microRNA-132 regulates gene expression programs involved in microglial homeostasis

Summary microRNA-132 (miR-132), a known neuronal regulator, is one of the most robustly downregulated microRNAs (miRNAs) in the brain of Alzheimer’s disease (AD) patients. Increasing miR-132 in AD mouse brain ameliorates amyloid and Tau pathologies, and also restores adult hippocampal neurogenesis and memory deficits. However, the functional pleiotropy of miRNAs requires in-depth analysis of the effects of miR-132 supplementation before it can be moved forward for AD therapy. We employ here miR-132 loss- and gain-of-function approaches using single-cell transcriptomics, proteomics, and in silico AGO-CLIP datasets to identify molecular pathways targeted by miR-132 in mouse hippocampus. We find that miR-132 modulation significantly affects the transition of microglia from a disease-associated to a homeostatic cell state. We confirm the regulatory role of miR-132 in shifting microglial cell states using human microglial cultures derived from induced pluripotent stem cells.

MiR-132 is downregulated in Alzheimer's disease (AD) brain Mir-132, mainly known for neuronal functions, regulates many aspects of AD pathology MiR-132 manipulation in mouse hippocampus reveals a role in microglial homeostasis MiR-132 shifts diseaseassociated to homeostatic microglia in iPSC-derived cultures INTRODUCTION miRNAs are endogenous small noncoding RNAs that primarily operate as part of the cellular gene silencing machinery. By binding complementary sequences in target messenger RNAs (mRNAs), miRNAs inhibit translation or induce mRNA degradation. 1-3 miRNAs can concomitantly regulate numerous targets and are, therefore, thought to mainly function as ''fine-tuners'' of core cellular functions, such as maintaining tissue homeostasis in the presence of stressors. 4,5 In the central nervous system (CNS), miRNAs control gene expression in diverse cellular populations. [6][7][8][9][10][11] The functional pleiotropy of miRNAs renders them, in principle, interesting candidates to treat complex multifactorial diseases, such as AD, as they can potentially ''hit'' multiple sensitive nodes of a pathogenic molecular cascade. [12][13][14][15][16] miRNA profiles are altered in the brain of AD patients, often from early disease stages onwards. [17][18][19][20][21][22] Several of these deregulated miRNAs directly regulate key molecular processes in AD pathogenesis including amyloidosis, Tau pathology, (neuro)inflammation, neuronal death, and memory impairment. 16 miR-132, one of the most abundant, brain-enriched miRNAs, and the most consistently downregulated miRNA in human AD brain, exemplifies the multi-targeting potential of miRNAs in the context of AD. 16,20,23,24 More specifically, miR-132 can ameliorate AD pathology by regulating a series of targets involved in amyloid plaque accumulation, Tau hyperphosphorylation and metabolism, neuronal survival, neurotrophic signaling, adult neurogenesis, and memory formation. 23-38 miR-132 expression variation could explain 6.7% of the observed variance in AD histopathology and mediate part of the effect of the AD polygenic risk score on global cognitive decline. 39,40 miR-132 supplementation in AD mouse models ameliorates amyloid and Tau pathology and improves memory performance. 23,26,32,33,35 Although these findings are indicative of the therapeutic potential of miR-132 in AD, it remains unclear how altering miR-132 in the brain would broadly impact molecular networks and biological cascades across different cell types.
Identifying molecular targets of any miRNA remains a significant challenge. While the primary effects of miRNAs involve inhibiting the translation of mRNAs, a large part of miRNA regulation results in alterations of the transcriptome. 2,3,41-45 Moreover, miRNAs exert modest effects on individual targets. It is assumed that the accumulation of such subtle changes in the expression of individual genes in a specific molecular

RESULTS
In silico identification of putative miR-132 targets in the brain Predicted mRNA targets of highly tissue-specific miRNAs are generally expressed in the same tissue as the miRNA. 60, 61 We initially employed an in silico approach to extract predicted miR-132 targets and characterize their expression in diverse cellular populations in the brain. We made use of the Encori database, 62 which provides the possibility to explore several target prediction algorithms together with AGO crosslinking immunoprecipitation sequencing (CLIP-seq) data 58,63 to identify targets of any microRNA of interest. We selected 384 predicted miR-132 targets based on the combination of at least 2 target prediction algorithms and a high-confidence threshold for AGO-CLIP-seq ( Figure 1A and Table S1). To functionally annotate the selected miR-132 predicted targets, we used the database for annotation, visualization and integrated discovery (DAVID) 64,65 and identified significantly enriched [Fisher's Exact test, false discovery rate (FDR)-adjusted p value <0.05] gene ontology (GO) biological processes and Kyoto encyclopedia of genes and genomes (KEGG) pathways. Not unexpectedly, regulation of transcription and translation and other rather generic functional categories known to be regulated by miRNAs, 42,52,66,67 were depicted. In addition, previously reported miR-132-regulated processes, such as stem cell pluripotency and proliferation, 28,29,35 axon extension, 34,36,68,69 and cellular homeostasis, autophagy and apoptosis, including FOXO signaling, 20,38,70,71 were identified among the significantly enriched pathways ( Figure 1B). We next asked in which brain cell types the in silico predicted targets of miR-132 are expressed. We assessed cell type-specific expression using a single-cell RNA sequencing (scRNAseq) dataset from mouse hippocampus. 72 Predicted miR-132 targets were expressed in neurons, oligodendrocytes, microglia, and astrocytes ( Figure 1C and Table S1), suggesting that miR-132 itself may also be similarly expressed in various cell types. We confirmed expression of miR-132 in oligodendrocytes, microglia, astrocytes, and neurons isolated from mouse hippocampus using magnetic activated cell sorting (MACS) and fluorescence-activated cell sorting (FACS) (Figures S1A, S1D, S1G, S1H, and S1J). Taken together, these data suggest a putative functional role for miR-132 in different cellular populations in the brain. To test this hypothesis, we set out to experimentally explore miR-132-regulated cellular targets in mouse hippocampus using a combinatorial proteomics and single-cell transcriptomics approach ( Figure 1D).

miR-132 depletion affects divergent biological pathways in different cellular populations of the hippocampus but effects on individual targets are small
To explore physiologically relevant effects of miR-132 regulation we first employed a loss-of-function approach. More specifically, to determine the impact of miR-132 knockdown in the hippocampus, we intracerebroventricularly (ICV) infused wild-type mice with antisense (miR-132 antagomiR, miR-132 KD) or corresponding control (Control) oligonucleotides ( Figure S1B) and assessed the effects on the proteome in bulk and on the transcriptome at single-cell resolution. The expression levels of miR-212, a cognate miRNA, which is co-transcribed with miR-132 and shares a similar mature sequence, did not change upon miR-132 KD ( Figure S1I). While miR-132 knockdown was consistent in isolated hippocampal oligodendrocytes, the decrease was less robust in astrocytes and microglia, whereas, unexpectedly, no reduction in miR-132 levels was detected in neurons, precluding the interpretation of any observed functional alterations in neuronal clusters ( Figure S1E). iScience Article For proteome profiling, whole hippocampal lysates were analyzed by label-free quantitative liquid chromatography-mass spectrometry using data independent acquisition (DIA). A total of 6,116 proteins were identified. Differential protein expression analysis between miR-132 KD and control hippocampi (Figure 2A and Table S2) revealed broad but subtle effects on protein levels, which did not retain statistical significance after correction for multiple testing (Benjamini-Hochberg), similarly to previous reports on miRNA-dependent regulatory effects. 73,74 Given that miRNAs are negative regulators of their targets, we selected the top 5% upregulated proteins (corresponding to 306 proteins). Next, we intersected this set with the previously extracted list of 384 in silico predicted miR-132 targets ( Figure 2B). On this basis, we identified 10 putative miR-132 targets of interest, which include proteins expressed in both neurons and glia or only in glial populations (Table S1).
We next performed scRNAseq analysis in miR-132 KD and control hippocampi by employing a droplet microfluidics platform (10X Genomics Chromium) (Table S3). Cells were sampled from whole hippocampus, while additional excitatory neurons and astrocytes were obtained by FACS from Thy1-YFP transgenic mice. After quality control and filtering, we recovered data from 19,705 high-quality single cell transcriptomes. Identification and possible function of miR-132 predicted targets in brain (A) Schematic representation of the identification strategy of predicted miR-132 targets, using an intersection of two target prediction algorithms and a highconfidence threshold for experimental data from AGO-CLIP-seq experiments. 62  We observed high abundance of oligodendrocytes, microglia and astrocytes in our dataset, whereas only small numbers of distinct neuronal subtypes were retrieved ( Figures 2C-2E), in-line with previous studies reporting the difficulty to isolate intact neuronal cells from the adult brain. 75,76 All acquired datasets were integrated using Harmony, 77 and the resulting integrated 14 clusters were annotated via established marker genes on uniform manifold approximation and projection (UMAP) plots ( Figures 2C, 2D, and S2C).
We performed differential gene expression analysis in miR-132 KD and control cells, using a two-sided Wilcoxon rank-sum test with Bonferroni correction for multiple testing. For each of the main identified cell types, we calculated the amount of significantly differentially expressed genes (DEGs) (Wilcoxon ranksum test, Bonferroni-corrected p value <0.05) ( Figure 2E and Table S3), and extracted potential miR-132 targets among the identified upregulated DEGs. Significant transcriptional changes and miR-132 targets were found in microglia and oligodendrocytes (microglia, 78 DEGs and 1 predicted target; oligodendrocytes, 160 DEGs and 3 predicted targets), while we were not able to depict potential miR-132 targets in the neuronal and astrocytic populations, in accordance with the limited numbers of DEGs (astrocytes, 2 DEGs; granule cells, 1 DEG) ( Figure 2E).
We then asked whether miR-132 knockdown exerts consistent effects on individual proteins and mRNAs in mouse hippocampus. To this end, we performed Pearson correlation analysis to compare directionality of the changes in protein and pseudo-bulked (upon in silico pooling of all cells) gene expression by plotting the logarithmic fold change (logFC) of gene/protein pairs shared between proteomics and transcriptomics ( Figure S2A). Expression profiles of individual proteins/mRNAs did not show concordant changes.
miR-132 overexpression alters gene expression programs in both neuronal and immunerelated pathways miR-132 overexpression could potentially elicit stronger effects in proteomic and transcriptomic profiles and is arguably more relevant in a therapeutic context. 81  iScience Article infused with corresponding control (Control) oligonucleotides (Figures S1C and S1I). We confirmed that average miR-132 levels increased in all main cell types ( Figure S1F), and hippocampi were subjected to bulk proteomics and scRNAseq, as before. For proteomics analysis, a total of 5,995 proteins were identified. Differential protein expression analysis comparing miR-132 OE and control did not result in significantly changing proteins after correction for multiple testing (Benjamini-Hochberg) ( Figure 3A and Table S2). We then selected the top 5% downregulated proteins (corresponding to 300 proteins) and identified 6 putative miR-132 targets by overlaying with the list of in silico predicted targets ( Figure 3B). Again, the identified proteins showed expression in both neuronal and glial populations in a previously published scRNAseq dataset in mouse hippocampus 72 (Table S1).

Characterization of putative miR-132 targets in the hippocampus
Next, we took a closer look into the predicted miR-132 targets identified in each of the datasets. Even though a total of 52 putative miR-132 targets were identified in all the datasets combined, only three predicted targets (Ptprc, Arf1, and Ssh2) were common between proteomics (miR-132 KD/OE combined) and transcriptomics (miR-132 KD/OE combined) ( Figures 4A, 4B, and Table S1). When considering the list of the 52 putative miR-132 targets, we observed an increase of experimentally identified putative targets expressed in microglia compared to the full set of 384 in silico predicted targets (12.9% in the experimental dataset compared to 4.8% in the in silico dataset) ( Figure 4C). For example, Ptprc, which is also one of the three common miR-132 targets between proteomics and transcriptomics, is a myeloid cell marker, showing specific microglial expression in the hippocampus. 72,[82][83][84] Together, these findings suggest that a subset of the identified putative miR-132 targets possibly drives gene expression programs in microglial populations.
In an attempt to understand the lack of correlation between the miR-132 targets identified in the proteomic and the transcriptomic datasets, we explored in more detail the type and prevalence of miR-132 binding sites in the 3 0 UTR of the putative targets. 1 First, we considered the proportions of binding site types present in targets that changed in the protein or RNA analysis and in predicted miR-132 targets that did not change in any of the datasets (non-changing targets) ( Figures 4D, 4E, and Table S1). Apart from a trend for an increase of 6-mer seeds in the experimentally identified targets compared to the non-changing subset, no striking differences in the proportion of binding site types across datasets were observed. We did, however, notice an overall lower abundance of perfect 8-mer seed sequences, and higher abundance of 6-mer seeds across datasets, confirming previous observations on miRNA seed type proportions in an in vitro large-scale RNA-seq dataset. 60 We then compared the total number of binding sites per target   Table S1). Again, the total amount of binding sites present per target did not differ across datasets, suggesting that seed (type) abundance cannot explain the differential effects of miR-132 on individual targets in our proteomics and transcriptomics datasets.

miR-132 regulates microglial activation in vivo
To characterize the microglial population present in our dataset, we subclustered all microglial cells and annotated them based on published datasets [85][86][87][88] (Table S4, ''ModuleScore''). We identified five distinct  iScience Article microglial subsets, annotated as homeostatic microglia, disease-associated microglia (DAM), interferonresponse microglia (IRM), cytokine-response microglia (CRM) and cycling microglia ( Figures 5A-5D, S3A, S3D, S3E, S3H, and Table S4). Even though we did identify a microglial subpopulation expressing IRMrelated genes (Ifitm3, Ifi27L2a, and Isg15) in the miR-132 OE dataset, this cluster did not recapitulate the full IRM response (and was therefore indicated as ''IRM-like''). Next, we interrogated the GO biological processes associated with the significantly enriched DEGs in each subcluster (Figures 5B, 5D, and Table S4). Consistent with their role in immune surveillance, homeostatic microglia-enriched genes were related to cell migration. DAM-related genes were associated with mitochondrial energy production and glucose metabolism, which is in line with their highly metabolically active phenotype and the metabolic shift from oxidative phosphorylation to glycolysis upon activation. [89][90][91][92][93] Both IRM and IRM-like populations were enriched for immune response-related pathways, while as expected, cycling microglia showed a strong enrichment for genes involved in cell cycle and division. CRM cells upregulated genes related to cell proliferation and transcription; at a closer look, we found them to also highly express a gene signature recently associated with ex vivo activated microglia 94 (exAM; Figures S3B, S3F, and Table S4), suggesting that they may represent a microglial subpopulation triggered by the cellular dissociation process involved in the scRNAseq workflow. Integration of the microglial datasets from the miR-132 KD and OE experiments ( Figure S3L) confirmed that cells from different samples and experiments are largely distributed across microglial subclusters, except for the cycling microglial population, which was only identified in the miR-132 OE experiment.
As miR-132 overexpression and knockdown triggered significant (Wilcoxon rank-sum test, Bonferroni-corrected p value <0.05) transcriptional responses in microglia (Table S4 for subcluster-specific expression changes), we next assessed whether these signatures could induce a phenotypic switch across microglial cell states. We therefore performed cell state composition analysis to determine the amount of cells present in each cell state, comparing miR-132 OE or miR-132 KD to their respective controls ( Figures 5E-5H). Interestingly, the DAM cluster was expanded upon miR-132 knockdown (scProportion test, Chi-squared test: p value = 0.0062) ( Figure S3C), whereas, reversely, it decreased upon miR-132 overexpression (scProportion test, Chi-squared test: p value = 0.0000004) ( Figure S3G). In addition, increased miR-132 levels resulted in significantly more cells in the IRM-like subcluster (scProportion Test, Chi-squared test: p value = 0.0007) (Figure S3G), suggesting a potential miR-132-regulated phenotypic shift from DAM to an IRM-like state. Next, we analyzed the DEGs (Wilcoxon rank-sum test, Bonferroni-corrected p value <0.05) and corresponding GO biological processes (Fisher's Exact test, FDR-adjusted p value <0.05) comparing miR-132 OE or KD to their respective controls, focusing on the DAM subcluster, which has high relevance for AD 85,88,95 (Figures 5I,  5J, S3I, S3J, and Table S4). Knockdown of miR-132 triggered significant upregulation of DAM-associated genes (e.g. Apoe, Fabp5, and Lgals3bp), and downregulation of homeostatic microglial genes (e.g. Csf1r, Hexb, and Cst3), consistent with the observed increase in the population of DAM cells. Interestingly, we  Figure S3 and  S3I, and S3J). miR-132-responsive transcripts in DAM showed enrichment in ribosomal genes associated with translation, in-line with previous reports profiling the DAM phenotype. 87,96 Apoe, one of the main risk factors of late-onset AD and a top expressor of the DAM transcriptional signature, was (even though not a predicted miR-132 target) among the top 4 most significantly upregulated genes upon miR-132 KD and, reversely, among the top 8 most downregulated genes after miR-132 OE in the microglial population, further supporting a role for miR-132 in gene expression programs regulating microglial cell state transitions ( Figure 5K). 85,88,[97][98][99] In addition, other AD genome-wide association studies (GWAS) genes were significantly differentially expressed in our microglial population upon miR-132 KD (Tpt1, p value = 5.3 10 -6 ) or miR-132 OE (Ms4a6d, p value = 0.018; Sdf2l1, p value = 0.006; Itgam, p value = 0.003; Siglech, p value = 0.048), suggesting an indirect regulation of AD risk genes by miR-132 ( Figure 5K). Even though only significant in the miR-132 KD experiment, Trem2, another critical risk factor for AD and key modulator of microglial function, 100-102 showed a trend for anticorrelation to miR-132 levels upon both overexpression and knockdown of miR-132 (Table S3). Of note, miR-132 depletion significantly induced inflammatory gene expression signatures also in IRM (Ptgs1, Hsp90b1) and homeostatic (Lgals3bp) microglial populations, while the expression of Ctsl, a DAM-associated gene, was significantly upregulated in homeostatic microglia. Reversely, miR-132 supplementation led to a significant reduction in inflammation-related genes in IRM-like (Hsp90ab1, Hsp90aa1, Hspa8) and homeostatic microglia (Cmtm8, Nav3, Tmem173) (Table S4).
To further understand the inhibitory effect of miR-132 on the DAM response, we performed single-cell trajectory analysis using Palantir. 103 Pseudotime ordering of the microglial population revealed the existence of two distinct phenotypic routes from homeostatic toward cellular states corresponding to DAM, IRM, and cycling microglia ( Figures 5L, 5M, and S3K). We then compared the expression profile of two DAM-associated genes (Apoc1 and Lgals3) along pseudotime between miR-132 KD or miR-132 OE cells and their corresponding controls ( Figure 5N). As expected, DAM gene expression collectively increased along the transcriptional route toward DAM. Interestingly, DAM gene expression levels were higher in miR-132 KD and lower in miR-132 OE cells compared to their controls, suggesting that, besides regulating the maintenance of the final pool of cells present in the DAM transcriptional profile, miR-132 also regulates DAM-related transcriptional programs during microglial transitions between homeostatic and DAM activation states. Employing the same approach as described for microglial cells, we also observed a decrease in the proportion of reactive oligodendrocytes upon miR-132 OE, an effect which was, however, not confirmed when miR-132 was knocked down (Figures S4, S5, and Table S5).
miR-132 shifts human iPSC-derived microglia from a disease-associated toward a homeostatic state We next validated our results in human iPSC-derived microglial cultures ( Figure 6A). miR-132 levels increased along microglial differentiation ($13-fold increase comparing macrophage precursors to iPSCs, $170-fold increase comparing microglia to iPSCs) ( Figure 6B), supporting a functional role for miR-132 in microglial maturation. Interestingly, the increase in miR-132 levels was similar to that observed in human iPSC-derived neurons, further pointing toward important endogenous functions of miR-132 in microglia ( Figure S1K). Notably, microglial cells in culture form a heterogeneous population, recapitulating several distinct transcriptional phenotypes linked to disease, and adapting in general a more activated phenotype. 86,104,105 To investigate the potential effects of miR-132 on cellular state transitions in human iPSCderived microglia, we treated differentiated microglia with a cholesterol-conjugated miR-132 mimic (or a control oligonucleotide) to increase miR-132 levels ( Figures 6A and 6C). No changes in miR-132 levels upon mock-treatment and in miR-212 levels upon miR-132 mimic treatment were observed (Figures S1L-S1M). To profile microglial states, we monitored the expression levels of markers of homeostatic microglia (P2RY12 and CX3CR1) and DAM (APOC1, CD9, and LGALS3) 85,86,106 ( Figures 6D and 6E). Increase of miR-132 levels significantly upregulated homeostatic microglial markers and downregulated DAM-associated genes, consistent with the miR-132-dependent anti-DAM response we observed in mouse hippocampus. Next, to gain insights into molecular targets that may be directly regulated by miR-132 in human microglia, we monitored the expression levels of several predicted miR-132 targets that we previously identified in the scRNAseq analysis ( Figure 4). Specifically, we focused on the predicted targets that were downregulated either in the microglial population (SSH2, PDIA3, DDX5, SEC62) or in the pseudo-bulked cells and were additionally highly expressed in microglia (MAF and CD164) ( Figure 6F). 72  iScience Article targets in human iPSC-derived microglia. Interestingly, DDX5, SSH2, and MAF, which have been previously linked to microglial homeostasis, [107][108][109] were all upregulated upon miR-132 overexpression ( Figure 6F), suggesting that they may be indirectly targeted by miR-132 to promote a homeostatic microglial phenotype. In addition, we assessed microglial morphological changes upon miR-132 increase, through immunolabeling of the microglial marker IBA1 ( Figure 6G). Control oligonucleotide-treated cells displayed a round, amoeboid morphology, typical of activated or reactive microglia in vitro. In contrast, microglia treated with the miR-132 mimic exhibited pronounced ramified morphology, as quantified by a significant increase in cellular endpoints ( Figure 6H), characteristic of homeostatic microglia partaking in immune surveillance through extension and retraction of mobile processes. 110,111 Together, these findings confirm that increase of miR-132 levels drives the transition of human microglia from a DAM-like state to a more homeostatic phenotype. Lastly, to gain further insights into miR-132-regulated pathways in iPSC-derived microglia, we monitored apoptosis-related genes, as the apoptotic biological process was one of the gene ontologies significantly enriched in microglia both in the miR-132 OE and KD scRNAseq experiments in mouse hippocampus ( Figures 2F and 3F). Caspase 1 (CASP1), a pro-apoptotic marker gene, was significantly upregulated upon miR-132 OE, while two anti-apoptotic regulators, BCL2 and XIAP, were downregulated compared to control-treated cells ( Figure 6I). Together, these data indicate that miR-132 may impact microglial phenotypes via-among others-the regulation of apoptotic programs both in mouse brain and in human microglial cultures.

DISCUSSION
Here, we employed a combinatorial proteomics and transcriptomics approach to profile broad regulatory effects of miR-132 in mouse hippocampus. Bulk proteomics analysis upon miR-132 overexpression and knockdown did not identify any significantly changing proteins. Also, transcriptional alterations in astrocytic and neuronal populations were limited. Yet, miR-132 modulation induced significant transcriptional responses in microglia and oligodendrocytes.
Currently, miRNA function is mostly inferred from studies using a ''one miRNA-one target'' approach. Yet, cumulative evidence from genome-wide studies suggests that miRNAs mainly function as regulators and fine-tuners of molecular networks. 50,112,113 In addition, miRNA-mediated repression of single targets is often relatively small (20-30%) and largely depends on the cellular context. 56,57,114,115 However, the profiling of broad miRNA-regulated gene networks (involved in cellular stress, apoptosis, inflammation, synaptic plasticity, and neuronal network activity) in mammalian brain has recently provided proof-ofconcept for the utility of network-based approaches to drug discovery in neurological diseases. 50,116 We observed limited effects on individual putative miR-132 targets upon miR-132 modulation in the mouse hippocampus. In addition, we did not depict significant correlation between individual transcripts and proteins across experimental conditions. Several factors could account for these findings, including -among others-differential (cell type-specific) effects of miR-132 on protein and mRNA targets, low overall correlation between proteome and transcriptome, 117 technical differences in sensitivity, resolution and noise iScience Article between the bulk proteomics, and scRNAseq approaches. Concordant changes in the mRNA and protein levels of miRNA targets are mostly observed for targets repressed by more than 50% (log2 > 0.58) by a miRNA, upon nearly perfect miRNA/mRNA binding 41 and primarily in in vitro systems. Notably, in both our proteomics and transcriptomics datasets, we observed an overall lower frequency of perfect 8-mer seed sequences in putative miR-132 targets, which could be indicative of potentially smaller repression rates, and thus, potentially explain the lack of consistent effects between proteome and transcriptome. In addition, the total amount or type of binding sites present per target did not differ across our datasets. These observations suggest that perhaps other factors (possibly related to the cellular microenvironment, the 3 0 UTR context and accessibility or the existence of non-canonical binding sites and alternatively spliced mRNA isoforms, and biological noise) may determine target prioritization for miR-132 targeting in the complex hippocampal tissue. 1,[118][119][120][121][122] In iPSC-derived microglia, two predicted miR-132 targets identified by scRNAseq in mouse hippocampus were significantly downregulated upon induction of miR-132 levels. SEC62 has been implicated in protein translocation in the endoplasmatic reticulum, 123 while CD164 positively affects chemotaxis. 124 Yet, how these changes contribute to the observed phenotypic shift still remains to be determined.
miR-132 has been previously proposed to act as a NeurimmiR, a collective term referring to miRNAs acting at the interface between the neuronal and the immune system. 125 With respect to its neuronal functions, miR-132 has been implicated in neuronal morphogenesis and survival, synaptic plasticity, and cognition. 23,28,31,33-35, 80 We recently reported a pro-neurogenic role for miR-132 in the adult hippocampus, where it regulates progenitor proliferation, neuronal differentiation, and maturation of newly born neurons, while also providing neurotrophic and neuroprotective support. 35 On the other hand, miR-132 overexpression suppresses the inflammatory response in monocytes, macrophages, and astrocytes in vitro, while it also represses the secretion of pro-inflammatory cytokines by targeting acetylcholinesterase in the periphery. [126][127][128] Interestingly, microglial miR-132 levels increase in human and rat brain after epileptic seizures, suggesting a possible role for miR-132 in the neuroinflammatory response under these conditions. 129 Yet, there is currently no evidence of direct regulation of the innate immune response by miR-132 in the brain. Cell type-specific expression profiling showed that miR-132 is highly expressed in hippocampal microglia already at baseline. Earlier evidence also suggested that miR-132 is among the top 14% most abundant miRNAs in microglia. 130 Taken that only the most highly expressed miRNAs in a given cell type have been proposed to significantly regulate targets in that particular cellular context, 131 miR-132 expression levels in microglia suggest that it may have the potential to effectively repress its targets and thereby exert a functional regulatory effect on microglial function. Indeed, miR-132 overexpression restricted the proportion of a disease-associated microglial population, while, reversely, miR-132 depletion significantly increased this microglial fraction. miR-132 levels were reversely associated with DAM gene expression signatures along unbiased pseudotime ordering of microglia from a homeostatic to a DAM cell state. Validation of these observations in human iPSC-derived microglia (where endogenous miR-132 levels increase approximately 200-fold compared to iPSCs), confirmed that miR-132 shifts human microglia from a disease-associated toward a homeostatic phenotype. Interestingly, an anticorrelation between miR-132 levels and inflammatory gene signatures was not only observed in DAMs, but also in other microglial subclusters, like homeostatic microglia, in vivo. We also observed significant changes in the levels of apoptosis-related genes in the iPSC-derived microglial cultures, suggesting that miR-132 could mediate its effects on microglial activation and homeostasis via -among others-regulating apoptosis in microglial (sub)populations. [132][133][134][135] Whether the impact of miR-132 regulation on microglial populations in vivo is partially also effectuated by non-cell-autonomous signaling from other cell types onto microglia, like for instance via neuronal exosomes, 136 cannot be ruled out. miR-132 overexpression experiments in neurons co-cultured with microglia could provide valuable insights in that regard. In addition, we observed a similar effect of miR-132 in oligodendrocytes, where we identified a reactive population characterized by high Serpina3n and C4b expression. 137,138 miR-132 overexpression resulted in a decrease of this reactive subcluster and repressed inflammatory genes within this population. Together, these findings support a role for miR-132 in the immune response in the hippocampus.
Recently, comprehensive single-cell transcriptomic studies in AD mouse models and human brain revealed complex activation profiles of distinct microglial cell states with significant roles in AD pathogenesis. [85][86][87][88]139,140 Low proportions of activated microglial subpopulations are also present in wild-type mouse brain, as we have observed, suggesting that these clusters represent physiological states of microglia. 88 DAMs are a microglial subset expressing many of the AD risk genes identified in genome-wide iScience Article association studies (GWAS), like TREM2 and APOE, the largest risk factor for AD. 85,88,97,101,102,[141][142][143] Interestingly, we observed a significant reverse association between the levels of miR-132 and those of ApoE and other AD risk genes in the whole microglial population, suggestive of the modulating effect of miR-132 on the DAM phenotypic response.
Our goal here was to understand which targets and biological processes miR-132 would affect in the hippocampus, as a first step toward exploring what would be the limitations of targeting miR-132 in AD. We and others have previously shown that miR-132 exerts a wide range of neuroprotective functions, ameliorates amyloid, and Tau pathologies and improves memory in AD mouse models. 23,32,33,35 While the current work mainly confirms that miR-132 supplementation can exert widespread effects across several cellular populations in the brain, it also suggests that such responses are likely going to be subtle. One possible avenue of concern is the regulatory effect of miR-132 overexpression on the equilibrium between different microglial cell states. Microglias are important carriers of genetic risk in AD 144 and it is, therefore, currently unclear whether fixing microglia in a more homeostatic phenotype would be beneficial or detrimental in the disease. [145][146][147][148] Targeting miR-132 overexpression or knockdown specifically to microglial (sub)populations in the brain may therefore be of particular biological and therapeutic relevance in the future.

Limitations of the study
Our study has three main limitations. First, given the relative subtle changes, limited sample size and interindividual variability could have partially confounded some of our observations. Proteomic changes upon miR-132 knockdown or overexpression did not retain significance after correction for multiple testing. For instance, inflammation-related proteins, like Lgals3bp, Ifit1/3, and Serpina3n, even though not significant, were among the top 5% anticorrelated to miR-132. While this may reflect the overall more subtle effect of miRNAs on proteome 43,149 than on the transcriptome, it could also be the result of the dilution of cell typespecific effects in whole hippocampal lysates (which contain different brain structures, like blood vessels and ventricles, and several other cell types, such as -among others-endothelial and epithelial cells, pericytes, macrophages), which might be overcome by using larger sample sizes. Most importantly, we did not observe many significant transcriptional changes in any of the neuronal populations analyzed via scRNAseq upon miR-132 OE (despite the robust miR-132 overexpression), which is in contrast with the well documented neuronal functions of miR-132, and could therefore most likely be attributed to the technical difficulties of isolating distinct neuronal subpopulations. This could have contributed to the relatively small numbers of neurons in our scRNAseq dataset, which may have compromised the power of the downstream analyses. In addition, during the process of neuronal isolation, axons and dendrites, which represent cellular compartments with high miR-132 expression, 36,150,151 may be lost, impacting the precise estimation of miR-132 (and putatively also target) levels both at baseline, but also after miR-132 manipulation. Moreover, the observed lack of miR-132 knockdown in neurons could potentially reflect a less productive oligonucleotide uptake, leading to inefficient target knockdown. 152 Some of the enriched biological pathways that were identified in the dataset and were related to neuronal function do not necessarily reflect cellautonomous signaling, as other cell types like microglia or oligodendrocytes might affect neuronal function. Interestingly, only small numbers of significant DEGs and no putative miR-132 targets were identified in the astrocytic cluster, despite processing adequate numbers of isolated cells (miR-132 KD, 3456; miR-132 OE, 3575) [i.e. comparable numbers to oligodendrocytes (miR-132 KD, 6004; miR-132 OE, 2928) and microglia (miR-132 KD, 5603; miR-132 OE, 2127)]. These observations may suggest that astrocytes are potentially less amenable to miR-132 regulation, at least under these experimental conditions.
Second, we have here employed wild-type mice and healthy donor iPSC-derived microglial cultures, in order to assess the regulatory output of miR-132 under physiological conditions. While essential for defining a therapeutic window in the future, our findings warrant further research into the immunomodulatory effects of miR-132 in an AD-relevant context.
Third, our proteomics and transcriptomics approaches do not enable discrimination between direct and indirect miR-132 effects. Binding assays (e.g., reporter assays, CLIP) can identify direct targets, yet do not offer any information on the functional output of these interactions. On the other hand, genomewide expression profiling of the transcriptome or the proteome upon miRNA modulation cannot discriminate between primary and secondary effects but does allow for global understanding of the biological pathways regulated by a miRNA. 42 Taking these into consideration, we here employed an integrative, multimodal approach to broadly profile the functional implications of miR-132 regulation in the     . Due to the differences in chemical modifications of the mimic and antagomiR oligonucleotides, different treatment times were employed for each experiment. [156][157][158] One week (miR-132 KD) or 48 h (miR-132 OE) after injection, mice were sacrificed and the injected right hippocampus was microdissected. For proteomics analysis, the hippocampi were snap frozen in liquid nitrogen and stored at À80 C until further processing. For transcriptomic analysis, mice were first perfused and isolated hippocampi were immediately further processed (see 'hippocampal cell dissociation').

Proteomics
Preparation of samples for mass spectrometry iScience Article 15 min and protein concentrations were measured by using Bradford assay. Samples were diluted by the addition of SDS to a final concentration of 5%, tris(2-carboxyethyl)phosphine (TCEP) to a final concentration of 10 mM and triethylammonium bicarbonate (TEAB) to a final concentration of 50 mM. Samples were alkylated in the dark for 1 h by the addition of 2-iodoacetamide (IAA) at 20 mM. Protein lysates were prepared for mass spectrometry using the s-trap mini plate format according to the manufacturer's instructions (Cat#NC1828287, Protifi). In summary, for each sample 50 mg of protein was loaded onto an s-trap 96-well plate. Captured protein was washed 5 times with 200 mL of wash buffer (90% methanol with 100 mM TEAB, pH 7.1). Proteins were digested by the addition of 5 mg of trypsin to each sample in 50 mM ammonium bicarbonate. Samples were digested for 2 h at 47 C. Once digestion was complete, peptides were eluted with 80 mL of 50 mM ammonium bicarbonate followed by 80 mL of 0.2% formic acid and lastly with the addition of 80 mL 50% acetonitrile with 0.2% formic acid. After the addition of each elution buffer, the plate was centrifuged at 1500 g for 2 min and the flowthrough was collected. Eluted peptides were dried by speedvac and suspended in 1% formic acid before quantification using the CBQCA assay (Cat#C6667, Invitrogen).

Mass spectrometry
For each sample, 1.5 mg of peptide was analyzed by DIA. Peptides were injected onto a nanoscale C18 reverse-phase chromatography system (UltiMate 3000 RSLC nano, Cat#ULTIM3000RSLCNANO, Thermo Fisher Scientific) and electrosprayed into an Orbitrap Exploris Mass Spectrometer (Thermo Fisher Scientific). For liquid chromatography, the following buffers were used: buffer A (0.1% (v/v) formic acid in Milli-Q water) and buffer B (80% (v/v) acetonitrile and 0.1% (v/v) formic acid in Milli-Q water). Samples were loaded at 10 mL/min onto a trap column (100 mm 3 2 cm, PepMap nanoViper C18 column, 5 mm, 100 Å , Cat#17294434, Thermo Fisher Scientific) equilibrated in 0.1% trifluoroacetic acid (TFA). The trap column was washed for 3 min at the same flow rate with 0.1% TFA then switched in-line with a resolving C18 column (75 mm 3 50 cm, PepMap RSLC C18 column, 2 mm, 100 Å , Cat#164563, Thermo Fisher Scientific). Peptides were eluted from the column at a constant flow rate of 300 nL/min with a linear gradient from 3% buffer B to 6% buffer B in 5 min, then from 6% buffer B to 35% buffer B in 115 min, and finally to 80% buffer B within 7 min. The column was then washed with 80% buffer B for 4 min and re-equilibrated in 3% buffer B for 15 min. Two blanks were run between each sample to reduce carry-over. The column was kept at a constant temperature of 50 C.
The data were acquired using an easy spray source operated in positive mode with spray voltage at 2.445 kV, and the ion transfer tube temperature at 250 C. The MS was operated in DIA mode. A scan cycle comprised a full MS scan (m/z range from 350 to 1650), with RF lens at 40%, AGC target set to custom, normalized AGC target at 300%, maximum injection time mode set to custom, maximum injection time at 20 ms, microscan set to 1 and source fragmentation disabled. MS survey scan was followed by MS/MS DIA scan events using the following parameters: multiplex ions set to false, collision energy mode set to stepped, collision energy type set to normalized, HCD collision energies set to 25.5, 27 and 30%, orbitrap resolution 30000, first mass 200, RF lens 40%, AGC target set to custom, normalized AGC target 3000%, microscan set to 1 and maximum injection time 55 ms. Data for both MS scan and MS/MS DIA scan events were acquired in profile mode.

Analysis of mass spectrometry data
Raw mass spec data files were searched employing Spectronaut (Biognosys) version 14.7.201007.47784 using the directDIA function. The following search settings were used: minimum peptide length 7, maximum peptide length 52, cleavage enzyme trypsin, maximum missed cleavages 2, protein and peptide FDR was set at 0.01, profiling and cross run normalization were disabled. Carbamidomethyl (C) was selected as a fixed modification while Acetyl (N-term), Deamidation (NQ) and Oxidation (M) were selected as variable modifications. Data were searched against a mouse database from Uniprot release 2020 06. This database consisted of all manually annotated mouse SwissProt entries along with mouse TrEMBL entries with protein level evidence and a manually annotated homologue within the human SwissProt database.

Hippocampal cell dissociation and fluorescence-activated cell sorting (FACS)
Following injection of miR-132 mimic/control oligonucleotides (miR-132 OE) or miR-132 antagomiR/control oligonucleotides (miR-132 KD), mice received an overdose of pentobarbital and were perfused with ice-cold aCSF. The aCSF was composed of (in mM): 93 NMDG, 2.5 KCl, 1.2 NaH2PO4, 30 NaHCO3, 20 HEPES, 25 glucose, 5 sodium ascorbate, 2 thiourea, 3 sodium pyruvate, 10 MgSO4, 0.5 CaCl2, and was ll OPEN ACCESS iScience Article Data processing Proteomics Sample KD1-1 was determined as an outlier in an initial principal component analysis and was thus removed from count and metadata tables, and then proteins not observed in at least 75% of samples in each experimental group were removed from the count table. Counts were RPM normalized before running differential expression analysis via Limma-Voom between miR-132 OE/KD groups and corresponding control groups. Voom estimates the mean-variance relationship and computes observation-level weights before a linear model is fit for each protein (additionally, the 75% protein detection threshold was decided based on voom mean-variance distribution). Empirical bayes moderated t-tests were run to determine differential expression comparing miR-132 KD/OE to corresponding control-injected samples.
Benjamini-Hochberg correction was applied to p values.

Single-cell RNA sequencing
Raw 10x Genomics data were processed using the Cellranger count command (v3.1.0), mapping to the GRCm38 assembly (ensembl 97 annotation). Filtered gene expression matrices were loaded into Scanpy (v.1.8.1) 159 and further filtered to remove cells expressing less than 500 genes or with greater than 15% of UMIs being assigned to mitochondrial genes, Scrublet was then used to remove cells likely to be doublets (doublet_score >0.25), and finally, samples were then combined into two datasets, miR-132 OE and miR-132 KD. Following QC and filtering, samples were normalized to a total of 10000 UMIs per cell and log normalized; highly variable genes were detected using default parameters and the data were scaled to unit variance with a mean of 0, values greater than 10 were clipped. A principal component analysis was performed on the highly variable gene matrix and Harmony 77 was used for batch correction between runs, 50 and 40 components were used respectively for miR-132 OE and miR-132 KD for downstream dimensionality reduction and clustering analyses. Clustering was performed at various resolutions using the Leiden algorithm and clusters were annotated with cell types based on their marker gene expression. Infiltrating immune cells segregated as distinct clusters and were removed from the analysis based on the expression of markers for monocytes (Cd44, Fn1, Gda), macrophages (Atf3, Cfp, Serpinb8), dendritic cells (Flt3) and lymphocytes (Lck). 84,160,161 Cells from cell types we did not intend to sort per RUN, were excluded from further analysis [RUN 1: all hippocampal cells; RUN 2: only neurons and astrocytes; RUN 3: all hippocampal cells; RUN 4: only neurons and astrocytes; RUN 5: only neurons and astrocytes (Table S3)]. After removal of these cells, the analysis was repeated starting with the detection of highly variable genes, 46 and 42 principal components were used respectively for miR-132 OE and miR-132 KD for filtered datasets.

Identification of cell subtypes and differences in cell proportions across conditions
Processed and annotated datasets (miR-132 OE and miR-132 KD) were loaded into the Seurat pipeline 162,163 using the SeuratDisk package (https://mojaveazure.github.io/seurat-disk/), for subclustering analysis. To better identify and annotate cell subtypes and states, the major cell types of interest (microglia and oligodendrocytes) were subset from the original dataset and processed separately using the subset function of Seurat. The Seurat package (v4. To determine changes in cell subtype proportions of miR-132 KD/OE versus control, the number of cells per subtype was normalized against the total number of cells into the respective major cell type per condition, implemented by Cluster_Stats_All_Samples function of scCustomize package. 164 Significant changes in proportions were determined using scProportionTest library (https://github.com/rpolicastro/ scProportionTest) and Chi-squared test. Proportional changes were considered significant at p value <0.05.

Pseudotime
Palantir (v1.1) 103 was used to perform pseudotime analysis. In short, microglial clusters along the lineage were extracted and an appropriate start cell was selected based on expression of markers expected early in the lineage, i.e. a microglial cell with the highest expression of homeostatic marker genes in this case. Palantir was then run on each subset using the Harmony corrected principal component analysis values as the base representation (n_components = 5, knn = 30, n_eigs = 3, num_waypoints = 500). After pseudotime calculation, Palantir objects were separated according to condition, in order to compare the effect of condition on pseudotime.

Pathway enrichment
To identify biological pathways regulated by miR-132 both at the level of the protein and RNA, we performed functional enrichment analysis using the Database for Annotation, Visualization and Integrated Discovery (DAVID). 64,65 For proteomics and single-cell transcriptomics (divided in oligodendrocytes, microglia, astrocytes, granule cells and hippocampal pyramidal neurons), differentially expressed proteins or genes were ranked based on their p value and the top 1000 up-and downregulated proteins or genes were used as input for pathway analysis. For functional analysis of the identified 384 predicted miR-132 targets, all 384 targets were used as input. Both GO Biological Process and KEGG pathways were processed and the pathways with an FDR-adjusted p value <0.05 (Fisher's Exact test, FDR-adjusted p values <0.05 considered significant) and a count of genes R1% were selected. Pathways were ranked based on their significance and the top 20 per condition (cell type for transcriptomics and proteomics) were selected for further processing. After pooling together pathways with high functional overlap, both gene count and significance of the largest subcategories were plotted.
To determine which biological pathways were specific to a certain cell state or type, we extracted all the marker genes per cell state/type that were upregulated and had an FDR-adjusted p value <0.05 (Fisher's Exact test, FDR-adjusted p values <0.05 considered significant) and uploaded those into DAVID. GO Biological Processes were selected based on significance (FDR-corrected p value <0.05) and only the top 5 pathways that were unique to that cell state/type were displayed.
After differential analysis comparing miR-132 KD/OE versus control in DAM microglia and reactive oligodendrocytes, pathway analysis was performed as described here. Briefly, DEGs were ranked on p value and the top 500 anti-correlating (upregulated after miR-132 KD and downregulated after miR-132 OE) genes were used as input in DAVID. The top 5 GO Biological Processes that had an FDR-adjusted p value <0.05 (Fisher's Exact test, FDR-adjusted p values <0.05 considered significant) were shown.

Identification and cell type-specific expression of miR-132 predicted targets
Identification of in silico predicted miR-132 targets was performed using Encori. Targets predicted by at least 2 out of 7 (microT, miRanda, miRmap, PITA, RNA22, PicTar, TargetScan) target prediction algorithms were intersected with experimental AGO-CLIP databases (using a high-confidence threshold, Table S1).
To determine the cell type-specific expression profile of miR-132 predicted targets, we used both our inhouse mouse hippocampus scRNAseq dataset and the previously published mouse dentate gyrus dataset. 72 Three independent scientists manually annotated the expression profile of the predicted targets based on relative expression distribution across cell types (http://linnarssonlab.org/dentate) and consensus expression was plotted (Table S1).

Microglial dataset integration
Microglia datasets from the two experiments (miR-132 OE and miR-132 KD) were integrated using the Seurat integration analysis (CCA algorithm) ( Figure S3L). In order to do so, genes that were repeatedly ll OPEN ACCESS iScience 26, 106829, June 16, 2023 29 iScience Article variable across two datasets were selected with the function SelectIntegrationFeatures. Subsequently, cross-dataset cell-pairs that were in a matched biological state were identified using the function FindIntegrationAnchors followed by integrating these two datasets with the function IntegrateData. Dimension reduction analysis was performed on the integrated datasets with the function RunPCA and RunUMAP. The clustering analysis was accomplished with the functions FindNeighbors and FindClusters.

AD GWAS gene analysis
To identify possible miR-132-regulated AD risk loci, we made use of a recent study 153 that reports a large genome-wide association meta-analysis of clinically diagnosed late-onset AD (94,437 individuals) to identify AD risk loci. MAGMA 165 first maps SNPs to genes based on genomic proximity, then tests for association of the gene to a trait. Here, Stage 1 summary statistics generated by Kunkle et al., 153 were analyzed with MAGMA v1.07, using the NCBI GRCh37 build and the European 1000 genomes as references. We set the annotation window (genomic region) at 35 kb upstream and 10 kb downstream of the gene boundaries. We then converted the list of genes output from MAGMA to their mouse orthologs using Ensemble Biomart Release 94, selecting only genes with strict one-to-one orthologs. Employing an adjusted p value cutoff of 0.01, we extracted a list of 571 GWAS genes (Table S4, ''AD GWAS genes''), and compared it to the significant DEGs identified in our microglial population in both miR-132 KD and miR-132 OE datasets.

miR-132 mimic treatment of human iPSC-derived microglia
Differentiated microglia were treated with a synthetic miR-132 mimic oligonucleotide (Batch ID 60374517, cholesterol-conjugated, Janssen Pharmaceutica) or corresponding control oligonucleotide at 1500 nM and collected after 1 week of treatment for RNA extraction or fixed on a coverslip with 4% PFA for immunocytochemistry.

RNA extraction, reverse transcription and semi-quantitative real-time PCR
RNA isolation from the mouse hippocampus was performed using the miRVana Paris Kit (Cat#AM1556, Life Technologies) according to the manufacturer's instructions. Briefly, using 350 mL of cell disruption buffer (containing protease and phosphatase inhibitors) the tissue was homogenized. After denaturation, acid phenol:chloroform was added and the samples were incubated and spun down, resulting in the formation of 2 phases. The upper aqueous phase was supplemented with 1.25 volumes of 100% ethanol and loaded on miRVana spin columns. Following consecutive washing steps, RNA was eluted from the column and concentration and purity were measured using NanoDrop. For the isolated brain cells and collected iPSCderived microglia, RNA was extracted using the miRNeasy Micro Kit (Cat#217084, Qiagen) following manufacturer's instructions.
Reverse transcription of miRNAs was performed with 100 ng RNA using the miRCURY LNA RT Kit (Cat#339340, QIAGEN). For the reverse transcription of mRNAs, 200 ng RNA was processed using the Superscript II reverse transcriptase (Cat#18064071, Thermo Fisher Scientific).
For miRNAs, real-time semi-quantitative PCR was performed using the Sybr Green mastermix (Cat#1708880, QIAGEN) and miRCURY LNA primers (Cat#204129, QIAGEN). The mean expression of two small-RNA housekeeping genes, U6 snRNA (Cat#203907, QIAGEN) and RNU5G (Cat#203908, QIAGEN), was used for normalization. Coding transcripts were analyzed using the SensiFast Sybr No-Rox kit (Cat#BIO-98020, Bioline) and GAPDH, UBC and PSMB4 were used as housekeeping genes for human mRNAs. Gapdh and Actin were used as housekeeping genes for mouse mRNAs. Primer sequences can be found in Table S6. Ct values were determined using the second derivative method and subsequently fold changes were calculated using the DDCt method.

Immunocytochemistry of iPSC-derived microglia
PFA-fixed microglial cells on coverslips, were initially washed with 1% (v/v) Triton X-100 in PBS and incubated with blocking buffer, consisting of 1% (v/v) Triton X-100, 10% (v/v) normal goat serum in PBS, for 2 h at room temperature. The primary antibody (anti-IBA1 (Rabbit), 1:100, Cat#019-19741, WAKO) incubation was performed in 0.3% (v/v) Triton X-100, 3% (v/v) normal goat serum in PBS at 4 C overnight, followed by incubation with the secondary antibody (Goat anti-rabbit 488, 1:500, Cat#111-487-003, Dylight) for 2 h in the dark at room temperature. Finally, cells were washed with 1% (v/v) Triton X-100 in PBS, stained in the dark with DAPI (1:5000) and mounted in Mowiol. iScience Article Image acquisition, processing and analysis Images of the miR-132-treated microglia were acquired using a Leica TCS SP8 X confocal microscope. Images were acquired at 203 magnification with an additional digital zoom of 2.66, speed of 100, and z stack size of 20 mm. For image processing, the FIJI software was employed. 166 To obtain overlaid images from different channels, we used the 'Merge Channel' function in FIJI. The number of endpoints per cell was quantified from immunofluorescent confocal images of IBA1-immunolabelled treated iPSC-derived microglia using ImageJ.

QUANTIFICATION AND STATISTICAL ANALYSIS
Statistical analysis was performed as indicated in the figure legends. Significance was set at (adjusted) p value threshold of 0.05.