Single-cell analysis identifies Ifi27l2a as a novel gene regulator of microglial inflammation in the context of aging and stroke

Microglia are key mediators of inflammatory responses within the brain, as they regulate pro-inflammatory responses while also limiting neuroinflammation via reparative phagocytosis. Thus, identifying genes that modulate microglial function may reveal novel therapeutic interventions for promoting better outcomes in diseases featuring extensive inflammation, such as stroke. To facilitate identification of potential mediators of inflammation, we performed single-cell RNA sequencing of aged mouse brains following stroke and found that Ifi27l2a was significantly up-regulated, particularly in microglia. The increased Ifi27l2a expression was further validated in microglial culture, stroke models with microglial depletion, and human autopsy samples. Ifi27l2a is known to be induced by interferons for viral host defense, however the role of Ifi27l2a in neurodegeneration is unknown. In vitro studies in cultured microglia demonstrated that Ifi27l2a overexpression causes neuroinflammation via reactive oxygen species. Interestingly, hemizygous deletion of Ifi27l2a significantly reduced gliosis in the thalamus following stroke, while also reducing neuroinflammation, indicating Ifi27l2a gene dosage is a critical mediator of neuroinflammation in ischemic stroke. Collectively, this study demonstrates that a novel gene, Ifi27l2a, regulates microglial function and neuroinflammation in the aged brain and following stroke. These findings suggest that Ifi27l2a may be a novel target for conferring cerebral protection post-stroke.

Both interferons (IFN) and certain types of viral infection upregulate I 27l2a expression in the brain 7,8 . In non-glial cells I 27l2a protein enhances in ammation by blocking the action of nuclear receptors (NR4A) that normally act to promote expression of anti-in ammatory genes 9,10 . In the current study, we further demonstrate that reducing I 27l2a expression provides signi cant reduction of neuroin ammation and brain infarct following ischemic stroke. Together, these studies provide compelling new evidence that targeting I 27l2a expression or function may mitigate brain injury and in ammation following ischemic stroke.
Results I 27l2a is highly upregulated after stroke in MG and aging signi cantly enhances this upregulation. To de ne the transcriptional signature across multiple cell types in the post-stroke brain, we performed scRNA-seq of young (3-month-old) and aged (20-month-old) male and female C57BL/6J mice subjected to permanent distal middle cerebral artery occlusion (pdMCAO) or sham surgeries ( Table 1). The pdMCAO stroke model was used since it produces both primary injury (cortical infarct) and secondary injury in the thalamus 2-weeks post-stroke 11 . Since the cortex and thalamus also feature clear increases in microgliosis and astrogliosis following stroke, we included a brain region containing both peri-infarct cortex and thalamus for scRNA-seq study 11 .
We evaluated differential gene expression (DGE) patterns, as an initial approach for determining how the molecular signature of cells within the young and aged brain change post stroke. To measure the effect of aging in stroked brains, we rst integrated young stroke (AGGR2) and aged stroke (AGGR4) data into a single analysis (Seurat package 12 ) for analysis. Then, a single integrated analysis (data integration, PCA, UMAP and clustering, and DGE) was performed. Visualization of this merged dataset of 21,092 cells from the aged and young stroke mouse brain through dimension reduction by uniform manifold approximation and projection (UMAP) identi ed eight clusters of unique cell types based on gene expression differences. Identities were assigned to each of the eight clusters using the expression of conserved cell type markers, including microglia (MG, n = 7180) (Trem2), oligodendrocytes (Oligo, n = 5149) (Plp1), endothelial cells (EC, n = 3943) (Cldn5), astrocytes (Astro, n = 1698) (Aldoc), lymphocytes (Lym, n = 1495) (Plac8), epithelial cells (Epi, n = 1192) (1500015O10Rik), vascular leptomeningeal cells (VLMC, n = 127) (Dcn) and vascular endothelial cells, venous (VECV, n = 89) (Pglyrp1) (Fig. 1a-b). Notably, the proportion of MG and Lym clusters were highly increased in the aged stroke brain, whereas the oligodendrocytes were reduced (Fig. 1c). These ndings are consistent with more extensive white matter injury in the aged stroke brain and correlate with increased MG-mediated neuroin ammation and lymphocyte in ltration. As we and others have demonstrated that MG are highly sensitive to in ammation and ischemic stress and act to regulate innate immunity in brains 3,13 , we focused our subsequent analysis on transcriptional changes within MG.
Interestingly, our unbiased analyses of 21,092 cells (combined from young and aged stroke brains) showed that I 27l2a was the most highly upregulated gene in MG clusters in aged stroke, compared to young stroke ( Table 2). The top ve genes that were signi cantly upregulated in MG in the aged stroke brain included MG related genes (Lgals3, Lyz2, Lgals3bp) and another interferon-stimulated gene (ISG), I tm3. Dot plots compared the expression level and percent of cells expressing the top ve genes upregulated in aged stroke versus young stroke (Fig. 1d). Within the MG cluster, there was a notable increase in the percentage of cells expressing I 27l2a between aged stroke and young stroke (62.6% vs 29.3%) (Fig. 1d). Total normalized expression of I 27l2a from all cells showed increased I 27l2a expression in cells of aged stroke, compared to young stroke (Fig. 1e). I 27l2a was more highly upregulated in MG of aged stroke brain, suggesting aging may act synergistically with ischemic stroke to promote I 27l2a expression in MG (Fig. 1f). While most of the I 27l2a-expressing cells belonged to the MG cluster, I 27l2a expression was also detected in Lym and VLMC populations (Fig. 1g). The VLMC showed increased I 27l2a expression with aged stroke, whereas stroke-induced I 27l2a expression in Lym was not markedly altered by aging. Other MG markers, such as Lgals3, I tm3 and Lgals3bp, were also upregulated in MG and other cells following stroke (Fig. 1h-j). In contrast, there was no synergistic effect of aging with stroke on C1qa expression, a MG marker gene (data not shown). As expected, a known marker of activated MG, Cst7, was increased in aged stroke compared to young stroke brain (Extended data Fig. 1a), con rming that MG were more highly activated in aged stroke brains than in young stroke brains. In addition, we found signi cant upregulation of Apoe and Lyz2, while Aif1 level appeared only slightly increased in MG in aged stroke compared to young stroke (Extended Data Fig. 1bd). Taken together, the scRNA-seq data suggest an age-dependent upregulation of I 27l2a, which occurs predominantly in MG after stroke.
scRNA-seq revealed that aging itself is su cient to increase I 27l2a transcripts in MG. Following our nding that I 27l2a is upregulation following stroke in an age-dependent manner, we next sought to determine if aging alone impacts I 27l2a expression. Thus, we compared young and aged sham brains, integrating the young and aged sham operated samples (young sham -AGGR1, aged sham -AGGR3). Eight clusters were identi ed (Extended data Fig. 2a and 2b), including oligodendrocytes (Oligo) (Plp1), MG (C1qa), EC (Cldn5), astrocytes (Astro) (Gpr37l1), epithelial cells (Epi) (Ttr), lymphocytes (Lym) (Nkg7), vascular smooth muscle cells, arterial (VSMCA) (Des), and B cells (CD79a). To determine how aging affects the transcriptional landscape in MGs, we compared the expression and percent of cells expressing the previously identi ed top ve MG genes (I 27l2a, Lgals3, I tm3, Lyz2, and Lgals3bp) in young and aged sham brains. All 5 genes that were upregulated in MG from aged stroke brain were also increased by aging alone (Extended Fig. 2c). Notably, I 27l2a transcript levels signi cantly increased with aging, as did Rps27rt. C1qa, on the other hand, was not dramatically altered between young and aged sham animals (Extended data Fig. 2d). We also con rmed the increased expression of I 27l2a in MG, Lym and B cell clusters in aged brains (Extended data Fig. 3a). The violin plots revealed modest upregulation of I 27l2a in MG in aged brain compared to young brains. Interestingly, we found signi cant age-dependent upregulation of ribosomal protein genes such as Rpl35, Rps27rt, and Rps28. This agedependent upregulation of Rps27rt in all clusters, including MG and Lym (Extended data Fig. 3b), suggested aging-mediated changes in ribosomal complex composition in MG. Expression of two other genes associated with activated MG (Aif1 and Il-1b) were also modestly increased with aging (Extended data Fig. 3c-d). We repeated the same analyses comparing sham to stroke for aged (Extended data Fig. 4, 5, 6) and young cohorts (Extended data Fig. 7, 8). Together, these data suggest a synergistic effect of aging and stroke on I 27l2a expression.
Disease-associated microglia (DAM) are present in the aged brain, and signi cantly increased following stroke. DAM are a recently discovered sub-population of MG found in the brains of various neurodegenerative diseases, such as AD, PD and ALS (REFs). We asked if DAMs are increased in the aged stroke brain. A recent study reported that homeostatic genes, such as C1qa, Ctss, Hexb, and Csf1r are not upregulated during the transition of MG into DAM, whereas other MG related genes (Spp1, Cst7, Lpl, and Itgax) are highly upregulated in DAM 14 . To determine whether a DAM-like MG subpopulation is increased in the aged or stroke brain, we compared the number and relative percentage of DAM in sham and stroke.
The DAM subtype was de ned by the elevated expression of Aif1, Spp1, Cst7, and Lpl among all MG ( lter applied: Aif1 high and Spp1 high and Cst7 high and Lpl high , threshold by count). As expected, we did not detect DAM (0.0%) in young sham brains. However, we a small number of DAM in aged sham samples (0.9%) (Extended data Fig. 9a) shows that stroke increased the percent of DAM in both the young and aged brain (10.2% in young stroke vs 17.9% in aged stroke). These data show that MGs are converted to DAMs during aging, but are DAMs are signi cantly increased following stroke. I 27l2a is inversely correlated with DAM cell phenotype. We further analyzed our scRNA-seq data (aged sham and aged stroke) to determine whether I 27l2a plays a role in microglial activity, particularly in DAM. First, we examined whether I 27l2a expression correlated with expression of DAM-related markers, such as Lpl, Spp1, Cst7 and Itgax. We subset MG into 4 different sub-clusters based on their levels of I 27l2a expression (normalized I 27l2a expression: 0.3-0.99, 1-1.99, 2-2.99, 3+). We consistently observed a negative correlation between I 27l2a and DAM related gene expression (Lpl, Spp1, Cst7) (Extended data Fig. 9b).
Furthermore, segregating MG into either an I 27l2a "high" or "low" expressing cells followed by correlation analysis with known DAM genes and MG homoeostatic genes revealed a negative correlation between I 27l2a expression and phagocytosis/DAM related genes (Lpl, Spp1, Itgax, Cst7, and Tyrobp), but not homeostatic genes or other MG genes (Extended data Fig. 9c). Comparing the expression of DAM genes and homeostatic genes between "low" and "high" I 27l2a MG subpopulations revealed that DAMrelated transcripts were signi cantly reduced in I 27l2a "high" MG. However, homeostatic genes (Aif1, C1qc, Hexb, and Gapdh) were either not changed or slightly increased (Extended data Fig. 9c). Furthermore, MG genes which are down-regulated in DAMs (Csfr1, Olfml3, Trem119, and P2ry13) were increased in I 27l2a "high" MG (Extended data Fig. 9c). Together, these data show a strong negative correlation between I 27l2a expression and a mature DAM transcriptional signature in MG.
Regional I 27l2a expression with natural aging. While scRNA-seq showed that I 27l2a transcripts are enriched in MG and other cell types (e.g. Lym, VLMC) in both young and aged stroke brains, we lacked any data on whether there was a regional basis for these changes within the brain (e.g. within the primary injury in the cortex or within the secondary injury region occurring within the thalamus). The cortex and thalamus were extracted from the brains of naïve young (3 months, n = 4) and aged male mice (18-20 months, n = 4) to determine if there were regional differences in I 27l2a expression. Notably, I 27l2a mRNA was signi cantly upregulated in the aged thalamus (Fig. 2a, p < 0.05). I 27l2a expression also approached upregulation in the cortex in aged brains (p = 0.23). These ndings agree with our earlier scRNA-seq nding and suggest normal aging increases I 27l2a expression in the brain. In addition, MG related genes Il-1b, Cst7, and Tyrobp were markedly increased in either the thalamus or cortex of aged brains, further supporting an age-dependent increase in MG activation ( Fig. 2b-d). Transcripts for C1qb and Lpl, two genes which are known to be associated with microglia phagocytosis, were indistinguishable between the two stages ( Fig. 2e-f). These data indicate that I 27l2a is induced along with other genes associated with proin ammatory MG phenotype in the aged brain.
Regional and temporal expression of I 27l2a in aged stroke brain. To provide regional and temporal expression of I 27l2a and other MG-related genes following stroke, we analyzed young and aged thalamus and cortex by qRT-PCR at 3 and 14 days post-stroke. As expected, I 27l2a was signi cantly elevated at three days (cortex) and two weeks (cortex and thalamus) after stroke, compared to sham ( Fig. 2i). We evaluated two other genes associated with MG activation (Cst7) and reparative phagocytosis (Tyrobp), and which were found to be elevated in our scRNA-seq analysis. Both Cst7 and Tyrobp were increased in cortex and thalamus by two weeks post-stroke, but not by 3 days (Fig. 2g-h). These ndings suggest that I 27l2a expression is associated with the earlier phase of MG activation following stroke. The delayed expression in thalamus re ects the slower progression of the secondary injury mechanism.
To provide spatial context to I 27l2a expression at the single-cell level, we pro led I 27l2a mRNA transcripts on mouse brain sections using single-molecule in situ hybridization (RNAscope). Probing for I 27l2a in aged sham and stroke brains revealed elevated transcripts in the peri-infarct area at 2 weeks post-stroke compared with sham-operated controls ( Fig. 2j-l). Combining RNAscope for I 27l2a with immunostaining for Iba1 con rmed that the majority of I 27l2a transcript is present in activated MG in the peri-infarct region of the aged brain (Fig. 2m).
MG represent the predominant source of I 27l2a expression after stroke. Our analyses of stroked brains at post stroke day (PSD) 3 and PSD 14 revealed a signi cant increase in I 27l2a mRNA. To determine whether MG represented the predominant source for the increased I 27l2a expression, we used PLX5622 treatment to deplete MG in mice prior to inducing stroke. PLX5622 is a CSF1R antagonist that eliminates CNS-resident MG 15 . CSF1R mediated signaling is required for MG survival and proliferation 16,17 . Mice were treated with PLX5622 for seven days. On day 7 of administration of PLX5622, pdMCAO was performed. The PLX5622 diet was continued for 3 days after stroke surgery to prevent repopulation by MG. At PSD 3, brains were isolated and analyzed by qRT-PCR (ipsilateral hemisphere) and immunostaining (contralateral hemisphere). As a control, mice were fed normal diet (ND) for the same period (Fig. 3a). Notably, I 27l2a mRNA level was signi cantly reduced by 86% in PLX-stroked brains (ipsilateral hemisphere), compared to ND-stroked brains (Fig. 3b, p < 0.05). The effectiveness of PLX5622 to eliminate MG in brains was con rmed by Iba1 immuno uorescence ( Fig. 3c-d) in the contralateral hemisphere. PLX5622 treatment resulted in a profound decrease in the number of MG in brains (Fig. 3d).
Moreover, PLX5622 treatment signi cantly reduced Tmem119 expression in brains after stroke, compared to naïve or normal diet administered brains (Fig. 3e, p < 0.05, compared to naïve and ND-stroke). These data indicate that the induction of I 27l2a after stroke is primarily dependent on the MG population in the brain.
MG induce I 27l2a/IFI27L2 expression with in ammatory stimuli. We next used cultured MG to evaluate the potential for in ammatory mediators to promote I 27l2a expression. First, we used mouse primary MG collected from the mixed glial cell culture obtained from P2 pups. Primary MG were treated with TNFα (20 ng/mL) and IFN-γ (20 ng/mL) for 24 hours (to measure mRNA level of I 27l2a) and 48 hours (to measure protein level of I 27l2a by ELISA with cell lysate). Both mRNA (Fig. 3f) and protein levels ( Fig. 3g) of I 27l2a were signi cantly increased with treatment.
Given these results, we next tested if IFI27L2 protein was increased in the brains of patients that featured neuroin ammation. Sections from the brains of deceased patients without neurological disease (n = 2, female) and from stroke patients (n = 3, female) who also demonstrated cerebral amyloid angiopathy (CAA) pathology and tauopathy, in which neuroin ammation (microgliosis) is prevalent. Immunohistochemistry showed signi cant IFI27L2 expression in the stroke brain samples but low expression in age-matched control samples (Fig. 3j, representative of n = 2-3). Together, these data show the responsiveness of I 27l2a (murine) and IFI27L2 (human) to in ammatory stimulation and the presence of elevated IFI27L2 in brain of patients with multiple forms of neuroin ammatory disease.
Differential expression of I 27l2a in subtypes of microglia (MG) and macrophage (MΦ) populations in the aged brains following stroke. Given the extensive heterogeneity evident within MG and MΦ, we subjected the aged scRNA-seq datasets to more granular analysis to determine if I 27l2a expression pro les correlated with different functional roles. We ultimately identi ed a total of 28 clusters from brain cells of aged sham and aged stroke mouse brains, eight clusters of which were assigned an MG or monocyte/MΦ identity based on the expression of conserved cell markers (Extended data Fig. 10). Two MG homeostatic clusters were identi ed based on the expression of MG genes such as Siglech, Tmem119, Gpr34, P2ry12, and Selplg. These MG were annotated as Siglech homeostatic MG and P2ry12 homeostatic MG. We also identi ed two different MG that appeared to be in an activated status (Rag + activated MG and Tyrobp + activated MG). Two Monocyte-Macrophage populations were also identi ed. We also found the disease-associated MG (DAM) like cluster showing high expression of Lpl, Itgax, Cst7, and Spp1. Note that these genes also correlate with the microglial genes and lipid metabolism genes upregulated in DAMs in other neurodegenerative diseases, such as AD 14,18 . Since we found that stroke and aging increase the expression of I 27l2a in MG, and that I 27l2a expression is negatively correlated with DAM genes, we asked whether expression levels and degrees of I 27l2a gene induction from sham to stroke in DAM would be different from MG in other sub-clusters. We therefore compared the degree of I 27l2a gene induction among MG sub-clusters in aged sham versus stroke brains (Table 3). Among the non-homeostatic MG clusters, I 27l2a induction in DAM (1.7 fold) is lower than any of the other activated MG.
I 27l2a expression is su cient to promote MG activation. Given the induction of I 27l2a in MG in aged brains and following stroke, we sought to elucidate the functional role of I 27l2a in MG-mediated neuroin ammation. Changes in microglial morphology is an early, quanti able sign of in ammation in MG and MG functionality. Thus, we asked if I 27l2a expression alone (without additional in ammatory mediators) could induce a pro-in ammatory morphology in MG. We infected a murine microglial cell line (Sim-A9 cells) with a lentivirus where the Cx3cx1 promoter drove the expression of I 27l2a and an eGFP reporter, or a lenti-eGFP control. At 5 days post-infection, quanti cation of cell morphology showed that induction of I 27l2a expression caused an increase in the percentage of cells with a small, rounded shape (to a more amoeboid morphology or de-rami cation) compared to lenti-eGFP control ( Fig. 4a-b). Interestingly, MG with higher I 27l2a expression (using eGFP intensity as a surrogate maker) showed more dramatic morphological changes compared to cells that had low I 27l2a/eGFP expression (Fig. 4c). These results provide direct evidence that I 27l2a alone can initiate MG activation, even in basal conditions (i.e. in ammatory stimuli are not required).
I 27l2a induces ROS production. Earlier reports showed evidence for I 27l2a localization in mitochondria within non-CNS cells 19 . We also detected increased IFI27L2 in the peri-nuclear membrane and in mitochondria in HMC3 cells (not shown), leading us to question whether I 27l2a could mediate mitochondrial dysfunction in MG. Thus, we asked if I 27l2a expression alone could initiate the reactive oxygen species (ROS) generation in activated MG. We used CellROX Red and MitoSox. First, we utilized CellROX Red, a detector of most ROS species, to determine if I 27l2a expression induces ROS production in Sim-A9 cells in unstimulated conditions. Quanti cation by ow cytometry revealed that I 27l2a overexpression alone promotes a signi cant increase in ROS production (Fig. 4d, as expressed in median uorescence intensity, MFI, Ctrl: lenti-eGFP control, I 27l2a: lenti-I 27l2a-eGFP, n = 4, * p < 0.05). Next, we only analyzed GFP positive cells, representing those with successful transduction. The ROS level was greater in I 27l2a expressing cells compared to eGFP only control cells (Fig. 4e, Ctrl: lenti-eGFP control, I 27l2a: lenti-I 27l2a-eGFP, n = 4, * p < 0.05).
We checked more speci cally if mitochondria contribute as an I 27l2a-induced ROS source using Mitosox dye (speci c indicator of mitochondria-derived ROS). I 27l2a overexpression resulted in a signi cant increase in mitochondria generated ROS level (Fig. 4f) and the percentage of Mitosox + cells (Fig. 4g). The "no-virus" cells (No) showed negligible effect on ROS levels. These data indicate that I 27l2a expression alone can cause ROS generation in mitochondria in activated MG, implying a causative role of I 27l2a in mitochondrial dysfunction in MG. I 27l2a hemizygous deletion is protective from ischemic brain injury in mice. Given our nding that increased I 27l2a expression alone is su cient to promote microglial activation, we asked if limiting I 27l2a expression could reduce microglial activation and brain injury following stroke. We used a permanent distal middle cerebral artery occlusion (pdMCAO) stroke model in WT and I 27l2a +/-(Het) mice (2-3 month old, male). At post-stroke day (PSD) 3, the infarct volume was signi cantly reduced in Het, compared to WT brain ( Fig. 5a-b, n = 5 or 6, * p < 0.05). The area of activated MG (Iba1) was also reduced in the primary injury region at PSD 14 ( Fig. 5c-d, n = 5 or 6, * p < 0.05). The pdMCAO model 20 is also a well-established model for evaluating secondary injury in stroke; signi cant gliosis develops in the ipsilateral thalamus several days after the primary injury. We and others have shown signi cant gliosis in the ipsilateral thalamus 1 or 2 weeks following stroke 11,21,22 . Therefore, to evaluate the role of I 27l2a in secondary thalamic injury, we examined thalamic gliosis in WT and Het mice (2-3 months old, male) at PSD 14. Evaluation of the ipsilateral thalamus revealed signi cant reduction in both microgliosis (Fig. 5ef, n = 6, * p < 0.05) and astrogliosis (Fig. 5g-h, n = 6 ** p < 0.01) in Het mice compared with WT. Note that the reduced injury in Het mice is not due to developmental differences in MCA territory. Analysis of vascular territory between WT and full I 27l2a KO revealed no difference (Extended Data Fig. 11, n = 6, p = 0.19). Together, these ndings indicate that reducing I 27l2a expression can reduce primary and secondary injury associated with ischemic stroke, likely through attenuation of the microglial-mediated in ammatory response.

Discussion
We used scRNA-seq to explore the effects of aging and stroke at the cellular level in the brain. As a result of these studies, we identi ed I 27l2a as a gene that demonstrated signi cant age-dependent upregulation in the post-stroke brain. This novel initial nding led to further study related speci cally to where and when I 27l2a was upregulated in the brain and to the functional role of I 27l2a in aging, stroke, and other neurodegenerative conditions. From these studies, we now present the following major new ndings: 1) I 27l2a is highly upregulated in MG following stroke, particularly in aged brain. 2) I 27l2a is mildly upregulated by aging alone in MG. 3) Upregulation of I 27l2a following stroke occurs predominantly in MG. 4) I 27l2a expression and upregulation following stroke varies by MG subtype. 5) I 27l2a expression is inversely correlated with gene markers of DAM cell phenotype. 6) Expression of I 27l2a alone promotes MG activation and mitochondrial ROS production. 7) Reducing I 27l2a expression provides reduced MG activation and ischemic injury in an ischemic stroke model. When considered as a whole, we now propose that in ammatory stress (caused by the aging process, ischemic stroke or other) initiates I 27l2a gene expression predominantly in MG, which then enhances and propagates in ammatory damage throughout the brain. Further, our data suggest that the level of I 27l2a expression in MG may serve as a molecular switch that triggers pro-in ammatory phenotypes and dampens reparative phenotypes in aging and following stroke. We discuss what is known about I 27l2a function and elaborate on our major ndings below.
Interferons and interferon mediated signaling were originally identi ed as antiviral 23 , anti-proliferative and immunomodulatory mechanisms 24 that were induced by viral infection. These pathways are most commonly known to play pivotal roles in host defense against viral infection. Accumulated evidence has also shown a critical role for interferon signaling (especially Type I IFN, α and β) in regulating neuroin ammation in aging and diseased brains, such as in the AD and stroke brain [25][26][27][28] .
However, how or if IFN signaling (e.g. Type I or Type II) modulates I 27l2a expression directly in stroke brain, especially in aged brain, has not been previously described. It has been shown that stimulation with IFNα and β, known activators of IFN type I signaling, can induce I 27l2a expression in cortical neurons 8 and adipocytes 29 . However, it was unclear whether canonical Type I (α, β) or Type II (γ) IFNs could induce I 27l2a expression in microglia in a dish or in damaged brain to induce in ammation. Moreover, analysis of the I 27l2a promoter region (human ortholog, Isg12b) failed to show interferon-stimulated response elements (ISREs), which are thought to be required for interferon stimulated gene induction 29 . These ndings suggested an alternative, non-canonical interferon-independent pathway for I 27l2a regulation, such as by microRNA or signaling via other foreign DNA/RNA sensing receptors. Indeed, our scRNA-seq data supports the notion of an interferon-independent pathway. We observed that while I 27l2a expression is markedly upregulated in MG, other representative Isg (Mx1, Mx2, I family such as I 27, I 35, and Ifnb1, etc.) known to be upregulated by IFN response, especially type I Interferons (IFNα and β), were not measurably changed in MG. These ndings suggested that another pathway (e.g. an IFN I independent pathway or combined signaling with IFN response and other intrinsic cellular signaling caused by ischemic or hypoxic insults) might be involved in the acute/chronic I 27l2a induction in MG. Such a scenario could be explained by the existence of other molecular hubs that relay the downstream signals to ultimately induce I 27l2a expression in MG after stroke. This possibility is supported by our scRNA-seq data showing signi cant upregulation of interferon regulatory factor 7 (IRF7) in MG from aged brain following stroke, whereas other IRFs were not markedly changed (data not shown). Moreover, in the series of in vitro experiments with primary MG treated with pro-in ammatory cytokines, a positive correlation was found between I 27l2a and IRF7, implying that IRF7 signaling may contribute to I 27l2a expression in activated MG. Interestingly, using TRANSFAC, a tool for transcriptional analysis, we found a putative IRF7 binding motif in the promoter of I 27l2a, suggesting that IRF7 may act as a key transcription factor to induce the I 27l2a gene expression in the in ammatory situation in microglia and other cells in brains. Further experiments will be required to speci cally test the role and involvement of IRF7-mediated transcriptional regulation on I 27l2a expression.
Outside of the CNS, a limited number of reports have suggested a role for I 27l2a in facilitating in ammation though its interaction with other cellular partner proteins. During conditions of in ammation, it was shown that the I 27l2a protein is rapidly expressed and interacts with nuclear receptor 4A (NR4A) family members. The NR4A family is thought to support expression of multiple genes involved in attenuating in ammation in various kinds of cells [30][31][32] . Binding between I 27l2a and NR4A in the nucleus results in the export of NR4A to the cytosol, and thus the removal of a driver of antiin ammatory and cytoprotective gene expression 10 . This model of NR4A regulation could also explain the novel role of I 27l2a in MG after stroke. In support of this possibility, a separate study showed that MG-speci c Nr4a1 knockout alone promoted in ammation and microglial activation as well as increased pathology in the experimental autoimmune encephalomyelitis mouse model (EAE) 33 . Interestingly, Nr4a1 was also shown to play a critical role in maintaining an anti-in ammatory state of macrophages via attenuating NF-kB mediated pro-in ammatory gene expression 34 . Moreover, it was revealed that if Nr4a1 is deleted in myeloid cells such as macrophages, more pro-in ammatory cytokines are produced 35 . It was recently shown that NR4A may also regulate phagocytosis via Mer tyrosine kinase (MerTK) gene expression in the cardiac repair process 36 . MerTK is a member of the MER/AXL/TYRO3 receptor kinase family, which is known to regulate phagocytic capacity in MG and MΦ. If other phagocytosis-related genes such as Axl and anti-in ammatory cytokines are also direct targets for NR4A, it is possible that lowering the expression level of I 27l2a might boost MG phagocytic capacity or promote phenotypical changes to DAM via promoting reparative phagocytosis related genes such as Axl. Indeed, our scRNA-seq data showed an inverse correlation between I 27l2a and Axl, indicating that lower I 27l2a expression may be a characteristic of the non-in ammatory MG phenotypes. We also found higher expression of Nr4a1 in DAM and one of the homeostatic MG clusters. Overall, our data support the novel working model that I 27l2a regulation of Nr4a1 contributes to the phenotypic polarization of microglia in natural aging and in brain pathology. If our model is correct, the interaction of I 27l2a and Nr4a1 would represent a novel therapeutic target for reducing brain in ammation.
The other reported mechanism by which I 27l2a acts involves regulation of apoptosis. Studies in activated MG and other cells showed that I 27l2a can be shuttled to the mitochondria membrane, where it initiates a mitochondria-dependent apoptosis process 37,38 . Our study also supports this possibility wherein I 27l2a can act as an initiator for mitochondrial dysfunction by producing ROS. The mode of action of I 27l2a may also be regulated by its subcellular destination. With regard to the potential to trigger apoptosis and in ammation, we speculate that with more severe or prolonged MG activation, I 27l2a accumulation at the mitochondrial membrane might initially contribute to the MG in ammatory change, and then later trigger MG apoptosis. One intriguing hypothetical scenario is that high I 27l2a expression or mitochondrial targeting of I 27l2a contributes to the eventual termination of in ammatory MG. At this time, however, the potential role of I 27l2a in regulating apoptosis of MG after stroke has never been explored.
Our comparison of young and aged non-stroke brains (shams) showed an age-dependent upregulation of I 27l2a in MG, Epi, and B-cell populations in the brain. Expression of I 27l2a in these cell populations went from undetectable expression in young brain to moderate expression in aged brain. How aging promotes increased I 27l2a expression in these clusters is unknown. Since I 27l2a is among the known interferon stimulated genes (Isg) that can be upregulated by IFN-mediated pathways (viral infection or by in ammatory pathways 39,40 ) chronic low-level activation of any of these pathways might promote the observed age-dependent increase in I 27l2a expression. However, given that the mice were housed in a speci c pathogen free (SPF) facility, it is most likely that the increase we observed is due to low-level chronic in ammation that is known to exist in the aged brain 41,42 .
Expression of I 27l2a differed among the MG sub-clusters in the basal level of I 27l2a expression in aged sham brain and degree of I 27l2a upregulation following ischemic stroke. Expression levels of I 27l2a in sham brain were low in homeostatic subclusters and resulted in less upregulation in the poststroke brain compared with activated MG subclusters. The activated subclusters showed 2.1-3.5 fold higher I 27l2a expression compared with homeostatic subclusters following stroke. The greater expression level in activated MG suggests a potential role for I 27l2a in microglial activation and proliferation. I 27l2a expression in two other MG subclusters is discussed further below. Our data showed an inverse correlation between I 27l2a expression and reparative MG genes, which are now recognized as DAM genes. This clear relationship in aged stoke brain suggests that I 27l2a could also be a key determinant for inducing DAM phagocytic activity or DAM phenotypical changes from nonactivated or activated MG. It was further notable that the inverse correlation held up with groups of genes that are related to maintaining MG homeostasis and the DAM phenotype. These ndings suggest that I 27l2a expression level may contribute to expression of genes in MG related to reparative and phagocytic function. Future study will be required to determine if the reduced I 27l2a expression is causative versus merely correlative for this MG phenotype.
In summary, using unsupervised scRNA-seq, we have found a signi cant increase in I 27l2a expression in MG following stroke, with particular upregulation in the aged stroke brain. Our data further show that mild I 27l2a upregulation even occurs with aging alone. We present evidence for a new model of MG phenotype regulation, wherein I 27l2a acts as a novel molecular regulator of microglial phenotypical changes and function. Based on the data we present here, we propose that elevated expression of I 27l2a contributes to a pro-in ammatory MG phenotype (producing more ROS and proin ammatory cytokines in MG) and reduced I 27l2a enables a non-in ammatory or phagocytic phenotype. Also we speculate that the functional role of I 27l2a is at least partly through negative regulation of Nr4a1mediated gene transcription. In total, these ndings suggest that targeting of I 27l2a expression or I 27l2a protein function in MG could be a novel strategy for regulating neuroin ammation in aging, stroke, or other neurodegenerative diseases to promote better functional recovery.

Methods
Animals. All procedures were performed in accordance with NIH guidelines for the care and use of laboratory animals and were approved by the Institutional Animal care and use committee of the Permanent distal middle cerebral artery occlusion (PDMCAO) model. C57BL/6J mice of both sexes were used for scRNA-seq at 11-14 weeks or 18-22 months of age. PDMCAO was induced by permanent ligation of the right distal middle cerebral artery (MCA) using a micro-coagulator (Accu-temp) 11 . Mice were anesthetized with iso urane (4% induction and 2% maintenance in air ow) and body temperature was maintained at 37°C by feedback-controlled heating pad and rectal temperature probe. Bupivacaine (0.25% at 1ml/kg) was injected subcutaneously (s.c.), prior to any skin incision 43  Research Diets Inc. PLX5622 was administrated for 7 days prior to the PDMCAO procedure and continued for 3 days after stroke. At 3 days after surgery, brains were isolated and ipsilateral hemisphere was used for RNA isolation and qRT-PCR analysis. The contralateral hemisphere was used for immunostaining.
Brain sample preparation for single-cell RNA sequencing. We processed brains from young and aged mice subjected to either sham or PDMCAO surgeries (Table 1). 14 days post PDMCAO (or sham surgery), Anesthetized mice were transcardially perfused with heparinized PBS (10 U/mL). Brains were removed from the skull and sliced coronally into 3 mm-thick blocks (from a region spanning + 1 to -2 mm from bregma), covering the cortical infarction and secondary thalamic injury site 11 . The brain slice was then minced with a razor blade and subjected to the brain tissue dissociation protocol (Miltenyi Biotec, Gladbach, Germany). Minced tissue was then incubated a collagenase/dispase mixture (150 µL of 1 mg/mL in 2 mL) for 30 minutes at 37˚C in a gentleMACS Octo Dissociator (Miltenyi Biotec, Bergisch Gladbach, Germany) using the pre-installed program for adult brain dissociation. Myelin was removed using debris removal solution. Red blood cells were lysed and removed with red blood cell lysis solution.
The nal cell suspension was stained with trypan blue and live cells were counted using Countess II FL Automated Cell Counter (Thermo Fisher scienti c, USA). Sequencing data processing and analysis. The cell ranger pipeline (10X genomics, Pleasanton, CA) was utilized to map the sequences to mouse reference genome (mm10), and to process barcode containing sequence data, aligning the read and generating feature barcode matrices that could be further processed by the Seurat package 12 using R. We also used cellranger aggr pipeline to combine outputs from multiple samples into one output le. AAGR1 (5,706 total cells analyzed) was a combined population consisting of "sham brains of young male and young female" mice. AAGR3 (5,174 total cells analyzed) was a combined population consisting of "sham brains of aged male and aged female" mice. AAGR2 (12,866 total cells analyzed) was a combined population consisting of "Stroke brains from young male and young female" mice. AAGR4 (total 8,226 cells analyzed) was a combined population consisting of "Stroke brains from aged male and aged female". Reads were mapped to the mm10 murine transcriptome (10X genomics, Pleasanton, CA). We used Seurat 3.1 44 to analyze scRNA-seq data for clustering, and DEG identi cation between clusters and between the two groups. Brie y, log-normalization using NormalizeData was utilized. Feature counts for each cell were divided by total counts for the cell and multiplied by the scale factor (10,000). Then using log1p, data was natural log-transformed. The Uniform Manifold Approximation and Projection (UMAP) dimensional reduction technique (UMAP) was used for dimensional reduction and clustering was carried out using FindNeighbors and FindClusters with the resolution parameter either at 0.02 (generating 6-7 clusters) or at 1 (generating 25-27 clusters).
Conserved cell type markers in each cluster were identi ed by using FindConservedMarkers. The name and level of genes that were differentially expressed in each cluster was determined using FindMarkers. Metadata and normalized read count data was extracted from Seurat objects and fed into Excel to further identify the critical genes (top 10 genes or top 50 genes) that were up-and down-regulated in each cluster and to nd the correlation between levels of I 27l2a and other MG genes.
Single molecule in situ hybridization (RNAscope). The RNAscope uorescent multiplex assay (Advanced Cell Diagnostics, Newark, CA, USA) was performed according to manufacturer's instructions with 2 week post-stroke brains of aged mice (18-20 months) and aged sham to probe I 27l2a transcripts in brain cells. The murine I 27l2a probe was designed by ACD Biosystems, based on their own criteria. Brain sections (PFA xed, 30-µm thickness) from the 2-week post stroke brain and sham brains of aged mice (18-20 month old) were hybridized with I 27l2a probes for 2 hours at 40˚C. At the same time, ACD 3-plex positive control and negative control probes were incubated on one brain section to con rm signal speci city. The probes were ampli ed according to the manufacturer's instructions and labeled with Opal-570 Red uorophore (Akoya Biosciences, Marlborough, MA, USA). DAPI was used to label nuclei. Images were taken with a uorescent microscope (Leica DMi8 uorescence microscope system, Leica Biosystem, IL, USA) and a confocal microscope (Leica TCS SPE confocal system, Leica Biosystem, IL, USA). Multiple images were captured with the 10X objective covering the hemisphere and stitched to generate a single image (Leica LAS X software). Brain processing and immunostaining. For detecting Iba1 in PLX-or normal diet-treated mouse brains and gliosis (Iba1 and Gfap) in the thalamus following stroke, we performed immunostaining as previously described 11 . Cardiac perfusion with PBS, followed by 4% PFA (paraformaldehyde in PBS) were performed to clear the blood in brains. Perfused brains were then submerged in 30% sucrose in PBS for 24 hours at 4ºC prior to sectioning at 30 µm thickness (Micron HM 450, Thermo Fisher Scienti c, Waltham, MA, U.S.A.). Sections corresponding to − 2 mm from bregma, which contain hippocampus and thalamus, were washed with PBS, incubated with blocking buffer (10% goat serum, 0.3% Trion X-100 in PBS), and then incubated overnight at 4ºC with the following primary antibodies: Rabbit anti-Iba1 antibody (1:200) (Wako Pure Chemical, Japan), mouse anti-GFAP antibody-cy3 (1:500) (Millipore Sigma, MO, USA). We used either donkey anti-rabbit IgG-Alexa 594 or 488 (1:200, Thermo Fisher Scienti c, Waltham, MA, USA) to recognize rabbit anti-Iba1 antibody. Sections were incubated with DAPI (4′, 6diamidino-2-phenylindole) to label nuclei. Images were obtained using a Leica TCS SPE confocal system and a Leica DMi8 uorescence microscope system (Leica Biosystem, IL, USA). Images were captured using a 10X objective. Higher magni cation images of selected regions were collected using 20X or 40X objectives. Image analysis was performed using Image J software (National Institutes of Health).
Lentivirus infection and ROS measurement by ow cytometry. Control lentivirus (Cx3cr1-IRES-eGFP, initial titer-1.55×10 8 TU/ml) and I 27l2a expressing lentivirus (Cx3cr1-I 27l2a-IRES-eGFP, initial titer-1.07×10 8 TU/ml) were generated (GeneCopoeia, Rockville, MD, USA) and these virus went through in-house quality control and validation. Sim-A9 cells, a microglia-like cell line, was transduced with control lentivirus (eGFP alone) or I 27l2a expressing lentivirus at 2 MOI using polybrene (Millipore Sigma, St. Louise, MO, USA). Real-Time quantitative RT-PCR. To validate the ndings of scRNA-seq data, we performed qRT-PCR. Brains from naive young (3 mons) and aged mice (18-20 mons), or brains from sham and stroked mice (PSD 3 or PSD 14) were harvested and dissected to obtain both cortex and thalamus. For the PLX5672 treatment experiment, the ipsilateral hemisphere was collected instead. Total RNA was puri ed with TRIzol™ Reagent (Thermo Fisher Scienti c, Waltham, MA, USA) using the RNeasy Mini Kit (Qiaqen, Germantown, MD, USA) according to the manufacturer's instructions. Purity of RNA (> 1.7 at 260/280) and concentration of puri ed RNA were measured by Nano-drop Spectrometer and 1 µg of RNA was used μ to generate cDNA with iScript™ Reverse Transcription Supermix (Bio-Rad, Hercules, CA, USA). The SsoAdvanced Universal SYBR Green Supermix (Bio-Rad, Hercules, CA, USA) was used to detect newly ampli ed amplicons with a C1000 Touch Thermal Cycler CFX96 Real-Time System (Bio-Rad, Hercules, CA, USA). The PCR cycles were as follows: initial denaturation at 95˚C for 30 sec, followed by 40 reaction cycles of 95˚C for 5 sec, 56˚C for 10 sec, and 72˚C for 10 sec. To quantify relative gene expression, we used the ΔΔCt method using Ct values for the gene of interest normalized to GADPH. Data was expressed as fold change relative to control samples. Primer sequences are provided in Extended data Table 1.
Statistical data analysis. Statistical data analysis was performed using Prism 7.0.3 (GraphPad Software, San Diego, CA, USA) and R in Rstudio environment with p < 0.05 considered statistically signi cant. Data are presented as the mean ± standard error of the mean (SEM), and analyzed using an unpaired t-test (for two group comparisons) or a one-way ANOVA with Tukey post-hoc test for multiple comparisons. Tables   Table 1 scRNA-seq samples analyzed      MG represent the primary source of I 27l2a upregulation following stroke. Depletion of brain MG with PLX5622 eliminates the stroke-induced increase of I 27l2a expression. Brain I 27l2a expression is reduced in PLX5622 treated mice after stroke. (a) Summary of experimental timeline and procedures.

Declarations
PLX5622 treatment is used to deplete the brain MG population. The brain hemisphere ipsilateral to the stroke was used for quantitative real-time PCR analysis, while the contralateral hemisphere was sliced for  ROS levels (f. MFI, g. % of MitoSox+ cells) detected by Mitosox was increased in I 27l2a lentivirus infected cells, compared to control (n=3-6, * p<0.05, ** p<0.01, one-way ANOVA with Bonferroni's multiple comparison test). Figure 5