Fetal brain response to maternal inflammation requires microglia

ABSTRACT

In utero infection and maternal inflammation can adversely impact fetal brain development. Maternal systemic illness, even in the absence of direct fetal brain infection, is associated with an increased risk of neuropsychiatric disorders in affected offspring. The cell types mediating the fetal brain response to maternal inflammation are largely unknown, hindering the development of novel treatment strategies. Here, we show that microglia, the resident phagocytes of the brain, highly express receptors for relevant pathogens and cytokines throughout embryonic development. Using a rodent maternal immune activation (MIA) model in which polyinosinic:polycytidylic acid is injected into pregnant mice, we demonstrate long-lasting transcriptional changes in fetal microglia that persist into postnatal life. We find that MIA induces widespread gene expression changes in neuronal and non-neuronal cells; importantly, these responses are abolished by selective genetic deletion of microglia, indicating that microglia are required for the transcriptional response of other cortical cell types to MIA. These findings demonstrate that microglia play a crucial durable role in the fetal response to maternal inflammation, and should be explored as potential therapeutic cell targets.

Highlighted Article: A comprehensive transcriptional atlas of microglia development demonstrating that microglia are necessary for widespread cortical gene expression changes triggered by in utero inflammation.

INTRODUCTION

Maternal infections and inflammation can disrupt fetal brain development and lead to significant neurological and psychiatric problems in affected offspring (Yamashita et al., 2003; Hornig et al., 2018). Although some pathogens directly infect fetal brain cells, widespread maternal immune system activation and cytokine release in the absence of direct central nervous system (CNS) infection in the fetus are also detrimental to neurodevelopment (Kim and Shresta, 2016). Maternal fever, elevated serum cytokine levels and infection with pathogens that do not cross the placenta all increase the risk of problems such as schizophrenia and autism in affected offspring (Adams et al., 1993; Buka et al., 2001; Brown et al., 2004; Ellman et al., 2010; Hornig et al., 2018). The molecular mechanisms that underlie this association are not well understood.

Maternal immune activation (MIA) in rodents is a well-characterized model for in utero inflammation in which pregnant mice are injected with an immune stimulant such as polyinosinic:polycytidylic acid (polyI:C) – a double-stranded RNA molecule that mimics viral infection (Hameete et al., 2020). The offspring of mice subjected to MIA display autism-like phenotypes that are most pronounced in male mice, including abnormal ultrasonic vocalizations and altered interactions with novel objects and mice (Dunaevsky and Bergdolt, 2019). MIA affects dendritic spine density and cortical lamination, and leads to global changes in protein expression in the developing cortex via activation of the integrated stress response (Coiro et al., 2015; Wu et al., 2018; Kalish et al., 2021). However, the specific cellular targets of MIA in the developing brain that mediate autism-like behavioral phenotypes observed in juvenile and adult mice are unknown.

Microglia, the resident phagocytic cells of the nervous system, are poised to respond to a variety of early insults and express receptors for some of the cytokines implicated in MIA, including interleukin-17a (IL-17a) and interleukin 6 (IL-6) (Smith et al., 2007; Choi et al., 2016). Microglia originate from myeloid precursors in the yolk sac and begin to populate the embryonic mouse cortex around embryonic day 9 (E9) (Alliot et al., 1991). Recent studies have demonstrated that microglia play a significant role in the cellular pathophysiology of congenital brain infection with Zika virus and cytomegalovirus, and that MIA impacts microglial motility and the ability to respond to immune stimuli in adulthood (Sakao-Suzuki et al., 2014; Lum et al., 2017; Cloarec et al., 2018; Ozaki et al., 2020; Xu et al., 2020; Hayes et al., 2022). However, it is unknown how MIA affects microglial gene expression in utero or whether microglia are necessary for the transcriptional changes observed across other embryonic cell types in the developing brain in response to MIA (Kalish et al., 2021). Here, we show that MIA alters embryonic and juvenile microglial gene expression profiles, and that microglia are necessary for mediating the effects of maternal inflammation on neighboring neurons and neuronal progenitors in the fetal cerebral cortex.

RESULTS

Pathogen and cytokine receptor expression across cell types in the developing cerebral cortex

We sought to confirm the expression of receptors relevant to maternal infection and inflammation in microglia, and to identify other cell types that may directly respond to these stimuli. We first interrogated a previously generated single-cell RNA sequencing (scRNA-seq) dataset for expression in embryonic cerebral cortex of receptors for pathogens known to cross the placenta and cause fetal infection (termed TORCH pathogens) and of receptors for cytokines involved in the inflammatory response to maternal infections (Smith et al., 2007; Scott et al., 2008; Choi et al., 2016; Chamera et al., 2020; Bordt et al., 2021; Di Bella et al., 2021; Gordon, 2021; Megli and Coyne, 2022; Taglauer et al., 2022) (Fig. 1A). To generate this dataset, the full thickness of the developing somatosensory cortex was isolated and all embryonic cortical cell types were profiled with high temporal resolution from E12.5 through P1 (Di Bella et al., 2021). We found that receptors implicated in MIA were robustly expressed in microglia, and absent in progenitor and neuronal cell types, including Tlr3, the receptor for Poly(I:C), and Il17ra, the receptor for IL-17a, which plays a crucial role in mediating behavioral phenotypes in MIA (Choi et al., 2016). We calculated a gene module score to quantify cumulative expression of TORCH and cytokine receptors over time, henceforth referred to as CytoTORCH module score (Fig. 1B and Materials and Methods). Microglia displayed a higher score than any other cell type in the developing cortex [P<0.05 for ANOVA and for pairwise comparisons between microglia and all other cell types using Tukey's Honest Significance tests (HSTs)]. Several non-neuronal cell types, including pericytes and endothelial cells, also displayed a relatively high CytoTORCH module score. Overall, these data suggest that microglia are poised to act as immediate responders to a broad range of inflammatory and infectious insults in the developing cortex.

Fig. 1.

Brain transcriptional changes after MIA require microglia. (A) Simplified schematic of sample processing pipeline from Di Bella et al. (2021) and bar plots of TORCH pathogen and cytokine receptor expression (mean log counts per million) in each cell type across development. (B) Violin plot of CytoTORCH module score for each cell cluster (MG, P<0.05 for ANOVA and for pairwise comparisons between microglia and all other cell types using Tukey's HST). (C) Overview of experimental design (top) and UMAP visualization of snRNAseq data from the developing somatosensory cortex of E15.5 wild-type and Csf1r^−/− mice after saline or Poly(I:C) maternal injection at E12.5 (two or three embryos per condition from a total of four pregnant dams, 42,736 total cells) (bottom). Cells are colored by cell type assignment. (D) Dot plots of log(fold change) in gene expression between MIA and saline for cell types with significant changes at E15.5. DEG analysis was performed using NEBULA. AP, apical progenitors; IP, intermediate progenitors; ImmN, immature neurons, MN, migrating neurons; (UL/DL) CPN, (upper layer/deep layer) callosal projection neurons; SCPN, subcerebral projection neurons; CThPN, corticothalamic projection neurons; LIV, layer IV stellate neurons; IN, interneurons; CR, Cajal-Retzius cells; Astro, astrocytes; CGC, cycling glial cells; Epend; ependymal cells; OL, oligodendroglia; MG, microglia; Peri, pericytes; EC, endothelial cells; VLMC, vascular and leptomeningeal cells. Created with BioRender.com.

Microglia mediate the response to immune activation of other cell types in the developing cortex

To determine whether microglia directly impact how other embryonic neocortical cell types respond to maternal inflammation, we performed single-nucleus RNA sequencing (snRNA-seq) of the developing somatosensory cortex in Csf1r null mice and wild-type littermate controls after MIA or sham saline injection at E12.5 (Fig. 1C). Csf1r null mice have grossly normal embryonic brain development but lack microglia because the encoded receptor, CSF1R, is required for microglial development (Erblich et al., 2011). Microglia are the only cell types in the CNS known to express CSF1R; therefore, the effects of Csf1r knockout on the brain can be attributed to the loss of microglia (Raivich et al., 1998; Sierra et al., 2007). We focused our analysis on the developing somatosensory cortex, due to a previous detailed developmental characterization of the cell types in this region and the demonstrated abnormalities in the somatosensory cortex in the setting of MIA and autism (Khan et al., 2015; Reed et al., 2020). We analyzed male mice only, given the sex-dependent effects of MIA at both the molecular and behavioral level (Xuan and Hampson, 2014; Gogos et al., 2020). An appropriate maternal inflammatory response to Poly(I:C) injection was confirmed via cytokine array profiling (Fig. S1A). Gene expression was analyzed at E15.5, to allow sufficient time (72 h) for microglia to express cell signaling factors that may affect surrounding cell types. After quality control and doublet filtering (Table S1 and Materials and Methods), we analyzed the expression profiles of 42,736 cells. We clustered the cells using the Louvain algorithm and annotated the clusters post hoc, based on expression of canonical marker genes. We identified 16 clusters, with identities and transcriptional profiles corresponding to nine previously described cell types at E15.5 (Di Bella et al., 2021), including apical progenitors (APs) and intermediate progenitors (IPs), migrating neurons, major subtypes of cortical projection neurons and interneurons, and Cajal Retzius cells (Fig. S1B). As expected, we observed a small cluster representing microglia in wild-type animals; this cluster was the only cluster that expressed Csf1r, and this cluster was nearly abolished in Csf1r-null mice (Fig. S1C,D). The absence of microglia in Csf1r-null mice was confirmed by immunohistochemistry (Fig. S1E). MIA did not alter non-microglial cell type distributions in wild type or Csf1r-null mice (binomial generalized linear mixed effects model, Fig. S1D).

Differentially expressed genes (DEGs) between MIA and saline control for each cell type were identified by negative binomial mixed model using a large-sample approximation (NEBULA) (He et al., 2021). In agreement with previous findings, MIA led to significant gene expression changes across multiple cell types in the developing cortex (Fig. 1D) (Kalish et al., 2021). We found that eliminating microglia via Csf1r knockout abolished most of these effects. Similar results were obtained by analyzing DEGs via DESeq2 using pseudobulked counts (Fig. S1F and Materials and Methods).

We examined MIA-induced, microglia-dependent gene expression changes in neurons and neuronal progenitors. We found several expression changes in genes involved in neuronal development (Table S2). For example, in wild type but not Csf1r-null mice exposed to MIA, migrating neurons showed increased expression of Flrt1, which encodes a cell-cell adhesion molecule that regulates neuronal migration and neurite outgrowth (Wheldon et al., 2010; Del Toro et al., 2017). Migrating neurons, corticothalamic projection neurons (CThPNs) and callosal projection neurons (CPNs) showed increased expression of Tenm4, a risk gene for schizophrenia that encodes a transmembrane protein that regulates axon guidance (Hor et al., 2015; Yi et al., 2021).

APs showed increased expression of Auts2, a marker of frontal cortical identity that is involved in synapse formation and axon elongation (Pang et al., 2021). We performed Gene Ontology (GO) analysis of DEGs using the enrichGO() function from clusterProfiler to more broadly query molecular pathways and functions that are altered by in utero inflammation in a microglia-dependent manner. GO analysis demonstrated that immune activation led to microglia-dependent upregulation of genes involved in neuronal fate commitment, axon guidance and non-canonical Wnt signaling (Fig. S1G).

We observed that MIA in the presence, but not absence, of microglia induced widespread downregulation in expression of genes involved in proliferation, energy metabolism and protein catabolism (Fig. S1H). The expression of multiple proteasomal subunits was reduced, including Psma6 (migrating neurons, CThPNs, CPNs, APs and interneurons), Psmb6 (migrating neurons, CThPNs, CPNs, APs and interneurons), Psmb2 (APs and migrating neurons) and Psmb4 (CPNs and APs). There was also a reduction in hemoglobin gene expression, including Hbb-bs (CPNs, APs and IPs), Hbb-bt (APs, IPs, migrating neurons and CPNs), Hba-a1 (CPNs, APs, IPs and migrating neurons) and Hba-a2 (APs and migrating neurons), suggesting widespread alterations in cell energetics. Furthermore, MIA reduced the expression of genes involved in positive regulation of viral processes in wild type but not Csf1r-null mice. For example, in CThPNs, CPNs and migrating neurons, MIA reduced expression of Tsg101, which encodes an ESCRT complex member that mediates viral budding, whereas expression of Banf1, which is involved in retroviral intermolecular integration, was reduced in APs and CPNs (Jamin and Wiebe, 2015; Leis et al., 2021). Additionally, in APs, IPs, SCPNs, CThPNs, CPNs, migrating neurons and interneurons, MIA resulted in downregulation of Ppia, which functions in the replication and infectivity of several viruses, and regulates the host type I interferon response to viral infections (Hoffmann and Schiene-Fischer, 2014). All of these responses were abrogated in the Csf1r-null mice. Taken together, these findings show that microglia are necessary for widespread gene expression changes that occur in multiple embryonic CNS cell types in response to maternal inflammation.

Single-cell transcriptional atlas of embryonic microglia

To fully delineate microglial molecular identity in the developing somatosensory cortex and to ask whether specific microglial substates could be responsible for mediating the effects of MIA, we built a comprehensive scRNA-seq atlas of purified microglia across embryonic cortical development. Previous studies of microglial transcriptional heterogeneity have included limited embryonic ages and brain regions, precluding a detailed understanding of lineage relationships and substate-specific roles in development and disease in the context of dorsal cortical development (Matcovitch-Natan et al., 2016; Hammond et al., 2019; Li et al., 2019). We performed scRNA-seq on wild-type microglia isolated from the prospective mouse somatosensory cortex throughout cortical development: embryonic day (E)12.5 and E13.5 (birthdate of layer 5 and 6 excitatory neurons), E14.5, E15.5 and E16.5 (birthdate of layer 2/3 and 4 excitatory neurons), E18.5 and postnatal day (P)1 (end of neurogenesis and transition to gliogenesis). To reduce ex vivo microglial activation, we performed nonenzymatic single-cell isolation under cold conditions (Hammond et al., 2019; Stogsdill et al., 2022) (Fig. 2A and Materials and Methods). Prospective mouse somatosensory cortical tissue was minced and dounce-homogenized, and microglia were labeled using antibodies against CX3CR1, CD11b and CD45, and purified via fluorescence-activated cell sorting (FACS) (Fig. S2A). Overall, we collected 831 to 6932 cells that passed quality filters per timepoint (Table S1). We clustered the cells using the Louvain algorithm, manually removed non-microglial clusters (which made up less than 3% of collected cells) and annotated the clusters post hoc (see Materials and Methods for details). We observed minimal expression of markers of ex vivo microglial activation (Marsh et al., 2022) (Fig. S2B).

Fig. 2.

Comprehensive atlas of embryonic microglia. (A) Overview of experimental design for isolation of microglia for scRNA-seq. (B) UMAP visualization of microglia scRNAseq data from the developing somatosensory cortex of wild-type mice (combined UMAP representing 11 samples of 5-10 pooled embryos each, n=21,135 total microglia). Cells are colored by microglia cluster assignment. (C) UMAP plots from B colored for expression of canonical, proliferating and substate-specific microglial marker genes. (D) Heat map of the most differentially expressed genes in each microglia substate, grouped by cluster. (E) Multiplex RNA in situ hybridization (RNAScope) at E15.5 of key markers for individual microglial substates and the pan-microglial marker Fcrls (Fcrl2). Images are from the following dorsal cortical regions: ventricular zone (Ube2c and Irf7), subventricular zone (Spp1), and intermediate zone (Ms4a7, Fabp5 and Hmox1). Scale bars: 5 μm. (F) Bar charts of microglial substate proportions over developmental time. (G) A force-directed weighted graph created by FLOWMAP algorithm representing microglia substate transitions. Cells are grouped by substates and each gray line represents one likely transition between cell types due to transcriptional similarity. The largest number of transitions occur between Ms4a7⁺ and maturing microglia, and between proliferating and maturing microglia. (H) Over-represented microglial substate transitions across time created using the BioCircos R package. The inner circle represents each substate, while the outer circle shows the distribution of ages of cells within each substate. Lines connecting substates depict transitions that happen more likely than chance when taking into account the number of cells within each substate, as calculated by a Fisher's exact test (FDR adjusted P<0.01). Line thickness represents the odds ratio of a transition, as calculated using Fisher's exact test.

Our analysis revealed 11 microglial clusters present during embryonic corticogenesis (Fig. 2B). All clusters expressed canonical microglial markers such as Crybb1 and Trem2 (Fig. 2C). We did not find evidence of Hoxb8 expression in any cluster at any timepoint (De et al., 2018). Clusters 1, 3, 5, 6 and 11 were transcriptionally similar to one another, and expressed markers of embryonic microglia previously described in studies of normal microglial development and maturation (Matcovitch-Natan et al., 2016; Hammond et al., 2019). Thus, these clusters were grouped together as likely representing a single substate: ‘maturing microglia’. The remaining clusters (2, 4, 7, 8, 9 and 10) expressed canonical microglial markers and also displayed high expression of identifying marker genes that were minimally expressed in other clusters; these clusters were annotated as unique microglial substates (Fig. 2C,D). We confirmed microglial expression of key substate markers in the dorsal cortex by RNA in situ hybridization (Fig. 2E). Substate diversity increased over time, and substates with more unique gene expression profiles (clusters 7, 8, 9 and 10) displayed an overall higher relative abundance in later embryonic ages (Fig. 2F).

A cluster of BAM-like microglia appears early in corticogenesis

Microglia cluster 4 expressed border-associated macrophage (BAM) markers (Ccr1, Dab2 and Mrc1) and Ms4a family chemosensory genes, including Ms4a6c, Ms4a6b and Ms4a7 (Fig. S3A,B). We therefore annotated this cluster as ‘Ms4a7-expressing microglia’. We confirmed that Ms4a7-expressing microglia are present in the embryonic cortical plate in addition to the intermediate zone at E15.5 by RNA in situ hybridization (Fig. 2E, Fig. S3C). Microglia are derived from the same yolk-sac progenitors as BAMs, which appear in the meninges and vasculature around E10.5 (Prinz and Priller, 2014). It is thought that local environmental signals trigger initiation of a transcriptional cascade that results in microglial identity after progenitor cells enter the brain parenchyma. Microglia that express BAM markers have been reported in the E14.5 mouse brain, but it is not known when these cells first appear, how long they persist or what the lineage relationship is between them and other microglia present in the embryonic and early postnatal brain (Hammond et al., 2019). In our dataset, cluster 4 microglia were present throughout embryonic development, even at the earliest age profiled (E12.5), before the emergence of most other microglial substates (Fig. 2F). To gain further understanding of the maturation and function of Ms4a7-expressing microglia throughout corticogenesis, we performed differential gene expression using DESeq2 with age treated as a continuous variable of interest, identifying genes with significant changes in expression over time within cluster 4 (Fig. S3D and Materials and Methods). GO analysis of the resulting up- and downregulated genes revealed that cluster 4 microglia have a more proliferative and metabolically active phenotype early in development. As corticogenesis progresses, cluster 4 microglia display increased expression of genes involved in immune and neurodevelopmental functions (Fig. S3E,F).

Embryonic microglial proliferation has two peaks

Microglia cluster 2 was characterized by high expression of markers of active cell cycling, such as Ube2c, Hist1h1b (Hif5) and Mki67, and was therefore annotated as ‘proliferating microglia’ (Fig. S4A,B). Cluster 2 peaked in overall proportion of microglia early during neurogenesis, at ∼E14.5, decreased at intermediate ages and increased again at P1, suggesting that birth may act as an immune stimulus for microglial proliferation (Fig. 2F). Using RNA in situ hybridization, we confirmed the presence of proliferating Ube2c-expressing microglia in the dorsal cortical ventricular zone and, more rarely, in the subventricular and intermediate zones, at E15.5 (Fig. 2E, Fig. S4C). We did not observe Ube2c-expressing microglia in the cortical plate. Proliferating (cluster 2) microglia expressed BAM markers such as Ms4a6c and Pf4 at early developmental ages only (E12.5-E13.5). At later ages, cluster 2 microglia showed increased expression of markers of cytokine production and of immune cell differentiation and activation (Fig. S4D-F). Thus, over the course of embryonic development, proliferating microglia may also take on more of the immune functions classically attributed to microglia. Alternatively, maturing microglia or other subspecialized substates may revert to a proliferative state during later stages of corticogenesis, while retaining aspects of a more differentiated transcriptional signature.

Microglial substate diversity increases throughout embryonic development

We observed four clusters (clusters 7, 8, 9 and 10) that did not express BAM markers and that increased in proportion in the late embryonic period (Fig. 2E,F). These clusters displayed high expression of genes and gene modules unique to each substate (Fig. 2D). Cluster 10 was characterized by high expression of osteopontin (Spp1), a pro-inflammatory phosphoglycoprotein that is secreted by microglia under stress conditions, such as in stroke, neurodegenerative disorders and multiple sclerosis (Rosmus et al., 2022). The gene expression profile of cluster 10 resembled the expression profile of axon tract-associated microglia previously described in the corpus callosum and cerebellar white matter of the P4/5 mouse brain (Hammond et al., 2019). Cluster 7 shared several marker genes with cluster 10 but lacked high Spp1 expression. Cluster 7 microglia displayed upregulation of the fatty acid binding protein Fabp5 and several unique pro-inflammatory cytokines, including Ccl3 and Ccl4. Cluster 9 was characterized by high expression of heme oxygenase (Hmox1), an enzyme that catalyzes the rate-limiting step of heme degradation (Fig. 2D). Microglial Hmox1 is thought to reduce inflammation after lipopolysaccharide (LPS) exposure and to improve neurorecovery after intraventricular hemorrhage, but the role of microglial Hmox1 in normal brain development is unknown (Parada et al., 2015; LeBlanc et al., 2016). Cluster 8 microglia displayed upregulation of several interferon-response genes, including Irf7, Ifitm3 and Ifit3. These genes are involved in regulation of the inflammatory response and inhibition of the viral life cycle. The transcriptional profiles of cluster 8 and 7 microglia resembled those of specific populations of Ifitm3⁺ and Ccl4⁺ microglia, respectively, that have previously been reported in the aged brain (Hammond et al., 2019).

Given that clusters 7, 8, 9 and 10 appeared later in developmental time, lacked BAM marker expression and expressed markers of mature microglia, we hypothesized that Ms4a7-expressing microglia may precede these later substates in a lineage relationship. We analyzed our scRNA-seq data using FLOW-MAP, a graph-based algorithm for trajectory mapping that incorporates sequential timepoint information (Ko et al., 2020) (Fig. 2G). FLOW-MAP analysis suggested that Ms4a7-expressing microglia are most likely to have a direct lineage relationship with maturing microglia and with Irf7- and Spp1-expressing substates (Fig. 2H). Hmox- and Ccl3-expressing microglia are most likely to transition directly from Spp1-expressing microglia (Fig. 2H). FLOW-MAP analysis did not suggest a consistent, unidirectional lineage pathway between microglial substates, likely reflecting the dynamic nature of microglial activation, proliferation and dormancy in the presence of variable external stimuli. Overall, these data constitute a comprehensive map of microglial substates throughout corticogenesis, with high temporal and cell type resolution, and will serve as a resource for future studies of microglial function in both normal development and in disease states.

All microglial substates highly express receptors for TORCH pathogens and cytokines across development

To understand whether specific microglial substates are poised to respond to inflammatory and infectious insults during brain development, we quantified expression over time of TORCH pathogen receptors and cytokine receptors using our single-cell transcriptional atlas of developing microglia (Fig. 3A). We observed that all microglial clusters displayed relatively high, sustained expression of multiple receptor types from E12.5 through P1, and all clusters had high CytoTORCH module scores (Fig. 3A,B). We compared expression of specific CytoTORCH receptors in microglia and other cortical cell types using the single-cell atlas of cortical development employed above (Di Bella et al., 2021). Some receptors, such as Tlr2 and Tlr12, demonstrated negligible expression in neurons and neural progenitors, but were highly expressed in multiple microglial substates, suggesting a possible primary role for microglia in mediating the corresponding infections (Figs 1A and 3A). TLR2 is a receptor for Treponema pallidum, which causes congenital syphilis. This disease is associated with many neurological problems, including microcephaly, ventriculomegaly, seizures, developmental delays and intellectual disability (Bale, 2009). Tlr2 showed a gradual increase in expression over embryonic development in most microglial substates, and no significant expression in other embryonic cortical cell types. TLR12 functions as a receptor for Toxoplasma gondii, the parasite that causes congenital toxoplasmosis, which is associated with intracranial calcifications, hydrocephalus, seizures, developmental delay and learning disabilities (Bale, 2009). Tlr12 expression was highest in proliferating microglia (cluster 2), maturing microglia (cluster 1) and Ms4a7⁺ microglia (cluster 4). Tlr12 was also highly expressed in pericytes and endothelial cells (Fig. 1A), suggesting that these cells may help mediate CNS entry. Other receptors displayed high expression across many cell types in the embryonic mouse brain. For example, Axl, a receptor for Zika virus, was highly expressed in most microglial substates (Fig. 3A), in pericytes and endothelial cells, and in many cells early in the excitatory neuronal lineage, including APs and IPs (Fig. 1A). These patterns of receptor expression suggest that some pathogens can directly target multiple cell types in the fetal brain parenchyma, whereas others initially act on microglia, which may in turn affect surrounding cell types.

Fig. 3.

Microglial substates transcriptional changes after MIA. (A) Bar plot of TORCH pathogen and cytokine receptor expression (mean log counts per million) in each microglia cluster across development using a developmental microglial atlas. (B) Violin plot of CytoTORCH module score for each microglia cluster. (C) Overview of experimental design (top) and volcano plots (bottom) showing differentially expressed genes in microglia from the embryonic prospective somatosensory cortex after maternal Poly(I:C) or saline injection. A total of 30,985 cells were sequenced, comprising six pooled samples from a total of 37 embryos for saline-treated controls, and six pooled samples from a total of 32 embryos for MIA. Genes are color coded according to enrichment (log2FC>0.15 and −log10Benjamini-Hochberg adj. P<0.05). Dotted lines represent enrichment cut-offs. Top signature genes are annotated. Log fold-change and P-values were generated with a generalized linear model with Benjamini-Hochberg correction using DESeq2. (D) Bar plot showing embryonic microglial substate proportions after maternal injection of saline when compared with Poly(I:C). A generalized mixed effects binomial models showed no significant changes in substate proportions between treatment groups. Error bars represent one standard deviation in substate proportion. Created with BioRender.com.

Over time, microglia displayed higher and more sustained expression of receptors for systemic inflammatory stimuli and Poly(I:C) than most other cell types in the developing brain (P<0.05 for ANOVA test and for pairwise comparisons between microglia and all other cell types using Tukey's HST). Tlr3, which encodes the receptor for Poly(I:C), was expressed throughout development in all microglia except for one maturing microglia cluster (cluster 11). Tlr3 was moderately expressed in astrocytes, endothelial cells and pericytes, primarily in late embryonic development, and minimally expressed in neurons and neural progenitors. All microglial substates displayed high and sustained expression of receptors for cytokines known to be involved in the response to MIA, including Il17ra and Il6ra. All microglial substates also highly expressed receptors for a subset of cytokines involved in maternal fever, including IL-6, IL-8 and interferon gamma (Romero et al., 2016; Lagodka et al., 2022). These receptor patterns suggest that microglia, when compared with most other cell types in the dorsal cortex, are more poised to respond to MIA and systemic inflammatory stimuli during fetal development.

Single-cell transcriptional changes in microglia in response to maternal immune activation

We asked how embryonic microglia respond to an in utero inflammatory stimulus at the single-cell level, hypothesizing that MIA leads to upregulation of genes involved in the inflammatory response. We performed scRNA-seq of FACS-isolated microglia from the prospective somatosensory cortex at E13.5 and E15.5, after MIA or sham saline injection at E12.5 (Fig. 3C). In total, we collected 35,794 scRNA-seq profiles, with 12,046 to 23,748 cells per timepoint. We clustered the cells as described above, manually removed non-microglial clusters and annotated the clusters post hoc (see Materials and Methods and Fig. S5A). This resulted in 30,985 cells in 11 microglial clusters suggestive of seven unique substates that closely mirrored the clusters from our analysis of normal embryonic microglial development (Fig. S5B). We found no changes in microglial substate proportions at E13.5 or E15.5 after MIA using a binomial generalized linear mixed effects model (Fig. 3D).

DEGs at each timepoint in all microglia and in each substate between MIA and saline control were identified by DESeq2 using pseudobulked counts and a negative binomial generalized linear model. We first assessed MIA-induced single-cell gene expression changes at E13.5, 24 h after the immune stimulus (Fig. 3C). When analyzing all E13.5 microglia together, we observed an upregulation of pro-inflammatory genes, including Ccl12, Jund, Isg15, and Clec12a after MIA. GO analysis (see Materials and Methods for details) also demonstrated a significant enrichment in genes involved in ‘Response to Interferon-Beta’ (Fig. S6A). MIA led to a reduction in expression of genes involved in proliferation when all microglia were grouped and analyzed together, with associated GO terms including ‘Chromatin organization’ and ‘Gene silencing’ (Fig. S6B). Analyzing the response to MIA in each cluster separately, we detected significant DEGs in maturing, proliferating and Ms4a7⁺ microglial substates, but not in other substates (Table S3). This result was likely related to the small number of microglia in many embryonic substates (Fig. 2F). Maturing microglia displayed the largest number of upregulated genes, many of which were associated with GO terms related to energy metabolism and immune activation, such as ‘positive regulation of leukocyte migration’ and ‘myeloid dendritic cell activation’ (Fig. S6A). Overall, 24 h after an immune stimulus, most microglia in the embryonic dorsal cortex displayed gene expression changes suggestive of increased activation and chemotaxis, and we did not find evidence of a specific selectively activated microglial substate.

We next analyzed single-cell gene expression changes in microglia at E15.5, 3 days after MIA (Fig. 3C). The widespread upregulation of genes involved in microglial activation that had been present at E13.5 did not remain at E15.5 (Table S4). Significantly enriched GO terms for upregulated genes included ‘aging’ and ‘ATP synthesis coupled electron transport’ (in proliferating microglia). Significantly enriched GO terms for downregulated genes included ‘response to endoplasmic reticulum stress’ (maturing microglia) and ‘proteasomal ubiquitin-independent protein catabolic process’ (maturing microglia) (Fig. S6B). Overall, gene expression changes in microglia 72 h after MIA were suggestive of persistent alterations in energetics and protein catabolism, mirroring effects seen at E15.5 in non-microglial cell types (Fig. 1A,B).

MIA-exposed microglia display persistent gene expression changes in juvenile mice

Prenatal MIA alters social behaviors in adolescents and was recently demonstrated to impair the microglial response to immune challenges in adult animals (Choi et al., 2016; Hayes et al., 2022). We hypothesized that persistent alterations in gene expression would be detectable in microglia from juvenile mice exposed to MIA during embryonic development. We performed scRNA-seq of FACS-isolated microglia from P14 somatosensory cortex after MIA or sham saline injection at E12.5 (Fig. 4A). After quality control and doublet filtering (see Materials and Methods and Table S1), we analyzed the expression profiles of 28,474 cells. We clustered the cells using the Louvain algorithm, manually removed non-microglial clusters (Jordão et al., 2019; Stogsdill et al., 2022) and annotated the clusters post hoc (Fig. 4A). We observed seven clusters of microglia that comprised five distinct substates (Fig. 4B). Clusters 1, 2 and 3 microglia expressed canonical microglial markers and homeostatic genes, and were thus annotated as a single substate: ‘homeostatic microglia’. The other clusters expressed subsets of non-homeostatic genes and were annotated as additional microglial substates. Cluster 4, ‘Apoe⁺ microglia’, was characterized by high Apoe expression. Cluster 5, ‘inflammatory microglia’, expressed pro-inflammatory cytokines including Ccl3 and Ccl4, and the cytokine receptors Cd9 and Cd63, which are also associated with AD plaques (Clayton et al., 2021). Cluster 6, ‘cycling microglia’, was composed of microglia expressing genes involved in proliferation and the cell cycle, while cluster 7, ‘innate immune microglia’, had high expression of genes involved in the innate immune response, including interferon-responsive genes such as Ifitm3 and Bst2. Microglial substate proportions did not differ significantly at P14 between MIA and saline injection (Fig. 4C).

Fig. 4.

Persistent transcriptional changes in microglia after MIA. (A) Overview of experimental design and UMAP visualization of microglia scRNA-seq data from the somatosensory cortex of P14 mice after maternal injection of saline or Poly(I:C) at E12.5. In total, we sequenced 28,474 microglia from a total of six P14 mice, two of which were treated with Poly(I:C) and four of which were treated with saline. Cells are colored by cluster. (B) UMAP plots from A colored for expression of homeostatic, inflammatory, innate immune, Apoe⁺ and cycling substate-specific microglial marker genes. (C) Bar plot showing P14 microglial substate proportions after maternal injection of saline when compared with Poly(I:C). Generalized mixed effects binomial models showed no significant changes in substate proportions between treatment groups. Error bars represent one standard deviation in substate proportion. (D) Volcano plots showing differentially expressed genes in microglia from the P14 somatosensory cortex after maternal Poly(I:C) or saline injection. Genes are color coded according to enrichment (log2FC>0.15 and -log10Benjamini-Hochberg adj. P<0.05). Dotted lines represent enrichment cut-offs. Top signature genes are annotated. Log fold-change and P-values were generated with a generalized linear model with Benjamini-Hochberg correction using DESeq2. (E) In situ hybridization for Hes1 and Lars2 in P14 control (saline injected at E12.5) and MIA [Poly(I:C) injected at E12.5] coronal sections. C1qa was used as microglia marker. Quantification showing higher levels of Hes1 expression in saline compared with MIA (data are mean±s.e.m, from n=3 mice, between 697 and 458 microglia for each group; unpaired t-test: Hes1, P=0.0004; Lars2, P=0.4092). Scale bars: 5 µm. Frequency distribution showing percentage of cells for each speckle count (n=3 mice, between 697 and 458 microglia for each group; Kolmogorov–Smirnov test: Hes1, P=0.0017; Lars2, P=0.6019). Created with BioRender.com.

DEGs between MIA and saline control were assessed in each cluster and in all microglia grouped together using DESeq2 and pseudobulked counts (see Materials and Methods for more details). At P14, a small subset of genes was significantly differentially expressed between MIA- and saline-exposed microglia. For example, the transcription factor hairy and enhancer of split 1 (Hes1) and the mitochondrial leucyl-tRNA synthetase Lars2 both demonstrated decreased expression in MIA-exposed when compared with saline-exposed microglia (Fig. 4D and Table S5). RNA in situ hybridization in P14 cortex confirmed significantly decreased expression of Hes1, but not of Lars2, in microglia after MIA exposure when compared with saline control (Fig. 4E). Downregulation of Hes1 in microglia has been demonstrated in microglia after treatment with anti-inflammatory compounds, suggesting the possibility of a prolonged, compensatory anti-inflammatory response in microglia in the weeks after MIA (Yao et al., 2022; Khoshnavay Foumani et al., 2023). Further investigation is needed to understand the functional implications of this finding, including whether downregulation of Hes1 may underlie the long-lived decrease in immune reactivity observed in adult microglia after exposure to MIA in utero (Hayes et al., 2022).

DISCUSSION

Much remains unknown regarding how maternal inflammation during pregnancy leads to adverse neuropsychiatric outcomes. Our findings demonstrate that microglia are necessary for the response of many embryonic cortical cell types to in utero inflammation, including neurons and neuronal progenitors. Microglia can play opposing, context-dependent roles in different types of brain pathology, sometimes helping promote recovery and pathogen clearance, while at other times exacerbating inflammation and enabling viral replication (Faustino et al., 2011; Fernández-López et al., 2016; Cloarec et al., 2018; Xu et al., 2020; Sarieva et al., 2023). When considering microglia as a potential future therapeutic target, it will be essential to determine whether the net impact of the microglial response in the setting of maternal inflammation is beneficial or harmful to the developing brain.

Our atlas of developing microglial gene transcription suggested that most microglia in the embryonic brain are capable of responding to a variety of infectious and inflammatory stimuli over a prolonged time window. High expression of CytoTORCH receptors was also observed in pericytes and endothelial cells, reflecting the important role of these cell types in pathogen entry into the CNS. Despite relatively uniform expression of CytoTORCH receptors across microglia, many substates bore distinct transcriptional signatures, and we observed rapid changes in substate proportions at high temporal resolution. Ms4a7-expressing microglia were present at the youngest ages studied, and FLOW-MAP analysis was consistent with this substate possibly representing an early link in the lineage relationship between yolk sac progenitors and other microglial substates. Four substates (clusters 7, 8, 9 and 10) appeared around E15.5 and increased in proportion thereafter, suggesting the potential for increased subspecialization of microglial function in later stages of corticogenesis. Three of these later-appearing clusters (7, 8 and 10) displayed gene expression profiles similar to those of microglia in the aged or diseased brain (Hammond et al., 2019; Masuda et al., 2019). These results suggest that developmental gene transcriptional programs could become reactivated in microglia later in life in response to environmental or genetic factors. Future studies using genetic labeling and conditional knockout methods are needed to unravel substate-specific roles in development and disease, and to confirm the lineage relationships between Ms4a7-expressing microglia and other substates.

We queried the temporal dynamics of the microglial response to in utero inflammation, observing that embryonic microglia upregulated pro-inflammatory and interferon-response genes 24 h after an immune stimulus. These effects mirror microglial profiles previously described in vitro after Poly(I:C) or LPS stimulation, and in vivo in the setting of ‘pseudo-TORCH’ syndromes, multiple sclerosis and brain ischemia (Das et al., 2015; McDonough et al., 2017). The immediate pro-inflammatory response subsided by E15.5, and was followed by more subtle, sustained changes in microglial gene expression that persisted through P14, including a downregulation of Hes1 expression. It is important to note that although RNA transcription generally correlates with protein expression, RNA sequencing may not accurately reflect simultaneous protein abundance (Reimegård et al., 2021). Additionally, many factors, including posttranslational modifications, and mRNA and protein stability, impact when and how changes in RNA transcription affect cell function. Further studies are needed to determine the functional consequences of the transcriptional changes reported here. Recently, it has been demonstrated that microglia exposed to MIA during development display an altered ability to respond to inflammatory stimuli in adulthood (Hayes et al., 2022). These findings suggest that the transcriptional changes we observed at P14 may reflect a persistent protective state induced by early exposure to inflammatory stimuli.

Taken together, our findings confirm the remarkable heterogeneity of microglia during development and show that microglia are essential for the response of other cell types in the fetal brain to in utero inflammation. A mechanistic understanding of how such responses are established may inform future therapeutic strategies to reduce the likelihood of adverse neuropsychiatric outcomes in fetuses exposed to maternal infections or inflammatory insults.

MATERIALS AND METHODS

Experimental model

Mouse lines

Developing microglia atlas and microglial transcriptional response to MIA experiments were performed using C57Bl/6N wild-type mice from Charles River (strain code 027). The Csf1r^−/− mouse line was procured from Jackson Laboratories (stock 028064) and was maintained on a C57Bl/6N background from Charles River (strain code 027).

Mouse husbandry

All animal procedures were approved by the Harvard University IACUC and conducted in compliance with institutional and federal guidelines. See supplementary Materials and Methods for further details.

Genotyping

Csf1r mice were genotyped using the following primers: CSF1R forward primer, TCCCAGCATTAGGCAGCCT; CSF1R reverse primer, GCCACCATGTGTCCGTGCTT. As we used only male mice for snRNAseq experiments with the Csf1r^−/− mouse line, sex was also determined for embryonic mice via amplification of the Sry gene on chromosome Y using the following primers: Sry forward primer, ACAAGTTGGCCCAGCAGAAT; Sry reverse primer, GGGATATCAACAGGCTGCCA. The sex of wild-type embryos for scRNAseq experiments (microglial atlas and microglial response to MIA) was not distinguished.

Experimental methods

Maternal immune activation

Maternal immune activation was induced by intraperitoneal (IP) injection of pregnant dams at E12.5 with poly(I:C), as previously described (Chow et al., 2016). See supplementary Materials and Methods for further details.

RNA in situ hybridization, immunohistochemistry and quantifications

RNA in situ hybridization was performed using the ACDBio Multiplex Fluorescent V2 Assay kit according to the manufacturer's protocol (ACDBio, 323100). Details of probes used and experimental methods are included in the supplementary Materials and Methods. For immunohistochemistry, tissue was blocked and permeabilized for 2 h at room temperature in blocking solution (PBS with 0.5% BSA, 0.3% Triton X-100 and 2% donkey serum), and incubated overnight at 4°C in blocking solution with rabbit anti-Iba1 (1:100, Wako, 019-19741). After washing off the primary antibody with PBS, secondary antibody incubation was carried out for 2 h at room temperature in blocking solution with donkey anti-rabbit Alexa Fluor 546 (1:500, Thermo Fisher, A10040).

Imaging and quantification

RNA scope images were obtained on a Zeiss 700 or LSM900 inverted confocal microscope and quantified using Fiji by an investigator blinded to the experimental condition (Schindelin et al., 2012). All confocal imaging was captured using Zen Black software (Zeiss, v2.3). Fluorescence imaging of Iba1 immunostaining was performed using a Zeiss Axio Imager 2 upright microscope using Zeiss Zen Software. For RNAscope quantitative analysis, mRNA speckles per microglia were counted across the cortical plate. For Iba1 immunostaining, Iba1⁺ cells were counted across the cortical wall (from pia to corpus callosum). See supplementary Materials and Methods for further details.

Tissue dissection for RNA sequencing

Brain dissection was performed with RNAse-free technique in ice-cold Hybernate-E (Brainbits) for embryonic tissue or in ice-cold Hibernate-A (Brainbits) for postnatal tissue. The somatosensory (S1) cortex (or prospective S1 cortex for embryonic tissue) was dissected and meninges removed. Tissue was immediately processed as described below for scRNA-seq experiments or flash frozen in liquid nitrogen and stored at −80°C until genotyping was completed for snRNA-seq experiments. See supplementary Materials and Methods for further details.

Single-cell suspension preparation and FACS

Live microglial suspensions from embryonic prospective S1 cortex and postnatal S1 cortex blocks were prepared for fluorescence activated cell sorting (FACS) and maintained under ice-cold conditions as previously described, and as detailed further in the supplementary Materials and Methods (Stogsdill et al., 2022). After cell dissociation, three primary antibodies, anti-Cd11b-PE (BioLegend, 101208, 1:200), anti-Cd45-APC/Cy7 (BioLegend, 103116, 1:200) and anti-Cx3cr1-APC (BioLegend 149008, 1:200) were incubated with the cells at 4°C for 15 min. The labeled cells were washed, resuspended and subjected to FACS on a Beckman Coulter MoFlo Astrios EQ Cell Sorter to isolate Cd11b⁺, Cx3cr1⁺, Cd45^low microglia, and immediately processed for single-cell GEM formation.

Single-nucleus suspension preparation and FACS

Prospective S1 cortices of E15.5 embryos after MIA at E12.5 were dissected and flash-frozen in liquid nitrogen and stored at −80°C. Tail samples were simultaneously obtained from each embryo followed by DNA extraction and genotyping (see Genotyping section above) for sex and Csf1r genotype. Frozen tissue blocks from male Csf1r mutant and wild-type embryos were processed to obtain a single-cell suspension using papain digestion (15 min) (Papain dissociation kit, Worthington), following the manufacturer's protocol. Nuclei were sorted on a Beckman Coulter MoFlo Astrios EQ Cell Sorter and immediately processed for single-cell GEM formation. See supplementary Materials and Methods for further details.

Single-cell RNA sequencing and single-nucleus RNA sequencing

Single-cell RNA sequencing (scRNA-seq) libraries were prepared with the Chromium Single Cell 3′ Kit v3.1 (10x Genomics, PN-1000121). Sorted live dissected microglia suspended in FACS buffer or sorted nuclei (15,000 nuclei/sample) suspended in Nuclei Solubilization Buffer were loaded onto a v3.1 Chromium Single Cell 3′ G Chip (10x Genomics, PN-1000127) and processed through the Chromium controller. scRNA-seq libraries from different dissected samples were pooled based on molar concentrations and expected captured cell counts, then sequenced on a NovaSeq (Illumina) instrument with 28 bases for read 1, and 91 bases for read 2. The mean reads per cell in each sample was 35K. Single-nucleus RNA sequencing (snRNA-seq) libraries were sequenced on a NovaSeq (Illumina) instrument with 26 bases for read 1, 91 bases for read 2 and 8 bases for the i7 index. The mean reads per nucleus was >24K with a minimum sequencing saturation of >40.5%.

Single nucleus RNA-seq data pre-processing and clustering

We aligned raw sequences to mm10 and produced cell-by-gene count matrices for each sample using 10X Genomics Cell Ranger 6.0.1. We used Seurat v4.0.0 to remove cells with fewer than 200 expressed genes or fewer than 500 UMIs. We scaled each sample independently before integrating all samples into a single Seurat object using Seurat's SelectIntegrationFeatures(), FindIntegrationAnchors() and IntegrateData() functions. We performed principal component analysis (PCA) on the 2000 genes used for dataset integration, and used the top 30 components for Shared Nearest Neighbor (SNN) Louvain clustering with a resolution parameter of 0.4, as well as for Uniform Manifold Approximation and Projection (UMAP) for visualization. We found 17 clusters, and manually annotated the cell type identity of each cluster based on marker genes found using Seurat's FindAllMarkers() function. For downstream analysis, we collapsed clusters that shared cell type labels. Clusters with fewer than 200 cells across all samples were ignored in differential gene expression analyses and GO term enrichment between experimental groups.

Differential expression analysis and GO enrichment

We used NEBULA to perform pairwise differential expression analyses within each cell type between Csf1r-null mice with MIA versus saline control, as well as between wild-type mice with MIA versus saline control. We used sample identity as a random effect, MIA treatment versus control as a fixed effect, and UMI count per cell as the offset. We performed Benjamini-Hochberg (BH) P-value adjustment and set a false discovery rate (FDR) threshold of 0.05. We achieved similar results when analyzing differentially expressed genes using DESeq2 and pseudobulked counts. For each cell type, genes were split into several lists based on whether they were significantly upregulated, significantly downregulated or not significantly altered in response to MIA in Csf1r-null mice and in wild-type mice according to our NEBULA analyses. We used the ClusterProfiler R package to perform GO term enrichment analyses on each of these gene lists. For lists with significantly enriched GO terms, we collapsed redundant and closely related GO terms using ClusterProfiler's simplify() function with a similarity cutoff of 0.4.

Purified microglia scRNA-seq data pre-processing and clustering

We aligned raw sequencing reads from all purified microglia scRNA-seq datasets against mm10 and produced cell-by-gene count matrices using Cell Ranger v6.0.0. We collected and processed 11 samples from untreated wild-type mice: two samples from each of the timepoints E13.5, E14.5, E15.5 and P1, and a single sample from each of the timepoints E12.5, E16.5 and E18.5. We merged these 11 samples into a single Seurat object and normalized them using SCTransform() before performing PCA, and then UMAP on the top 30 components. We then performed SNN Louvain clustering with a resolution parameter of 0.3 on the top 30 components, yielding 12 clusters. One of the clusters showed high expression of Sox11 and other neuronal markers, and lacked expression of microglia markers, so we excluded it from downstream analysis as neuronal contamination, resulting in the final 11 clusters. Additionally, we collected and processed 12 embryonic microglial scRNA-seq samples after treating pregnant dams with Poly(I:C) (MIA) or saline injections (see Table S1 for number of pooled embryos collected for each age and condition) and six samples from pups at P14 (see Table S1 for number of pups per sex per condition). We removed cells with fewer than 500 expressed genes or a mitochondrial gene content higher than 20%. We merged all 12 embryonic samples into a single Seurat object and all postnatal samples into a second object using merge(), then normalized them using SCTransform(). We clustered cells using the SNN Louvain algorithm on the top 30 principal components, using a resolution of 0.5 for the embryonic cells and 0.8 for the postnatal cells. We removed one neuronal cluster from the embryonic dataset and two neuronal clusters from the postnatal dataset as contamination. We manually annotated the remaining clusters as microglial substates based on marker genes derived for each cluster using Seurat's FindAllMarkers() function. We derived cell cycle scores using Seurat's CellCycleScoring() function. Because our MIA versus saline purified microglial samples each included cells from both males and females, and because we observed sexual dimorphism in microglial expression profiles, we used Xist and Tsix expression to identify the most likely sex of individual cells.

Expression changes over time in developing microglia

To assess changes in expression over the course of development within each microglial substate, we summed counts from all cells from each sample within each substate to create a pseudobulked counts matrix. We assigned a numeric age value to each developmental timepoint corresponding to the number of days since E12.5 (e.g. E12.5=0, E18.5=6, etc.). We then used DESeq2 to perform differential expression analyses with age as a continuous variable.

Differential expression in microglia in response to immune activation

We summed all counts for female (Xist+ and/or Tsix+) and male cells of each sample within each microglial substate to derive pseudobulk expression values. We passed those pseudobulked count matrices to DESeq2 to perform differential expression analyses between MIA and control independently for each substate at each timepoint, using sex and sequencing batch as co-variates. For each microglial subtype and timepoint, we performed GO term enrichment using ClusterProfiler() on the lists of significantly (FDR<0.1) upregulated and downregulated genes. For lists with significantly enriched GO terms, we collapsed redundant and closely related GO terms using the simplify() function with a similarity cutoff of 0.4.

CytoTORCH module score calculation

We passed a list of TORCH and cytokine receptors (Itgav, Dag1, Axl, Tyro3, Itgb1, Itgb3, Pdgfra, Pvrl1, Pvrl2, Mog, Tlr1, Tlr2, Ccr5, Tlr3, Tlr7, Tlr12, Cdh1, Met, Nlrp3, Cx3cr1, Cxcr4, Cxcr3, Cxcr2, Ccr1, Il17ra, Il1r1, Il1r2, Ifngr1, Ifngr2, Il4ra, Il6ra, Il15ra, Il2rb and Il2rg) to Seurat's AddModuleScore function which calculates the average expression levels for the requested genes subtracted by the aggregate expression of randomly selected control features (see Seurat documentation for details).

Inferring transitions between substates over time in developing microglia

We used the FLOWMAPR package (Ko et al., 2020) to infer cell state transitions over time in developing microglia. We used the top 12 principal components from our merged developing microglia dataset (described above) as input, and clustered the cells within each timepoint (E12.5-P1) into 693 nodes using FLOWMAPR's k-means algorithm. We used FLOWMAPR in ‘single’ mode to construct a force-directed graph based on Euclidean distance between these nodes, only allowing edges between nodes belonging to the same or consecutive timepoints. To determine whether transitions between each pair of substates were more likely than random chance, for each microglial substate we examined all possible edges originating from nodes with within that substate and terminating in valid nodes within other substates as a population for a series of Fisher's exact tests: for each target substate, we defined categories for contingency tables based on whether each possible edge within the population terminates within the target substate and whether each possible edge was actually found by FLOWMAPR. We visualized the odds ratios and P-values for transitions between each pair of microglial substates using the BioCircos package in R.

Statistics and reproducibility

To test whether MIA altered the relative proportions of each microglial substate, we used the lme4 R package to construct binomial generalized linear mixed effect models. For each substate, we compared two models. For both models, a cell’s membership in that substate was used as the binary response variable, and the litter of origin was used as a random effect. One model included MIA treatment as an additional fixed effect, the other did not. We used the ANOVA function in R to calculate a P-value for the significance of including treatment as a fixed effect. We used identical methods to test for changes in cell type proportions between MIA and NaCl in wild-type and Csf1r^−/− mice.

Although no statistical methods were used to predetermine sample size, the sample sizes are similar to those used in the field (Hammond et al., 2019; Di Bella et al., 2021; Kalish et al., 2021). Animals were randomized into treatment groups with similar numbers per group when possible. Investigators were not blinded during analysis of sequencing data due to the nature of the experiments. Statistical analyses were performed using R, and we controlled for multiple comparisons where appropriate using a corrected P value cutoff of <0.05. Information regarding specific statistical comparisons and packages is listed above.

Supplementary Material

Table S1.

Sample details and quality control information for RNA sequencing experiments

Table S2.

List of genes differentially expressed in cortical cell types at E15.5 after maternal immune activation, with and without microglia present

Table S3.

List of genes differentially expressed in microglia at E13.5 after maternal immune activation

Table S4.

List of genes differentially expressed in microglia at E15.5 after maternal immune activation

Table S5.

List of genes differentially expressed in microglia at P14 after maternal immune activation

Peer review history

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.202252.reviewer-comments.pdf