1. Introduction

Multiphoton microscopy is a powerful technique for deep tissue visualization with submicron optical resolution and it offers unique tools for biology and medicine [1,2]. Ultrafast lasers that emit in the near-infrared spectral region are the key enabling technology for multiphoton microscopy due to the high peak power of pulses produced by such systems. The most popular commercially available sources, titanium-sapphire (Ti:S) lasers, are stable, robust and widely tunable (700-1000 nm). However, their significant cost typically limits their use to specialized labs or as shared resources. Additionally, some fluorophores used in multiphoton microscopy are not efficiently excited within the tuning range of Ti:S lasers.

Fiber lasers have emerged in recent years as alternative sources of ultrashort pulses for multiphoton microscopy. Several groups have demonstrated low-cost, home-built mode-locked fiber oscillators and showed that these sources are adequate for multiphoton microscopy [3–5]. However, the homemade sources’ pulse energies and pulse durations remain limiting factors for high-performance deep tissue microscopy. There are two main approaches for high-energy pulse amplification to increase the power of mode-locked oscillators: 1) linear amplification based on the chirped-pulse amplification (CPA) technique and 2) nonlinear pulse amplification [6,7]. In general, there is a tradeoff between pulse duration, pulse energy, and the complexity/robustness of the system. In a typical linear fiber CPA system very high pulse energy is achievable, but the pulse duration is limited to above approximately 200 fs by gain narrowing and residual dispersion mismatch between the stretcher and compressor [6]. In nonlinear amplifiers (mostly based on self-similar pulse amplification or pre-chirp managed amplification) the amplified pulses can be directly compressed below 100 fs [8–11]. However, achieving optimal performance with nonlinear amplifiers often requires balancing several parameters, which increases the complexity of the design and operation [10,12].

Recently, Sidorenko et al. proposed and demonstrated a new short-pulse amplification scheme, termed Gain-Managed Nonlinear (GMN) amplification [13]. In GMN amplifiers, a relatively narrowband seed pulse experiences up to 100-fold spectral broadening, but surprisingly, is still compressible to the transform-limited duration with a standard grating compressor. This is the result of the interplay of nonlinearity, dispersion, and the longitudinally-varying gain spectrum as the pulse propagates. In the final pulse, the effects of nonlinear spectral broadening and dispersion are balanced by gain-shaping. To date, state-of-the-art GMN systems with large mode area Yb fiber have achieved sub-40-fs pulses with energy as high as 1.2 $\mathrm{\mu}$J [14]. The maximum pulse energy is limited by stimulated Raman scattering. The simplicity and robustness of the GMN amplifiers were demonstrated in several applications [15–17].

Here we report an evaluation of the performance of a GMN amplifier for deep-tissue two-photon excited fluorescence (2PEF) imaging in vivo. We present a robust and stable ultrashort pulse laser system that can be built entirely from off-the-shelf commercial components. The source consists of a single amplification stage seeded by a simple fiber oscillator and delivers 58-nJ and 33-fs pulses at 1080 nm. We conducted two imaging experiments to demonstrate the utility of this laser for 2PEF imaging. We imaged individual cells labelled with multiple fluorescent markers with several spectral slices of the broadband amplified pulse and demonstrated higher-purity unmixing of the detected spectral signal than was possible when using a single spectrally narrow excitation source. We also performed in vivo imaging in a mouse that was labelled with multiple fluorescent markers with the GMN amplifier, and we compare the results to those obtained with a chirped-pulse fiber amplifier that has been used successfully for 2PEF imaging in earlier works [18,19].

2. Gain-managed nonlinear amplifier

2.1 Amplifier design

The source is based on the recently demonstrated scheme of the GMN amplifier and includes three major parts (Fig. 1): a homemade mode-locked oscillator, the GMN amplifier, and a pulse compressor [13]. The fiber oscillator has a ring-cavity architecture and operates in the normal dispersion regime [20]. Half-wave and quarter-wave plates control the polarization rotation effect which initiates and sustains mode-locking [21]. The grating (G1) and the collimator (C2) together act as a tunable spectral filter, which is a common design in mode-locked fiber lasers [22,23]. This filter plays a crucial role in suppressing oscillation in the conventional high-gain wavelength window around 1035 nm and forces the oscillator to emit around 1020 nm. Operating the oscillator at 1020 nm is beneficial for seeding the GMN amplifier in the blue part of Yb gain spectrum [13]. While we have found a 1020 nm seed to result in the best performance, seed wavelengths between 1020 nm and 1070 nm may be used. The gain fiber is cladding-pumped with a laser diode at 976 nm coupled to multimode fiber through the pump/signal combiner. A polarization insensitive Faraday isolator ensures unidirectional pulse evolution inside the ring cavity. The polarizing beam splitter serves as the output coupler. The oscillator delivers 250 mW of average power at 31-MHz repetition rate. A more detailed discussion on the construction of such oscillators can be found in the literature [3].

 

Fig. 1. Schematic of the laser system. Pulses are generated in the ring cavity, dissipative soliton oscillator. These pulses are spectrally filtered and seed the GMN amplifier. After amplification, the pulses are compressed with a standard grating compressor. HWP: half-wave plate; QWP: quarter-wave plate; PBS: polarizing beam splitter; G: grating; C: pigtailed fiber collimator; ISO: isolator; BPF: bandpass filter; YDF: ytterbium-doped fiber; LD: laser diode; M: mirror; PM: pick-off mirror.

Download Full Size | PDF

The amplifier is constructed from a pair of pigtailed collimators (WaveSource Photonics PL1060L-R18-E02-AR_40), a pump/signal combiner (Thorlabs PMC1060H2 or equivalent) and 3 meters of cladding pumped Yb-doped fiber with 10-$\mathrm{\mu}$m core and 125-$\mathrm{\mu}$m cladding (Coherent PLMA-YDF-10/125-VIII) (Fig. 1). The gain fiber is co-pumped with a multimode diode at 976 nm (BWT K976AAHRN027.00WN1N-10522B20ENA). In a previous study, Sidorenko et. al. established that the GMN amplifier is not very sensitive to seed parameters and that even variations as large as 50% in seed energy and 40% in seed bandwidth did not affect its performance (as shown in Fig. 7 of [13]). It is important to note that for fast SPM spectral broadening to occur at the beginning of the GMN pulse evolution, the seed pulse should be close to the transform-limited duration. More recently, a comprehensive study of the effect of seed parameters such as temporal and spectral shapes, repetition rate, and pulse energy on GMN amplifiers was conducted [24]. To optimize the oscillator pulse for seeding the GMN amplifier the output of the oscillator is spectrally filtered with a 4-nm bandpass filter. This is necessary because the pulses from the oscillator are chirped to approximately 8 ps, and GMN amplification relies on rapid spectral broadening in the beginning of the gain fiber, which in turn requires the seed pulse to be close to its transform limited duration. Spectral filtering of the chirped oscillator pulse significantly reduces the pulse duration and produces an appropriate seed pulse for the GMN amplifier. The second Faraday isolator ensures no back coupling from the amplifier to the oscillator.

2.2 Characterization of the laser system

Previous literature on GMN amplifiers suggests that these systems should be highly suitable for use in multiphoton imaging applications because they produce high-energy (100 nJ – 1.2 $\mathrm{\mu}$J) and short (~30 fs – 40 fs) pulses with simple pulse compressors [13,14]. However, it is important to characterize the particular amplifier both spectrally and temporally as nonlinear imaging performance depends sensitively on the peak power of the pulses. This laser delivered 2.6 W of average power which remained stable over at least 6 hours (the duration of the longest imaging session in this work), at which point the laser was turned off. The pulses generated are highly broadband ($\sim$150-nm at 1/e), which is both similar to previously demonstrated GMN amplifiers and significantly more broadband than other types of ytterbium doped fiber amplifiers (Fig. 2(a)). The chirped pulses from the amplifier must be compressed for high-performance multiphoton imaging. After a four-pass transmission grating compressor (in the Treacy configuration, LightSmyth 16-869) the usable power was 1.8 W.

 

Fig. 2. Measured laser spectrum (a) and compressed pulse measured with FROG (b). Insets: measured and retrieved FROG traces.

Download Full Size | PDF

While the broad bandwidth of the amplifier output implies a short transform-limited pulse duration, it is important to be able to easily compensate the spectral phase with a pulse compressor as has been done with other GMN amplifiers. The pulse was characterized after the pulse compressor with intensity frequency-resolved optical gating (FROG) (Fig. 2(b)) [25]. With the external grating pair the pulse was compressed to 33 fs when the gratings are separated by $\sim$10 mm. This pulse duration is shorter than typical commercial systems used for 2PEF imaging, including standard Ti:S oscillators ($\sim$70 fs) and fiber CPA systems ($\sim$350 fs). For fixed pulse energy and average power, a shorter pulse duration leads to higher signal-to-noise ratio (SNR) in 2PEF and is thus desirable, although third- and higher-order dispersion of the microscope objective should also be considered for very short pulses [26].

We note that for this specific amplifier significant trial and error was not required to obtain pulses at this energy and duration. However, previous experience suggests that this is not always the case. Experimental performance is typically limited by either excessive Raman scattering in the fiber or through complex wave-breaking phenomena. Excessive Raman scattering is usually due to the amplifier fiber being too long (we have typically found 2 m - 3 m to be appropriate) or by leaving too long a fiber pigtail on the output collimator (less is better, and with more than ~10 cm we have noticed performance degradation with other GMN amplifiers). Wave-breaking dynamics typically result from an anomaly involving the seed pulse. This is most commonly due to the seed not being close to transform-limited in duration (whether due to sub-pulses or a chirp) or by not launching the seed fully into the slow axis of the amplifier (a sub-pulse in the fast axis can interfere with the GMN evolution).

3. Imaging experiments

The pulse parameters achieved by this amplifier are attractive for nonlinear microscopy. In particular, the pulses are significantly more broadband than pulses typically used in 2PEF, which enables very short compressed pulse durations or multi-wavelength excitation for superior separation of fluorescence from distinct fluorophores (using multiple fluorophores enables identification of distinct biological structures). We performed two different imaging experiments to demonstrate this. In the first, we image HeLa cells in vitro labelled with several different fluorescent protein fusion constructs and compare the quality of signal separation in our unmixed images with single pulse and multi-wavelength excitation, with the expectation that more information (excitation at different wavelengths) would lead to higher-purity unmixing. In the second experiment, we compare the deep-tissue imaging performance of the GMN amplifier (with its full bandwidth and shortest possible pulse duration) with that of a standard commercial fiber amplifier.

3.1 Cell imaging

HeLa cells were grown in Dulbecco’s Modified Eagle’s medium (DMEM, Invitogen) supplemented with 10% fetal bovine serum with 1% antibiotics, and maintained at 37$^\circ$C in a humidified incubator with 5% CO$_2$. Cells were plated at a density of 5$\cdot 10^4$ cells in a 35-mm cell culture imaging dish with a glass coverslip bottom (ibidi). After 24 hours in a dish and reaching about 70% confluence, the cells were transfected with Lipofectamin 3000 according to the manufacturer’s instructions (Invitrogen). The plasmids used for labeling cellular organelles were as follows: tdKatushka2-CENPB-N-22 (Addgene #56033) for the centromeres, tdTomato-Calreticulin-N-16 (Addgene #58074) for the endoplasmic reticulum, and pQC NLS YFP IX (Addgene #37341) for the nucleus.

The cells were fixed in 0.5% paraformaldehyde for 5 minutes, followed by three washes in phosphate buffered saline, and then imaged through a glass coverslip. For cell imaging, we used a 63x water immersion objective with a numerical aperture (NA) of 1.2 (Zeiss) to image using the full laser bandwidth. We identified a cell that contained three fluorescently labeled organelles and recorded one slice of images across the four microscope detection channels. The detection channels were about 40 nm wide and centered at emission wavelengths of 540 nm, 579 nm, 623 nm, and 660 nm. We split the full laser bandwidth into four sections 20 nm wide while using the dispersion compensation optimal for compression of the full-bandwidth pulse. We recorded a slice of images of the same cell across all detection channels using the spectral selections sequentially (Fig. 3(a)).

 

Fig. 3. Plasmid labeled, fixed HeLa cell imaged sequentially using a GMN amplifier at excitation wavelength slices of 1025 nm, 1050 nm, 1075 nm, and 1100 nm; a) raw images of the cell acquired with each spectral slice displayed across the four detection channels of wavelengths; 540/40 nm (center wavelength/bandwidth), 579/34 nm, 623/32 nm, and 660/40 nm. b) Single label images of the unmixed data showing the individual cell features that were identified based on their fluorescent label; centromeres labeled in Katushka2, endoplasmic reticulum (ER) labeled in tdTomato, and nucleus labeled in yellow fluorescent protein (YFP). Column i shows data unmixed using raw images acquired by the single laser spectral slice (1075-nm) highlighted with a red rectangle in panel a. Column ii shows the data unmixed using raw data acquired from all laser wavelengths. The bottom row shows the composite images of the unmixed data labeled with false color; centromeres in green/yellow, ER in blue and the nucleus in red. c) Emission spectra of the three labels, Katushka2, tdTomato, and YFP that were extracted from pure label pixels visually identified in the imaged regions of interest. Scalebar: 10 $\mathrm{\mu}$m.

Download Full Size | PDF

For data processing and analysis, we performed a background subtraction, extracted basis spectra from the image data, and used the basis spectra to linearly unmix the data. The background signal for each image was calculated as the mean of all pixels below the 10th percentile value, set visually to best separate the signal from the background for this sample. We visually identified regions of interest (ROIs) containing pixels that had only one pure fluorescent label. From these ROIs we extracted basis spectra that we used for linearly unmixing the data (Fig. 3(c)). In each ROI, we used the average of the pixels that had intensity values greater or equal to the 98-percentile value of all pixels in the ROI to generate the basis spectra for the corresponding label. We unmixed both by using only raw data acquired using the 1075-nm pulse (Fig. 3(bi)) and by using the raw data acquired using all four pulses centered at 1025 nm, 1050 nm, 1075 nm, and 1100 nm (Fig. 3(bii)). The 1075-nm pulse was selected for separate unmixing because all three organelles were most visible with this wavelength excitation. Unmixing using raw data from each of the four excitation spectral slices yielded clearer separation and higher contrast between labels compared to using 1075-nm acquired data alone (Fig. 3(b)).

3.2 In vivo imaging

We imaged the cortex of a live transgenic mouse to demonstrate the deep-tissue tissue imaging capabilities of the GMN amplifier. Images were recorded using the full bandwidth of the GMN amplifier. The same images were record with a commercially available ytterbium fiber laser (Satsuma, Amplitude Systèmes) that has been used successfully for two-photon imaging. With each laser, images were recorded with 100-mW power transmitted through the microscope objective. This power corresponds to pulse energies of 3.2 nJ for the GMN amplifier and 14.5 nJ for the Satsuma (due to their repetition rates of 31 MHz and 6.9 MHz, respectively). The imaging experiments are not intended as a competition between the lasers, but rather as a demonstration that the new GMN amplifier can achieve similar imaging performance as the established chirped-pulse amplifier. All animal procedures were approved by the Cornell Institutional Animal Care and Use Committee (IACUC) and were performed under the guidance of the Cornell Center for Animal Resources and Education (CARE). Breeders were purchased from The Jackson Laboratory (CX3CR1-GFP #8451, Thy1-YFPH #3782) and were crossed to make a CX3CR1-GFP and Thy1-YFP double positive mouse on a C57BL/6J genetic background. Chronic cranial window surgery was performed on a 24-week old mouse.

About two weeks after the craniotomy surgery, we imaged the mouse using a 25x 1.05 Numerical Aperture (NA) water immersion objective (Olympus, XLPLN25XWMP2). The mouse was anesthetized and injected with 50 $\mathrm{\mu}$L of Texas Red dye retro-orbitally to label the blood vessels. We imaged the mouse 15 minutes later. During the imaging process we identified an ROI in which we could visualize both blood vessels and neurons. For each laser, we recorded image stacks of 550-$\mathrm{\mu}$m depth at 1-$\mathrm{\mu}$m intervals beginning from 20-$\mathrm{\mu}$m below the cortex surface. The images were recorded across four microscope channels about 40-nm wide each and centered at emission wavelengths of 540 nm, 579 nm, 623 nm, and 660 nm.

For processing and analysis, we did a background subtraction and then extracted basis spectra from the data stacks. We used the basis spectra to perform linear unmixing on the data. We calculated the background signal for each stack image as the mean of all pixel values below the 25th percentile value (where the cutoff is set to best isolate the signal from the background for this sample) and visually identified ROIs containing pixels that had only one pure fluorescent label. From these ROIs we extracted basis spectra that we used for linear unmixing of the data.

Figure 4 displays the results obtained with the Satsuma and GMN lasers. In general, we observed comparable imaging performance. We were able to note the presence of most of the same neuron cell bodies with the GMN amplifier as we were with the Satsuma laser around 450 $\mathrm{\mu}$m below the cortex. Likewise, similar signal levels were produced when visualizing the Texas Red-labelled blood vessels. However, we were not able to observe the GFP-labelled microglia with either laser, possibly due to the excitation wavelengths of these two lasers.

 

Fig. 4. Images of a live mouse cortex taken using the two lasers (Satsuma and GMN amplifier). Blood vessels were exogenously labeled with Texas Red dye, while neurons were transgenically labeled by yellow fluorescent protein. a) 3D rendering of the data stack taken using the Satsuma laser (top) and GMN amplifier (bottom) for a depth of 450 $\mathrm{\mu}$m at intervals of 1 $\mathrm{\mu}$m beginning at 50-$\mathrm{\mu}$m below cortex surface. b) Maximum projections of image slices across 10 $\mathrm{\mu}$m of the data stack, at depths of 250, 350, and 450 $\mathrm{\mu}$m for the Satsuma laser (left column) and GMN amplifier (right column). c) Plotted line profiles across blood vessels (cyan lines on images in part b) and across neurons’ cell bodies or dendrites (yellow lines on images in part b). d) The spectra of the laser sources, Satsuma (left) and GMN (right) used for the live imaging of the mouse cortex. Scale bar: 50 $\mathrm{\mu}$m.

Download Full Size | PDF

4. Discussion

We have demonstrated that the broad bandwidth and high pulse energy of the GMN amplifier allows for dye separation in cell cultures beyond that which can be expected using a single narrower-bandwidth pulse. In addition, the short compressed pulse duration enables imaging up to 450 $\mathrm{\mu}$m deep in vivo, similar to the performance with the chirped-pulse amplifier. With some modifications to the source presented here, it will be possible to achieve deeper imaging and/or higher signal-to-noise ratio. 2PEF signals depend on the product of the average and peak powers of the excitation pulses. With higher pulse energy lower repetition rate is usually desirable for very deep imaging [27]. Lower repetition rate is achievable either by designing the seed oscillator to operate at a lower repetition rate or by using an acousto-optic modulator to pick pulses before the amplifier. A GMN amplifier using large-mode-area fiber generates pulses with at least an order of magnitude greater energy than the laser described here with similarly short pulse duration [14].

Laser stability is also an important factor to consider. The stability of short-pulse amplifiers is often limited by the stability of the seed oscillator. The oscillator that seeds the GMN amplifier employs nonlinear polarization rotation to start and stabilize pulse formation. It remained mode-locked for several weeks without adjustment, and the output power was stable during imaging sessions. However, nonlinear polarization rotation is not fundamentally stable; environmental perturbations such as changes of temperature or mechanical stress of the fibers eventually disrupt mode-locking. A Mamyshev oscillator is a qualitatively different type of femtosecond laser which is compatible with polarization maintaining fiber and is therefore environmentally stable. The seed laser may be replaced with an environmentally-stable Mamyshev oscillator, and in the future it may be possible to replace both the seed laser and amplifier with a high-energy Mamyshev oscillator [28]. Alternatively, GMN amplification has been demonstrated with a commercial laser as the seed (Orange, Menlo Systems), which promises to be a highly accessible way to build GMN amplifiers [24].

5. Conclusion

We have presented a simple, compact and inexpensive fiber laser that enables deep-tissue 2PEF imaging performance. We also have shown that the broad bandwidth of the amplifier allows for multiple-wavelength excitation, resulting in superior dye separation in imaging as compared to imaging with a single narrow-bandwidth pulse. We expect to be able to further improve imaging performance by optimizing the repetition rate of the system and by using larger-mode-area fiber in order to increase the pulse energy, as has been previously demonstrated. We believe that amplifiers and lasers based on gain-managed amplification offer significant opportunities to reduce the cost and complexity of high-performance multiphoton microscopes.