Astrocytes exhibit localized Ca²⁺ microdomain (MD) activity thought to be actively involved in information processing in the brain. However, functional organization of Ca²⁺ MDs in space and time in relationship to behavior and neuronal activity is poorly understood. Here, we first show that Adeno-Associated Virus (AAV) particles transfer anterogradely from axons to astrocytes. Then we use this axo-astrocytic AAV transfer to express genetically encoded Ca²⁺ indicators at high contrast circuit-specifically. In combination with two-photon microscopy and unbiased, event-based analysis we investigated cortical astrocytes embedded in the vibrissal thalamocortical circuit. We found a wide range of Ca²⁺ MD signals, some of which were ultrafast (≤300 ms). Frequency and size of signals were extensively increased by locomotion but only subtly with sensory stimulation. The overlay of these signals resulted in behavior dependent maps with characteristic Ca²⁺ activity hotspots, maybe representing memory engrams. These functional subdomains are stable over days, suggesting subcellular specialization.

All experimental procedures were approved by the OIST Institutional Animal Care and Use Committee (IACUC) in an Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC International) accredited facility.

Animals

Male, 1 to 3-month-old C57/BL6-J mice (Japan Clea) were used. Experiments were performed during the dark period of the 12/12-h dark/light cycle.

Surgeries

Mice were anesthetized by an intraperitoneal (ip) injection of a mixture of Medetomidine (0.3 mg/kg), Midazolam (4 mg/kg), and Butorphanol (5 mg/kg) in 0.9% saline solution. Carprofen (5 μg/g; ip), Dexamethasone (2 μg/g; intramuscular) and Buprenorphine (0.1 μg/g; subcutaneous) were administered to reduce inflammation, immune response, and pain, respectively. Following anesthesia induction, the eyes of the mouse were covered with mycochlorin eye ointment (Sato Pharmaceutical Co., Ltd). The mouse head hair was removed with a hair trimmer and hair removal cream (Veet®). The head of the animal was fixed on a stereotactic apparatus and its skin was sterilized with iodine. If further anesthesia was needed, 1% isoflurane was administered.

In preparations without chronic window implantation, a single midline sagittal incision of the scalp was performed. The exposed skull was cleaned with the local anesthetic lidocaine. The injection coordinates were based on the Paxinos’ Mouse Brain Atlas (53), and previous calibration injections. The skull was thinned at the desired coordinates using a dental drill (OS-40, Osada) with a diamond drill bit (TRUSCO, T2-532M). A sterile glass pipette with a broken tip was used to puncture the thinned skull at the coordinates established to minimize the exposure of the brain. Using beveled glass pipettes (10-15μm tip diameter), AAVs were injected into ventral posteromedial nucleus (VPM; 1.8 mm posterior, 1.7 mm lateral, 3.5 mm below dura) and for some experiments additionally in barrel cortex (BX) (1.8 mm posterior, 3.2 mm lateral, 0.5 mm below dura) for intersectional strategies. AAVs were injected at a rate of ~70 nl/ 5 min by air pressure application. The pipette was left in place for at least 5 min before and after injection. After injections, the cut skin was closed using superglue. Animals were housed individually and allowed to recover for at least one week, unless otherwise stated.

For chronic window implantation surgeries (54), the skin was sterilized with iodine before cutting it to expose the skull. The skull was cleaned with lidocaine solution (2% in saline). The region above the injection site in BX was marked, and a circle of ~2 mm radius was drawn around the marking. The skull surrounding the marked circle was thinned with the drill. The bone above the intended craniotomy was gently lifted after gluing a wooden cotton swap with superglue gel to the bone, to expose the dura without damaging it (55). Any bleeding was cleaned carefully with Carprofen soaked gelfoam (Pfizer). AAVs were injected into VPM only or both VPM and BX (intersectional strategy) using beveled glass pipettes (10-15 μm tip diameter), at VPM: 1.8 mm posterior, 1.7 mm lateral, 3.5 mm below dura, and BX: 1.8 mm posterior, 3.2 mm lateral, 0.5 mm below dura. In all intersectional strategies, AAVs were first injected in the VPM followed immediately be a second injection in BX. AAVs were injected at a rate of ~70 nl/ 5 min by air pressure application. The pipette was left in place for at least 5 min before and after injection. A 5-mm diameter glass coverslip was placed on the craniotomy and fixed to the bone with super glue. Additionally, an aluminum head plate was mounted and fixed with dental cement (Super-Bond). Super-Bond was also used to cover any exposed skull. The animals were individually housed and allowed to recover for at least one week.

Adeno-associated viruses

Intersectional strategy – astrocytes: To assess whether AAV1 tags astrocytes anterograde the injection site, 140 nl (1:1 ratio) AAV2/1-CMV-Pl-Cre-rBG (1.2x10¹³ GC/mL) and AAV2/1-hSyn-TurboRFP-WPRE-rBG (3.9x10¹³ GC/mL) (University of Pennsylvania Viral Vector Core) were co-injected in VPM and 100 nl AAV2/5-GFaABC1D-Flex-Lck-GCaMP6f-WPRE-SV40 (1.0x10¹³ GC/mL) (Sirion Biotech) was injected in BX.

For sparse labelling of BX astrocytes with membrane tagged GCaMP6f for in vivo imaging we injected: 140nl AAV2/1-CMV-Pl-Cre-rBG (1.2x10¹³ GC/mL; University of Pennsylvania Viral Vector Core) in VPM and 140nl AAV2/5-GFaABC1D-Flex-Lck-GCaMP6f-WPRE-SV40 (1.0x10¹³ GC/mL; Sirion Biotech) in BX.

Intersectional strategy – neurons: To assess if a similar intersectional strategy can be used to label neurons, 140 nl (1:1 ratio) AAV2/1-CMV-Pl-Cre-rBG (1.2x10¹³ GC/mL) and AAV2/1-hSyn-TurboRFP-WPRE-rBG (3.9x10¹³ GC/mL) (University of Pennsylvania Viral Vector Core) were co-injected in VPM and 100nl AAV9-Syn-Flex-GCaMP6f-WPRE-SV40 (2.8x10¹³ GC/mL) (University of Pennsylvania Viral Vector Core) was injected in BX.

Single VPM injection strategies: To test whether injection in VPM alone is sufficient for inducing cell labelling in BX and what elements of the AAV vector are needed to do so, 140 nl (1:1 ratio) of different AAV combinations were injected only in VPM.

Simple strategy: AAV2/1-hSyn-TurboRFP-WPRE-rBG and AAV2/1-CAG-GCaMP6f-WPRE-SV40 (1.33x10¹³ GC/mL) (University of Pennsylvania Viral Vector Core).

Combinatorial strategy: AAV2/1-CMV-Pl-Cre-rBG and AAV2/1-CAG-Flex-eGFP-WPRE-bGH (9.16x10¹³ GC/mL) (University of Pennsylvania Viral Vector Core). Or: AAV2/1-CMV-Pl-Cre-rBG and AAV2/1-CAG-Flex-tdTomato-WPRE-rBG (4.65x10¹³ GC/mL) (University of Pennsylvania Viral Vector Core).

For sparse labelling of BX astrocytes and neurons along with thalamocortical axons for in vivo imaging, we injected 120nl (1:1:1 ratio) AAV2/1-CMV-Pl-Cre-rBG (1.2x10¹³ GC/mL), AAV2/1-CAG-Flex-GCaMP6f-WPRE-SV40 (1.33x10¹³ GC/mL) and AAV2/1.hSyn-TurboRFP-WPRE-rBG (3.9x10¹³ GC/mL) (University of Pennsylvania Viral Vector Core) in VPM.

AAV capsid tracking experiment

To track AAV capsids we used two groups of mice (n=3 each). Each mouse was injected twice, i.e. in each VPM (left and right) once: The first group got the first injection 12 days before the perfusion and the second injection into the contralateral VPM 1 h before the perfusion. The second group received its first injection 24 h before the perfusion and the second the injection into the contralateral VPM 1 h before the perfusion. Injections 1 h before perfusion were used as positive controls for the Anti-VP1 antibody to label AAV capsids at the injection site, within 1 h post injection. One injection of 140 nl, AAV2/1-CMV-Pl-Cre-rBG and AAV2/1-CAG-Flex-tdTomato-WPRE-rBG (1:1 ratio) was injected per VPM of each mouse.

Brain slice preparation. Mice were deeply anesthetized with a mixture of Medetomidine (0.3mg/kg), Midazolam (4 mg/kg), and Butorphanol (5 mg/kg) in 0.9% saline solution and transcardially perfused with PBS followed by PLP (4% paraformaldehyde, 1.2% lysine, and 0.2% periodate in 0.1 M phosphate buffer) fixation. The brains were extracted and stored in PLP at 4ºC for a minimum of 48 h. For extended storage PLP was replaced with PBS. A vibratome (VT1000S, Leica) was used to cut 100 μm thick coronal slices. The slices were mounted on glass slides with Mowiol and stored at 4º C.

For immunohistochemistry preparation, PLP fixed brains were cryoprotected using 20% sucrose solution in PBS overnight. The brains were then trimmed and embedded in optimal cutting temperature (OCT) compound (Tissue-Tek). The samples were frozen at -80º C for 1 h. A cryotome (Leica, CM3050 S) was used to cut 50 μm thick coronal slices at -15º C which were immediately transferred into wells filled with PBS.

Immunohistochemistry - AAV capsid tracking. We used immunohistochemistry to tag AAV capsids with anti-VP1 antibody (Antibodies-online, ABIN933221). Brain sections cut with the cryotome were washed in PBS and then incubated in 20% normal goat serum in permeabilization solution (0.3% Triton-X-100, 0.05% sodium azide, PBS) for 1 h. The serum was replaced with 1:20 mouse Anti-AAV1 monoclonal antibody diluted in permeabilization solution. The slices were incubated for 24 h at 4º C. The samples were washed with PBS and incubated in goat anti-mouse polyclonal secondary antibody Alexa Fluor 488 (abcam, ab150113) 1:200 (diluted in permeabilization solution) for 2-3 h, in the dark, at room temperature. The brain slices were washed with PBS, mounted on glass slides with DAPI Antifade Mounting Medium (VECTASHIELD), and stored at 4ºC.

Immunohistochemistry – S100β and NeuN staining. We used anti-S100β antibody (rabbit monoclonal, EP1576Y, abcam, ab52642) as an astrocyte specific marker. 50 μm thick coronal slices (see brain slice preparation) were prepared from mice sacrificed 3 weeks after intersectional injection of 140 nl (1:1 ratio) AAV1-CMV-Cre and AAV1-hSyn-TurboRFP in VPM and 100 nl AAV5-GFaABC1D-FLEx-Lck-GCaMP6f in BX (see intersectional strategy – astrocytes). The brain sections were washed 4 times in PBS and then incubated in 20% normal donkey serum in permeabilization solution (0.3% Triton X-100, 0.02% sodium azide, PBS) for 1 h. The serum was replaced with 1:500 anti-S100β antibody diluted in permeabilization solution. The slices were incubated for 24 h at 4º C. The samples were washed 4 times with PBS and incubated with the secondary antibody (donkey anti-rabbit polyclonal antibody, Alexa Fluor 647, abcam ab150075) 1:200 (diluted in permeabilization solution) for 2-3 h, in the dark, at room temperature. The brain slices were washed 4 times with PBS, mounted on glass slides with DAPI Antifade Mounting Medium (VECTASHIELD), and stored at 4º C.

We used anti-NeuN antibody (rabbit monoclonal, EPR12763, abcam ab177487) as a neuron specific marker. 50 μm thick coronal slices (see brain slice preparation) were prepared from mice sacrificed 3 weeks after intersectional injection of 140 nl (1:1 ratio) AAV1-CMV-Cre and AAV1-hSyn-TurboRFP in VPM and 100nl AAV9-Syn-FLEx-GCaMP6f in BX (see intersectional strategy – neurons). The brain sections were washed 4 times in PBS and then incubated in 20% normal donkey serum in permeabilization solution (0.3% Triton X-100, 0.02% sodium azide, PBS) for 1 h. The serum was replaced with 1:1000 anti-NeuN antibody diluted in permeabilization solution. The slices were incubated for 24 h at 4º C. The samples were washed 4 times with PBS and incubated with secondary antibody (donkey anti-rabbit, polyclonal antibody, Alexa Fluor 647, abcam ab150075) 1:200 (diluted in permeabilization solution) for 2-3 h, in the dark, at room temperature. The brain slices were washed with PBS, mounted on glass slides with DAPI Antifade Mounting Medium (VECTASHIELD), and stored at 4º C.

Confocal imaging and analysis in fixed slices. LSM 510 META ConfoCor3 (Carl Zeiss) and LSM 710 (Carl Zeiss) confocal microscopes were used to image fixed brain slices. LSM 510 was used for all ex vivo imaging, except for antibody-labelled capsid samples in the cortex. For LSM 510, a 488 nm argon laser and a 561 nm Diode-Pumped Solid-State (DPSS) laser were used to excite green fluorophores (eGFP, GCaMP6f, Alexa Fluor 488) and red fluorophores (tdTomato, TurboRFP) respectively. A 405 nm diode was used to excite DAPI. Immuno-stained AAV capsids in BX were imaged with LSM 710 using a 63x /N.A. 1.46 a-Plan-Apochromat oil objective.

Quantification of capsid puncta density in BX. tdTomato⁺ astrocyte and neuron cell bodies (n =15) were randomly selected from 3 brain slices mainly in layer 2/3 and 4 (n = 3 mice) ~1700 μm lateral of the injection tract. We captured 3D image stacks of 67.5 x 67.5 μm² field of view (16 bit, 1024 x 1024 pixel), 0.7 μm interspaced from each other at 0.84 AU and averaged 8 times. Alexa-Fluor 488 and tdTomato were excited using a 488 nm Argon laser and 543 nm Helium-Neon laser, respectively. For control, 3D stacks with the same parameters were acquired on contralateral hemisphere BX (~1700 μm lateral; ipsilateral the control VPM injection). To quantify the number of capsid puncta, images were first converted to 8 bit and a 1-1-1-pixel 3D Gaussian filter was applied. Brightness, contrast, and threshold parameters were established empirically and kept constant. Fluorescent puncta of 0.1 - 0.5 μm² in size were automatically counted using Fiji (49) plugin Analyze Particles. The size-range parameter was based on empirical observations of puncta sizes since they could be easily identified by eye. Unpaired two tailed t-tests were used to compare the puncta density between a) BX cell bodies (n = 15) vs their local background (n=12) and b) BX background vs background of contralateral BX, that was devoid of tdTomato⁺ cell bodies (1 h after injection of its ipsilateral, control VPM; n = 10). The sampling volume was calculated to be equal to the image area x 0.7 μm x number of Z sections. Puncta density was calculated to be the number of puncta / sampling volume.

Quantification of cell types. A high-resolution confocal microscope LSM 880 Airyscan (Carl Zeiss) was used to image fixed brain slices. A 488 nm argon, 561 nm Diode-Pumped Solid-State (DPSS) and 633 nm helium-neon lasers were used to excite GCaMP6f, TurboRFP and Alexa Fluor 647 respectively. A 405 nm diode was used to excite DAPI. Images were taken with a 20x objective.

Astrocyte test: we randomly selected the nuclei (DAPI⁺) of BX cells expressing GCaMP6f with a cloud-like morphology in anti-S100β stained brain slices (n=3/mouse, 3 mice). For each cell, we captured a 3D image stack of 142 x 142 x 7 μm (1024 x1024 pixel, averaged 4 times, 1 μm interspaced). We recorded which cells were S100β ⁺ and which ones were not. We reported the percentage of S100β ⁺ and S100β ^– cells.

Neuron test: we randomly selected a field of view with cell bodies expressing GCaMP6f in BX in anti-NeuN stained brain slices (n=3/mouse, 3 mice). For each slice, we randomly selected a volume of 283 x 283 x 25 μm, 1μm interspaced (1024 x 1024 pixel, averaged 4 times). In this volume we counted the GCaMP6f⁺ cell bodies (~6 cells per slice) and identified which cells were NeuN ⁺ and which ones were not. We reported the percentage of NeuN ⁺and NeuN ^– cells.

Quantification of astrocyte and neuron density in BX. We used brain slices from mice injected using intersectional strategies to label astrocytes (see intersectional strategy – astrocytes, n = 3 slices/mouse, 4 mice) or neurons (see intersectional strategy – neurons, n = 3 slices/mouse, 3 mice). We used confocal-tile scanning to image GCaMP6f⁺ cells in the BX. We manually counted all the cell bodies in the BX column with the most GCaMP6f⁺ cells using Zen Lite software (ZEISS). We divided the number of cells by the volume of each column to estimate the density of astrocytes and neurons. Furthermore, we divided each column into L1, L2/3, L4, and L5/6 based on (56) (~total width: 1125 μm, L1: 37 μm, L2/3: 234 μm, L4: 259 μm and L5/6: 595 μm, scaled). We divided the number of labelled cells by the volume of each layer to estimate the density of astrocytes and neurons in each preparation respectively. The characteristic barrel-like projection patterns of TurboRFP⁺ thalamocortical neurons was used as assistance in delineating cortical columns and layers.

Two photon imaging in behaving animals

One week after chronic cranial window implantation, mice were habituated by periodic handling (~20 min daily for ~1 week) and exposure to head-fixation on a cylindrical treadmill (r = 7.4 cm). A vibrissa stimulus was introduced by inserting a wooden rod (toothpick segment) in the treadmill to intercept (one stimulus per rotation, equaling one stimulus every 46.5 cm) the vibrissae contralateral the injection site while the animal run. Animals were briefly anesthetized with 2% isoflurane before head fixation and left to rest and recover from anesthesia for ~10 min before two-photon (2P) imaging. In vivo 2P microscopy was used to image Ca²⁺ signaling ~3 weeks after injection (depending on the expression of GCaMP6f). A CCD camera and infrared light source (920 nm, Thorlabs) were used to monitor behavioral activity (focused on the vibrissa pad of the animal) and vibrissa stimulation by the wooden rod. A rotary encoder (E6A2, Omron) was used to record locomotion.

In vivo imaging was performed using a custom-built combined wide field and 2P microscope (MOM, Sutter Instruments) with a 25x/NA 1.05 water immersion objective with 2 mm working distance (Olympus). A Ti:sapphire femtosecond pulsed laser (Vision II, Coherent) was used to excite fluorescence at 950 nm (power measured after the objective < 60 mW). The back aperture of the objective was underfilled to elongate the point spread function (PSF) of excitation to ~4-5 μm in axial direction and 1 μm in the imaging plane (39). A resonant scanner was used to acquire images at 30.9 Hz. Fluorescence was detected by two GaAsP photomultiplier tubes (Hamamatsu) in the spectral range of 490–560 nm (green) and 570–640 nm (red) separated by a 565-nm dichroic mirror (Chroma). Commercial software (MScan, Sutter Instruments) controlled the microscope, analog channels, and CCD camera. Imaging was done ~300 μm below dura (L 2/3) in BX with 512 x 512 pixels and 94 x 94μm² field of view. To image the same astrocyte repeatedly, a 5x/NA 0.25 air objective (Zeiss) was used followed by the 25x objective to map the blood vessel pattern on the brain surface directly above the astrocyte of interest. To find the same z-plane over consecutive days (1 - 3 days after the first recording) we used depth information, autofluorescent puncta patterns, the morphology of the astrocyte, and the blood vessel pattern passing through the astrocyte.

Multiple pole experiment. 3 more mice were used by a different experimenter to investigate the effect of high frequency stimulation on astrocyte Ca²⁺ signaling. Mice were habituates (see above) on a treadmill with 23-poles equally interspaced from each other (2 cm) along the circumference stimulating the vibrissae contralateral of the imaging site. 1 astrocyte per mouse was recorded as before, 2 times/day across 2 consecutive days (n = 4 recording/ astrocyte, 12 recordings total). During the imaging, mice were free to run on the treadmill with either 1-pole or 23-poles. After the 1^st recording (~1 h long), poles were added or subtracted accordingly, and the same astrocyte was recorded for ~1 more h. Event characteristics (frequency, amplitude, duration, size) were collected during these recordings. We compared the events collected during (a) locomotion without pole interaction with vibrissa, (b) locomotion with 1-pole interaction (within 1.5s of interaction) and (c) locomotion on 23-pole treadmill. We divided the mean event characteristics obtained during each state by the respective mean event characteristics obtained during the respective recording while the animal was at rest. We performed this normalization to minimize inter-recording variability of baseline activity. A notable difference was a cutoff frequency of 3 Hz used to compensate for lower signal-to-noise recording to minimize bleaching and photo toxicity. 30.9 Hz recordings were averaged 9 times instead of 3 times like other experiments to increase the S/N ratio sufficiently to ensure correct event detection by AQuA.

In vivo data analysis

Behavioral data extraction. A CCD camera was used to detect vibrissa stimuli. The sharp change in infrared light intensity detected by the camera at the region of the vibrissa pad contralateral the injection side indicated the transit. The vibrissa pad region was manually selected in Fiji (49). A threshold light intensity binarized the vibrissa stimulation events (0 = no vibrissa interception, 1 = vibrissa interception).

The behavioral state of the animal was determined as a binary signal: running or resting. Movement of the mouse on the treadmill was extracted from the analog signal of the rotary encoder. If the running mouse stopped for less than 1.5 s and then continued running, its state was considered continuous running. If the mouse stopped for more than 1.5 s, its state was defined as at rest. If the wooden rod intercepted the vibrissae, it was defined as a vibrissa stimulus. In most cases, the animal was running when the wooden rod intercepted its vibrissae, defined as vibrissa stimulation (VS). Sometimes, the animal would stop and explore the wooden rod with its vibrissae during rest, defined as vibrissa exploration (VE).

Preprocessing. Two-photon microscopy (2P) movies (30.9 Hz), rotary encoder analog signals, and animal behavior movies were synchronously recorded using MScan software. Proprietary video format files (MDF, Sutter Instruments Inc.) were converted to Tiff using commercial software (MView, Sutter Instrument). Time lapse recordings were preprocessed using Fiji (49). Movement artifacts in the imaging plane was corrected using the TurboReg (50) plugin (batch mode, translation mapping) and a custom macro to automate the process. Videos were then visually inspected to confirm movement artifact correction. A 3D Gaussian filter with sigma = 1 pixel in x, y, and t was applied to all 2P movie stacks using Fiji. The 30.9 Hz raw video was binned to 10.3 Hz.

AQuA processing. AQuA (Automatic Quantification and Analysis) (23) was used for unbiased identification and characterization of astrocyte Ca²⁺ MD signals. AQuA applies machine learning techniques to model Ca²⁺ events in an event-based, data driven way that does not impose a priori assumptions about the data. All preprocessed time lapse movies used for AQuA processing had a nominal spatial resolution of ~0.18 μm/pixel (optical resolution based on the PSF of the microscope: 1.0μm; therefore, 5-fold oversampling) and a temporal resolution of 10.3 Hz. AQuA performed further preprocessing by applying a Gaussian filter (smoXY = 2 SD). AQuA estimated the standard deviation of the noise (sigma). A conservative ΔF/F threshold was set for signal detection equal to 4 times sigma (thrARScl = 4). Also, events with lower ΔF/F than 20% of peak F were discarded (minShow1 = 0.2). AQuA excluded any Ca²⁺ signals composed of less than 6 pixels (minSize = 6 pixel) and less than 0.2s in duration (seedRemoveNeib = 2). Events smaller than 2μm² (2x PSF) were also subsequently filtered out. No pixels were removed close to the imaging boundary (regMaskGap = 0). A temporal threshold (thrTWScl = 2) was set to discriminate between several events occurring in the same spatial location (defined as delta = thrTWScl x sigma). Voxels (x, y, t) with higher values compared to their spatial and temporal neighbors are termed seeds. To determine whether neighboring pixels to each seed are similar enough to be included in the signal, a growing z-threshold relating to noise is used (thrExtZ = 2). The assigned pixels to seeds form super voxels which in turn can be combined to form super events. This is classified by AQuA using three parameters: rising time uncertainty (cRise = 2), slowest delay propagation (cDelay = 2) and propagation smoothness (gtwSmo = 1). A z-score threshold was set to distinguish events from noise (zThr = 2). AQuA also compensates for possible bleaching that can occur during long recordings. This is done by cutting the video into sub-stacks (cut = 200 frames) to calculate baseline fluorescence (F0) through a moving average filter (movAvgWin = 25 frames). Unmentioned parameters were set to default. The same parameters were used for all single-astrocyte Ca²⁺ MD signal analysis.

AQuA was used to extract four event characteristics: size, amplitude, duration, and frequency of events. Size refers to the maximum spatial extent (or spread) of the Ca²⁺ signal (in μm²), amplitude to maximum ΔF/F of the signal, duration to the maximum temporal extent of the signal (in seconds), and frequency refers to the number of events detected over time. Each event is described as a 2D binary footprint, whose area (basic.area) represents the event size (μm²). The peak ΔF/F of each event (curve.dffMax2) was used as amplitude (ΔF/F). Duration was calculated as the time between the starting (loc.t0) and stopping frames (loc.t1) of each event.

Statistics. Statistics were performed using Python (3.7). The normality of the distributions was assessed using the Shapiro–Wilk test for data < 50 datapoints or Kolmogorov-Smirnov test for data with > 50 datapoints. If the p value was less than 0.05 (chosen alpha level) we rejected the null hypothesis and considered the data tested not to be normally distributed. A straight line of the quantile-quantile (QQ) plot was also used as evidence of a normal distribution. Normal distributed data was analyzed using unpaired t-test (two independent groups), paired t-test (two paired groups), one-way ANOVA (multiple independent group comparison) or one-way repeated measures ANOVA (multiple groups, within group comparison). Non-parametric data was analyzed using Kruskal-Wallis test (multiple independent groups). To counteract the problem of multiple comparisons we used Tukey’s honestly significant difference (HSD) for normally distributed data. For non-normal distributed data, we used the more conservative Bonferroni correction. Data was plotted as mean ± 95% confidence interval (CI). Only two-tailed tests were used, and the significance threshold was set to p < 0.05.

Data fitting was also performed in Python. Skewed normal probability density functions (scripy.stat.skewnorm) were fitted to the data and characterized based on (51). To plot and compare complementary cumulative distribution functions, we used the ‘powerlaw’ Python package (57). We visualized and compared power law, exponential, and lognormal fits to the empirical event-size data. For this fitting we set a minimum Ca²⁺ event size threshold of 5 μm².

Astrocyte Ca²⁺ MD signal analysis. For the analysis of Ca²⁺ MD signals, we used 6 astrocytes labelled with membrane tagged GCaMP6f. Two of these astrocytes were imaged a second time 1 day later, and another one 3 days later. For the analysis, all 9 recordings were used unless otherwise stated. Mice transitioned between rest and run states during all recordings. During 3 out of 9 recordings, mice didn’t stop to explore the vibrissa stimulus. Therefore, no data was gathered during vibrissa exploration state for these astrocytes. Four Ca²⁺ event characteristics were analyzed: amplitude (ΔF/F), duration (s), size (μm²), and frequency (number of events/t, t = time in seconds or min) during the four behavioral states: rest, run, vibrissa stimulation (VS), and vibrissa exploration (VE). We compared the means of event characteristics (between group analysis) for all events (n = 50787) from all astrocyte recordings (n = 9) during rest (n = 29637), run (n = 15123), vibrissa stimulation (n = 2906) and vibrissa exploration (n = 3121). Event characteristics were grouped and averaged per astrocyte recording to control for variability between astrocyte recordings and recording time. We compared the mean event characteristics of all recordings (within group design, paired t-test). All events (n = 50787) from all astrocyte recordings (n = 9) were used to investigate the probability distributions of event characteristics (amplitude, duration, size) during different states. The continuous recoding times (n =9) were: 22, 35, 40, 40, 75, 75, 79, 79, and 106 min long.

To compare event characteristics during state transitions, all recordings were aligned to a transition point. The transition points were rest-to-run (226 transitions: 9-s intervals, 4263 Ca²⁺ events), run-to-rest (177 transitions: 9-s intervals, 3089 Ca²⁺ events), and vibrissa stimulation during run (279 transitions: 6-s intervals, 5333 Ca²⁺ events). Transition events were only selected if the behavioral state was stable during the pre- and post-transition period. For example, during a rest-to-run transition the animal was at rest for at least 3 s followed by at least 6 s of continuous running. The onset time (loc.t0) of each event was used to determine the Ca²⁺ MD signal characteristics in relation to the time of state transition.

Dual axon-astrocyte Ca²⁺ signal analysis. By co-injecting AAV1-CMV-Cre, AAV1-CAG-Flex-GCaMP6f and AAV1-hSyn-TurboRFP (see adeno-associated viruses’ section) in the VPM we labelled BX astrocytes with GCaMP6f and thalamocortical axons (VPM to BX) with GCaMP6f and TurboRFP (n = 3 mice, 3 astrocytes). Because of sparse astrocyte labelling, some axon boutons (identified as functional puncta expressing both GCaMP6f and TurboRFP) were found inside the astrocyte territory and some outside (see Fig. S4B). We used the Ca²⁺ activity of axon boutons outside the astrocyte territory as controls to confirm thalamocortical neuron activity indeed increases with vibrissa stimulation and locomotion as expected.

The Ca²⁺ activity of axon bouton signals outside the gliapil during behavioral state transitions was extracted with AQUA. AQUA parameters were optimized for extracting axon activity. The minimum event size cutoff threshold was set to >0.74 μm (minSize = 4), the spatial smoothing level was set to 1 SD (SmoXY = 1), the active threshold scale was set to 6 SD of noise (thrARScl = 6). Also, 50 pixels close to the image border (regMaskGap = 50) were removed. Signals represent distinct points of [Ca²⁺]_i elevation, not distinct axons. We investigated changes in mean frequency (number of events/s) of axon Ca²⁺ signals during behavioral state transitions like we did previously with astrocytes. The behavioral state transitions investigated and their respective axon Ca²⁺ signals were: rest to run (174 intervals, 8613 events), run to rest (52 intervals, 2274 events) and whisker stimulation during run (756 intervals, 28232 events). Using this preparation (GCaMP6f not membrane tagged, no AAV injection in BX) we also re-tested, if astrocytes responded with [Ca²⁺]_i elevation to whisker stimulation (756 intervals, 6091 events). We selected astrocyte processes (inside gliapil) constrained within manually drawn ROIs (n = 35 from 3 astrocytes) devoid of TurboRFP labelled axon boutons to minimize contamination from axon activity. We tested again, how the mean frequency (number of events/s) of Ca²⁺ signals initiated within these ROIs changed during whisker stimulation (756 intervals, 6091 events).

Astrocyte activity heatmaps. Astrocyte activity maps refer to overall activity patterns observed in astrocytes during long recording times (typically tens of min) represented in the form of a heatmap. A heatmap represents the proportion of events in space detected by summing the 2D event footprints detected by AQuA in time.

Heatmap generation. Motion corrected, two-photon Ca²⁺ imaging data of single astrocytes were processed by AQuA. The Ca²⁺ signals were summarized as 2D footprints and binarized to pixel value = 1. All binarized events were summed in time while maintaining their spatial coordinates to develop activity heatmaps for each astrocyte recording. Some heatmaps were limited to Ca²⁺ events occurring during specific behavioral states (rest, run, VS, VE, all states). In such cases, only frames associated with the respective behavioral state were considered. Heatmaps were normalized to 1 min and to their respective maximum pixel value unless otherwise noted.

Simulation of random distributions. Simulations of activity heatmaps were created by randomly distributing binarized Ca²⁺ signals onto a mask corresponding to the real, recorded astrocyte. Ca²⁺ signals were approximated as ellipses of random orientation, center, and eccentricity. These ellipses were incrementally added within the bounds of the astrocyte previously imaged (binarized maximum projection mask). The area of each ellipse was randomly sampled from the list of event areas corresponding to the respective, real astrocyte recording. The total area of all ellipses in each random simulation was equal to the total area of real Ca²⁺ events of their respective astrocyte recording. For example, when the total area of ellipses added to the simulation was equal to the total area of real event 2D footprints then no more ellipses were added.

Heatmap comparison. Heatmap hotspot patterns were compared using the total continuous recording, video segments corresponding to the state of the animal, or subsequences of the respective video recording (sequences) for within day or between day comparisons.

Within day. To compare activity heatmaps within same day recordings and between their respective simulations (random distributions), 2D image cross correlation (512 x 512 pixel; Pearson cross correlation) was used. One of the images was translated to a position (translation vector) of maximum correlation. The maximum Pearson correlation coefficient (PCC) was used as the correlation metric between the images.

Between days. To correlate activity heatmaps between days we created an activity mask to establish the boundaries of cell activity throughout the whole recording. The mask array had binary x, y values where 1 represents an event taking place. 2D cross correlation was used to find the maximum correlation between a stationary sample mask and one translated by a move vector. This allowed us to align the masks of the sample pairs and use 2D correlation between the heatmap pairs.

Sequences. Sections of continuous video recordings are referred to as sequences. 2D correlation between heatmaps generated from 70-min-long recordings and their shorter subsequences (t = 2, 5, 10, 15, 25, 30, 35, 70 min) were used to determine the approximate recording time needed to capture a reliable Ca²⁺ activity map. 70-min movies were created by omitting frames after 70 min in 5 recordings longer than 70 min.

Similarly, 2D correlation between the activity heatmaps of rest state sequences of the same astrocyte (within day) were used to determine the stability of heatmaps over individual recordings. The sequences were created from 5 astrocyte movies, 70 min each. Each video was split into 3 subsequences of equal duration corresponding to the rest state of the animal. Their respective heatmaps were 2D correlated to each other (3 subsequences per astrocyte correlated, 5 astrocytes, 15 correlations). Alternatively, subsequences were correlated with the total running activity heatmap of their respective astrocyte (state comparison, 3 correlations per astrocyte, 5 astrocytes). Between day correlation of subsequence heatmaps (n = 12) was obtained from the same astrocytes (n = 2) but during different day recordings (day 0 and day 1, 18 correlations between days).

For corresponding references see the manuscript.

The analysis code and guide on how to use the data can be found here: https://github.com/Achilleas/aqua-py-analysis

Readme files are included in the Github repository above. Also Readme files for each figure can be found in supplemental information (Data_curation_SA) uploaded on Zenodo.