Measuring capillary flow dynamics using interlaced two-photon volumetric scanning

Introduction

Capillary blood flow is essential for supplying the brain with oxygen and other nutrients needed to support the high metabolic demand required for normal brain function. ¹ Numerous studies have shown an association between reduced cerebral blood flow and cognitive decline,^2,3 including changes in microvasculature structure and flow.^4–6 More recently, transient disruptions in flow at the single capillary level were identified as a potential mechanism that contributes to cerebral blood flow changes driving neurological deficits.^7,8 In addition to increasing vascular resistance, capillary flow models suggest that increased heterogeneity of capillary transit times, which stalls would induce, reduces the efficiency of the microvascular network.^9–11 However, experimental observations of both spatial and temporal capillary flow speed variability across a large volumetric field of view (FOV) with numerous capillaries in the microvascular network is missing. This is because traditional multiphoton microscopy used to measure red blood flow speeds in capillaries^12,13 uses line scans along the axis of a capillary to observe the passage of RBCs through a single capillary. As a result, many of the capillary flow speeds reported are from measurements of flow speeds where a capillary is monitored for as little as 30 seconds, and in a sequential manner. Though capillary flow speed measurements have been extended to measure RBCs in multiple capillaries simultaneously by employing different scanning sequences^14–16 the short axial point spread function limits the depth of field; thus only a limited number capillaries can be measured at a single depth in a given time.

Recently, two photon imaging with axially extended Bessel-foci has been shown to greatly increase the volume rate of imaging sparsely labeled samples by allowing for the capture of fluorescent signal in a 3D volume from a single 2D scan. Coupled with a resonant galvo scanner, Fan et al. ¹⁷ showed video rate tracking of RBCs as they flow single file along the capillaries. But tracking speeds of up to 2 mm/s require imaging rates of ∼100 Hz in a resonant galvo-galvo raster scanning system. At these speeds, the imaging field of view is sacrificed as lower pixel resolution is needed to reach high frame rates, and the short pixel dwell times require high excitation powers and limit the ability to image deep into the brain.

Alternatively, OCT provides volumetrically resolved angiograms from a 2D scan using the endogenous contrast that arises from moving RBCs.^18,19 Doppler^20,21 and speckle decorrelation^22–24 OCT imaging have been used to quantify RBC speeds within capillaries but have imaging rates of 10 sec per frame or more.^21,25 This is not suitable for imaging fast flow speed changes within capillaries such as changes due to RBC stalls, which occur on the order of a few seconds or less. ²⁶ Furthermore, OCT lacks the specificity provided by fluorescent labeling that permits identification of the roles of different types of blood cells (erythrocytes, leukocytes, and thrombocytes) ⁷ and vessel wall components (e.g. pericytes) in dynamic flow changes^4,27–30 and disruptions.^27,30

Our work overcomes these challenges by integrating a novel scanning pattern called “stagger scan” achieved with linear galvo scanners and a low cost extended Bessel-foci two photon imaging system. Using the stagger scan method we track RBC movement and developed two techniques, Frame Pair (F-Pair) and Frame Tilt (F-Tilt), to measure RBC velocities with average speeds ranging from 0 mm/s to 1.5 mm/s in capillaries across a 237 μm by 237 µm by 120 µm volume. The measured speeds with our method were validated against the traditional line scanning method ¹² for single capillaries. With our approach, RBC speeds across a capillary network were extracted to quantify the spatial and temporal variability in flow speeds, including the ability to capture changes in flow speed due to RBC stalling events both within the stalling and adjacent capillaries. The extracted capillary flow dynamics and distributions matches well with previously published reports^5,11 while the observed point prevalence of stalls was higher than previously reported^8,26 due to the increased temporal resolution of our system. Lastly, we demonstrate that our imaging depth and ability to extract flow speeds is background limited at roughly 400 μm deep.

Methods

Two photon microscope and Bessel foci generator

Measurements were performed with a custom designed, galvo-galvo, two-photon microscope with an axicon based Bessel foci-generator. A Ti-Sapphire Laser (Insight X3,Specatra Physics) tuned to 920 nm was used for excitation with an electro-optic modulator (Conoptics 350-105-20) for fast power control.

Following,^33–35 an axicon (Eskima Optics, Model 131-1279) and a Thorlabs achromat (FL = 250 mm) is used to generate the annular ring, which is then imaged onto the first beam steering galvanometer inside the microscope using a Thorlabs achromat (FL = 100 mm) and a Sill Optics lens assembly (S4LFT0089/094) in a 4f configuration. A Thorlabs BE02-05-B beam expander is used to tune the excitation beam diameter before the axicon, which allows us to tune the length of the extended foci from 50 μm to 500 μm without changing optics (see Figure 1(a)).

OPEN IN VIEWER

Figure 1.

(a) Scan sequence for a typical raster scan, all lines are scanned sequentially before returning to the first line and scanning another image. (b) Scan sequence for a stagger scan. 2 images are constructed simultaneously, with the line jump s controlling the number of lines jumped when alternating between the 2. In this case, the time between frames is dramatically reduced. (c) Schematic of the Bessel beam two-photon microscope. The axicon (AX) and lens L1 are used to generate the annular illumination required to generate a Bessel like focus, which is imaged onto a plane conjugate to the back pupil plane of the objective of a conventional two-photon microscope using lenses L2 and L3. The Bessel focus creates an axially extended focus shown on the right with the fluorescence intensity trace plotted below showing a full width half maximum of 100 µm. (d) Comparison of the same ROI imaged with a Gaussian (left) and Bessel (right) focus. The Bessel focus allows for imaging a 3D volume during a single 2D raster scan, visualizing significantly more capillaries in the same field of view (scale bar: 100 µm) and (e) The staggered scan pattern creates an image containing two interlaced images (left, line phase offset added to make 2 images more clear). This can be separated into a pair of frames (right) generated with a time delay of a few line periods. 512 × 512 pixels with a paired frame delay (PFD) 12.24 ms. Scale bar 100 µm.

The microscope consists of two decoupled linear galvanometers (Thorlabs GVS001) imaged onto each other using two Sill Optics lens assembly (S4LFT0089/094) in a 4f configuration. They are then imaged to the back pupil plane of the objective (Olympus 25x, 1.0 NA) using a Sill Optics lens assembly (S4LFT0089/094) and a custom designed Plossl tube lens comprised of two Thorlabs achromates (AC508-500-AB). The objective is mounted on a 1 mm, fast piezo Z translation stage for axial scanning (nPoint nPFocus 1000).

For detection of the 2P fluorescent emission, a long-pass primary excitation/emission dichroic (SEMROCK DI03-R785) is placed between the tube lens and objective to reflect the fluorescently emitted light to two cooled PMTs (Hamamatsu H74422-40/50) with emission filters for green (SEMROCK FF01-502/60) and red (SEMROCK FF01-607/70) fluorescence respectively.

The microscope is controlled using NI industrialized PXIe with an embedded controller (NI PXIe-8880) chassis running ScanImage software. ³⁶ Galvo, pockel cell, and piezo control voltages are generated by ScanImage and sent to via a data acquisition card (NI PXIe-6358) and BNC break out connector box (NI BNC-2110). The photocurrent from the two PMT’s are converted to voltage using Femto DHPCA-100 trans-impedance amplifiers and collected, along with galvo position feedback, by a second data acquisition card (NI PXIe-6358) and BNC break out connector (NI BNC-2110).

Stagger scanning, paired frame rate, and measurable velocities

In conventional unidirectional raster scanned imaging, each line in the image is scanned sequentially until a full frame is generated (Figure 1(a)). The next frame then begins again at the first line and this sequence is iterated to produce a time series of images. The time delay between two frames is a product of the line rate and the number of lines per image. In a stagger scanned image (Figure 1(b)), pairs of lines are acquired sequentially with each pair having a fixed line offset, or line jump, that is equal to an integer number of lines. In this manner, two images of the same field of view are constructed simultaneously rather than sequentially. For example, the first pair of lines scanned in the image would represent lines 3 and 1 in image frame (Figure 1(b)). The next pair of lines in the image would represent lines 4 and 2 in a normal raster scanned image, and so on until all of lines in the scanned image are scanned twice before a single stagger scan frame is complete. So, in a single stagger scanned image, there are two interlaced frames with a pairwise frame delay (PFD) of

PFD = 2 s + 1 f L

(1)

where f L is the line acquisition rate including flyback overhead in hertz, and s is the line jump. Since s is a small number, a PFD on the order of milliseconds can be created at typical raster scanning rates. The total scan period for one stagger scan image becomes

T scan = 2 N + s f L

(2)

which is roughly equal to the time required for two raster scanned images, but with significantly less time between the two frames. This short time between the pair of frames rate enables tracking of fast-moving RBC’s travelling within capillaries.

To demonstrate the utility of RBC tracking with stagger scanning, we compare the theoretical maximum RBC velocities that can be measured along a 50 µm length of capillary between traditional raster scanning versus stagger scanning for a 256 × 256-resolution image (Supplemental Figure 1). In the ideal scenario, a single RBC flows along a visible length of capillary between two full image frames and the maximum measurable speed is determined by the length of the capillary and time between the two frames. Assuming a galvo line rate is at 900 Hz, a traditional uni-directional raster has a maximum measurable RBC velocity of ∼0.17 mm/s, while in the same scenario, the theoretical maximum RBC velocity with a line jump (s) of 2 is roughly 9 mm/s, which is fast enough to resolve any capillary RBC velocities.

Another advantage of stagger scanning with traditional linear galvos over using resonant scanning galvos for tracking fast dynamics is our ability to easily control the pixel dwell times when acquiring the image. With a longer pixel dwell time, more fluorescence is generated and translates to a higher contrast between the fluorescent plasma and RBCs. This is particularly important when using an extended axial focus, where excitation power is distributed over a large volume compared to traditional Gaussian foci; as well as imaging deep into scattering tissue.

Figure 1(c) shows an illustration of our axicon based Bessel beam generator module used in our custom built 2 P microscope, with an elongated axial full width half maximum of 100 µm. Comparing the 2P angiograms captured with the extended axial focus with a conventional Gaussian focus (Figure 1(d)), angiograms generated with the elongated foci observe significantly more capillaries in a single scan. Then, using a staggered scan, the de-interlaced images create a time series of frame pairs with milliseconds paired frame delay (Figure 1(e)) that is tunable by changing the line rate of the acquisition, f L , and the line jump, s, between repeated line scans (equation (1)). Within each frame pair, the short time delay allows us to observe RBC shadows as they flow along the length of the capillary (Figure 2(a)) and estimate RBC flow velocities in multiple capillaries using the configured paired frame delay (PFD) and the measured distances the RBCs travelled along the capillary. We call this method of velocity estimation F-Pair.

OPEN IN VIEWER

Figure 2.

(a) The short temporal delay between frame pairs allows for the tracking of RBCs as they flow along a capillary. In fast flowing capillaries (green arrow), RBCs flow a much large distance between frame pairs compared to slower moving capillaries (orange arrow). Taking a centerline profile, the distance Δx the RBC flows can be used to calculated speed using equation (3). Scale Bar 100 µm and (b) Comparison of capillaries with different amounts of RBC slewing during image scanning. In the yellow ROI the RBCs were flowing left to right, while in the blue ROI the flow was right to left, as determined by the direction of the slant. By measuring the angle of the slew relative to the capillary axis, the speed can be estimated using equation (6). Lines indicate the fast (F) and slow (S) axis of scanning. Scale bar 50 µm.

In addition, the apparent shape of an RBC in a given capillary appears slew due to the RBC flow velocity and the slow axis raster scan velocity (Figure 2(b)). Faster RBC velocities create a steeper slewed shape (Figure 2(b)), while slower velocities create a shallower slew shaped. By using this angle of the slew and the known slow axis line rate, an additional method called F-Tilt can be used to estimate flow speed to track RBC flow as previously described in.^14–16

RBC velocity measurements using F-Pair and F-Tilt

From the image pair that is deinterlaced from a stagger scan image, there are two methods for estimating RBC speeds. In the F-Pair method, a capillary with an RBC shadow that is visible in both images of the frame pair is identified. The pixel position of the RBCs along the length of the capillary is defined as the center of the RBC shadow in the capillary (Figure 2(a). From the pixel positions in each frame pair, the velocity of the RBC along the capillary is simply

v = △ l p x × p x PFD

(3)

where △ l p x is the pixel shift between the two intensity traces in pixels, p x is the pixel step size in µm/pixel, and PFD is the paired frame delay in seconds. For F-pair to produce reliable speed measurements, RBCs must be uniquely identifiable, otherwise aliasing of the RBC shadows will produce erroneous results.

In the F-tilt method, RBC movement in the capillary will result in an image where the shadow appears skewed in raster scanned systems. Like the image distortion seen with a rolling camera shutter, ³⁷ the faster an RBC is moving through a capillary that is parallel to the fast axes of the scanning, the more the RBC shape is tilted/shifted in the image.^14–16 For a capillary with flow aligned perfectly with the fast axis of scanning, the RBC speed is estimated from

v = △ X △ Y * p x * f L 2

(4)

where △ X is the distance the RBC moved along the length of the capillary in pixels, △ Y is the number of lines it took to scan over the capillary width, and f L is fast axis line rate in lines/second. When the capillary’s axis does not align perfectly with the imaging fast axis, equation (4) must be modified to account for the orientation of the capillary and the imaging fast axis

v = 1 2 p x * f L tan θ 1 cos ( α ) + sin ( α ) / tan θ

(5)

where α is the angle between the capillary and the fast axis accounts for the capillary orientation differences and θ is the angle between the RBC shadow and the capillary axis.

According to this equation, F-tilt velocity estimates will have large uncertainties when the angle between the capillary and the fast axis is large. To minimize this error, F-tilt measurements for estimating RBC velocity should be applied to a pair of orthogonally oriented and sequentially acquired stagger scan images.

Results

Validation of flow speed measurements

To quantify the flow speed accuracy obtained with F-Tilt and F-Pair from the stagger scanning method, we compared to velocities calculated using both methods to a one second line scan interleaved between the stagger scans. Measurements taken from 24 different capillaries, with speeds ranging from 0–2 mm/s as measured using the line scanning technique. Figure 3(b) shows accurate F-pair velocity estimates in comparison to the line scanned estimates for capillaries with slower speeds, but F-Pair tends to underestimate velocities at higher speeds. This is because tracking movement using the frame pairs can underestimate higher speeds due to aliasing of the RBC shadows. Figure 3(c) compares the RBC velocity extracted using F-tilt against the velocity estimated by line scanning. Using F-Tilt, RBC velocities at higher speeds match velocities estimated by line scans and overestimates RBCs at slower speeds because shallower slew angles are harder to estimate from the image. Increasing the sampling resolution would give a better angle estimate at the expense of imaging speed. Therefore, we used of a combination of the two methods(See Supplemental Note), taking the F-pair based estimate at slower speeds, and F-Tilt based estimate at faster speeds to better ensure accurate reported speeds (Figure 3(d)). We also found that accounting for the orientation of the capillary using equation (6) is critical for an accurate F-Tilt estimation (Supplemental Figure 2).

OPEN IN VIEWER

Figure 3.

(a) Summary of the flow speed validation. For each sequence, the F-pair and F-Tilt measurement was compared to the line scan of a selected capillary interleaved between orthogonal stagger scans. (b–d) Comparison of speed estimates to interleaved line scan speeds. Black line indicates ideal y = x (speeds match), blue line indicate linear fit of the results. Horizontal and vertical scale bars indicate standard deviation. (b) Comparison of the F-Pair speed with the line scan speed across capillaries. Fit: y = 0.81× +0.14. (c) Comparison of the F-Tilt speed with the line scan speed across capillaries. Fit: y = 0.97 ×−0.05 and (d) Line Scan Speed compared to a combined F-pair and F-Tilt estimate. For each time point, F-Tilt speeds >0.4 mm/s were used, and F-pair speed used in all other cases. Fit: y = 1.01 × + 0.01.

Tracking of capillary flow dynamics

Using a combination of F-pair and F-Tilt for estimating flow speeds, we track the flow dynamics within capillaries with a range of flow speeds (Figure 4(a) and (b)), including in capillaries where flow undergoes abrupt changes when it stalls. Figure 4(d) shows an example time course of capillary flow speed estimated using our method compared to line scanning, including where a stoppage or “stall” and recovery of capillary flow occurred during the validation measurements. The similarity in reported flow speeds demonstrates that our approach can capture these dynamics when compared to a line scan taken of a single stalling capillary. We found the variability in flow measured in a capillary (measured as the coefficient of variation) to be well matched between line scans and stagger scans (Supplemental Figure 3), despite the fact that the line scans and stagger scans were performed sequentially and flow speed changes could occur when switching between the scans Figure 4(e) to (h). However, in the case of staggered scanning, it is not limited to serially tracking single capillaries, as is the case with line scanning.

OPEN IN VIEWER

Figure 4.

Tracking of Capillary Flow Dynamics. Comparison of the speed using the F-pair/F-Tilt estimate to the line scan speed averaged over 1 second in a (a) low speed capillary (b) high speed capillary (c). Capillary oriented at a 18.6° angle with respect to the fast axis of scanning and (d) Capillary whose flow stalled during the measurement (e–f). Capillaries where flow changes and variability are similar between stagger and line scans (g–h) and Capillaries where fast changes in speed occurred between the stagger and line scans taken in sequence, though average speeds are similar.

OPEN IN VIEWER

Figure 5.

Comparison of flow speeds and flow variability in capillaries where F-pair and F-Tilt estimates could be obtained across 4 healthy adult mice (a) Mean capillary flow speed (b) Standard deviation of flow in the same capillaries across the 2-minute measurements.

Tracking of capillary flow distributions

From modeling of capillary blood flow, ⁹ increases in transit time heterogeneity is associated with reduced oxygen delivery and has been observed in older mice where hypoxic pockets formed. ⁵ Using staggered scanning, we measured capillary flow speed in 4 fields of view between 3 healthy adult mice for 2 minutes per volume. Overall, we extracted speed estimates from typically 1/3 of the visible capillaries (an average of 10 capillaries per dataset). Capillaries were typically excluded from analysis due to only have a short visible length, due to being on the edge of the image or having spatial overlap with other capillaries. Lack of clear red blood cells in the capillaries due to high speed or poor signal to noise was also the cause in some cases.

Flow speed across an average of 10 capillaries per field of view was extracted and flow heterogeneity was estimated using the mean and standard deviation of flow both within and across the capillaries (40 capillaries in total). We found the mean flow speed across all capillaries to be 0.57 +/− 0.25 mm/s; an inter-capillary flow variability (the coefficient of variation of mean flow speed across capillaries) to be 0.42; and an intra-capillary variability (coefficient of variation of flow speed in a single capillary) to be 0.44 +/− 0.2. Our measured inter capillary variability is similar to that reported using point scanning ¹¹ and line scanning, ⁵ which reported an inter-capillary (spatial) heterogeneity of ∼38% and ∼50% reported in young mice respectively. Interestingly, our measured intra capillary variability was significantly higher ¹¹ and indicates that the variability in flow speed within a given capillary is similar to the variability across the capillary network, at just under 50% of the mean flow speed. The implication is that across even a few minutes, capillary flow speed can vary significantly within a capillary, at a similar magnitude to the variability seen across the capillary network. The long, continuous measurement of capillaries with stagger scan allows us to observe these different capillary flow dynamics, since we do not need to change focus or measure capillaries one at a time.

Estimation of capillary stalling parameters and effects

In addition to the capture of capillary flow speeds using staggered scanning, our Bessel beam based two-photon system provides an improved ability to catch stalling events compared to previously used techniques. Using standard raster scanning, we were able to image up to a 713 × 713 × 120 µm volume of the micro-vasculature at a rate of 0.57 Hz, generating angiograms where flow in >200 capillaries are visible on average (Figure 6(a)). This volume rate is a significant improvement over previously used OCT based angiograms (because of the shorter pixel dwell time), as well as standard two-photon microscopy which requires serially scanning in depth to acquire a volume. We used this large FOV to capture stalling events that occurred over a 10-minute time period (Figure 6(b)) and compared them to what has been reported previously using OCT-based methods.^8,26 Stalls were manually identified and classified as when an RBC shadow remained stationary for at least two sequential images (Figure 6(c)).

OPEN IN VIEWER

Figure 6.

Capturing of capillary stalling event and flow dynamics (a) MIP of a 713 × 713 × 120 µm volume that was monitored using standard raster scanning for the identification of stalling events. Red highlighted segments represent capillaries that were identified to have stalled during the 10-minute measurement. (b) Example of two capillaries that experience stalling, indicated by the blue and yellow arrows. Solid arrows denote frames where the capillary is flowing, and hollow arrows denote times where RBCs are stalled in the capillary scale bar 20 µm. (c) Stallogram showing the time points where each of the 18 identified stalls occurred. Black indicates time when flow was stalled and (d) Speed traces from a 10-minute stagger scan where a capillary stalled for longer than a minute, 4 frame running average. The connected capillary (blue arrow) flow speed increase during periods of slow/no flow in the stalling capillary (red). Pearson Correlation −0.31.

We found an incidence comparable to that found with OCT based angiograms (Supplementary Table 1), and significantly larger than rates that have been observed with conventional two-photon microscopy^7,38 since capillaries were monitored continuously for the full measurement time. The cumulative stall duration of an individual capillary was also found to be similar, though the point prevalence (average number of capillaries stalled at a given time) was found to be close to double. This is likely due to the higher frame rate of our imaging enabling us to identify shorter duration stalls than observable with the ∼9 seconds per image with the OCT angiograms.

Previous work^7,8,26 only quantified the occurrence rate of these events, but not their effect on nearby flow speeds. Using stagger scanning we were able to observe a change in flow speed in a capillary due to nearby stalling. Figure 6(d) shows an ROI when we capture a stalling of flow in a capillary that lasts for ∼1 minute. During the time, we can see a significant increase in flow in a connected capillary during the stall, before returning to a lower speed when flow resumes in the stalled capillary.

Finally, by using higher dwell times with our system, generating more fluorescence, we are better able to image capillary flow deep into the brain (Figure 7). RBC shadows are clear even after shifting the focus of the Bessel up to 350 µm below the surface. Below this (Figure 7(d)), capillaries are visible but the poor signal to noise means RBCs are not visible. We used F-Tilt to estimate speed at the maximum imaging depth (Figure 7(d)) in two representative capillaries (Figure 7(d) inset; Figure 7(f)) across ∼1 minute of raster scanning, showing speed dynamics deep in the capillary network. However, the larger pixel size in these raster images leads to reduced sensitivity to small speed changes (equations (4) and (5)), and the lower slow axis line rate means a steeper slewing effect occurs at lower speeds, limiting the potential maximum speeds. Therefore while this demonstrates that stalling and flow speed data can be extracted across multiple cortical layers ¹³ with our system, stagger scans with higher resolution would offer a more robust estimate of flow speed.

OPEN IN VIEWER

Figure 7.

Resolving capillary flow across depth. Single 713 × 713 × 120 μm volumes raster scanned with a 4 μs pixel dwell time at an approximate depth range of (a) 0–120 μm (b) 100–220 μm (c) 200–320 μm (d) 320–420 μm, and (e) 400–520 μm. Scale Bar 100 μm Inset: 50 μm and (f) Speed measurement taken using F-Tilt for the capillaries indicated in the inset in (d).

Discussion

Previous studies of capillary flow dynamics, particularly around transient changes in flow, have been hindered by limited throughput or low frame rates. To overcome this, we built a custom two-photon Bessel beam microscope to enable efficient and reliable tracking of capillary flow speeds and capillary stalling events. Combined with our staggered scan pattern, we can track flow dynamics across the capillary network simultaneously at ∼1 Hz. The use of a staggered scan pattern to create frame pairs allowed us to capture and identify the flow of individual RBCs as they flow along a capillary while maintaining high dwell times to ensure a good signal to noise ratio. Rather than being limited to scanning single capillaries or multiple capillaries in sequence, tracking can be done across numerous capillaries simultaneously. Using this technique, we show very reliable estimates of flow speed at speeds less than ∼ 0.4 mm/s. Above this speed, this approach can underestimate speed in some cases, due to the aliasing of multiple red blood cell shadows that are present in the capillary. To help address this limitation, we use the angle of the RBC as it appears in the capillary to estimate its speed (F-Tilt). We found this better able to estimate faster flow speeds, but less accurate at lower flow speeds where the slewing effect was harder to distinguish. A combined approach was therefore used, taking the F-Tilt estimate for speeds above 0.4 mm/s, and the F-pair estimate for anything else. This resulted in reliable tracking over a large range of relevant flow speeds in capillaries across the field of view. With this combined approach, we were able to quantify flow speeds and similar levels of flow variability when compared to line scanning of individual capillaries (Supplemental Figure 3). We also demonstrated that F-Tilt analysis could also be performed on conventional raster scans (Figure 7(f)), which could also apply to previously acquired data in any raster scanning system where the line rate is known. Compared to previous studies using a similar approach, we also demonstrated the ability to track much faster capillary flow speeds and have a higher frame rate.^15,16

To leverage the ability for simultaneous flow speed measurements, we performed our scanning approach in 237 × 237 × 120 μm volumes and looked at flow variability across 40 capillaries in awake mice (Figure 5(a)). Our results show capillary flow speed can vary significantly within a capillary over the course of 2 minutes (Figure 5(b)), with a mean coefficient of variation (COV) of 0.44. We also found that the spatial variability (measured by the COV) to be very similar to the COV of mean speeds across the capillaries measured, and comparable to values reported with capillary line scans and point scan measurements.^5,11 Understanding changes in flow heterogeneity across the capillary network due to capillary stalling, vascular changes associated with aging ⁵ and Alzheimer’s, ³⁹ and vascular remodeling due to stroke^40,41 will be key to mapping disease progression and the effectiveness of treatment. The ability to quantify both spatial and temporal variability in flow speeds simultaneously will provide important information on the health of the network. Further, while the present work measured capillary flow speed in awake resting mice, our technique is easily applicable to measuring flow speed responses and variability to stimulus, and is a part of ongoing/future studies. In addition, by increasing the number of capillaries where flow speeds can be measured, we also greatly increase the throughput for experiments. For the study of capillary stalling for example, stalling events occur randomly making line scanning ill-suited since it is limited to one or a few capillaries. This would require many sequential measurements (Supplementary Figure 4) in order to capture even a small number of these events. Further, line scans are sensitive to mouse motion during longer imaging acquisitions. Instead, we can monitor a region of interest that contains many capillaries in parallel (Supplementary Figure 4) and analyze capillaries that stall during our measurements. Further, we can more easily study the effects of stalling on capillaries nearby the stall (Figure 6). While this has been done with line scanning to study the effects of permanent occlusions,^42,43 the transient nature of capillary stalling requires the measurement of capillaries in parallel to capture similar dynamics. Similarly, while OCT offers the ability to monitor flow speed in a large volume of capillaries label free (Supplementary Figure 4), the temporal resolution is not sufficient to capture the faster dynamics around stalling events. ²¹

We also demonstrated that our system is capable of monitoring hundreds of capillaries for the occurrence of stalling (Figure 6) using a large field of view conventional raster-scanning pattern and identifying capillaries whose flow was disrupted for periods of time (Supplementary Table 1). These scans could then be followed up with stagger scans to measure flow dynamics. Since capillaries that stall have a high probability to stall again, ²⁶ capillaries that stall during the large FOV survey scan are likely to stall again during a follow-up stagger scan measurement. Spatial and temporal flow variability can then also be quantified as a function of the stalling rate in a given region. This approach will greatly improve the ability to understand the impact of these stalls on flow distributions in the capillary network, as well as test the effects of functional activation and pharmaceutical intervention on capillary stalling and flow heterogeneity.

We implemented our scan pattern in a configurable arbitrary line scan pattern in ScanImage, a widely used open-source software for two photon microscopes. The scan pattern itself requires no additional or specialized hardware, only control of the galvo positions during image acquisition. We also used off the shelf linear galvos, so any scanning-based system should be able to achieve the line rates used. This flexibility means there is also great opportunity for tunability and optimization. For example, at our current line rate, we found the maximum measurable speed to be ∼2 mm/s, making the majority of capillary flow speeds measurable.^11,12 Above this speed, we typically are unable to resolve individual RBCs in the capillary, limiting the ability to make a speed measurement using either F-pair or F-Tilt. This limitation could be overcome by using a scanner that can achieve higher line rates, decreasing the time taken to scan over the capillary as the RBC flows through. However, the use of higher line rate always comes with the tradeoff of shorter dwell times and worse signal to noise ratio (SNR). Alternatively, bi directional scanning could be easily implemented to further increase the effective line rate of the scan, though extra care would be needed to avoid any scan field warping or interlacing artifacts between the frame pairs. Image resolution, pixel dwell time, and the paired frame delay are also easily controllable.

The ease of implementation makes it readily applicable for anyone interested in studying capillary flow or other faster dynamics where conventional frame rates are not sufficient. For example, three photon microscopy is limited due to need to use low repetition rate sources (typically 1–2 MHz), capping the maximum line rate of scanning.^44–46 The Bessel beam module ⁴⁵ and line rates used in our studies are compatible in this case and could easily enable high throughput studies of capillary flow speeds in deep brain regions. This is of particular importance as microvascular structure and flow dynamics can differ across brain regions,^11,47 as well as changes associated with aging ⁶ and pathology. ⁴⁸

While not required to use our approach, the extended depth of field offered by a Bessel focus increases the number of capillaries that can be measured simultaneously. This allowed us to monitor on average 10 capillaries in a single field of view, including bifurcations (Figure 6(d)). The axicon based design we used to generate the Bessel focus is also a low-cost solution, which can be added in both custom^33,35 and commercial systems.^34,49

The main limitation of our approach is that the F-pair and F-Tilt based analysis currently relies on manual determination of RBC displacement and tilt angle. This is not only cumbersome for processing many capillaries and long measurements but subject to experimenter-to-experimenter variability. We are currently seeking to automate this process, using our validation measurements as a training set. This will also allow for a broader exploration of the parameter space of scanning line rates and paired frame delays to best assess flow speeds (See Supplemental Note).

In summary, we developed a scanning approach with a low cost Bessel beam two-photon microscope and standard linear galvanometers that allows two different methods for robust tracking of flow speed across numerous capillaries simultaneously and up to 400 µm below the cortical surface. We then applied these techniques and quantified both spatial and temporal flow variability in awake mice, as well as demonstrated the utility of the Bessel beam microscope in capturing stalls in capillary flow.

Supplemental Material