2.1.

Animals, Viruses, and Surgery

All animal experimentation was carried out with approval from the UK Home Office and University of Oxford Animal Welfare and Ethical Review Board. Three wildtype (C57Bl/6) mice and three mice transgenically labeled with GCaMP6s in excitatory neurons [B6;DBA-Tg(tetO-GCaMP6s)2Niell/J x CamK2a-tTa(AI94)] were used in this study.

Briefly, a cranial window, consisting of two #1 thickness cover glasses (3 and 4 mm diameter) adhered to one another using UV-curing optical adhesive, was implanted over the somatosensory cortex ($-1.9\mathrm{mm}$ anterior and $+3.8\mathrm{mm}$ lateral from Bregma) of each mouse during anaesthetized recovery surgery. During surgery, $\sim 800\mathrm{nl}$ of C1V1-Kv2.1 virus (AAV2/9-CaMKIIa-C1V1-t/t-kv2.1-mScarlet; $\sim 2\times {10}^{12}\mathrm{GC}/\mathrm{ml}$ diluted 1:5) was injected at $300\mu \mathrm{m}$ deep below the pial surface. In the case of wildtype mice, GCaMP6s virus (AAV2/1-syn-GCaMP6s-WPRE-SV40; $\sim 2.5\times {10}^{13}\mathrm{GC}/\mathrm{ml}$ diluted 1:10 in sterile phosphate-buffered saline) was also injected at the same time. The animals were allowed to recover for at least 3 weeks before imaging, which also allowed enough time for the virus to express in neurons.

2.2.

Microscopes and Lasers

The microscope used in this study was a Scientifica HoloStim3D (from Scientifica, United Kingdom), equipped with a fixed wavelength 920 nm laser for imaging (Axon 920 from Coherent Inc. or FemtoFiber Ultra 920 from Toptica Photonics) and a 1030 nm fixed wavelength laser for photostimulation (Satsuma ${\mathrm{HP}}^2$ from Amplitude Laser), each with their own dedicated scan path. The HoloStim3D was equipped with a spatial light modulator (SLM; $1920\times 1152$ calibrated for 1064 nm from Meadowlark Optics), which allowed for deflection of the photostimulation beam axially to the imaging plane without affecting the $z$ position of the imaging plane in the sample. ScanImage acquisition software (MBF Bioscience, Williston, Vermont, United States), written in MATLAB, was used to control all imaging and photostimulation parameters. The microscope was equipped with a 16× magnification objective (Nikon 16×, 0.8 NA, N16XLWD-PF from Thorlabs), which was used throughout all calibrations, measurements, and experiments.

2.3.

Varying and Measuring Optical Point Spread Function

The effective numerical aperture of the photostimulation beam path was altered by cropping the $1920\text{pixels}\times 1152\text{pixels}$ phase mask loaded onto the SLM. The undeflected zeroth order beam at the center of the photostimulation field of view (FOV) was blocked with an optical flat and a small piece of metal, so a single-spot hologram could be measured from the first order deflections. The PSF of the photostimulation beam was estimated by imaging $0.2\mu \mathrm{m}$ beads (TetraSpeck Microspheres T7280 from ThermoFisher Scientific) that were dried onto a microscope slide using ethanol. Images for PSF estimation were acquired at 16× magnification with $256\times 256\text{pixels}$, $0.1\mu \mathrm{m}$ pixel size, $1\mu \mathrm{m}$ between slices in the axial dimension, and 5 mW average power. Individual beads that were bright and did not get damaged by the imaging process were cropped and chosen for PSF estimation. The resultant stacks were registered using the FIJI³⁴ plugin StackReg³⁵ to remove any tilt in the PSF. The MetroloJ³⁶ plugin for ImageJ was used to fit a Gaussian curve to each individual bead in the axial and lateral dimensions. The curve was manually inspected for a good fit between the model and the data. At least three beads were imaged for each PSF to ensure that results were consistent.

2.4.

Photostimulation

To determine whether fast versus slow spiraling was better, fast spiraling was done using $25\times 10\mathrm{ms}$ spirals, whereas slow spiraling used $1\times 250\mathrm{ms}$ spiral. For the comparison, each neuron was measured with both fast and slow spiralings, in a random order. Slow spiraling was used for all experiments. 6 mW average laser power was used for photostimulation in these experiments.

Subsequently, to find the best photostimulation laser power, different neurons were targeted for each power of 2, 3, 4, and 5 mW average power, and no neuron was repeated with multiple laser powers.

For all further experiments, the 1030 nm photostimulation laser was used at 3 mW average power as measured by a sensor (S130C from Thorlabs) at the objective. The power was measured separately for each axial position and for each effective numerical aperture adjustment. The repetition rate of laser pulses was set to 2 MHz, with $\sim 350\mathrm{fs}$ pulse width and $\sim 4300\mathrm{W}$ peak power, and the energy per pulse was $\sim 1.5\mu \mathrm{J}$. The SLM was kept at a temperature of $\sim 25°\mathrm{C}$.

Each axial position for the photostimulation beam was controlled by adjusting the diffraction from the SLM to focus the hologram deeper or shallower relative to the imaging plane. The first-order deflection was $\sim 100\mu \mathrm{m}$ lateral to the zeroth order. The zeroth order was blocked at a conjugate plane with a piece of metal to reflect/absorb the beam. The exact positions were calibrated and measured by burning a spot into a plastic slide (Autofluorescent Plastic Slides from Chroma) at each depth. The spots ranged axially from $-60\mu \mathrm{m}$ to $+60\mu \mathrm{m}$ relative to the imaging plane in steps of $15\mu \mathrm{m}$ with 9 steps in total.

Using separate galvanometers from the imaging path, the single-spot hologram was spiraled with five concentric revolutions across the target neuron with a diameter of $10\mu \mathrm{m}$ at 3 mW average power. Repeats were carried out every 10 s to maintain photostimulation efficacy, with three repeats per neuron per axial position. The interval between each axial position was at least 30 s. Stimulations alternated between the “standard” direction from $-60$ to $+60\mu \mathrm{m}$ or the “reverse” direction from $+60$ to $-60\mu \mathrm{m}$, and the optical resolution was also alternated. Each neuron received stimulation in only one direction with only one optical resolution.

2.5.

Imaging

Responses to photostimulation were imaged at 30 Hz using resonant scanning and a wavelength of 920 nm. Images consisted of $512\times 512\mathrm{pixels}$ with a pixel size of $\sim 1.25\mu \mathrm{m}$. Two separate detectors (multialkali PMTs) were used to collect emission from C1V1-Kv2.1-mScarlet and GCaMP6s. Group delay dispersion (GDD) was optimized for the imaging laser by finding the highest fluorescent signal in a uniformly fluorescent sample while adjusting the GDD. To reduce imaging-induced photoactivation of C1V1, a large imaging FOV of $650\mu \mathrm{m}$ square was chosen to reduce the time that the imaging beam dwelt on each neuron. An average imaging laser power of 20 mW was chosen to further reduce the effects of imaging-induced photoactivation, and this power level was determined by scanning at different imaging powers and measuring the change in GCaMP fluorescence over a 30-s period. A power of 20 mW resulted in no ramp-up of activity across the duration of imaging (data not shown).

During in vivo experiments, animals were head-fixed while awake, and images were acquired at $150\mu \mathrm{m}$ deep in the somatosensory cortex (estimated layer 2). The cranial window was flattened by tipping/tilting the animal holder so that the window was perpendicular to the objective lens, resulting in less distortion of the images from refraction.

All imaging and photostimulation was synchronized using the voltage input/output recording software, PackIO.³⁷ A voltage trigger was used to begin imaging, and photostimulation was manually triggered from ScanImage. Both the frame clock and galvanometer output were recorded to detect spiral initiation times relative to imaging frames.

2.6.

Data Analysis

The voltage recordings of the frame clock and photostimulation trial onsets were used to find the trial start times. Mean raw fluorescence trials were calculated across the three repeats for each neuron, as described in the rest of this paragraph. Imaging data from different axial photostimulation positions were aligned to each other using ImageJ StackReg plugin to ensure that there was no drift between different acquisitions in the same FOV. The normalized change in fluorescence was calculated using the ImageJ Image Calculator tool as follows: $\mathrm{\Delta }F/F=(F-F_0)/F_0$, where $F$ is the raw fluorescence and $F_0$ is the mean $F$ in the baseline period (1.5 s before the stimulation onset). A freehand region of interest was drawn around the stimulated neuron using the ImageJ freehand drawing tool, and the mean $\mathrm{\Delta }F/F$ of all of the pixels was taken after importing the data back into Python. Control ROIs that were not photostimulated were drawn $\sim 100\mu \mathrm{m}$ away from the targeted neuron with the same shape as the neuronal ROI and analyzed in the same way. The response for each trial was taken as the mean $\mathrm{\Delta }F/F$ of the period post-stimulation (from 0 to 1.5 s after the photostimulation ended). Any neuron that did not respond at a single-axial position with a mean of $\ge 0.35$ $\mathrm{\Delta }F/F$ in the 1.5 s post-photostimulation was removed as it suggested that the photostimulation was unsuccessful. This was based on the average response to stimulations in the 3 mW condition of the laser power photostimulation tests [0.37 $\mathrm{\Delta }F/F$ versus 0.15 $\mathrm{\Delta }F/F$ for sample versus control regions; Figs. 2(b) and 2(c)]. There is no photostimulation artifact reduction strategy to prevent fluorescence from the photostimulation laser excitation appearing in the imaging channel for the duration of the photostimulation period. Overall, 15/31 neurons were included for the large OPSF, and 17/38 neurons were included for the small one.

Due to the uncertainty of the exact position of the neurons in the axial dimension, the photostimulation precision curves were offset so that the maximum response $z$ offset was relabeled as $0\mu \mathrm{m}$. No curve was shifted by more than $15\mu \mathrm{m}$ (one SLM $z$ offset position). The photostimulation precision for the different optical resolutions was measured as the FWHM (the width of a horizontal line intercepting the signal at half the prominence of the peak of the curve) using the find_peaks and peak_widths functions as part of SciPy’s signal processing module for Python. Standard ($-60$ to $+60\mu \mathrm{m}$) and reverse ($+60$ to $-60\mu \mathrm{m}$) photostimulation response curves were combined by flipping the reverse curves, so the positions matched the standard condition.

For $\mathrm{\Delta }F/F$ timeseries plotting, a rolling average of 5 frames was used. All error bars are $\pm 95\%$ confidence interval, unless otherwise stated.

2.7.

Modeling

MATLAB was used to generate mock data for an axial OPSF from the experimental results. For each OPSF, a standard deviation was calculated from the FWHM value, and a Gaussian distribution was generated from random numbers using that standard deviation. The data were normalized to the maximum values, and a moving average filter of 5 was applied. Cell sizes were modeled as binary square waves, in which the $y$ axis was 1 across the entire cell diameter on the $x$ axis. Each OPSF was convolved with the square wave of each cell size using the convolve function. The code and processed data are freely available to repeat these measurements and can be found at https://github.com/Packer-Lab/OPSF_vs_PPSF

3.1.

Varying the Photostimulation OPSF Axial Spread Using a Spatial Light Modulator

We used an SLM to vary the size of the photostimulation OPSF and control its $z$ position relative to the targeted neuron. The phase masks being applied to the SLM active area resulted in a single focused laser spot at the sample, roughly $100\mu \mathrm{m}$ from the center of the photostimulation FOV. To elongate the point spread function axially, we cropped the phase masks [Fig. 1(a)], effectively underfilling the back aperture of the objective. We imaged submicron fluorescent beads using the 1030 nm photostimulation beam of the SLM path to measure the axial extent of the OPSF produced by these phase masks. The “small” OPSF had an axial FWHM of $8.3\mu \mathrm{m}$, and the “large” OPSF had an FWHM of $18.6\mu \mathrm{m}$ [large OPSF, $n=3$ beads; small OPSF, $n=4$ beads; Fig. 1(b)]. We offset the $z$ position of the OPSF in $15\mu \mathrm{m}$ steps from $-60$ to $+60\mu \mathrm{m}$ relative to the imaging plane by creating a new phase mask for each position. This was confirmed by burning the resultant holograms into a fluorescent plastic slide and measuring the $z$ position of the burns relative to the imaging plane using the motorized stage (data not shown). In summary, we created large and small OPSFs that we could position in three dimensions to target individual neurons and test the effect of axial OPSF size on photostimulation precision.

Fig. 1

Varying the axial spread of the photostimulation OPSF using an SLM. (a) Example phase masks applied to the SLM active area. These phase masks were used to vary the axial spread of OPSFs; the examples are from the $0\mu \mathrm{m}$ photostimulation $z$ position. (b) Mean Gaussian curves $\pm 95\%$ CI (confidence interval) fitted to the $z$ profile of $0.2\mu \mathrm{m}$ fluorescent beads imaged using the photostimulation OPSF (large OPSF, $n=3$ beads; small OPSF, $n=4$ beads). The FWHM is indicated (dashed line), showing the axial extent of the OPSF. Inset is example images from a bead imaged with the “large” (blue outline) and “small” (orange outline) OPSFs. The images are $xz$ projections; brightness and contrast have been adjusted separately for the two images to highlight the shape.

3.2.

Optimizing Photostimulation for Detection of Nonsaturated Neuronal Responses

We needed to detect small changes in the neuron’s response to photostimulation to produce a curve of neuronal response versus the $z$ offset of the photostimulation beam. To do this, we optimized the average power and frequency of photostimulation to prevent saturation of the neuronal response (as measured using calcium indicators). In this study, we used the normalized change in fluorescence ($\mathrm{\Delta }F/F$) of a common calcium indicator, GCaMP6s, as a proxy for neuronal response. We expressed GCaMP6s in neurons in the somatosensory cortex of mice alongside a perisomatic red-shifted opsin, C1V1-Kv2.1, conjugated to the fluorescent protein mScarlet in pyramidal cells only [Fig. 2(a)]. We identified single mScarlet-positive pyramidal neurons in the field of view to target for photostimulation.

Fig. 2

Optimizing photostimulation parameters for the detection of nonsaturated neuronal responses. (a) Left: example two-photon fluorescent image in layer 2 of the mouse somatosensory cortex. Neurons are labeled through viral expression with GCaMP6s and C1V1-Kv2.1-mScarlet (solid box, targeted neuron; dashed box, control region). Top-right: zoom of the targeted neuron showing the C1V1-Kv2.1-mScarlet labeling. An example spiral of $10\mu \mathrm{m}$ diameter is overlaid on the targeted neuron. Bottom-right: trial-averaged image of the targeted neuron in units of normalized change in fluorescence ($\mathrm{\Delta }F/F$) averaged across a period of 1.5 s post-photostimulation. (b) Top: mean $\mathrm{\Delta }F/F$ traces $\pm 95\%$ CI for all neurons stimulated at each power level (different neurons for each power level; 2 mW, 3 mW, $n=10$ neurons each; 4 mW, 5 mW, $n=9$ neurons each). Bottom: mean $\mathrm{\Delta }F/F$ traces $\pm 95\%$ CI for all control regions (dashed box in the left panel of (a)). (c) Mean $\mathrm{\Delta }F/F$ traces $\pm 95\%$ CI for all neurons in each spiral condition (every cell was tested with both spiral conditions, $n=24$ cells). Bottom: voltage traces from the $x$ galvanometric mirror that spirals the photostimulation beam showing the single-slow spiral ($1\times 250\mathrm{ms}$) and the fast repeated spirals ($10\times 25\mathrm{ms}$). The response post-photostimulation was not significantly different (independent $t$-test for $1\times 250\mathrm{ms}$ versus $10\times 25\mathrm{ms}$: $t=-1.06$, $p=0.293$).

First, we wanted to determine if the opsin was more effectively activated by repeatedly scanning the neuron with 10 fast spirals (25 ms) or a single slow (250 ms) spiral. For each neuron, we spiraled the photostimulation beam over the cell without any $z$ offset while simultaneously imaging the GCaMP6s response at 920 nm. We stimulated the same neurons with fast and slow spiral scans, alternating which stimulation was tested first for each neuron. There was no significant difference in calcium response between slow spiraling and multiple fast spirals [Fig. 2(c)]. However, we did this test at 6 mW average power, which was potentially already a saturating power. We decided to use slow spirals ($1\times 250\mathrm{ms}$) for the experiments as this produced a slightly lower response (nonsignificant; independent $t$-test for $1\times 250\mathrm{ms}$ versus $10\times 25\mathrm{ms}$: $t=-1.06$, $\mathrm{p}=0.293$).

We adjusted the average power of the photostimulation beam in 1 mW steps from 2 to 5 mW as we noticed that 6 mW was resulting in very large, possibly saturating responses ($>3\mathrm{\Delta }F/F$). Photostimulation of each neuron was repeated for three trials at each power level. The results showed that 4 and 5 mW average power of the photostimulation beam achieved similar responses, potentially indicating that the neurons had reached a maximal response at these powers [Fig. 2(b)]. We chose 3 mW average power for all further photostimulation experiments as the responses at 2 mW were too close to the noise floor estimated from nonphotostimulated control regions ($100\mu \mathrm{m}$ from the photostimulation beam; see Sec. 2).

In summary, we chose a photostimulation laser power that produced submaximal responses at the central photostimulation plane ($z=0\mu \mathrm{m}$) to ensure that we could accurately sample the curve of neuronal GCaMP6s response at different photostimulation $z$ offsets.

3.3.

Measuring the Effect of OPSF Size on Neuronal Photostimulation Precision

We wanted to test whether the axial spread of the OPSF affected the precision with which neurons can be photostimulated in vivo. To do this, we stimulated single neurons using different photostimulation $z$ offsets with two different OPSFs. We again used mice expressing GCaMP6s and C1V1-Kv2.1 in neurons of the somatosensory cortex [Fig. 2(a)]. We varied the $z$ offset of the photostimulation OPSF while simultaneously imaging the GCaMP response from the targeted neuron at $z=0\mu \mathrm{m}$, repeating three trials at each photostimulation $z$ position. To account for any effects of opsin desensitization due to repeated exposure, the photostimulation $z$ offset was either sequentially moved from $+60$ to $-60\mu \mathrm{m}$ (“standard” direction) or from $-60$ to $+60\mu \mathrm{m}$ [“reverse” direction; Fig. 3(a)]. Each neuron was only tested for a single OPSF size and a single direction (standard or reverse). This was done to ensure that as many different neurons were sampled as possible and to prevent as far as possible the potential for photodamage caused by excessive repeated trials. The resultant response from each neuron was recorded and the normalized change in fluorescence ($\mathrm{\Delta }F/F$) was measured post-stimulation ($n=8$ cells for each direction for small OPSF, $n=7$ cells for reverse, $n=8$ cells for standard for the large OPSF). The neuronal responses were highest at positions around $z=0\mu \mathrm{m}$ and dropped to undetectable levels at the extreme $z$ positions [Figs. 3(b) and 3(c)], implying a complete estimation of the entire response curve.

Fig. 3

Accuracy of neuronal photostimulation in mouse somatosensory cortex layer 2 is not affected by variations in commonly used OPSF sizes. (a) Schematic of the different directions of photostimulation $z$ travel. The photostimulation OPSF was stepped sequentially through the $z$ dimension from top-to-bottom ($+60$ to $-60\mu \mathrm{m}$ in $15\mu \mathrm{m}$ steps; “standard”) or bottom-to-top (“reverse”) by changing the hologram produced by the SLM. Three trials were repeated at each $z$ position before moving to the next one. (b) Example mean $\mathrm{\Delta }F/F$ images of post-photostimulation responses (mean of 1.5 s after stimulation end) for a targeted neuron that was photostimulated at each $z$ position while being imaged at a fixed imaging plane ($z=0\mu \mathrm{m}$). $\text{Scale bar}=20\mu \mathrm{m}$. (c) Left: $\mathrm{\Delta }F/F$ traces from a targeted neuron showing responses at each photostimulation $z$ position. Right: the $\mathrm{\Delta }F/F$ values in the period 1.5 s post-photostimulation were averaged to produce a neuronal response curve for axial photostimulation precision. (d) Left: curves showing mean post-photostimulation responses $\pm 95\%$ CI at each photostimulation $z$ position for the large OPSF, for the two different directions (different cells for each direction; reverse, $n=7$ cells; standard, $n=14$ cells). The standard and reverse conditions were not statistically different (independent $t$-test for standard versus reverse at each $z$ position: $p>0.05$). Middle: curves from the small OPSF (reverse, $n=11$ cells; standard, $n=14$ cells). Right: mean normalized photostimulation response $\pm 95\%$ CI versus photostimulation $z$ position, a summary of the previous four curves for the large and small OPSFs. The half-width at half-maximum is indicated (dashed line) and not different for the two curves (independent $t$-test for large versus small OPSFs at each $z$ position: $p>0.05$).

We compared the response curves from the standard and reverse directions of both OPSF sizes. Neurons that failed to respond with $>0.35$ $\mathrm{\Delta }F/F$ in at least one $z$ position were removed, so only responsive neurons were included (based on average responses from sample versus control regions; see Sec. 2.6). Additionally, due to the variation in the axial position of the imaging plane relative to the center of the targeted neuron soma, we aligned individual response curves by their maximum response [Fig. 3(d)]. The maximum offset of a curve was one $z$ position ($15\mu \mathrm{m}$), which is about the size of a cell body and fits with the expected degree of uncertainty about the axial position when imaging neuronal cell bodies, that is, the identification of a cell’s axial position in the imaging plane was difficult to determine at a low zoom (required to avoid imaging-induced photostimulation; see Sec. 2.5), so the $z=0\mu \mathrm{m}$ plane may have been at the very top or bottom of a cell. The standard and reverse directions were not significantly different from each other for either OPSF size (independent $t$ test for standard versus reverse at each $z$ position: $p>0.05$), so they were combined to improve the statistical power. Similarly, the two halves of the curve (above and below the cell) were combined; however, we do note that the shape of the curve above the neuron was steeper for both OPSF sizes. The FWHM of the photostimulation response curve for the large and small OPSFs was 44 and $42\mu \mathrm{m}$, respectively [independent $t$-test for large versus small OPSFs at each $z$ position: $p>0.05$; Fig. 3(e)].

Our results show that the PPSF does not depend on the axial extent of the OPSF for the OPSFs that we measured and the biological system that we utilized (pyramidal cells in layer 2 of mouse neocortex). To further understand this result, we modeled the contribution of the cell size to the axial extent of the PPSF (Fig. 4). We used the width of a square pulse to model the cell diameter and convolved it with the axial extent of the large and small OPSFs to give the PPSF [Fig. 4(b)]. The results show that, as the cell diameter increases beyond the size of the OPSF axial FWHM, the PPSF axial FWHM begins to depend solely on the cell diameter [Fig. 4(c)]. Beyond a certain size ($\sim 30\mu \mathrm{m}$), both OPSF sizes converge on the same PPSF, breaking the dependence of PPSF on OPSF.

Fig. 4

Modeling the physiological PSF for different cell shapes. (a) Hypothesis that the PPSF results from the convolution of the cell shape and the optical PSF shape. (b) The cell diameter was used to model the cell shape as a sphere, and the axial FWHM from the fluorescence signal of the OPSF was used to model the contribution of the OPSF. The cell diameter was modeled as a square wave, and the axial FWHM was a Gaussian. (c) A plot of the relationship between PPSF and cell diameter for the large (blue stars) and small (orange stars) OPSFs. Experimental results from this study (gray) and other literature using Kv2.1 (green) are included. The line of equality is also plotted (solid black line).