T cells modulate the microglial response to brain ischemia

Neuroinflammation after stroke is characterized by the activation of resident microglia and the invasion of circulating leukocytes into the brain. Although lymphocytes infiltrate the brain in small number, they have been consistently demonstrated to be the most potent leukocyte subpopulation contributing to secondary inflammatory brain injury. However, the exact mechanism how this minimal number of lymphocytes can profoundly affect stroke outcome is still largely elusive. Here, using a mouse model for ischemic stroke, we demonstrated that early activation of microglia in response to stroke is differentially regulated by distinct T cell subpopulations. Acute treatment with engineered T cells overexpressing IL-10 administered into the cisterna magna after stroke induces a switch of microglial gene expression to a profile associated with pro-regenerative functions. These findings substantiate the role of T cells in stroke with large impact on the cerebral inflammatory milieu by polarizing the microglial phenotype. Targeting T cell-microglia interactions can have direct translational relevance for further development of immune-targeted therapies for stroke and other neuroinflammatory conditions. Summary The crosstalk between brain infiltrating T cells and microglia in response to stroke remains elusive. Benakis et al. report that transcriptional signature of the stroke-associated microglia is reprogrammed by distinct T cell subpopulations. Engineered T cells overexpressing IL-10 administered four hours after stroke reinitiate microglial function inducing a pro-regenerative environment.


122
The microglial reaction to stroke causes a gradual shift from the homeostatic transcriptomic profile to 123 a reactive state. In order to capture differences in the microglia transcriptome along its transition phase,  Fig. 2d). We then clustered genes into groups of correlating and anti-correlating 5 genes and investigated the activation of these gene sets along the identified trajectory path 2 in stroke 136 condition only ( Supplementary Fig. 1d-f). Gene sets which were significantly different between WT and 137 Rag1 -/mice after stroke revealed that the absence of lymphocytes significantly reduces microglial 138 genes associated with macrophage activation state (G6: Foxp1, Syk), whereas genes associated with 139 cytokine/chemokine stimulus were enriched (G3: Il1b, Tnf, Csf1, Ccl2) in Rag1 -/in comparison to WT 140 microglia (Fig 1g).

141
Figure 1. T cells influence microglia morphology and transcriptomic signature. a Morphological analysis of microglia and transcriptomic profile of sorted microglia were performed 5 d after stroke in wild-type (WT) and Rag1 -/mice. b Top, representative images of IBA1+ microglial cells in the perilesional region (900 µm distal to the infarct border, cortical layer 4). Bottom, threedimensional (3D) reconstruction of microglia in WT and Rag1 -/mice. c Morphological analysis of microglia in the peri-infarct area (ipsi) and in the contralateral hemisphere (contra) for two representative features: sphericity and branch length (µm) in WT (orange) and Rag1 -/-(blue) mice. Each individual mouse is represented in the plots by one color (4 mice/condition), each dot corresponds to one microglial cell; ns, non significant; ****, P < 0.0001. Wilcoxon rank sum test with continuity correction and Bonferroni post-hoc correction for multiple testing. d CD45+CD11b+ cells were sorted from the ipsilateral hemisphere 5 d after stroke in WT and Rag1 -/-(3 mice/condition) and RNA was isolated for single cell RNA sequencing (10x Genomics). e Uniform manifold approximation and projection 2D space (UMAP) plots of 2345 CD45+CD11b+ cells colored by conditions (left) and by 14 distinct transcriptional clusters (right and Supplementary Fig. 1). f Clustering of the microglia subset color-coded by conditions (left) and into homeostatic and reactive microglia (right). g Selected gene sets of highly correlated and anti-correlated genes based on trajectory inference analysis in stroke condition ( Supplementary Fig. 1d-f). Mean gene set activation score in WT and Rag1 -/cells, selected marker genes and top enriched gene ontology pathways associated to each gene set. Gene sets were classified by p-value (the lowest p-value at the top, asterisks (*) indicate significant difference between genotype in stroke condition) and by similar pathways, such as: pathways related to inflammation (dark blue), pathways related to DNA/RNA regulation (blue) and lipid pathways (light blue). 6 142 143 Figure 2. Analysis of microglia isolated from WT and Rag1 -/in naïve and stroke conditions using the Smart-Seq2 platform. a Correlation between the 291 pools of microglia. Hierarchical clustering identified 7 clusters, named from S1 to S7. The amount of samples is in between parentheses. The 1000 most variable genes were considered. The color represents the Pearson correlation coefficient between the samples. b-d UMAP plots showing the samples colored by clusters (b), genotype (c) and condition (d) and their associated cell distribution of genotypes and conditions among the sample clusters. e Average expression of gene clusters (defined in panel g) within each sample cluster. Scale is the average of the log normalized expression. f Box plots of the gene clusters in each conditions and genotypes. Grey boxes highlight significant difference between WT and Rag1 -/in stroke condition. *, P < 0.05; **, P < 0.01; ***, P < 0.001. Nonparametric Kruskal Wallis test followed by multiple non-parametric Wilcoxon t-tests and Bonferroni post-hoc correction for multiple testing. g Gene clusters and associated pathway enrichment analysis. Hierarchical clustering identified 7 gene analysis. Hierarchical clustering identified 7 gene clusters, named G1 to G7, only the significantly regulated gene clusters G1 and G3 between WT and Rag1 -/in stroke condition are shown. Each barplot shows the pathway enrichment analysis for the genes included in the gene clusters.
Next, we aimed to validate our scRNA-seq findings by an independent transcriptomic platform using 144 Smart-seq2 profiling of sorted pools of 50 microglia which provides cell type specificity and higher 145 mRNA capture than scRNA-seq (Picelli et al., 2014). We confirmed a predominant effect of stroke on 146 the microglial transcriptome (Fig. 2a, b), in which lymphocyte-deficiency affected the abundance of 147 specific microglia subsets (Fig. 2c-e). In order to identify the pathways specifically regulated in these 148 subsets, we performed a 'Sample-Gene cluster' correlation analysis and identified two gene sets that 149 were significantly regulated between WT and Rag1 -/mice after stroke (i.e., G1 and G3; Fig. 2f, g).

150
Specifically, we detected a decrease of gene expression associated with glial cell migration and 151 leukocyte differentiation (gene cluster G3), and an increase in neutrophil degranulation and cytokine 152 production (G1) in lymphocyte-deficient mice (Fig. 2g), supporting our results from the single cell  (Fig. 3a). The transcriptional profile of microglia induced by TREG 164 cells was more similar to vehicle treated Rag1 -/mice (named control (CT)), than microglial gene 165 expression induced by TH1, as shown in the heatmap and volcano plots of the differentially expressed 166 genes (P < 0.05 and |fold change| > 3) with 51 and 20 microglial genes regulated in TH1 or TREG conditions 167 compared to control injection, respectively (Fig. 3b, c). Gene ontology analysis of the differentially up-168 regulated genes revealed TH1-dependent pathways associated with antigen presentation, response to 169 cytokines and regulation of type I interferon whereas TREG-dependent microglial genes were associated 170 with chemotaxis (Fig. 3d). These results demonstrate the potency of T cell subpopulations to 171 differentially skew the microglial transcriptome towards distinct phenotypes previously associated with 172 different cellular functions. In particular, we found that TH1 polarized microglia toward an antigen-173 immunocompetent phenotype (Cd74) and expression of interferon response-related genes (Irf7). This 174 profile of microglial response was previously associated with a pronounced immune response during 175 the later stages of neurodegeneration (Mathys et al., 2017). In contrast, TREG cells promoted the 176 expression of chemokines/cytokines in microglia (Ccl2, Ccl7, Cxcl10), which can have either pro-177 regenerative or detrimental effects such as the regulation of leukocyte chemotaxis to the injured brain ). In addition, after experimental stroke, the TH1-mediated effects on the microglial 181 transcriptomic profile were associated with an increase of Trem2 expression, a key marker of disease-8 associated microglia in various brain disorders, in comparison to microglia primed by TREG cells (Fig.   183 3e). These transcriptomic differences in microglia related to the in vivo TH1 or TREG cell exposure was 184 also reflected by the difference in the morphology of microglia between these conditions. Microglia 185 displayed a reactive state as shown by a more spherical and less branched morphology in TH1 cell-186 injected compared to TREG-injected mice (Fig. 3f).

188 189
Figure 3. TH1 and TREG cells influence microglia gene expression after stroke. a Naïve CD4 cells were polarized in vitro to TH1 or TREG phenotype ( Supplementary Fig. 3a). One million cells (TH1 or TREG cells) or vehicle (control, CT) were injected into the cisterna magna (CM) in Rag1 -/mice 24 h after stroke induction. Microglia cells were sorted from the ipsilesional hemisphere and RNA was extracted. Gene expression analysis was performed using the Neuroinflammation Panel profiling kit on the Nanostring platform. In a second set of experiment, 100 µm coronal sections were proceeded for smFISH or microglia morphology. b Heatmap representation of microglia gene expression between conditions: control (CT; vehicle administration of PBS), TH1 or TREG. c Up-and down-regulated differentially expressed genes between either isolated microglia from TH1-(top) and TREG-(bottom) microglia from TH1-(top) and TREG-(bottom) treated Rag1 -/mice relative to control condition (microglia isolated from Rag1 -/mice treated with vehicle, genes are color-coded accordingly to a p-value < 0.05 and |fold change| > 3). d Pathway analysis was performed for the up-regulated genes in each condition using the ClueGO package from Cytoscape. e Higher amount of Trem2 mRNA puncta (red) per Cx3cr1-positive (green) in P2ry12-labelled microglia (white) in TH1-treated mice in comparison the TREG-treated mice. DAPI (blue) was used as nuclear dye. Scale bar = 10 µm. f Morphological analysis of IBA1+ microglia in the ipsilateral (900 µm distal to the infarct border, cortical layer 4) and contralateral hemisphere, as shown in the representative coronal section. Sphericity score and branch length (µm) of microglia treated with TH1 (orange) or TREG cells (green). Each individual mouse is represented in the plots by one color (3 mice/condition), each dot corresponds to one microglial cells; ns, non significant; *, P < 0.05; **, P < 0.01; ****, P < 0.0001. Wilcoxon rank sum test with continuity correction and Bonferroni post-hoc correction for multiple testing.
Interestingly, these morphological changes were not only restricted to the ipsilesional hemiphere but 190 were also observed in the contralateral hemisphere, suggesting possible brain-wide effects of 191 differentiated THELPER cells injected to the cerebrospinal fluid (CSF) compartment. In accordance, we 192 found that intra-CM injection of eGFP-labelled TH1 cells to Rag1 -/mice after stroke were primarily 193 recruited to the ischemic brain parenchyma, but were additionally also localized in border tissues 194 including the meninges, and some CM-injected cells even circulated and could be detected in the 195 spleen. (Fig. 4a, b and Supplementary Fig. 3b). Together, these findings support that polarized T cells   investigated whether eTc-IL10 treatment affected stroke outcome but did not find any difference in 205 infarct volumes between conditions (Fig. 4c). This is in accordance with the concept of early ischemic 206 lesion formation in stroke which is not being affected by the delayed immunological mechanisms 207 (Dirnagl et al., 1999). In contrast, mice receiving eTc-IL10 injection in the CM had a substantially 208 improved functional outcome at 48 h after stroke as shown by a reduced forelimb assymetry (Fig. 4d).

209
This might reflect the particular impact of inflammatory pathways and specifically cytokine secretion such as spine pruning, phagocytosis and complement activation (Fig. 4f). This anti-inflammatory effect 219 of eTc-IL10 treatment on microglia was confirmed by a reduction of Trem2 mRNA in Cx3cr1+ microglia 220 from eTc-IL10 compared to vehicle-treated mice (Fig. 4g). Since we observed a down regulation of 221 genes associated with synapse pruning (C1qa, C1qc), microglia activation (P2ry12, Cx3cr1) and  showing CD4+eGFP+ cells isolated from the brain (ipsilateral and contralateral hemispheres), meninges and spleen (the detailed gating strategy is shown in Supplementary Fig. 3b). The graph represents the percentage of eGFP+TH1 cells relative to the total number of cells injected in the CM (10 6 eGFP+TH1 cells). b Coronal section showing eGFP+TH1 cells in the meninges. Insert 1 indicates a representative photomicrograph of eGFP+TH1 cells counterstained with an eGFP-booster (magenta), cell nuclei are stained with DAPI (blue). The magnified images of white boxed area show eGFP+TH1 cells injected into the CM are located in the meninges. c Infarct volumes at 5 d after stroke in WT C57BL/6J mice treated by CM administration of either T cells secreting IL-10 (eTc-IL10, 10 6 naïve CD4+ cells transfected with a plasmid overexpressing IL-10, Supplementary Fig. 3c, d) or vehicle (aCSF) 4 h after stroke induction. d Percentage of assymetry in independent forepaw use ("0%" indicates symmetry) in mice treated with vehicle or eTc-IL10; *, P < 0.05, ANOVA with Šídák's multiple comparisons test. e Heatmap representation of ipsilateral brain gene expression between vehicle and eTc-IL10 treated mice 5 d after stroke. f Selected gene ontology annotations for the 50 genes that were up-regulated (top) and down-regulated (bottom) in the whole ipsilateral brain tissue of eTc-IL10 treated mice in comparison to vehicle treated mice. g smFISH analysis of brains from eTc-IL10 treated mice showed a reduction of Trem2 mRNA puncta per Cx3cr1-positive microglia in the peri-infarct region in comparison to vehicle treated mice; **, P < 0.01, Mann-Whitney U test.

Discussion 229
The cellular constituents of the acute neuroinflammatory response to stroke has been well

237
Here, we established a mechanistic link between T cells and microglial function and showed the distinct

258
Therefore, we aimed to take a different approach for the localized and sustained production of IL-10 at 259 the inflamed peri-lesional brain parenchyma. For this we took advantage of the potent capability of T 260 cells to be specifically recruited and accumulated to the ischemic lesion site in order to deliver IL-10 with functional recovery after stroke. Interestingly, eTc-IL10 cells did not exclusively invade the injured 265 brain, but were also located in the meningeal compartment and could additionally contribute to 266 functional recovery by resolving inflammation at these broder structures or providing IL-10 to the brain 12 parenchyma along CSF flow. This concept is in accordance with previous observations of meningeal 268 immune cell accumulation after stroke  and that meningeal T cell-derived 269 cytokines may enter the brain via CSF flow and paravascular spaces (Iliff et al., 2012).

270
An important finding in this study was the observation that IL-10 overexpression in T cells modulated

291
Taken together, we have been able to demonstrate that brain-invading T cells can specifically "fine-

489
Reads were mapped to the mouse genome assembly reference from Ensembl (mm10/GRCm38). thresholds were determined for each sample separately. Raw counts of a cell were normalized by total 498 counts neglecting highly expressed genes which constitute more than 5% of total counts in that cell.

499
Then, counts were log-transformed (log(count+1)). These processed and normalized count matrices 500 were used as input for all further analyses.

502
For the full data set and the microglia subset first a single-cell nearest-neighbor graph was computed 503 on the first 50 independent principal components. Principle components were calculated using the

511
Trajectories from homeostatic to reactive microglia were inferred with partition-based graph

521
Finally, to obtain an activation score per cell for a given gene set, cell scores were computed as 522 described by (Satija et al., 2015) and implemented in Scanpy in the tl.score_genes functionality.

523
Differential activation of gene sets between WT and Rag1 -/samples was determined by a Wilcoxon 524 rank-sum test. To identify genes differentially regulated along the inferred cellular trajectory a 525 differential gene expression test (Welch t-test with overestimated variance) between the root and end Jupyter notebooks with custom python scripts for scRNA-seq analysis will be made available in a github 534 repository upon publication (https://github.com/theislab/). 10X Genomics and Smart-seq2 data have 535 been submitted to GEO and will be made available upon publication.

537
Library preparation for Smart-seq2 platform

538
The 96-well plates containing the sorted pools were first thawed and then incubated for 3 min at 72°C

559
Processing and analyses of Smart-seq2 data 560 BCL files were demultiplexed with the bcl2fastq software from Illumina. After quality-control with 561 FastQC, reads were aligned using rnaSTAR (Dobin et al., 2013) to the GRCm38 (mm10) genome with 562 ERCC synthetic RNA added. Read counts were collected using the parameter "quantMode GeneCounts"           . Bottom, threedimensional (3D) reconstruction of microglia in WT and Rag1 -/mice. c Morphological analysis of microglia in the peri-infarct area (ipsi) and in the contralateral hemisphere (contra) for two representative features: sphericity and branch length (μm) in WT (orange) and Rag1 -/-(blue) mice. Each individual mouse is represented in the plots by one color (4 mice/condition), each dot corresponds to one microglial cell; ns, non significant; ****, P < 0.0001. Wilcoxon rank sum test with continuity correction and Bonferroni post-hoc correction for multiple testing. d CD45+CD11b+ cells were sorted from the ipsilateral hemisphere 5 d after stroke in WT and Rag1 -/-(3 mice/condition) and RNA was isolated for single cell RNA sequencing (10x Genomics). e Uniform manifold approximation and projection 2D space (UMAP) plots of 2345 CD45+CD11b+ cells colored by conditions (left) and by 14 distinct transcriptional clusters (right and Supplementary  Fig. 1). f Clustering of the microglia subset color-coded by conditions (left) and into homeostatic and reactive microglia (right). g Selected gene sets of highly correlated and anti-correlated genes based on trajectory inference analysis in stroke condition ( Supplementary Fig. 1d-f). Mean gene set activation score in WT and Rag1 -/cells, selected marker genes and top enriched gene ontology pathways associated to each gene set. Gene sets were classified by p-value (the lowest p-value at the top, asterisks (*) indicate significant difference between genotype in stroke condition) and by similar pathways, such as: pathways related to inflammation (dark blue), pathways related to DNA/RNA regulation (blue) and lipid pathways (light blue).    Gene Cluster G3 -down-regulated in stroke Rag1 -/microglial cells analysis. Hierarchical clustering identified 7 gene clusters, named G1 to G7, only the significantly regulated gene clusters G1 and G3 between WT and Rag1 -/in stroke condition are shown. Each barplot shows the pathway enrichment analysis for the genes included in the gene clusters.

Figure 2. Analysis of microglia isolated from WT
and Rag1 -/in naïve and stroke conditions using the Smart-Seq2 platform. a Correlation between the 291 pools of microglia. Hierarchical clustering identified 7 clusters, named from S1 to S7. The amount of samples is in between parentheses. The 1000 most variable genes were considered. The color represents the Pearson correlation coefficient between the samples. b-d UMAP plots showing the samples colored by clusters (b), genotype (c) and condition (d) and their associated cell distribution of genotypes and conditions among the sample clusters. e Average expression of gene clusters (defined in panel g) within each sample cluster. Scale is the average of the log normalized expression. f Box plots of the gene clusters in each conditions and genotypes. Grey boxes highlight significant difference between WT and Rag1 -/in stroke condition. *, P < 0.05; **, P < 0.01; ***, P < 0.001. Nonparametric Kruskal Wallis test followed by multiple non-parametric Wilcoxon t-tests and Bonferroni post-hoc correction for multiple testing. g Gene clusters and associated pathway enrichment  Trem2 mRNA puncta/microglia *** up-regulated pathways associated with TH1-inducing microglia gene expression up-regulated pathways associated with TREG-inducing microglia gene expression microglia from T H1 -(top) and T REG -(bottom) treated Rag1 -/mice relative to control condition (microglia isolated from Rag1 -/mice treated with vehicle, genes are color-coded accordingly to a p-value < 0.05 and |fold change| > 3). d Pathway analysis was performed for the up-regulated genes in each condition using the ClueGO package from Cytoscape. e Higher amount of Trem2 mRNA puncta (red) per Cx3cr1-positive (green) in P2ry12-labelled microglia (white) in T H1 -treated mice in comparison the T REG -treated mice. DAPI (blue) was used as nuclear dye. Scale bar = 10 μm. f Morphological analysis of IBA1+ microglia in the ipsilateral (900 μm distal to the infarct border, cortical layer 4) and contralateral hemisphere, as shown in the representative coronal section. Sphericity score and branch length (μm) of microglia treated with T H1 (orange) or T REG cells (green). Each individual mouse is represented in the plots by one color (3 mice/condition), each dot corresponds to one microglial cells; ns, non significant; *, P < 0.05; **, P < 0.01; ****, P < 0.0001. Wilcoxon rank sum test with continuity correction and Bonferroni post-hoc correction for multiple testing.   Flow cytometry analysis and whole skull-brain coronal sections of 10 6 eGFP+T H1 cells injected into the cisterna magna (CM) of Rag1 -/mice 24 h after stroke. Samples were collected 4 h after CM injection for further analysis: a Flow cytometry plots showing CD4+eGFP+ cells isolated from the brain (ipsilateral and contralateral hemispheres), meninges and spleen (the detailed gating strategy is shown in Supplementary Fig. 3b). The graph represents the percentage of eGFP+T H1 cells relative to the total number of cells injected in the CM (10 6 eGFP+T H1 cells). b Coronal section showing eGFP+T H1 cells in the meninges. Insert 1 indicates a representative photomicrograph of eGFP+T H1 cells counterstained with an eGFP-booster (magenta), cell nuclei are stained with DAPI (blue). The magnified images of white boxed area show eGFP+T H1 cells injected into the CM are located in the meninges. c Infarct volumes at 5 d after stroke in WT C57BL/6J mice treated by CM administration of either T cells secreting IL-10 (eTc-IL10, 10 6 naïve CD4+ cells transfected with a plasmid overexpressing IL-10, Supplementary Fig. 3c, d) or vehicle (aCSF) 4 h after stroke induction. d Percentage of assymetry in independent forepaw use ("0%" indicates symmetry) in mice treated with vehicle or eTc-IL10; *, P < 0.05, ANOVA with Šídák's multiple comparisons test. e Heatmap representation of ipsilateral brain gene expression between vehicle and eTc-IL10 treated mice 5 d after stroke. f Selected gene ontology annotations for the 50 genes that were up-regulated (top) and down-regulated (bottom) in the whole ipsilateral brain tissue of eTc-IL10 treated mice in comparison to vehicle treated mice. g smFISH analysis of brains from eTc-IL10 treated mice showed a reduction of Trem2 mRNA puncta per Cx3cr1-positive microglia in the peri-infarct region in comparison to vehicle treated mice; **, P < 0.01, Mann-Whitney U test. Fig. 4 -Benakis et al.