Multi-photon stimulation can occur when two or more photons at longer wavelengths are absorbed nearly simultaneously by a molecule into an energy gap spanning their combined energy and where angular momentum is conserved [1]. The discovery of this process made possible the development of two-photon laser scanning microscopy (LSM) [2] which has resulted in an exceptional method of exciting fluorescence for imaging in many areas of science. Among the reasons for its widespread use are that it enables non-invasive study and better penetration into biological specimens such as mouse skin tissue due to the use of longer wavelengths [3]. Also, the use of temporally compressed femto-second pulses having a high instantaneous intensity peak allows a greater limit of incident power than single-photon fluorescence before saturation occurs [4]. Another advantage is that photo-bleaching begins first in the focal plane where beam intensity is maximized and decreases rapidly outside of the plane, unlike in single-photon fluorescence where it occurs everywhere along the axial direction equally [5] and, to a lesser degree, throughout the light-cone of illumination. Furthermore, in two-photon microscopy, virtually all the returning photons, even the scattered ones, originate from a tight volume around the focal point which provides good optical sectioning without the need for a spatial filter such as a pinhole. The implication of this is that two-photon microscopy can observe more deeply into turbid samples than other microscopy systems. Finally, the different fluorescent response of various dyes, tissues, and cell structures eases their identification. In contrast to single-photon fluorescent microscopy, where the excitation and emission spectra often overlap, two-photon excitation typically occurs at wavelengths well separated from the emission spectra which enhances signal separation via optical filtering. As such, two-photon LSM systems have been developed with an ever growing range of capabilities to enhance biological studies [6].

Current commercial multi-photon and confocal laser scanning single focal point microscopes typically use conventional galvanometer or resonant scanners. This method limits their capture rates to a range of 30~40 frames per second (FPS) for 512 by 512 pixels in the high-speed variant of most commercial systems (Scientifica Multiphoton Resonant System, Olympus FVMPE-RS, Nikon A1R HD, Leica TCS SP8 MP, etc), with at least one system capable of imaging at up to 45 FPS (Thorlabs Bergamo II series). Higher imaging rates can be achieved in a variety of ways including by replacement of the fast-scan axis scanner with an acousto-optical deflector, an electro-optical deflector, a digital micro-mirror device, a spatial light modulator, or a rotating polygon mirror. In one example, a confocal laser scanning microscopy system, custom-built by our group using a rapidly rotated 128-facet polygon mirror, demonstrated that imaging rates up to 200 FPS for 512 by 512 pixels can be achieved [7].

The primary aim of this paper is to demonstrate the advantage achieved in imaging temporal resolution for two-photon microscopy resulting from the implementation of a custom-built high-speed single-point 2D scanning system. The need for faster scanning is seen in studies such as two-photon excitation neuronal stimulation [8], examining peripheral micro-circulation in blood flow [9], and other rapid phenomena. High blood flow in capillary vessels is a prime example of a rapid biological process, as well as an important subject of rheological studies, so tracking fluorescently labeled cells moving at high velocity was selected as a useful two-photon imaging target to demonstrate the novelty of our system at high rates of capture.

The schematic of the custom-built high-speed laser scanning microscope system used in this study is illustrated in Fig. 1. It consists of our previous confocal laser scanning microscopy system’s core components [7] along with an upgrade to include a dual channel two-photon system. The Ti:Sa laser cavity (Ti-S 800B, Leeoptics, Korea) is pumped by a 532 nm continuous wave diode-pumped solid state laser (finesse-7W, Laser Quantum, UK) and it is capable of being tuned to produce mode-locked femto-second pulses over a range of 730 nm to 890 nm. A rotating polygon mirror (MPA, DT-72-250-025/SA34, Lincoln Laser, USA) provides the fast-axis scanning while a galvanometer-driven mirror (dynAxis L, ScanLab, USA) provides the slow-axis scanning. The objective lens has a magnification of 40× and a NA of 1.2 (Carl Zeiss, Germany, C-Apochromat, water immersion) with a field of view (FOV) of 220 μm × 220 μm when a full 512 × 512 pixel frame-size is used. Light returning from the sample is captured by the objective lens, passes through it, and has its path split by a dichroic mirror into wavelengths above and below 705 nm. The longer wavelengths, consisting primarily of reflected pulses, are de-rastered and then redirected by a pellicle beam-splitter to a lens which focuses them through a 100 μm pinhole to be captured by a photo-multiplier tube (PMT) (R9110, Hamamatsu Photonics, Japan) for confocal imaging. The shorter wavelengths, which are comprised of two-photon excited fluorescence, are separated by another long-pass dichroic mirror at 552 nm, filtered to narrow ranges of wavelengths of interest, and captured as two separate channels by PMTs (R9110, Hamamatsu Photonics, Japan). The photo-current from the PMTs are converted to a voltage signal by custom-built amplifiers. These consist of cascading an inverting operational amplifier (op amp) circuit into a non-inverting op amp one. Both circuits use the Texas Instruments OPA847 op amp and designs for these circuits were taken from its data sheet with only a minor variation to each. Together, they produce a signal gain of 1110, while amplifying noise by a factor of only 584, and have a composite bandwidth of 120 MHz. This bandwidth was chosen due to it being approximately twice that of the frame grabber’s (Solios eA/XA, Matrox Imaging, Canada) maximum data acquisition rate in order to ensure accurate sampling of the signal. Rendering the data into images is performed by software (Matrox Imaging Library 9.0, Matrox Imaging, Canada) and then recorded for analysis.

The original LSM system constructed by our group shared the same scanning and optical components as that of Veilleux et al. [10]. Prior to the polygonal mirror scanner, the laser beam diameter is ~2 mm, which gets increased by two beam expanders in order to fill more of the back of the objective lens. The first beam expander uses lenses of 40 and 75 mm focal lengths while the second uses 100 and 150 mm focal length lenses. In order to upgrade that system’s image acquisition rate from 30 FPS, at a 512 × 512 pixel frame size, we have previously implemented faster beam steering optical components. The first change was to replace the galvanometer by a faster one (dynAxis L, ScanLab, USA). The second modification was to exchange the 36-facet polygon mirror with a 72-facet one, which can provide a scanning rate in the fast-scan direction (X-axis) of approximately 66,000 lines per second using the motor’s maximum rotation speed of 55 × 10³ rotations per minute (RPM). Using this maximum, we could theoretically implement 512 × 512 pixel image scanning with a frame rate of up to 120 FPS. However, in actuality, the vertices of the polygon mirror split the scanning beam which results in non-uniform illumination at both sides of the acquired image on the fast-scan axis. Thus, the achievable frame rate is less than 120 FPS because of the necessity of having to add extra pixels to scan which contain the areas of non-uniform illumination and can be cropped from the imaging. In the software settings, these areas for cropping are referred to as the back porch and the front porch. In our system, we have used 32 pixels for each thereby necessitating a scan size of 576 pixels for an image of 512 pixels on the fast-scan axis.

Another factor affecting the imaging speed arises from the performance of the galvanometer. A depiction of the control signals and the resultant rotation angle of the mirror is shown in Fig. 2. Horizontal synchronization (H-sync) pulses are initiated when a photodiode detects a beam sweeping by from a laser diode reflected off the polygon mirror. These pulses are counted up to the number of horizontal lines required for scanning a single frame. Each subsequent pulse drives a stepwise ramp cycle in the voltage applied to the galvanometer which causes the mirror to rotate by set angular displacements. When the maximum count is reached, it is reset and another pulse initiates vertical synchronization (V-sync) for the next frame. Simultaneously, the mirror is driven to accelerate its rotation to return to its starting position in as short a time as possible. The result is a non-linear response occurring whenever the beam position is returned to the start of its scan on the slow-scan axis. The duration of this effect becomes increasingly pronounced as the cycle time becomes shorter for higher imaging speeds. This portion of the scan should be cropped from the image and can be done either by using fewer pixels along the slow-scan axis in the image or increasing the number of pixels in the scan. Again, as for the fast-scan axis, this is set by the frame-grabber software in the digitizer configuration format (DCF) file as another back and front porch.

In the current study, we used three different frame rates: 30, 90 and 180 FPS. At 30 FPS, the slow-scan axis was set to 576 pixels and a back porch and front porch of 32 pixels each used to yield a 512 × 512 pixel frame image. When imaging at 90 FPS, the galvanometer produces a dramatically increased proportion of non-linear response during its cycle each time it returns to the frame reset position. It was found to be necessary to increase the scan size to 648 pixels, using a back porch of 32 pixels and a front porch of 104 pixels, in order to maintain a 512 × 512 pixel (220 μm × 220 μm FOV) frame image. Alternatively, keeping the same scan size of 576 pixels, it was found that using a back porch and front porch of 48 pixels each, a defect-free image of 512 × 480 pixels (220 μm × 206 μm FOV) is formed.

In the case of 180 FPS imaging, the non-linear response time of the galvanometer, being driven by the sawtooth pattern, dominates the cycle and cannot be accounted for by increasing the back and front porches. To achieve this imaging speed we increased the frame rate by reducing the number of slow-scan axis y-lines, resulting in a FOV with a 512 × 220 pixel (220 μm × 95 μm FOV) setting, which provided a good imaging speed and frame size for the in vivo mouse model.

The selection of the dichroic mirrors and filters in the laser scanning microscopy system was made following an investigation of achieving two-photon signals from fluorescently labeled erythrocytes. Based on our preliminary study, the cell marker 1,1-dioctadecyl-3,3,3',3'-tetramethylindocarbocyanine (DiI) (C59H97ClN2O4, CAS#: 41085-99-8, V22885, Thermo Fisher Scientific Inc., Invitrogen^TM) was chosen. The manufacturer’s report of its spectral characteristics in methanol indicates an emission band approximately between 525 and 700 nm with a peak at 565 nm. As such, a 573~613 nm band-pass filter was effective at isolating the emission from DiI and captured by the two-photon PMT2 channel. We observed an easily detectable DiI fluorescent emission within the auricle skin of a mouse at a central wavelength of 740 nm using ~15 mW of power on the sample. Moreover, there are a wide range of tissues in skin which contain auto-fluorescent proteins that exhibit a large cross-section to two-photon excitation at this wavelength. Among these are nicotinamide adenine dinucleotide (phosphate) hydride (NAD(P)H), flavin adenine dinucleotide (FAD), lipoamide dehydrogenase (LipDH), and collagen, where the first has an emission peak at about 460 nm and the last three have emission peaks at or near to 525 nm [11-14]. Due to having a greater spectral displacement in emission peak wavelength from DiI, NAD(P)H was selected as the target of the two-photon PMT1 channel using a 435~455 nm band-pass filter. For this channel, a beam power of ~45 mW in the focal plane of the objective lens was required in order to generate a sufficiently clear auto-fluorescent two-photon signal. A level of 50 mW was the maximum power condition prior to the onset of photo-thermal bubble formation. The third PMT channel imaged the de-rastered reflectance confocal signal for reference against non-labeled cells.

Three to six week old BALB / cAnNCrl mice were used for imaging in vivo biological processes. Immobilization of the subject in preparation for imaging was achieved via anesthetic injection into the abdominal cavity of 0.02 ml of Avertin (2,2,2-tribromomethanol) per gram of body mass. After a state of unconsciousness was verified by the righting reflex, the subject’s ear was flatly affixed to a cover slip with clear cosmetic ultrasonic gel. All procedures were performed in accordance with the animal protocol approved by the Kyungpook National University Institutional Animal Care and Use Committee (2018-0065).

The erythrocyte labeling protocol used in our lab is an adaptation of the one advised by Vybrant^TM in combination with the one published by Tkaczyk et al. [15]. It was devised with the goal of increasing the fraction of labeled cells in the sample. In the first stage, 20 μL of DiI is added to 200 μL of PBS, and in order to reduce particulate size, it is rigorously agitated mechanically three or more times for 5 minutes over the course of at least an hour, and left in a warm bath of 35°C when not being agitated. Concurrently, 200~250 μL of blood is drawn from mice via cardiac puncture, placed in a vial which in turn is topped off to 3 mL with PBS, and centrifuged at 300 RPM for 10 minutes to separate the erythrocytes, serum, PBS, and buffy coat. The non-erythrocyte layers are then removed by pipette, and washing of the remaining erythrocytes is done by performing a double cycle of adding PBS to fill to a volume of 3 mL, centrifuging at 300 RPM for 10 minutes, and removing the top layer of PBS via a pipette or by careful decanting. A quantity of 12 μL of erythrocytes is stained with the DiI-PBS solution for 30 minutes at 35°C. The cells are then washed three times to remove unincorporated DiI particulates by topping off the solution to 3 mL with PBS, centrifuging it at 300 RPM for 10 minutes, and removing the non-erythrocyte layers by pipette. Finally, 200 μL of sterile 0.9% NaCl (saline solution) is used to re-suspend the cells. Comparisons of confocal and fluorescent imaging of the solution yielded a range of 50~80% of cells responding to two-photon excitation. For experimental observation, 8 μL of the labeled-cell solution was injected intravenously into a subject mouse via a tail vein.

Investigating the capabilities of our high-speed two-photon microscopy system was separated into three tasks. In the first task, imaging of typical auto-fluorescent tissue structures was performed at a variety of depths in ex vivo skin of a mouse’s auricle in order to qualitatively demonstrate that our high-speed two-photon microscopy system maintains a reasonable signal-to-noise ratio relative to slower systems. The structures chosen to feature were the strata corneum, granulosum, spinosum, and basale/dermis, and the articular cartilage layer at sites both with and without a hair follicle. In a mouse auricle, the stratum corneum ranges from 0~10 μm in depth, the granulosum and spinosum are 10~15 μm deep, the basale is 15~25 μm deep, the dermis is 25~60 μm deep, and the articular cartilage is 60 μm or more deep [17].

The results of auto-fluorescent imaging in ex vivo mouse auricle skin tissue, using our LSM system’s fastest scanning rate of 180 FPS, are shown in Fig. 3. The two-photon excitation was achieved using a center wavelength of 740 nm with a spectral pulse width of 9 nm FWHM and 45 mW of power after the objective lens. The individual images are composed of an average of 30 frames each. Subsequently deeper strata are shown from top to bottom. Both sides show tissue structures down to the articular cartilage layer with the right side including a hair follicle. We observe that the high scanning speed does not yield significant additional signal noise to reduce the clarity of the images when compared to two-photon auto-fluorescent mouse skin studies conducted at slower scanning speeds [18].

In the second task, in order to demonstrate the efficacy of the tracking DiI-labeled ex vivo mouse erythrocytes moving at a known high speed with our two-photon excitation LSM, the stage was mechanically translated at a constant velocity while imaging was done at 30, 90, and 180 FPS. Subsequently, the velocity was then obtained by image analysis to verify its accuracy in comparison to the velocity set by the motorized sample translation stage. The maximum speed of cells capable of being tracked was set by imposing the constraint of position separation between subsequent frames to be less than the average erythrocyte diameter in a human. At 7.3 μm, this is larger than in a mouse at 5.8 μm [19], but it is still a reasonable constraint for cell tracking in the current study.

Imaging was achieved by two-photon excitation at a central wavelength of 740 nm, a FWHM spectral pulse width of 9 nm, and 45 mW of power after the objective lens. Using samples of blood which had undergone the DiI-labeling process, but were imaged ex vivo directly, it can be seen that some unattached DiI particles remain but we note that the erythrocytes are readily distinguishable by their form, exhibited a strong two-photon excited fluorescence, and were abundantly present. Thus, they qualified as acceptable candidates for velocity tracking.

Figure 4 shows the imaging of cells being translated at the maximum speed possible, 1200 μm/s, which satisfies the condition for tracking. For comparison, the imaging rates are shown at 30, 90, and 180 FPS using a 512 × 220 pixel FOV (220 μm x 95 μm). The ratio of the inverse of the displacements is 1:3.01:6.03 which is also in good agreement with the ratio of the imaging rates. The absolute difference between the induced speed and the derived speed was about 1% at all three imaging speeds. For a reasonable level of path continuity at this speed, an imaging rate of 180 FPS is needed to show path positions separated by less distance than twice the diameter of a typical erythrocyte.

In order to examine the feasibility of applying our high-speed two-photon microscopy system to capturing rapid in vivo phenomena, imaging was done in the auricle skin of a healthy mouse using DiI-labeled erythrocytes as targets moving within capillary blood flow. In this third and final phase, the imaging sites were recorded using 30, 90, and 180 FPS at locations which contained cells moving at velocities visually estimated as typically slow, medium, and fast to demonstrate the advantage gained in temporal resolution at higher frame rates. The speeds of the DiI-labeled erythrocytes were evaluated by measuring their displacement in pixels between consecutive frames, converting this to a physical distance based on the dimensions of the FOV, and dividing this by the time interval between the frames. The input beam was set to a center wavelength of 740 nm with a spectral pulse width of 9 nm FWHM. The power was reduced to 15 mW after the objective lens to avoid photo-thermal damage in the living skin tissue, thereby extending the time duration available for observation of in vivo blood flow.

Figure 5 depicts a representative sample site of capillary blood flow being imaged at a rate of 180 FPS. All imaging of blood flow in this study was conducted in the basale and derma strata which typically ranges from 15 to 60 um in the auricle skin of a mouse. The composite image in Fig. 5(a) shows a single DiI-labeled erythrocyte in blue. The red portion of the image is due to a reflectance confocal signal and includes cells and other tissue within the image plane set by the pinhole position. The inclusion of the confocal channel was done to demonstrate that the single DiI-labeled cell depicted is contained within blood flow. An outline of the blood vessel wall is shown by red dashed lines in Figs. 5(b) and 5(c), with the first showing only the confocal signal and the latter the DiI signal. The discrepancy in the cell geometries seen between these figures are a result of the confocal image plane being slightly below the incoming beam’s focal plane.

Figure 6 shows a two-photon excited DiI-labeled in vivo erythrocyte traveling at ~130 μm/s through a blood vessel within the auricle skin of a mouse. Imaging rates of 30, 90, and 180 FPS are depicted in columns from left to right. Individual frames show, from top to bottom, consecutive positions of DiI-labeled erythrocytes which are moving towards the left. Similarly to the ex vivo case, capturing moving in vivo cells deep in skin tissue in overlapped positions between frames is also accomplished even at 30 FPS. Using higher rates of 90 and 180 FPS yields better resolution in path position which may be useful in cases where the cell is following a sharp turn.

Figure 7 shows a two-photon excited DiI-labeled in vivo erythrocyte traveling at ~600 μm/s through a blood vessel within the auricle skin of a mouse. Imaging rates of 30, 90, and 180 FPS are depicted in columns from left to right. Individual frames show, from top to bottom, consecutive positions of DiI-labeled erythrocytes which are moving towards the left. As was shown in the ex vivo case, capturing moving in vivo cells deep in skin tissue in positions between frames, that are separated by the average diameter of a cell or less, is also accomplished here at 90 FPS and not at 30 FPS. Using the higher rate of 180 FPS yields better resolution in which sequential path positions even overlap significantly.

Figure 8 shows a two-photon excited DiI-labeled in vivo erythrocyte traveling at ~1200 μm/s through a blood vessel within the auricle skin of a mouse. Imaging rates of 30, 90, and 180 FPS are depicted in columns from left to right. Individual frames show, from top to bottom, consecutive positions of DiI-labeled erythrocytes which are moving towards the left. In this case, just as was demonstrated ex vivo, capturing in vivo cells moving deep within skin tissue, that are separated by the average diameter of a cell or less in positions between frames, is also accomplished here only at 180 FPS.

In this study, we successfully implemented a high-speed two-photon single focal point laser scanning microscopy system by utilizing faster scanning elements. On the fast-scan axis, a 72-facet polygon mirror was mounted on a motor capable of rotating at a maximum rate of about 55 kRPM. This enabled a 3.8-fold increase in the number of lines scanned per second, from 17,280 in our previous microscopy configuration, to 65,934 lines per second. In addition, a faster galvanometer-mounted mirror was used to achieve an imaging rate of 90 FPS instead of the standard 30 FPS found in most commercial laser scanning microscopes operating at the 512 × 512 pixel frame size. We used this capture rate to image fluorescently marked erythrocytes moving in blood flow at an intermediate speed. For faster blood flow, the imaging speed can be conveniently increased by reducing the scan angle of the galvanometer, which reduces the FOV. This method was employed to set the image acquisition rate to 180 FPS with a 512 × 220 pixel FOV. Thus, we were able to accurately track relatively fast moving cells at a speed of about 1.2 mm/s using our two-photon laser scanning microscopy system due to the improved scanning speed.

The frame acquisition rate can be enhanced further by continuing to narrow the total deflection angle of the slow-scan axis galvanometer. Typical commercial systems can achieve frame acquisition rates in excess of 400 FPS by using a 512 × 32 pixel frame size. The controller in our two-photon LSM could be modified to operate at this frame size and it is expected that an estimated frame rate in excess of 1200 FPS would be achieved, which should allow cells moving at up to 8 mm/s to be accurately tracked.

One method previously employed by our group to exceed 90 FPS full frame size imaging, to up to 200 FPS, was to replace the 72 facet polygon mirror with a 128 facet one and use bi-directional scanning for the galvanometer mounted mirror [7]. While this represents the optimum imaging rate configuration of our scanning system, it was achieved at the cost of a reduced clear aperture for each facet as well as a decreased scan angle, in comparison to our 72-facet polygon mirror. For an incoming beam of a given width, the vertices of a 128-facet polygon mirror dwell in the beam for a significantly greater fraction of the line scan duration. This leads to a reduction in image quality due to the fading of illumination at both sides of the frame, thereby decreasing the effective FOV. In addition, bi-directional scanning yields an alternating time duration for each overlapping pixel between consecutive images which makes particle velocity tracking problematic. For these reasons it was decided that returning to singledirectional scanning of the slow-scan axis and using the 72 facet polygon mirror for the fast-scan axis, with high rate imaging to be done by narrowing the total deflection angle of the galvanometer, would be the most expedient method of carrying out this study.

The usefulness of our two-photon LSM to examine rapid processes in biological tissue could be increased further by extending the range of wavelengths the laser source is capable of producing in order to be able to capture other endogenous fluorescent processes. As an example, two-photon excitation, at 590 nm, of the tryptophan contained in leukocytes has been used to good effect for particle tracking at an imaging rate of 30 FPS [19]. Another study showed fluorescent emission from hemoglobin in erythrocytes when two-photon excitation was applied in the Q-band at a central wavelength of 600 nm [20]. By utilizing the fast scanning method employed in our laser scanning microscopy system in combination with femto-second laser sources capable of operating at shorter central wavelengths, we expect that two-photon excited label-free high-velocity particle tracking in rapid blood flow could be realized.

Furthermore, we anticipate that our two-photon high-speed LSM could be used to better quantify the general characteristics of turbulent blood flow with methods such as cell-assisted particle image velocimetry (PIV). The frequency of particle markers per volume needed for the application of meaningful PIV analysis should be higher than was utilized in this study. To get enough data for vector field determination, in vivo PIV studies have used reflectance confocal LSM imaging to capture all the cells moving in the focal plane [10]. It is expected though, that with a sufficient density per volume of target cells, we can conduct PIV studies using purely our two-photon system. This has an immediate advantage in that collagen can be selectively observed to provide a clear image of the blood vessel walls, unlike in reflectance confocal studies, which should allow for a better analysis of boundary interaction with the cells.

In conclusion, we developed a two-photon laser scanning microscopy system, capable of high-speed imaging, by improving the scanning elements with the use of a rapidly rotating 72-facet polygon mirror and a fast conventional galvanometer. We demonstrated the efficacy of this system by imaging and accurately tracking in vivo labeled erythrocytes in blood flow moving at speeds of up to ~1200 μm/s. In doing so, we established that an imaging rate of 180 FPS is necessary in order to maintain accurate tracking between concurrent frames. Two-photon images using a 512 × 512 pixel frame size were captured with good spatial resolution at a rate of up to 90 FPS. For faster moving erythrocyte tracking, the imaging rate was increased to 180 FPS using a reduced 512 × 220 pixel frame size. In addition to the benefit of high-speed imaging, we showed that our single-laser source dual-channel twophoton LSM is capable of capturing typical auto-fluorescent tissue structures and performing cell tracking in blood flow in an inexpensive and straight-forward manner using a live mouse model. It is expected that such a system has potential use in fast dynamic phenomena in rheological, skin absorption, and neuronal stimulation clinical and biological studies.