1. Introduction

Light-sheet (LSI) and selective plane illumination (SPI) imaging have conferred significant speed, phototoxicity, and photobleaching gains in 3D fluorescence microscopy [1–9]. These gains emanate primarily by reconfiguring the illumination at a direction that is orthogonal to the detection axis, such that only specimen sections coinciding with the imaging plane are being illuminated [10,11]. In this way, the illuminated sections are projected more efficiently onto multi-pixel detectors, while the unnecessary specimen illumination is greatly reduced. Since the emergence of the first LSI and SPI schemes, several modifications have been reported that improve spatiotemporal resolution [2,5,7,9], simplify hardware configurations [12–16], and integrate additional imaging modalities [17]. Concomitantly, considerable increases in the imaging field-of-view (FOV) of LSI have been also reported by using either optical lattices [2] or quasi-lattices [3]. These modalities harness the self-healing properties of Bessel [18] and transversely accelerating Airy beams [19], thus, generating imaging FOVs that greatly exceed the Rayleigh length of Gaussian beams.

More recently, chemical LSI and SPIM imaging schemes relying on spontaneous and coherent Raman scattering have also been reported [20–26]. Like fluorescence, these developments accelerate image acquisition and reduce specimen irradiance in comparison to point-detection confocal microscopy. Such gains are particularly pertinent to Raman imaging given the ultralow conversion efficiency of pump to Raman-shifted photons (∼10⁻⁸) that require long integration times and > 10 mW/µm² irradiance levels [27–31]. Conversely, LSI and SPI enable speed gains, which are approximately equal to the number of detection pixels where the illumination sheet is projected [11] (Fig. 1(a)). These gains have been previously quantified to be more than fivefold in comparison to confocal microscopy [20]. Further, hyperspectral light-sheet Raman imaging has also been demonstrated by adopting tunable laser excitation [21], a spectrometer with an entrance slit parallel to the light-sheet [23], Fourier transform spectral imaging [20], and tunable filters [25].

 

Fig. 1. (a) Increasing imaging throughput rates by evolving point-scanning (e.g., confocal) (i) to light-sheet illumination using Gaussian (ii) and Airy (iii) beams; light red represents the imaging sensor. (b) The light-sheet setup employed in this work highlighting the orthogonally arranged illumination (IO) and detection (DO) objectives, the illumination source (tunable CW Ti: sapphire ring laser), and the image projection onto a sCMOS sensor following band-pass filtering (BPS). (c) The Raman spectrum of a PS particle, with the breathing mode of the aromatic carbon ring (∼1000 cm⁻¹) highlighted by the dotted rectangle. (d) The same band as in (c) deciphered by conventional spectroscopy (left y-axis, solid lines) and Airy light-sheet imaging (right y-axis, red points); each shade of blue corresponds to specific bands that are imaged with an Airy beam by keeping the detection frequency constant while tuning the excitation wavelength at the noted values.

Download Full Size | PDF

Despite these key advances, existing Raman LSI and SPIM systems rely on light-sheets that are generated by relaying Gaussian beams onto cylindrical lenses or scanners [20–25]. In this context, the Rayleigh length of the Gaussian beam restricts both the size and uniformity of the resulting imaging FOV [20]. To address this shortcoming, broader Gaussian beams could be adopted, at, however, lower illumination efficiency and reduced optical sectioning capabilities [3]. Multiple exposure angles [32] or tunable focusing [33,34] have also been applied in fluorescence imaging to circumvent the Rayleigh length limitation; these approaches, however, increase the irradiance times and yield poor heat dissipation that can be detrimental to sample physiology [35]. Similar to fluorescent volumetric imaging, non-diffracting Bessel and Airy beams can also alleviate the reduced size and uniformity of Gaussian-based light-sheets and enable optical sectioning at large FOVs [2,3,8]; these approaches, however, have not yet been demonstrated in Raman light-sheet microscopy.

Here, we show that a light-sheet created by digitally scanning an Airy beam enables spontaneous Raman imaging with FOVs that are both uniform and up to tenfold larger than what Gaussian beams can confer. Critically, this FOV increase translates to an equivalent increase in imaging throughput rates due to the more efficient projection of the illumination onto a multi-pixel detector, as displayed in Fig. 1(a). Among several key applications, these gains in throughput rates pertain particularly to the imaging of larger biological objects (e.g., embryos and biofilms) or to single-cell biology exercises that demand multiple observations towards statistical significance. Bessel beams also enable FOV increases in comparison to Gaussian beams; however, the diffraction-free path of Bessel beams is considerably shorter (> 4x) than that of the Airy beam [3]. Using a standard light-sheet setup integrated with an inverted microscope (Fig. 1(b)), we demonstrate that the FOV gains of the accelerating Airy beam (Fig. 1(a)) enable rapid 3D Raman imaging of polystyrene (PS) particles. In this context, a 2D image required just 300 msec integration times at 70% illumination efficiency and 1.4 mW/µm² irradiance using a scientific CMOS camera. We also demonstrate hyperspectral Raman imaging with 25 cm⁻¹ resolution by adopting a tunable, ultra-narrow, linewidth laser (Fig. 1(c), (d)).

2. Airy light-sheet Raman imaging

To generate the Airy beam, we illuminated a spatial light modulator (MSP 1920-400-800-HSP8, Meadowlark Optics) with a 6 mm diameter continuous wave Gaussian laser beam (Fig. 1(b)). For illumination, we employed an ultra-narrow linewidth (2 MHz bandwidth, < 10⁻³ cm⁻¹ at λ = 750 nm) Titanium: Sapphire ring laser (Matisse CR, Spectra Physics) pumped with a 532 nm DPSS laser (Millenia ev, Spectra Physics). This source emitted 1.5 W average power that was continuously tunable in the 700 nm – 800 nm range. Given the linear nature of Raman scattering, we ensured the stability of the laser power with a motorized variable attenuator in a closed-loop format (VA-BB-2-CONEX, Newport). With this element in the beam path, it was possible to maintain less than 1% power fluctuations for several hours.

The SLM was conjugated at the back aperture of the illumination objective and exhibited the cubic-polynomial phase profile [19]: ΔΦ(x,y) = $\frac{\mathrm{\alpha }}{3}({{\textrm{x}^3} + {\textrm{y}^3}} )$, where α is the characteristic Airy modulation depth parameter and x/y denote the cartesian coordinates on the SLM plane. To generate the light-sheet, we scanned the Airy beam orthogonally to its propagation direction (i.e., y-axis in Fig. 1(b)) using a 2D MEMS mirror (2.8 mm diameter, Mirrorcle Technologies) with a 200 Hz frequency at a full scanning range. The optical setup used to generate the Airy light-sheet and relay it on the sample is detailed in Fig. 2 and in [17]. Samples were positioned on a customized motorized stage that enabled the Airy beam to access the sample orthogonally to the detection axis (z-axis in Fig. 1(b)). Integrated with a standard inverted microscope, the stage scanned the sample in three directions using a piezoelectric (IPZ-3150, Applied Scientific Instrumentation) and two linear (LS-50, Applied Scientific Instrumentation) stages. The piezo and linear stages, and the MEMS mirror were automated through a programmable controller (TG-1000-8, Applied Scientific Instrumentation) and Micro-Manager 1.4 [36].

 

Fig. 2. The detailed optical set-up for Airy light-sheet imaging; the table lists all the components, including the lenses’ focal lengths and positions. The lightly blue shaded areas correspond to the portion of the optical path that was modeled, as further detailed in Section 3.1.

Download Full Size | PDF

In all experiments, a long working distance objective (20x/0.42, Mitutoyo) focused the Airy light-sheet on the sample. The Raman signal was projected through a 20x/0.4 or a 40x/1.1 objective lens (Leica) and a series of custom bandpass filters centered at 821 nm (total OD ≥ 30 at λ ≤ 790 nm, 1.5 nm bandwidth, Alluxa) on to a sCMOS camera (ORCA-Flash 4.0, Hamamatsu). The filters were positioned directly in front of the sCMOS camera at moderate angles (∼ 0.5°) between them, enabling approximately 25 cm⁻¹ spectral resolution. This level of resolution was adequate to perform hyperspectral imaging by scanning the laser wavelength with a Δλ = 0.2 nm step while maintaining the filter response (and position) constant [21]. Lower resolution (∼ 200 cm⁻¹) Raman imaging was also possible by using a filter set (OD ≥ 30 at λ ≤ 790 nm, Alluxa) centered at 812 nm with a 14 nm bandwidth.

3. Results

3.1 Throughput rates

We quantified the imaging throughput rates by determining both computationally and experimentally the imaging field-of-view (FOV) and diffraction-free length of the Airy beam (Fig. 1(a)). Experimentally, we illuminated a polydimethilsiloxane (PDMS, Sylgard 184, Dow Corning) slab with a non-scanning Airy beam at λ = 733 nm and collected the corresponding Raman signal at 1,414 cm⁻¹. Here, we specifically selected a non-scanning beam to ensure parallel visualization of the beam’s acceleration. Subsequently, we quantified the beam diameter (approximated by e⁻¹) through the lateral intensity distribution (y-axis in Fig. 1(b)) at various propagation distances (x-axis in Fig. 1(b)). Computationally, we adopted a numerical model to quantify the propagation of the Airy beam that we previously detailed in [17]. Briefly, the model applies the Fast Fourier Transform Beam Propagation Method (FFT-BPM) to compute the electric field of a monochromatic wave at various propagation distances. For this, we used a Gaussian illumination of the SLM cubic phase as input and included all lenses of Fig. 2 (blue shaded area) using the thin lens approximation.

As expected [19] and previously demonstrated [3], varying the characteristic α parameter controlled the uniformity and size of the Raman image FOVs (Fig. 3(a)). Specifically, we both predicted and verified up to a 260 µm diffraction free for α = 35 mm⁻³ and a 6 mm diameter Gaussian beam prior to SLM modulation (Fig. 3(b)). This translates to 260 × 600 µm² FOVs at 20x magnification (or 260 × 320 µm² at 40x) by scanning the illumination beam (along the y-axis, Fig. 1(b)) in combination with the abovementioned sCMOS camera (6.5 µm pixel). In comparison, an unmodulated Gaussian beam of similar diameter yielded tenfold smaller diffraction-free paths and, as such, a tenfold smaller FOV and imaging throughput rates.

 

Fig. 3. (a) Panel representing the phase masks projected on the SLM, the resulting Airy beam cross-sectional intensity profiles (yz plane, Fig. 1(b)) at the focus, and the diffraction-free propagation distance (xy plane, Fig. 1(b)); both white bars represent a 20 µm length scale. (b) The computed and experimentally determined FOV dependence on α; inset table displays the resulting beam diameters (i.e., the diameter of the Airy main lobe for α ≠ 0). (c) The computed and experimentally determined peak irradiance dependence on α (normalized with the respect α = 0 mm⁻³), as well as the resulting Raman signal (data points and error bars represent the mean and s.d. of n = 3 locations).

Download Full Size | PDF

Further, varying the phase modulation parameter α altered the intensity cross-section of the Airy beam. In this context, higher α’s yielded broader main lobes of the Airy beam with its side-lobes positioned further away from the main lobe and each other (Fig. 3(a), (b)). This intensity reconfiguration occurs despite the identical total irradiance of all Airy and Gaussian beams, and enables substantially lower peak irradiance at increasing α’s. Indicatively, a 7-fold peak irradiance reduction was predicted and measured for α = 2.48 mm⁻³ with respect to the Gaussian beam (Fig. 3(c)). While such irradiance reductions are advantageous in Raman microscopy [37] and live cell imaging [3], they can also reduce the intensity of the Raman signal (Fig. 3(c)). In this context, we observed that the Raman signal decrease is slower than the peak irradiance for increasing α’s (indicatively, a 2-fold signal reduction occurs at α = 2.48 mm⁻³ despite the 7-fold peak irradiance reduction). This is because while all Airy and Gaussian beams are expected to excite the same number of molecules, the peak irradiance decrease for increasing α’s is partially recovered by collecting photons from larger volumes [3]. This recovery is facilitated by employing detection objectives of specific numerical apertures, as detailed in the following sub-section.

3.2 Contrast

While the imaging contrast is intrinsically high in fluorescence light-sheet microscopy, it is challenging to achieve high contrast in Raman imaging due to the low Raman scattering cross-sections. To quantify the contrast levels in Airy light-sheet Raman imaging, we adopted 1 µm diameter (PS) particles (Bangs Laboratories), a standardized sample in Raman imaging [23,37,38] (Fig. 4(a)), embedded in an agarose gel [17]. Further, we employed a scanning (∼12 µm scanning width at ∼5 kHz rates) Airy beam with α = 2.48 mm⁻³ and λ = 757.8 nm, and collected the (broadband) Raman signal at the PS breathing mode 1,000 cm⁻¹). It is worth noting that a higher α parameter would enable higher FOVs (Fig. 3(b)), albeit it would require higher average powers to maintain the same levels of peak irradiance (indicatively, an α = 35 mm⁻³ would require a 3-fold higher average illumination power than α = 2.48 mm⁻³). We then quantified contrast as C = [I_signal – I_background] / [I_signal + I_background] by selecting the appropriate regions of interest (ROIs) in the acquired Raman images.

 

Fig. 4. (a) 3D Raman image of a 1 µm diameter PS particle using the 40x/1.1 objective following processing for background correction and denoising with median (1 pixel) and Gauss (σ = 1) filtering; 3D image reconstruction was performed with VolumeViewer (ImageJ). (b) The dependence of particle imaging contrast on integration time (Δt) at three distinct integration density levels using the 20x/0.4 objective. (c) The same as in (b), albeit using the 40x/1.1 objective.

Download Full Size | PDF

We observed that Airy light-sheet microscopy yields high contrast levels at relatively short integration times. Specifically, the 20x/0.4 detection objective enabled 60% contrast levels [39] at just 300 msec integration times per 2D plane (Fig. 4(b)) and 1.4 mW/µm² irradiance (20 mW after the illumination objective). Importantly, the low NA and long depth-of-field (∼ 6 µm) of this optic enabled 74% overall efficiency (computed as the fraction of signal collected by the detection objective and used to form the final image, as originally described in [3]) in comparison to the 100% bestowed by the Gaussian beam. This value was found to compare well with the computed Airy beam intensity profile (70% efficiency). Comparable contrast levels were also obtained at 100 msec integration times and 1.4 mW/µm², or at 300 msec and 0.4 mW/µm² using the 40x/1.1 objective (Fig. 4(c)). For this objective, however, the overall efficiency decreased to 25% (22% using the computed Airy beam profile) in comparison to the Gaussian beam given its shorter depth-of-field (∼1 µm). Overall, these findings indicate that the enhanced throughput rates and efficiency, as well as the reduced peak irradiance of the Airy light-sheet do not limit the resulting Raman imaging contrast levels.

3.3 Resolution

Like fluorescence LSI, the lateral resolution (i.e., x-y plane in Fig. 1(a)) in Raman light-sheet microscopy depends on the NA of the detection objective. Concomitantly, the axial resolution (i.e., z-axis in Fig. 1(a)) depends both on the NA of the detection objective and the waist of the illuminating light sheet [1]. We confirmed this for the 20x/0.4 and 40x/1.1 objectives using 0.5 µm and 0.2 µm diameter PS particles (Bangs Laboratories), respectively. For both measurements, we employed 1.4 mW/µm² irradiance, α = 2.48 mm⁻³, 12 µm wide scanning ranges (along the y-axis in Fig. 1(b)), and averaged n = 5 distinct particle observations. Importantly, we did not detect the Airy side lobes in the y-axis (Fig. 1(b)) due to temporal scanning. Further, we also did not detect the Airy side lobes along the z-axis due to their lower intensity than the main lobe, as also previously reported specifically in two-photon microscopy [40]. As displayed in Fig. 5, the 20x/0.4 detection objective yielded lower planar (1.43 ± 0.15, mean ± s.d.) and axial (3.78 ± 0.61) resolution than the 40x/1.1 (0.49 ± 0.03 planar and 0.49 ± 0.05 axial). The axial resolution for the 20x/0.4 objective, however, could be in part recovered via deconvolution. In this context, and unlike previous one-photon fluorescence reports, the absence of detectable Airy side-lobes enabled the implementation of a standard deconvolution approach (AutoQuant X3, MetaMorph, 7.8, Molecular Devices) without implementing a computational [3] or experimental [17] point spread function.

 

Fig. 5. (a) Planar (x-axis, Fig. 1(b)) and (b) axial (z-axis, Fig. 1(b)) of the 20x/0.4 detection objective. Data points represent the mean experimental measurements of n = 5 particles with 0.5 µm diameter (error bars represent the s.d.), while the legend displays the planar resolution before (red) and after (blue) deconvolution (through the average full-width half maximum, FWHM, ± s.d.); inset depicts the same, albeit for the 40x/1.1 detection objective and 0.2 µm diameter particles.

Download Full Size | PDF

3.4 Hyperspectral imaging

Finally, we explored the utility of Airy light-sheet microscopy in performing hyperspectral imaging. Using the narrowband filter (λ_center = 821 nm), we obtained the Raman spectra of a uniform PDMS slab by scanning the ultranarrow linewidth laser at Δλ = 0.2 nm steps in the 728 nm – 744 nm region (Fig. 6) with approximately a 25 cm⁻¹ resolution. A PS particle (1 µm) yielded similar results by scanning the excitation frequency within the 754 nm – 759.2 nm region (Fig. 6, inset). In all measurements, the Airy beam was spatially scanned within a 12 µm range (along the y-axis in Fig. 1(b)) and its power stability was ensured using the variable attenuator. For both samples, the acquired hyperspectral imaging data were found to be in good agreement with the Raman spectra of the same samples using a conventional spectrometer (785 nm excitation, 50 mW average power, 250-2350 cm⁻¹ frequency range with 6 cm⁻¹ resolution, 10 sec integration times using the EZRaman-M system by EnWave Optronics equipped with a linear CCD array with 2048 pixels of 14 µm x 200 µm size, TE cooled to −15°C). These conventional Raman spectra were acquired independently (i.e., in a different setup) using the same samples.

 

Fig. 6. Comparison of the Airy light-sheet hyperspectral Raman imaging (blue points) with the corresponding Raman spectra (red lines) collected with a standard spectrometer for PDMS and PS (inset). Both PDMS and PS data sets were acquired with the 20x/0.4 objective, with the data points representing the average of n = 5 observations and the error bars denoting the respective standard deviation. The images (highlighted by the dark arrows) correspond to the respective PDMS Airy Raman images collected at the indicated Raman shifts.

Download Full Size | PDF

4. Conclusion

In summary, we demonstrated the application of an Airy beam in spontaneous Raman light-sheet microscopy. The self-healing properties of the Airy beam enabled Raman imaging with tenfold greater imaging FOVs and, thus, throughput rates than conventional light-sheet Raman imaging [20–25] at ∼1 µm quasi-isotropic 3D resolution and more than 70% imaging efficiency. Further, Airy light-sheet Raman imaging was effective in rapid, high-contrast, imaging at low irradiance levels, requiring just 300 msec per 2D image of PS microparticles at 1.4 mW/µm². These performance metrics greatly surpass the throughput rates of point-scanning or diffraction-limited light-sheets methods [20–25], as well as the efficiency of line-scan schemes [38]. Similarly, the irradiance levels required by Airy light-sheet Raman imaging are substantially lower than what is required for Coherent Anti-Stokes and Stimulated Raman imaging (∼W/µm²) that also necessitate the spatio-temporal overlap of two, typically pulsed, beams [37,41]. We also demonstrated hyperspectral Raman imaging by tuning the excitation frequency within a spectrally bound detection window.

Importantly, Airy light-sheet Raman imaging is compatible with multimodal imaging, including quantitative-phase imaging (QPI) and light-sheet fluorescence microscopy that we recently demonstrated [17]. One such representative example is displayed in Fig. 7(a) for a 1 µm PS particle illuminated by 561 nm (for fluorescence), 758 nm (for Raman), and white-light for QPI by spatial light interference microscopy (SLIM) [42]. As such, we anticipate the advantages of the presented method will find applications in investigations where QPI informs about organelle location and cellular dry-density, fluorescence elucidates the 3D dynamics of tagged-proteins, and Raman quantifies the metabolic dynamics and nutrient fate in conjunction with the use of Raman-tags (e.g., alkyl entities [43]), heavy water [44], or deuterium labeled glucose [45]). For the latter, one related Raman imaging exercise of live cells tagged with deuterated glucose is displayed in Fig. 7(b), demonstrating the possibility of tracking the subcellular fate of glucose.

 

Fig. 7. (a) Representative image of a 1 µm polystyrene particle embedded in PDMS imaged by fluorescence (left, 40x/1.1 objective) Raman (middle, 40x/1.1 objective), and QPI (right, 40x/0.6, PH2). Calibration bars correspond to counts (fluorescence and Raman) and radians (QPI). (b) Examples of live-cell (Yarrowia lipolytica, Po1 g) imaging by phase-contrast (PC) and Raman using the 20x/0.4 objective. Specific for Raman (5 second integration times), cells were first grown in rich media (YPD) for 24 hrs and then transferred to nitrogen-depleted media (YSM) for approximately 100 hrs. YSM contained deuterium labelled glucose (glucose-D₇, 20 g/L) yielding distinct Raman bands within the cell silent region (2000cm⁻¹–2300 cm⁻¹ as we [46] and others [45] have reported). We probed a sub-section of these bands using 700 nm excitation and the broadband filter (centered at 812 nm with a 14 nm bandwidth). All Raman images in this figure were processed by a median filter (1-pixel radius) and Gauss blurring (σ = 1), and were acquired at α = 2.48 mm⁻³ and a 12 µm wide scanning ranges (∼5 kHz).

Download Full Size | PDF