1. Introduction

Coherent anti-Stokes Raman scattering (CARS) microscopy has been demonstrated to be a powerful tool for non-invasive, chemical imaging of biological systems [1-7]. This technique obviates the need for labeling the sample because it relies on molecular vibrations for contrast. When the difference in the frequencies of the pump and Stokes fields incident on the sample is in resonance with the sample molecular vibration frequency, a strong CARS signal is emitted. Due to the coherent nature of the CARS signal and nonlinear dependence on the excitation intensity, CARS microscopy permits relatively fast data acquisition times and has inherent 3-dimensional sectioning capability.

Despite its well acknowledged power and utility, CARS microscopy is not yet a widely adopted, commercially available biomedical imaging tool. One of the primary reasons for this is that it is technologically complex and involves investing in expensive light sources for generating the pump and Stokes beams. Examples of such expensive, yet close to ideal light sources range from two synchronized picosecond lasers [1] to the current state-of-the-art synchronously pumped OPO system [2]. A fiber based approach involving a single broadband laser source and photonic crystal fiber (PCF) [4–6] to generate both pump and Stokes pulses for CARS, has been developed recently with the view to reducing costs. In particular, broadband, multiplex CARS spectroscopy has been demonstrated using the supercontinuum (SC) generated in a photonic crystal fiber (PCF) with a single zero dispersion wavelength (ZDW) [5, 6]. An alternate scheme for CARS microspectroscopy was proposed recently [7], in which a picosecond fiber laser generated the pump pulse while a PCF with two widely separated ZDWs, was used as a frequency tunable Stokes source. It is notable that these efforts [4–7] involving PCF for the Stokes pulse have not been successful at generating CARS images with short data acquisition times for studying dynamic living systems. The reason for this is that the SC generated in a PCF with one ZDW involves soliton fission and exhibits significant noise amplification [8-10] at low frequencies that are important for imaging. In contrast, SC generation in a PCF with two closely lying ZDWs results from a combination of self-phase modulation and phase matched four-wave mixing and not soliton fission [11]. Consequently, the noise amplification through modulation instabilities is suppressed, resulting in very well-behaved, low noise amplitude and phase of the SC. Such a PCF was recently employed for ultrahigh resolution optical coherence tomography at 800 nm and 1300 nm [12, 17].

We present in this paper, our results involving the use of a PCF with two closely lying ZDWs for generating the low noise Stokes pulse and its application in CARS microscopy when combined with a femtosecond pump pulse. The current prototype CARS microscope provides a low cost alternative for imaging lipid-rich structures in biological specimens with short data acquisition times. The SC output of the PCF was experimentally characterized as a function of the average input power, pump wavelength, pulse duration and polarization of the input pulse. The possibility of extending the CARS imaging for molecular vibration frequencies below 2400 cm^-1 by spectral shaping of the Stokes pulse, was also explored.

2. PCF with two closely lying ZDWs for Stokes pulse

The PCF (NL-1.4-775-945, Crystal Fibre A/S, Denmark) used for generating the Stokes pulse has a parabolic dispersion profile similar to the one used by Hilligsoe and colleagues [11], and has two closely lying ZDWs at 775nm and 945nm. It is however 12.5 cm long and is housed inside a hermetically sealed package (Femtowhite, Crystal Fibre A/S) that improves ease of light coupling and enhances long term power stability. The pitch of the fiber is 0.98 μm and it has a relative hole size of 0.54. The fiber core diameter is 1.5 μm with a corresponding effective area of ~1.3 μm² and a high nonlinear coefficient γ of 0.15 W^-1m^-1. The numerical aperture of the fiber is 0.49 at 800 nm and the fiber is single mode but not polarization maintaining.

Light from a tunable Ti:sapphire laser (Tsunami, Spectra-Physics, Mountain View, CA) producing ~ 65 fs pulses at 80 MHz repetition rate was split into two arms: pump and Stokes arms as shown in Fig. 4, Section 3.. A Faraday isolator was inserted in the beam path immediately after the laser, for avoiding back reflection into the laser. This however stretched the pulse (dispersion of ~ 6600 fs²), so a custom-built prism compressor consisting of two prisms was added in the Stokes arm. This enabled pulse compression and the ability to control duration of the input pulse launched into the PCF. Due to the non-perfect alignment of the prism compressor, the pulse after compression, had residual amounts of spatio-temporal distortions. This resulted in the measured pulse duration of ~ 100 fs (instead of ~ 65 fs) out of the prism compressor. An intensity autocorrelator (FR-103XL, Femtochrome Research Inc., Berkeley, CA) was used for the measurement of the pulse duration.

Light was coupled into the PCF with a coupling efficiency of ~ 40 - 50 % by means of an aspheric lens (Thorlabs, Newton, NJ) having a focal length of 4.5 mm. The measured spectral output of the PCF as a function of the average pump power coupled into the core of the PCF is shown in Fig. 1. The excitation wavelength of 810 nm (~ 100 fs pulse duration) lies within the anomalous dispersion region indicated by the vertical lines. For this measurement, light from the PCF was collected by butt-coupling a multimode fiber at the output end of the PCF and was analyzed using an optical spectrum analyzer (OSA). Fig. 1 shows that when output power is greater than 55 mW, the spectral density is mainly located in two peaks centered in the visible and infra-red (IR) regions accompanied by almost total depletion inside the anomalous dispersion region. The ratio of intensities between the two peaks was found to depend on the efficiency of collection of the light output from the PCF into the multimode fiber into the OSA. With increasing power, the outer edges of the SC shift outward in agreement with observations reported earlier [11]. It was also found that the spectral intensity at 1040nm increased. This new observation is particularly important since this region centered at 1040 nm with a bandwidth of ~ 53 nm was used as the Stokes pulse for CARS imaging of lipids. When spatially and temporally overlapped with the pump pulse at 810 nm (12346 cm^-1) the difference between the pump and Stokes frequency bands was in the range of ~2484 cm^-1 - 2965 cm^-1. The Raman vibrational frequency of the C-H bonds in lipids at 2840 cm^-1 lies in this range as discussed in Section 3 below. Average input power of 300mW at 810 nm was typically used to excite the SC for the CARS microscopy experiment. This corresponds to average power of ~140 mW or peak power of ~ 18 kW for a pulse of ~100 fs duration propagating inside the core of the PCF. Under this condition, average power measured in the Stokes pulse was ~ 10 mW.

 

Fig 1. (Color online) Measured spectral output of the PCF as a function of average power coupled into the core. The vertical straight lines indicate the position of the two ZDWs. Excitation pulse is at 810 nm and is ~ 100 fs wide.

Download Full Size | PDF

The significance of spectral shaping of the SC output of the PCF by changing input pulse parameters has been pointed out in the literature [13, 10]. The possibility of controlling spectral shape of the SC for CARS microscopy by tuning the input pulse wavelength is considered here for the first time. The excitation wavelength was tuned inside the anomalous dispersion region in steps of 5 nm from 810nm to 915 nm, while keeping the pulse duration, average input power and coupling efficiency constant.

 

Fig. 2. (Color online) Dependence of SC on pump wavelength. Output power in the PCF is constant at 140 mW and pulse duration is ~ 100 fs.

Download Full Size | PDF

Figure 2 illustrates the raw experimental data at representative wavelengths of 810 nm, 850 nm and 880 nm. Note that the spectral intensity changes at different excitation wavelengths are not absolute since the coupling of output power from the PCF into the collection fiber changed in between data sets. It is evident from Fig. 2 that the double peaked spectrum is maintained when the input pulse wavelength is increased. However there is a significant increase in the energy content of the IR peak, especially at ~ 1040 nm when pump wavelength is at 880 nm. This experimentally confirms for the first time the predictions of results reported earlier [11] for the dependence of SC output on pump wavelength, which were based only on simulations. When average input power of ~ 300 mW is used, the average output power measured in a bandwidth of ~ 53 nm centered at 1040 nm is ~ 20 mW at 850 nm and 23 mW at 880 nm as opposed to 10 mW at 810 nm. Note that the frequency difference between the Stokes band around 1040 nm and pump pulse at 850 nm and 880 nm, approximately equals the Raman shifts of 2100 cm^-1 and 1650 cm^-1 respectively. This result is particularly favorable for CARS microscopy, since this would enable imaging of aggregates of deuterated lipids (at 2100 cm^-1) and proteins (1650 cm^-1) corresponding to vibration frequencies of the C-D and amide I vibration bands, respectively.

 

Fig. 3. (Color online) Dependence of SC on frequency chirped (indicated as up and down chirp) pulses. Pump wavelength is at 810 nm and output power of the PCF is 140 mW.

Download Full Size | PDF

The SC output did not change significantly when the input pulse duration was varied. This experimentally confirms the theoretical results reported earlier for such dependence [11], which were based primarily on simulation. However as shown in Fig. 3 and in contrast to simulation results in Ref. 15, our experimental results indicate that there is no significant difference in the spectral output of the PCF when the pump pulse is positively (up chirp) or negatively (down chirp) chirped. The chirp values were estimated using a standard formula [16] and varied by changing the distance between the two prisms in the prism compressor. A pulse with duration of ~100 fs launched at the input of the PCF was found to acquire a positive chirp [11] and resulted in a measured pulse duration of ~185 fs after propagation through the PCF. A half wave plate positioned in front of the focusing lens allowed control of the polarization of the input pulse injected into the PCF. A slight redistribution of spectral intensity occurred within 45 degrees for every 10 degree of rotation in polarization. The polarization was adjusted for obtaining maximum power in the Stokes band for the CARS microscopy experiments. This dependence on polarization can be attributed to the appreciable amount of birefringence induced in the PCF due to slight irregularities in the diameter, position, pitch and shape of the air holes inside the cladding of the PCF [17]. The output from the PCF was found to be slightly polarized as well.

The process of SC generation in this PCF with two closely lying ZDWs has been explained by Hilligsoe and colleagues [11] as follows. Self-phase modulation is the dominant broadening mechanism and provides seed wavelengths for four-wave mixing which leads to efficient depletion of spectral intensity between the two ZDWs. Additional non-degenerate four wave mixing also contributes to extending the depleted region between the main peaks beyond the region between the ZDWs [11]. The role of four-wave mixing in the spectral broadening was questioned by Frosz et al in a subsequent paper [14] where they demonstrated the role of generation of dispersive waves followed by soliton self frequency shift (SSFS) when the separation between the ZDWs is large enough [14]. Additional work on the fundamental mechanisms of SC generation in such fibers is necessary to fully understand the discrepancies in these studies. However, since the anomalous dispersion region is very narrow (165 nm) in the PCF considered in this work, we believe that the theory of soliton generation does not apply [14] and that a combination of self-phase modulation and four-wave mixing moves most of the power out of the anomalous dispersion region. In the absence of soliton fission in the PCF with two closely lying ZDWs, the SC contains much less noise compared to SC from conventional PCFs with 1 ZDW.

Noise characteristics of SC generated in a PCF with 1 ZDW have been extensively studied in the literature [8-10]. The significant amplitude noise, well in excess of shot noise, has been shown to be comprised of two components: a broadband or white-noise component, and a low-frequency component. The low-frequency component of the amplitude noise results from amplification of the technical noise (i.e., all noise excluding shot noise) on the input Ti:sapphire laser pulse, and can be reduced through the use of a highly stabilized pump laser [8]. In contrast, the substantial white-noise component of the SC results predominantly from the amplification of the quantum shot noise on the input laser pulse and thus represents a fundamental limit to the SC stability [9]. The amplification of the input pulse shot noise has a strong dependence on pulse energy, pulse duration and wavelength. Its physical origin lies in the extreme sensitivity of the spectral broadening mechanism involving the process of soliton fission in particular, to input pulse noise [10]. The noise in the SC in a PCF with 1 ZDW was experimentally measured by Corwin et al. [8, 9]. The spectrally filtered, 8 nm bandwidth of the SC was directed to either an IR or a visible photodetector. The resulting electrical signal was fed into an electrical spectrum analyzer, where the RF noise power above the detector noise floor was measured. The typical RF noise power in the raw data was found to be ~ 45 dB below the signal power level. The relative intensity noise (RIN) deduced from such a measurement corresponded to pulse-to-pulse amplitude fluctuations of 7% - 50% depending on the values for the input pulse energy and chirp.

A noise measurement similar in procedure to that described above was performed by Hilligsoe et al. [11] for a PCF with two closely lying ZDWs. For every 2 nm bandwidth of the SC in the range of 600 nm to 700 nm, the signal power at 76 MHz was compared to the maximum noise feature in the spectrum. The spectra were however seen to be flat and without noise over the entire frequency range. An upper limit on the noise level was established to be at least 60 dB below the signal power at all wavelengths. This lower noise level in the SC compared to that in the PCF with 1 ZDW is consistent with the fact that the SC generation process in PCF with two ZDWs does not involve soliton fission and hence no amplification of shot noise comes into play. This result has important implications for laser scanning microscopy and in particular, CARS microscopy. In most applications involving imaging of biological samples, the low signal levels dictate that pixel dwell time is typically ≥ 1 μs. This implies that any fluctuations in the intensity of the excitation light in the low frequency region (≤ 1 MHz) will be detrimental to image quality. The effect of the light source fluctuations can be overcome to a limited extent by increasing the pixel dwell time or by averaging over many frames. However this is not a practical approach for applications involving imaging of dynamic processes in live biological samples. The quieter light source involving the PCF with two closely lying ZDWs is thus ideal for CARS microscopy with short data acquisition times as proven in Section 4 below.

3. CARS microscopy setup

A schematic of the experimental setup of our CARS microscope is shown in Fig. 4. Light from a Ti:sapphire laser producing ~ 65 fs pulses at 80 MHz repetition rate was split into two arms marked as pump and Stokes in Fig. 4. A Faraday isolator (Electro-Optics Technology, Inc., Traverse City, MI) was included immediately at the output to reduce back reflections into the laser. The pump arm (degenerate with the probe) proceeds to the sample via a variable delay line. The pump pulse acquires a temporal chirp of ~300 fs after passing through the Faraday isolator. Its measured bandwidth was ~12 nm (~180 cm^-1). In principle this would imply low spectral resolution, since typical sample Raman line widths in the “fingerprint” region are ~ 10 - 15 cm^-1 [1]. This would reduce the resonant to non-resonant CARS signal for imaging in such bands [1]. However in the case of lipid rich samples, such as myelin or adipocyte cells, the effective band width of the band of Raman vibrational frequencies in the CH vibrational region (2,800-3000 cm^-1) is large (~130 cm^-1) [3, 18]. This wide spectral shape is indicative of the chemical composition which includes many vibrational bands, including the aliphatic CH₂ and the vinyl CH stretches [3]. Thus, in spite of the large pump pulse bandwidth, we were able to obtain very large resonant to non-resonant CARS signal (20:1) as evident from the images in Section 4 below. In order to extend the CARS imaging to the fingerprint region, we will resort to chirped CARS or spectrally compressed chirped CARS microscopy for reducing the spectral width of the pump pulse [6, 19].

 

Fig. 4. (Color online) Schematic of the CARS microscope setup. PCF- Photonic Crystal Fiber with two ZDWs; Iso - Faraday Isolator; DM- Dichroic mirror; Com -Prism compressor; SM - Scanning mirror assembly; HW - Half wave plate; BP - Band Pass filter; BS - Beam Splitter; Obj - Objective; PMT - Photo Multiplier Tube; Fiber - multimode fiber to guide the CARS signal into the PMT.

Download Full Size | PDF

The second arm (Stokes) consists of a pulse compressor and the PCF with two closely lying ZDWs as described in Section 2 above. Recompression of pulses was performed by a prism compressor before the pulse was injected into the PCF. This compensated for the dispersion due to the Faraday isolator. The SC output from the PCF was recollimated and filtered (Chroma Technologies, Brattleboro, VT) to produce the Stokes pulse centered at ~ 1040nm with a bandwidth of ~ 53nm. This particular frequency band was chosen so that the difference between the pump and Stokes frequencies corresponds to the Raman vibrational frequency (~ 2800 to 3000 cm^-1) of the C-H bonds associated with lipid-rich structures such as myelin. The input pulse at 810nm, 300mW average power and pulse duration of 100 fs when launched into the PCF generated about 10 mW of average power in the Stokes pulse after bandpass filtering. The measured spectra corresponding to the pump and Stokes pulses are shown in the insets in Fig. 4. The pump and Stokes beams were combined and spatio-temporally overlapped inside the sample using a dichroic mirror and a home built laser scanning microscope system. This system utilizes an XY pair of galvanometer mounted mirrors and relay lenses in the standard optical configuration of a laser scanning microscope [20]. The Zeiss Plan Neofluar (40x, 1.3 NA) oil immersion microscope objective used to focus the pump and Stokes light has about 15% transmission at these wavelengths. This resulted in a measured average power at the sample of ~ 15 mW (0.2 nJ pulse energy) for the pump beam and ~ 0.5 mW (0.006 nJ) for the Stokes beam. The forward-directed CARS signal was collected by an Olympus 40x, 0.8 NA water immersion, microscope objective and was filtered by a 700nm short pass filter (Chroma Technologies). This light was coupled into a multimode fiber and detected by a photomultiplier tube.

4. CARS imaging results

As mentioned above, positive dispersion in the Faraday isolator temporally stretches the laser pulse so that the pump pulse duration is ~ 300 fs. In the Stokes arm, an input pulse that is ~ 100 fs wide acquires a slightly positive chirp after emerging from the PCF such that the Stokes pulse duration is ~ 185 fs. This configuration involving the overlap of a temporally narrow, broad band Stokes pulse with a temporally wide, narrower band pump pulse is similar to the configuration of chirped pulse CARS [19]. When the time delay between the pump and Stokes pulse is varied, different frequencies from the pump bandwidth are involved in the CARS generation process, resulting in a shift of the center wavelength in the CARS spectrum. For CARS imaging, the pump and Stokes pulses were temporally overlapped by finding time zero using the delay line. CARS signal from the oil used for oil immersion objective was optimized by tuning the optical alignment in the beam paths. The ratio of the resonant CARS signal obtained from oil to the non-resonant CARS signal obtained from glass is > 55:1.

Some proof of principle CARS images taken in the forward direction from various types of samples are presented in Fig. 5. The forward CARS image of isolated unstained live dorsal root axons from rat obtained in about 5.5 seconds is shown in Fig. 5(a). Bright parallel bands seen in the image correspond to large resonant CARS signal from myelin that surrounds the axons which are ~ 10 μm in diameter. Although the axial resolution in the image can be further improved, interesting features such as the node of Ranvier and a Schmidt-Lanterman incisure [21] indicated by a line arrow and a block arrow, respectively, can be still be seen along the myelin in the bottom half of this image. Fig. 5(b) shows the unaveraged CARS image of lipid droplets seen as bright circular spots inside unlabeled 3T3 L1 cultured adipocyte cells. The single frame collection time for the 256×256 pixels image was ~ 5.5 seconds. Figs. 5 (c and d) are forward CARS images of sebacious gland (c) and fat-producing adipocyte cells (d) about 20-30 μm in size, taken at different thicknesses at a depth of at least 100 µm inside the tissue of a mouse ear. Fig. 5(e) is an image of the same adipocyte cells as in (d), but with the Stokes signal blocked. A weak signal probably dominated by two photon excitation fluorescence from endogenous fluorophores in tissue surrounding the fat cells, is evident in the image. This also confirms that the signal in Fig. 5 (d) is mainly due to CARS. The note worthy features in all of these CARS images are that the pixel dwell time is very short at ~ 84 μs per pixel and that the signal originating from lipid rich regions is typically found to be a factor of 10 - 20 greater than background levels.

 

Fig. 5. Proof-of-principle forward - CARS images at Raman shift of 2840 cm^-1 from (a) isolated unstained live rat dorsal root axons (b) lipid droplets in unlabeled 3T3 L1 adipocyte cell culture (c) sebaceous gland (in pseudo colour) in a mouse ear at a depth of ~ 40 μm and (d) adipocyte cells at a depth of 100 μm in a mouse ear imaged in the forward mode through several hundred microns of tissue. (e) same as in (d) but with Stokes beam blocked.

Download Full Size | PDF

5. Discussion and conclusion

A practical implementation of a cost effective, low noise source for CARS microscopy with short data acquisition times, is demonstrated in this paper. It is based on using a single femtosecond Ti:sapphire laser source and makes use of a PCF with two closely lying ZDWs to produce the Stokes pulse for CARS microscopy. Unlike the PCF with 1 ZDW or two widely separated ZDWs employed in earlier work [4-7], the SC generated in the PCF with two closely lying ZDWs does not involve soliton fission and is hence inherently more stable. The special dispersion profile of this type of PCF enables generation of very stable, low noise Stokes pulses, resulting in high contrast CARS images with short acquisition times of 84 μs per pixel. In contrast, the SC generated in a PCF with one ZDW exhibits higher noise content [8,9] at low frequencies that are important for CARS imaging.

The particular parameters such as wavelength, average power, pulse duration and polarization for input light pulse launched into the PCF, were determined with the aim of generating optimum Stokes pulses for CARS microscopy. We showed that it is possible to control the spectral shape of the SC by tuning the pump wavelength of the input pulse and that the energy content of the Stokes pulse filtered out of the IR part of the SC increased with increase in pump wavelength. Due to the single source based design of our CARS microscope, increasing the pump wavelength up to ~ 915 nm enables access to Raman shifts in the “fingerprint” region below 1800 cm^-1, up to about 1200 cm^-1.

Although not shown here, application of this PCF with two closely lying ZDWs for CARS multiplex spectroscopy [4 - 7, 19] is relatively straightforward. For example, the spectrally narrow pump pulse can be generated by means of spectral compression of a chirped pulse [6]. In that case, the whole IR band of the SC interacts with the spectrally narrow pump pulse to generate multiplex CARS spectra.

Proof-of-principle CARS images at ~ 2840 cm^-1 Raman shift, obtained with our CARS setup demonstrated high signal to background levels. Signal originating from lipid rich regions was typically found to be a factor of 10 - 20 greater than background levels. The non resonant background excited due to the broad bandwidth pump does not limit CARS imaging of lipid rich biological materials as demonstrated in this report. This is due in part to the fact that the Raman active CH vibrational region in lipid-rich structures has a large effective band width of ~130 cm^-1 [3, 18].

Another application involving this PCF with two closely lying ZDWs will be very useful to explore in conjunction with CARS microscopy. The IR or visible part of the SC spectrum can be used as a low noise source for two-photon excitation fluorescence (TPEF) of exogenous fluorophores such as green fluorescence protein [22] or several endogenous fluorophores such as elastin, NADH and aromatic amino acids like tryptophan, tyrosine and phenylalanine [23]. This can be combined with CARS microscopy as well as other nonlinear optical techniques such as second harmonic generation (SHG) and sum frequency generation (SFG) for multi-nonlinear optical imaging [21, 22]. In conclusion, the high-contrast low noise CARS imaging capability demonstrated in this paper could be fairly readily adapted to most currently installed two-photon excited fluorescence microscopy systems, resulting in major cost savings. Such an instrument could extend the two-photon fluorescence excitation capabilities of conventional systems to include simultaneous label-free, non-bleaching and chemically selective CARS imaging of fixed and live biological samples.