1. Introduction

Nonlinear optics in conjunction with ultrashort laser pulses and scanning microscopy has created very useful imaging tools in biology and material science. Various nonlinear optical effects such as second harmonic generation (SHG), third harmonic generation (THG), two-photon or multi-photon excited fluorescence (TPEF or MPEF), four-wave mixing (FWM), coherent anti-Stokes Raman scattering (CARS), stimulated Raman scattering (SRS), etc. can provide intensity-dependent contrasts where each modality is sensitive to specific symmetries, structures or chemical species. Among these, SHG [1], THG [2], TPEF or MPEF [3] and CARS [4–6] microscopy have become very popular lately for fast acquisition imaging applications. One of the unique advantages that the nonlinear optical microscopy brings about over the linear microscopy is that, in the prior, because of the nonlinear intensity-dependence, the signal is generated from only the focus region and thereby allows three-dimensional (3D) sectioning of thick samples without the need of a pinhole as done in confocal microscopy. One beam geometries that include SHG, THG and TPEF or MPEF are the simplest ones among these nonlinear optical microscopy techniques and have been realized for investigations of microstructures in material and biosciences. The SHG technique is highly sensitive to molecular symmetry, whereas THG technique can pick up interfaces or boundaries on the surface and underneath due to inherent nonlinear refractive index-dependence of the optical signal. These have been proven very useful in imaging the noncentrosymmetric arrays of molecules such as collagen fibers found in muscles and other tissues [7], and the cell boundaries or lipid water interfaces [8], respectively. Single ultrashort pulse generated degenerate-FWM signals provide an additional complementary nonlinear coherent contrast mechanism and have been used in microscopy for the detection of low concentrations of chromophores and hemoglobin [9]. Vibrational contrast imaging is necessary to identify specific molecules for label-free microscopy. These measurements are typically obtained using spontaneous Raman spectroscopy or more efficiently by using CARS and SRS for high molecular concentrations, both of which rely on the nonlinear Raman interaction of light with molecules [10]. In particular, CARS based microscopy has been very successful in addressing many questions in medical research and diagnostics, e.g., cancer detection [11], tissue engineering [12] and skin biology [13].

The SHG, THG, FWM and CARS are scattering based techniques which are very well suited for in-vivo measurements as there is no energy absorption and have been shown not to negatively affect cell viability [8]. Therefore, coupling of these nonlinear optical imaging modalities into a single microscope can prove beneficial in a variety of applications requiring complimentary information since each modality brings a different specificity of the sample details [14–17].CARS is a third-order optical process that requires pump and Stokes photons to excite a vibrational mode and probe photons to generate the measurable anti-Stokes signal. While SHG, THG and degenerate FWM require high peak intensities attainable from temporally short pulses, fast CARS measurements have been realized using multiple, synchronized pulses from multiple picosecond sources in conventional techniques by ensuring spatio-temporal overlap in the interaction volume. Conventional experimental schemes for CARS or SRS are complex and require a minimum of two color laser beams. Hence, multimodal nonlinear optical microscopy incorporating these two nonlinear Raman scattering techniques have previously been obtained using multiple, synchronized pulses from multiple sources [16–18]. Recently, fast scanning CARS and multimodal microscopy using a single laser source, yet with multi-beam arrangement for excitation was demonstrated [19,20]. In such studies, the Stokes beam is taken either from the same broad band-width laser source or generated in a nonlinear fiber. The two beams are then spatially and temporally overlapped in the sample by properly adjusting the alignment and the time-delays in the setup. Here, CARS microscopy for a specific Raman mode is achieved by spectral focusing where the chirp in either of the two beams (pump/probe and Stokes) is the control parameter. To reduce the complexity of the CARS systems further and ensure inherent spatio-temporal overlap, there have been developments towards single-pulse, single-beam CARS techniques employing broadband-width femtosecond laser pulses which provide the necessary pump, Stoke, and probe photons [21,22]. Although single-pulse CARS has been demonstrated in either selective excitation or multiplexed schemes [22], these measurements typically use complex pulse shaping devices that require precise calibration and alignment. Furthermore, these techniques require either full spectral acquisitions (multiplexed measurements) or iterative measurement (single detector measurements) over many excitations to resolve vibrational modes therefore inhibiting the incorporation in fast scanning microscopy applications.

The current work is about development and realization of CARS based multimodal nonlinear optical imaging microscope where nonlinear optical signals generated in the SHG, THG, CARS and FWM processes are acquired simultaneously and used for rapid noninvasive label-free bioimaging. We present a simple scheme to acquire single-pulse CARS measurements using a single detector, thereby allowing us perform rapid scanning microscopy in virtually alignment-free experimental configuration consisting of a single broadband-width femtosecond pulse laser source and passive elements without the need of a pulse shaper. Our single pulse CARS approach is not only practically simple, but also it can be easily integrated into any nonlinear optical microscope supporting the band-width of a broad-band laser source. We have developed a single-pulse CARS based multimodal nonlinear optical microscope (MNLOM) enabling us concurrently measure, for the first time, the SHG, THG, CARS, TPEF and FWM signals from single pulses to accomplish rapid scan imaging of biological specimens. All these modalities independently have already been proven to be essential and successful for noninvasive, label-free bioimaging. Combining these modalities into a single setup was made feasible as we resolve the single-pulse CARS signal while preserving the spectral profile of the excitation pulses and therefore are able to obtain multimodal measurements from all these nonlinear optical processes which occur simultaneously.

The rapid, single-pulse CARS based multimodal imaging technique can address the current problem in structural biology, namely, the lack of an integrated technique for fast imaging to assess morphology, organic matrix organization, and mineralization quantification simultaneously in a single cost-effective setup. The most commonly measured diagnostic indicators in bone diseases such as osteoporosis and osteogenesis imperfecta [23], mechanical properties such as tensile strength, elastic modulus, bone loading and age [24], susceptibility to fracture [25], and classification are mineral localization, density and extracellular orientation matrix, e.g., collagen, which can be assessed using the proposed integrated technique. We demonstrate the applicability of our multimodal microscope on two biological samples, a canine femur bone and a rat tail tendon chosen for phosphate mineral localization, molecular (collagen) orientations and bulk structural and morphological investigations which until now have been observed using multiple acquisitions from multiple techniques such as time intensive confocal Raman and high-energy photons based measurements. Furthermore, due to the fact that the broad band-width femtosecond laser pulses are compatible also with high-resolution optical coherence tomography (OCT), the MNLOM will be able to produce wide-field OCT images by simply incorporating an OCT arm into the setup. These OCT scanned images can be zoomed-in online to very narrow regions for sub-micron resolution nonlinear optical images and thereby making the device potentially a unique platform to perform both wide-field and sub-micron resolution optical microscopy for rapid, non-invasive, label-free structural and chemical-sensitive imaging. For completeness, we have performed OCT scans on our samples at present using an independent OCT setup, and discussed the feasibility of incorporating such an OCT arm into our microscope to build a hybrid MNLOM-OCT imaging platform.

2. Methods

An image from a nonlinear optical microscope is the contrast image of the sample formed by collecting the corresponding photons generated in the light-matter interaction and dependent nonlinearly on the incident laser intensity in various possible ways. Image acquisition is achieved by raster scanning the sample across the focal point and point-by-point signal collection from the illuminated focal regions. The multimodal imaging is achieved by simultaneous acquisition of various nonlinear optical signals in a simplified and convenient way equipped with a single laser source and a single microscope platform. The second-order (third-order) optical susceptibility χ⁽²⁾ (χ⁽³⁾) of the material under investigation mediates the interaction between two (three) incident photons to generate a high-energy photon as depicted in the schematic energy level diagram in Figs. 1(a) and 1(b), which if detected sensitively carries the information about the symmetry of the material susceptibility and hence can be used for structural characterization. Nonlinear Raman scattering is also mediated by the third order susceptibility χ⁽³⁾ where the polarizability of the molecules changes with the spatial coordinate variations in presence of three incident photons under the phase matching condition, a general form of which is the nonresonant electronic four wave mixing in the absence of any resonant vibrational level. In Fig. 1, we have also shown the schematic energy level diagram for some of the possible multi-photon processes taking place in resonant CARS and FWM. Figure 1(c) considers the situation with a narrow-band pump and probe, and a broad-band stokes pulse such that only a single vibrational level gets coherently excited along with a large number of nonresonant levels represented by the dashed lines creating the large background signal. In Fig. 1(d), the four wave mixing process is purely nonresonant in nature as no real vibrational level has been addressed by the three incident laser fields and the blue shifted light is generated due to electronic processes only. Therefore, for broad band-width laser pulses, a combination of processes represented in Figs. 1(c) and 1(d) contribute to the total four wave mixing signal which has a very small contribution from resonant vibrational levels as compared to the large nonresonant contribution, referred as the resonant CARS and the FWM signals, respectively, later on in the paper.

 

Fig. 1 Schematic energy level diagram for the multi-photon processes; (a) second harmonic generation, (b) third harmonic generation, (c) a combination of resonant and nonresonant coherent anti-Stokes Raman scattering, and (d) purely nonresonant electronic four wave mixing which contributes to a broad-band background signal in the spectral domain or an ultrashort temporal component at zero in the time-domain. By applying a narrow band-width pump and probe (green), and a broad band-width stokes photons (red), there are three cases of the four wave mixing signal generation, namely, the resonant excitation of one vibrational level with frequency Ω along with a large number of nonresonant levels represented by the dashed lines in (c) and only nonresonant contribution in (d). The four wave mixing process presented in (d) is purely nonresonant in nature as no real vibrational level has been addressed by the three incident laser fields. A combination of processes represented in (c) and (d) contribute to the total four wave mixing signal in the experiments and hence the total signal has a very small resonant contribution as compared to the large nonresonant signal.

Download Full Size | PDF

The heart of our multimodal nonlinear optical microscope is the single pulse CARS technique which relies on two passive elements for resonant CARS generation and selective detection, respectively. In the single pulse CARS scheme, we use the concept of single shot multiplexed excitation and detection which is typically achieved by defining a narrowband probe in the excitation pulse and measuring the energy shift in the spectrally resolved signal from the actual probe location by the energy of the coherently excited vibrations [21,22]. Similarly, we can detect specific Raman mode by controlling the enhancement of a particular spectral region in the presence of the resonant vibrational mode. The required spectral modulation of the excitation pulse is accomplished by using a spectrally tunable narrow notch feature created by a resonant photonic crystal slab (RPCS) [26] or a narrowband notch filter. By creating a notch in the excitation pulse spectrum, the pulse characteristics such as the overall intensity or the extent of the total pulse spectrum that is useful for all other nonlinear processes such as SHG, THG, TPEF or FWM is minimally altered. However, by doing so one gets enabled to resolve the coherently excited CARS spectra. To explain how this becomes possible in our scheme, we have to look at the frequency-domain and the time-domain pictures together as presented in Fig. 2(a) using simulated pulses for both the unshaped and notch-shaped cases. The effect of the notch in a Gaussian laser pulse (Fig. 2(a)) in the frequency domain is to create a small magnitude component in the time-domain which is time-delayed with respect to the temporal extent of the main pulse. This longer component in the time-domain is largely dependent on the width of the notch feature achievable in the experiments. In all the CARS schemes, the probe pulse needs necessarily to be spectrally narrow (that corresponds to picosecond or wider component in time-domain) so that the CARS spectra can be resolved close to the natural Raman line-width of the excited modes. The cartoon in Fig. 2(b) correctly depicts the situation in our single pulse CARS experiments having a spectrally broad laser pulse providing the necessary pump and stokes photons and a narrow notch feature acting as the delayed probe to resolve the coherently excited Raman modes in the system. Narrow well-defined resonant CARS features, blue-shifted from the probe location by the vibrational frequencies Ω_i of the excited Raman modes are generated riding on the broad-band FWM background spectrum. However, the detected CARS lines have line-widths limited by the notch width.

 

Fig. 2 Single pulse CARS scheme explained. (a) Simulated traces of Gaussian femtosecond (fs) laser pulses in wavelength and the time domain compared for the pulses as they are vis-à-vis shaped using a notch filter. The effect of a narrow notch feature in the spectral domain results into a picosecond delayed component in the pulse in the time-domain. (b) Schematic energy level picture of the broad-band pump and stokes pulses coherently exciting vibrational modes with frequencies Ω_i which are frequency-resolved using a narrow probe. (c-d) Actual experimental results on toluene showing the strong Raman vibrational bands coherently excited and spectrally resolved using the single pulse CARS scheme. The notch-shaped laser pulse spectra for three different angular positions of the notch filter are shown in the upper panel of (c) and the corresponding FWM spectra are shown in the lower panel of (c) along with that for the notch location for which there are no resonant features in detected spectral window (black dashed curve). (d) Normalized difference FWM spectra showing the spectrally resolved CARS features at the expected frequencies. These features move with the excitation notch location (the peaks are connected by dashed lines for a reference). For the probe notch location at 772 nm, the CARS features appear at 716, 727 and 706 nm which correspond to Raman frequencies of ~1010 cm⁻¹, 793 cm⁻¹ and 1202 cm⁻¹, respectively.

Download Full Size | PDF

An experimental demonstration of performing vibrational spectroscopy with the above described single pulse CARS scheme is presented in Figs. 2(c) and 2(d) where three coherently excited Raman bands of toluene have been spectrally resolved. Figure 2(c) shows the spectrum of the excitation pulses shaped with the notch filter for three probe locations in the upper panel and the corresponding FWM spectra in the bottom panel. In the FWM spectra, notch shaped CARS features can be easily seen which move with the movement of the excitation notch location (shown by the dashed lines connecting the peaks in Fig. 2(d)). Typical linear Raman spectrum of toluene has about six vibrational modes in the finger-print region at frequencies of ~1010 cm⁻¹ (stronger), 793 cm⁻¹(strong), 505 cm⁻¹ (medium), 1202 cm⁻¹ (medium), and 615 cm⁻¹ (weak), ordered as per their intensity in the spectrum. The normalized difference FWM spectra for each of the three probe locations are presented in Fig. 2(d) showing the spectrally resolved CARS features at the expected frequencies where the spectral line-width of each mode is more or less the same as the width of the excitation notch, i.e., of ~1.4 nm in our case. For the excitation notch location at 772 nm, the CARS features appear at 727, 716 and 706 nm (much weaker feature at 706 nm becomes observable after taking a difference spectrum) which correspond to the normal modes of vibration of toluene at frequencies ~793, 1010, and 1202 cm⁻¹, respectively. Therefore we are able to see only three of them which appear in the experimental spectral window. Commercially, very narrow notch filters are available which can provide very fine notches (better than 0.5 nm full-widths at half maximum) at the right experimental wavelengths and hence there is scope for better resolved resonant CARS detection in our single pulse CARS scheme.

The use of an RPCS as a passive shaping element has been demonstrated already in multiplexed, single-pulse CARS spectroscopy [26]. While this technique simultaneously resolves multiple vibrational modes, it requires spectrally resolved measurements limiting its applicability in fast scan microscopic imaging experiments. To alleviate the need for spectral imaging devices, we use an additional RPCS or narrow filter to reflect a narrow-band of the forward propagating FWM signal tuned for a particular resonant CARS mode. Here on, we will refer to the first notch filter in the excitation path as the excitation notch and the other in the detection path as the detection notch. By tuning the location of the notch feature in the excitation pulse spectrum, we can enhance the detected signal in the narrow-band reflected region in the presence of a Raman scatterer. To achieve fast scanning and low integration times, we mounted the excitation notch filter on a galvanometer and rapidly modulated it at KHz rates around its mean position by a small amount. We measured the spectral enhancement by collecting the narrowband light reflected off the detection notch filter, with a photomultiplier tube (PMT) and a lock-in amplifier set to the modulation frequency of the excitation notch. The measured signal is proportional to the resonant signal as the non-resonant background is subtracted in real-time by the lock-in amplifier. This allows for acquisition at least an order of magnitude faster than previously reported using spectrally resolved detection [26]. The forward propagating light from the sample is spectrally selected and detected at various PMTs for simultaneous multimodal imaging experiments. Moreover, the CARS spectrum of the resonantly excited Raman modes can also be generated in our multimodal imaging setup without using a spectrometer. For this, we simply fix the spectral location of the detection notch (at optical frequency ω_d) and scan the location of the excitation notch (at optical frequency ω_pr) via the voltage applied to the galvanometer mounted excitation notch filer. The signal magnitude at the detection notch is plotted as the function of the difference between the locations of the excitation and detection notches (ω_d-ω_pr = ω_vibration). By modulating the excitation notch around its mean and collecting the signal reflected off the detection notch at the same modulation frequency using the lock-in amplifier not only avoids the background FWM contribution, it also improves the signal to noise immensely. For constructing the CARS map of a sample for a specific resonant vibrational mode, signal is collected point by point for fixed excitation and detection notches while scanning the sample on a translational stage. Although the excitation notch filter has little effect on the excitation pulse, creating a notch at different spectral locations over part or full of the laser pulse spectrum results in small changes in the pulse energy, for example, minimum for the notch located near the tail and maximum when it is near the peak of a Gaussian pulse spectrum. This variation in pulse energy is also modulated at the modulation frequency of the resonant signal and therefore creates a background that can be normalized by subtracting the modulated nonresonant FWM signal while measuring the multiplex CARS spectrum by scanning/moving the excitation notch location over substantial part of the excitation pulse. For the experimental results described in this paper, we only modulate the excitation notch around its fixed mean position by the extent of its spectral width, e.g., ~772 ± 0.7 nm, and perform the resonant CARS imaging for a fixed detection notch. This leads to negligible modulation of the overall pulse energy incident onto the sample and hence the strength of all the nonlinear signals is least affected.

3. Experiments

The experimental setup is shown in Fig. 3 which consists solely of passive optical elements. Broadband-width (790 ± 75nm) femtosecond pulses are generated with a home-built Ti:Sapphire (Ti:S) oscillator running at ~80 MHz pulse-repetition rate. The pulses are then dispersion-compensated using an ultrafast-pulse compressor (either a grating compressor or a prism compressor) to achieve nearly transform limited pulses at the sample point, followed by spectral modulation using an optical notch filter (NF) which essentially places a sharp notch on the pulse spectrum with its location movable by galvanometer mount. The pulses are then long-wave pass filtered using a 735nm filter and by cutting in the spectral focal plane of the pulse compressor such that no laser light was observed in the CARS detection window.

 

Fig. 3 Layout of the multimodal nonlinear optical microscopy setup used for simultaneous second harmonic generation (SHG), third harmonic generation (THG), resonant coherent anti-Stokes Raman scattering (CARS), and four wave mixing (FWM) micro-spectroscopic imaging of biological samples. The forward propagation light scattered through the sample is spectrally selected by various dichroic mirrors (DM), short-pass filters (SPF) or band-pass filters (BPF) before it is detected at the corresponding photomultiplier tube (PMT). The resonant detection of a CARS signal is achieved by a pair of notch filters (NF), one tunable excitation notch and one fixed detection notch. The spectrally resolved and time-integrated light is detected using a spectrometer and PMTs, respectively. This experimental system is capable of producing simultaneous two photon excited fluorescence (TPEF) micrographs in the epi-direction.

Download Full Size | PDF

Microscope Design –The light transmitted through the long-wave pass filter was focused on to the sample using a 20X/0.4 objective lens. The bright-field light and TPEF were collected in the epi-direction using a CCD camera and a photomultiplier tube, respectively. A fused silica UV condenser (16 mm focal length, 25 mm clear aperture) was used to collimate the forward propagating light containing the SHG, THG, FWM and CARS photons before being spectrally separated using various and dichroic mirrors (DM) and short wave pass filter (SPF) or band wave pass filters (BPF), and detected using photomultiplier tubes (PMTs) in combination with either current preamplifiers (Stanford Research model SR570) or lock-in amplifiers (Signal Recovery model 7265). The THG light was reflected off a custom ultraviolet dichroic mirror (EKSMA Optics HR>95% 250-280), filtered using a short wavelength band-pass filter centered at 265 ± 15nm before the signal collection at a PMT (HamamatsuR7154). After the UV dichroic, part of the transmitted light was reflected off a dichroic mirror (Semrock FF510Di HR>99% 350-480) and filtered with a band-pass filter centered at 400nm (Chroma HQ400/40M-2P) for SHG signal detection at a PMT (Hamamatsu R4220). The remaining of the transmitted light containing the entire resonant and nonresonant CARS signal is short pass filtered at 730nm (Omega 3RD730) filter. An RPSC notch filter similar to that used in modulation of the excitation pulse was used to reflect a tunable narrow spectral region to be detected at a PMT (Hamamatsu R955) for resonant CARS signal collection. The remaining light that consist the nonresonant FWM signal was sent to either a PMT (Hamamatsu R3896)or to a liquid nitrogen cooled CCD (Symphony) coupled high-resolution spectrometer (JobinYvonTriax320) for time-integrated detection or spectrally resolved CARS measurements, respectively. The PMT signals for the SHG and THG were amplified using current pre-amplifiers and that for the CARS and FWM using lock-in amplifiers. All measurements were recorded using an analog to digital converter data acquisition PCI card (National InstrumentsDAQ-6024E). As shown in Fig. 3, the TPEF signal generated in the epi-direction can also be collected on another PMT and used for simultaneous imaging. However, we have limited our experimental demonstrations in this paper only for the forward propagating signals.

Galvanometer mounted excitation notch filter -The location of the notch is easily tunable by changing the relative angle of the notch filter to the collimated incident beam. To rapidly change the spectral location of the enhanced CARS signal in the presence of a Raman scatter, we mounted the excitation notch filter on a galvanometer that we controlled using a computer addressed function generator (Agilent 33210A). To obtain full CARS spectra, we fixed the detection notch filter (typically near the maximum signal) and linearly scanned the excitation notch filter throughout the entire bandwidth of the laser pulse. The reflected light from the detection notch was detected on the respective PMT. As the excitation notch filter was scanned/moved over the pulse spectrum, it was rapidly modulated around the mean notch position by its spectral width (~1.4 nm), e.g., modulating the notch located at 772 nm in the range of 772 ± 0.7 nm, which corresponds to scanning the dc zero-point offset of a sinusoidal voltage function driving the galvanometer. The fast modulation of the notch allows the resonant CARS signal to be picked up by lock-in amplifier at the modulation frequency and record the multiplex CARS spectrum without the need of a spectrometer. The spectral location of the excitation notch filter was calibrated as a function of the dc voltage so that the optical frequency of the coherently excited Raman modes could be readily resolved from the scans. For a specific Raman vibrational mode to be used in vibrational contrast CARS imaging as described in our present paper, the excitation notch was fixed at one spectral location such that the frequency shift between the excitation and the detection notches corresponds to the frequency of the desired Raman mode.

Sample preparation -Bone sections were taken from a 1-year old, skeletally mature male dog of mixed breed [27] without evidence of skeletal disease. The left femur was removed from soft tissues using a surgical blade and gauze. The central region (diaphysis) of the femur was separated from the epiphysis and mounted on a low speed diamond saw. Transverse sections of 3mm thickness were cut under water irrigation. The sections were further reduced to a thickness of ~100µm with emery paper. The bone plates were inspected under a polarized light microscope. Round concentric structures were visualized confirming that the sample bone belonged to the osteonal bone type. The collagen fibers were prepared from an adult (3-6 months) Norwegian female rat-tail following the procedures described previously [28]. Single collagen fibers (diameter ~0.6 mm) were dissected and extensively washed in TNC buffer (50mM TRIS, pH 7.4, 150mM NaCl, 10mM CaCl₂, 0.02% NaN₃) to remove the macroscopic scrap pieces of tissues and the excess proteases.

Microscopy –A demonstration of performing vibrational spectroscopy with the simple setup of Fig. 3 is already presented in Fig. 2 on toluene. High-resolution multimodal imaging was performed by collecting time-integrated signals using PMTs rather than a spectrometer. To preserve the beam profile, we achieved 3D scanning by moving the sample in the transverse plane across the focal point using a 2D piezo stage (Mad City Labs, Inc. Nano-Bios) and translating the objective lens along the direction of light propagation (z) using a piezo driven linear stage (Nano-Bios) in a closed loop feedback system. To compensate for the non-linear movement due to inertia, we measured the stage position during scanning and reconstructed the images to restore spatial integrity. Fast scanning CARS imaging is achieved by fixing the spectral location of the detection notch (at frequency ω_d) and scanning the location of the modulated probe frequency (ω_pr) via the excitation notch, as explained already in the last section. The notch spectra have full width at half maximum of ~1.4nm.The time-integrated signals in all the four imaging modalities of SHG, THG, resonant CARS and FWM were collected as the sample is raster scanned to construct the corresponding images simultaneously. In current experiments, the pixel dwell time or the signal integration time for each pixel was limited by the maximum operating frequency of the excitation notch galvanometer (typically 1 ms), i.e., the lock-in amplifier must be allowed to integrate for longer than the modulation period (typical lock-in time-constant was 5 ms). For many biological samples which are not rigid enough to be moved by the translational stage without being damaged or deformed, one would prefer to use a pair of galvanometric mirrors for scanning the laser beam over the immobile sample. This, of course, could significantly improve the speed of our multimodal imaging setup.

4. Results

Simultaneous measurements in the four imaging modalities, i.e., the SHG, THG, CARS and FWM on the canine femur bone and the rat tail tendon are presented here to demonstrate the single-pulse CARS based multimodal imaging capability of our custom-built nonlinear optical microscope. All these modalities are set for the forward propagating light. For the fixed detection notch filter we place the excitation notch filter at an angular position such that the light reflected off from the detection notch is tuned only for a specific resonant CARS mode at which we require to perform the vibrational contrast imaging. Under such a condition, the PMT placed for CARS signal detection receives maximum signal allowing us perform background free selective CARS imaging of the sample with the help of lock-in amplifier. The FWM imaging is performed by collecting the remaining FWM generated light on another PMT and the lock-in amplifier. Spectrally selected light in the SHG and THG region is collected at the respective PMTs and amplified using current pre-amplifiers. For these experiments, a 735 nm long-wave pass filter (HQ735LP) in the excitation beam path and another 730 nm short-wave pass filter (3RD 730SP) in the CARS detection beam path were used to completely filter out the incident light at the detection unit. All the experimental results presented here have been obtained using excitation power flux of ~10 mW/cm². The lateral spatial resolution in our microscope is about ~0.4 micron, mainly limited by the numerical aperture (NA = 0.4) of the objective lens we have used for the present experiments. However, since we employ the signal generated in various nonlinear optical processes for imaging, the spatial resolution is improved further by the order of that particular nonlinear process.

Canine femur bones - Here we demonstrate measurements to simultaneously assess phosphate mineralization, collagen, and bulk morphology in canine femur mid-shafts. We first obtained a full CARS spectrum from a thin slice of the sample to identify the presence of calcium phosphate (PO₄) band namely, the O-P-O symmetric stretch vibrational mode at frequency of ~960 cm⁻¹ and then fixed both the excitation and detection notches to obtain the CARS map of the sample. We acquired images at off resonance frequencies also which resulted in far less contrasts confirming that the vibrational contrast in our recorded CARS images of the bones were indeed generated due to the phosphate mode. Figure 4(a) shows the recorded spectra of forward propagating FWM generated light from the bone sample taken for two positions of the excitation notch such that for one there is just the nonresonant FWM background signal (lower curve) and for the other there is a resonant CARS feature superimposed on the nonresonant background (upper curve) in the observation window of the spectrometer. The two curves have been vertically shifted from each other for clarity. The contribution due to only the coherently excited CARS modes is obtained by taking a difference between the two recorded normalized spectra as shown in the lower panel of Fig. 4(a). The CARS spectrum of the sample clearly shows the strong O-P-O stretching mode at ~960 cm⁻¹ and was used for the CARS imaging of the sample.

 

Fig. 4 (a) As recorded spectra (upper panel) of the forward propagating four wave mixing generated light from the canine femur bone sample shown for two positions of the excitation notch location such that for one there is just the nonresonant background FWM signal (lower curve) and for the other there is a resonant CARS feature superimposed on the nonresonant background in the observation window of the spectrometer, and the normalized difference FWM spectrum (lower panel) revealing the resonant CARS feature at a frequency of ~960 cm⁻¹. (b) Microscope generated real images of the bone sample set for simultaneous SHG, THG, FWM and the resonant CARS imaging at frequency 960 cm⁻¹ as marked in the corresponding images. The results in each modality have been drawn with a single color, i.e., CARS in red, SHG in blue, THG in gray and FWM in coral, between the minimum (color black) and the maximum (color white). The scanned area is 250μm × 250μm with 1μm pixel size. The black dot near the top edge in the images (as marked in the THG image) is the Haversian canal in the bone which provides a good landmark for imaging around it. Scale bar is 50μm. The interlinking of atleast four Osteons can be identified in the images which have been marked with dashed lines in the THG image.

Download Full Size | PDF

The imaging results are shown in Fig. 4(b) for each of the four modalities as marked in the respective images over a scanned area of 250μm × 250μm with 1μm resolution. It is clear from the images that one gets different contrasts over the complete scanned area due to chemical and structural selectivity in the four different nonlinear optical processes thus providing complementary information acquired from the tissue microstructures in the bone. The sample was imaged in the cortical bone region (see the OCT image on a larger length scale discussed later) where parts of few Osteons (or the Haversian system) can be visualized interlinked with each other as marked in the THG image in Fig. 4(b). The THG image provides the general structural morphology since interfaces and boundaries or the optical refractive heterogeneities within the scanned region of the sample contribute to the contrast. The Haversian canal appears as a small black dot near the upper end of the images (marked by a half circle in the THG image).The wall of the Osteon consists of concentric rings approximately 3-7µm thick alterations according to the chemical composition and structural organization of collagen fibrils. They appear woven in a young bone and are less organized or randomly distributed. With age they are desorbed and replaced by lamellar bone made up of preferentially orientated collagen fibrils. The collagen molecules mineralized with hydroxyl apatite crystals represent mineralized collagen fibrils. Since SHG contrast arises mainly due to these ordered structures of the collagen fibrils, we can see that they are aligned tangentially to the boundary of the Osteon. It is known from the literature that O-P-O bonds are oriented along the long axis of the collagen fibrils [29]. Possibly this could also be inferred from the CARS image together with the SHG image in Fig. 4(b). The small dark black structures in Fig. 4(b) are attributed to cartilage lacunae which are cavities containing cartilage cells and surrounded by lamellae of calcified matrix. For the FWM image in Fig. 4(b), we also used the lock-in detection, but collected the scattered light in a much broader spectral window between 700 and 730 nm while the resonant CARS feature due to the 960 cm⁻¹ Raman band was located at ~718 nm (Fig. 4(a)). The similar contrast seen in both the CARS and FWM images of the bone in Fig. 4(b) probably hints that the FWM signal is dominated by the strong and spectrally broad Raman band at ~960 cm⁻¹. We did confirm that the contrast in the CARS image was due to only the 960 cm⁻¹ band by simply moving the excitation notch away from resonance.

The SHG and THG images were also recorded in a sequence of various depths by moving the focal plane along the sample thickness. This provides clearer view of the complimentary structural micro-morphology of the sample brought out by these two imaging modalities due to differences in the origin of the contrasts. The results taken over scan area of 250μm × 250μmand on a region different from that in Fig. 4 are presented in Fig. 5 where each subsequent image has been taken at 5μm deeper into the sample than the image before it. We find that the magnitude of the two signals changes oppositely. This means that the local density and distribution of the second harmonic and third harmonic scattering centers is varying along the thickness which might arise if the Osteon and other structures appearing on the cross-sectional view are twisted or bent along the bone length in the observed region. The contrast in THG image arises when the illuminating beam is focused at a small inclusion or at an interface between two materials and vanishes completely from most homogeneous or isotropic bulk samples. So it is helpful in providing the depth-resolved inhomogeneities also and can be complimented by the OCT on a much larger length scales allowing overlapping and co-registration of the two techniques.

 

Fig. 5 SHG and THG contrast images of canine femur bone sample taken over an scan area of 250µm × 250µm recorded at various depth positions (z) of the focal plane inside the sample relative to its first position (z = 0 for the left most image).

Download Full Size | PDF

Collagen-rich Rat tail tendon – As a second example for the realization of the applicability of our multimodal imaging tool, we choose a collagen-rich tendon from a rat tail. Collagen is a structural protein that forms the major component of the extracellular matrix providing mechanical integrity to different tissues and organs in all the metazoa, e.g., in bone, tendon and skin. Collagen type-I is the most abundant form of collagen which is classified as the fibril-forming collagen [28]. The diameter and length of these fibrils vary depending on the anatomical location. Rat tail tendon has been a widely investigated biological system in various experiments to understand the structure of collagen molecules and their organization into fibrils, the size and orientations of the fibrils and their role in the biomechanical properties of the tissues [28,30–33]. Nonlinear optical microscopic experiments on rat tail tendon have also been performed bringing useful insights in many previous studies [34–36].

Chemical and structural morphology specific nonlinear optical images of the rat tail tendon over a scanned area of 250μm × 250μm taken simultaneously using the four modalities of our custom-built MNLOM are shown in Fig. 6. For the resonant CARS imaging, we fixed the excitation and detection notches for Raman frequency of ~1260 cm⁻¹ which is close to the frequency of amide-III band of protein and can be considered as Raman marker of unordered and triple helical structures of collagen type-I and type-IV [37–39]. The usual amide-I band near 1665 cm⁻¹ arises from the C = O stretching vibrations of the protein backbone, and are related to collagen type-I chains in the fibrils [39]. However, this band is outside the sensitivity of our current experimental setup. Many cable-like, parallel fiber bundles can be visualized clearly in the CARS, SHG and FWM images. The THG signal is very strong near the edges of the sample. From an enlarged view of an SHG map as shown by the dashed square box of length 50μm in Fig. 6, we have estimated the density of the collagen fibrils in our rat tail tendon to be ~0.2/μm or an average diameter of the fibril to be ~5μm.

 

Fig. 6 Nonlinear optical images of a rat tail tendon taken simultaneously using the resonant CARS for collagen imaging at ~1260 cm⁻¹, FWM, SHG and THG imaging modalities of the custom-built MNLOM platform. The scale bar is 50μm. For estimating the areal density of the collagen fibrils, a square area (dashed in yellow) of length 50μm is chosen within the SHG image of the sample.

Download Full Size | PDF

Clearly, in the above we have seen that the spectral scanning of the CARS by a narrow-band notch filter offers significant simplification over setups using two beams; hence it could provide compact, easy to use and highly cost-effective CARS microscopy. Furthermore, other single-beam nonlinear optical microscopy techniques, SHG, THG, FWM and TPEF are easily integratable in the same platform to provide a simple multimodal microscope.

Wide-field scanning by OCT –Optical coherence tomography is a noninvasive optical imaging method based on white light interferometry which primarily detects the heterogeneities in the refractive index profile in a sample under investigation [40,41]. In contrast to other optical microscopy techniques, where the resolution, either axial or lateral, is restricted by the numerical aperture of the objective and by the wavelength of the light, the axial and the transverse resolutions in OCT are decoupled. For the later, the axial resolution δz is proportional [42] to λ₀²/Δλ with λ₀ being the central wavelength and Δλ, the spectral band-width of the laser. Therefore, ultra-broad band-width femtosecond pulses from the Ti:S lasers enable ultrahigh resolution OCT with resolution close to that of histology. The high-resolution microscopic structural images that we obtain using THG imaging can really be complimented with the wide-field scanned images from the OCT because both the techniques are sensitive to refractive index inhomogeneities in the sample.

For the purpose of completeness, we performed wide-field scanning spectral domain OCT (SD-OCT) measurements on our samples using a hybrid CARS-OCT setup with details already given in [43] and keeping the laser parameters similar to what we have used in the nonlinear optical imaging experiments discussed above. Large area 3D wide-field view of the canine femur bone from OCT scans performed over an area of 8.4mm × 4.4mm are shown in Fig. 7. On this comparatively much larger scale (a few mm), the inhomogeneities in biological samples usually arise from the distribution of different cell types. The 3D imaging of bone poses some difficulties since the bone matrix is highly scattering. Nonetheless, the different appearance between cortical and trabecular bone regions can be visualized and a 3D cross-section can be made. As can be seen from Fig. 7, the trabecular bone has the typical honeycomb network and is located in the interior of the bone sample. It is surrounded by the much denser cortical bone that forms the outer shell of the bone (Fig. 7). The Osteons or the Haversian systems and, the Haversian canal with blood vessels and sympathetic nerve fibers appearing as black holes, can be visualized clearly. The nonlinear optical images discussed and presented already in Fig. 4 were taken in a region within the yellow square in the OCT image in Fig. 7. Similarly, the results from the wide-field OCT scanning on the rat tail tendon obtained using our custom-built SD-OCT setup are presented in Fig. 8 for the front and the top views. A total of 4.2mm × 1.1mm area was scanned. The parallel stripes like features are related to collagen fiber bundles.

 

Fig. 7 Wide-field prescreening of the canine femur bone obtained using SD-OCT setup described in [43] covering 8.4mm × 4.4mm area of the sample surface: enface view summed up over all depth information (top) and cross-sectional slice (bottom) of size 1024 pixels × 512 pixels.

Download Full Size | PDF

 

Fig. 8 Wide field SD-OCT images of the rat tail tendon taken using custom-built SD-OCT setup described in [43] for the front view (a) and top view (b). A total 4.2mm × 1.1mm area of the sample surface was covered.

Download Full Size | PDF

There is significant demand for an imaging modality that seamlessly permits extraction of localized chemical specificity and tissue morphology on a micron scale to allow for functional imaging of physiologic and pathologic processes. Therefore a combination of technologies revealing morphological and functional parameters is highly attractive for investigation of dynamic processes in a wide range of fields from developmental biology to pathogenesis and clinical treatment. The different and partially costly light sources required for CARS and OCT have traditionally presented a barrier for the combination of these two techniques into a single, biologically applicable system. Nevertheless, these imaging devices have the ability to extract highly-contrasted molecular fingerprints of small volumes to monitor physiologic activity at high specificity (CARS) or can quickly visualize large three-dimensional structures with a wealth of morphologic, but only limited functional information (OCT). Multimodal CARS-OCT designs have been proposed earlier and demonstrated the potential for nonlinear optical contrast enhancement [44,45]. Nonetheless, highly sophisticated setups with regenerative amplifier and optical parametric amplifier (OPA) for detection of a single vibrational mode or the need of knowledge of the exact composition of the samples prevented widespread applications of these approaches. Also, few other multimodal nonlinear optical microscopes have been developed and combined with OCT. For example, the combination of SHG and OCT with a femtosecond Ti:S laser already dates back to 2004 [46]. The desired features of such a multimodal imaging platform are fast wide-field acquisition with high sensitivity and specificity as well as high resolution and reliability at reasonable costs to reveal morphological and molecular changes approaching cellular resolution. Combining the single-beam/single-pulse CARS based multimodal nonlinear optical microscope as presented in this paper and the SD-OCT into a single platform, a highly valuable and cost-effective device is feasible to be made as both of these require a single broad band-width femtosecond laser source. In the simplest case, while switching between the two modes one would need to change the objective lens and the scanning mirrors to take care of the size of the field of view and the imaging speeds in the two modalities. However, parallel acquisitions of OCT scans with tissue morphology down to the micrometer scale and localized chemical specificity at cellular levels are achievable. Such a multimodal NLOM-OCT device is hoped to significantly impact on noninvasive functional tissue diagnostics.

5. Conclusions and outlook

In summary, we have presented a single-pulse CARS technique capable of single detector signal acquisition suitable for rapid scanning microscopy. This is achieved by using two, carefully tuned, spectral notch filters to modulate and detect the resonant CARS signal in a lock-in amplifier. In contrast to shapers, which typically have ~50% efficiency, the notch filters transmit >99% of the pulse energy with minimal effect to the pulse and are, therefore, suitable for multimodal detection. We have demonstrated this CARS technique for spectroscopy and microscopy and also incorporated the SHG, THG, and FWM for simultaneous multimodal imaging of tissues with spatial co-registration and faster integration times. While we demonstrate this technique with millisecond integration times, this was limited by the response time of the piezo translational stage and the maximum galvanometer frequency which can be further increased resulting in µs rates for strongly scattering modes. As the proof of concept, we have realized the applicability of our novel multimodal imaging setup to characterize a canine femur bone and a rat tail tendon based solely on intrinsic sources of contrast. Molecular sensitivity of CARS permits to distinguish similarly looking tissues and can be used locally for label-free imaging of various important bio-molecules at better than a micron resolution and at fast image acquisition rates. SHG imaging is extremely valuable for imaging collagen and is capable of providing information about orientation, crystallinity and morphology. THG provides structural information which is intrinsically background-free. Because SHG and THG do not involve excitation of molecules, hence they do not suffer from photo-toxicity effects or photo-bleaching, both of which are critical in the case of fluorescence microscopy including the two-photon fluorescence microscopy. All the above factors bring the single-pulse, single-beam CARS-based multimodal microscope described here on similar footing with other state-of-art techniques in terms of high-resolution fast scanning imaging performance. Clearly, the unique advantages of our setup are single-pulse, single beam approach that is nearly alignment-free, compact, easy to use and cost-effective.

The ultra-broad band-width femtosecond pulse Ti:S laser driving our multimodal nonlinear optical microscope is highly suitable source for SD-OCT as well. Therefore, a combined MNLOM-OCT device can be envisioned by integrating an OCT arm into our multimodal microscope that will allow accomplish a highly valuable and more importantly, a cost-effective device for noninvasive label-free imaging in biology and medicine. It will bring together the molecular structural and vibrational contrast imaging capability for tissue micro-morphology by the MNLOM and the depth resolved wide-field scanning of larger volumes by the OCT for pre-screening in parallel acquisitions. Still several technological challenges will have to be overcome in order to make such a quadruple imaging platform applicable for thick or strongly absorbing tissues and the limitations in depth information.