{"paper_id":"227ee058-1508-4696-8ea3-202d6a47c8db","body_text":"Page 1 of 37 \n  Research Article \n \n \n \n 5 \n \n \n \nLayer-5 Pyramidal Cell tLTD Requires Astrocytic Ca2+ \nand CB1 Receptor Signaling 10 \nAiri Watanabe (渡辺愛利)1,2, Connie Guo1,2, Shawniya Alageswaran1,2,  \nP. Jesper Sjöström1,3 \n1) Centre for Research in Neuroscience, Department of Neurology and Neurosurgery, \nDepartment of Medicine, Brain Repair and Integrative Neuroscience Program, The \nResearch Institute of the McGill University Health Centre, Montreal General Hospital, 15 \nMontreal Quebec, H3G 1A4, Canada \n2) Integrated Program in Neuroscience, McGill University, 3801 University Street, \nMontreal, Quebec H3A 2B4, Canada \n3) Correspondence: jesper.sjostrom@mcgill.ca  \n 20 \n \n \n \n \n 25 \n \n \n \nAddress for editorial correspondence: \nProf P J Sjöström 30 \nCentre for Research in Neuroscience \nDepartment of Neurology and Neurosurgery \nThe Research Institute of the McGill University Health Centre \nMontreal General Hospital \n1650 Cedar Ave, room L7-225 35 \nMontreal Quebec H3G 1A4, Canada \nMob: +1-438-826-1971 \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 2 of 37 \nAbstract \nTiming-dependent long-term depression (tLTD) is a form of spike timing-40 \ndependent plasticity (STDP) that has been attributed to presynaptic NMDA and CB1 \nreceptors at synapses between layer 5 (L5) pyramidal cells (PCs) in visual cortex. Here, \nwe asked whether astrocytes, known for gliotransmission and tripartite synapse \nformation, are also required. Using quadruple whole-cell recordings in acute slices from \nC57BL/6 mice, we found that L5 PC → PC tLTD was abolished by the glial metabolic 45 \ninhibitor sodium fluoroacetate. Disrupting astrocyte Ca2+ signaling through loss-of-\nfunction approaches such as AAV-mediated expression of CalEx and BAPTA loading of \nastrocyte networks consistently prevented tLTD. Optogenetic activation of astrocyte Gq \nsignaling during tLTD induction abolished tLTD and led to potentiation. Since tLTD \nrelies on endocannabinoid (eCB) signaling and astrocytes express CB1 receptors, we 50 \nconditionally deleted CB1 receptors from astrocytes and found that this manipulation \nabolished tLTD. Taken together, our results show that L5 PC → PC tLTD requires \nastrocyte Ca2+ signaling and CB1 receptor activation. These findings suggest that \nastrocyte-dependent control of STDP may represent a general principle across circuit and \nsynapse types. 55 \nSignificance Statement \nSpike timing-dependent plasticity (STDP) enables neural circuits to adapt based on \nthe precise timing of activity, with timing-dependent long-term depression (tLTD) \npromoting competition and stabilization. While tLTD is often considered a two-factor \nprocess involving pre- and postsynaptic neurons, we show that astrocytes form a key 60 \nthird element. Using targeted loss-of-function approaches, we demonstrate that tLTD at \nlayer-5 pyramidal cell synapses in visual cortex requires astrocyte Ca2+ signaling and \nCB1 receptors. Unexpectedly, optogenetic stimulation of astrocyte Gq signaling during \ninduction blocked depression and instead triggered potentiation. These findings call for a \nrevision of existing models of STDP and suggest that astrocyte-dependent plasticity may 65 \nreflect a general regulatory principle across cortical circuits. \nKeywords \nAstrocytes, spike timing-dependent plasticity, tLTD, calcium signaling, \nendocannabinoid signaling, optogenetics, cortical microcircuits  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 3 of 37 \nIntroduction 70 \nSynaptic plasticity is considered a cornerstone of memory formation (Bliss and \nCollingridge, 1993; Malenka and Bear, 2004; Nabavi et al., 2014) and a key mechanism \nin the developmental refinement of neural circuits (Katz and Shatz, 1996; Cline, 1998; \nSong and Abbott, 2001). This foundational idea is often traced back to Donald Hebb \n(1949) and is captured by the slogan “cells that fire together wire together” (Shatz, 1992), 75 \nwhich highlights the role of correlated activity of connected neurons in strengthening \nsynapses. \nIn more recent years, researchers have emphasized the role of temporal ordering of \nneuronal activity in shaping synaptic changes, a process known as spike timing-\ndependent plasticity, or STDP (Feldman, 2012; Markram et al., 2012). Early studies 80 \nrevealed that when a presynaptic neuron fires within a ten-millisecond before a \npostsynaptic partner, the connection is strengthened through timing-dependent long-term \npotentiation (tLTP), but when the order is reversed, the connection weakens through \ntiming-dependent long-term depression (tLTD) (Markram et al., 1997; Bi and Poo, 1998; \nZhang et al., 1998; Feldman, 2000; Sjöström et al., 2001). This bidirectional form of 85 \nplasticity thus extends Hebb’s original idea by incorporating tLTD, which enables key \ncomputational features such as synaptic competition (Song et al., 2000) and network \nstabilization (Song and Abbott, 2001). \nBoth classical Hebbian plasticity and STDP are examples of two-factor learning \nrules, meaning they depend on coordinated activity in presynaptic and postsynaptic 90 \nneurons (McFarlan et al., 2023). However, research over the past couple of decades have \nhighlighted that other cells — including astrocytes in a tripartite synaptic structure (Perea \net al., 2009) — can also contribute to long-term plasticity at specific synapse types \n(Henneberger et al., 2010; Valtcheva and Venance, 2016; Adamsky et al., 2018). \nWe previously demonstrated that tLTD at synapses between visual cortex L5 PCs 95 \nrequires simultaneous activation of endocannabinoid CB1 receptors and presynaptic \nNMDA receptors (Sjöström et al., 2003). Based on this, we proposed a model in which \nboth receptor types are located at release sites in L5 PC axons (Sjöström et al., 2003; \nDuguid and Sjöström, 2006). However, more recent work at neocortical L4 → L2/3 \nsynapses (Min and Nevian, 2012) as well as at hippocampal CA3 → CA1 synapses 100 \n(Andrade-Talavera et al., 2016; Falcón-Moya et al., 2020) demonstrated that tLTD relies \non CB1 receptor signaling in astrocytes. This raised the possibility that at L5 PC → L5 \nPC connections, the CB1 receptors relevant for tLTD are not those located in L5 PC \naxons (Katona et al., 2006) but those in nearby astrocytes (Busquets-Garcia et al., 2018). \nWe therefore revisited our original model of L5 PC → L5 PC tLTD (Sjöström et 105 \nal., 2003; Duguid and Sjöström, 2006), testing whether astrocyte signaling is also \nrequired. Using a range of targeted interventions, we found that astrocytes are in fact \nnecessary for the induction of L5 PC → PC tLTD. Specifically, we found that astrocyte \nsignaling via Ca2+, Gq protein, and cannabinoid CB1 receptors critically determines L5 \nPC → L5 PC tLTD. Our findings suggests that astrocyte-mediated control of STDP may 110 \nbe a general principle that holds across circuit and synapse types. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 4 of 37 \nMethods \nAnimals and Ethics Statement \nThe animal study was reviewed and approved by the Montreal General Hospital \nFacility Animal Care Committee and adhered to the guidelines of the Canadian Council 115 \non Animal Care. Mice were kept on a 12h light and 12h dark cycle, provided with \nenrichment such as bio-huts and chew blocks, and given food and water ad libitum.  \nC57BL/6J mice (referred to as WT here) were obtained from The Jackson \nLaboratory (JAX strain 000664). Hemizygous Aldh1ll-Cre mice (Tien et al., 2012) were \nobtained from The Jackson Laboratory (JAX 023748) and maintained by crossing to WT 120 \nanimals. Homozygous Emx1-Cre mice (Gorski et al., 2002) were also obtained from The \nJackson Laboratory (005628). Cryorecovered heterozygous CB1f mice (JAX 036107) \nwere used to conditionally delete the cannabinoid receptor 1 Cnr1 gene (Marsicano et al., \n2003). Cnr1fl/+ mice that were cryorecovered were crossed to obtain Cnr1fl/fl mice. \nTo determine mouse genotypes, tail biopsy and paw tattooing were performed prior 125 \nto postnatal day 6 (P6). Genomic DNA was extracted using standard methods, and \ngenotyping was carried out using Jackson Laboratory-recommended primers (Aldh1l1-\nCre: 14314, 18707, 18708; CB1f: 57165, 57166). PCR reactions were performed using \nHotStarTaq DNA Polymerase kit (QIAGEN, 203203) and dNTPs from \nInvitrogen/Thermo Fisher (18427–013). 130 \nAcute Slice Preparation \nMale and female mice aged P11 – P28 were anesthetized with isoflurane and \nswiftly decapitated once the hind-limb withdrawal reflex was lost. The brain was rapidly \nextracted and submerged in ice-cold (<4°C) artificial cerebrospinal fluid (ACSF), \ncontaining in mM: 125 NaCl, 2.5 KCl, 1.25 NaH2PO4, 26 NaHCO3, 25 glucose, 1 MgCl2 135 \nand 2 CaCl2. ACSF solution was constantly bubbled with 95% O2/5% CO2 and \nosmolality was adjusted to 338 mOsm, measured using Model 3300, Model 3320 or \nOsmo1 osmometers (Advanced Instruments Inc., Norwood, MA, USA). Oblique coronal \n300 µm-thick slices were cut using a Campden Instruments 5000 mz-2 vibratome \n(Lafayette Instrument, Lafayette, IN, USA). Slices were incubated at ~33°C in ACSF for 140 \n15–30 min, then allowed to recover at room temperature (~23°C) for at least one hour \nafter slicing before recording. \nFor mice aged ~P16 to ~P21, slices were sectioned in low-Ca2+ ACSF (2 mM \nMgCl2, 1 mM CaCl2). For mice ~P21 and older, slicing was performed in sucrose-based \nsolution containing (in mM): 200 sucrose, 2.5 KCl, 1.0 NaH2PO4, 2.5 CaCl2, 1.3 MgCl2, 145 \n47 glucose, and 26.2 NaHCO3. These slices were also recovered in low-Ca2+ ACSF prior \nto recording in standard ACSF as described above. \nElectrophysiology \nWe carried out experiments with ACSF heated to 33°C (resistive inline heater, \nScientifica Ltd, Uckfield, UK; or TC-324C + SH-27B, Warner Instruments, Holliston, 150 \nMA, USA). Temperature was continuously recorded and verified offline, and recordings \nwere truncated or excluded if outside the 32°C to 34°C range.  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 5 of 37 \nPatch pipettes (typically 4 – 7 MΩ) were pulled from fire-polished glass capillaries \n(BF150-86-10, Sutter Instruments, Novato, CA, USA) using a P-1000 puller (Sutter \nInstruments, Novato, CA, USA) and filled with internal solution (in mM: KCl, 5; K-155 \nGluconate, 115; HEPES, 10; Mg-ATP, 4; Na-GTP, 0.3; Na-Phosphocreatine, 10; \nBiocytin, 0.1% w/v; adjust with KOH to pH 7.2; adjusted with sucrose to 310 mOsm). \nDepending on the experiment, the internal solution also included one of the \nfollowing: Alexa Fluor 594 (15–40 µM; A10428, Life Technologies), Alexa Fluor 488 \n(20–100 µM; A10436, Life Technologies), BAPTA (20 mM; B1204, Fisher Scientific), 160 \nor Fluo-5F (200 µM; F14221, Fisher Scientific). \nWhole-cell recordings were obtained using either BVC-700A (Dagan Corporation, \nMinneapolis, MN, USA) or Model 2400 (A-M Systems, Carlsborg, WA, USA) \namplifiers, with signals filtered at 5 kHz and sampled at 40 kHz using PCIe-6323 boards \n(NI, Austin, TX, USA) controlled by MultiPatch custom software (https://github.com/pj-165 \nsjostrom/MultiPatch.git) running in Igor Pro 8 or 9 (WaveMetrics Inc., Lake Oswego, \nOR, USA). \nL5 PCs were targeted for patching based on their large and characteristically \npyramid-shaped somata. Cell identity was further assessed via a series of somatic current \ninjections (500 ms; –0.3 to +0.7 nA in 0.2 nA steps), which evoked spiking patterns 170 \ntypical for L5 PCs. Morphological confirmation was obtained post hoc using two-photon \n(2p) imaging (see below). Astrocytes were identified by preincubating slices in ACSF \ncontaining sulforhodamine 101 (1 µM; S7635, Millipore Sigma, ON, Canada) for \n5 minutes at room temperature. SR101-labeled astrocytes were then visualized by 2p \nmicroscopy and targeted for whole-cell recording and/or imaging. Astrocytes were loaded 175 \nwith Fluo-5F for at least 15 min before Ca2+ imaging. \nTo find monosynaptically connected pairs among L5 PCs, for which connectivity is \nonly a sparse 10 – 15% (Song et al., 2005; Chou et al., 2024), we employed quadruple \nwhole-cell recordings to test up to twelve candidate connections simultaneously. Four \nneurons were identified and approached with four pipettes. Gigaohm seals were formed 180 \nsequentially, followed by successive break-in. To screen for synaptic connections, five \nspikes were evoked at 30 Hz using 1.3 nA current injections in each cell, repeated up to \n20 times at 20-second inter-sweep intervals. In a subset of experiments, synaptic \nresponses were evoked using either extracellular stimulation or a targeted optogenetic \nmethod described below (optomapping). For extracellular stimulation, the pipette was 185 \nfilled with filtered ACSF and placed approximately 100 µm from the patched cell to \nrecruit local inputs. \nPlasticity Protocols \nPrior to drug application or plasticity induction, a baseline period of 7–20 minutes \nwas recorded. During this period, bursts of 5 spikes were evoked using trains of 5-ms-190 \nlong ~1.3-nA current pulses delivered at 30 Hz every 20 seconds. We induced tLTD \nusing a post-before-pre spike pairing protocol consisting of five postsynaptic spikes at \n20 Hz, timed either 10 ms or 25 ms before presynaptic stimulation, repeated 15 times with \n10-second intervals. This protocol is known to reliably elicit tLTD at L5 PC → PC \nsynapses (Sjöström et al., 2003). Following the induction, the baseline 30-Hz spike 195 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 6 of 37 \npattern was resumed for up to 75 minutes. All recordings were performed in current \nclamp. A subset of plasticity experiments (Fig. 3) were carried out with 2p optogenetic \nactivation (Chou et al., 2024), as previously described (Chou et al., 2025), see below for \nadditional details. \nRecordings with unstable baselines were excluded based on a t-test of Pearson’s r 200 \nfor response amplitudes (Lalanne et al., 2016). Input and series resistance were assessed \nusing a 250-ms-long hyperpolarizing test pulse of -25 pA. Recordings were excluded or \ntruncated if input resistance changed by more than 30%, or if resting membrane potential \nshifted by more than 8 mV. Recordings shorter than 20 minutes post induction or drug \nwash-in were excluded from analysis. 205 \nThe locus of plasticity, i.e., whether plasticity was expressed pre- or \npostsynaptically, was assessed using paired-pulse ratio (PPR) and coefficient of variation \n(CV) analyses (Brock et al., 2020). PPR was calculated as the ratio of the second EPSP to \nthe first, or EPSP2 / EPSP1. The change in PPR, or ΔPPR, was computed by subtracting \nPPR measured after drug wash-in or plasticity induction from the baseline PPR, or 210 \nPPRafter - PPRbefore. If the presynaptic cell spiked twice for individual 5-ms-long current \npulses, the experiment was excluded from PPR analysis. For PPR and CV analyses, \nexperiments with EPSPs weaker than 0.3 mV were excluded, to ensure sufficient signal \nto noise. \nNeonatal Viral Injection 215 \nAll adeno-associated viruses (AAVs) were aliquoted on arrival and stored at –80 °C \nuntil use. The following constructs were used for neonatal intracerebral injections: \nAAV2/5-GfaABC1D-mCherry-hPMCA2w/b (Addgene 111568-AAV5), AAV9-CAG-\nDIO-ChroME-ST-P2A-H2B-mRuby3 (Addgene 108912-AAV9), AAV1-GfaABC1D-\nOptoGq-eYFP (Adamsky et al., 2018), AAV5-GfaABC1D-Cre-4x6T (Addgene 196410-220 \nAAV5), and AAV5-pCAG-FLEX-EGFP-WPRE (Addgene 51502-AAV5). \nThe following viral titers were used (in genome copies per milliliter, GC/ml): \n1.20×1013 for hPMCA2w/b, 2.7×1012 for ChroME, 1.60×1012 to 1.60×1013 for OptoGq, \n1.60×1013 for GfaABC1D-Cre, and 1.50×1013 for FLEX-EGFP. \nIntracerebral injections were performed in P0–P2 mice, targeting visual cortex 225 \n(Kim et al., 2014; Chou et al., 2024). Mice were cryoanesthetized by placing them on \naluminum foil over ice for approximately 10 minutes, then secured in a neonatal \nstereotaxic apparatus. Injections were delivered using a 33-gauge needle (point style 4, \n30° bevel, 25.4 mm length) mounted on a 10 µl gas-tight syringe (Hamilton Instruments, \nReno, NV, USA), controlled by a syringe pump (Model 70-4507, Harvard Apparatus, 230 \nHolliston, MA, USA). \nThe injection site was located at 0.00 mm anterior-posterior and 1.1 mm medial-\nlateral relative to lambda. Injections were delivered at three depths along a single needle \ntrack (0.2, 0.15, and 0.1 mm below the pial surface) during stepwise retraction of the \nneedle, at a constant rate of 0.25 µl/min. At each depth, volumes of 0.35 µl were infused 235 \nfor hPMCA2w/b, GfaABC1D-Cre, and FLEX-EGFP. For ChroME, 0.3–0.4 µl was \ninjected; for OptoGq, 0.2–0.4 µl was used. Following injection, pups were placed under a \nheat lamp and monitored until spontaneous movement resumed. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 7 of 37 \nTwo-Photon Imaging \nWe used custom-built imaging workstations for 2p imaging, as previously 240 \ndescribed (Abrahamsson et al., 2017), using either a Chameleon Ultra II (Coherent, PA, \nUSA) or a Mai Tai HP (Spectra-Physics, CA, USA) tunable Ti:Sa laser. Lasers were \ntuned to wavelengths optimal for visualization of the dye at hand, e.g., 750 nm for \nmCherry and Alexa 488, 820 nm for Alexa-594, 1040 nm for mRuby, 920 nm for eYFP, \netc. 245 \nLaser power was monitored using a power meter and controlled via a half-wave \nplate and polarizing beam-splitting cube (Thorlabs GL10-B and AHWP05M-980). Gating \nwas achieved with a mechanical shutter (Thorlabs SH05/SC10), and scanning was \nperformed using 6215H galvanometric mirrors (3 mm; Cambridge Technology, Bedford, \nMA, USA). 250 \nFluorescence was detected using bialkali photomultipliers (2PIMS-2000-20-20, \nScientifica Ltd., Uckfield, UK) through an Olympus LUMPlanFL N 40×/0.80 objective, \nor via substage-mounted R3896 photomultipliers (Hamamatsu, Bridgewater, NJ, USA) \ncollected through an Olympus Achromatic/Aplanatic condenser (NA 1.40). Emission \nlight was split using a Semrock FF665 dichroic, and laser light was blocked with a 255 \nSemrock FF01-680 filter. Red and green fluorescence channels were separated with a \nChroma t565lpxr dichroic in combination with Chroma ET630/75M and ET525/50M \nemission filters. \nLaser-scanning Dodt contrast was generated using custom optics by collecting \ntransmitted laser light through the acute slice with a spatial filter and diffuser placed in a 260 \n1× telescope, focused onto an amplified photodiode (Thorlabs PDA100A-EC). Signals \nfrom photomultipliers and the photodiode were digitized using PCIe-6374 boards \n(National Instruments, Austin, TX, USA) and acquired with ScanImage (2019–2024) \nrunning in MATLAB (MathWorks, Natick, MA, USA). \nAstrocyte Ca2+ activity was imaged using 2p excitation at 780 – 930 nm to visualize 265 \nchanges in Fluo-5F fluorescence. Each movie was taken at a single focal plane for 150 \nseconds using 512 × 512-pixel frames acquired at 2.11 Hz. Movies were taken before and \nafter drug wash-in or tLTD induction. Ca2+ signals were measured as dG/R, a change in \ngreen Fluo-5F fluorescence normalized to red SR101 fluorescence. For the detection of \nCa2+ events, at least 10 regions of interests (ROIs) of were manually selected and baseline 270 \nwas set at the 10 – 20 frames with the weakest Ca2+ signal. We low-pass filtered at 0.2 Hz \nto eliminate high-frequency noise. Ca2+ events were detected by a simple thresholding \nalgorithm set to detect signal at 0.5 – 1 sigma above background noise (Watanabe et al., \n2023). Mean Z-scores of Pearson’s r for the correlation of all ROIs were used to measure \nthe correlation of Ca2+ activity within cells (Watanabe et al., 2023). Astrocytes that 275 \ndepolarized to >-40 mV during imaging, cells with low signal to noise ratio whereby at \nleast 10 ROIs could not be selected, and movies with significant drift that could not be \ncorrected for (e.g., Z-plane drift) were excluded from analysis. Imaging data were \nanalyzed using LineScan Analysis (https://github.com/pj-sjostrom/LineScanAnalysis) \nrunning in Igor Pro. 280 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 8 of 37 \nOptogenetics \nWe used 2p optogenetic stimulation, or optomapping (Chou et al., 2024), to recruit \npresynaptic cells onto a patched postsynaptic PC in acute slices from Emx1Cre/Cre mice \ninjected with AAV-ChroME. A Mai Tai HP Ti:Sa laser tuned to 1040 nm was used to \nevoke action potentials in ChroME-expressing neurons via spiral scanning with jScan 285 \nsoftware (https://github.com/pj-sjostrom/jScan) running in Igor Pro 9. Each opsin-\nexpressing candidate cell was stimulated with two spiral scans 500 ms apart, while \nmonitoring the postsynaptic response. \nTo exclude directly activated postsynaptic PCs, responses with onset latency <1 ms \nwere discarded (Chou et al., 2024). Upon identifying a synaptic connection, only the 290 \nconnected presynaptic cell was repeatedly stimulated to record a ~10 min baseline, induce \ntLTD, and record a post-induction period, following electrophysiology parameters \ndescribed above. Responses <0.3 mV EPSP amplitude were excluded. The power of the \nTi:Sa laser was set to ~55 mW as measured at the objective back aperture. \nTo activate the light-sensitive Gq-coupled receptor Opto-α1AR in C57BL/6J mice 295 \ninjected with OptoGq virus, we used 445-nm laser light (Industrial 600 mW Blue Laser \nModule, eBay.ca, Item ID: 181041391795). Light was delivered via the 40× objective in \n45 ms pulses at 20 Hz during tLTD induction. The 445-nm beam was combined with the \n2p laser using a Semrock FF665 dichroic and controlled via the MultiPatch software to \nallow synchronized stimulation with electrophysiology recordings. The power of the 445-300 \nnm laser was set to ~13 mW as measured at the objective back aperture. \nPharmacology \nAcute slices were preincubated in ACSF containing 5 mM sodium fluoroacetate \n(NaFAC; MP Biomedicals, CA, USA) for 30 min prior to recordings; NaFAC was not \npresent during experiments. Arachidonyl-2-chloroethylamide (ACEA; Cayman 305 \nChemical, MI, USA) was used at 125 nM in ACSF and was washed in following baseline \nacquisition. \nBiocytin Staining and Confocal Imaging \nTo preserve patched neurons for biocytin histology, patch pipettes were slowly \nretracted after recordings to allow resealing of the membrane. Acute slices were fixed 310 \novernight at 4 °C in 4% paraformaldehyde and then transferred to 0.01 M phosphate-\nbuffered saline (PBS) for storage (up to 2 weeks) prior to staining. \nFor biocytin-only staining, slices were washed 4× for 10 min each in 0.01 M Tris-\nbuffered saline (TBS) with 0.3% Triton X-100, blocked for 1 h in the same buffer \ncontaining 10% normal donkey serum (NDS; 017-000-121, Jackson ImmunoResearch), 315 \nthen incubated overnight at 4 °C with 1:200 Alexa Fluor 647–conjugated streptavidin \n(S32357, ThermoFisher Scientific) in TBS with 0.3% Triton X-100 and 1% NDS. Slices \nwere then washed 4× for 10 min in TBS. \nFor combined biocytin and immunofluorescence staining, slices were washed 4× \nfor 5 min in 0.01 M PBS with 1% Triton X-100, blocked for 1.5 h in PBS with 0.3% 320 \nTriton X-100 and 10% NDS, and incubated overnight at 4 °C with 1:500 Alexa Fluor \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 9 of 37 \n647–streptavidin and 1:1000 monoclonal rat anti-mCherry antibody (M11217, \nInvitrogen) in PBS containing 1% NDS and 0.3% Triton X-100. The next day, slices \nwere washed 3× in the same buffer and incubated for 2 h with 1:1000 Alexa Fluor 488–\nconjugated donkey anti-rat secondary antibody (712-585-150, Jackson 325 \nImmunoResearch), followed by three additional washes. Slices were mounted on \nSuperFrost Plus slides (1255015, ThermoFisher) using ProLong Glass Antifade Mountant \n(P36984, ThermoFisher), and the edges of the coverslip were sealed with clear nail polish \nafter curing. \nImage stacks were acquired using Zeiss LSM 780 or 880 confocal microscopes at 330 \n10× or 20× magnification with ZEN2010 software (ZEISS International). We used the \nfollowing laser lines: 633 nm for Alexa Fluor 647, 561 nm for mRuby, and 488 nm for \nAlexa Fluor 488 and for EGFP. \nStatistics \nUnless otherwise noted, results are reported as mean ± standard error of the mean 335 \n(SEM). Boxplots show medians and quartiles, with whiskers denoting extrema. \nSignificance levels are indicated as * p < 0.05, ** p < 0.01, and *** p < 0.001. \nFor multiple comparisons, pairwise testing was performed only if ANOVA \nindicated significance at the p < 0.05 level. Standard ANOVA was used for \nhomoscedastic data and Welch’s ANOVA for heteroscedastic data, based on Bartlett’s 340 \ntest (p < 0.05). Pairwise comparisons employed two-tailed Student’s t-tests for equal \nmeans. If an F-test for equality of variances indicated p < 0.05, the unequal variances t-\ntest was used instead. Bonferroni’s correction was always used to adjust for multiple \ncomparisons. For data that was not normally distributed (e.g., Supp. Fig. 2B), we relied \non the Kruskal-Wallis test instead of ANOVA, and Mann-Whitney U test instead of 345 \nStudent’s t-test. \nFor CV analysis, φ was defined as the angle relative to the diagonal demarcation \nline and Wilcoxon signed rank test was used to compare φ relative to the diagonal (Brock \net al., 2020). Changes in 1 𝐶𝑉𝑛𝑜𝑟𝑚\n2⁄  were compared using a one-sample Student’s t test \nversus 1. 350 \nStatistical tests were carried out in Igor Pro 9 (WaveMetrics Inc., OR, USA). Linear \nmixed-effects models were implemented in RStudio 2024.12.0+467 and 2025.09.2+418 \n(Posit Software, Boston, MA, USA), with Tukey’s method for post hoc adjustments. For \nthe mixed-effects models, age and genotype were treated as fixed effects, and cell as \nrandom effect. To account for log-normal synaptic strength distribution (Chou et al., 355 \n2024), EPSP amplitudes were log-transformed prior to parametric statistical comparisons. \nResults \nAstrocytes are involved in L5 PC → PC tLTD \nTo test whether astrocytes contribute to L5 PC → PC tLTD, we preincubated acute \nvisual cortex slices with the glial metabolic inhibitor NaFAC (Swanson and Graham, 360 \n1994). Because L5 PC → PC connectivity rate is low (~10-15%, Song et al., 2005; Chou \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 10 of 37 \net al., 2024), we used quadruple whole-cell recordings (Fig. 1A) to find monosynaptic \nconnections (Lalanne et al., 2016). In control slices, L5 PC → PC connections reliably \nunderwent tLTD. However, in the presence of NaFAC, tLTD was abolished (Fig. 1B–E). \nNaFAC-treated recordings showed stable responses (after/before: 103 ± 4%, n = 6), 365 \ncomparable to control stability recordings without NaFAC (100 ± 5%, n = 8, p = 0.61; \npooled in Fig. 1D, E), suggesting that acute synaptic transmission was not directly \nimpaired. However, overall experimental success was lower in NaFAC-treated slices \n(42% inclusion rate of recordings) than in controls (65%), suggesting possible adverse \neffects on slice viability. 370 \nIn untreated slices, we next examined the locus of tLTD expression using PPR and \nCV analyses (Brock et al., 2020). A positive ΔPPR indicates decreased release \nprobability, consistent with a presynaptic locus of expression. Similarly, because \nnormalized 1/CV2 scales with release probability, a reduction in 1/CV2 also supports \ndecreased presynaptic release (Brock et al., 2020). Consistent with these criteria, both 375 \nPPR and CV analyses pointed to a presynaptic expression of tLTD (Fig. 1F, G), in \nagreement with previous reports (Sjöström et al., 2003). \nThese findings suggested that astrocytes may be required for L5 PC → PC tLTD, \nalthough we could not rule out potential off-target effects of NaFAC, including neuronal \ntoxicity. 380 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 11 of 37 \n \nFigure 1 | Astrocyte function is required for tLTD at L5 PC → PC synapses. \n(A) Representative quadruple whole-cell recording used to probe monosynaptic L5 \nPC → PC connections in acute slices. Neurons were filled with Alexa 594 for \nvisualization. 385 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 12 of 37 \n(B) Example tLTD experiment in a control slice. Gray bar indicates the tLTD \ninduction period. Inset: representative averaged EPSPs before and after induction. Scale \nbars: 25 ms, 0.1 mV. \n(C) Example experiment in a NaFAC-treated slice showing abolished tLTD. \nInduction protocol and format as in (B). 390 \n(D) Time course of normalized EPSP amplitudes from pooled control recordings \n(black), tLTD in NaFAC-treated slices (red), and tLTD in untreated slices (blue). Black \nbar: induction period. \n(E) Magnitude of plasticity across groups demonstrates that NaFAC abolished \ntLTD (after/before Pooled Ctrls 101 ± 3%, tLTD NaFAC 104 ± 7%, tLTD 74 ± 5%; 395 \nANOVA p < 0.001). Boxplots show medians and quartiles; whiskers denote extremes. \n(F) Changes in paired-pulse ratio (ΔPPR) were consistent with presynaptically \nexpressed tLTD (Sjöström et al., 2003). ANOVA p < 0.05. ΔPPR of controls (-0.01 ± \n0.03) and tLTD NaFAC (-0.11 ± 0.07) were indistinguishable (t-test p = 0.23) and were \ntherefore pooled for comparison with tLTD (0.15 ± 0.08, t-test p < 0.05). 400 \n(G) Coefficient of variation (CV) analysis supported a presynaptic locus of \nexpression for tLTD (φ = 21 ± 3°; Wilcoxon signed rank p < 0.05) (Brock et al., 2020). \n \nAbolishing astrocyte Ca2+ signaling disrupts tLTD \nTo further investigate the role of astrocytes in L5 PC → PC tLTD, as suggested by 405 \nthe NaFAC experiments, we examined whether impairing astrocyte Ca2+ signaling would \nalter plasticity expression. We selectively expressed the human plasma membrane Ca2+ \nATPase isoform 2w/b (hPMCA2w/b, also known as CalEx) in astrocytes by injecting \nAAV-pZac2.1-GfaABC1D-mCherry-hPMCA2w/b (Yu et al., 2018) into neonatal visual \ncortex. CalEx constitutively exports cytosolic Ca2+, thereby disrupting intracellular Ca2+ 410 \ndynamics (Yu et al., 2018; Institoris et al., 2022; Guayasamin et al., 2025). Experiments \nwere performed at P15 – P22, when astrocyte CalEx expression was robust (Fig. 2A, B).  \nUsing quadruple whole-cell recordings, we identified monosynaptically connected \nL5 PC → PC pairs in CalEx-expressing slices (Fig. 2C). Under these conditions, the post-\nbefore-pre induction protocol failed to elicit tLTD, and EPSP amplitudes remained stable, 415 \nindistinguishable from non-induction control recordings (Fig. 2D, E). These results show \nthat astrocyte Ca2+ signaling is required for L5 PC → PC tLTD. \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 13 of 37 \n \nFigure 2 | tLTD relies on astrocyte Ca2+ signaling. 420 \n(A) Neonatal AAV injections to express CalEx (AAV2/5-GfaABC1D-mCherry-\nhPMCA2w/b) were performed at P0–P2 in C57BL/6 mice. Electrophysiology was \nconducted at P15 – P22. \n(B) Sample expression of CalEx in visual cortex, visualized via mCherry \nfluorescence following neonatal AAV injection. 425 \n(C) Representative quadruple whole-cell recording used to probe for monosynaptic \nPC → PC connections in CalEx-expressing slices. Neurons were filled with Alexa 594. \nPresynaptic PC1 was connected to postsynaptic PC2.  \n(D) Example electrophysiological trace from the same PC1 → PC2 connection as in \n(C). Gray shading denotes the tLTD induction period. Inset scale bars: 25 ms, 0.2 mV. 430 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 14 of 37 \n(E) Ensemble time course of normalized EPSP amplitudes before and after \ninduction, comparing tLTD induction (red) and control recordings (black) in CalEx-\nexpressing slices, showing no tLTD. \n(F) Population averages showed that plasticity was indistinguishable across the two \ngroups (after/before Control 104 ± 6% vs tLTD CalEx 109 ± 4%; t-test p = 0.54). 435 \n \nTo complement the global Ca2+ extrusion approach, we also assessed tLTD \nexpression while loading individual astrocytes with the Ca2+ chelator BAPTA, taking \nadvantage of the fact that 20 mM BAPTA can diffuse across gap-junctions throughout the \nastrocyte syncytium (Watanabe et al., 2023). To facilitate the search for synaptic 440 \nconnections while one pipette was used to patch an astrocyte, we performed experiments \nusing 2p optomapping (Chou et al., 2024) instead of quadruple whole-cell recording. We \npreviously demonstrated that 2p optogenetics can be reliably used to study long-term \nplasticity (Chou et al., 2025). \nNeonatal Emx1Cre/Cre mice were injected with AAV9-CAG-DIO-ChroME-ST-P2A-445 \nH2B-mRuby3 in visual cortex to express ChroME opsin in excitatory cells (Fig. 3A). \nSlices were preincubated with sulforhodamine 101 to label astrocytes for targeted \npatching under 2p imaging. In L5, PCs were patched, and EPSPs were evoked by \nactivating ChroME-expressing candidate presynaptic cells using spiral scans at 1040 nm \n(Fig. 3B). A neighboring astrocyte was patched 40 – 100 µm from the PC and loaded 450 \nwith BAPTA via the patch pipette. When tLTD was induced with astrocytic BAPTA \nloading, tLTD failed (Fig. 3). Together with the CalEx data, these results support the \nconclusion that astrocyte Ca2+ signaling is essential for tLTD. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 15 of 37 \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 16 of 37 \nFigure 3 | Astrocyte Ca2+ buffering abolishes tLTD. 455 \n(A) ChroME was expressed in cortical excitatory neurons by neonatal viral \ninjection (P0–P2) in Emx1Cre/Cre mice. Experiments were conducted at P16–P20. \n(B) Left: experimental setup showing a patched L5 pyramidal cell (Alexa 488, \ngreen), ChroME-expressing neurons (mRuby3, red), and a nearby astrocyte patched and \nloaded with BAPTA (Alexa 488, green). Numbers indicate presynaptic neurons scanned 460 \nvia 2p spiral stimulation. Solid circles met inclusion criteria; dashed circles were \nexcluded during analysis. Right: sample postsynaptic response traces from nine \npresynaptic cells. \n(C) Sample recording without astrocytic BAPTA illustrating typical amount of \ntLTD. Grey shading indicates induction period. Inset scale bars: 25 ms, 1 mV. 465 \n(D) Left: sample tLTD recording with astrocytic BAPTA loading showing reduced \nplasticity. Induction protocol and format as in (C). Right: associated confocal image \nshowing the biocytin-filled PC and astrocyte, with mRuby-labeled ChroME-expressing \ncells in red. \n(E) Ensemble averages of tLTD with or without astrocytic BAPTA. Recordings 470 \nwith EPSP amplitudes <0.3 mV were excluded due to poor signal-to-noise ratio. \n(F) Astrocytic BAPTA prevented tLTD (after/before tLTD 78 ± 6% vs tLTD with \nBAPTA 96 ± 2%; t-test p < 0.05). \n \nManipulating astrocyte G-protein abolishes tLTD 475 \nGq proteins coupled to astrocytic CB1 receptors are implicated in Ca2+ elevation in \nresponse to eCBs (Navarrete and Araque, 2008). We therefore asked if perturbing \nastrocytic Gq signaling would alter tLTD. We therefore expressed the opto-α1AR opsin \n(Adamsky et al., 2018) — a light-sensitive Gq-coupled receptor — in visual cortex \nastrocytes to enable temporally precise optogenetic manipulation (Fig. 4A).  480 \nIn acute slices from OptoGq-expressing mice, we patched L5 PCs and evoked \nsynaptic responses using extracellular stimulation (Fig. 4B). This approach was used \ninstead of quadruple patch or optomapping because one acquisition board was dedicated \nto light stimulation and spectral overlap with neuron-targeted opsins had to be avoided. \nDuring tLTD induction, 445-nm laser pulses were used to activate OptoGq in astrocytes 485 \n(Fig. 4C), which altered the direction of plasticity (Fig. 4D, E). PPR and CV analyses \nsuggested a presynaptic locus of tLTD expression (Fig. 4F, G). These results indicate that \ndisrupting astrocytic Gq signaling abolishes tLTD. \nTo characterize the plasticity shift, we applied a linear mixed-effects model \nincluding initial EPSP amplitude, animal age, and sex as covariates. EPSP changes in the 490 \ntLTD + light and light-only conditions were statistically indistinguishable (p = 0.93, \nSupp. Fig. 1). The pooled light group showed a negative correlation between EPSP \nchange and initial amplitude (Supp. Fig. 1A). CV analysis (Supp. Fig. 1B) indicated a \nmixed pre- and postsynaptic locus of expression. The negative correlation between EPSP \nchange and initial amplitude (Debanne et al., 1999; Sjöström et al., 2001; Hardingham et 495 \nal., 2007), along with the CV analysis suggesting a mixed pre- and postsynaptic locus of \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 17 of 37 \nexpression (Sjöström et al., 2007), is characteristic of LTP. Taken together, these data \nsuggest that activation of astrocytic Gq signaling prevents tLTD and instead promotes \nLTP. \n 500 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 18 of 37 \nFigure 4 | Disrupting astrocyte Gq signaling prevents tLTD. \n(A) Timeline of neonatal viral injections (P0–P2) with AAV1-GfaABC1D-OptoGq-\neYFP to express a light-sensitive Gq-coupled receptor (OptoGq) in visual cortex \nastrocytes. Electrophysiological experiments were performed at P16–P26. \n(B) Example image showing a patched L5 pyramidal neuron filled with Alexa 594 505 \n(magenta) surrounded by eYFP-labeled astrocytes expressing OptoGq (yellow). Dotted \nline indicates position of extracellular stimulation pipette. \n(C) Sample recording from a tLTD + light experiment. The shaded gray bar \ndenotes the period of plasticity induction as well as of 445-nm light delivery. Inset: \nrepresentative EPSPs before and after induction. Scale bars: 25 ms, 2 mV. 510 \n(D) Time course of normalized EPSP amplitudes for control light-only recordings \n(black), tLTD + light (red), and tLTD (blue). Black bar indicates induction period. \n(E) Light activation of astrocytic Gq signaling prevented tLTD and shifted \nplasticity toward LTP (after/before light 143 ± 18%, tLTD + light 130 ± 15%, tLTD 68 ± \n5%; ANOVA p < 0.05). 515 \n(F) ΔPPR values were consistent with presynaptically expressed tLTD (Sjöström et \nal., 2003). ANOVA p < 0.001. ΔPPR values in light (-0.03 ± 0.08) and tLTD + light \nconditions (0.05 ± 0.03) were indistinguishable (t-test p = 0.38) and were therefore \npooled for comparison with tLTD (0.43 ± 0.1; t-test p < 0.05). \n(G) CV analysis in the tLTD group was consistent with a presynaptic locus of 520 \nexpression (φ = 16 ± 3°, Wilcoxon signed rank p = 0.06). In addition, 1 𝐶𝑉𝑛𝑜𝑟𝑚\n2⁄  was \nreduced (39 ± 12%, t-test p < 0.01, n = 5), suggesting reduced release probability (Brock \net al., 2020). \n \nACEA induces LTD at L5 PC → PC synapses via astrocyte Ca2+ signaling 525 \nIn our previously proposed working model of tLTD (Sjöström et al., 2003), \nplasticity induction requires activation of presynaptic CB1 receptors by retrograde eCB \nsignaling in conjunction with presynaptic spiking. Consistently, LTD at L5 PC → PC \nsynapses could be induced by bath application of the synthetic CB1 receptor agonist \nACEA, which mimicked and occluded tLTD (Sjöström et al., 2003). 530 \nWe replicated this ACEA-induced LTD in control slices (Fig. 5A, B). In contrast, \nwhen ACEA was applied in slices expressing CalEx selectively in astrocytes, LTD was \nabolished (Fig. 5C-E). The loss of ACEA-induced LTD was observed only in mice older \nthan P20, likely reflecting the time required for sufficient CalEx expression. \nPPR analysis revealed no detectable differences across conditions (Fig. 5F), 535 \nsuggesting that paired-pulse dynamics were not altered by ACEA application or CalEx \nexpression. In contrast, CV analysis suggested that ACEA reduced release probability \n(Fig. 5G). Higher-frequency stimulation may more effectively deplete the readily \nreleasable pool to reveal larger PPR changes due to ACEA. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 19 of 37 \nTogether, these results suggest that astrocyte Ca2+ signaling is necessary for 540 \ncannabinoid-mediated LTD, paralleling its requirement for tLTD. These findings further \nsupport the involvement of astrocytes in L5 PC → PC tLTD. \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 20 of 37 \nFigure 5 | Astrocyte Ca2+ signaling is needed for eCB-mediated L5 PC → PC LTD. 545 \n(A) Representative quadruple whole-cell patch recording used to probe \nmonosynaptically connected L5 PC → PC connections in slices from control and CalEx-\ninjected mice. Neurons were filled with Alexa 594. \n(B) Sample experiment showing a reduction in EPSP amplitude following ACEA \nwash-in (grey) in a control slice. Inset scale bars: 0.5 mV, 25 ms. 550 \n(C) Sample experiment showing no EPSP reduction after ACEA wash-in (grey) in \na slice with CalEx-expressing astrocytes. Inset scale bars: 1 mV, 25 ms. \n(D) Ensemble time course of normalized EPSP amplitudes, comparing control with \nno ACEA wash-in (black), ACEA wash-in in control slices (blue), and ACEA wash-in in \nCalEx expressing slices (red). Solid black line indicates the period of ACEA application. 555 \n(E) Population data showed reduced EPSPs following ACEA wash-in in control \nslices (after/before Control 98 ± 7%, ACEA 70 ± 5%) but not in slices with astrocyte \nCalEx expression (101 ± 4%, ANOVA p < 0.01). \n(F) PPR did not detectably differ across groups (ΔPPR Control -0.06 ± 0.03, n = 3, \nACEA 0.13 ± 0.2, n = 5, ACEA CalEx 0.02 ± 0.04, n = 5, ANOVA p = 0.6). 560 \n(G) CV analysis of the ACEA group was consistent with presynaptic expression (φ \n= 11 ± 1°, Wilcoxon signed rank p < 0.05). In addition, 1 𝐶𝑉𝑛𝑜𝑟𝑚\n2⁄  was reduced (59 ± 8%, \nt-test p < 0.01, n = 6), indicating reduced release probability (Brock et al., 2020). \n \nAstrocyte-Specific Deletion of CB1 Receptors Abolishes tLTD 565 \nAstrocytes possess CB1 receptors (Navarrete and Araque, 2010), so we tested if \nthey are needed for L5 PC → PC tLTD. To specifically delete CB1 receptors in \nastrocytes, we either crossed CB1f and Aldh1l1-Cre mice, or we injected visual cortex of \nCB1f mice with AAV5-GfaABC1D-Cre-4x6T (Fig. 6A). To tag astrocytes, we co-\ninjected a Cre-dependent EGFP reporter (Fig. 6A, B). 570 \nIn homozygous CB1fl/fl mice for which the Cnr1 gene was completely deleted in \nastrocytes, tLTD could not be induced (Fig. 6D, E, F). In contrast, heterozygous CB1fl/+ \nmice, which retained one functional copy of the Cnr1 gene, exhibited normal tLTD (Fig. \n6C, E, F). The degree of tLTD were indistinguishable when CB1 receptors in CB1fl/+ \nmice were deleted via AAV (after/before 68 ± 17%, n = 3) or when crossed with 575 \nAldh1l1-Cre mice (after/before 69 ± 12%, n = 4, t-test p = 0.96; pooled in Fig. 6E – H). \nPPR and CV analysis also confirmed a presynaptic locus of tLTD expression in CB1fl/+ \nmice (G, H). \nThese findings show that tLTD relies on astrocytic CB1 receptors, and that partial \nCB1 receptor deletion does not disrupt this form of plasticity. Consistent with impaired 580 \ntLTD and a reduced capacity for synaptic weakening, age-matched EPSP amplitudes \nwere furthermore larger in CB1 knockout and CalEx conditions (Supp. Fig. 2). \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 21 of 37 \n \nFigure 6 | Astrocyte-specific deletion of CB1 receptors prevents L5 PC → PC tLTD  585 \n(A) Neonatal AAV injections were performed at P0–P2 in heterozygous or \nhomozygous CB1f mice, or CB1f mice were bred with Aldh1l1-Cre mice. Experiments \nwere conducted at P19–P23. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 22 of 37 \n(B) Representative quadruple-patch experiment used to assess monosynaptic PC → \nPC connections (blue, Alexa 594) in slices with astrocyte-specific Cnr1 deletion (red, 590 \nEGFP). Presynaptic PC1 was connected to postsynaptic PC2. \n(C) Example recording from tLTD induction experiments in Cnr1fl/+ slice. Gray bar \nindicates the tLTD induction period. Inset scale bars: 25 ms and 0.2 mV. \n(D) Example recording from the same PC1 → PC2 connection as in (B), showing \nfailed tLTD in a Cnr1fl/fl slice. Gray bar indicates the tLTD induction period. Inset scale 595 \nbars: 25 ms and 0.2 mV. \n(E) tLTD was robust in Cnr1fl/+ (blue) but not in Cnr1fl/fl slices (red). \n(F) Homozygous astrocyte-specific CB1 receptor deletion abolished tLTD (Cnr1fl/fl: \n114 ± 7% vs 100%, one-sample t-test p = 0.1), whereas heterozygous deletion did not \n(Cnr1fl/+: 68 ± 9% vs 100%, one-sample t-test p < 0.05; genotype comparison t-test p < 600 \n0.01). \n(G) ΔPPR for Cnr1fl/+ group was consistent with presynaptically expressed tLTD \n(ΔPPR Cnr1fl/+ 0.14 ± 0.08, n = 6 vs Cnr1fl/fl -0.14 ± 0.07, n = 4, t-test p < 0.05). \n(H) CV analysis in the Cnr1fl/+ group showed most data points below the identity \nline, consistent with a presynaptic expression (φ = 13 ± 6°, Wilcoxon signed rank p = 605 \n0.09). In addition, 1 𝐶𝑉𝑛𝑜𝑟𝑚\n2⁄  was reduced (45 ± 9%, t-test p < 0.01, n = 6), suggesting \nreduced release probability (Brock et al., 2020). \n \nCB1 receptor activation alters astrocyte Ca2+ dynamics \nWe argued that if tLTD relies on astrocytic CB1 receptor signaling, astrocytic Ca2+ 610 \nactivity may be altered by ACEA and by tLTD. \nUsing 2p Ca2+ imaging (Fig. 7A – C), we found that ACEA increased astrocyte \nCa2+ event frequency and reduced event duration compared to control (Fig. 7D). In \naddition, Ca2+ activity became more spatially coordinated within individual astrocytes, as \nreflected by increased correlations across regions of interest (Fig. 7E). These changes are 615 \nconsistent with coordinated activation following CB1 receptor stimulation. \nHowever, in astrocytes patched <100 µm from neurons that underwent tLTD \ninduction, there were no detectable changes in Ca2+ event frequency (before: 63 ± 14 vs \nafter: 70 ± 9 , n = 3; paired t-test p = 0.5) or duration (21 ± 9 s vs 16 ± 6 s , n = 3; paired \nt-test p = 0.5). In addition, Ca2+ activity did not synchronize, as reflected by unchanged 620 \ncross correlation across regions of interest (Δ mean Z-score for tLTD: 0.6 ± 0.2, n = 3 vs \nControl: 1.4 ± 0.5, n = 3; t-test p = 0.1).  \nThis contrasts with previous reports that astrocyte Ca2+ activity increases during \ntLTD induction (Min and Nevian, 2012) and indicates that, under our conditions, tLTD \ndoes not measurably alter astrocyte Ca2+ dynamics. However, this does not exclude the 625 \npossibility that synapse-specific astrocyte Ca2+ microdomains were engaged during tLTD \ninduction but were too small, too brief, or too spatially restricted to be resolved under our \nrecording conditions. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 23 of 37 \nIn conclusion, CB1 receptor activation robustly alters astrocyte Ca2+ dynamics, \nwhereas tLTD induction does not produce detectable changes under our experimental 630 \nconditions.  \n \nFigure 7 | CB1 receptor activation alters astrocyte Ca2+ dynamics. \n(A) Fluo-5F Ca2+ signals (green, arrowheads) were recorded as a 150-second-long \nmovie (see Videos S1 and S2) of this sample astrocyte. During offline analysis, twelve 635 \nROIs (red, top) were manually selected. Comparing to relatively quiet sample frame a \n(bottom), Ca2+ transients are visible in in sample frames b, c, and d (arrowheads). \n(B) Based on ROIs in (A), fluorescence was quantified as dG/R sweeps before and \nafter ACEA wash-in. Vertical dashed lines denote the timepoints of frames a through d \ndepicted in (A), bottom. Note synchrony across most ROIs at timepoint d. 640 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 24 of 37 \n(C) Sample ROI cross-correlation matrices before and after ACEA wash-in for \nastrocyte in (A). For each matrix, a single mean Z-score for Pearson’s r was calculated as \na metric for how correlated the activity of a cell was.  \n(D) The number of Ca2+ events was increased by ACEA (45 ± 4 vs 60 ± 7, n = 9; \npaired t-test p < 0.05) but not in controls (58 ± 4 vs 60 ± 10, n = 5; p = 0.8). The duration 645 \nof Ca2+ events, however, decreased during ACEA wash-in (24 ± 3 s vs 18 ± 3 s, n = 9; p \n< 0.05) but not in control (18 ± 3 s vs 21 ± 4 s, n = 5; p = 0.4). Amplitude did not differ \nfor either condition (ACEA 2.7 ± 0.7 vs 2.9 ± 0.6, n = 9, p = 0.7, and Control 3.1 ± 1 vs \n2.9 ± 1, n = 5, p = 0.6). Closed circles indicate sample astrocyte shown in (A). \n(E) ACEA increased event correlation (Δ mean Z-score: 7.3 ± 4, n = 7) compared 650 \nto controls (0.6 ± 0.1, n = 5; Mann-Whitney U test p < 0.05). Diamonds denote means. \nClosed circle indicates sample astrocyte shown in (A). \n \n \nFigure 8 | Updated working model of L5 PC → PC tLTD 655 \nLeft: The original model of tLTD proposed that coincident activation of presynaptic \nNMDA and CB1 receptors is required to induce tLTD at L5 PC → PC synapses \n(Sjöström et al., 2003). Right: We now update this framework to include a central role for \nastrocytes, in which astrocytic CB1 receptors and Ca2+ signaling mediate the induction of \ntLTD. While in L4 → L2/3 tLTD requires glutamate release from astrocytes (Min and 660 \nNevian, 2012), tLTD at other synapses requires different gliotransmitters, such as D-\nserine (Andrade-Talavera et al., 2016; Andrade-Talavera et al., 2024; Martínez-Gallego \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 25 of 37 \net al., 2024). Pre: presynaptic terminal, Post: postsynaptic terminal, Astro: astrocyte, \nbAP: back propagating action potential, eCB: endocannabinoid, CB1R: cannabinoid \nreceptor type-1, NMDAR: NMDA receptor, Glu: glutamate.  665 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 26 of 37 \nDiscussion \nHere, we revisited the cellular basis of tLTD at visual cortex L5 PC → PC \nsynapses. Although this plasticity has been attributed to presynaptic NMDA and CB1 \nreceptors, our findings reveal that astrocytes are a critical third player. Using \ncomplementary loss-of-function approaches, we found that astrocyte Ca2+ signaling and 670 \nastrocytic CB1 receptors are essential for tLTD induction. Moreover, optogenetic \nactivation of astrocytic Gq signaling during induction abolished tLTD and shifted the \noutcome toward potentiation, highlighting a bidirectional astrocytic influence.  \nAstrocytes are required for L5 PC → PC tLTD  \nWe found that astrocytes are essential for inducing L5 PC → PC tLTD in mouse 675 \nvisual cortex. Disruption of glial metabolism with NaFAC — a glia-specific Krebs cycle \ninhibitor (Swanson and Graham, 1994) — abolished tLTD, consistent with prior reports \nthat NaFAC impairs both LTD (Zhang et al., 2008) and LTP (Fossat et al., 2012; Han et \nal., 2015; Boddum et al., 2016). To rule out nonspecific effects, we also employed more \ntargeted astrocyte-specific approaches. 680 \nWe interfered with astrocyte Ca2+ signaling using two orthogonal strategies: the \nplasma membrane Ca2+ ATPase CalEx (Yu et al., 2018), and intracellular loading of the \nfast Ca2+ chelator BAPTA. Both manipulations reliably abolished tLTD, confirming that \nastrocyte Ca2+ transients are required. BAPTA is expected to target fast, local astrocyte \nCa2+ signals (Tsien, 1980), while CalEx dampens global Ca2+ transients in astrocytes by 685 \nextruding cytosolic Ca2+ (Yu et al., 2018). Their convergent effects suggest that both \nlocal and global astrocytic Ca2+ signals contribute to tLTD. \nWe found that buffering Ca2+ in astrocytes within ~100 µm of the postsynaptic \nneuron was sufficient to abolish tLTD, consistent with BAPTA spreading via gap \njunctions throughout the syncytial astrocyte network (Watanabe et al., 2023). This 690 \noutcome agrees with prior studies in cortex and hippocampus (Andrade-Talavera et al., \n2016; Andrade-Talavera et al., 2024; Martínez-Gallego et al., 2024). \nOur findings thus revise our earlier model of L5 PC → PC tLTD, which posited \nthat coincident activation of presynaptically located CB1 and NMDA receptors drove \ntLTD (Sjöström et al., 2003; Duguid and Sjöström, 2006). Instead, our new results 695 \nposition astrocytes as critical intermediaries in this form of STDP (Fig. 8). \nAstrocytic Gq signaling promotes potentiation \nOptogenetic activation of astrocytic Gq signaling disrupted tLTD and drove what \nappeared to be LTP, as evidenced by increased EPSP amplitude, an inverse correlation \nwith baseline strength (Sjöström et al., 2001), and a shift in CV consistent with mixed 700 \npre- and postsynaptic expression (Sjöström et al., 2007). These observations suggest that \nastrocytes not only gate plasticity but can shift its polarity. \nIt is possible that Gq signaling biases gliotransmitter release toward LTP-promoting \ncues such as D-serine or glutamate. Astrocyte-driven LTP following Gq-DREADD \nactivation has been reported to be NMDA receptor-dependent (Adamsky et al., 2018). 705 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 27 of 37 \nOur results are broadly consistent with this and additionally suggest that astrocytes, via \nGq activation, can override tLTD and instead promote potentiation. \nThis effect appears to be context- and circuit-dependent. While astrocyte Gq \nactivation triggered LTP at our visual cortex L5 PC → PC synapses, Gq activation \ninduces LTD at corticostriatal synapses (Cavaccini et al., 2020), highlighting the 710 \nspecificity of astrocyte-mediated modulation. Other modes of astrocyte activation — \nincluding melanopsin stimulation (Mederos et al., 2019) and cAMP elevation (Zhou et \nal., 2021) — can also induce LTP. Multiple astrocytic signaling pathways may thus \nconverge to facilitate potentiation. \nTogether, these findings expand current models of STDP by showing that 715 \nastrocytes are not passive but active integrators in plasticity. \nAstrocyte CB1 receptors mediate cannabinoid-dependent tLTD \nWe identified astrocyte CB1 receptors as an essential locus of eCB signaling in \ntLTD. Bath application of ACEA, a synthetic analog of anandamide, mimicked and \noccluded tLTD (Sjöström et al., 2003). This aligns with recent findings that anandamide 720 \npreferentially activates astrocytic CB1Rs, while 2-arachidonoylglycerol preferentially \nengages neuronal CB1Rs (Noriega-Prieto et al., 2025). The sensitivity of L5 PC → PC \nsynapses to ACEA, along with the abolition of tLTD by astrocyte-targeted CalEx, \nsuggests that the relevant eCB signaling is mediated by astrocytes rather than neurons. \nCrucially, conditional deletion of CB1 receptors from astrocytes abolished tLTD, 725 \nsimilar to disrupting astrocytic Ca2+ signaling. Astrocytes thus decode activity-dependent \neCB release via their CB1 receptors that could be Gq-coupled (Navarrete and Araque, \n2008), mobilize intracellular Ca2+, and communicate back to neurons through the release \nof gliotransmitters, functioning as critical intermediaries between pre- and postsynaptic \ncompartments. 730 \nBoth deletion of CB1 receptors from astrocytes and disruption of astrocytic Ca2+ \nsignaling resulted in increased strength of L5 PC → PC synapses. This is consistent with \nimpaired tLTD and a reduced capacity for synaptic weakening, although contributions \nfrom other disruptions of astrocyte function cannot be excluded. \nAstrocytic CB1 receptors are increasingly recognized as modulators of plasticity 735 \nand behavior. They have been implicated in D-serine regulation and recognition memory \n(Robin et al., 2018), in stress resilience (Dudek et al., 2025), and in context-dependent \ninflammatory signaling (Colomer et al., 2026). The latter study proposed that distinct \nastrocytic CB1 receptor pools located at the plasma membrane or within mitochondria \nmay mediate opposing effects depending on the inflammatory milieu. 740 \nOur findings add to this growing body of work, highlighting that CB1 receptor-\ncentric plasticity models not involving astrocytes may need to be revisited. \nFunctional implications of tripartite tLTD \nOur findings place astrocytes as key contributors to cortical tLTD, suggesting a \nfunctional tripartite L5 PC → PC synapse (Perea et al., 2009). This supports three-factor 745 \nlearning rules (McFarlan et al., 2023), where a modulatory signal influences Hebbian \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 28 of 37 \nplasticity (Pawlak et al., 2010). Astrocytes are ideally suited for this third role, integrating \nlocal activity and neuromodulatory context to gate plasticity outcomes (Falcón-Moya et \nal., 2020; Martinez-Gallego et al., 2022; Lefton et al., 2025). \nAstrocyte involvement may help explain why STDP varies across synapse types 750 \n(Larsen and Sjöström, 2015; McFarlan et al., 2023). Differences in astrocyte coupling, \nreceptor profiles, and brain state sensitivity provide a flexible substrate for context-\ndependent circuit tuning (Min and Nevian, 2012; Andrade-Talavera et al., 2016; Letellier \net al., 2016; Falcón-Moya et al., 2020; Letellier and Goda, 2023). \nMoreover, astrocyte calcium dynamics track global states like arousal and sleep 755 \n(Poskanzer and Yuste, 2016). Tripartite tLTD thus enables participation in behaviorally \nrelevant circuit reconfiguration. By identifying astrocytes as indispensable for tLTD, our \nwork reframes this plasticity as a tripartite learning process. \nCaveats and limitations \nThis study was conducted in acute slices from juvenile mice. It remains to be 760 \ndetermined whether our findings generalize to more mature developmental stages or in \nvivo conditions.  \nThe OptoGq experiments relied on extracellular stimulation, which in neocortex \nmay non-specifically recruit multiple synaptic pathways (Dragsted et al., 2025). Given \nthe well-established synapse-type specificity of cortical plasticity (Larsen and Sjöström, 765 \n2015; McFarlan et al., 2023), such non-specificity could confound interpretation. \nHowever, the majority of our experiments targeted identified L5 PC → PC connections, \neither using paired recordings (Lalanne et al., 2016) or 2p optogenetic stimulation (Chou \net al., 2024). This allowed us to link observed plasticity effects specifically to the L5 PC \n→ PC synapse type.  770 \nAstrocyte Ca2+ imaging was acquired at 2.11 Hz, sufficient to capture global Ca2+ \ntransients in processes, but likely insufficient to detect fast, localized microdomain events \nin fine astrocytic branches. Functionally important Ca2+ signals may therefore have gone \nundetected. \nWhile PPR and CV analyses suggested a presynaptic locus of expression, these are 775 \nindirect measures (Brock et al., 2020). The downstream pathways through which \nastrocytic CB1 receptor activation modulates presynaptic function remain unclear and \nwarrant further investigation. \nConclusions and outlook \nOur study challenges canonical neuron-centric models of synaptic plasticity by 780 \ndemonstrating that astrocytes decode eCB signaling to enable tLTD (Fig. 8). By \nidentifying astrocytes as active participants, our results support the emerging tripartite \nframework in which glia dynamically gate synaptic change (Perea et al., 2009; \nHenneberger et al., 2010). More broadly, our findings place astrocytes within three-factor \nlearning rules, in which modulatory signals shape synaptic plasticity (Pawlak et al., 2010; 785 \nGerstner et al., 2018; McFarlan et al., 2023) and add to the growing body of work \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 29 of 37 \nshowing the involvement of astrocytes in STDP (Min and Nevian, 2012; Andrade-\nTalavera et al., 2016; Andrade-Talavera et al., 2024; Martínez-Gallego et al., 2024). \nThe finding that astrocytic Gq signaling controls synaptic plasticity suggests that \nastrocytes are local arbiters of synaptic modification, potentially linking it to network 790 \nstate or neuromodulatory context (Poskanzer and Yuste, 2016). Such mechanisms may \nadd computational power, for instance to address the credit assignment problem. \nFuture work will clarify how astrocytic CB1 receptor activation modulates synaptic \nfunction, determine the gliotransmitters involved, and test whether astrocyte-dependent \ntLTD operates in vivo. Given links to eCB signaling and glial dysfunction in disease, 795 \nthese insights may inform future therapies targeting circuit pathology. \nReferences \nAbrahamsson T, Chou CYC, Li SY, Mancino A, Costa RP, Brock JA, Nuro E, \nBuchanan KA, Elgar D, Blackman AV, Tudor-Jones A, Oyrer J, Farmer WT, Murai KK, \nSjöström PJ (2017) Differential Regulation of Evoked and Spontaneous Release by 800 \nPresynaptic NMDA Receptors. Neuron 96:839-855 e835. \nAdamsky A, Kol A, Kreisel T, Doron A, Ozeri-Engelhard N, Melcer T, Refaeli R, \nHorn H, Regev L, Groysman M, London M, Goshen I (2018) Astrocytic Activation \nGenerates De Novo Neuronal Potentiation and Memory Enhancement. Cell 174:59-71 \ne14. 805 \nAndrade-Talavera Y, Duque-Feria P, Paulsen O, Rodríguez-Moreno A (2016) \nPresynaptic Spike Timing-Dependent Long-Term Depression in the Mouse \nHippocampus. Cereb Cortex 26:3637-3654. \nAndrade-Talavera Y, Sánchez-Gómez J, Coatl-Cuaya H, Rodríguez-Moreno A \n(2024) Developmental Spike Timing-Dependent Long-Term Depression Requires 810 \nAstrocyte d-Serine at L2/3-L2/3 Synapses of the Mouse Somatosensory Cortex. The \nJournal of Neuroscience 44:e0805242024. \nBi GQ, Poo MM (1998) Synaptic modifications in cultured hippocampal neurons: \ndependence on spike timing, synaptic strength, and postsynaptic cell type. Journal of \nNeuroscience 18:10464-10472. 815 \nBliss TV, Collingridge GL (1993) A synaptic model of memory: long-term \npotentiation in the hippocampus. Nature 361:31-39. \nBoddum K, Jensen TP, Magloire V, Kristiansen U, Rusakov DA, Pavlov I, Walker \nMC (2016) Astrocytic GABA transporter activity modulates excitatory \nneurotransmission. Nature Communications 7:13572. 820 \nBrock JA, Thomazeau A, Watanabe A, Li SSY, Sjöström PJ (2020) A Practical \nGuide to Using CV Analysis for Determining the Locus of Synaptic Plasticity. Frontiers \nin Synaptic Neuroscience 12:11 10.3389/fnsyn.2020.00011. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 30 of 37 \nBusquets-Garcia A, Bains J, Marsicano G (2018) CB(1) Receptor Signaling in the \nBrain: Extracting Specificity from Ubiquity. Neuropsychopharmacology 43:4-20. 825 \nCavaccini A, Durkee C, Kofuji P, Tonini R, Araque A (2020) Astrocyte Signaling \nGates Long-Term Depression at Corticostriatal Synapses of the Direct Pathway. The \nJournal of Neuroscience 40:5757. \nChou CYC, Droogers WJ, Lalanne T, Fineberg E, Klimenko T, Owens H, Sjöström \nPJ (2025) Postsynaptic spiking determines anti-Hebbian LTD in visual cortex basket 830 \ncells. Front Synaptic Neurosci 17. \nChou CYC, Wong HHW, Guo C, Boukoulou KE, Huang C, Jannat J, Klimenko T, \nLi VY, Liang TA, Wu VC, Sjöström PJ (2024) Principles of visual cortex excitatory \nmicrocircuit organization. The Innovation 6:1-11. \nCline HT (1998) Topographic maps: developing roles of synaptic plasticity. Curr 835 \nBiol 8:R836-839. \nColomer T et al. (2026) Astrocyte CB1 receptors drive blood-brain barrier \ndisruption in central nervous system inflammatory disease. Journal of \nNeuroinflammation. \nDebanne D, Gähwiler BH, Thompson SM (1999) Heterogeneity of Synaptic 840 \nPlasticity at Unitary CA3–CA1 and CA3–CA3 Connections in Rat Hippocampal Slice \nCultures. The Journal of Neuroscience 19:10664. \nDragsted LK, Alageswaran S, Suvrathan A, Sjöström PJ (2025) Extracellular \nstimulation in neocortex lacks specificity. microPublication Biology. \nDudek KA et al. (2025) Astrocytic cannabinoid receptor 1 promotes resilience by 845 \ndampening stress-induced blood–brain barrier alterations. Nature Neuroscience. \nDuguid I, Sjöström PJ (2006) Novel presynaptic mechanisms for coincidence \ndetection in synaptic plasticity. Curr Opin Neurobiol 16:312-322. \nFalcón-Moya R, Pérez-Rodríguez M, Prius-Mengual J, Andrade-Talavera Y, \nArroyo-García LE, Pérez-Artés R, Mateos-Aparicio P, Guerra-Gomes S, Oliveira JF, 850 \nFlores G, Rodríguez-Moreno A (2020) Astrocyte-mediated switch in spike timing-\ndependent plasticity during hippocampal development. Nat Commun 11:4388. \nFeldman DE (2000) Timing-based LTP and LTD at vertical inputs to layer II/III \npyramidal cells in rat barrel cortex. Neuron 27:45-56. \nFeldman DE (2012) The spike-timing dependence of plasticity. Neuron 75:556-855 \n571. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 31 of 37 \nFossat P, Turpin FR, Sacchi S, Dulong J, Shi T, Rivet JM, Sweedler JV, Pollegioni \nL, Millan MJ, Oliet SH, Mothet JP (2012) Glial D-serine gates NMDA receptors at \nexcitatory synapses in prefrontal cortex. Cerebral Cortex 22:595-606. \nGerstner W, Lehmann M, Liakoni V, Corneil D, Brea J (2018) Eligibility Traces 860 \nand Plasticity on Behavioral Time Scales: Experimental Support of NeoHebbian Three-\nFactor Learning Rules. Front Neural Circuits 12:53. \nGorski JA, Talley T, Qiu M, Puelles L, Rubenstein JL, Jones KR (2002) Cortical \nexcitatory neurons and glia, but not GABAergic neurons, are produced in the Emx1-\nexpressing lineage. J Neurosci 22:6309-6314. 865 \nGuayasamin M, Depaauw-Holt LR, Adedipe, II, Ghenissa O, Vaugeois J, \nDuquenne M, Rogers B, Latraverse-Arquilla J, Peyrard S, Bosson A, Murphy-Royal C \n(2025) Early-life stress induces persistent astrocyte dysfunction associated with fear \ngeneralisation. Elife 13. \nHan H, Peng Y, Dong Z (2015) D-Serine rescues the deficits of hippocampal long-870 \nterm potentiation and learning and memory induced by sodium fluoroacetate. \nPharmacology Biochemistry and Behavior 133:51-56. \nHardingham NR, Hardingham GE, Fox KD, Jack JJB (2007) Presynaptic Efficacy \nDirects Normalization of Synaptic Strength in Layer 2/3 Rat Neocortex After Paired \nActivity. Journal of Neurophysiology 97:2965-2975. 875 \nHebb DO (1949) The Organization of Behavior. New York: Wiley. \nHenneberger C, Papouin T, Oliet SH, Rusakov DA (2010) Long-term potentiation \ndepends on release of D-serine from astrocytes. Nature 463:232-236. \nInstitoris A, Vandal M, Peringod G, Catalano C, Tran CH, Yu X, Visser F, \nBreiteneder C, Molina L, Khakh BS, Nguyen MD, Thompson RJ, Gordon GR (2022) 880 \nAstrocytes amplify neurovascular coupling to sustained activation of neocortex in awake \nmice. Nat Commun 13:7872. \nKatona I, Urban GM, Wallace M, Ledent C, Jung KM, Piomelli D, Mackie K, \nFreund TF (2006) Molecular composition of the endocannabinoid system at \nglutamatergic synapses. J Neurosci 26:5628-5637. 885 \nKatz LC, Shatz CJ (1996) Synaptic activity and the construction of cortical circuits. \nScience 274:1133-1138. \nKim JY, Grunke SD, Levites Y, Golde TE, Jankowsky JL (2014) \nIntracerebroventricular viral injection of the neonatal mouse brain for persistent and \nwidespread neuronal transduction. Journal of visualized experiments : JoVE:51863. 890 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 32 of 37 \nLalanne T, Abrahamsson T, Sjöström PJ (2016) Using Multiple Whole-Cell \nRecordings to Study Spike-Timing-Dependent Plasticity in Acute Neocortical Slices. \nCold Spring Harb Protoc 2016. \nLarsen RS, Sjöström PJ (2015) Synapse-type-specific plasticity in local circuits. \nCurr Opin Neurobiol 35:127-135. 895 \nLefton KB, Wu Y, Dai Y, Okuda T, Zhang Y, Yen A, Rurak GM, Walsh S, Manno \nR, Myagmar B-E, Dougherty JD, Samineni VK, Simpson PC, Papouin T (2025) \nNorepinephrine signals through astrocytes to modulate synapses. Science 388:776-783. \nLetellier M, Goda Y (2023) Astrocyte Calcium Signaling Shifts the Polarity of \nPresynaptic Plasticity. Neuroscience 525:38-46. 900 \nLetellier M, Park YK, Chater TE, Chipman PH, Gautam SG, Oshima-Takago T, \nGoda Y (2016) Astrocytes regulate heterogeneity of presynaptic strengths in hippocampal \nnetworks. PNAS 113:E2685-2694. \nLiao D, Jones A, Malinow R (1992) Direct measurement of quantal changes \nunderlying long-term potentiation in CA1 hippocampus. Neuron 9:1089-1097. 905 \nMalenka RC, Bear MF (2004) LTP and LTD: an embarrassment of riches. Neuron \n44:5-21. \nMarkram H, Gerstner W, Sjöström PJ, eds (2012) Spike-timing-dependent \nplasticity: a comprehensive overview. Lausanne, Switzerland: Frontiers Media SA. \nMarkram H, Lübke J, Frotscher M, Sakmann B (1997) Regulation of synaptic 910 \nefficacy by coincidence of postsynaptic APs and EPSPs. Science 275:213-215. \nMarsicano G, Goodenough S, Monory K, Hermann H, Eder M, Cannich A, Azad \nSC, Cascio MG, Gutierrez SO, van der Stelt M, Lopez-Rodriguez ML, Casanova E, \nSchutz G, Zieglgansberger W, Di Marzo V, Behl C, Lutz B (2003) CB1 cannabinoid \nreceptors and on-demand defense against excitotoxicity. Science 302:84-88. 915 \nMartinez-Gallego I, Perez-Rodriguez M, Coatl-Cuaya H, Flores G, Rodriguez-\nMoreno A (2022) Adenosine and Astrocytes Determine the Developmental Dynamics of \nSpike Timing-Dependent Plasticity in the Somatosensory Cortex. J Neurosci 42:6038-\n6052. \nMartínez-Gallego I, Coatl-Cuaya H, Rodríguez-Moreno A (2024) Astrocytes 920 \nmediate two forms of spike timing-dependent depression at entorhinal cortex-\nhippocampal synapses. eLife 13:RP98031. \nMcFarlan AR, Chou CYC, Watanabe A, Cherepacha N, Haddad M, Owens H, \nSjöström PJ (2023) The plasticitome of cortical interneurons. Nat Rev Neurosci 24:80-97. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 33 of 37 \nMederos S, Hernández-Vivanco A, Ramírez-Franco J, Martín-Fernández M, 925 \nNavarrete M, Yang A, Boyden ES, Perea G (2019) Melanopsin for precise optogenetic \nactivation of astrocyte-neuron networks. Glia 67:915-934. \nMin R, Nevian T (2012) Astrocyte signaling controls spike timing-dependent \ndepression at neocortical synapses. Nat Neurosci 15:746-753. \nMontgomery JM, Pavlidis P, Madison DV (2001) Pair Recordings Reveal All-930 \nSilent Synaptic Connections and the Postsynaptic Expression of Long-Term Potentiation. \nNeuron 29:691-701. \nNabavi S, Fox R, Proulx CD, Lin JY, Tsien RY, Malinow R (2014) Engineering a \nmemory with LTD and LTP. Nature 511:348-352. \nNavarrete M, Araque A (2008) Endocannabinoids mediate neuron-astrocyte 935 \ncommunication. Neuron 57:883-893. \nNavarrete M, Araque A (2010) Endocannabinoids potentiate synaptic transmission \nthrough stimulation of astrocytes. Neuron 68:113-126. \nNoriega-Prieto JA, Falcón-Moya R, Noeker JA, Cai R, Fundazuri UB, Eraso-Pichot \nA, Cai S, Guttipatti P, Belisle L, Rodríguez-Moreno A, van der Stelt M, Li Y, Cheer JF, 940 \nMarsicano G, Kofuji P, Araque A (2025) Distinct endocannabinoids specifically signal to \nastrocytes or neurons in the adult mouse hippocampus. Nature Neuroscience. \nOswald AM, Reyes AD (2008) Maturation of intrinsic and synaptic properties of \nlayer 2/3 pyramidal neurons in mouse auditory cortex. J Neurophysiol 99:2998-3008. \nPawlak V, Wickens JR, Kirkwood A, Kerr JND (2010) Timing is not Everything: 945 \nNeuromodulation Opens the STDP Gate. Front Synaptic Neurosci 2:146. \nPerea G, Navarrete M, Araque A (2009) Tripartite synapses: astrocytes process and \ncontrol synaptic information. Trends Neurosci 32:421-431. \nPoskanzer KE, Yuste R (2016) Astrocytes regulate cortical state switching in vivo. \nPNAS 113:E2675-2684. 950 \nRobin LM et al. (2018) Astroglial CB1 Receptors Determine Synaptic D-Serine \nAvailability to Enable Recognition Memory. Neuron 98:935-944.e935. \nShatz CJ (1992) The developing brain. Scientific American 267:60-67. \nSjöström PJ, Turrigiano GG, Nelson SB (2001) Rate, timing, and cooperativity \njointly determine cortical synaptic plasticity. Neuron 32:1149-1164. 955 \nSjöström PJ, Turrigiano GG, Nelson SB (2003) Neocortical LTD via coincident \nactivation of presynaptic NMDA and cannabinoid receptors. Neuron 39:641-654. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 34 of 37 \nSjöström PJ, Turrigiano GG, Nelson SB (2007) Multiple forms of long-term \nplasticity at unitary neocortical layer 5 synapses. Neuropharmacology 52:176-184. \nSong S, Abbott LF (2001) Cortical development and remapping through spike 960 \ntiming-dependent plasticity. Neuron 32:339-350. \nSong S, Miller KD, Abbott LF (2000) Competitive Hebbian learning through spike-\ntiming-dependent synaptic plasticity. Nat Neurosci 3:919-926. \nSong S, Sjöström PJ, Reigl M, Nelson S, Chklovskii DB (2005) Highly nonrandom \nfeatures of synaptic connectivity in local cortical circuits. PLoS biology 3:e68. 965 \nSwanson RA, Graham SH (1994) Fluorocitrate and fluoroacetate effects on \nastrocyte metabolism in vitro. Brain Research 664:94-100. \nThomazeau A, Rannio S, Brock JA, Wong HH-W, Sjöström PJ (2025) Neocortical \nlayer-5 tLTD relies on non-ionotropic presynaptic NMDA receptor signaling. eLife \n14:RP106284. 970 \nTien AC, Tsai HH, Molofsky AV, McMahon M, Foo LC, Kaul A, Dougherty JD, \nHeintz N, Gutmann DH, Barres BA, Rowitch DH (2012) Regulated temporal-spatial \nastrocyte precursor cell proliferation involves BRAF signalling in mammalian spinal \ncord. Development 139:2477-2487. \nTsien RY (1980) New calcium indicators and buffers with high selectivity against 975 \nmagnesium and protons: design, synthesis, and properties of prototype structures. \nBiochemistry 19:2396-2404. \nValtcheva S, Venance L (2016) Astrocytes gate Hebbian synaptic plasticity in the \nstriatum. Nat Commun 7:13845. \nWatanabe A, Guo C, Sjöström PJ (2023) The Developmental Profile of Visual 980 \nCortex Astrocytes. iScience:106828. \nYu X, Taylor AMW, Nagai J, Golshani P, Evans CJ, Coppola G, Khakh BS (2018) \nReducing Astrocyte Calcium Signaling In Vivo Alters Striatal Microcircuits and Causes \nRepetitive Behavior. Neuron 99:1170-1187 e1179. \nZhang LI, Tao HW, Holt CE, Harris WA, Poo M (1998) A critical window for 985 \ncooperation and competition among developing retinotectal synapses. Nature 395:37-44. \nZhang YP, Oertner TG (2007) Optical induction of synaptic plasticity using a light-\nsensitive channel. Nature methods 4:139-141. \nZhang Z, Gong N, Wang W, Xu L, Xu TL (2008) Bell-shaped D-serine actions on \nhippocampal long-term depression and spatial memory retrieval. Cerebral Cortex 990 \n18:2391-2401. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint \n\nPage 35 of 37 \nZhou Z, Okamoto K, Onodera J, Hiragi T, Andoh M, Ikawa M, Tanaka KF, \nIkegaya Y, Koyama R (2021) Astrocytic cAMP modulates memory via synaptic \nplasticity. Proceedings of the National Academy of Sciences 118:e2016584118. \n 995 \nAcknowledgements \nWe thank Inbal Goshen and her lab for the generous gift of OptoGq AAV. We \nthank Alanna Watt, Keith Murai, Aparna Suvrathan, Anne McKinney, Jonathan Britt, \nAlfonso Araque and his lab, Sabine Rannio, Elia Painchaud-Lakatos, Meghan Appleby, \nand Mackenzie Fredericks for help and insights. We thank Christina Chou for guidance 1000 \non optomapping, Jialin Ding and Zoe Harrington for help with genotyping, staining, and \nimaging. We thank all other members of the Sjöström lab for support and useful \ndiscussions. We thank the Molecular Imaging Platform at the Research Institute of the \nMcGill University Health Centre and facility staff for their contribution to this \npublication. Neonatal injection schematic in Fig. 2, 3, 4, 6 was created in BioRender. 1005 \nWatanabe, A. (2026) https://BioRender.com/ipt45k5 \nFunding Statement \nThis work was supported by CFI LOF 28331 (PJS), CIHR OG 126137 (PJS), CIHR \nNIA 288936 (PJS), FRQS CB 254033 (PJS), NSERC DG 418546-2 (PJS), NSERC DG \n2017-04730 (PJS), and NSERC DAS 2017-507818 (PJS). AW was in receipt of an FRQS 1010 \nAward 335452. CG was supported by NSERC USRA 2023-584320, FRQNT BPCA \n342221, RI-MUHC Studentship, CIHR CGSM 524050, and FRQS Award BF1 346922. \nSA was supported by FRQNT 360969, HBHL, and the RI-MUHC Studentship. The \nfunders had no role in study design, data collection and interpretation, or the decision to \nsubmit the work for publication. 1015 \nConflict of Interest Statement \nThe authors declare that the research was conducted in the absence of any \ncommercial or financial relationships that could be construed as a potential conflict of \ninterest. \nAuthor Contribution 1020 \nAW, CG, and PJS conceptualized the study and designed the experiments. AW and \nCG carried out experiments. AW, CG, SA, and PJS conducted formal analysis. PJS wrote \ncustom software. AW, CG and PJS wrote the manuscript. \nData Availability Statement \nThe raw data supporting the conclusions of this manuscript will be made available 1025 \nby the authors, without undue reservation, to any qualified researcher. \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719669doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}