CD51 delineates functionally distinct astrocyte subtypes in neural circuits

Using fluorescence-activated cell sorting (FACS)-based approaches in the adult mouse brain, our previous studies identified CD51, the αV integrin subunit encoded by Itgav, as a cell surface marker capable of subdividing subsets of astrocytes with unique molecular properties⁴. Whether CD51-expressing astrocytes are endowed with unique properties in vivo in functioning neural circuits remains unclear. We therefore generated a mouse line that enables the selective targeting of CD51⁺ astrocytes in the mouse brain. To achieve this, we engineered a CD51-FLP BAC (bacterial artificial chromosome) mouse line and intercrossed the CD51-FLP founder line with Rosa-FSF-tdTomato mice (where FSF denotes FRT-stop-FRT), such that the reporter is expressed in the presence of FLP activity (CD51-FLP;Rosa-FSF-tdTomato) (Fig. 1a). Analysis of this mouse line revealed expression of CD51 protein in tdTomato-expressing cells, indicating successful targeting of this population (Extended Data Fig. 1c–e). tdTomato expression was also evident in a subset of endothelial cells, but rarely in other cells types (Extended Data Fig. 2a,b). Importantly, tdTomato expression in astrocytes was highly concordant with immunoreactivity to an anti-CD51 antibody (Extended Data Figs. 1c–e and 2c,d). To assess the extent to which the CD51-FLP;Rosa-FSF-tdTomato line targets astrocytes in the brain of 8-week-old mice, we co-stained for the pan-astrocyte marker SOX9 and evaluated the constituency of CD51⁺ (tdTomato⁺SOX9⁺) and CD51⁻ (tdTomato⁻SOX9⁺) astrocytes across a host of brain regions (Fig. 1b,c and Extended Data Fig. 1a,b). This analysis revealed the presence of CD51⁺ and CD51⁻ astrocytes in varying proportions throughout the brain, with CD51⁺ astrocytes representing approximately 19% of hippocampal astrocytes, 13% of olfactory bulb astrocytes and 38% of cortical astrocytes (Fig. 1c). Notably, our finding that 19% of astrocytes in the hippocampus are CD51⁺ is consistent with publicly available scRNA-seq data from the adult mouse hippocampus¹¹ (Extended Data Fig. 1f). These data indicate that the CD51-FLP mouse can distinguish CD51⁺ and CD51⁻ astrocytes in the adult brain.

We next sought to assess the cellular properties of CD51⁺ and CD51⁻ astrocyte populations in the cortex and hippocampus, focusing on cell morphology and Ca²⁺ activity. The CD51-FLP mouse line allows us to use the FLP-FRT system to conditionally express reporters using intersectional viral approaches in CD51⁺ or CD51⁻ astrocytes^12,13. Combining these FLP-based viral tools with the CD51-FLP mouse, we can label CD51⁺ and CD51⁻ astrocyte subpopulations with tdTomato (Fig. 1d). Accordingly, we injected the cortex or hippocampus of 4-week-old CD51-FLP mice, collected mouse tissue at 8 weeks of age and assessed expression of the conditional tdTomato reporter. Using the tdTomato reporter to assess morphology, we found that CD51⁺ astrocytes in the cortex exhibit enhanced morphological complexity compared with CD51⁻ astrocytes; by contrast, CD51⁺ astrocytes in the hippocampus exhibited comparatively decreased complexity (Fig. 1e,f). Using similar approaches, we generated conditional adeno-associated viruses (AAVs) that enabled us to selectively target CD51⁺ and CD51⁻ astrocytes with GCaMP6, an optical sensor that reliably reports Ca²⁺ activity¹⁴ (Fig. 1g). Using the same workflow as described above, we used two-photon imaging to measure Ca²⁺ activity in ex vivo slices from the cortex or hippocampus and found that CD51⁺ astrocytes in the cortex exhibit enhanced spontaneous Ca²⁺ activity compared with CD51⁻ astrocytes (Fig. 1h,i). By contrast, in the hippocampus, CD51⁻ astrocytes exhibited elevated spontaneous Ca²⁺ activity compared with CD51⁺ astrocytes (Fig. 1j,k). Together, these studies indicate that CD51⁺ and CD51⁻ astrocytes are endowed with unique morphological and physiological properties within and across the cortex and hippocampus.

The identification of unique properties for CD51⁺ and CD51⁻ astrocytes in the adult brain led us to next examine whether they also exhibit unique developmental characteristics, such as proliferation and timelines of expansion. Using postnatal cortical development as a model, in conjunction with the CD51-FLP;Rosa-FSF-tdTomato mouse line, we evaluated the constituency of CD51⁺ and CD51⁻ astrocytes during the postnatal day 1 (P1) to P28 developmental interval when astrocytogenesis occurs in the cortex^15,16,17. This analysis revealed a steady increase in the proportion of CD51⁺ astrocytes between P1 and P21, with the relative number of tdTomato⁺SOX9⁺ astrocytes peaking at about 40% by P21 (Fig. 1l–n). This finding was conserved across individual cortical layers (Extended Data Fig. 2e). These observations suggest that CD51⁺ cells undergo a proliferative expansion during this interval. Given that around 50% of cortical astrocytes are generated by local proliferation outside the germinal zones and in the brain parenchyma between P1 and P21 (refs. ^15,16,18), we next assessed the relative proliferation rates of these populations. Using EdU-labelling approaches (Fig. 1l), we found that CD51⁻ astrocytes are highly proliferative at P1, which is followed by enhanced proliferation of CD51⁺ astrocytes at P7–21 (Fig. 1m–o). Immunostaining for the cell cycle markers Ki-67 and phosphorylated histone H3 yielded the same pattern of results (Extended Data Fig. 2f–k). These data are consistent with our observation of increased numbers of CD51⁺ astrocytes during this interval, indicating that CD51⁺ astrocytes proliferate and expand in the brain parenchyma during the P1–21 interval. By contrast, CD51⁻ astrocytes appear to be predominantly generated earlier in development and exhibit less proliferative capacity during the later stages of the P1–21 interval.

We next sought to understand the origin of CD51⁺ astrocytes. Analysis of developing astrocyte RNA-seq data¹⁹ revealed that expression of Itgav emerged after birth in astrocyte lineage cells (Extended Data Fig. 3a). Concordantly, using CD51-FLP;Rosa-FSF-tdTomato mice, we found that SOX2⁺ precursors within the ventricular/subventricular zone did not express tdTomato (CD51) at the onset of cortical gliogenesis at embryonic day 16.5 (E16.5), but a subset was tdTomato positive at P1 (Extended Data Fig. 3c–e). These data suggest that CD51⁺ astrocytes are specified after the onset of gliogenesis. Notably, we identified binding motifs for the glial development transcription factor nuclear factor I-A (NFIA) in upstream regulatory regions of Itgav (Extended Data Fig. 3g), suggesting that NFIA could be involved in regulating CD51 expression and forming CD51⁺ astrocytes. To test this, we generated Aldh1l1-creER;NFIA-flox;Rosa-LSL-tdTomato mice, which enabled tamoxifen-inducible deletion of Nfia and expression of tdTomato in astrocytes (hereafter, Nfia conditional knockout (cKO); Extended Data Fig. 3h). Aldh1l1-creER;Rosa-LSL-tdTomato mice were used as controls. We injected pups with tamoxifen within 24 h of birth, when relatively few CD51⁺ astrocytes are present in the cortex (Fig. 1n). At P28, we found that Nfia-cKO mice had a significantly smaller proportion of CD51⁺ astrocytes in the cortex compared with the control mice (Extended Data Fig. 3h–l), indicating that NFIA regulates the development of CD51⁺ astrocytes.

Postnatal astrocyte development is regulated in part by extrinsic cues through neuronal-activity-dependent mechanisms^20,21. To test whether neuronal activity affected the proportions of CD51⁺ or CD51⁻ astrocytes, we used G_q-DREADD (designer receptors exclusively activated by designer drugs) stimulation of cortical excitatory neurons from P7–21 adapted from our previous study²¹ (Extended Data Fig. 3m–r). However, we found that this manipulation did not affect the proportions of astrocytes that were CD51⁺ or CD51⁻. Thus, the origin of the diverse CD51⁺ and CD51⁻ astrocyte populations that we have identified appears to be predominantly dependent on cell-intrinsic genetic programs rather than neuronal activity.

Having found that CD51⁺ and CD51⁻ astrocytes exhibit unique properties, we next examined whether these populations demonstrate distinct molecular profiles. To achieve this, we intercrossed the CD51-FLP;Rosa-FSF-tdTomato line with the Aldh1l1-GFP mouse line (Fig. 2a). This approach enabled us to FACS-isolate CD51⁺ (GFP⁺tdTomato⁺) and CD51⁻ astrocytes (GFP⁺tdTomato⁻) from the cortex and hippocampus of 8-week-old mice (Extended Data Fig. 1c) and perform transcriptomic analysis using RNA-seq. Bioinformatics analysis revealed that the isolated CD51⁺GFP⁺ and CD51⁻GFP⁺ populations exhibit molecular signatures of astrocytes and demonstrate transcriptomic profiles that are distinct from one another (Extended Data Fig. 4a–d,f and Supplementary Table 1). Moreover, comparison with FACS-isolated astrocyte populations from our previous studies⁴ revealed a strong correlation between CD51⁺GFP⁺ astrocytes and population C, which also expresses CD51 (Extended Data Fig. 4e). Next, we assessed Gene Ontology (GO) terms enriched in CD51⁺ astrocytes from the cortex, finding genes associated with synaptic function, synapse assembly and interactions with neurons enriched in this population (Fig. 2b,c and Supplementary Table 2). Given that astrocytes have essential roles in synapse formation^22,23, we next assessed the expression of known synaptogenic genes in CD51⁺ and CD51⁻ astrocytes, finding enriched expression of glypican family members in CD51⁺ astrocytes (Fig. 2d and Extended Data Fig. 5). In summary, these data indicate that CD51⁺ and CD51⁻ astrocytes have unique molecular features.

The foregoing transcriptomic analyses indicate that CD51⁺ astrocytes are endowed with unique synaptogenic properties. Moreover, our previous in vitro studies suggest that CD51⁺ populations engender enhanced synapse formation on cultured neurons⁴. Together, these observations led us to examine the capacity of CD51⁺ and CD51⁻ astrocytes to promote synapse formation in vivo in the developing postnatal cortex. Previous studies have shown that chemogenetic, DREADD-based (hM3Dq) stimulation of astrocytes promotes neuronal circuit activity^24,25. These collective observations led us to examine whether DREADD-based stimulation of CD51⁺ or CD51⁻ astrocytes can promote synapse formation during postnatal cortical development. To test this, we combined conditional AAV tools with FLP-on or FLP-off elements regulating hM3Dq expression with the CD51-FLP mouse lines, such that, in one condition, hM3Dq is expressed exclusively in CD51⁺ astrocytes and, in the other condition, it is expressed exclusively in CD51⁻ astrocytes (Fig. 2e–h). Following our established protocol for chemogenetic stimulation in the developing cortex²¹, we injected AAVs at P0–2, treated the mice with 5 mg per kg of clozapine N-oxide (CNO) (or saline as a control) from P7–21 to stimulate astrocytes and collected mouse tissue at P21 (Fig. 2e). We next used immunostaining to assess synapse numbers in layers II/III of the cortex, finding that DREADD-based stimulation of CD51⁺ astrocytes promoted a robust increase in the formation of excitatory synapses (Fig. 2i,j), as well as inhibitory synapses (Extended Data Fig. 6a–d). Notably, stimulation of CD51⁻ astrocytes did not significantly enhance excitatory or inhibitory synapse formation (Fig. 2i,j and Extended Data Fig. 6a–d). In parallel, we performed electrophysiological recordings from layer II/III at P21 and found that stimulation of CD51⁺ astrocytes led to a concordant increase in synaptic activity, while stimulation of CD51⁻ astrocytes did not (Fig. 2k,l). Notably, after CD51⁺ astrocyte stimulation, we observed an increase in synaptic current frequency, but not amplitude, which is consistent with an increase in the number of synapses (Fig. 2k,l). Critically, our astrocyte stimulation paradigm did not alter neuronal numbers or layering, indicating the observed phenomenon is not due to alterations in the constituency or organization of cortical neurons (Extended Data Fig. 7). hM3D⁺ astrocytes robustly expressed FOS 90 min after the last CNO injection, validating that our repeated stimulation paradigm enabled persistent astrocyte activation (Extended Data Fig. 8a–c). Moreover, repeated CNO administration in the absence of hM3D did not affect synapse numbers (Extended Data Fig. 8d–f). These data indicate that stimulation of CD51⁺ astrocytes selectively promotes synapse formation in the developing cortex.

These observations led us to investigate the mechanisms by which CD51⁺ astrocytes promote synapse formation in the developing cortex. Our transcriptomic analysis identified members of the glypican family as demonstrating enriched expression in cortical CD51⁺ astrocytes^26,27 (Fig. 2d). Using viral approaches in CD51-FLP mice to label CD51⁺ and CD51⁻ astrocytes, we found that glypican 6 (GPC6) protein expression is enriched in CD51⁺ astrocytes (Extended Data Fig. 5c). This led us to next examine whether Gpc6 expression in CD51⁺ astrocytes is necessary for the enhanced synaptogenic effect we observed (Fig. 2i–l). To achieve this, we intercrossed the CD51-FLP line with the Rosa-LSL-Cas9 line generating the CD51-FLP;Rosa-LSL-Cas9-eGFP mouse line (where LSL denotes loxP-stop-loxP, for Cre-dependent Cas9-eGFP expression). Using a similar paradigm as described above, mice were injected with an AAV containing single guide RNAs (sgRNAs) targeting Gpc6 (omitted in the controls), an AAV containing an FLP-inducible astrocyte-specific Cre and the abovementioned AAV driving FLP-dependent astrocyte-specific hM3Dq expression (Fig. 2m). Together, this system enabled us to delete Gpc6 from and subsequently stimulate CD51⁺ astrocytes. Mice were injected at P1, treated with CNO beginning at P7 and collected at P21 (Fig. 2m); subsequent analysis of GPC6 expression revealed robust and selective KO in CD51⁺ astrocytes expressing Cre (Extended Data Fig. 6e–g). We next assessed synapse formation through immunostaining and found that KO of Gpc6 eliminated hM3Dq-induced increases in excitatory synapse formation (Fig. 2n,o). Furthermore, overexpression of Gpc6 with AAV-GFAP-Gpc6 injected intracortically between P1 was sufficient to increase synapse numbers measured at P21 (Extended Data Fig. 8g–k). RNA-seq analysis of hM3D–mCherry⁺CD51⁺ astrocytes following our CNO or saline injection paradigm (Extended Data Fig. 9) revealed an upregulation of pathways related to protein trafficking to distal processes, suggesting that G_q stimulation may promote delivery of GPC6 to astrocyte leaflets where it drives synapse formation. Collectively, these data indicate that CD51⁺ astrocytes selectively promote synapse formation and use GPC6 as part of this mechanism.

The preceding finding established that CD51⁺ and CD51⁻ astrocyte populations have distinct functional properties towards synaptogenesis in the developing cortex. We next sought to expand our functional comparison between CD51⁺ and CD51⁻ astrocytes to other brain regions, associated circuits and stages of life, focusing on the adult hippocampus. Transcriptomic analysis of hippocampal CD51⁺ astrocytes indicates that they are also enriched in synaptogenic and synaptic transmission-related genes when compared with CD51⁻ astrocytes (Fig. 2c,d and Extended Data Fig. 4a–c). These observations, coupled with the synaptogenic function of CD51⁺ astrocytes in the developing cortex, led us to hypothesize that they may also selectively influence hippocampal circuits and associated memory behaviours. To test this, we combined the CD51-FLP mouse with the conditional AAV tools that enable selective expression of hM3Dq in CD51⁺ or CD51⁻ astrocytes (Figs. 2f and 3a), injecting the hippocampus of 8-week-old mice. Key controls included using AAVs containing hM3Dq that did not selectively target CD51⁺ or CD51⁻ astrocytes, as previous studies have shown that pan-astrocytic stimulation of hippocampal astrocytes can enhance hippocampal circuit function^24,25. As a frontline experiment, we collected mouse tissue 4 weeks after injection, generated hippocampal slices and evaluated long-term potentiation (LTP). Using subthreshold stimulation (40 Hz, 10 pulses) of the Schaffer collateral, we found that stimulation of pan-astrocytes, CD51⁺ astrocytes and CD51⁻ astrocytes were all capable of inducing LTP (Extended Data Fig. 10c,d). In parallel, we generated another set of these experimental groups and performed a series of memory behaviours, finding that DREADD-based stimulation led to enhanced memory performance in each group (Extended Data Fig. 10e,f).

In the hippocampus, around 80–85% of astrocytes are CD51⁻, while about 15–20% of astrocytes are CD51⁺ (Fig. 1c). Our observation that stimulation of CD51⁺ astrocytes yielded the same enhancement of LTP and memory performance as CD51⁻ astrocytes, despite four to sixfold less representation in the hippocampus, suggests that CD51⁺ astrocytes are significantly more effective at enhancing hippocampal circuit plasticity than CD51⁻ astrocytes. To test this, we repeated these experiments using a dilution of the GFAP-FRT-hM3Dq-mCherry-FRT AAV that targets CD51⁻ astrocytes. We found that a 1:50 dilution restricts expression in CD51⁻ astrocytes to around 15–20% of all astrocytes, matching the proportion of CD51⁺ astrocytes in the hippocampus (Extended Data Fig. 10a,b). Similar to the above studies, we injected CD51-FLP mice with undiluted versions of pan-astrocytic and CD51⁺ targeting virus, as well as the 1:50 dilution of the CD51⁻ targeting virus at 8 weeks and collected mouse tissue 4 weeks later for LTP studies, while also conducting memory behaviour analysis on a separate cohort at the same timepoint. Notably, these studies revealed that stimulation of around 20% of CD51⁻ astrocytes was incapable of inducing LTP at subthreshold levels or enhancing memory performance, while stimulation of CD51⁺ astrocytes induced LTP and enhanced memory performance (Fig. 3d–g). Consistent with these findings, analysis of basal synaptic activity of CA1 neurons revealed that stimulation of CD51⁺ astrocytes promotes synaptic activity, while targeting around 20% of CD51⁻ astrocytes did not (Fig. 3b,c). Additional studies revealed no difference in paired-pulse ratio, indicating that stimulation of CD51⁺ astrocytes enhances synaptic input and connectivity (Extended Data Fig. 11). Furthermore, the levels of LTP, synaptic activity and memory performance after stimulation of CD51⁺ astrocytes was comparable to pan-astrocyte stimulation (Fig. 3c–g and Extended Data Fig. 10c–f). Collectively, these data indicate that CD51⁺ astrocytes are more effective than CD51⁻ astrocytes at promoting hippocampal circuit plasticity and associated memory behaviours.

Our studies in the developing cortex and adult hippocampus indicate that CD51⁺ astrocytes have specialized functions that facilitate neural circuit activity. To further understand the mechanisms by which this population engenders circuit function, we focused on CD51 itself, which is encoded by the gene integrin alpha V²⁸. We used the adult hippocampus as a model system in which to assess how manipulation of CD51 affects circuit function. To delete Itgav in hippocampal astrocytes, we used similar approaches used in the GPC6 experiment (Fig. 2m), except in these studies we injected Rosa-LSL-Cas9 mice with AAVs to express (1) a gRNA targeting Itgav; and (2) cre under control of the Gfap promoter (Fig. 4a). Littermates not carrying the Rosa-LSL-Cas9 allele were used as controls. When put together, this system enabled us to selectively express Cas9 and the Itgav-targeting gRNA in hippocampal astrocytes. Accordingly, we injected the hippocampus of 8-week-old LSL-Cas9 mice (and wild-type littermates) and collected mouse tissue 1 month later, confirming selective deletion of Itgav in hippocampal astrocytes (Fig. 4b). Using a similar workflow to that described above, we generated mice in which Itgav has been knocked out in hippocampal astrocytes, collected mouse tissue 4 weeks after injection and assessed LTP and basal synaptic activity. Using theta-burst stimulation to evoke LTP, we found that deletion of astrocytic Itgav led to a loss of potentiation (Fig. 4c,d), coupled with a significant reduction in the basal levels of synaptic activity in CA1 neurons (Fig. 4e–g). Using a parallel cohort of mice generated using the same workflow, we conducted a series of memory behavioural assays, finding that mice with Itgav-KO exhibited significant defects in memory performance (Fig. 4h,i). These data suggest that CD51 contributes to hippocampal circuit function.

We complemented these KO studies with Itgav-overexpression experiments, using an AAV-based system to overexpress HA-tagged CD51 (or mCherry control) in hippocampal astrocytes from Aldh1l1-GFP mice (Fig. 4k). To test this system, we injected the hippocampus of Aldh1l1-GFP or wild-type mice at 8 weeks of age, collected mouse tissue 1 month later and confirmed robust overexpression of HA–CD51 specifically in astrocytes (Fig. 4l and Extended Data Fig. 12). We next repeated this experiment on wild-type mice, using LTP and basal synaptic properties as our frontline experiments on mouse tissue collected 1 month after AAV injection. For the LTP studies, we used subthreshold stimulation (40 Hz, 10 pulse) of the Schaffer collateral and found that overexpression of CD51 in hippocampal astrocytes elicited LTP (Fig. 4m,n). Analysis of synaptic properties under these conditions revealed an increase in synaptic activity as well, coupled with no changes in paired-pulse ratio, consistent with enhanced synaptic input and connectivity (Fig. 4o–q and Extended Data Fig. 11). Parallel studies on memory behaviours revealed a congruent increase in both spatial- and fear-memory behaviours after CD51 overexpression (Fig. 4q,r). Taken together, these studies indicate that astrocytic CD51 protein has a direct role in regulating hippocampal circuit activity and contributes to the specialized synaptic functions of CD51⁺ astrocytes.

To further understand how CD51 influences hippocampal circuit function, we overexpressed HA–CD51 in both hippocampal and cortical astrocytes, and performed immunoprecipitation–mass spectrometry (IP–MS) to identify prospective interacting partners specific to each region. Analysis of the protein interactome for CD51 revealed region-specific interactions (Extended Data Fig. 13 and Supplementary Table 3). Focusing on CD51-interacting proteins that are specific to the hippocampus, we found that CD51 interacted with NMDA receptors (GRIN1), GABA transporters (SLC32A1), Na⁺/K⁺ channels (HCN1) and SLC16A7 (also known as MCT2). Interactions with these proteins is consistent with a role for CD51 in modulating hippocampal circuit function.

Astrocytes have central roles in every form of neurological pathology²⁹. As CD51⁺ and CD51⁻ astrocytes exhibit unique functional properties during cortical development and in functioning neural circuits, we next examined whether these populations are endowed with distinct characteristics under pathological conditions. We focused these studies on cortical stroke injury because astrocytes are known to contribute to the repair process, of which synaptic remodelling is a key component^30,31. We first evaluated core features of CD51⁺ and CD51⁻ astrocyte populations in a cortical photothrombotic model of ischaemic stroke^32,33. Using our FLP-dependent AAVs in conjunction with the CD51-FLP mouse (Fig. 1d), we assessed the morphological properties of peri-infarct CD51⁺ and CD51⁻ astrocytes after stroke. These studies revealed that peri-infarct CD51⁺ astrocytes displayed an enhanced morphological complexity compared with CD51⁻ astrocytes at 7 and 21 days after stroke (Fig. 5a,b). Moreover, we observed an accumulation of CD51⁺ astrocytes at the lesion border, suggesting that CD51⁺ astrocytes preferentially form the astrocyte border (Extended Data Fig. 14a). This finding was further supported by spatial transcriptomics after stroke in which elevated expression of Itgav is observed at the lesion border³⁴ (Extended Data Fig. 14b). These observations, coupled with our findings that CD51⁺ and CD51⁻ astrocytes exhibit different proliferative capacities during postnatal cortical development (Fig. 1n,o) led us to assess their proliferative response after stroke, which is part of the astrocyte response to injury³². We induced strokes in CD51-FLP;Rosa-FSF-tdTomato mice, labelled proliferating cells with EdU from days 3–7 after stroke and assessed proliferative responses at 21 days after injury (Fig. 5c). We found that CD51⁺ astrocytes (SOX9⁺tdTomato⁺) were more proliferative (EdU⁺) at the infarct border compared with CD51⁻ astrocytes (SOX9⁺tdTomato⁻) (Fig. 5d,e). We observed similar results when assessing proliferation by staining for the cell cycle markers Ki-67 and phosphorylated histone H3 7 days after stroke (Extended Data Fig. 14c). Furthermore, we again observed an accumulation of CD51⁺ astrocytes at the lesion border (Fig. 5d). Together, these data indicate that CD51⁺ and CD51⁻ astrocytes exhibit unique cellular responses to cortical stroke, which parallel their distinct properties in the developing cortex.

Given the distinct properties of CD51⁺ and CD51⁻ astrocytes responding to stroke, we next examined whether these two populations differentially interacted with synapses in peri-infarct cortex. To accomplish this, we used the Förster resonance energy transfer (FRET)-based neuron-astrocyte proximity assay (NAPA), which enables measurement of astrocyte–synapse interactions at synaptic distance scales³⁵. NAPA involves expressing extracellular membrane-targeted GFP (FRET donor) in astrocytes and presynapse-targeted mCherry (FRET acceptor) in a defined neuronal afferent projection. In this assay, FRET signals are reflective of synaptic-scale proximity between GFP-expressing astrocytes and mCherry-expressing synapses. We injected the cortex, targeting regions that would later surround the infarct, of two groups of CD51-FLP mice with AAVs encoding the astrocytic component (NAPA-a) controlled by Flp-dependent elements to specify expression in either CD51⁺ or CD51⁻ astrocytes (Fig. 5f,g). In both groups, we labelled thalamocortical projections with the neuronal component (NAPA-n). We subjected mice to strokes 3 weeks after AAV injections and prepared tissue for FRET imaging in peri-infarct cortex 7 days later (Fig. 5f). Peri-infarct CD51⁺ astrocytes had significantly greater FRET signal than CD51⁻ astrocytes (Fig. 5g,h). These data suggest that CD51⁺ astrocytes exhibit closer interactions with synapses in remodelling peri-infarct regions after stroke.

The preceding studies indicate that DREADD-based stimulation of CD51⁺ astrocytes selectively promotes synapse formation in the developing cortex (Fig. 2), leading us to examine whether such a manipulation would similarly affect synaptic remodelling after stroke, which is linked with functional recovery³¹. Using the CD51-FLP mouse line and FLP-dependent AAVs previously described (Fig. 2e,f), we expressed hM3Dq (or lck-smV5 as an inert control) in CD51⁺ or CD51⁻ cortical astrocytes in regions that would later surround the infarct, and induced strokes 4 weeks later (Fig. 5i,j). Beginning 7 days after stroke, we treated mice with CNO or saline twice daily until 21 days after stroke (Fig. 5i,j). During this timeline, we assessed motor function using the grid walk assay at three key timepoints: (1) before stroke; (2) 1 week after stroke before commencing CNO or saline treatment; and (3) 3 weeks after stroke (after 2 weeks of CNO or saline treatment) (Fig. 5i). We found that stimulation of CD51⁺ astrocytes resulted in enhanced recovery of motor function in comparison to the controls and stimulation of CD51⁻ astrocytes (Fig. 5k). Importantly, there was no difference in the lesion volume between groups (Fig. 5l and Extended Data Fig. 14d). We next assessed synapse numbers in the peri-infarct cortex after stimulation, finding that CD51⁺ astrocyte stimulation increased numbers of excitatory and inhibitory synapses, while there were no observable differences in synapse formation after CD51⁻ astrocyte stimulation (Fig. 5m–o and Extended Data Fig. 14e–g). We observed no differences in per-infarct vessel density, neuron numbers, GFAP⁺ immunoreactivity or IBA1⁺ immunoreactivity (Extended Data Fig. 15). These observations suggest that the functional recovery observed after CD51 stimulation is the result of increased synapse remodelling leading to enhanced restoration of circuit function. Collectively, these data indicate that stimulation of CD51⁺ astrocytes selectively enhances functional recovery after stroke and that the unique synaptogenic properties of CD51⁺ astrocytes are conserved across postnatal development and repair after brain injury.

The nature of astrocyte diversity in the CNS has remained enigmatic owing to the paucity of tools that enable targeting of specific subpopulations, coupled with a lack of robust and clearly defined functional benchmarks^1,2. In our study, we generated a mouse line and associated viral tools that enable selective targeting and manipulation of CD51⁺ and CD51⁻ astrocyte subpopulations in the developing, mature and injured brain. We found that CD51⁺ astrocytes are functionally distinct from CD51⁻ astrocytes and have specialized functions that support synaptogenesis in the developing and injured cortex, along with memory circuit function in the hippocampus. Critically, comparing the functional properties between subpopulation of cells is essential for operationally defining distinct subpopulations and our newly developed tools enable such direct comparisons between CD51⁺ and CD51⁻ astrocytes. Inferenced heterogeneity from single-cell transcriptomic or developmental patterning data does not incorporate functional evaluation, as mRNA profiles cannot predict physiological outputs for astrocytes^1,2,3. Parallel studies identified genes that exhibit restricted expression in subsets of astrocytes and demonstrated roles for these genes in core astrocyte functions; however, they do not directly compare the functional properties between prospective astrocyte subpopulations^9,10. In contrast to neuronal diversity, which has clearly defined physiological, molecular and morphological criteria, defining astrocyte diversity is not as clear cut. Our findings indicate that astrocyte subpopulations exhibit graded functional properties, with CD51⁺ astrocytes being more efficient at promoting interactions with functioning neural circuits under normal and pathological conditions. These observations suggest that the specialized properties of CD51⁺ astrocytes are driven by interactions with neurons, insinuating that additional, functionally distinct astrocyte subpopulations may arise from interactions with vascular or immune populations.

Our finding that CD51⁺ astrocytes engender synapse formation in the postnatal cortex, in part through elevated expression of GPC6, and exhibit unique proliferative properties suggests that these functional properties are specified early in development. Indeed, we found CD51⁺ populations initially appear in the ventricular/subventricular zone around P1 and their genesis is tied to intrinsic transcription factor programs, further suggesting that they represent a specified astrocyte subtype with distinct functional properties. These findings are consistent with observations in the spinal cord indicating that developmental patterning mechanisms generate distinct subtypes of astrocytes³⁶. By contrast, clonal analysis of cortical astrocytes suggests that morphological diversity of astrocytes is linked to environmental cues, occurring after migration from the germinal zones³⁷. The extent to which the functional properties of CD51⁺ astrocytes are shaped by environmental cues or are pre-ordained is an important future direction, where clonal analysis of this population may provide additional insight. While neuronal activity did not affect the generation of CD51⁺ astrocytes in the cortex, it remains possible that their functional properties exhibit plasticity and are tied to specific neural circuits in other brain regions. In the mature brain, CD51⁺ astrocytes and Itgav itself robustly support hippocampal circuit function, while also executing parallel roles in repairing synapses and circuit function after cortical stroke injury. Roles for astrocytes in hippocampal circuits and memory are well known^38,39,40; however, our findings suggest that a select subset of astrocytes are sufficient to promote memory. These results parallel recent findings demonstrating that ensembles of learning-associated astrocytes function within memory engrams and are sufficient to promote memory recall¹³. Collectively, these observations raise the possibility that CD51⁺ astrocytes are preferentially recruited to memory engrams in the hippocampus and, more broadly, that CD51⁺ astrocytes may specifically participate in engrams associated with other forms of learning⁴¹.

Animals

Mice were used in compliance with protocols approved by the Institutional Animal Care and Use Committee at Baylor College of Medicine. Mice were housed with food and water available ad libitum under a 12 h–12 h light–dark cycle in a 20–22 °C and 40–60% humidity environment. The following transgenic strains were used: Rosa-CAG-FSF-tdTomato (Jackson Laboratory, 032864), Aldh1l1-GFP (RRID: MMRRC_011015-UCD), Rosa-CAG-LSL-Cas9-eGFP (Jackson Laboratory, 024857), Nfia-flox⁴², Aldh1l1-creER (Jackson Laboratory, 029655), Rosa-CAG-LSL-tdTomato (Jackson Laboratory, 007914) and CD51-FLP. The CD51-FLP mouse was generated using BAC transgenesis, whereby the FLP cassette was inserted into the Itgav locus immediately after the ATG start codon in exon one. The reference transcript used for the BAC modification is NM_008402.3. The BAC was introduced through pronuclear injection, in collaboration with Cyagen. All mice were on a C57BL/6 background. Transgenic strains were crossed as described in the main text. We used 2–6-month-old mice of both sexes and randomly allocated littermates to groups where possible. All experiments and analyses were conducted blinded to group allocation. For stimulation of G_q-coupled signalling through hM3Dq, mice were injected intraperitoneally with CNO (Abcam, ab141704) dissolved in saline (0.5 mg ml⁻¹ for pups and 1 mg ml⁻¹ for adults). For multiday dosing, injections were done twice daily, spaced by 8 h. For acute injections, CNO was administered 30 min before the start of behavioural testing. Tamoxifen was dissolved in corn oil and injected subcutaneously at 100 mg per kg.

Plasmid and virus production

The following constructs were previously reported¹³: GFAP-fDIO-tdTomato-4x6T, GFAP-FSF-GCaMP6, GFAP-FSF-hM3D-mCherry, GFAP-cre-mCherry and GFAP-fDIO-cre. Several constructs were obtained from Addgene for cloning or AAV preparation: GFAP-hM3D-mCherry (a gift from B. Roth; Addgene, 50478).

GFAP-lck-smV5-4x6T (a gift from S. T. Carmichael; Addgene, 196416), gfaABC1D-cyto-GCaMP6f (a gift from B. Khakh; Addgene, 52925), pZac2.1-GfaABC1D-NAPA-A-SV40 (a gift from B. Khakh; Addgene, 92281) and pZac2.1-hSynapsin1-NAPA-N-SV40 (a gift from B. Khakh; Addgene, 92282). The following constructs were newly constructed using restriction and insertion cloning or Gibson assembly: GFAP-FRT-tdTomato-FRT, GFAP-FRT-GCaMP6-FRT, GFAP-FRT-hM3D-mCherry-FRT, GFAP-Itgav-HA-4x6T, GFAP-Gpc6, GFAP-fDIO-NAPA-a, GFAP-FRT-NAPA-a-FRT. The Itgav ORF was obtained from SinoBiological (MG50156-CY). The Gpc6 ORF was obtained from Origene (MR208930). 4x6T microRNA targeting sequences were included in several constructs to improve astrocyte specificity¹². All new constructs were verified by whole-plasmid sequencing. Plasmids driving gRNA expression under control of the U6 promoter were designed using CHOPCHOP (https://chopchop.cbu.uib.no/) to identify target sequences for SaCas9 within exons of target genes (Gpc6 and Itgav) and avoid potential off-target effects. AAVs were packaged by the Optogenetics and Viral Vectors Core at the Jan and Dan Duncan Neurological Research Institute. AAV2/9 was produced at titres ranging from 5 × 10¹¹ to 7 × 10¹² genome copies per ml.

AAV injection surgery

For adult injections, mice were anaesthetized with isoflurane (3% induction, 2% maintenance) in oxygen and placed in a stereotaxic frame. The scalp was shaved and cleaned with alternating applications of 70% ethanol and betadine. After a midline scalp incision, burr holes were drilled in the skull and 250–500 µl of AAV was injected in each target site using a pulled pipette attached to a Drummond Nanoject III. The pipette was left in place for 2–5 min after each injection before being slowly removed. Injections were targeted bilaterally to the hippocampus (1.7 mm posterior, 1.7 mm lateral and 1.8 mm ventral relative to Bregma) or multiple sites in the cortex (2.0 mm anterior, 2.0 mm lateral and 0.6 mm below the pial surface; 0.5 mm anterior, 3.2 mm lateral and 0.6 mm below the pial surface; 1.0 mm posterior, 2.5 mm lateral and 0.6 mm below the pial surface; 0.5 mm anterior, 0.5 mm lateral and 0.6 mm below the pial surface; 1.5 mm posterior, 0.6 mm lateral and 0.6 mm below the pial surface). For NAPA experiments, AAV-hSyn-NAPA-n was injected into the thalamus to target thalamocortical projections (1.7 mm posterior, 1.6 mm lateral, 3.2 mm below the pial surface). Combinations of AAVs for each experiment are listed in the main text and figures. After injections, the scalp was sutured closed. Then, 1 mg per kg of buprenorphine slow release was given subcutaneously 30–60 min before surgery. For postnatal injections, P0–2 pups were briefly anaesthetized by hypothermia and manually injected at 1–3 cortical sites with 0.5–1 μl of virus using a Hamilton syringe with a 32-gauge needle.

Slice recording

Mice were deeply anaesthetized with isoflurane. After decapitation, the brain was rapidly removed from the skull and submerged in ice-cold cutting solution containing 130 mM NaCl, 24 mM NaHCO₃, 1.25 mM NaH₂PO₄, 3.5 mM KCl, 1.5 mM CaCl₂, 1.5 mM MgCl₂ and 10 mM D(+)-glucose, pH 7.4. The solution was continuously bubbled with 95% O₂/5% CO₂. After trimming the brain, 300 μm coronal slices containing the cortex or hippocampus were prepared using a vibratome and transferred to extracellular ACSF containing 130 mM NaCl, 24 mM NaHCO₃, 1.25 mM NaH₂PO₄, 3.5 mM KCl, 1.5 mM CaCl₂, 1.5 mM MgCl₂ and 10 mM D(+)-glucose, pH 7.4. The slices were incubated at room temperature for at least 1 h before being transferred to a recording chamber continuously perfused with ACSF at a flow rate of 2 ml min⁻¹. The recording chamber was mounted onto the stage of an upright Olympus microscope and visualized using a 60× water-immersion objective (NA = 0.90) with infrared differential interference contrast optics.

LTP was induced by theta-burst stimulation (10 trains of 4 stimuli at 100 Hz with a 200 ms interval) delivered to the Schaffer collateral pathway. Subthreshold stimulation consisted of 10 pulses delivered at 40 Hz with a pulse width of 0.1 ms. Stimulus intensity was adjusted in each slice to evoke an fEPSP slope corresponding to approximately 40–50% of the maximal response. fEPSPs were recorded in the CA1 stratum radiatum using a recording pipette filled with 1 M NaCl solution. The fEPSP slope (mV ms⁻¹) was measured and normalized to the average baseline response. The magnitude of potentiation was calculated by averaging the responses during the final 5 min of recording. Whole-cell recordings were performed using a Multiclamp 700B amplifier (Molecular Devices) controlled by pClamp v.10.7 software from cortical pyramidal neurons and hippocampal CA1 neurons. The holding potential was −60 mV, and pipette resistance was typically 5–8 MΩ. For EPSC recordings, pipettes were filled with an internal solution containing 135 mM CsMeSO₄, 8 mM NaCl, 10 mM HEPES, 0.25 mM EGTA, 1 mM Mg-ATP, 0.25 mM Na₂-GTP and 30 mM QX-314, with pH adjusted to 7.2 using CsOH (278–285 mOsm). For IPSC recordings, pipettes were filled with an internal solution containing 135 mM CsCl, 4 mM NaCl, 0.5 mM CaCl₂, 10 mM HEPES, 5 mM EGTA, 2 mM Mg-ATP, 0.5 mM Na₂-GTP and 30 mM QX-314, with pH adjusted to 7.2 using CsOH (278–285 mOsm). sEPSCs were recorded in the presence of the GABA_A receptor antagonist bicuculline (20 μM, Tocris). sIPSCs were recorded in the presence of ionotropic glutamate receptor antagonists APV (50 μM, Tocris) and CNQX (20 μM, Tocris). All electrophysiological experiments were performed at room temperature. Electrical signals were acquired using a Digidata 1440A data acquisition system and a Multiclamp 700B amplifier (Molecular Devices) controlled by pClamp v.10.7 software. Signals were digitized at 10 kHz (100 μs sampling interval) and low-pass filtered at 2 kHz. Recorded currents were analysed using Clampfit v.10.7 software.

Photothrombotic stroke

Mice were anaesthetized with isoflurane (3% induction, 2% maintenance) in oxygen and placed into a stereotaxic frame. Body temperature was maintained with a heated pad for the duration of anaesthesia. The scalp was shaved and cleaned with alternating applications of 70% ethanol and betadine. Focal infarcts were induced through the intact skull by making a scalp incision, administering rose bengal (0.15 ml, 15 mg ml⁻¹, intraperitoneal) and illuminating the skull 2 mm right from bregma with a Schott KL 200 light source for 15 min though a 3 mm aperture. This location targeted the forelimb area of motor cortex and surrounding sensorimotor areas³².

Behavioural assays

Grid walk

The grid walk assay was used to assess motor deficits in the left forelimb after stroke damage in the right forelimb area of the motor cortex and the surrounding cortical areas³². Mice were placed onto a wire grid with 1 cm openings and allowed to walk spontaneously while recorded from below for 3–6 min, allowing recording of at least 100 steps. Incorrectly placed steps of the impaired forelimb were counted relative to the total number of steps counted. Incorrect steps were defined as steps during which the paw missed the wire grid and went through a hole or when the wrist or forearm, rather than the paw, were used for support. Mice were acclimatized to the task with a 5 min training session in which they were placed onto the grid and allowed to walk spontaneously. Tests were conducted before stroke and 7 and 21 days after stroke.

Contextual fear conditioning

The fear-conditioning chamber was a 25 cm × 35 cm arena with a metal grid floor. The arenas had different patterns on some walls as visual cues. Context A was cleaned with 35% isopropyl alcohol, and context B was cleaned with 70% ethanol to give the contexts different scents. A fan provided approximately 65 dB of ambient noise for both chambers. Mice were recorded from above during all of the behavioural sessions. Mice were allowed to explore the chamber for the first 120 s. Then, 1.5 mA foot shocks lasting 2 s were delivered at 120 s, 180 s and 240 s after mice were initially placed in the chamber. Mice were removed from the chamber 60 s after the final foot shock and returned to their home cage. Videos were manually scored for freezing behaviour by a blinded investigator. Freezing was defined as the absence of movement, except for breathing-related movement, lasting more than 1 s.

Novel-place recognition

Mice were placed into a square arena (40 cm × 40 cm) with two identical objects (cylinders, 12 cm tall) in adjacent quadrants. The mice were trained on three consecutive days for 10 min each day to learn the position of the objects. The arena had distinct shapes on certain walls as orientation cues. On the testing day, the location of one object was moved to a different quadrant. Mice were placed into the arena for 5 min and recorded from above. The discrimination index, which is a measure of spatial memory performance, was calculated as: (time spent investigating an object in a new place − time spent investigating an object in its original place)/total time exploring either object.

Calcium imaging and analysis

Acute slices were prepared as described above. Slices were maintained in oxygenated ACSF with a perfusion system. Calcium traces were recorded using a two-photon microscope (LSM 7MP, Zeiss) equipped with a Coherent Chameleon Ultra (II) Ti-sapphire laser tuned to 900 nm and a ×20, 1.0 NA water-immersion Zeiss objective lens. GCaMP signals were recorded for 3–5 min per trial at 1,024 × 1,024 pixel resolution and 1 Hz temporal resolution from astrocytes at depths of approximately 30 µm below the surface. All multiphoton imaging experiments were done within 4 h of slicing. Images were quantified using GECIquant¹⁴, ImageJ (v.54f) and Clampfit (v.10.7) software. The region-of-interest detection for soma and main processes or microdomains was done in a semi-automated manner using the GECIquant algorithm as previously described³⁵. The amplitude, area under the curve and frequency values of GCaMP signals were computed using Clampfit. Spike width was calculated as the full width at half-maximum for each spike.

Histology

Mice were perfused with PBS followed by 4% paraformaldehyde in PBS. Brains were collected and post-fixed overnight in 4% paraformaldehyde at 4 °C and then sectioned at a thickness of 35 µm using a vibratome (Leica VT1200s) or cryoprotected in 20% sucrose before embedding and sectioning at a thickness of 35 µm using a cryostat. For immunostaining, the sections were washed three times in PBS and incubated for 1 h in PBS with 0.25% Triton X-100 and 10% donkey serum (blocking solution), then incubated overnight at room temperature with primary antibodies. The following primary antibodies were used: rabbit anti-BRN2 (1:1,000, Cell Signaling, 121375), rabbit anti-CD51 (1:500, Invitrogen, SC56-07), rabbit anti-CD51 (1:500, Sigma-Aldrich, AB1930), rat anti-FOS (1:5,000, Synaptic Systems, 226 017), rabbit anti-Ctip2 (1:500, Abcam, ab18465), rabbit anti-Foxp2 (1:500, Abcam, ab16046), mouse anti-gephyrin (1:2,000, Synaptic Systems, 147 011), chicken anti-GFAP (1:1,000, Abcam, ab4674), rabbit anti-GFAP (1:1,000, DAKO, Z0334), mouse anti-GPC6 (1:100, RnD Systems, AF1053), rabbit anti-HA (1:500, Roche, 11867423001), goat anti-IBA1 (1:500, Abcam, ab5076), rabbit anti-Ki-67 (1:500, Abcam ab66155), rabbit anti-OLIG2 (1:500, Chemicon, ab9610), rabbit anti-NeuN (1:2,000, Millipore, MABN140), rabbit anti-NFIA (1:500, Sigma-Aldrich, HPA006111), rabbit anti-phospho-histone H3 (1:300, Cell Signaling, 3377), rabbit anti-RFP (1:500, Rockland, 600-401-379), rabbit anti-PSD95 (1:500, Invitrogen, 51-6900), rabbit anti-SOX9 (1:500, Millipore, AB5535), human anti-V5 (1:2,000, Absolute Antibodies, AB00136-10.0), guinea pig anti-VGAT (1:1,000, Synaptic Systems, 131 004), guinea pig anti-VGlut1 (1:5,000, Millipore, AB5905), guinea pig anti VGlut2 (1:2,000, Synaptic Systems, 131 004). DyLight 649 tomato lectin was used to label vasculature (1:50, Vector Laboratories, DL-1178-1). The sections were then washed three times in PBS followed by incubation with secondary antibodies (1:500) in blocking solution for 1 h. The following secondary antibodies were used: Alexa-Fluor-647-conjugated donkey anti-chicken (Jackson ImmunoResearch, 703-605-155), Alexa-Fluor-594-conjugated donkey anti-guinea pig (Jackson ImmunoResearch, 706-585-148), Alexa-Fluor-647-conjugated donkey anti-guinea pig (Jackson ImmunoResearch, 706-605-148), Alexa-Fluor-594-conjugated donkey anti-human (Jackson ImmunoResearch, 709-585-149), Alexa-Fluor-488-conjugated donkey anti-mouse (Jackson ImmunoResearch, 715-545-151), Alexa-Fluor-488-conjugated donkey anti-rabbit (Jackson ImmunoResearch, 711-545-152), Alexa-Fluor-594-conjugated donkey anti-rabbit (Jackson ImmunoResearch, 711-585-152), Alexa-Fluor-647-conjugated donkey anti-rabbit (Jackson ImmunoResearch, 711-605-152). The sections were washed three additional times in PBS before mounting and coverslipping. EdU was detected using the EdU Click-iT kit according to the manufacturer instructions (Thermo Fisher Scientific, C10640).

Imaging and analysis

Images were acquired using the Zeiss LSM 880 laser-scanning confocal microscope with a ×20 air, ×40 oil or ×63 oil objective, or the Keyence BZ-X800 epifluorescence microscope. Confocal images were taken with 1–2 µm steps in the z plane and 512 × 512, 1,024 × 1,024 or 2,048 × 2,048 pixel resolution. Images were quantified using ImageJ/FIJI (v.1.54f) in a blinded manner. Cell numbers were quantified manually using the multi-point selection tool. Fluorescence intensity was quantified using the mean pixel value measured within regions of interest. All analyses were done on regions devoid of injection-related damage or inflammation.

For calculating lesion volume after stroke, every tenth section encompassing the entirety of the lesion was stained with NeuroTrace deep-red fluorescent Nissl stain according to the manufacturer’s instructions (Thermo Fisher Scientific, N21483). Low-magnification images were acquired using a ×2 objective. As previously described, cortex was manually traced and area measured in ImageJ, and cortical volume in each hemisphere was calculated using Cavalieri’s method as the sum of area of cortex × distance between section planes. Lesion volume was determined as the volume of intact cortex – volume of injured cortex.

For FRET analysis in NAPA experiments⁴³, confocal images were collected using excitation lines 488 and 561 nm. Three channels of emission were collected: green with 488 nm excitation (donor), red with 561 nm excitation (acceptor) and red with 488 nm excitation (raw FRET). All images were collected using two-frame averaging. Detector gains, amplifier offsets and laser power were adjusted to maximize the range of signal for each channel and were kept consistent across all samples, including single colour controls (NAPA-a or NAPA-n only). Samples were imaged and analysed in 3D to account for spatial distribution of FRET signals. FRET analysis was conducted using the PixFRET ImageJ plugin following a previously described protocol⁴⁴. The number of FRET puncta was normalized to astrocyte volume and the density of NAPA-n signal to account for natural variation in astrocyte size and thalamocortical projection labelling.

FACS

Mice were perfused with ice-cold saline and cortex or hippocampi were dissected on ice. Regions from three mice were pooled per replicate and dissociated as previously^12,13. Dissociated cells were sorted based on GFP and tdTomato expression using a Becton Dickinson FACSAria II with BD FACSDiva software. The samples were sorted using a 70 μm nozzle with 70 psi of pressure. After singlet gating with forward and side scatter, singlet events were assessed for GFP and tdTomato fluorescence. Non-fluorescent brain tissue was used as a negative control to set the double negative population gate. Single colour controls were used to establish ideal voltage, potential compensation and gating. The GPF signal was detected with a 488 nm laser, through a 505 nm long-pass filter and 530/30 detector. tdTomato signal was detected with 561 nm laser, through a 570 long-pass filter and 585/51 detector. In total, 100,000 events of each type were collected into 1.5 ml tubes containing Qiagen lysis buffer. The samples were then frozen at −80 °C until RNA extraction.

Library preparation and sequencing

RNA was extracted from sorted cells using RNeasyMicro Kit (4004, Qiagen). cDNA synthesis and sequencing libraries with 8-bp single indices were constructed from 10 ng RNA using the Revelo RNA-seq High Sensitivity kit (30184149, Tecan). The resulting libraries were validated using the Standard Sensitivity NGS Fragment Analysis Kit (DNF-473-0500) on a 12-Capillary Fragment Analyzer and quantified using the Quant-it dsDNA assay kit (Q33120). Equal concentrations (2 nM) of libraries were pooled and processed for paired-end sequencing (150 reads each direction), targeting at least 40 million reads per sample on the Illumina NovaSeq X Plus according to the manufacturer’s instructions.

RNA-seq analysis

Raw RNA-seq reads from hippocampal and cortical CD51⁺ and CD51⁻ astrocyte samples were aligned to the mouse reference genome (mm10) using STAR⁸. Raw read count matrices were generated from the aligned reads. Normalization and differential expression analyses were conducted using DESeq2 (v.1.40.2)⁴⁵. Variance-stabilized transformation was applied to normalized counts for visualization and downstream analyses. Principal component analysis (PCA) was used to assess sample-level clustering by CD51 status and brain region. Differential expression analyses for CD51⁺ or CD51⁻ astrocyte samples were performed separately for the cortex and hippocampus. Genes with an adjusted P < 0.05 and an absolute log₂-transformed fold change of greater than 1 were considered significantly differentially expressed. Volcano plots were generated with ggplot2 to visualize differential expression results. Gene set enrichment analyses were performed using fgsea (v.1.26.0)⁴⁶ and GSVA (v.1.50.0) packages. Pre-ranked fgsea analyses were based on DESeq2 test statistics to evaluate enrichment of astrocytic and neuronal gene sets derived from curated gene lists (Supplementary Table 1). Single-sample gene set enrichment analysis (ssGSEA) was performed using GSVA to compute enrichment scores for astrocyte- and neuron-enriched gene sets, as well as for previously defined astrocyte populations (A–E)⁴. The resulting ssGSEA scores were visualized using pheatmap and ggplot2 with hierarchical clustering. Functional enrichment of differentially expressed genes (DEGs) was assessed with Enrichr using the GO Biological Process database^47,48. Terms with adjusted P < 0.05 were considered significantly enriched. Enrichment results were summarized and visualized as dot plots using –log₁₀[p] values. Quality control was conducted using fastQC (v.0.10.1) and MultiQC (v.0.9). Reads were mapped to the mouse genome mm10 assembly using STAR (v.2.5.0a). RNA-seq data were analysed and plotted as previously described. DESeq2 (v.1.20.0) was used for both differential gene expression analysis (Wald test) and read count normalization. We defined DEGs as those with expression fold change > 1.5 at P < 0.01. GO terms associated with DEGs were determined using Enrichr. Motif analysis was performed using Hypergeometric Optimization of Motif Enrichment (HOMER, v.4.10) to identify transcription factor motifs enriched near the transcription start site of DEGs. scRNA-seq analysis of hippocampus from mice aged 6 months was previously described⁴⁸ and data were accessed from NCBI GEO GSE224398. RNA-seq data of developing astrocytes were accessed from NCBI GEO GSE158329. Spatial transcriptomics data were previously reported, and data were accessed from http://www.jianglab.cn/StrokeMap.

IP–MS

CD51-interacting proteins were identified by IP–MS. Hippocampus and cortex overexpressing HA–CD51 was collected, cells were dissociated and lysed by probe sonication in lysis buffer containing 50 mM Tris-HCl (pH 8.0), 150 mM NaCl, 0.5% NP-40, 5 mM EDTA and protease inhibitor cocktail (GeneDepot). The lysates were clarified by centrifugation at 20,000g and precleared with protein G Dynabeads. Precleared lysates were incubated with 2 μg anti-HA antibody (Cell Signaling Technology), followed by capture with protein G Dynabeads (Thermo Fisher Scientific). Anti-HA immunoprecipitates were washed three times with lysis buffer, three times with PBS and once with 50 mM ammonium bicarbonate. Bound proteins were digested on-bead with 0.5 μg trypsin/Lys-C for 16 h at 37 °C. Peptides were desalted using C18 tips, vacuum-dried and reconstituted in 0.1% formic acid, and 20% of each sample was subjected to analysis using liquid chromatography coupled with tandem MS using the Evosep One system coupled to the timsTOF Ultra 2 mass spectrometer (Bruker Daltonics). Peptides were separated using the Whisper Zoom 20 method, and spectra were acquired in ddaPASEF mode. Raw data were processed with FragPipe v.22.0 and searched against the UniProt Mus musculus reference proteome database (UP000000589) using the label-free quantification workflow.

Statistics and reproducibility

Sample sizes and statistical tests are noted in the figure legends. Significant statistical tests were defined as P < 0.05. GraphPad Prism (v.10) was used for all statistical analyses other than those used for RNA-seq analysis. Independent samples were compared with two-tailed t-tests. Variance was assessed with F tests, and Welch’s corrected t-tests were used when variance was significantly different. One-way and two-way ANOVA and nested t-tests were used as noted in the legends. Post hoc tests were used following significant ANOVA, as noted in the legends. We found no sex-dependent effects. All experiments were repeated independently at least three times with similar results.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

1. Khakh, B. S. & Deneen, B. The emerging nature of astrocyte diversity. Annu. Rev. Neurosci. 42, 187–207 https://doi.org/10.1146/annurev-neuro-070918-050443 (2019).

   Article  CAS  PubMed  Google Scholar 

2. Kwon, W., Williamson, M. R. & Deneen, B. A functional perspective on astrocyte heterogeneity. Trends Neurosci. 48, 691–705 https://doi.org/10.1016/j.tins.2025.06.009 (2025).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

3. O’Dea, M. R. & Hasel, P. Are we there yet? Exploring astrocyte heterogeneity one cell at a time. Glia 73, 619–631 https://doi.org/10.1002/glia.24621 (2025).

   Article  CAS  PubMed  Google Scholar 

4. John Lin, C. C. et al. Identification of diverse astrocyte populations and their malignant analogs. Nat. Neurosci. 20, 396–405 https://doi.org/10.1038/nn.4493 (2017).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

5. Burda, J. E., Bernstein, A. M. & Sofroniew, M. V. Astrocyte roles in traumatic brain injury. Exp. Neurol. 275, 305–315 https://doi.org/10.1016/j.expneurol.2015.03.020 (2016).

   Article  CAS  PubMed  Google Scholar 

6. Khakh, B. S. & Sofroniew, M. V. Diversity of astrocyte functions and phenotypes in neural circuits. Nat. Neurosci. 18, 942–952 https://doi.org/10.1038/nn.4043 (2015).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

7. Lacoste, B., Prat, A., Freitas-Andrade, M. & Gu, C. The blood–brain barrier: composition, properties, and roles in brain health. Cold Spring Harb. Perspect. Biol. 17, a041422 https://doi.org/10.1101/cshperspect.a041422 (2025).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

8. Endo, F. et al. Molecular basis of astrocyte diversity and morphology across the CNS in health and disease. Science 378, eadc9020 https://doi.org/10.1126/science.adc9020 (2022).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

9. Ollivier, M. et al. Crym-positive striatal astrocytes gate perseverative behaviour. Nature 627, 358–366 https://doi.org/10.1038/s41586-024-07138-0 (2024).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

10. de Ceglia, R. et al. Specialized astrocytes mediate glutamatergic gliotransmission in the CNS. Nature 622, 120–129 https://doi.org/10.1038/s41586-023-06502-w (2023).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

11. Park, H. et al. Single-cell RNA-sequencing identifies disease-associated oligodendrocytes in male APP NL-G-F and 5XFAD mice. Nat. Commun. 14, 802 https://doi.org/10.1038/s41467-023-36519-8 (2023).

12. Gleichman, A. J., Kawaguchi, R., Sofroniew, M. V. & Carmichael, S. T. A toolbox of astrocyte-specific, serotype-independent adeno-associated viral vectors using microRNA targeting sequences. Nat. Commun. 14, 7426 https://doi.org/10.1038/s41467-023-42746-w (2023).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

13. Williamson, M. R. et al. Learning-associated astrocyte ensembles regulate memory recall. Nature 637, 478–486 https://doi.org/10.1038/s41586-024-08170-w (2025).

    Article  ADS  CAS  PubMed  Google Scholar 

14. Srinivasan, R. et al. Ca²⁺ signaling in astrocytes from Ip3r2^−/− mice in brain slices and during startle responses in vivo. Nat. Neurosci. 18, 708–717 https://doi.org/10.1038/nn.4001 (2015).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

15. Ge, W. P., Miyawaki, A., Gage, F. H., Jan, Y. N. & Jan, L. Y. Local generation of glia is a major astrocyte source in postnatal cortex. Nature 484, 376–380 https://doi.org/10.1038/nature10959 (2012).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

16. Akdemir, E. S., Huang, A. Y. & Deneen, B. Astrocytogenesis: where, when, and how. F1000Res 9, 233 https://doi.org/10.12688/f1000research.22405.1 (2020).

17. Freitas-Andrade, M. et al. Astroglial Hmgb1 regulates postnatal astrocyte morphogenesis and cerebrovascular maturation. Nat. Commun. 14, 4965 https://doi.org/10.1038/s41467-023-40682-3 (2023).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

18. Rurak, G. M. et al. Sex differences in developmental patterns of neocortical astroglia: a mouse translatome database. Cell Rep. 38, 110310 https://doi.org/10.1016/j.celrep.2022.110310 (2022).

    Article  CAS  PubMed  Google Scholar 

19. Park, S. et al. NR3C1-mediated epigenetic regulation suppresses astrocytic immune responses in mice. Nat. Commun. 16, 8330 https://doi.org/10.1038/s41467-025-64088-5 (2025).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

20. Williamson, M. R., Murali, S. & Deneen, B. Development and diversity of astrocytes. Development 152, dev204705 https://doi.org/10.1242/dev.204705 (2025).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

21. Cheng, Y. T. et al. Inhibitory input directs astrocyte morphogenesis through glial GABA_BR. Nature 617, 369–376 https://doi.org/10.1038/s41586-023-06010-x (2023).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

22. Chung, W. S., Allen, N. J. & Eroglu, C. Astrocytes control synapse formation, function, and elimination. Cold Spring Harb. Perspect. Biol. 7, a020370 https://doi.org/10.1101/cshperspect.a020370 (2015).

    Article  PubMed  PubMed Central  Google Scholar 

23. Allen, N. J. & Eroglu, C. Cell biology of astrocyte-synapse interactions. Neuron 96, 697–708 https://doi.org/10.1016/j.neuron.2017.09.056 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

24. Adamsky, A. et al. Astrocytic activation generates de novo neuronal potentiation and memory enhancement. Cell 174, 59–71 https://doi.org/10.1016/j.cell.2018.05.002 (2018).

    Article  CAS  PubMed  Google Scholar 

25. Nagai, J. et al. Hyperactivity with disrupted attention by activation of an astrocyte synaptogenic cue. Cell 177, 1280–1292 https://doi.org/10.1016/j.cell.2019.03.019 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

26. Allen, N. J. et al. Astrocyte glypicans 4 and 6 promote formation of excitatory synapses via GluA1 AMPA receptors. Nature 486, 410–414 https://doi.org/10.1038/nature11059 (2012).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

27. Bosworth, A. P. et al. Astrocyte glypican 5 regulates synapse maturation and stabilization. Cell Rep. 44, 115374 https://doi.org/10.1016/j.celrep.2025.115374 (2025).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

28. Sims, M. A. et al. Cloning and characterisation of ITGAV, the genomic sequence for human cell adhesion protein (vitronectin) receptor alpha subunit, CD51. Cytogenet. Cell Genet. 89, 268–271 https://doi.org/10.1159/000015631 (2000).

    Article  CAS  PubMed  Google Scholar 

29. Escartin, C. et al. Reactive astrocyte nomenclature, definitions, and future directions. Nat. Neurosci. 24, 312–325 https://doi.org/10.1038/s41593-020-00783-4 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

30. Williamson, M. R. et al. Subventricular zone cytogenesis provides trophic support for neural repair in a mouse model of stroke. Nat. Commun. 14, 6341 https://doi.org/10.1038/s41467-023-42138-0 (2023).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

31. Kim, S. Y., Hsu, J. E., Husbands, L. C., Kleim, J. A. & Jones, T. A. Coordinated plasticity of synapses and astrocytes underlies practice-driven functional vicariation in peri-infarct motor cortex. J. Neurosci. 38, 93–107 https://doi.org/10.1523/JNEUROSCI.1295-17.2017 (2018).

    Article  CAS  PubMed  Google Scholar 

32. Williamson, M. R., Fuertes, C. J. A., Dunn, A. K., Drew, M. R. & Jones, T. A. Reactive astrocytes facilitate vascular repair and remodeling after stroke. Cell Rep. 35, 109048 https://doi.org/10.1016/j.celrep.2021.109048 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

33. Laug, D. et al. Nuclear factor I-A regulates diverse reactive astrocyte responses after CNS injury. J. Clin. Invest. 129, 4408–4418 https://doi.org/10.1172/JCI127492 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

34. Han, B. et al. Integrating spatial and single-cell transcriptomics to characterize the molecular and cellular architecture of the ischemic mouse brain. Sci. Transl. Med. 16, eadg1323 https://doi.org/10.1126/scitranslmed.adg1323 (2024).

    Article  CAS  PubMed  Google Scholar 

35. Octeau, J. C. et al. An optical neuron-astrocyte proximity assay at synaptic distance scales. Neuron 98, 49–66 https://doi.org/10.1016/j.neuron.2018.03.003 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

36. Bosquez Huerta, N. A. et al. Sex-specific astrocyte regulation of spinal motor circuits by Nkx6.1. Cell Rep 44, 115121 https://doi.org/10.1016/j.celrep.2024.115121 (2025).

    Article  CAS  PubMed  Google Scholar 

37. Clavreul et al. Cortical astrocytes develop in a plastic manner at both clonal and cellular levels. Nat. Commun. 10, 4884 https://doi.org/10.1038/s41467-019-12791-5 (2019).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

38. Haydon, P. G. & Carmignoto, G. Astrocyte control of synaptic transmission and neurovascular coupling. Physiol. Rev. 86, 1009–1031 https://doi.org/10.1152/physrev.00049.2005 (2006).

    Article  CAS  PubMed  Google Scholar 

39. Haydon, P. G. & Nedergaard, M. How do astrocytes participate in neural plasticity? Cold Spring Harb. Perspect. Biol. 7, a020438 https://doi.org/10.1101/cshperspect.a020438 (2014).

    Article  PubMed  PubMed Central  Google Scholar 

40. Kofuji, P. & Araque, A. Astrocytes and behavior. Annu. Rev. Neurosci. 44, 49–67 https://doi.org/10.1146/annurev-neuro-101920-112225 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

41. Serra, I. et al. Astrocyte ensembles manipulated with AstroLight tune cue-motivated behavior. Nat. Neurosci. 28, 616–626 https://doi.org/10.1038/s41593-025-01870-0 (2025).

    Article  CAS  PubMed  Google Scholar 

42. Scavuzzo, M. A. et al. Pancreatic cell fate determination relies on notch ligand trafficking by NFIA. Cell Rep. 25, 3811–3827 https://doi.org/10.1016/j.celrep.2018.11.078 (2018).

    Article  CAS  PubMed  Google Scholar 

43. Badia-Soteras, A., Octeau, J. C., Verheijen, M. H. G. & Khakh, B. S. Assessing neuron-astrocyte spatial interactions using the neuron-astrocyte proximity assay. Curr. Protoc. Neurosci. 91, e91 https://doi.org/10.1002/cpns.91 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

44. Dobin, A. et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29, 15–21 https://doi.org/10.1093/bioinformatics/bts635 (2013).

    Article  CAS  PubMed  Google Scholar 

45. Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550 https://doi.org/10.1186/s13059-014-0550-8 (2014).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

46. Korotkevich, G. et al. Fast gene set enrichment analysis. Preprint at bioRxiv https://doi.org/10.1101/060012 (2021).

47. Chen, E. Y. et al. Enrichr: interactive and collaborative HTML5 gene list enrichment analysis tool. BMC Bioinform. 14, 128 https://doi.org/10.1186/1471-2105-14-128 (2013).

    Article  Google Scholar 

48. Kuleshov, M. V. et al. Enrichr: a comprehensive gene set enrichment analysis web server 2016 update. Nucleic Acids Res. 44, W90–W97 https://doi.org/10.1093/nar/gkw377 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

Download references

Author notes

1. Michael R. Williamson

   Present address: Program in Neurosciences and Mental Health, The Hospital for Sick Children and Department of Physiology, University of Toronto, Toronto, Ontario, Canada

2. These authors contributed equally: Junsung Woo, Michael R. Williamson

Authors and Affiliations

1. Center for Cell and Gene Therapy, Baylor College of Medicine, Houston, TX, USA

   Junsung Woo, Michael R. Williamson, Dongjoo Choi, Ekin Akdemir, Thea Zlatkov, Kwanha Yu, Wookbong Kwon, Ehson Maleki, Yeunjung Ko, Kaitlyn Sanchez, Hussein Maleki & Benjamin Deneen

2. Center for Cancer Neuroscience, Baylor College of Medicine, Houston, TX, USA

   Junsung Woo, Michael R. Williamson, Dongjoo Choi, Ekin Akdemir, Thea Zlatkov, Kwanha Yu, Wookbong Kwon, Ehson Maleki, Yeunjung Ko, Kaitlyn Sanchez, Hussein Maleki, Akdes Serin-Harmanci & Benjamin Deneen

3. Department of Neurosurgery, Baylor College of Medicine, Houston, TX, USA

   Junsung Woo, Michael R. Williamson, Dongjoo Choi, Kwanha Yu, Wookbong Kwon, Akdes Serin-Harmanci & Benjamin Deneen

4. Developmental Biology Program, Baylor College of Medicine, Houston, TX, USA

   Ekin Akdemir & Benjamin Deneen

5. Department of Neuroscience, Baylor College of Medicine, Houston, TX, USA

   Thea Zlatkov

6. Program in Immunology and Microbiology, Baylor College of Medicine, Houston, TX, USA

   Yeunjung Ko

7. Lester and Sue Smith Breast Center, Baylor College of Medicine, Houston, TX, USA

   Jong Min Choi

Contributions

J.W., M.R.W. and B.D. conceived the project and designed the experiments. J.W. and M.R.W. generated the AAV viral constructs, performed brain injections, behavioural studies, analysis of brain tissue and RNA-seq studies. J.W. performed electrophysiological recordings and M.R.W. performed the stroke experiments. E.A. generated the CD51-FLP mouse lines. D.C., Y.K. and A.S.-H. assisted with the RNA-seq studies and bioinformatics analysis. T.Z., K.Y., E.M., W.K. and H.M. assisted with analysis and imaging of brain tissue. J.M.C. assisted with the MS–IP experiment. K.S. assisted with animal husbandry. M.R.W., J.W. and B.D. wrote the manuscript.

Corresponding author

Correspondence to Benjamin Deneen.

Competing interests

The authors declare no competing interests.

Peer review information

Nature thanks Baptiste Lacoste, Jeffrey Rothstein and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions