2.1.

Picosecond Laser with High-Speed Wavelength Scanning System

We found that the development of an ideal tunable laser source for realization of fast spectral CRS microscopy still remains a challenge. We decided to construct a CARS system with two picosecond mode-locked Ti:sapphire lasers, one of which was a high-speed tunable laser (Mega Opto Co. Ltd., Saitama, Japan). The high-speed tunable laser was equipped with an AOTF inside its cavity for tuning the wavelength electrically.³¹ The AOTF was a piezoelectric device made from a birefringent crystal that alters the diffracted wavelength of incident light by applying suitable radio frequency waves without the need for any mechanical parts.³² Therefore, the laser could be rapidly tuned from 800 to 940 nm with a typical pulse duration of 10 ps within $1\mathrm{ms}/\text{wavelength}$ change. An average laser power $>300\mathrm{mW}$ was high enough for obtaining CARS images with multifocus scanning.¹⁵

Figure 1 shows a schematic diagram of the modified AOTF laser for synchronization with another picosecond mode-locked laser. A prism pair was provided to passively reduce the dispersion of the laser cavity. An end mirror attached on a piezoelectric actuator and a pair of parallel plates driven by galvanometer motors were provided to actively compensate for changes in the repetition frequency.

Fig. 1

Modified acousto-optic tunable filter (AOTF) laser with a moveable end mirror attached on a piezoelectric actuator and a pair of parallel plates on galvanometer motors.

The repetition frequency of a mode-locked laser depends on the cavity length. During the wavelength scanning of the AOTF laser from 800 to 940 nm, the repetition frequency varied greatly, by a value of 44 kHz. This mainly occurred due to the refractive index dispersion of the AOTF crystal, even though a prism pair (SF-10) was placed inside the laser cavity. The relationship between the repetition frequency, $f$, and the laser cavity length, $L$, is given by $f=c/2L$, where $c$ is the speed of light. A large change in the repetition frequency leads to difficulty in pulse synchronization; thus, compensation is required to maintain the repetition frequency.

We attached the end mirror of the laser cavity on a piezoelectric actuator (Model 840.10, 15 μm, Physik Instrumente GmbH & Co.KG, Karlsruhe, Germany) to compensate for small changes in the repetition frequency and to synchronize both lasers. Unfortunately, its controllable range of 300 Hz was not enough for compensation of the 44-kHz repetition frequency change. To achieve larger compensation, we placed a pair of parallel plates (BK7, $12\times 4.3\times 10{\mathrm{mm}}^3$) mounted on galvanometer motors (LSA-20B-30, Harmonic Drive Systems, Tokyo, Japan) inside the laser cavity. The parallel plates were rotated in opposite directions by the same angular increment, which was rapidly and precisely controlled by the galvanometer motors. This configuration allowed repetition frequency compensation of 65 kHz, which was sufficient for our purposes.

2.2.

Synchronization System for Picosecond Mode-Locked Lasers

Both laser pulses in CARS microscopy must spatio-temporally overlap. To achieve spatial overlapping, one can adjust the beam pathways in the optical setup. The degree of temporal overlapping is determined by timing jitter during synchronization of both laser pulses. Reduction of timing jitter is very important for improving the signal-to-noise ratio of the CARS image. We employed two types of timing-jitter detection systems to synchronize the lasers: an electronic phase detector for coarse control of synchronization and a balanced cross-correlator for fine control (Fig. 2).^33,34

Fig. 2

Synchronization system: timing jitter detection system, proportional-integral-differential (PID) and digital signal processor (DSP) for feedback control to galvanometer motors and piezoelectric actuator controllers in the AOTF laser.

The output from the electronic phase detector represented picosecond-order temporal differences between the two laser pulses.³³ This signal was fed back to the galvanometer motors via a digital signal processor (DSP) controller (C6713DSK, Texas Instruments Inc., Texas) to rotate the pair of parallel plates to compensate for variations in the repetition frequency caused by wavelength scanning and to synchronize both the lasers. The electronic phase detector output depended on the time difference and amplitudes of the two input pulses. Therefore, we introduced an autogain amplifier (AD8368, Analog Devices Inc., Boston) after each photodetector (Fig. 3) to keep the amplitudes stable. The autogain amplifier generated a 224 mV rms electrical output signal for any photodetector signal with an amplitude $>63\mathrm{mV}$ rms.

Fig. 3

Electronic phase detector with autogain amplifier. PD1, $\mathrm{PD}2=\text{photodetectors}$; L1, $\mathrm{L}2=\text{lenses}$.

The output from the balanced cross-correlator obtained by optical detection was used for fine control of synchronization because it could detect the femtosecond-order temporal differences between the two laser pulses.³⁴ Since the stability of the balanced cross-correlator output highly depended on the balance of cross-correlation signals detected by two-photon detectors, it was necessary to conserve equal separation of the pump and Stokes pulses to both detectors. As shown in Fig. 4, a polarizing beam splitter (PBS052, Thorlabs Inc., New Jersey) together with two achromatic half-wave plates (AHWP05M-980, Thorlabs Inc., New Jersey) enabled us to easily control and maintain the balance of the transmitted and reflected intensities of the incident pulses so that the balanced cross-correlator could minimize the timing jitter significantly. The output from the balanced cross-correlator was fed back to the piezoelectric actuator via an analog proportional-integral-differential (PID) controller (T-PID01Z, Turtle Industry Co. Ltd., Ibaraki, Japan) and to the galvanometer motors.

Fig. 4

Balanced cross-correlator with polarizing beam splitter and half-wave plates. $\mathrm{PBS}=\text{polarizing beam splitter}$; HWP1, $\mathrm{HWP}2=\text{half-wave plates}$; TPD1, $\mathrm{TPD}2=\text{two-photondetectors}$; OL1, $\mathrm{OL}2=\text{objective lenses}$; MP1, $\mathrm{MP}2=\text{dichroic mirrors}$ with highly reflective mirrors.

2.3.

Fast Spectral CARS Imaging System

We developed a fast spectral CARS microscopy system by adopting a multifocus CARS configuration with a microlens array scanner and used the AOTF laser described above as one of the laser sources in the multifocus CARS imaging system (Fig. 5). Details of the multifocus microscopy system have been described in our previous article.³⁵ The system covers the Raman spectral range from 400 to $2260{\mathrm{cm}}^{-1}$ with a pump beam at 775 nm and from 1600 to $3470{\mathrm{cm}}^{-1}$ with a pump beam at 709 nm.

Fig. 5

Fast spectral coherent anti-Stokes Raman scattering (CARS) imaging system with microlens array and high-speed tunable laser.

Personal computers (PC 1 and PC 2) were mainly used for controlling the AOTF laser operation and image acquisition, respectively. Before starting spectral imaging, the AOTF laser pulses and the picosecond mode-locked laser pulses should be synchronized. This synchronization was conducted by employing a phase-locked loop mechanism using the electronic phase detector and the balanced cross-correlator as parts of the pulse synchronization system. It was also necessary to store proper amplitude and frequency values of the RF wave in the AOTF driver for determining the wavelength change of the AOTF laser. During the wavelength scanning, PC 1 sent trigger pulses to the AOTF driver to start the wavelength changes. Wavelength changes were performed rapidly within 1 ms or less, and the pulse synchronization system attempted to recover the synchronization within 20 ms. Then, the DSP sent a trigger signal to an electron multiplying CCD (EM-CCD) camera (Ixon, Andor Technology Plc, Belfast, United Kingdom) for image acquisition, and the images were stored in PC 2. The total time required to complete imaging at every wavelength change should include both the image acquisition and the resynchronization process. The wavelength change, the resynchronization, and the image acquisition processes were repeated until the scanning process was completed.

3.1.

Fast Spectral CARS Imaging of Lipid Droplets in 3T3-L1 Adipose Cells

We demonstrated the performance of our CARS microscopy system for imaging at rates faster than video rates ($\sim 33\mathrm{ms}/\mathrm{frame}$). We tuned the Raman vibration to $2850{\mathrm{cm}}^{-1}$ for imaging lipid droplets in 3T3-L1 adipocytes and acquired CARS images at $10\mathrm{ms}/\text{frame}$. The sample stage was moved during imaging. CARS images observed at some time points are shown in Fig. 6. Movie 1 presents a recording of a 100-fps imaging process, showing timestamps.

Fig. 6

(a)–(d) CARS images of lipid droplets in 3T3-L1 adipocytes from different locations taken at $2850{\mathrm{cm}}^{-1}$ (Movie 1, QuickTime, 1.1 MB) [URL: http://dx.doi.org/10.1117/1.JBO.18.9.096009.1]. $\text{Acquisition rate}=100\mathrm{fps}$ ($\text{exposure time}=9\mathrm{ms}/\mathrm{image}$). $\text{Scale bar}=10\mu \text{m}$. The color scale in Movie 1 is the same as in Fig. 6.

Spectral CARS images of 3T3-L1 adipocytes were observed by scanning the wavelength of the AOTF laser every 0.5 nm from 875.0 to 905.0 nm, which corresponded to a Raman shift of 2663.21 to $3049.41{\mathrm{cm}}^{-1}$. For a 100-ms exposure time, the imaging process requires $120\mathrm{ms}/\text{image}$ to cover the wavelength change, laser resynchronization after the wavelength change, and image acquisition. To produce final images, we removed the contributions of dark noise and the fundamental lights by subtraction. We also considered intensity fluctuations of the AOTF laser during wavelength scanning and took this into account by division. Figure 7 shows spectral CARS images of 3T3-L1 adipocytes at (a) $2663.21{\mathrm{cm}}^{-1}$, (b) $2845.31{\mathrm{cm}}^{-1}$, and (c) $2896.26{\mathrm{cm}}^{-1}$, which were assigned to off-resonant vibration, ${\mathrm{CH}}_2$ symmetric stretching vibration, and ${\mathrm{CH}}_3$ stretching vibration, respectively. Movie 2 presents a recording of spectral CARS images, showing the Raman shift. A CARS spectral profile in Fig. 7(d) was reconstructed by averaging the intensity in a $5\times 5$-pixel area of interest in Movie 2 [indicated by white arrow in Figs. 7(a)–7(c)]. The reconstructed CARS spectral profile included both resonant and nonresonant contributions and it agrees with the results in previous reports.^25,41,42 However, lipid abundantly contained CH bonds and they were highly accumulated in the lipid droplets. It led to the generation of large CARS signal that was relatively higher than the nonresonant background.

Fig. 7

Spectral CARS images of 3T3-L1 adipocytes at (a) $2663.21{\mathrm{cm}}^{-1}$, (b) $2845.31{\mathrm{cm}}^{-1}$, and (c) $2896.26{\mathrm{cm}}^{-1}$, assigned to nonresonant vibration, ${\mathrm{CH}}_2$ symmetric stretching vibration, and ${\mathrm{CH}}_3$ stretching vibration, respectively, taken with fast spectral CARS imaging system. $\text{Exposure time}=100\mathrm{ms}/\mathrm{image}$. $\text{Total acquisition time}=120\mathrm{ms}/\mathrm{image}$ included the wavelength change, laser resynchronization after the wavelength change, and exposure time (Movie 2, QuickTime, 435 kB) [URL: http://dx.doi.org/10.1117/1.JBO.18.9.096009.2]. (d) CARS spectral profile reconstructed from area of interest (averaged over $5\times 5$ pixels) indicated by white arrow in (a)–(c). $\text{Scale bar}=20\mu \text{m}$.

3.2.

CARS Imaging by Fast-Switching Between CD and CH Stretching Bands

Intracellular lipids can be easily detected using CARS microscopy because they contain abundant CH bonds, which have a strong Raman vibration around $2850{\mathrm{cm}}^{-1}$. Lipids cover a large group of biochemical structures with various functions and complex metabolism profiles. Therefore, it is important to distinguish lipid subspecies. One potential solution for distinguishing them is to use a deuterated lipid that consists of CD bonds instead of CH bonds on its carbon chain. A deuterated lipid has a Raman vibration that is significantly downshifted to the silent region around $2100{\mathrm{cm}}^{-1}$ where no interference from biological specimens occurs.^11,43–47

A stable, rapid switching process between two or more different wavelengths is needed for fast imaging to obtain the CARS images of different structures at various Raman shifts. We observed a sample of colocalized deuterated and nondeuterated stearic acids for a rapid switching experiment using our fast spectral CARS microscopy system. Each fatty acid was separately prepared as a stock solution of 40 mM in methanol. First, 5 μL of the deuterated stearic acid solution was dropped onto a glass-based dish (12 mm, AGC Techno Glass Co. Ltd., Shizuoka, Japan). The solvent easily evaporated in the room temperature and the deuterated stearic acid crystals were formed. Then, 5 μL of the nondeuterated stearic acid solution was added. With such small amount of solution, the solvent of the added solution was inadequate to dissolve the deuterated stearic acid crystals because it quickly evaporated. Therefore, we were able to obtain the colocalized deuterated and nondeuterated stearic acids crystals. This sample enabled us to detect sufficiently high Raman signals of ${\mathrm{CH}}_2$ stretching and ${\mathrm{CD}}_2$ stretching at 2850 and $2100{\mathrm{cm}}^{-1}$, respectively [Figs. 8(a) and 8(b)]. For that purpose, the AOTF laser was rapidly switched between 833 and 888 nm, and CARS images were acquired within $520\mathrm{ms}/\text{image}$. The merged CARS image indicated the locations where the deuterated and nondeuterated stearic acids were independently and colocatedly recrystallized [Fig. 8(c)]. Hence, the CD bond serves as a powerful chemical tag in CARS microscopy. For instance, in the detection of lipid uptake metabolism by cells, the exogenous lipid can be prepared using its deuterated isotope. The exogenous lipid should be easily detectable because naturally no deuterated isotope presents inside the cells. Thus, it is possible to observe lipid dynamics and metabolism in cells.

Fig. 8

Fast-switching CARS images of recrystallized sample of deuterated and nondeuterated stearic acid mixture at (a) $2100{\mathrm{cm}}^{-1}$ and (b) $2850{\mathrm{cm}}^{-1}$, assigned to ${\mathrm{CD}}_2$ and ${\mathrm{CH}}_2$ symmetric stretching vibrations. $\text{Exposure time}=500\mathrm{ms}/\mathrm{image}$. $\text{Total acquistion time}=520\mathrm{ms}/\mathrm{image}$ included the wavelength change, laser resynchronization after the wavelength change, and exposure time. (c) Merged CARS images from (a) and (b). (d) Spontaneous Raman spectra of each fatty acid. $\mathrm{Red}=\text{deuterated stearic acid}$, $\mathrm{green}=\text{nondeuterated stearic acid}$. $\text{Scale bar}=20\mu \text{m}$.