretching, proving its vibrational origin. Thus, the excitation spectrum of SREF loyally maps out the line shape of the corresponding Raman vibrational mode in the ground electronic state of Rh800 [Figs. 2(c)–2(e)]. As a negative control, pump or Stokes alone only exhibits broad excitation spectra.

In agreement with the expected high sensitivity from the 2011 proposal, single-molecule vibrational imaging was achieved without the need of plasmonic enhancement [Figs. 2(f) and 2(g)]. Linear concentration dependence is confirmed with a superb sensitivity readily down to 10 nM. We then dispersed Rh800 molecules in a polymer film and acquired SREF spectral images of single molecules by sweeping the pump wavelength across the vibrational resonance of the nitrile stretching. Consistently more pronounced signal is observed at on-resonant 2236 cm⁻¹. The peak position and narrow linewidth all resemble the solution SREF spectrum. Quantitatively, about 4 photons/ms were detected from the single-molecule image. This agrees with the solution measurement and our theoretical estimation. Further rigorous evidence was provided for single-molecule imaging of isotopologues of Rh800 dyes [Figs. 2(d)–2(g)]. Hence, such a hybrid spectroscopy combines the best of fluorescence and Raman, merging these two fundamental processes into one new three-photon process.

As illustrated in Fig. 2(c), the excitation of SREF is often accompanied by other competing linear and nonlinear excited fluorescence processes, which results in complex fluorescence backgrounds. Whether the desired SREF signal can be detected with a good signal-to-noise ratio is determined by the relative level between the pure SREF signal and the background. In a sense, this background had hindered the experimental progress in this line for nearly 40 years. As summarized below, there are three underlying physical principles that collectively govern the success of SREF spectroscopy.

First, the electronic pre-resonance excitation should be adopted for the SRS process so that the SREF signal will not be overwhelmed by the competing two-photon fluorescence. As shown in Fig. 1(b), the frequency of the probe beam (ω_p) is always set to approach the electronic resonance of the fluorophore. Hence, the two-photon absorption processes involving the probe beam are inevitably under the electronic pre-resonance condition [Fig. 3(a)] and therefore are strongly enhanced.²⁰ This is likely the reason behind the overwhelming two-photon excited fluorescence observed in the first unsuccessful attempt performed in 1983 on perylene dye.²⁰ In order for SREF to compete favorably with the two-photon fluorescence, our solution is to set the wavelengths of the pump beam (ω_p) and Stokes beam (ω_s) close to the probe beam or, for simplicity, use the pump beam as the probe beam at the same time [Fig. 3(b)] so that the SRS process can also be significantly enhanced (by up to five orders of magnitude) under the electronic pre-resonance condition. Therefore, near IR dyes were chosen in our first demonstration of SREF,²² which shares the same spirit as the design and execution of electronic (pre-) resonance SRS imaging.¹² With this configuration, one should anticipate the SREF signal at the comparable level as the two-photon fluorescence background²² (details in the  Appendix). In contrast, the 1983 experiment and the 2011 proposal did not consider the electronic pre-resonance condition in the SRS pumping step.

Second, as approaching the electronic resonance also results in the anti-Stokes one-photon absorption [Fig. 3(c)], one needs to make sure that the SREF signal will not be overwhelmed by the anti-Stokes fluorescence background, both in spectroscopy and in microscopy. To this end, we found that the resonance depth should be carefully controlled in spectroscopy, and rigorous confocal detection should be applied in microscopy instrumentation. For example, the thermal population of the C=C skeletal vibrational mode (typically ∼1600 cm⁻¹) of a fluorophore is about $eℏωkT≈4×10−4$ at room temperature. Suppose an efficient upconversion of SRS transition by the probe beam, one needs to achieve an effective SRS excitation cross section not less than 10⁻⁴ of the single-photon absorption cross section in order for the SREF signal not to be overwhelmed by the anti-Stokes fluorescence background. Experimentally, such an effective SRS excitation cross section has been demonstrated by electronic pre-resonance SRS on a large amount of visible and near-IR chromophores with the electronic detuning even larger than 1600 cm⁻¹.^6,12 However, the anti-Stokes fluorescence background increases much faster than that of the SRS pumping rate as the electronic resonance approaches.²³ It is because the anti-Stokes fluorescence follows the Boltzmann statistics,^24,25 which increases exponentially as the pump detuning decreases, whereas the SRS transition is modeled by the Albrecht A-term,^9,26 which follows the negative fourth power of the pump detuning (rigorous analysis can be found in the Supporting Information of Ref. 23). As a result, the signal-to-background ratio of SREF detection is rather sensitive to the electronic resonance depth.

Under the two-beam configuration [Fig. 3(b), the pump beam is also used as the probe beam], we demonstrated that successful SREF excitations of the C=C skeletal mode or higher energy Raman modes can be achieved only when the total excitation energy [(ω_p − ω_s) + ω_p] is set within less than the 20-nm range above the 0–0 transition of the fluorophore [Fig. 3(d)].²³ As a rule-of-thumb criterion, it is generally safe to set the total excitation energy between the 0–0 transition line and the absorption peak of the fluorophore.^22,23 The current mainstream picosecond-pulsed OPO system (APE picoEmerald S) used for SRS microscopy can provide a fixed Stokes beam at ∼1031 nm and a tunable pump beam that covers the 760 nm–960 nm range. Under the above criterion, fluorophores with conjugated nitrile moieties and their absorption peaks close to ∼695 nm (such as Rh800) can have the nitrile mode (∼2235 cm⁻¹) successfully excited when the pump beam is tuning to ∼838 nm, and the fluorophore with absorption peak close to ∼760 nm (such as ATTO740) can have its C=C skeletal mode (∼1640 cm⁻¹) successfully excited when the pump beam is tuning to ∼882 nm.²² In addition to spectroscopy consideration, proper microscopy instrumentation is also needed to minimize this anti-Stokes fluorescence background. Because the three-photon-excited SREF signal can only occur inside the tight laser focal volume, whereas one-photon-excited anti-Stokes fluorescence can occur through the entire laser excitation path, rigorous confocal detection must be applied especially for thick-sample measurements. Otherwise, the SREF signal would be completely overwhelmed by the out-of-focus anti-Stokes fluorescence background, which we had encountered in our own experiments. In our setup, a pinhole size two times of the Airy unit (corresponding to the pump wavelength) has been proved to be sufficient to suppress the anti-stokes fluorescence background when the SREF excitation is set in the proper electronic pre-resonance range [see Fig. 3(d)].

The third important factor is the coupling efficiency between the stimulated Raman transition and the fluorescence emission, which can be optimized by adjusting the excitation pulse width. For room-temperature fluorescence spectroscopy excited by picosecond pulses, because the fluorescence readout only represents the excited state population at the end of the pulsed excitation and the vibronic coherence decays within the sub-picosecond time scale,²⁷ the rate-equation limit is naturally fulfilled. Therefore, it is feasible to model the SREF process by rate equations of a three-level system.^19,22,28 Under the steady-state approximation, the coupling efficiency between the stimulated Raman transition and the fluorescence emission can be simply represented as²² 

where $τv$ is the lifetime of the vibrational state, τ_pulse > $τv$ is the pulse duration of the picosecond pulsed lasers, and η is fluorescence quantum yield of the fluorophore. To increase the coupling efficiency without losing the spectral resolution needed for vibrational spectroscopy, the shortest pulse for SREF excitation is around 1 ps for the Fourier transform-limited pulse laser system (bandwidth ∼10 cm⁻¹). Clearly, because the fluorescence lifetime is orders of magnitude longer than the time constant of the three-level system, a longer excitation pulse will not contribute to a larger SREF signal but only will result in stronger photobleaching after the steady-state is achieved. Even under this optimal condition, a fairly large portion of stimulated Raman transition is unproductive (i.e., not yielding the fluorescence excitation), considering the sub-picosecond-level lifetime for a typical Raman mode (i.e., lifetime-broadened with ∼10 cm⁻¹ bandwidth).¹⁷ 

When the above three criteria are fulfilled, successful SREF excitation of a given fluorophore can be routinely achieved. Indeed, soon after our first demonstration on near IR fluorophore of Rh800 and its isotopologues, we have extended SREF to a broad range of commercial fluorophores in the visible range.²³ The results not only establish the generality of SREF spectroscopy for a wide range of molecules but also reveal a tight window of proper electronic pre-resonance for the stimulated Raman pumping process. Our theoretical modeling and further experiments on newly synthesized dyes also support the obtained insights.

As a newly observed spectroscopy, the instruments involved in SREF are far from optimized. Currently, the main technical challenge is the lack of proper laser sources. As a double-resonance technique, at least two laser tuning freedoms are required to fulfill both the vibrational resonance and the proper electronic pre-resonance requirements in SREF [Figs. 3(b) and 3(d)]. The proof-of-concept demonstrations were based on a near-IR OPO system (with synchronized ω_p and ω_s) optimized for SRS applications. This commercial system only enables successful SREF excitation of limited Raman modes of some red and near-IR fluorophores whose spectroscopic properties fulfill the total excitation energy requirement [Fig. 3(d)].^22,23 To enable efficient SREF excitation of more fluorophores widely used in biology and chemistry,^29,30 a two-beam or even three-beam free-tunable picosecond-pulsed laser source should be ideal, which is commercially unavailable unfortunately. On the other hand, SREF excitation only requires sub-nJ picosecond laser pulses,²³ which is relatively easy to achieve with the well-developed fiber laser techniques³¹ and nonlinear frequency conversion.^32,33 For example, super-continuum generation with the recompressible dispersion profile could be a suitable and flexible source after simple pulse shaping.³⁴ Adopting a more suitable and flexible laser source will enable SREF investigation of essentially any fluorophores across the entire spectrum.

In addition to the laser source, the detector side can also be optimized. As a three-photon process, SREF is relatively weak when compared with direct (one-photon) resonance fluorescence. The typical photon count rate of the pure SREF signal is at the level of ∼1% of that of the saturated resonance fluorescence.²² To achieve high detection sensitivity, the Si avalanche photodiode (Si-APD) based single-photon counter rather than the photomultiplier tube for conventional confocal fluorescence microscopy is employed to achieve high-quantum-yield detection. However, the APD has a relatively low maximum count rate (∼10 000 counts/ms), which strongly affects the signal linearity and limits the imaging speed of SREF microscopy. More advanced detectors, such as superconducting nanowire single-photon detector (SNSPD),³⁵ can have much better performance for all the mentioned parameters but are expensive and may require liquid helium cooling during operation.

The successful observation of (single-molecule) SREF of a wide range of fluorophores opens the door for a variety of exciting opportunities for molecular spectroscopy, biophysics, and biomedical imaging. Until now, all the discussions are based on steady-state picosecond-pulse excitation. The transient excitation method based on successive femtosecond pulses, as those already demonstrated on the IR counterpart,^18,36 could be another possible way to realize or even improve SREF. Its well-established Fourier-spectroscopy based detection manner should naturally enable background-free hyperspectral vibrational spectroscopy and imaging at the single molecule level. This could allow time-dependent vibrational spectroscopy.

In addition, the narrow line shape of the SREF excitation spectrum can be harnessed to break the “spectral crowding” (i.e., less than five fluorescent colors can be simultaneously resolved) of the conventional fluorescence microscopy,³⁷ leading to a powerful technique toward super-multiplexing SREF imaging. The superior sensitivity (by about 100 times) of SREF than electronic pre-resonance SRS imaging could be particularly advantageous for detecting low-abundance biological targets and for high-speed high-throughput measurement. For this super-multiplexing application of SREF, a new probe palette of bright fluorophores conjugated with suitable vibrational groups (such as nitrile bonds and alkyne bonds) needs to be developed to match with the laser excitation source.

Moreover, when coupled with the vibrational solvatochromism and Stark effect,^38,39 SREF can be used for sensing applications with high sensitivity, as recently demonstrated in spatially-resolved mapping of water states inside single mammalian cells and electric field imaging at the interface of aqueous microdroplets.^40,41 Environment-sensitive vibrational probes, which are often detected by infrared spectroscopy such as FTIR,^42,43 are playing an important role here. Compared to traditional infrared spectroscopy, SREF would offer much higher detection sensitivity and spatial resolution. This line of application has potential to open up new discoveries of chemical science and molecular and cellular biophysics.

The authors thank L. Shi, L. Wei, Y. Shi, N. Qian, Y. Miao, X. Zhu, X. S. Xie, and L. E. Brus for helpful discussions. W.M. acknowledges support of R01 Grant (Award No. GM128214) and R01 Grant (Award No. GM132860) from the National Institute of Health.

Data sharing is not applicable to this article as no new data were created or analyzed in this study.

All small-signal two-photon processes can be uniformly described by the famous Kramers–Heisenberg formula in second-order perturbation theory.^44,45 Define the second-order tensor α for the two-photon transition from an initial state $i$ to a final state $f$ induced by two light modes,

Here, +ω₁ and +ω₂ correspond to absorption processes and −ω₁ and −ω₂ to emissions. k and l represent the three-dimension cartesian coordinates; ε_i, ε_m, and ε_f are the corresponding energy eigenvalues of the vibronic states |i⟩, |m⟩, and |f⟩ of the molecule, respectively. $μfmk$ is the k component of the transition dipole moment μ_fm. Γ_m is the homogeneous linewidth of the intermediate state.

The stimulated Raman scattering originates from α(ω₁, −ω₂). Let us just consider the simplest case in which the transition is induced by two single-mode light fields (with photon numbers N₁ and N₂). Then, the stimulated Raman transition rate is

Here, $ϵ0$ is the vacuum permittivity. V is the field normalization volume. ρ is the state density of the state quasi-continuum at the final state |f⟩. e₁ and e₂ are the corresponding unit polarization vector of the field. Similarly, α(ω₁, ω₂) corresponds to the two-photon absorption process. The corresponding transition rate is

As indicated by the uniform description, the two-photon absorption rate and the stimulated Raman transition rate excited by the same fields on the same molecules should be within the same order of magnitude when both processes are set to resonance [i.e., $ℏ(ω1−ω2)$ and $ℏ(ω1+ω2)$ match the energy gaps between two specific vibronic states and the ground state, respectively]. The difference just comes from the Franck–Condon factors $∑vf|vm$ (⁠$v$ for the vibrational states of the nuclei) for each transition dipole moment μ_fm under the Born–Oppenheimer approximation and should be small because of the sum over all states (i.e., an average over a full spectrum of Franck–Condon factors is achieved).