Nonlinear optical properties of biological macromolecules have in recent years been exploited as contrast mechanisms in microscopy.¹ Probed by multiphoton scattering events, e.g., second-harmonic generation (SHG) and coherent anti-Stokes Raman scattering (CARS), they enable label-free chemical imaging of living cells with high 3-D resolution. Assemblies with noncentrosymmetric molecular arrangements, e.g., fibrillar polymers and structural proteins are known to be SHG-active,² and collagen content in tumors,³ cellulose fibers in artificial blood vessels,^{4, 5} and the detailed arrangement of myosin filaments in muscle tissues,⁶ have successfully been monitored by SHG microscopy. In CARS microscopy,⁷ contrast is achieved by probing molecular bonds via Raman-active vibrational transitions, such as carbon-hydrogen bonds in the acyl chains of lipids, resulting in a strong signal. Accumulation in preadipocytes,⁸ storage in single- and multiple-cell organisms,^{9, 10} and free fatty acids in cancer-cell membranes,¹¹ are all examples of lipid studies proving the usefulness of CARS microscopy. Thus, multiple highly relevant applications of nonlinear microscopy have been brought forward in scientific literature. Still, most studies are conducted by a few research groups world-wide that have the competence and advanced equipment required. This is particularly evident for CARS microscopy, involving the coupling of two high-power, short-pulsed laser beams into the microscope with the requirement of precise spatial and temporal overlap in the probe volume.

Early experimental configurations for CARS microscopy were based on temporal synchronization of two separate femtosecond (fs) or picosecond (ps) laser sources.⁷ By the launch of a broadly tunable optical parametric oscillator (OPO), pumped by a high-power solid-state laser,¹² inherent synchronization of two picosecond beams was achieved, resulting in a more robust laser source. Still, regular adjustment of the OPO is required. Clearly, to be able to fully focus on the biological problems addressed, there is a great need for laser sources offering simpler operation. One promising route is the use of fiber-based lasers, which require less alignment of the laser itself for operation, deliver high quality laser beams, and can be built more cost effectively. The first attempts toward fiber-based CARS microscopy used a photonic-crystal fiber (PCF) for replacement of the tunable laser source,^{13, 14} though still requiring a conventional femtosecond laser for pumping the fiber. Another semifiber laser concept was based on a Yb-doped fiber oscillator in series with a cladding-pumped gain fiber emitting enough power to pump a tunable OPO.¹⁵ Both concepts require optical alignment beyond the competence of a nonspecialist. A first all-fiber-based system was presented by Andresen using a gain fiber laser pumping a PCF, allowing for CARS microspectroscopy but with powers too low for imaging.¹⁶ Recently, CARS spectroscopy and microscopy were demonstrated by Marangoni ¹⁷ and Pegoraro ¹⁸ using spectral compression of the femtosecond output from all-fiber laser systems. This approach increases the spectral selectivity for femtosecond pulse CARS, though including more external optics and lowering the overall power.

In this work we present a compact laser system composed of a femtosecond Er-doped fiber source. Without external pulse compression, we show that the femtosecond output alone allows for chemically specific CARS imaging in a straightforward way with a minimum of free-space optics. The laser system emits high enough power for simultaneous CARS and SHG imaging, i.e., multimodal nonlinear microscopy, here demonstrated for the important biological model systems yeast and C. elegans. This extension of a recently launched commercial single-beam laser source has proven useful for two-photon fluorescence microscopy.¹⁹ The frequency-doubled fundamental of the Er-doped pump laser, together with the output of a pumped highly nonlinear fiber (HNLF), allows for different combinations of nonlinear microscopy schemes. The CARS and SHG imaging capabilities of the femtosecond Er-doped fiber laser system are directly and quantitatively compared to the performance of the widely used picosecond Nd:Van laser-pumped OPO system.

The two independent femto- and picosecond laser systems were combined with a single microscope unit. The compact femtosecond light source (FFS series, Toptica Photonics AG, Graefelfing, Germany) is based on a mode-locked Er-doped fiber oscillator that generates $100\text{-}\mathrm{fs}$ pulses at $1.56\mu \mathrm{m}$ with a repetition rate of $82\mathrm{MHz}$ . Two fiber amplifier stages seeded by the same oscillator generate two perfectly synchronized output beams for the CARS experiment: the frequency-doubled output with $80\text{-}\mathrm{mW}$ average power at $782\mathrm{nm}$ serves as a pump/probe beam, and the tunable IR continuum $\left(990\text{to}1100\mathrm{nm}\right)$ generated in the HNLF, with $20\text{-}\mathrm{mW}$ output power, serves as a Stokes beam. The picosecond laser system, consisting of a solid state pump laser and an OPO, is described in more detail in Enejder, Brackmann, and Svedberg.²⁰ Briefly, the system consists of a Nd:Van laser (picoTRAIN, High-Q Laser GmbH, Bregens, Austria) that delivers $7\text{-}\mathrm{ps}$ pulses at $1064\mathrm{nm}$ with a repetition rate of $76\mathrm{MHz}$ . A fraction of the $1064\text{-}\mathrm{nm}$ laser beam constitutes the Stokes beam, while the rest is used to synchronously pump a ring-cavity OPO (Levante, APE GmbH, Berlin, Germany) that provides the pump/probe beam tunable in the range of $785\text{to}845\mathrm{nm}$ . The femto- and picosecond laser beams, initially following separate beam paths, were aligned into the same inverted microscope (Nikon TE-2000) via the beam scanning unit (Nikon C1) and focused on the sample by the objective (Nikon Plan Fluor $40\times$ , NA 1.30). A flip-mounted mirror positioned in front of the scanner allowed for sequential femto- and picosecond imaging. The signal was collected by an aspherical lens (NA 0.68) in the forward direction and detected by a photomultiplier tube (PMT) (Hamamatsu R6357) equipped with bandpass filters, allowing on-and-off resonance measurements in the CH region using both systems. The powers of the two laser systems were adjusted to give approximately the same CARS signal level on the PMT for a given detector voltage, and powers measured before the microscope were around $5\mathrm{mW}$ and $300\mathrm{mW}$ for the femto- and picosecond beams, respectively. The powers at the sample position are typically 10% of these values due to attenuation in the microscope. For excitation of the $\mathrm{C}{\mathrm{H}}_2$ stretching frequency at $2845{\mathrm{cm}}^{-1}$ with both laser systems, the femtosecond continuum was centered at $1007\mathrm{nm}$ and the picosecond OPO was tuned to $817\mathrm{nm}$ . The laser line widths of the femtosecond beams are $40\mathrm{nm}$ for the continuum and $5\mathrm{nm}$ for the frequency doubled; the line widths of both picosecond beams are $0.3\mathrm{nm}$ . Intrinsically larger line widths are expected for a laser system of shorter pulse duration; however, the fiber laser continuum also obtains additional spectral broadening from dispersion in the HNLF.

The CARS microscopy capabilities of the fiber laser system are shown in Fig. 1 together with a direct comparison with the picosecond system. Figures 1 and 1 show images of living yeast cells at the $\mathrm{C}{\mathrm{H}}_2$ stretching frequency obtained with the femto- and picosecond systems, respectively, using the specified laser wavelengths. Qualitatively, the images look similar, with the intracellular lipid droplets clearly visible for both cases. Although a higher lipid droplet signal is obtained in the femtosecond image compared to the picosecond image, the latter shows a better signal versus background from the surrounding cell medium. The images differ in contrast by a factor of $\sim 2.5$ , as indicated by the profile below each image. The difference in contrast can be explained by the broader spectral profile of the femtosecond laser, resulting in a higher amount of laser power used for excitation outside the Raman vibrational band, and contributing to a higher unspecific nonresonant background.

Fig. 1

Direct comparison of CARS images of live yeast cells obtained with (a) the femtosecond and (b) the picosecond laser system. Below each image an intensity line trace along the path indicated in the images is shown.

To directly compare the spectral selectivity of the femto- and picosecond lasers, CARS images of a sample composed of tripalmitin crystals and polystyrene spheres were collected. Tripalmitin is a saturated triacylglyceride showing a $\mathrm{C}{\mathrm{H}}_2$ symmetric stretch mode at $2845{\mathrm{cm}}^{-1}$ , while polystyrene has a symmetric stretch vibration of the C–H bonds in the phenyl ring at $3070{\mathrm{cm}}^{-1}$ . Figure 2 shows CARS images collected at the two frequencies using both laser systems. While Figs. 2 and 2 clearly show that the spectrally broad excitation profile of the femtosecond system generates signal from both sample components when tuned to any of the resonances, the picosecond images, Figs. 2 and 2, show a clear on-and-off resonance behavior for both sample components. The influence of laser line width on the spectral selectivity of CARS microscopy is well known, and broadband excitation with femtosecond pulses results in less specific imaging.⁷ Based on the measured laser emission spectra, the femtosecond system probes a $\sim 400\text{-}{\mathrm{cm}}^{-1}$ -wide band of nonresonant CARS signals together with a small resonant contribution, whereas the picosecond system effectively excites the resonant CARS signal with a $\sim 5\text{-}{\mathrm{cm}}^{-1}$ narrow profile inside the range of the probed vibrational Raman band with a line width of $50{\mathrm{cm}}^{-1}$ in the CH region. CARS images of yeast cells obtained with the femtosecond system at 2845 and $3070{\mathrm{cm}}^{-1}$ are shown in Figs. 3 and 3 , respectively. Again the high background in the images is due to the broad spectral signature of the femtosecond system, and the CARS signal in lipid droplets is $\sim 1.25$ times higher on $\mathrm{C}{\mathrm{H}}_2$ resonance compared to off. Thus, the selectivity obtained with the femtosecond system is sufficient for imaging of lipids probing their vibrations of high cross sections.

Fig. 2

CARS images of a mixed sample of polystyrene spheres and tripalmitin crystals measured in the CH region. The upper image pair (a) and (b) is measured with the femtosecond laser system at 2845 and $3070{\mathrm{cm}}^{-1}$ , respectively. The lower image pair (c) and (d) is measured at the corresponding frequencies using the picosecond laser system. Image size, 100×100 μm.

Fig. 3

CARS images of live yeast cells obtained with the femtosecond fiber laser system. Measured at (a) $2845{\mathrm{cm}}^{-1}$ and (b) $3070{\mathrm{cm}}^{-1}$ . Below each image an intensity line trace along the path indicated in the images is shown.

The versatility of the fiber-based laser system also allowed for SHG imaging using the frequency-doubled fiber output at $782\mathrm{nm}$ . Figure 4 shows a SHG image of blood vessels from sheep and Fig. 4 the corresponding image acquired with the picosecond OPO at $817\mathrm{nm}$ . The collagen fibers in the tissue give strong SHG signals, easily generated by both laser systems, and the line traces below the images show that they provide the same amount of detail, although the average excitation power is 60 times lower for the femtosecond system. In contrast to many other fiber lasers, the high output power from the Er-doped fiber even permits multimodal nonlinear microscopy, as shown by the combined CARS and SHG images of C. elegans [Figs. 4 and 4]. This is an important quality, allowing colocalization of structures with different chemical composition and biological functions. The images provide similar qualitative information, but in this more complex organism, it is clear that the limited specificity for CARS imaging of lipids using the femtosecond system [Fig. 4] makes lipid droplet quantification difficult compared to the corresponding picosecond image [Fig. 4]. This is further demonstrated by the intensity profiles shown below the images. Whereas the lipid droplets in the femtosecond CARS image tend to disappear in the background, the SHG signals generated by the myosin of the nematode muscle tissue still result in high image quality. A laser system combining the improved specificity with picosecond pulses and the facilitated operation of a fiber laser has recently been presented, though with low output powers that could be a limiting factor for its use in SHG and multimodal microscopy.²¹

Fig. 4

SHG images of blood vessels from sheep excited by (a) femtosecond pulses and (b) picosecond pulses. Superimposed images of CARS and SHG of C. elegans obtained with (c) the femtosecond and (d) the picosecond system. The red color represents CARS and the blue color represents SHG. (Color online only.)

Although variations in image contrast due to spectral selectivity, no difference in spatial resolution could be detected between the two systems, which mainly is governed by the excitation wavelength and microscope optics, and often in practice, by the pixel dimension. As indicated by the arrows in Fig. 4, the SHG images exhibit fine details of collagen fibers with dimensions of $\sim 400\mathrm{nm}$ , i.e., close to the diffraction limit, and the CARS images clearly show submicron-sized lipid droplets. The SHG and CARS signals probed inside C. elegance nematodes gave similar optical information throughout the sample, indicating no practical difference in penetration depths of the lasers systems.

In conclusion, the femtosecond fiber laser provides a compact and user friendly excitation source for multimodal nonlinear microscopy. Compared to a picosecond system used as a standard for CARS microscopy, the femtosecond system generates images with similar visual information at significantly lower average excitation power. This is beneficial for biological imaging, since photodamage related to linear absorption is minimized while the risk of nonlinear photoinduced changes that benefit from high peak intensities is on the same level as for picosecond excitation.²² The laser powers here corresponds to peak intensities well below those reported to induced photodamage in cultured cells.²³ For CARS microscopy, we demonstrate that limited specificity is achieved, though sufficient for lipid structures in a simple chemical environment, i.e., biological model systems such as lipid bilayer vesicles and yeast cells. A variety of methods can be implemented to increase the signal sensitivity in CARS microscopy by suppressing the nonresonant background.²⁴ In the case of femtosecond excitation, relatively complex time-resolved detection can utilize the different temporal behavior of the resonant and nonresonant contributions to completely remove the nonresonant part, although also significantly attenuating the resonant part.²⁵ Nevertheless, with technical improvements of the laser in continuum tunability, allowing CARS background measurements in the silent region, and filtering of the continuum emission using suitable bandpass filters to reduce nonresonant excitation, the femtosecond Er-fiber laser has the potential to become a future basic excitation source for nonlinear microscopy.