Single-cell resolution analysis of the crosstalk between chemogenically activated astrocytes and microglia

Astrocytes are the most common glial cell type in the brain, yet, it is unclear how their activation affects the transcriptome of neighboring cells. Engineered G protein-coupled receptors (GPCRs) called Designer Receptors Exclusively Activated by Designer Drugs (DREADDs) enable selective activation of specific cell types, such as astrocytes. Here, we combine activation of astrocytes in the hippocampus and cortex of healthy mice with single-cell RNA sequencing. Our data show that long-term activation of astrocytes dramatically alters the transcriptome of astrocytes and microglia. Genes that were differentially expressed in Gq-DREADD-activated astrocytes are involved in neurogenesis and low-density lipoprotein particle biology, while those in the microglia were involved in lipoprotein handling, purinergic receptor activity, and immune cell migration and chemotaxis. Furthermore, network analysis showed that Gq-DREADD-mediated activation in astrocytes resulted in an upregulation of genes involved in the GPCR signaling pathways and calcium ion homeostasis, confirming astrocyte activation. This dataset will serve as a resource for the broader neuroscience community, and our findings highlight the importance of studying transcriptomic alterations in microglia after astrocyte activation in vivo.


1
Single cell RNA sequencing identifies 15 transcriptionally different cell types 2 2 2 To determine how activation of astrocytes affects the transcriptome of astrocytes and 2 2 3 neighboring microglia, we performed scRNA seq on cells isolated from the hippocampus and 2 2 4 cortex of four-month-old C57BL/6J mice. These mice were injected with AAV5-GFAP-hM3D(Gq) 2 2 5 at 8 weeks of age ( Figure 1A) and treated with CNO or saccharin for 8 weeks. Two samples per 2 2 6 condition were sequenced with an average of 442M reads per sample, or an average of 37,000 2 2 7 reads per cell. About 56% of all reads were exonic, while 29% were intronic and only 5% were 2 2 8 intergenic. A total of 34,736 cells passed the quality control filters, 16,107 cells in the CNO 2 2 9 treated group and 18,629 cells in the saccharin group. Unsupervised clustering resulted in the 2 3 0 identification of 15 different cell clusters ( Figure 1D, Supplementary Figure 1C and 1D). The cell 2 3 1 clusters were annotated manually using marker genes that were conserved between both 2 3 2 conditions as astrocytes, microglia, neurons, endothelial cells, ependymal cells, and mixed glia 2 3 3 (Supplementary Data 1) (Arneson et al., 2018;McKenzie et al., 2018). We observed an 2 3 4 enrichment of glial cells with 96% of all cells detected annotated as either astrocytes, microglia 2 3 5 or mixed glia.  When assessing the cell distribution between the CNO treated and the saccharin groups 2 3 7 across different cell types, we noticed significant differences in the percentage of cells in the 2 3 8 mixed glia I (30.66% in the saccharin group vs. 45.33% in the CNO group, p ≤ 0.0001), mixed 2 3 9 glia II (29.62% in the saccharin group vs. 14.87% in the CNO group, p ≤ 0.0001), microglia I 2 4 0 (12.37% in the saccharin group vs. 16.74% in the CNO group, p ≤ 0.05) and microglia II clusters 2 4 1 (11.40% in the saccharin group vs. 6.61% in the CNO group, p ≤ 0.05, Figure 1E, 2 4 2 Supplementary Figure 1D and Supplementary Table 1). A smaller difference was observed in 2 4 3 the astrocyte cluster (7.42% in the saccharin group vs. 5.31% in the CNO group, Figure 1E, 2 4 4 Supplementary Figure 1D and Supplementary Table 1).

4 5
Chronic activation of astrocytes results in a transcriptionally different astrocyte type 2 4 6 Since Gq-DREADD was only expressed in astrocytes, we analyzed the effect of long-term Gq-2 4 7 DREADD activation by comparing the genes that were differentially expressed between the 2 4 8 saccharin group and the CNO group in the astrocyte cluster. We identified 396 genes (p < 0.05) 2 4 9 that were either up-or downregulated in the CNO group compared to the saccharin group 2 5 0 ( Figure 2A). Gene enrichment and functional annotation analysis showed that transcripts that 2 5 1 were upregulated in the CNO group are involved in adenylate cyclase activator activity and the 2 5 2 regulation of cell motility and migration, while the downregulated genes were involved in gene 2 5 3 expression and translation (Supplementary Data 1). Network analysis was performed using 2 5 4 MetaCore TM . We found that, among the genes that were upregulated after activation of Gq-2 5 5 DREADD with CNO, 16 of them were in a 50-gene network represented by CMKLR1, 2 5 6 SPARCL1, Rich1, Transcobalamin II, OLFML3 ( Figure 2B and 2C). The top processes linked to 2 5 7 this network were GPCR signaling pathway, ionotropic glutamate receptor signaling pathway, 2 5 8 adenylate cyclase inhibiting GPCR signaling pathway, cellular calcium ion homeostasis, and 2 5 9 calcium ion homeostasis ( Figure 2B). Since these cellular processes are pathways regulated by   Figure 2C shows that the genes involved in the network shown in Figure   To investigate how long-term astrocyte activation alters their gene expression in more 2 7 0 detail, we re-clustered the astrocyte cluster ( Figure 2D). Sub-clustering of the astrocytes 2 7 1 showed that cell types observed in the saccharin group (clusters 0, 2 and 3) were nearly absent 2 7 2 from the CNO group, and vice versa (clusters 1 and 4, Figure 2D). This was confirmed by   Table 2). Genes that were differentially expressed between the different 2 7 7 astrocytic sub-clusters, as well as gene enrichment and functional annotation analysis of these 2 7 8 genes showed the presence of transcriptionally and functionally different astrocytes ( Figure 2F).

7 9
The clusters in the saccharin group (cluster 0, 2 and 3) are enriched for genes involved in 2 8 0 regulation of neuronal death, neuroinflammatory response and regulation of synaptic 2 8 1 organization, while those of the CNO clusters (cluster 1 and 4) are enriched for genes involved 2 8 2 in low-density lipoprotein particles, central nervous system development, gliogenesis, and glial  Even though Gq-DREADD was only expressed in astrocytes, we observed a 35% increase and 2 8 6 a 42% decrease in the number of cells in the microglia I and II clusters, respectively, after 2 8 7 astrocyte activation with CNO ( Figure 1E and Supplementary Table 1). In addition, it is known 2 8 8 that the increase of intracellular Ca 2+ after astrocytic activation results in calcium waves that can 2 8 9 Differentially expressed genes in these two clusters were involved in G protein-coupled 3 1 0 purinergic nucleotide receptor activity, the innate immune system, antigen processing and 3 1 1 presentation, and neuronal cell death (Supplementary Data 3). In addition, clusters 0, 3 and 5 3 1 2 were more common in the saccharin group, while clusters 1 and 2 were more pronounced in the 3 1 3 CNO group (Figure 3E and Supplementary Table 3). Clusters 1 and 2 showed an enrichment for 3 1 4 functional terms involving ribosomal processes, while no gene ontology analysis was performed 3 1 5 for clusters 0, 3 and 5 due to the low number of differentially expressed genes (Supplementary 3 1 6 Data 3). Gene expression analysis showed a relative upregulation of homeostatic microglial 3 1 7 genes in clusters 1, 4 and 5, while they were downregulated in clusters 0, 2, 3 and 6 ( Figure  3 1 8 3F).

1 9
Re-clustering of the microglia II cell cluster resulted in the identification of seven sub-3 2 0 clusters. Three sub-clusters were mainly observed in the saccharin group (clusters 0, 2 and 3), 3 2 1 two were more common in the CNO group (clusters 1 and 4), and two others (clusters 5 and 6) 3 2 2 were present in both groups ( Figure 3G, 3H and Supplementary Table 3). Differentially 3 2 3 expressed genes in clusters 0, 2 and 3 were involved in G protein-coupled purinergic nucleotide 3 2 4 receptor activity, the innate immune system and cell death (Supplementary Data 4). Clusters 1 3 2 5 and 4, on the other hand, were enriched for transcripts involved in the innate immune system, 3 2 6 lysosomal functions, synaptic pruning, respiratory chain complex, and response to lipoprotein 3 2 7 particles (Supplementary Data 4). A strong downregulation of homeostatic microglial genes was 3 2 8 observed in cluster 6, while an upregulation was observed in cluster 5 ( Figure 3I).

2 9
In the microglia III cluster, gene enrichment and functional analysis of the differentially 3 3 0 expressed genes showed an enrichment for transcripts involved in the immune system and cell 3 3 1 killing, while a downregulation of transcripts involved in monocyte differentiation was observed 3 3 2 (Supplementary Figure 3A and Supplementary Data 1).
Re-clustering of the microglia III cluster resulted in five sub-clusters ( Figure 4A).

4
Although all of them were observed in both groups, cluster 0 was six times more prominent in 3 3 5 the saccharin group (56.08%) compared to the CNO group (8.95%), and cluster 2 was nine 3 3 6 times more pronounced in the CNO group (36.24%) than in the saccharin group (3.97%, Figure  3 3 7 4A, 4B and Supplementary Table 3). Gene ontology and functional analysis of the differentially 3 3 8 expressed genes pointed towards similar functions for all cells in the different clusters. The 3 3 9 differentially expressed genes in clusters 0 and 2 were involved in G protein-coupled receptor 3 4 0 binding and the migration of immune cells like neutrophils, granulocytes, and macrophages 3 4 1 (Supplementary Data 5). Again, we observed a strong downregulation of homeostatic genes in 3 4 2 one cluster, cluster 4, compared to the other clusters ( Figure 4E).

4 3
Interestingly, the genes that were downregulated in the microglia IV cluster after 3 4 4 astrocyte activation were involved in G protein-coupled purinergic nucleotide receptor activity 3 4 5 and ATP synthesis coupled electron transport. The genes that were upregulated in the CNO 3 4 6 group compared to the saccharin group were enriched for terms involving ribosomal processes 3 4 7 (Supplementary Figure 3B and Supplementary Data 1).

4 9
Clusters 2 and 3 were only observed in the saccharin group, and thus, were completely absent 3 5 0 from the CNO group (Figure 4C, 4D and Supplementary Table 3). No gene ontology and 3 5 1 functional analysis was performed for the differentially expressed genes in cluster 2 due to the 3 5 2 low number of differentially expressed genes. However, the differentially expressed genes in 3 5 3 cluster 3 were shown to be involved in gene expression, translation and ribosomal processes 3 5 4 (Supplementary Data 6). Cluster 0 (14.67% in the saccharin group vs. 45.09% in the CNO 3 5 5 group) and cluster 1 (0.80% in the saccharin group vs. 48.61% in the CNO group) were more  IV, we observed again that one cluster, cluster 0, showed a downregulation of homeostatic gene 3 6 0 expression compared to the other four clusters ( Figure 4E).

6 1
Lastly, we noticed that genes involved in cell surface receptor signaling pathway and 3 6 2 response to lipids were upregulated in the microglia V cluster. The downregulated genes were 3 6 3 enriched for gene ontology terms including ribosomal processes (Supplementary Figure 3C  The re-clustering of the fifth microglial cluster resulted in four different cell clusters 3 6 6 (Supplementary Figure 3D). Clusters 0 and 2, showed a 1.7-fold increase and a 50% decrease 3 6 7 in cell numbers between the saccharin and the CNO groups, respectively (Supplementary 3 6 8 Figure 3E and Supplementary Table 3). The gene enrichment and functional annotation of the 3 6 9 differentially expressed genes for cluster 0 showed that these cells were responsive to 3 7 0 cytokines, while cluster 2 showed an involvement in major histocompatibility complex (MHC) 3 7 1 binding. In addition, due to the high number of ribosomal genes that were differentially 3 7 2 expressed, both clusters also showed an involvement in ribosomal processes (Supplementary 3 7 3 Data 7). Same as with microglia I-IV, we noticed that a big difference in expression levels of 3 7 4 homeostatic microglial genes. Cluster 0, 2 and 3 showed relatively low expression of 3 7 5 homeostatic genes, while cluster 1 showed an increased expression of these genes compared 3 7 6 to the other three clusters (Supplementary Figure 2F).

7 7
Re-clustering of the mixed glia cell clusters showed the presence of microglia-like, 3 7 8 astrocyte-like and oligodendrocyte-like cells 3 7 9 We annotated two cell clusters as 'Mixed Glia' since it was not clear from their conserved 3 8 0 marker genes whether these cells should be annotated as astrocytes (e.g. Cpe and Apoe), 3 8 1 microglia (e.g. Actb) or oligodendrocytes (e.g. Dpysl2; Figure 5A   Differentially expressed genes in the mixed glia I cluster showed that chronic activation 3 8 5 of astrocytes by the administration of CNO downregulated genes involved in apoptotic and ATP 3 8 6 metabolic processes ( Figure 5C and Supplementary Data 1). Genes that were upregulated after 3 8 7 CNO treatment are involved in immune cell differentiation and migration, cytokine production, 3 8 8 and response to lipids ( Figure 5C and Supplementary Data 1). In the mixed glia II cluster, gene 3 8 9 enrichment and functional annotation of differentially expressed genes demonstrated that 3 9 0 downregulated genes were also involved in apoptotic and ATP metabolic process processes 3 9 1 while the genes that were upregulated after Gq-DREADD-mediated activation of astrocytes 3 9 2 were mainly involved in gene expression and immune-related processes ( Figure 5F and 3 9 3 Supplementary Data 1). Based on gene ontology analysis, we expect the presence of both 3 9 4 astrocytes and microglia in the mixed glial clusters. 3 9 5 To determine which type of cell might be responsible for the observed effects, we re-3 9 6 clustered both mixed glia clusters. Re-clustering of the mixed glia I cluster resulted in eight 3 9 7 different sub-clusters while the re-clustering of the mixed glia II cluster resulted in six sub-3 9 8 clusters ( Figure 5D and 5G). Cells belonging to cluster 2 in the mixed glia I group were more Re-clustering of the mixed glia II cluster showed more dramatic effects of Gq-DREADD-4 1 0 mediated activation of astrocytes with CNO on the different sub-clusters than was observed in 4 1 1 the mixed glia I cluster. Cluster 0 (56.90% in the saccharin group vs. 11.66% in the CNO group), 4 1 2 cluster 1 (37.12% in the saccharin group vs. 0.87% in the CNO group) and cluster 4 (3.13% in 4 1 3 the saccharin group vs. 0.62% in the CNO group) disappeared almost entirely after CNO 4 1 4 treatment while cluster 2 (2.69% in the saccharin group vs. 66.26% in the CNO group) and 4 1 5 cluster 3 (0.05% in the saccharin group vs. 20.45% in the CNO group) appeared after CNO To the best of our knowledge, this is the first study to show transcriptional changes in glia that 4 2 5 occur after long-term activation of astrocytes with the Gq-DREADD in an in vivo model.

2 6
Previous studies have expressed the Gq-DREADD specifically in astrocytes using the GFAP 4 2 7 promotor and showed an increase in intracellular Ca 2+ after activation with CNO (Agulhon et al.,  C57BL/6J mice to determine the effect of long-term activation of astrocytes through Gq-4 3 2 DREADD. Network analysis of the genes that were upregulated after Gq-DREADD activation 4 3 3 with CNO in astrocytes showed an involvement of these genes in the GPCR signaling pathway 4 3 4 and calcium homeostasis ( Figure 2B and 2C). These findings confirmed the transcriptionally 4 3 5 activated state of the astrocytes after long-term activation of Gq-DREADD with CNO. 4 3 6 Furthermore, our results demonstrate the presence of transcriptionally different astrocyte 4 3 7 populations after Gq-DREADD activation with CNO compared to the control group. Although the 4 3 8 up-or downregulation of differentially expressed genes was relatively low, we could clearly 4 3 9 observe transcriptionally different astrocytic sub-clusters ( Figure 2F). While some clusters were 4 4 0 predominantly present in the control group (clusters 0, 2 and 3, Figure 2D and 2E), others were 4 4 1 mainly observed after activation with CNO (clusters 1 and 4, Figure 2D and 2E). We also 4 4 2 observed that our astrocyte population in the CNO group did not show the transcriptional profile 4 4 3 of reactive astrogliosis that is typically seen in neurodegenerative diseases (Supplementary 4 4 4 Figure 2D) (Kraft et al., 2013). Furthermore, we also noticed that these Gq-DREADD activated 4 4 5 astrocytes did not show the transcriptional profile of other, recently identified astrocyte sub-4 4 6 types such as PAN reactive, A1-specific, A2-specific, disease-associated astrocytes, or 4 4 7 astrocytes seen in acute injury and chronic neurodegeneration (Supplementary Figure 2A-C) 4 4 8 (Das et al., 2020;Habib et al., 2020;Liddelow et al., 2017). Therefore, we hypothesize that 4 4 9 long-term activation of astrocytes changes their transcriptional profile to a, thus far than in the CNO group (clusters 0, 2 and 3, Figure 2E) were involved in the regulation of several 4 5 7 processes such as transport, neuronal cell death, and synapse structure and activity 4 5 8 (Supplementary Data 2). The cell clusters that were more abundant in the CNO group (clusters 4 5 9 1 and 4, Figure 2E) were important for low-density lipoprotein homeostasis and central nervous 4 6 0 system development. Strikingly, gene ontology showed that the differentially expressed genes 4 6 1 from the CNO sub-clusters were enriched for glial cell development, astrocyte differentiation and 4 6 2 positive regulation of neuronal apoptotic processes (Supplementary Data 2). Furthermore, a 4 6 3 recent study (Batiuk et al., 2020) determined the transcriptomic profile of astrocytes in the 4 6 4 hippocampus and cortex of C57BL/6J mice at post-natal day 56 and showed the existence of 4 6 5 five transcriptionally different astrocytic clusters. We looked at the transcriptomic profile of these 4 6 6 five astrocyte types in our data and were only able to link our astrocyte sub-cluster 4 to their 4 6 7 astrocyte cluster 5 (Supplementary Figure 2E). These astrocytes seem to be enriched in cortical 4 6 8 layers 2/3 and 5, and in the dentate gyrus of the hippocampus (Batiuk et al., 2020). While the 4 6 9 astrocytes enriched in the saccharin group show a homeostatic phenotype (Barres, 2008), the 4 7 0 phenotype of the astrocytes enriched in the CNO group is less clear. Previous research has 4 7 1 shown that the transcriptomic profile of astrocytes varies in response to different "insults", e.g. 4 7 2 long-term activation (Anderson et al., 2014;Hamby et al., 2012). Therefore, additional research 4 7 3 is needed to elucidate the exact molecular phenotype of the Gq-DREADD activated-astrocytes. 4 7 4 Furthermore, the transcriptome of neighboring microglia was also dramatically affected 4 7 5 by the long-term activation of astrocytes (Figures 3 and 4). We observed microglial clusters that 4 7 6 were absent in the control group and dominating in the CNO group ( Figures 3D, 3G, 4A and 4 7 7 4C). We also observed that the transcriptomic changes that occurred in microglia after 4 7 8 astrocytic Gq-DREADD-mediated activation were different than those in control microglia 4 7 9 ( Supplementary Data 3-7). After long-term activation of astrocytes, the microglial transcriptome 4 8 0 suggests a role in lipoprotein particle processes, and migration and chemotaxis of immune cells. Additionally, we observed that certain microglial sub-clusters showed an increased 4 8 7 expression of genes that are known to characterize homeostatic microglia, while other sub-4 8 8 clusters showed a decreased expression of these genes ( Figure 3F, 3I, 4E and Supplementary 4 8 9 Figure 3F). To determine the transcriptomic profile of these microglial sub-cluster, we 4 9 0 determined whether these clusters showed the expression profile of disease-associated 4 9 1 microglia (DAM). This subgroup of microglia were identified while studying the immune cells at  Figure 4), resulting in the hypothesis that DAM cells may not always be linked 4 9 5 to amyloid pathology and thus are not always disease-associated. Our data hypothesizes that 4 9 6 DAM genes are induced following astrocyte activation. This would implicate that the microglial 4 9 7 response is secondary to the astrocytic response in AD, which can have major implications for 4 9 8 AD research. However, further research is needed to elucidate what causes the induction of 4 9 9 DAM genes after long-term activation of astrocytes in a healthy mouse model. 5 0 0 We further observed a 48% increase and 50% decrease in cell numbers in the two 5 0 1 largest cell clusters, mixed glia I and mixed glia II respectively, after Gq-DREADD-mediated 5 0 2 activation of astrocytes. All cell clusters were annotated manually based on marker genes that 5 0 3 were conserved between the saccharin group and the CNO group and gene lists present in 5 0 4 literature (Arneson et al., 2018;McKenzie et al., 2018). Still, about 60% of the cells in our 5 0 5 analysis were annotated as "Mixed Glia". A possible explanation for this phenomenon is that we 5 0 6 analyzed the cortex and hippocampus together, while it is well established that both astrocytes 5 0 7  al., 2020). Gene ontology analysis of differentially expressed genes in both mixed glial clusters 5 0 9 pointed towards the presence of mainly microglia. The dominance of microglia genes was 5 1 0 confirmed when we looked at the expression of cell-type-specific marker genes for microglia 5 1 1 (e.g. Cx3cr1,P2ry12,Tmem119,Aif1,Olfml3,Ccl3,Itgam), astrocytes (e.g. Aldh1l1, Atp1b2, 5 1 2 Aqp4, Sox9, Slc4a4, Mlc1) and oligodendrocytes (e.g., Nfasc, Kndc1; Supplementary Figure 5) 5 1 3 (Pan et al., 2020). These difficulties in annotating cell clusters in brain shows the urgency for 5 1 4 reliable automatic annotation tools. 5 1 5 Previous research also showed that activation of astrocytes in the mouse hippocampus 5 1 6 using the Gq-DREADD influences the excitability of neurons. The increased intracellular Ca 2+ 5 1 7 levels stimulated glutamate release which resulted in activation of neuronal NMDA receptors 5 1 8 (Durkee et al., 2019). Since less than 1% of our total cell population subsists of neuronal cells, 5 1 9 we were unable to confirm this neuronal activation at the transcriptomic level. Additional 5 2 0 research with a different cell isolation protocol is needed to study the effect of long-term 5 2 1 activation of astrocytes on the transcriptomic profile of neurons.

2 2
Lastly, our single cell dissociation protocol shows a dramatic enrichment of glial cells 5 2 3 and a relative depletion of neurons. Possible explanations for this glial enrichment are (1) the 5 2 4 fact that we dissociated brain tissue of four months-old mice. Previous studies mostly perform 5 2 5 scRNA seq in younger mice, while single nucleus RNA sequencing is more common in adult 5 2 6 mice (Hook et al., 2018;Loo et al., 2019;Zeisel et al., 2015;Zhou et al., 2020). The central 5 2 7 nervous system of older mice is more complex than that of younger mice, making it harder to 5 2 8 dissociate it into single cells, and as a result some cell types can get over-or underrepresented 5 2 9 (Darmanis et al., 2015;Grindberg et al., 2013;Krishnaswami et al., 2016;Lake et al., 2016;   our dissociation protocol at 4°C instead of 37°C. Research has also shown that the enzymatic 5 3 3 dissociation of brain tissue at 37°C results in an upregulation of inflammatory genes in microglia.

3 4
The phenomenon was not observed with a mechanical dissociation at 4°C (Bennett et al., Heatmap showing the differential expression of homeostatic microglia genes in microglia 6 2 5 clusters III and IV. 6 2 6