Astrocytes appear in the cortex at birth, and migrate, proliferate, and mature throughout the first postnatal month (Ge et al., 2012), coincidently with the stages of synapse development. To determine the transcriptomic profile of astrocytes at these stages (P7, P14, P28, and adult, P120), we used the astrocyte Ribotag mouse model to isolate mRNA for bulk RNA sequencing (B6N.129-Rpl22^tm1.1Psam/J crossed to B6.Cg-Tg(Gfap-cre)73.12Mvs/J – labeled Astrocyte-RiboTag) (Figure 1, Figure 1—figure supplement 1, Figure 1—source data 1; Boisvert et al., 2018; Chai et al., 2017). In this mouse line, cre recombinase drives the expression of an HA-tagged ribosomal subunit, enabling purification of cell-type-specific ribosomes and associated mRNA for analysis by RNA sequencing (Figure 1A and B). We found a significant enrichment in astrocytic genes over other cell types in the mRNA isolated by HA immunopurification (IP; labeled Astro) compared to total VC mRNA (input; IN; Figure 1—figure supplement 1A). Furthermore, immunostaining analysis of VC sections at P28 shows a high overlap between the HA ribosome tag and the astrocyte marker S100β, but not with other cell-type markers (Figure 1—figure supplement 1B-D), with more than 95% of S100β-positive cells also positive for the HA tag, suggesting high astrocyte coverage of our mRNA isolation, consistent with our previous analysis in the adult (Boisvert et al., 2018). The complete RNA sequencing dataset is available in a searchable format online (http://igc1.salk.edu:3838/astrocyte_transcriptome/).

To assess if there are broad changes in the transcriptomic profiles of astrocytes across development, we performed principal component analysis (PCA; Figure 1C). This showed that P7 and P14 astrocytes form distinct clusters, while P28 and P120 astrocytes cluster together. To investigate this further, we analyzed the number of differentially expressed genes (DEGs; fragments per kilobase of exon per million mapped fragments [FPKM] > 1; false discovery rate [FDR] < 0.05) between each age group. DEGs are classified into total genes (all changes detected), genes that are expressed by astrocytes (IP/input > 0.75), and genes that are enriched in astrocytes (IP/input > 3; Figure 1D, for definitions, see also Boisvert et al., 2018). The largest number of DEGs is between P7 and P28 (~6000 total genes), and smallest numbers between P14 and P28 (~1000 total genes), and P28 and P120 (~2000 total genes). Analysis of astrocyte-enriched genes (IP/input > 3) showed that ~60% of all astrocyte-enriched genes are significantly changed from P7 to P14, while only ~20% are changing between P28 and P120 (Figure 1E). This shows that most changes in astrocyte gene expression are occurring between the first and second postnatal weeks, a time of transition from synapse formation to synapse maturation, and in the VC, from spontaneous to visually evoked neuronal activity.

To determine the different astrocyte functions at each age, we focused on the astrocyte-enriched gene lists (~1000 total genes FPKM >1, astro IP/input > 3). To identify common genes expressed at all ages, as well as astrocyte-enriched genes that are most highly expressed at each age, we performed Gene Ontology (GO) analysis, focusing on Biological Process (BP) terms (Figure 1F–H, Figure 1—figure supplement 1F and G). This showed that genes common to all ages are enriched in GO terms related to cholesterol processing and serine synthesis, confirming the previously established important role of astrocytes in regulating brain cholesterol (Orth and Bellosta, 2012; Figure 1G). Analysis of GO terms unique to each age showed that at P7 astrocyte genes are enriched in GO terms related to cortical development, while at P14 astrocyte genes are enriched in GO terms related to Wnt and BMP signaling pathways. Conversely, adult astrocytes (P120) are enriched in terms related to regulation of extracellular matrix assembly and contact inhibition (Figure 1H). In all we found both the number of genes and the GO terms common to all ages constitute about 50% of all terms identified for each age, while genes and GO terms unique to each age consist less than 10% of all terms (Figure 1—figure supplement 1F and G). These results demonstrate that astrocytes perform their core functions across all developmental stages similarly, with several age-specific functions that turn on and off depending on the developmental stage and the environment in the VC.

An important analysis to be performed with this dataset is determining the temporal expression changes of known astrocyte genes, for example, to identify potential astrocyte markers that are either global or age specific. We found that Apoe and Cst3 are the most highly expressed genes in astrocytes at all ages (peak FPKM ~10,000; Figure 1F, Figure 1—source data 1), while amongst highly expressed astrocyte genes (FPKM >100) Lars2 is the most astrocyte-enriched gene at all ages (astro IP/ input ~40; Figure 1—figure supplement 1E). These genes are highly expressed in adult astrocytes across multiple brain regions (Morel et al., 2017), thus can be potentially utilized to mark or target astrocytes globally. The expression of another well-known astrocytic gene Aldh1l1 (Morel et al., 2017; Figure 1I) is stable across all ages, making it an optimal marker for astrocytes during both development and in the adult (Srinivasan et al., 2016). On the other hand, S100b expression is upregulated later in development, making it more suitable to mark adult astrocytes (Figure 1I). For genes that encode proteins important for astrocyte function, we found that the metabotropic glutamate receptor Grm5 (mGluR5) is most highly expressed at P7 and then declines with maturation, confirming previous reports (Catania et al., 1994; Sun et al., 2013), while the glutamate transporter Slc1a2 (Glt1) and the connexins (Gja1, Gjb6) are significantly upregulated from P14 onwards (Figure 1I).

Our goal is to characterize how astrocytes regulate synapses, so we next used the astrocyte Ribotag dataset to investigate the developmental expression changes of key synapse-regulating genes (Figure 1J). These include astrocyte-secreted thrombospondins (Thbs), which induce silent synapse formation. The family members expressed by astrocytes in the CNS show divergent expression, with Thbs1 being significantly higher at P7 than later ages, whereas Thbs4 is significantly higher at P120 than P7 (Figure 1J). This temporal expression profile fits with previous studies that have demonstrated important roles for Thbs1 in initial synapse formation at P7 (Christopherson et al., 2005), and suggested roles for Thbs4 in the adult brain (Benner et al., 2013). Similarly, glypican (Gpc) family members have a divergent expression. Though redundant in culture for their ability to regulate immature synapse formation (Allen et al., 2012), Gpc4 and Gpc6 show different temporal expression in vivo. While Gpc4 is most highly expressed at P7 and gradually declines with maturation, Gpc6 peaks at P14–P28. Gpc5, a glypican family member with yet unknown function, has low expression at P7, and is significantly increased at all later ages. Astrocyte-secreted chordin-like 1 (Chrdl1) regulates synapse maturation and its expression peaks at P14, confirming previous analysis (Blanco-Suarez et al., 2018; Figure 1J). These temporal expression profiles are not limited to factors that promote synapse formation. Astrocyte phagocytic receptors involved in synapse elimination, Megf10 and Mertk (Chung et al., 2013), significantly increase in expression between P7 and P14, coincident with the initiation of synapse elimination.

Our RNA sequencing analysis demonstrates that in the developing VC synapse-regulating genes show differential temporal expression patterns. However, whether these levels are equal across astrocytes in all cortical layers throughout development is unknown. During the first postnatal weeks, when astrocytes regulate synapse development, they are still migrating and dividing (Ge et al., 2012). Yet, how developing astrocytes populate each of the cortical layers, and their ratio relative to other cortical cells across development has not been quantified. To address this, we utilized the well-established mouse line where astrocytes express GFP under the Aldh1l1 promoter (Cahoy et al., 2008; Dougherty et al., 2010; John Lin et al., 2017; Stogsdill et al., 2017; Tien et al., 2012). Immunostaining of brain sections from Aldh1l1-GFP mice at P7 and P28 with antibodies against known astrocyte markers ALDH1L1, S100β, and SOX9 showed high overlap between GFP and marker-positive cells, suggesting the majority of astrocytes in the VC express GFP in this mouse line (Figure 2—figure supplement 1A–C), further validating its usage. We quantified astrocyte numbers within each of the six neuronal layers at the developmental time points, which correspond to astrocyte and synapse development throughout the first postnatal month (P1, P4, P7, P14, P28; Figure 2A–C). At birth (P1), very few astrocytes are present, comprising 0.5–2% of the total cell number in the VC (represented as GFP-positive cells as a percentage of all cells marked by the nuclear dye DAPI), with a significantly higher percentage of astrocytes in deeper layers than upper layers (Figure 2C and D, Figure 2—source data 1A). The astrocyte percentage increases with development in all cortical layers, peaking at P21–P28. At this time, astrocytes are ~10% of total cell number in layers (L) 2–6, and ~50% in L1 (Figure 2C and D, Figure 2—source data 1A). As the brain develops, the distance between cells grows to accommodate the increase in cell size and complexity, as evident by a significant decrease in DAPI-positive nuclei per mm² of VC that occurs from P1 to P14–P28 (Figure 2—figure supplement 1D, Figure 2—source data 1A). Despite this decrease in total cell density, the proportion of astrocytes remains constant in all cortical layers and across ages, with the exception of L1–2/3, where it is significantly lower at P1 (Figure 2C and E). This stability in astrocyte density is likely explained by new astrocytes still being generated in the weeks after birth (Ge et al., 2012).

Having established the developmental profile of astrocytes, we next determined if there are layer-specific developmental changes in mRNA expression of synapse-regulating genes by performing single-molecule fluorescent in situ hybridization (smFISH; RNAscope) on brain sections of Aldh1l1-GFP mice. We probed for seven genes that regulate distinct aspects of synaptogenesis: active synapse-regulating – glypicans (Gpc) 4, 5, 6; synapse maturation regulating – chordin-like 1 (Chrdl1); and silent synapse-regulating – thrombospondins (Thbs1, 2, 4) (Figure 2F–M, Figure 2—figure supplement 2, Figure 2—source data 1B and C). Expression of each gene (represented as total area of thresholded signal in µm²; labeled as thresh area) was analyzed within the territory of GFP-positive astrocytes in each cortical layer at similar time points as above: P4, P7, P14, and P28, when most alterations in astrocyte transcriptome (Figure 1) and synapse development occur (Farhy-Tselnicker and Allen, 2018), and excluding P1, when the numbers of astrocytes in the cortex are very low. A negative control probe was used to determine the minimal signal threshold of detection (Figure 2—figure supplement 2H and I).

We first analyzed glypicans, factors that promote active synapse formation. Bulk RNAseq (Ribotag) showed that Gpc4 expression by astrocytes is highest at P7 and gradually declines with maturation (Figure 1J, Figure 1—source data 1A). Layer-specific analysis, however, shows that these differences are driven by astrocytes in L1: Gpc4 expression decreases between P7 and P14 only in L1 astrocytes, staying stable across development in all other layers (thresh area [μm²] L1: P4 3.1 ± 0.5; P7 4.1 ± 0.1; P14 1.7 ± 0.2; P28 1.1 ± 0.3; Figure 2F and G, Figure 2—source data 1B and C). Gpc6, on the other hand, peaks at P14–P28 in bulk sequencing. Layer-specific analysis showed that this increase occurs in astrocytes between P7 and P14 in L2–5, with significant upregulation in L5, and remaining high at P28 (thresh area [μm²] L5: P4 2.5 ± 0.4; P7 2.5 ± 0.4; P14 4.4 ± 0.4; P28 3.1 ± 0.4; Figure 2H and I, Figure 2—source data 1B and C). Gpc5 is strongly upregulated at P14 and remains high at P28 in astrocytes in all layers, matching the bulk RNAseq data (thresh area [μm²] L1: P4 0.5 ± 0.1; P7 1.3 ± 0.2; P14 5.9 ± 0.7; P28 5.3 ± 1.1; Figure 2J and K, Figure 2—source data 1B and C).

The expression of the synapse maturation factor Chrdl1 peaks at P14 in the bulk RNAseq data (Figure 1J, Figure 1—source data 1A). Spatial analysis revealed that this increase from P7 to P14 is layer-specific, with the largest increase in Chrdl1 occurring in upper layer astrocytes with significant upregulation in L2/3 (thresh area [μm²] L2/3: P4 4.1 ± 0.2; P7 4.1 ± 1; P14 9.9 ± 0.6; P28 7.6 ± 1; Figure 2L and M, Figure 2—source data 1B and C). Further, at its peak of expression, Chrdl1 is highest in L1–4 astrocytes compared to L5–6, demonstrating a heterogeneous expression across layers, as previously reported (Bayraktar et al., 2020; Blanco-Suarez et al., 2018). Finally, we analyzed astrocyte thrombospondins, factors that induce silent synapse formation (Figure 2—figure supplement 2B-G). Thbs mRNA levels in the VC are much lower at their peak expression than glypicans or Chrdl1, consistent with our bulk RNAseq results (Figure 1J, Figure 1—source data 1A), and previous studies showing low thrombospondin expression in the resting state, and an upregulation by learning or injury (Nagai et al., 2019; Tyzack et al., 2014). Nevertheless, we observed an increase in Thbs2 expression in L4–5 at P28 (thresh area [μm²] L5: P4 0.7 ± 0.1; P7 0.7 ± 0.19; P14 1 ± 0.1; P28 1.8 ± 0.3; Figure 2—figure supplement 2D and E, Figure 2—source data 1B); and an increase in Thbs4 at P28 in all layers (thresh area [μm²] L1: P4 0.1 ± 0.01; P7 0.1 ± 0.1; P14 0.3 ± 0.1; P28 1.9 ± 0.5; Figure 2—figure supplement 2F and G, Figure 2—source data 1).

The seven astrocyte synapse-regulating genes we analyzed using smFISH all regulate formation and function of excitatory glutamatergic synapses. In the case of Gpc4, Gpc6, and Chrdl1, they regulate glutamatergic synapse development by primarily affecting localization of AMPA glutamate receptor (AMPAR) subunits (Allen et al., 2012; Blanco-Suarez et al., 2018). Previous studies in the rat somatosensory cortex have identified developmental alterations in AMPAR subunit expression (Brill and Huguenard, 2008; Kumar et al., 2002); however, their spatio-temporal expression patterns in the developing mouse VC, and whether these correlate with the expression of astrocytic genes that regulate them, have not been systematically analyzed (Figure 3A and B). To detect postsynaptic AMPA glutamate receptors, we stained for GLUA1 and GLUA2 subunits (Figure 3C–F, Figure 3—source data 1). To detect the corresponding glutamatergic presynaptic terminals, we stained for VGLUT1 to identify local cortico-cortical connections (Figure 3G and H, Figure 3—figure supplement 1A, Figure 3—source data 1) and VGLUT2 to identify thalamocortical connections (Figure 3I and J, Figure 3—figure supplement 1B, Figure 3—source data 1; Fremeau et al., 2001). As expected based on the literature, GLUA1 levels peak at earlier time points than GLUA2 (Figure 3C and D, Figure 3—source data 1). GLUA2 immunoreactivity significantly increases from P7 to P14 in L1–5, and then remains stable to P28 (Figure 3E and F, Figure 3—source data 1). At all ages, the levels of GLUA1 and GLUA2 are significantly higher in L1 than all other layers, consistent with L1 being rich in synaptic connections (Figure 3—source data 1; Douglas and Martin, 2004). VGLUT1 immunoreactivity greatly increases between P7 and P14 in all cortical layers and remains stable at later ages (Figure 3G and H, Figure 3—figure supplement 1A, Figure 3—source data 1). VGLUT2 levels steadily increase from P1 to P14, and then remain stable. VGLUT2 immunoreactivity is significantly higher in L1 and L4 than other layers at all ages, consistent with these being thalamic innervation zones and previous findings in rat cortex (Figure 3I and J, Figure 3—figure supplement 1B, Figure 3—source data 1; Boulland et al., 2004; Conti et al., 2005; Fremeau et al., 2001; Lopez-Bendito and Molnar, 2003). Taken together, these data show that synaptic proteins show complex spatio-temporal expression patterns, with multiple significant changes occurring between P7 and P14, as also observed for astrocyte gene expression.

In summary, the transcriptomic analysis reveals significant changes in astrocyte gene expression across development, with the most prominent changes occurring between P7 and later ages, a time between synapse initiation and maturation. We have further determined the spatio-temporal expression profile of key astrocyte synapse-regulating factors, identifying divergent developmental and layer-specific expression patterns within the same families of genes. These findings strongly suggest that astrocyte expression of synapse-regulating genes is closely tied to the developmental stage of the cortex, which features both changes in neuronal and astrocyte activities across development. In addition to revealing important information about the developmental changes in expression of synapse-regulating genes, these transcriptomic data can be further utilized to inform further studies on astrocyte development. The complete dataset and GO term list are available in Figure 1—source data 1.

Having found broad differences in astrocyte expression of synapse-regulating genes across cortical layers, we next asked what are the possible physiological mechanisms that regulate these layer-specific alterations between P7 and P14 in the developing VC in vivo. Our experiments using cultured astrocytes and neurons show that GPC4 protein secretion from astrocytes is significantly reduced in the presence of neurons (Figure 4—figure supplement 1A, Figure 4—figure supplement 1—source data 1) or when astrocytes are incubated with neurotransmitters including glutamate, adenosine, or ATP (Figure 4—figure supplement 1B, Figure 4—figure supplement 1—source data 1). Others have reported a similar effect on Gpc4 mRNA expression (Hasel et al., 2017), suggesting that neuronal activity can influence expression and release of synapse-regulating factors from astrocytes. Indeed, significant alterations in the activity patterns of both glutamatergic and GABAergic neurons occurs in the VC at around P14 (Espinosa and Stryker, 2012). Since the developmental expression changes in Gpc4 and Chrdl1 occurred mostly in the upper cortical layers innervated by thalamic neurons, we hypothesized that levels of these genes may be regulated by changes in the activity of thalamic neurons that occur upon eye opening at ~P12, when there is a transition from spontaneous to visually evoked activity in the retina that is relayed via the thalamus to the VC.

To investigate this, we generated mice where the release of glutamate from thalamocortical terminals in the VC is perturbed by knocking out the vesicular glutamate transporter VGlut2 (Slc17a6) from neurons in the dLGN. Knockout of VGlut2 has been previously shown to abolish presynaptic release of glutamate in VGlut2-expressing neurons, in full or conditional knockout mouse models (Wallén-Mackenzie et al., 2010). We crossed Slc17a6^f/f mice (Slc17a6^tm1Lowl/J; labeled as VGlut2 WT) to an RORα cre line (Rora^tm1(cre)Ddmo; Slc17a6^f/f;cre labeled as VGlut2 cKO), where cre recombinase is expressed in thalamic neurons including the dLGN (Figure 4A; Chou et al., 2013; Farhy-Tselnicker et al., 2017). Immunostaining experiments showed a significant decrease in VGLUT2 signal in the VC of VGlut2 cKO mice compared to WT at P14 (Figure 4B). Notably, VGlut2 cKO did not strongly affect the levels of VGLUT1, which marks cortico-cortical connections (Figure 4C; Wallén-Mackenzie et al., 2010), though a small but significant increase in VGLUT1 immunoreactivity was observed in L4, a major target of thalamocortical projections. This suggests that the normal upregulation in VGLUT1 that occurs at P14 across all other cortical layers (Figure 3) is either intrinsic to the cortical neurons and/or regulated by mechanisms other than dLGN-VC-evoked activity.

A lack of VGLUT2 immunoreactivity in the VC could also result from an absence of thalamic axons innervating their target regions. To test whether that is the case, we crossed RORα cre and VGlut2 cKO mice with a tdTomato reporter line (B6.C-Gt(ROSA)26Sor^{tm14(CAG-tdTomato)Hze/J}; labeled as tdTomato) to visualize dLGN axons (Figure 4A). All cre+ tdTomato+ groups (VGlut2 WT (Slc17a6^+/+), VGlut2 cHet (Slc17a6^f/+), and VGlut2 cKO (Slc17a6^f/f)) showed a comparable number and volume of tdTomato-labeled projections in L1 and L4 of the VC (Figure 4—figure supplement 2A and B). Analysis of VGLUT2 puncta colocalized with tdTomato-positive axons showed a significant decrease in number in VGlut2 cHet and VGlut2 cKO compared to WT (Figure 4—figure supplement 2A, C). These results show that in the VGlut2 cKO mice thalamic axons are present at their target layers in the VC but lack VGLUT2, as has been shown in studies performing similar manipulations, suggesting that they are functionally silent (Li et al., 2013; Zechel et al., 2016).

Does the lack of thalamocortical glutamate release influence the expression of synapse-regulating genes in astrocytes? To address this, we performed smFISH probing for Gpc4, Gpc6, or Chrdl1, along with a probe for the glutamate transporter Slc1a3 (Glast) (Ullensvang et al., 1997) to label astrocytes (Figure 4D–I, Figure 4—figure supplement 2D–J, Figure 4—source data 1). The number of cortical astrocytes marked by Slc1a3, and the expression level of Slc1a3 mRNA, is not affected by VGlut2 cKO in any cortical layer at P14, showing that gross astrocyte development proceeds normally in the absence of thalamic input (Figure 4—figure supplement 2D and E). During normal development, Gpc4 expression significantly decreases between P7 and P14 specifically in L1 astrocytes (Figure 2F and G). In VGlut2 cKO mice, this change no longer occurs (Figure 4D and E). Gpc4 expression is significantly increased in VGlut2 cKO compared to WT at P14 specifically in L1 astrocytes and unchanged in all other layers (thresh area [μm²]: L1 WT 1.3 ± 0.1; cKO 2.3 ± 0.2; Figure 4D and E, Figure 4—source data 1). Gpc6 expression is normally increasing in astrocytes in deep layers between P7 and P14 (Figure 2H and I). In the absence of thalamic glutamate release, Gpc6 is lower in astrocytes in layers 4 and 5 than in the WT at P14, showing that the normal developmental upregulation has been blocked (thresh area [μm²]: L4 WT 2.7 ± 0.4; cKO 1.5 ± 0.1; Figure 4F and G, Figure 4—source data 1). Similarly, Chrdl1 expression normally increases in upper layer astrocytes between P7 and P14 (Figure 2L and M); however, it is significantly decreased in VGlut2 cKO compared to WT specifically in astrocytes in upper layers (1–4), and not affected in deep layers at P14 (thresh area [μm²]: L1 WT 6 ± 0.4; cKO 4.6 ± 0.5; Figure 4H and I, Figure 4—source data 1). To determine if these alterations are due to the loss of thalamic neuron glutamate release at a specific developmental time (e.g., eye opening at ~P12), we analyzed the expression of Gpc4, Gpc6, and Chrdl1 at P7 (prior to eye opening). We found no difference in expression of any of these genes (Figure 4—figure supplement 2H–J), nor any difference in the number of cortical astrocytes marked by Slc1a3, or the expression level of Slc1a3 mRNA (Figure 4—figure supplement 2F and G). These results show that during VC development from P7 to P14 glutamate release from thalamic neurons regulates the developmental expression of astrocyte Gpc4, Gpc6, and Chrdl1 in a layer-specific manner.

Are there any consequences of decreased glutamate release and subsequent altered expression of astrocyte synapse-regulating genes on synaptic development? To address this, we performed immunostaining for pre- and postsynaptic proteins in VGlut2 cKO mice and their WT controls in L1 of VC at P14. We first analyzed cortico-cortical synapses marked by VGLUT1 and found that, as in the low-resolution characterization (Figure 4C), there is no change in VGLUT1 puncta in the absence of VGLUT2 (Figure 4J, K, N and O). In the case of GLUA1 containing AMPARs, which are regulated by Gpc4, we found a significant increase in the number of GLUA1 puncta and their colocalization with VGLUT1 in the VGlut2 cKO, correlating with the observed increase in Gpc4 (GLUA1 FC 1.24 ± 0.04; Figure 4J and L; Coloc FC 1.38 ± 0.07; Figure 4J and M). For GLUA2 containing AMPARs, which are regulated by Chrdl1, we found a significant decrease in both the number of GLUA2 puncta and the number of colocalized GLUA2-VGLUT1 puncta in VGlut2 cKO mice compared to WT, correlating with the observed decrease in Chrdl1 (GLUA2 FC 0.77 ± 0.04; Figure 4N and P; Coloc FC 0.75 ± 0.1; Figure 4N and Q). We asked if similar effects are also present at thalamocortical synapses. Since VGLUT2 is absent in cKO mice, we used Slc17a6^{f/f; RORαcre;tdTomato} (labeled as VGlut2 cKO; tdTomato) to label thalamic axons (Figure 4A, Figure 4—figure supplement 2A) and identified presynaptic active zones within tdTomato axons by immunostaining for the presynaptic marker Bassoon (Figure 4—figure supplement 2K and O). We found no difference in the number of Bassoon puncta colocalized with tdTomato between the WT and cKO mice, a further indication that synapses form in the absence of VGLUT2 (Figure 4—figure supplement 2K, L, O, P), and fitting with findings from mice that globally lack presynaptic release (Verhage et al., 2000). As is the case for cortico-cortical synapses, we found an increase in GLUA1 and colocalization of GLUA1 with Bassoon and tdTomato (GLUA1 FC 1.21 ± 0.09; Figure 4—figure supplement 2K and M; Coloc FC 1.36 ± 0.23; Figure 4—figure supplement 2K and N), and a decrease in total GLUA2 and GLUA2-Bassoon synapses (GLUA2 FC 0.70 ± 0.08; Figure 4—figure supplement 2O and Q; Coloc FC 0.65 ± 0.13; Figure 4—figure supplement 2O and R).

These results show that synaptic GLUA1 and GLUA2 levels are altered in VGlut2 cKO VC in the direction which follows the change in astrocytic expression of Gpc4 (which recruits GLUA1) and Chrdl1 (which recruits GLUA2) in L1. These correlated changes in astrocyte genes and synaptic proteins suggest a disruption in synapse maturation in the VC at P14 in the absence of thalamocortical glutamate release that may be mediated by astrocytes.

Since we observed that changes in thalamic glutamate release influence the expression of astrocyte synapse-regulating genes, we next asked how perturbing the astrocyte response to neurotransmitters affects the expression of Gpc4, Gpc6, and Chrdl1. Astrocytes express many neurotransmitter receptors, in particular GPCRs, and respond to neurotransmitters with increased intracellular calcium (Kofuji and Araque, 2021; Porter and McCarthy, 1997). In the case of somal increases in calcium, which have the potential to regulate expression of activity-regulated genes, most of this increase is mediated by the release of calcium from intracellular stores via IP3R2 (Itpr2) (Srinivasan et al., 2015). We therefore asked if blunting astrocyte calcium signaling by removing store-mediated calcium release using Itpr2 KO mice (Itpr2^tm1.1Chen; labeled Ip3r2 KO; Figure 5A and B) has an impact on the expression of synapse-regulating genes.

To determine this, we performed smFISH on the VC of P14 Ip3r2 KO and WT mice, marking astrocytes with a probe against Slc1a3 along with Gpc4, Gpc6, or Chrdl1. At P14, knocking out Ip3r2 does not affect the number of astrocytes or the expression levels of Slc1a3, showing astrocytes develop grossly normally when store-mediated calcium release is diminished (Figure 5—figure supplement 1A and B, Figure 5—source data 1, Figure 5—source data 2). However, loss of IP3R2 does affect expression of synapse-regulating genes. In the case of Gpc4, the mRNA level is reduced in astrocytes in all layers, with a significant decrease occurring in L1, 4, and 6 (thresh area [μm²]: L1 WT 1.5 ± 0.1; KO 1.1 ± 0.1; Figure 5C and D, Figure 5—source data 1). For Gpc6, there is no difference in the mRNA level between Ip3r2 KO and WT in astrocytes in any layer (Figure 5E and F, Figure 5—source data 1). Chrdl1 expression is increased in astrocytes in all layers, with a significant increase occurring in L1, 2/3, and 5 (thresh area [μm²]: L1 WT 7.3 ± 1.1; KO 10.6 ± 1; Figure 5G and H, Figure 5—source data 1). To ask if these alterations are present throughout development, we performed the same analysis at P7. As with P14, at P7 there is no change in astrocyte number or Slc1a3 mRNA signal, showing astrocytes develop grossly normally (Figure 5—figure supplement 1C and D). In the case of Gpc4 and Chrdl1, there is no difference in expression between Ip3r2 KO and WT at P7 (Figure 5—figure supplement 1E and G), whereas for Gpc6 there is a significant increase in the Ip3r2 KO restricted to L4 (Figure 5—figure supplement 1F). Therefore, in contrast to the layer-specific alterations in gene expression in the VGlut2 cKO mice, removing IP3R2 impacts astrocytes in all layers and does not strictly follow the developmental trajectory. This suggests a broad requirement for astrocyte calcium signaling in all astrocytes to maintain the correct level of gene expression, and that the signals to do this come from multiple sources and are not restricted to thalamic inputs.

Are there any consequences of diminished astrocyte calcium signaling and altered expression of synapse-regulating genes on synaptic development? As with the VGlut2 cKO, we addressed this by performing immunostaining for presynaptic terminals (VGLUT1 or VGLUT2) and postsynaptic AMPAR subunits (GLUA1 or GLUA2) in Ip3r2 KO mice and WT controls in L1 of VC at P14. We found that the numbers of both cortico-cortical presynaptic terminals marked by VGLUT1, and thalamocortical terminals marked by VGLUT2, are significantly decreased in Ip3r2 KO mice compared to WT, demonstrating a global deficit in synapse formation in the absence of astrocyte calcium signaling (VGLUT1 from GLUA1: FC 0.86 ± 0.04; Figure 5I and J; VGLUT1 from GLUA2: FC 0.87 ± 0.05; Figure 5M and N. VGLUT2 from GLUA1: FC 0.89 ± 0.05; Figure 5—figure supplement 1H and I; VGLUT2 from GLUA2: FC 0.80 ± 0.05; Figure 5—figure supplement 1L and M). For GLUA1 containing AMPARs, which are regulated by Gpc4, we found a significant decrease in the total number of puncta and their colocalization with VGLUT1 and VGLUT2 in the Ip3r2 KO, correlating with the observed decrease in Gpc4 mRNA (VGLUT1-GLUA1: GLUA1 FC 0.84 ± 0.03; Figure 5I and K; Coloc FC 0.70 ± 0.03; Figure 5I and L; VGLUT2-GLUA1: GLUA1 FC 0.83 ± 0.03; Figure 5—figure supplement 1H and J; Coloc FC 0.74 ± 0.03; Figure 5—figure supplement 1H and K). For GLUA2 containing AMPARs, which are regulated by Chrdl1, we found a significant increase in their number, correlating with the observed increase in Chrdl1 (VGLUT1-GLUA2: GLUA2 FC 1.22 ± 0.07; Figure 5M and O; VGLUT2-GLUA2: GLUA2 FC 1.14 ± 0.03; Figure 5—figure supplement 1L and N). The number of colocalized presynaptic puncta and GLUA2 is, however, unchanged, likely due to the opposing decrease in presynaptic puncta and increase in GLUA2 (VGLUT1-GLUA2 Coloc FC 1.09 ± 0.1; Figure 5M and P; VGLUT2-GLUA2 Coloc FC 1.00 ± 0.08; Figure 5—figure supplement 1L and O).

These results show that GLUA1 and GLUA2 levels are altered in Ip3r2 KO VC in the direction which follows the change in astrocytic expression of Gpc4 (which recruits GLUA1) and Chrdl1 (which recruits GLUA2). This strongly suggests that both astrocytes and neurons play an important role in regulating the expression of synapse-regulating genes, and subsequently AMPAR subunit protein levels, and the final expression levels arise from the complex interaction between these two cell types.

Having found that multiple synapse-regulating genes in astrocytes show layer-specific enrichment, and that these patterns are regulated by neuronal and astrocyte activity, we next asked if these findings are specific to synapse development, or if other astrocyte genes show a similar pattern. To address this using an unbiased approach, we performed single-nucleus RNA sequencing of glial cells isolated from the P14 VC of wild type, VGlut2 cKO, and Ip3r2 KO mice. To isolate the glial cell populations, we immunostained VC nuclei in suspension with an antibody against the neuronal marker NeuN and performed FACS to select the NeuN-negative population (Figure 6A). We used the Chromium 10X system to isolate individual glial nuclei and performed RNA sequencing to quantify mRNA levels (Figure 6A). This identified 22,781 cells in the VGlut2 condition (cKO and WT) (Figure 6B, Figure 6—figure supplement 1A) and 21,240 cells in the Ip3r2 condition (KO and WT) (Figure 6—figure supplement 1B and C). Initial clustering analysis determined 17 distinct cell populations in both models, with the majority of cells detected clustered within the main glial cell types: astrocytes, microglia, and oligodendrocyte lineage cells (Figure 6B, Figure 6—figure supplement 1A–C). Just two clusters enriched for neuronal markers are present (Hernandez et al., 2019), showing that the NeuN depletion had been successful. This dataset is available in a searchable format online (https://cells.ucsc.edu/?ds=mouse-astro-dev; Speir et al., 2021 ).

We focused our downstream analysis on astrocytes. A second round of unbiased clustering of the astrocyte population identified four groups (Figure 6C, Figure 6—figure supplement 2A) in each model and genotype. By comparing the genes enriched in each cluster with datasets in the literature, we determined these to anatomically correspond to upper (L1–2/3), middle (mid; L2/3–5), deep (L5–6) layer, and white matter (WM) astrocytes (Figure 6C, Figure 6—figure supplement 2A; Batiuk et al., 2020; Bayraktar et al., 2020; Lanjakornsiripan et al., 2018). Similar cell numbers were identified in each group across models and genotypes (number of nuclei in VGlut2 cKO model: upper 1114 WT; 1065 cKO, mid 1289 WT; 1163 cKO, deep 440 WT; 466 cKO, WM 273 WT; 268 cKO; Figure 6—figure supplement 2C). We also determined the fractions of astrocytes present in each group and found that this corresponds to the fractions we identified via anatomical cell counts (Figure 6—figure supplement 2D), showing that the process of nuclear isolation has captured astrocytes in levels that reflect their in vivo abundance. To validate the layer-enriched genes, we took advantage of publicly available in situ hybridization datasets generated by the Allen Brain Institute (Allen Brain Institute, 2008; available from https://developingmouse.brain-map.org/). Cross-referencing cluster-enriched genes identified in our study with P14 in situ hybridization datasets, we found that the gene Kcnd2 is expressed in mid cortical layers, Id3 is enriched in deep layers, and Gfap is enriched in white matter, matching the snRNAseq (Figure 6—figure supplement 2B).

Pairwise comparisons between each of the clusters using the WT astrocytes of each model as the input cells showed that the most robust differences are between the deep and upper layer astrocytes, with over 700 DEGs and up to sixfold log2 FC in expression level (Figure 6D, Figure 6—figure supplement 2E, Figure 6—source data 1A and B). On the other hand, upper and mid astrocytes are the most similar, with about 200 DEGs and twofold log2 FC maximal difference in mRNA level (Figure 6D, Figure 6—figure supplement 2E, Figure 6—source data 1A and B). GO terms analysis of genes enriched in each cluster identified between 300 and 600 terms significantly enriched per cluster, with 140 terms that are common to all four clusters, and between 40 and 120 terms that are unique to each cluster (Figure 6—figure supplement 2F, Figure 6—source data 1C). Terms with the highest gene ratio in the upper astrocyte cluster include pathways related to signal transduction and membrane biogenesis (Figure 6—figure supplement 2G), whereas mid astrocytes are enriched in terms related to GABAergic signaling and PSD95 clustering. Genes belonging to the deep astrocyte cluster are enriched in GO terms related to the regulation of pre- and postsynapse organization (Figure 6—figure supplement 2F), while WM astrocyte genes are enriched with pathways related to axonal guidance, maintenance, and signaling (Figure 6—source data 1C). Importantly, while there are differences in astrocyte gene expression across layer groups, these are mostly gradients of gene expression, suggesting that the astrocyte layer groups are on a continuum rather than distinct cell types (Figure 6C, Figure 6—figure supplement 2A; Bayraktar et al., 2020; John Lin et al., 2017).

Next, we asked how astrocyte marker, function, and synapse-regulating genes highlighted in the bulk RNAseq dataset (Figure 1G and H) are expressed across layers (Figure 6—figure supplement 2I and J). Overall, we found a positive correlation between levels of gene expression obtained by the two sequencing methods, meaning that genes that were shown to be highly expressed in the bulk dataset (such as ApoE) were also highly expressed in the snRNAseq dataset (Figure 6—figure supplement 2I). Unsurprisingly, sequencing of bulk RNA samples was more sensitive in detecting the low expressed genes, such as Gpc4, Tgfb1, and Thbs1, which were close to the detection threshold in the snRNAseq dataset, precluding statistical analysis. Nevertheless, plotting the total counts per million mapped reads (CPM) levels for Gpc4, Gpc6, and Chrdl1 revealed that they matched the spatial analysis performed at the same age (Figure 3). Gpc4 expression is lowest in the upper layer cluster, Chrdl1 expression is highest in the upper layer cluster, while Gpc6 levels are similar across all four populations (Figure 6—figure supplement 2J). For other astrocyte marker, function, and synapse-regulating genes detected in the snRNAseq dataset, most exhibited similar levels of expression in all layer groups, with some notable exceptions. For example, Gfap and Aqp4 expression is higher in deep and WM astrocytes than in upper and mid groups, while the expression of connexin 43 (Gja1) is highest in deep layer astrocytes compared to all other groups (Figure 6—figure supplement 2I). Taken together, these results show that wild-type astrocytes are transcriptomically diverse, but not distinct, in the developing VC, in accordance with previous studies in which astrocyte diversity was assessed at a similar developmental stage (Bayraktar et al., 2020).

Given that we found that astrocyte synapse-regulating genes are regulated by both neuronal and astrocyte activity, we next asked what other astrocyte genes are affected by these activity manipulations. To increase the power of our analysis, we combined the four astrocyte subpopulations into one group for each genotype and used this combined group to identify DEGs between the WT and KO. We found 61 DEGs for the VGlut2 cKO model and 131 DEGs for the Ip3r2 KO model (Figure 6E–H, Figure 6—source data 1D). Performing the same analysis on two other abundant glial populations, OPCs and microglia, showed 28 DEGs for OPCs and 24 DEGs for microglia in the VGlut2 cKO model, and 38 DEGs for OPCs and 29 DEGs for microglia in the Ip3r2 KO model (not shown), 20–50% of the astrocyte DEG level. This suggests that astrocytes are more sensitive to neuronal activity changes, as well as more profoundly affected by silencing their calcium activity. GO analysis of astrocyte DEGs in both models revealed a broad range of BPs, which go beyond terms that may be associated with synapse regulation, such as ‘response to stimulus,’ ‘cell communication,’ or ‘retrograde axonal transport.’ For example, DEGs in the VGlut2 cKO model are also enriched for BP terms related to nuclear envelope disassembly, and nitric oxide metabolism in the upregulated genes, and copper and zinc ion processing in the downregulated genes (Figure 6—figure supplement 3A, Figure 6—source data 1G). In the Ip3r2 KO model, upregulated genes are enriched for organic acid biosynthesis, while downregulated genes are enriched in histone modification-related BPs (Figure 6—figure supplement 3B, Figure 6—source data 1G).

Next, as our smFISH analysis of Gpc4, Gpc6, and Chrdl1 showed that they are regulated in opposite ways by neuronal and astrocyte activity (Figures 4 and 5), we asked whether the DEGs identified by snRNAseq show any overlap between the two models. We found a total of 19 DEGs common to both the VGlut2 cKO and Ip3r2 KO models (Figure 6H, Figure 6—source data 1D). Only one gene was commonly upregulated, and only three genes commonly downregulated (21% of total common DEGs), while the remaining 15 genes showed opposing changes, similar to Gpc4 and Chrdl1 (79% of common DEGs; Figure 6H heatmap). The same effect of opposing changes is observed when the overlap between the enriched GO terms was compared (Figure 6—figure supplement 3C). For example, the term ‘intracellular signal transduction’ (GO:0035556) is enriched in the downregulated gene list in the VGlut2 cKO model, whereas in the Ip3r2 KO model it is enriched in the upregulated gene list (Figure 6—figure supplement 3C, Figure 6—source data 1G).

Since we found that neuronal but not astrocyte activity is important for regulating layer-specific expression of Gpc4, Gpc6, and Chrdl1 (Figure 2, Figure 4 and Figure 5), we next asked if this is true for other DEGs identified in the snRNAseq dataset in each model (Figure 6I and J). To address this, we compared the list of neuronal or astrocyte activity-regulated genes with the list of layer-enriched genes from the WT (Figure 6I and J, Figure 6—source data 1E) and identified 43 common genes in the VGlut2 cKO model (70% of all cKO/WT DEGs; Figure 6I) and 101 common DEGs in the Ip3r2 model (77% of all KO/WT DEGs; Figure 6J). In contrast to our expectation of finding layer-dependent expression changes only in the VGlut2 cKO model, we found that in each model a subset of genes show a dysregulated layer expression when activity is altered. For example, expression of Mapk10 is lowest in WM astrocytes in the WT; however, in astrocytes from the VGlut2 cKO Mapk10 is upregulated in WM astrocytes compared to deep (Figure 6I). Similarly, the expression of Pde10a in the Ip3r2 KO model WT group is higher in the upper astrocyte group than other groups, while in the KO the level in that group is now low (Figure 6J). WT expression of the gene Gm47283 is higher in the upper than deep layer group, while in the VGlut2 cKO, Gm47283 is upregulated in all groups, and the relative expression between layers is maintained (Figure 6I). A similar pattern is observed in the Ip3r2 KO model for the gene Stk38I (Figure 6J). Thus, perturbation of neuronal or astrocyte activity influences the layer enrichment of some but not all genes, without causing gross rearrangement of overall astrocyte spatial identity.

Further, while the expression of Gpc4, Gpc6, and Chrdl1 shows opposite levels in VGlut2 cKO mice to those seen during normal development (Figure 4), astrocyte activity does not regulate these genes in the same way (Figure 5). To test if this is a general phenomenon beyond synapse-regulating genes, we took advantage of the bulk RNAseq dataset (Figure 1, Figure 1—source data 1A) and calculated the FC in gene expression of WT astrocytes between the time points P7 and P14 (FC P14/P7; Figure 6—figure supplement 3D and E, Figure 6—source data 1F). We compared the DEG lists of VGlut2 cKO and Ip3r2 KO models against the genes that are significantly up- or downregulated at P14 compared to P7 in the bulk data in search for common genes (Figure 6—figure supplement 3D and E). This identified 30 DEGs from the VGlut2 cKO dataset (52% of all cKO/WT DEGs, Figure 6—figure supplement 3D) and 57 genes from the Ip3r2 KO dataset (44% of all KO/WT DEGs, Figure 6—figure supplement 3E). Comparing the direction of expression changes between normal development and the VGlut2 cKO DEGs, we observed that 70% of common genes were regulated in the opposite direction (Figure 6—figure supplement 3D, middle panel), similar to Gpc4, Gpc6, and Chrdl1 (Figure 4), suggesting strong dependence of astrocyte developmental maturation on neuronal cues. On the other hand, Ip3r2 KO data showed 50% of genes displaying the same directionality as during normal development, and the other 50% showing the opposite regulation (Figure 6—figure supplement 3E). Taken together, these results show that during development astrocyte-neuron communication enacts global gene expression changes in astrocytes, going beyond synapse-regulating genes.

In the mouse cortex, astrocytes begin to be generated right before birth and populate the cortex throughout the first month of life (Farhy-Tselnicker and Allen, 2018; Ge et al., 2012). During this time, many changes are occurring in astrocytes, as well as in the synapses between neighboring neurons. We observed that the most significant change in astrocytes at the transcriptome level occurred between the first and second postnatal weeks (Figure 1). Similarly, an analysis of the synaptic proteome during development showed the largest difference between P9 and P15 (Gonzalez-Lozano et al., 2016), suggesting similar or overlapping regulatory mechanisms in both astrocytes and neurons. In addition to the transcriptomic changes, astrocyte numbers are also strongly regulated during development. Indeed, genes upregulated at P14 are uniquely enriched in GO terms related to cell proliferation and migration (Figure 1—source data 1B). Interestingly, the density of astrocytes remains fairly constant throughout development, suggesting that their expansion rate is correlated with the overall expansion of the brain tissue. The mechanisms that regulate these migration patterns are still unknown and seem to be largely unaffected by neuronal or astrocyte activity, as evident from the similar numbers of astrocytes within each cortical layer in both neuronal and astrocyte activity manipulation models tested here (Figure 4—figure supplement 2, Figure 5—figure supplement 1). Future studies will determine the factors or sets of factors that regulate the number and location of astrocytes within defined domains.

The diversity of neurons based on location, morphology, connectivity, and activity patterns has been extensively studied for decades, with multiple subtypes of excitatory and inhibitory neurons identified (Kepecs and Fishell, 2014; Migliore and Shepherd, 2005; Zeisel et al., 2015). For a long time, cortical protoplasmic astrocytes were viewed as a homogeneous population. However, recent studies looking in-depth at astrocyte heterogeneity using both bulk and single-cell sequencing approaches have shown that within the cortex astrocytes form a heterogeneous population (Batiuk et al., 2020; Bayraktar et al., 2020; Lanjakornsiripan et al., 2018). Unlike neurons, astrocytes do not fall into the six-layer categories, but rather exist on a gradient of transcriptomically separable yet overlapping groups. Indeed, our snRNAseq data shows that the biggest differences are between astrocytes of the upper and deep layer groups, while upper and mid-layer groups are the most similar (Figure 6C and D, Figure 6—figure supplement 2A and E). Nevertheless, we have identified several astrocyte population marker genes (such as Dcc or Kcnd2; Figure 6, Figure 6—figure supplement 2B, Figure 6—source data 1A and B), which are significantly enriched in one group over others. Before performing functional studies based on these genes, further characterization is required, for example, cross-referencing these genes with our bulk RNA sequencing dataset to identify astrocyte-enriched genes and performing immunohistochemistry (IHC) to determine if protein expression is also heterogeneous. These validated genes could then be used to target specific populations of astrocytes, similar to the methods employed for neurons, in order to manipulate astrocytes that interact with specific synapse types or circuits. Importantly, while blocking thalamocortical activity did alter the expression of multiple genes in astrocytes, it did not alter the layer patterning of the cells, showing that this is not a major factor in driving layer-enriched gene expression. Indeed, altering the identity of local cortical neurons by using Dab1 KO mice, in which cortical layer neurons are reversed, does alter astrocyte layer identity, suggesting a role for local cues (Lanjakornsiripan et al., 2018). Our findings further suggest that neuronal activity acts to fine-tune the level of astrocyte genes that are important for neuronal function, rather than determining their presence or absence. Functional studies are further needed to identify the precise neuronal activity patterns that govern astrocyte-neuron reciprocal communication.

Moreover, some of the synapse-regulating genes we profiled display layer-specific expression changes across development (Figure 2). We found a correlation between Chrdl1 upregulation in the upper layers with that of GLUA2, consistent with our previous findings regarding Chrdl1 regulation of GLUA2 levels (Blanco-Suarez et al., 2018). A more complex picture emerges for Gpc4, Gpc6, and GLUA1, the AMPAR subunit regulated by these factors (Allen et al., 2012). GLUA1 protein levels steadily increase across development, peaking in most layers at P7, and do not show downregulation at P14 in L1 (as was observed for Gpc4), or upregulation in deeper layers (as was shown for Gpc6). Still, changes in Gpc4 expression are contributing to the levels of GLUA1, as GLUA1 is affected in correlation with changes in Gpc4 expression in the neuronal and astrocyte activity-deficit models (Figures 4 and 5), and GLUA1 levels are reduced in the VC of Gpc4 KO mice (Farhy-Tselnicker et al., 2017). One possibility is that Gpc4 and Gpc6 may regulate GLUA1 levels at specific synapses, such as glutamatergic terminals onto interneurons in L1, or deep layer cortical neurons, making it hard to distinguish their specific effect when analyzing synapses as a group. Alternatively, they may be required to induce initial recruitment of GLUA1 to synaptic sites, but not for its maintenance, so GLUA1 levels remain stable when Gpc4 is downregulated with development.

Ever since the astrocyte-derived factors that promote synapse formation were identified, an outstanding question in the field has been, how are they regulated (Baldwin and Eroglu, 2017; Farhy-Tselnicker and Allen, 2018)? Is it astrocyte-intrinsic, or is it driven by changes in neuronal activity that occur as synapses develop? Our in vitro work, together with previously published studies, has provided evidence that neuronal activity can influence astrocyte gene expression and function at the synapse (Figure 4—figure supplement 1, Benediktsson et al., 2012; Bernardinelli et al., 2014b; Durkee and Araque, 2019; Hasel et al., 2017). However, how this occurs in the developing brain in vivo has not been systematically addressed. Here, we approached this question by perturbing the activity of thalamocortical projections (through VGlut2 knockout), with a goal of manipulating signals that have the potential to regulate astrocytic developmental gene expression under physiological conditions. Indeed, this perturbation resulted in attenuation of the developmental expression changes in astrocyte genes as well as AMPAR subunits at P14 but not at P7, suggesting a disruption in circuit maturation. A similar outcome on neuronal and astrocyte maturation was observed in studies employing visual deprivation methods (Albanese et al., 1983; Desai et al., 2002; Freire, 1978; Funahashi et al., 2013; Ishikawa et al., 2014; Ko et al., 2014; Müller, 1990; Stogsdill et al., 2017). Future strategies including manipulation of neuronal and astrocyte function using opto- or chemogenetic approaches will further elucidate the role of astrocyte-neuron interaction in circuit development and maturation.

An additional important regulation of astrocyte gene expression (including Gpc4 and Chrdl1) revealed here is by IP3R2-mediated astrocyte calcium activity, a central mechanism for astrocytic somal calcium increases, and as such, a likely candidate to modulate gene expression changes (Petravicz et al., 2014; Srinivasan et al., 2015; Yu et al., 2020). Interestingly, blunting store-mediated astrocytic calcium release (through Ip3r2 knockout) resulted in an opposite regulation of gene expression to the ones observed in the VGlut2 cKO mice and did not correspond to layer-specific developmental changes, suggesting a more global role of astrocyte calcium activity in the regulation of gene expression. This is likely due to the overall diminished signaling that occurs in this knockout model in response to multiple neurotransmitters that rely on calcium increases, and in future, systematic deletion of individual receptors will be necessary to tease out the impact of each neurotransmitter on astrocyte gene expression and synapse regulation. Notably, while we observed significant effects on Gpc4 and Chrdl1 mRNA levels, Gpc6 expression was unaltered in Ip3r2 KO VC at P14, suggesting distinct regulation of expression of the two glypican family members, and synapse-modulating genes in general. The mechanisms underlying these effects are unknown and likely involve activation of Ca²⁺-dependent cellular cascades and transcription factors. Indeed, several such genes were identified in our snRNAseq dataset including Mapk10 and Pde4b, and will form the focus of future studies. Previous work in the adult striatum found that increasing astrocytic calcium through activation of exogenous Gi-GPCRs using DREADDs induces transient synapse formation by regulating expression of thrombospondin 1, with no effect on Gpc4 or Chrdl1 (Nagai et al., 2019), suggesting specific dependence of age, region, and environmental cues on types of genes regulated by calcium activity. Furthermore, in addition to altered gene expression in astrocytes, we observed diminished levels of presynaptic markers VGLUT1 and VGLUT2 in Ip3r2 KO mice, suggesting that in addition to having intrinsic roles in regulating astrocytic gene expression IP3R2-mediated activity has a non-cell-autonomous effect on nearby neurons. These data may suggest that behavioral deficits observed in Ip3r2 KO mice could stem from the developmental alterations in synaptogenesis caused by lack of astrocyte calcium responses. Importantly, the activity-mediated regulation of astrocytic gene expression observed here appears to extend beyond synapse development as DEGs identified in the snRNAseq analysis of VGlut2 cKO and Ip3r2 KO mice span multiple cellular processes, including signal transduction, metabolism, and gene regulation (Figure 6—figure supplement 3A–C, Figure 6—source data 1G). Many of these genes may potentially be involved in regulating synapses, and further studies are required to characterize the most promising candidate genes identified here (such as Sltm, Gm47283). This will help to determine the precise role of thalamocortical and astrocytic calcium-dependent transcriptomic changes in regulating the different aspects of both astrocyte and synapse function.

In all, this study demonstrates that the correct formation of synapses and hence neuronal circuit connectivity depends on precise communication between neurons and astrocytes, where disruption in one cell type leads to disruption in the other and an overall dysregulation of synapse formation. It further shows that astrocyte regulation of synapses is intimately linked to environmental changes. Thus, an image of astrocyte identity emerges as highly plastic and dynamic cells, actively perceiving and responding to their environment. Future studies employing functional approaches such as electrophysiology, optogenetic manipulations, and behavior are needed to determine the precise nature of astrocyte plasticity and to further distinguish intrinsic and extrinsic influences on these cells, giving further insight into their function in both health and disease.

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Nicola J. Allen (nallen@salk.edu).

All animal work was approved by the Salk Institute Institutional Animal Care and Use Committee.

Samples were heated in reducing loading dye (Thermo # 39000) for 45 min at 55°C. For conditioned media, 10 µg/well was loaded; for tissue lysates, 20 µg/lane was loaded. Samples were resolved on 4–12% bis-tris or bolt gels (Invitrogen #NW04120) for 30–40 min at 150–200 V. Proteins were transferred to PVDF membranes at 100 V for 1 hr, then blocked in 1% casein (Bio-Rad #1610782) in TBS (Bioworld #105300272) blocking buffer for 1 hr at room temperature on a shaker. Primary antibodies were applied overnight at 4°C diluted in blocking buffer. The antibodies used were Rb anti-glypican 4 (Proteintech #13048-1-AP; 1:500), Rb anti-IP3R2 (a gift from Ju Chen lab, UCSD 1:1000), and Ms anti-tubulin (Thermo #MA5-16308 1:5000). The next day, membranes were washed 3 × 10 min with TBS-0.1% Tween and the appropriate secondary antibody conjugated to Alexa Fluor 680 (Molecular Probes) was applied for 2 hr at room temperature (dilution 1:10,000). Bands were visualized using the Odyssey Infrared Imager (LI-COR) and band intensity analyzed using the Image Studio software (LI-COR).

All data is presented as either mean ± s.e.m., scatter with mean ± s.e.m., or scatter with range, as indicated in each figure legend. Statistical analysis was performed using Prism software (GraphPad) unless otherwise stated in the text. Multiple group comparisons were done using one-way analysis of variance (ANOVA) with post hoc Tukey’s or Dunn’s tests. Pairwise comparisons were done by t-test. When data did not pass the normal distribution test, multiple comparisons were done by Kruskal–Wallis ANOVA on ranks and pairwise comparisons were done with the Mann–Whitney rank-sum test. p-Value ≤ 0.05 was considered statistically significant. Analysis was done blind to genotype. The sample sizes, statistical tests used, and significance are presented in each figure and figure legend.