The importance of chromatin-mediated processes in neurodevelopmental and intellectual disability disorders is well recognized. However, how chromatin dysregulation in glial cells impacts cognitive abilities is less well understood. Here, we demonstrate that targeted loss of the ATRX chromatin remodeler targeted to microglia leads to altered cell morphology, increased chromatin accessibility profiles, and de-repression of endogenous retroelements, triggering viral mimicry. In mice that lack ATRX in microglia, CA1 hippocampal neuron morphology and electrophysiological properties are affected, and the mice display specific behavioral deficits. These findings demonstrate that ATRX is required in microglia to preserve chromatin structure and maintain microglial homeostasis. Disruption of these functions elicits neuroinflammation and may contribute to the pathology of human neurological disorders caused by ATRX mutations.

Altered morphology and increased CD68 foci in ATRX-null microglia

We targeted Atrx deletion in microglia by mating tamoxifen-inducible Cx3cr1^ERT2 mice [26] and Atrx^LoxP mice [16], resulting in male progeny lacking ATRX in microglia (hereon referred to as ATRX miKO mice). The nuclear membrane Sun1GFP reporter gene (IMSR_JAX:021039) was also introduced to track the fate of Cre-expressing microglia [27]. Cre expression was induced by daily intraperitoneal injections of tamoxifen for five consecutive days starting at postnatal day (P)45 (Fig 1A). There was no difference in body weight between control and ATRX miKO mice at 2 and 3 months of age, nor was there a difference in Cre and Sun1GFP expression by quantitative RT-PCR, suggesting equivalent level of recombination in mice of both genotypes (S1A and S1B Fig). The efficiency of Cre recombination in microglia was evaluated by quantifying the number of cells expressing Sun1GFP and the microglial marker Ionized calcium binding adaptor molecule 1 (Iba1), revealing over 95% overlap in control and ATRX miKO mice across multiple brain regions (S1C Fig). Assessment of ATRX protein expression by immunofluorescence staining of brain sections shows that more than 90% of Sun1GFP⁺ cells lack ATRX expression across various brain regions in ATRX miKO mice (Fig 1B and 1C).

We next replaced the Sun1GFP with the tdTomato-Ai14 Cre-sensitive allele to allow morphological visualization of control and ATRX-null microglia in brain sections. Upon microscopic examination of tdTomato fluorescence [28], we initially noted that the soma of microglia lacking ATRX appear morphologically altered compared to control microglia. We categorized microglia soma structure as round, intermediate, or elongated (S1D Fig), and quantified each category and measured aspect ratios. The data confirms that ATRX-null microglia somas in the cortex are more elongated at 2 months, with a larger effect seen at 3 months (Figs 1D, S1E, and S1F). The nuclei themselves are also larger and appear rod-shaped in ATRX-null compared to control microglia in the cortex (S1G and S1H Fig). Hippocampal microglia somas and nuclei (CA1, CA2, CA3, and DG regions) are rod-shaped, but this only became evident at 3 months of age (Figs 1D and S1I). We next performed a more detailed analysis of microglia morphology using tracings of confocal z-stack images (Fig 1E). We detected a significant increase in branching of microglia processes in ATRX miKO mice, increased microglia convex hull volume, and increased length of outer projections (Fig 1F–1H). The maximum order of branching reached in the ATRX null microglia was also higher than in controls (Fig 1I). Overall, these results show that ATRX-null microglia are larger and hyper-ramified, similar to microglia exposed to the cytokine Interferon α [29,30].

We considered that the change in morphology of ATRX-null microglia might reflect a reactive state sometimes associated with upregulation of the lysosomal Cluster of Differentiation 68 (CD68) [4,31]. Analysis of stained brain cryosections shows that the number of CD68 foci in microglia is significantly increased in the cortex of ATRX miKO mice at 2 and 3 months of age, and that this effect has largely resolved by 6 months of age (Fig 1J and 1K). In hippocampal regions, CD68 staining was only significantly increased at 3 months in ATRX miKO mice (Figs 1K and S1J). We note that CD68 control levels are higher at 2 months compared to 3 months, although the expectation is of a progressive increase as the mice age. One possible explanation is a transient effect of injections performed at P45–P49. The density of microglia is largely maintained in the cortex and hippocampus of ATRX miKO mice, except for a decrease at 3 months of age (Figs 1L and S1K). Overall, these results indicate that loss of ATRX expression in microglia alters microglia morphology and leads to a transient increase in CD68, indicating a loss of homeostasis.

Transcriptome analysis reveals changes in cell proliferation, genome integrity, and activation of the innate immune response

We next evaluated whether transcriptomic changes in ATRX-null microglia might provide insight into specific abnormalities underlying the observed morphological changes. Sun1GFP⁺ microglia nuclei were obtained from the cortex and hippocampus of control and ATRX miKO mice by fluorescence-activated nuclei sorting (FANS; Fig 2A) [27,32]. Given that both control and ATRX miKO undergo FANS simultaneously, we assume that any effects on gene expression caused by the method itself would occur in cells of both genotypes and would not appear as differentially expressed. The method also bypasses several problematic artifacts: it doesn’t include cell dissociation steps that rupture cell processes, or nuclei staining that could affect chromatin structure. RNA-seq of sorted microglia identified 6,168 differentially expressed genes (DEGs) between control and ATRX-null microglia, 3,265 of which were upregulated, and 2,903 downregulated (Adj p < 0.05, S1 Table). Gene Set Enrichment Analysis (GSEA) of the upregulated DEGs identified biological process categories mainly related to cell proliferation (sister chromatid segregation, DNA replication, nuclear division, mitotic cell cycle), DNA damage repair (DNA metabolic processes, DNA repair, cellular response to DNA damage stimulus), and to the immune response (defence response) (Fig 2B and S2 Table). For downregulated DEGs, the top 10 enriched biological processes are related to synaptic transmission, neuronal morphogenesis, cell adhesion, and memory (Fig 2B). GSEA enrichment plots demonstrate that upregulated genes involved in the cell cycle (i.e., Ccnb1, Ccnd3, E2F2, E2F3, E2F7, E2F8, Cdc25c, Mcm10), DNA repair (mostly homologous recombination repair) and replication stability (i.e., Rad51, Brc1a, Fanca, Gen1, DNA2, Blm, DNA2) and the immune response and inflammatory signaling (i.e., Cd69, Cd72, Axl, Mx1, Mx2, Tnf, Stat2, Cxcl10), were significantly over-represented in ATRX-null compared to control microglia (Fig 2C). To validate these results, we performed immunofluorescence staining for γH2AX, a marker of DNA double-stranded breaks and Ki67, a marker of cell proliferation. The results confirm that ATRX-null microglia contain more DNA damage (Fig 2D). There was also a notable increase in proliferative Ki67+ ATRX-null microglia in the cortex and hippocampus at 2 months that largely (but not completely) resolves by 3 months of age (Fig 2E and 2F). Deconvolution analysis of the RNA-seq data was performed to estimate the relative proportions of microglia subpopulations in control and ATRX-null microglia based on published single-cell data [33]. Of the nine microglia subclusters identified, the inferred proportion of “activated response” and “interferon response” microglia subclusters is increased in ATRX miKO mice compared to controls, seemingly at the expense of homeostatic microglia, again suggesting that ATRX deletion causes an immune response in microglia (Fig 2G). A limitation of this in silico approach, however, is the inability to detect microglial cell states not represented in this published dataset.

Increased chromatin accessibility in ATRX-null microglia

To investigate whether the transcriptional changes upon loss of microglial Atrx are associated with altered chromatin accessibility, Sun1GFP⁺ microglia nuclei from the cortex and hippocampus were sorted and subjected to the assay for transposase-accessible chromatin followed by sequencing (ATAC-seq) (Fig 3A) [32]. Two different approaches, MACS2-DESeq2 and csaw-EdgeR identified a similar number of DARs with comparable genomic distribution (S2A and S2B Fig). We proceeded with the list of DARs identified using MACS2 for downstream analysis. A total of 32,936 DARs were identified (Adj p < 0.05), located within gene promoters, exons, introns, downstream regions, and distal intergenic regions (S2A Fig and S3 Table), the majority (94%) displaying increased accessibility in ATRX-deficient microglia compared to controls (Fig 3A). Accessibility heatmaps and violin plots of DARs display a marked increase in chromatin accessibility at genes, especially at the transcriptional start sites (TSS) (Fig 3B and 3C).

Intersection of the ATAC-seq and transcriptomic data shows that 2,945 of the total 6,182 DEGs (~47%) have a DAR at their TSS (Fig 3D), demonstrating that altered chromatin accessibility is often coupled with transcriptional effects. We also find that increased chromatin accessibility is associated with elevated expression of genes related to the immune response (Fig 3E), cell cycle, and DNA repair (S2C Fig). Within the immune response category, DEGs mainly belong to the innate immune response, interferon and cytokine signaling pathways, and a large majority of DEGs are significantly upregulated and their chromatin more accessible in ATRX-null microglia (Fig 3E).

We next characterized the transcription factor (TF) binding site profiles represented in the ATAC-seq data. We scanned all the annotated TF motifs and scored their differential binding using TOBIAS [34]. We identified 30 TFs with increased differential binding scores (top 2%) between control and ATRX miKO (Fig 3F and S4 Table). The accessibility of binding sites for TFs related to interferon signaling (IRF1, IRF7, IRF9, STAT1, STAT2, ELK1, MEF2C, and ZNF384) was increased ATRX miKO (Fig 3G), confirming broad instigation of an immune and interferon response in ATRX-null microglia.

Deletion of ATRX in microglia activates endogenous retroelements

Failure to suppress retroelements in the genome is known to trigger an immune response [35]. Given ATRX’s role in heterochromatin formation and previous reports of ATRX binding to retrotransposons in mouse embryonic stem cells [22,36], we investigated whether activation of the immune response in ATRX-null microglia is linked to the de-repression of retroelements. The analysis of the ATAC-seq data identified 11,155 DARs (Adj p < 0.05) in repetitive sequences across the genome between control and ATRX-null microglia, with >99% exhibiting increased accessibility (Fig 4A and S5 Table). These mainly belong to GC/C-rich repeats, simple repeats as well as long terminal repeats (LTR), especially ERVK and ERVL as well as non-LTR transposable elements (LINES and SINEs) (Fig 4B). From the RNA-seq data, we identified 20,189 differentially expressed retroelements (Adj p < 0.05) between control and ATRX miKO (S6 Table). Consistent with ATAC-seq, the majority of these were de-repressed in ATRX-null microglia (71%) and mainly belong to LTR class ERVK and ERVL, as well as LINEs and SINEs (Fig 4C and 4D). Applying RetroTector to our RNA-seq data revealed that a notable subset of the upregulated ERVs in ATRX-null microglia are full-length ERVs (3%, n = 99) and the majority (83%, n = 2,641) are solo LTRs (S6 Table). TE subfamily-based analysis was also performed and confirmed a strong de-repression of the LTR class (Figs 4E and S3A and S7 Table). Conversely, the deletion of ATRX in oligodendrocytes or astrocytes does not cause such an extensive de-repression of retroelements suggesting that retroelement suppression is particularly sensitive to ATRX loss in microglia (S3B and S3C Fig and S8 Table). Overall, sequencing analysis demonstrates that loss of ATRX in microglia leads to a marked de-repression of TEs.

Loss of ATRX engages a viral mimicry response

The expression of genes related to a viral response is significantly enriched in ATRX-null microglia and correlates with increased chromatin accessibility (Fig 4F and 4G). Both de-repression of retroelements and DNA damage have been shown to trigger a viral mimicry response through sensing of cytosolic dsRNA or dsDNA, respectively [37–40]. Bidirectional transcription of retroelements could result in the formation of double-stranded RNA (dsRNA) [41,42] that trigger the interferon response [43,44]. We identified 4,480 and 5,226 significant differentially expressed TEs (P-adjusted value < 0.05) for forward and reverse analysis, respectively, with 57 TEs significantly expressed from both strands (S9 Table). Examples of LINEs, LTRs, and SINES with increased sense and antisense transcripts in ATRX miKO microglia are shown in Figs 4H and S2D.

To determine whether excessive expression of retrotransposons results in dsRNA species that reach the cytoplasm, we established primary cell cultures and performed immunofluorescence staining with a J2 antibody that specifically detects dsRNA [45]. This revealed an accumulation of dsRNA in Ai14⁺ATRX-null microglia (Fig 4I). To identify some of the TE-derived dsRNAs that reside in the cytoplasm, RNA immunoprecipitation using J2-conjugated beads was performed. RT-qPCR analysis of top upregulated TEs (ERVB2_1-I MMdup5 and MLT1C_LTR-ERVL) showed that they are enriched in the cytoplasmic fractions of ATRX miKO but not control brain tissue (Fig 4J). ERVB2 is evolutionary younger and has a higher potential to directly induce a strong viral mimicry response due to possible transcription of longer, complementary RNA sequences.

Increased DNA damage and cytoplasmic dsRNA can engage the viral mimicry pathway. Indeed, multiple genes belonging to the DNA/RNA sensing pathways are overexpressed in ATRX-null microglia, including Cyclic GMP-AMP synthase (cGAS) and Retinoic acid-inducible gene I (RIG-1/Dhx58) (S4A Fig). cGAS senses dsDNA and produces cyclic guanosine monophosphate–adenosine monophosphate (cGAMP), which then binds Stimulator of interferon response CGAMP interactor 1 (STING1) [46]. RIG-1 senses dsRNA and interacts with the Mitochondrial antiviral-signaling protein (MAVS) [46]. Western blot analysis shows increased RIG-1 in the ATRX miKO cortex, hippocampus, and cerebellum in 2- and 3-month-old mice; cGAS is increased in the cortex and hippocampus at 2 months but not at 3 months while it is increased at both time points in the cerebellum of ATRX miKO mice (Figs 4K, S4B, and S4C). These results suggest a potential activation of the DNA- and RNA-sensing pathways in ATRXmiKO brains.

The cGAS or RIG-1 signaling cascades both lead to the phosphorylation and heterodimerization of the Signal transducer and activator of transcription 1 and 2 (STAT1, STAT2) [47]. Stat1 gene expression is significantly increased in ATRX-null microglia (S4A Fig), and we show that total STAT1 (t-STAT1) and phospho-STAT1 (p-STAT1) protein levels are increased in ATRX miKO cortical tissue compared to controls (Fig 4K). Downstream of STAT activation, the expression of numerous interferon-stimulated genes (ISGs) was increased in ATRX-null microglia (Fig 4L).

Recognition of nucleic acids by pattern-recognition receptors (PRRs), such as RIG-1 and cGAS, is essential for triggering an antiviral immune response by inducing the production of proinflammatory cytokines [48]. Given that several cytokine genes were upregulated in ATRX-null microglia nuclei, a cytokine/chemokine array analysis of cortical protein extracts was performed and corroborates elevated levels of numerous cytokines/chemokines in ATRX miKO mice, including CC motif ligand 2 (CCL2), CC motif ligand 5 (CCL5), CC motif ligand 12 (CCL12), CC motif ligand 17 (CCL17), C-X-C motif chemokine ligand 9 (CXCL9), and C-X-C motif chemokine ligand 10 (CXCL10) (Fig 4M and S10 Table). Note; however, that the extent of cytokine production detected by this assay could be amplified due to secondary astrocytosis. These data show that loss of ATRX in microglia activates the DNA and RNA-sensing pathways as well as an interferon and cytokine response.

Loss of microglial ATRX impacts the morphology and electrophysiological states of hippocampal CA1 neurons

Given that microglia sense neural activity and, in turn, can modulate neuronal morphology and activity [49], we next evaluated the impact of ATRX deficiency in microglia on hippocampal CA1 neurons. We introduced the Thy1GFP-M allele [50] in our model to achieve sparse fluorescence labeling of neurons, facilitating morphological assessments. Sholl analysis of CA1 neuronal tracings reveals significantly increased dendritic branching (Fig 5A) and increased density of filopodia spines in mid-apical dendrites (Fig 5B), with no difference in overall dendrite length (Fig 5C) when comparing control and ATRX miKO CA1 neurons. Since loss of homeostasis in microglia can influence dentate gyrus stem cells, we also looked for potential deleterious effects on this cell population at 3 months of age. No difference was detected in dentate granule progenitor proliferation, differentiation, or apoptosis, as measured by Ki67, Doublecortin (DCX), and activated caspase 3 staining (S5A–S5C Fig).

We then evaluated the neuronal membrane properties (i.e., intrinsic properties) as well as excitatory synaptic transmission in hippocampal CA1 pyramidal neurons. We first studied the subthreshold properties by artificially injecting a small-amplitude current in whole-cell current clamp recordings (Fig 5D). We found that CA1 neurons from ATRX miKO mice exhibit a lower membrane resistance (input resistance) compared to control mice (Fig 5E). A decrease of the input resistance can be due to an increase in the cell size (surface membrane area [51]). However, the membrane capacitance (Cm), a proxy for the surface membrane area, was similar between the ATRX miKO and control (Fig 5F), suggesting that CA1 neurons lacking ATRX have increased opening in membrane channels with little, if any, change in cell size. In response to the steps of depolarizing current injections, we observed a regular-spiking pattern as reported elsewhere for hippocampal pyramidal neurons [52] in both control and ATRX miKO mice (Fig 5G). The overall current-spike frequency relationship is similar between control and ATRX miKO mice (Fig 5H). Other firing properties, including latency and inter-spike interval, are also not different between control and ATRX miKO mice (Fig 5I). The examination of the action potential shapes elicited by the minimal current injection (rheobase) revealed that CA1 neurons from ATRX miKO mice exhibit an increased action potential amplitude compared to control mice (Fig 5J and 5K). The rheobase current and threshold were not different between control and ATRX miKO neurons (Fig 5L and 5M). We also studied spontaneous excitatory postsynaptic current (sEPSC) in voltage-clamp recordings (Fig 5N). The frequency was not different between control and ATRX miKO mice (Fig 5O). However, ATRX miKO mice showed a significant leftward shift in the sEPSC amplitude distribution (Fig 5P). The average sEPSC amplitude change did not reach a statistically significant decrease (Fig 5Q), suggesting that the synaptic change may occur in subpopulations of synapses and their amplitude changes may be masked when expressed as population average. These results demonstrate that microglial ATRX deficiency imparts noncell autonomous effects on CA1 neurons, altering their morphology and electrophysiological properties.

Loss of ATRX in microglia leads to memory deficits and anxiolytic effects in mice

We next assessed whether the molecular and cellular defects observed in ATRX-null microglia impact cognitive abilities. Details of statistical analyses of all behavior assays can be found in S12 Table. In the open field test, ATRX miKO and control mice travel an equivalent distance (Fig 6A), indicating that general locomotor activity is not affected. The total time spent in the center also does not differ between genotypes (Fig 6B).

Several anxiety tests were conducted, revealing noteworthy differences between the ATRX miKO and control groups. In the light-dark paradigm, ATRX miKO mice spend significantly more time in the lit area compared to control mice, suggesting reduced anxiety levels (Fig 6C). This finding is further supported in the elevated plus maze test during which ATRX miKO mice spend significantly more time in the open arm than control mice (Fig 6D). However, there is no significant difference between groups in the number of entries into different arms of the elevated plus maze (S6A Fig).

In the contextual fear conditioning test, ATRX miKO mice spend the same amount of time freezing as control mice (Fig 6E and 6F) and in the Y-maze, the percent spontaneous alternations and total arm entries were not different between genotypes (Fig 6G), indicating that contextual fear memory and working memory processes are intact in ATRX miKO mice.

In the Morris water maze task, ATRX miKO and control mice had equivalent latency to find the platform as well as speed and distance traveled over the four days of training (Figs 6H, S6B, and S6C). Short-term and long-term spatial memory were tested on day 5 and day 12, respectively.

On day 5, control and ATRX miKO mice spent significantly more time in the target compared to other quadrants, suggesting intact short-term spatial memory. However, the 12-day probe trial revealed that ATRX miKO mice have impaired long-term spatial memory. A two-way ANOVA revealed a significant interaction of genotype and time spent in the target versus non-target quadrants, and a post hoc test identified a significant difference in percent time spent in the target quadrant between control and ATRX aiKO mice (Fig 6I).

Finally, mice were evaluated in the novel object recognition memory test. During the habituation period, control and ATRX miKO mice equally explored the two identical objects (Fig 6J). When memory was tested (1.5 and 24 h), the effect of the ATRX miKO genotype on memory was only significant after 24 h, indicating a long-term memory deficit. However, when the data is expressed as discrimination index, the difference between ATRX miKO and control mice is not significant (Figs 6K and S6D). In summary, behavioral testing shows that mice lacking ATRX in microglia exhibit lower anxiety levels and impaired long-term spatial memory.

Animal husbandry and genotyping

Mice were exposed to the 12-h light/12-h dark cycles and fed regular chow and water ad libitum. ATRX^loxP mice were previously described [16], and two reporter lines (Sun1GFP (B6;129-Gt(ROSA)26Sor^{tm5(CAG-Sun1/sfGFP)Nat}/J, MGI:5614796, RRID: IMSR_JAX:021039) and Tomato-Ai14 (Ai14) allele (B6.Cg-Gt(ROSA)26Sor^{tm14(CAG-tdTomato)Hze}/J, MGI:3809524, RRID:IMSR_JAX:007914)) were bred with the ATRX^loxP mice [27,28]. Upon Cre-mediated recombination, Sun1GFP green fluorescent protein will label the nuclear membrane, while Tomato-Ai14 will express tdTomato protein with red fluorescence in the cytoplasm and nucleus. Male progeny with Atrx deficiency in microglia upon tamoxifen treatment was produced by mating ATRX^loxP;Sun1GFP or ATRX^loxP;Tomato-Ai14 females to males expressing Cre under the inducible promoter of Cx3Cr1^ER (B6.129P2(Cg)-Cx3cr1^{tm2.1(cre/ERT2)Litt}/WganJ, MGI:5617710, RRID:ISMR JAX:021160) [26]. To delete ATRX in astrocytes, C57BL/6 heterozygous Atrx^loxP females were mated with heterozygous B6 Glast-CreER (Tg(Slc1a3-cre/ERT)1Nat, IMSR JAX:012586, MGI:4430111) male mice that express inducible Cre recombinase under the control of the Slc1a3 (Glast) promoter [95]. Mating Atrx^loxP female mice to males expressing Cre recombinase under the control of the inducible Sox10 promoter [96] (CBA;B6-Tg(Sox10-icre/ERT2)388Wdr, MGI:5634390, RRID:IMSR JAX:027651), produced male progeny with Atrx deficiency in OPCs (Atrx^loxP;Sox10Cre or Atrx^Sox10Cre) upon tamoxifen treatment, as described by Rowland and colleagues [97]. Genomic DNA was extracted from an ear notch and genotyped as described previously [32]. The genotyping primers are listed in S11 Table. Mouse weight was checked with weighing balance at 2 and 3 months of age. Behavioral tests were performed using male mice of 3–6 months of age, starting from less demanding to more demanding ones (open field tests, light-dark box, elevated plus maze, Y-maze, novel object recognition, fear conditioning, and Morris water maze). All behavioral tests were performed on at least 15 animals per genotype between 9:00 AM and 4:00 PM. ARRIVE guidelines were followed: mouse groups were randomized, experimenters were blind to the genotypes, and software-based analysis was used to score mouse performance in all the tasks. The Animal Care and Use Committee of the University of Western Ontario approved all animal procedures in compliance with the Animals for Research Act guidelines of the province of Ontario, Canada (AUP-2021-049 and AUP-2021-064).

Tamoxifen administration

Tamoxifen (10 mg; Cat#T5648, Sigma) was dissolved in 100 µl 95% ethanol at 65 °C for 10 min and then diluted with 900 µl corn oil (Cat# C8267, Sigma). 45-day-old adolescent male mice were injected intraperitoneally daily with 2 mg tamoxifen for 5 consecutive days.

Immunofluorescence

Two- or three-month-old mice were trans-cardially perfused and fixed as described previously [97]. Fixed brains were then sectioned coronally at 8–10 μm thickness (Leica CM 3050S) on Superfrost slides (Thermo Fisher Cat# 22-037-246) and stored at −80 °C with a desiccant (VWR, 61161-319). For the immunofluorescence, slides were rehydrated in 1×PBS for 5 min followed by washing with wash buffer (1×PBS + 0.3% TritonX-100 (Millipore Sigma Cat# T8787)). Antigen retrieval was performed (except for CD68 staining) by incubating slides in 10 mM sodium citrate pH6 at 95 °C for 10 min. Cooled sections were washed in 1×PBS and then blocked with 5% goat serum (Millipore Sigma Cat# G9023) or donkey serum (Millipore Sigma Cat# D9663) in wash buffer and incubated with primary antibody overnight at 4 °C. After washing three times (5 min each) with wash buffer, slides were incubated with secondary antibody for 1 h at room temperature in the dark. Slides were washed twice for 5 min with washing buffer, and counterstained with 1 μg/mL DAPI (Millipore Sigma Cat# D9542) for 5 min followed by 1× PBS wash. Finally, sections were mounted with Permafluor (Thermo Fisher Cat# TA-006-FM) and imaged with an inverted microscope (DMI 6000b, Leica) equipped with a digital camera (ORCA-ER, Hamamatsu). Volocity (PerkinElmer Demo Version 6.0.1, RRID:SCR_002668) and Adobe Photoshop was used for image processing. All cell counts were performed in a blinded and randomized manner. Primary antibodies used for immunofluorescence were anti-ATRX (Santa Cruz Biotechnology, Cat# sc-15408, RRID:AB_2061023), anti-CD68 (Biorad, Cat#MCA1957, RRID:AB_322219), anti-Ki67 (abcam, Cat# ab15580, RRID:AB_443209), anti-IBA1 (Cedarlane, Cat# 019-19741, RRID:AB_839504), anti-IBA1 (Novus Biologicals, Cat# NB100-1028, RRID:AB_3148646), anti-γH2AX (Cell Signaling Technology Cat# 2577, RRID:AB_2118010), anti-GFP (Thermo Fisher Scientific, Cat# PA1-9533, RRID:AB_1074893), anti-DCX (NEB cat# 4604S), anti-NeuN (Millipore Sigma cat# MAB377), and anti-activated Caspase 3 (NEB cat# 9661S). The following secondary antibodies were used: goat anti-rabbit-Alexa Fluor 594 (1:800, Thermo Fisher Scientific, A-11012, RRID:AB_2534079), goat anti-rabbit-Alexa Fluor 488 (1:800, Thermo Fisher Scientific Cat# A-11008, RRID:AB_143165), donkey anti-sheep-Alexa Fluor 594 (1:800 Thermo Fisher Scientific Cat# A-11016, RRID: AB_2534083), goat anti-chicken-Alexa Fluor 488 (1:800, Thermo Fisher Scientific Cat# A-11039, RRID:AB_2534096), donkey anti-goat-Alexa Fluor 594 (1:500 Thermo Fisher Scientific Cat# A-11058, RRID:AB_2534105) and goat anti-rat-Alexa Fluor 488 (1:800, Thermo Fisher Scientific, A-11006, RRID:AB_2534074). Slides were washed twice for 5 min with 0.3% Triton X 1×PBS, followed by DAPI staining for 5 min, washed, and mounted with Permafluor Aqueous Mounting Medium (Fisher Scientific cat#TA-030-FM).

Analysis of microglia cell morphology

Tomato-Ai14 (Ai14) reporter mice (B6.Cg-Gt(ROSA)26Sor^{tm14(CAG-tdTomato)Hze}/J, MGI:3809524, RRID:IMSR_JAX:007914)) were bred with the ATRX^loxP mice to enable induction of tdTomato protein in the cytoplasm and nucleus in a Cre-dependent manner. Fixed brains were sectioned at 10 µm thickness using a cryostat or 50 µm thickness on a Vibratome Series 1000. Cryosections were counterstained with 1 μg/mL DAPI (Millipore Sigma Cat# D9542) for 5 min, washed in 1×PBS, mounted with Permafluor (Thermo Fisher Cat# TA-006-FM) and imaged with an inverted microscope (DMI 6000b, Leica) equipped with a digital camera (ORCA-ER, Hamamatsu). Volocity (PerkinElmer Demo Version 6.0.1, RRID:SCR_002668) and Adobe Photoshop was used for image processing. Microglia (Ai14⁺) were classified based on the shape of the soma and nuclei, i.e., round, elongated, and medium. Thicker sections were mounted and imaged on a laser scanning confocal microscope (Leica Stellaris 5). Z-stacks of 0.5 µm thickness were obtained of the hippocampal CA1 region. Projections were traced and analyzed in a blinded manner using Neurolucida 360 Studio (MBF Bioscience, V. 2023.1.1) and Neurolucida Explorer (MBF Bioscience, V. 2022.2.1), respectively. Convex Hull 3D was used to determine volume. In this analysis, a convex polygon is generated by connecting the tips of the distal branches of microglia and the volume of the polygon is reported. The maximum order of branching for each microglia was determined and the average number of microglia for each maximum branching order was arranged in bins and plotted. Data was obtained from 8 to 15 microglia per animal from 3 mice of each genotype.

Fluorescence-activated sorting of microglia nuclei

FANS was performed as described previously [32]. The cortex and hippocampus of 2-month-old mice were homogenized in 20 mM Tricine KOH, 25 mM MgCl₂, 250 mM sucrose, 1 mM DTT, 0.15 mM spermine, 0.5 mM spermidine, 0.1% IGEPAL-630, 1× protease inhibitor cocktail (Millipore Sigma Cat# 11873580001), 1 μL/mL RNase inhibitor (Thermo Fisher Scientific Cat# 10777019)). After dilution with homogenization buffer and filtering through a 40 μm strainer (Fisherbrand Cat#22363547), the samples were layered on the top of 1:1 volume cushion buffer (0.5 mM MgCl₂, 0.88 M sucrose, 0.5 mM DTT, 1× protease Inhibitor cocktail (Millipore Sigma Cat# 11873580001), 1 μL/mL RNase inhibitor (Thermo Fisher Scientific Cat# 10777019). Nuclei were then pelleted at 2,800g for 20 min at 4 °C and incubated for 10 min in sorting buffer (4% FBS, 0.15 mM spermine, 0.5 mM spermidine, 1× protease inhibitor cocktail (Millipore Sigma Cat# 11873580001) and 1 μL/mL RNase inhibitor (Thermo Fisher Scientific Cat# 10777019) in 1×PBS). After resuspending and filtering through a 20 μm strainer (PluriSelect Cat#431002060), nuclei were sorted using a Sony SH800 Cell sorter and Sun1GFP+ microglia nuclei were collected into a 1.5 ml tube.

RNA purification and RNA-seq library preparation

RNA extraction followed by RNA-seq library preparation was performed as described previously [32]. Briefly, sorted Sun1GFP+ microglia nuclei were collected directly in lysis buffer (supplemented with 2% β-mercaptoethanol) from the single cell RNA purification kit (NorgenBiotek Cat#51800). RNA extraction, with on-column DNase treatment, was performed by following the manufacturer’s instructions. 35 ng of total RNA was used to deplete rRNA (Ribo-off rRNA Depletion kit (H/R/M), Vazyme Cat#N406), followed by strand-specific RNA-seq library preparation using the VAHTS Universal V8 RNA-seq Library Prep Kit for Illumina (Vazyme Cat#NR605-01).

RNA-seq analysis

RNA-seq libraries were sequenced at Canada’s Michael Smith Genome Sciences Centre (BC Cancer Research, Vancouver, BC, Canada) using the Illumina HiseqX (Illumina, San Diego, CA), and 60−120 million paired-end reads (150 bp) were obtained for each library. RNA-sequencing analysis was performed as described previously [32,97]. Briefly, Trim galore v0.6.6 with the following parameters (‘–phred33 -length 36 -q 5 -stringency 1 -e 0.1‘) was used to trim the raw data, and HISAT2 version 2.0.4 [98] mapped the paired-end reads against the Mus musculus GRCm38.p6 (primary assembly downloaded from Ensembl). SAMtools [99] was then used to sort and convert SAM files. StringTie v.2.1.5 [100] was used to obtain gene and transcript abundance for each sample by providing read alignments and Mus musculus GRCm38 genome annotation as input. Tximport R/Bioconductor package was used to import transcript coverage and abundance into R, and the DESeq2 R/Bioconductor package [101] conducted a differential analysis of transcript count data. The independent hypothesis weighting method was used to weigh P-values and adjust for multiple testing using the procedure of Benjamini–Hochberg (BH). Finally, the Lancaster method was used to aggregate transcripts’ p-values. Gene ontology and gene cluster analysis were performed as described previously [97]. For GSEA, significant DEGs were ranked based on log2FoldChange. gseGO function from clusterProfiler v4.0.5 [102] was then used with the following parameters (ont = “BP”, OrgDb = org.Mm.e.g.,db, minGSSize = 10, maxGSSize = 500, eps = 1e-10, pvalueCutoff = 0.05, pAdjustMethod = “BH”) to perform GSEA. For single-cell deconvolution, reference-based decomposition mode from the R toolkit, Bisque v1.0.5 [103], was used to estimate cell composition from our bulk expression data with microglia single-cell data downloaded from GSE142267 [33].

Assay for transposase-accessible chromatin (ATAC)

ATAC-seq on sorted nuclei was performed as described previously [32]. Briefly, sorted Sun1GFP+ microglia nuclei were collected directly in 1× PBS and pelleted at 3,000 rpm for 5 min at 4 °C. The nuclei were resuspended directly in tagmentation buffer containing Tn5 transposase (Vazyme Cat#TD501) and the reaction mixture was placed at 37 °C for 45 min. After tagmentation, samples were purified using the QIAquick PCR Purification Kit (Qiagen Cat#28104) and amplified by PCR according to the TruePrep DNA kit instructions (Vazyme Cat#TD501). Quantitative PCR was performed to determine the optimal number of cycles (1/3 saturation), and the libraries were amplified with no more than 12 PCR cycles.

ATAC-seq analysis

On average, 100M paired-end reads (150 bp) were obtained for each library. Raw reads were trimmed with Trim Galore (v.0.6.6) with the following parameters (‘–phred33 -length 36 -q 5 -stringency 1 -e 0.1‘) and mapped to Mus musculus GRCm38.p6 using Bowtie2 v2.4.4 [104] with the default parameters. SAMtools v1.12 was then used to create and sort BAM files from the aligned reads recorded in SAM format. The duplicated reads were then marked with the MarkDuplicates function from Picard (v.2.26.3). The mitochondrial DNA reads, and blacklist regions of the genome were filtered out using Bedtools intersect (v.2.30.0). Two approaches were employed to identify differentially accessible regions (DARs). First, MACS2 v2.2.7.1 [105] (with parameters --nomodel -f BAMPE) was used to call the peaks for each sample. The peaks from all samples were merged into a set of non-redundant open regions with Bioconductor package soGGi v1.24.1. The peaks that were present in at least two samples were kept for downstream analysis. The featurecounts function from Rsubread v2.6.4 [106] was used to count the reads from each sample overlapping the consensus peak set and DESeq2 v1.32.0 was used to perform a differential analysis of accessible regions between ATRX miKO and control samples. Second, R package csaw (v.1.26.0) [107] was used to count the reads in 200 bp non-overlapping windows. Background noise was estimated by counting reads in 2,000 bp bins. We selected 200 bp windows that have a signal higher than log2(3) above the background. Windows were then merged if less than 100 bp apart but did not extend above 5kb width. EdgeR 3.40.2 [108] was then used to identify the DARs. Because MACS2 and csaw have more than 93% overlap in calling DARs (S2B Fig), only MACS2 was later used for downstream analysis. DARs (P-value < 0.05) in the genome were annotated using ChIPseeker v1.28.3 [109]. For TF footprinting, TF motifs were downloaded from the JASPAR CORE database [110]. The footprinting analysis was performed using merged BAM files of each condition and a merged set of peaks from all samples. We used TOBIAS v0.12.11 [34] to assess chromatin occupancy by TFs. TOBIAS ATACorrect was used to correct Tn5 insertion bias in input BAM files. Footprinting scores were then calculated using TOBIAS ScoreBigWig. We then used TOBIAS BINDetect to analyze the differential binding of TFs between ATRX miKO and control groups.

Intersection of RNA-seq and ATAC-seq

To generate heatmaps of expression levels, gene-level counts from the Stringtie results were transformed using the variance stabilizing transformation (VST) method in DESeq2. For the heatmap of chromatin accessibility, ATAC-seq signals at promoters of the genes (1,000 bp upstream and 1,000 bp downstream of TSS) were counted using the featureCounts function from Rsubread v2.6.4. Z-scores were calculated for expression or accessibility independently. Heatmaps were plotted using the Pheatmap package.

Analysis of repetitive elements

For the analysis of repetitive elements, we first took a broad, per locus approach. The repeatmasker annotation file for mm10 was downloaded from the UCSC table browser. Reads falling within the repeatmasker annotation were counted for both RNA and ATAC assays using the featureCounts function from Rsubread v2.6.4. For both RNA-seq and ATAC-seq, reads were aligned using HISAT2 v2.0.4 (RNA-seq) or Bowtie2 v2.4.4 (ATAC-seq) with default parameters. For quantification with featureCounts, only uniquely mapping, non-duplicate reads were counted (countMultiMappingReads = FALSE; --ignoreDup). The average rate of uniquely mapped reads assigned to the repeatmasker annotation was approximately 25% for both RNA-seq and ATAC-seq across all samples (control and miKO). Full mapping details, including the number of assigned, duplicate, no feature, ambiguous, and unmapped reads for each sample, are provided in S13 Table. Differential expression and accessibility of repeats were determined using DESeq2. For the bi-directional transcription of retroelements, mapped reads from RNA-seq data in BAM format were split into sense and antisense strand BAMs using a custom script. Reads falling within the UCSC repeatmasker annotation were counted for both strands separately using the featurecounts function from Rsubread. Bigwig tracks were generated using Deeptools11 bamCoverage v.3.5.2 with the parameters “-bs 25 –normalize using RPKM” and visualized in the UCSC genome browser. In a second approach, we performed TE expression analysis using TEtranscripts (version 2.0.3) (Jin and colleagues, 2015), a tool specifically designed for accurate TE quantification. We utilized the curated GTF file provided by the TEtranscripts authors, which excludes simple repeats and various RNA species (e.g., rRNAs, scRNAs, snRNAs, srpRNAs, and tRNAs), ensuring the analysis focused solely on transposable elements. RNA-seq reads were aligned to the GRCm38/mm10 reference genome using STAR aligner (version 2.7.3a) (Dobin and colleagues, 2013) with parameters optimized for TE analysis:

• --winAnchorMultimapNmax 100 to allow for multiple mappings of repetitive sequences.
• --outFilterMultimapNmax 100 to include reads mapping up to 100 locations.

TEtranscripts was run in multi-mode (--mode multi) to accurately assign multi-mapping reads, which is essential due to the repetitive nature of TEs. To assess bidirectional transcription, mapped reads from RNA-seq data in BAM format were split into sense and antisense strand BAMs using a custom script. Reads falling within the UCSC repeatmasker annotation were counted for both strands separately using featurecounts function from Rsubread. Differential expression analysis of repeats for forward and reverse strands was performed using DESeq2.

Identification and classification of upregulated LTR elements

We began with a list of TEs showing significant differential expression in our RNA-seq analysis. From this set, we selected the subset of LTR elements with positive log2 fold change (log2FC > 0), resulting in 3,202 upregulated LTR loci. For each of these loci, genomic coordinates were extended by ±15 kilobases (kb) relative to the reference mouse genome (mm10) to provide sufficient flanking sequence for identification of complete endogenous retrovirus (ERV) structures. Genomic sequences for the extended regions were retrieved using the BSgenome.Mmusculus.UCSC.mm10 package in R, and individual FASTA files were generated for each locus. These FASTA files were then analyzed using RetroTector [111], which was run in batch mode (one sequence per subdirectory) using the SweepDNA and SweepScript executors with default parameters. RetroTector classified each locus based on the presence and completeness of canonical retroviral motifs. Elements containing both flanking LTRs and all major internal domains were classified as full-length proviruses. Elements missing one or more of these internal domains, but with evidence of partial internal structure, were classified as partial proviruses. LTRs lacking substantial internal retroviral motifs were categorized as solo LTRs. Classification results were parsed from RetroTector chain files using a custom R script, which extracted annotation and structural information and merged the output with the original LTR annotation table.

Mixed glial primary culture

Cre recombination in pups was induced by IP injection of lactating mothers from postnatal days 1 to 3 (P1–P3). At P4, cortices were used to establish mixed glial cultures as previously described [97] with some modifications. Briefly, tissue was dissociated and incubated at 37 °C for 20 min in papain solution (Worthington Cat# LS003124) supplemented with Turbo DNase (AM2238). Serum-supplemented media was added to deactivate papain followed by a 10 min incubation at room temperature and trituration with a flame-polished Pasteur pipette until a homogeneous suspension was obtained. Samples were centrifuged at 300 g for 5 min, pellets were resuspended in 2 ml of DMEM supplemented with 10% FBS and Primocin (10 mg/ml, InvivoGen, Cat# ant-pm-1). Cells were added to Millicell EZSlide (Millipore, Cat#PEZGS0816) wells coated with poly-L-lysine (1 mg/ml) and incubated in supplemented media in an incubator at 35 °C and 5% CO₂ for 3 h to allow cells to attach before a full media change. Media was changed after 48 h and 5 days in vitro (DIV). At 7DIV, cells were fixed with 4% PFA at room temperature for 15 min, permeabilized with 0.25% Triton-X in RNase-free PBS, washed once with 0.05% tween 20 in PBS and blocked for 1 h at room temperature with 3% BSA in PBS. Cells were incubated with J2 antibody (1:200, SCICONS # 10010200) for 1 h, followed by two washes and incubation with goat anti-mouse-Alexa Fluor 488 (1:1000, Thermo Fisher Scientific Cat# A-11001, RRID:AB_2534069) for 1 h at room temp to detect dsRNA. The wells were counter-stained with DAPI before mounting.

dsRNA immunoprecipitation

Control and ATRX miKO forebrain and cerebellum were harvested, minced, and mixed with 1.2 ml cold lysis buffer (10 mM Tris pH7.0, 10 mM NaCl, 5 mM MgCl₂, 0.5% IGEPAL CA-630, 0.5% Triton X-100, 10U/ml DNAse I, Sigma, and 40 U/µl RNaseOut inhibitor, Thermo Fisher). The lysate was centrifuged at 3,000g for 3 min at 4 °C, and the supernatant transferred to a fresh tube and further centrifuged at 21,000g for 5 min at 4 °C. The supernatant (cytoplasmic fraction) was collected for RNA IP and incubated with 5 µl anti-dsRNA monoclonal antibody J2 (SCICONS # 10010200, Mouse/ IgG2a, Kappa, Thermo Fisher) or Mouse IgG2a, Kappa Monoclonal [MOPC-173]-isotype control (ab18413, Abcam) for 2 h at 4 °C. Protein G-Dynabeads (10003D, Thermo Fisher) were washed twice using RIP buffer (50 mM Tris, pH7.4, 100 mM NaCL2, 3 mM MgCl and 0.5% IGEPAL CA-630), and incubated with the sample-antibody mix for another hour at 4 °C. The dsRNA-antibody-Dynabeads complex was washed four times with RIP buffer. RNA was extracted using 1 ml Trizol and 200µl chloroform. After isopropanol precipitation and washing with 75% ethanol, RNA was resuspended in 25 µl RNase-free water cDNA was synthesized by using previous Reverse Transcriptase and protocol [17]. PCR primers used were as follows: ERVB2_1-I_MM-intDup 5, Forward 5′-AGCACATGTTGTCCAATCGG-3′, Reverse 5′-CTCTCCTCCAGAAACCAGGG-3′ (102 bp amplicon). ERVL-MaLR-MLT1C, Forward 5′-TGAGGATGGTGAGCTGAGTT-3′, Reverse 5′-TGCTCTCCACAGTGTCCATT-3′(124 bp amplicon).

Western blot analysis

Total protein was extracted from the cortex and hippocampus of 2-month-old mice using the RIPA buffer (150 mM NaCl, 1% NP-40, 50 mM Tris pH 8.0, 0.5% deoxycholic acid, 0.1% SDS, 0.2 mM PMSF, 0.5 mM NaF, 0.1 mM Na3VO4, 1× protease inhibitor cocktail (Millipore Sigma Cat# 11873580001). Homogenized samples were incubated on ice for 30 min and centrifuged at 12,000 rpm for 15 min at 4 °C. Bradford assay (BioRad Cat# 500–0006) was used to quantify the protein and 100 μg of protein was resolved on 10% SDS-PAGE gel. After transferring proteins to nitrocellulose membrane (BioRad Cat# 1620115), the membrane was blocked with milk-TBST (5% skimmed milk, 1×TBS and 0.1% Tween-20) and then incubated with anti-STAT1 (Cell Signaling Technology Cat# 9172, RRID:AB_2198300), anti-pSTAT1 (Cell Signaling Technology Cat# 9167, RRID:AB_561284), anti-RIG-1 (Cell Signaling Technology Cat# 3743, RRID:AB_2269233), anti-cGAS (Cell Signaling Technology Cat# 31659, RRID:AB_2799008), anti-Actin (Sigma-Aldrich Cat# A2066, RRID:AB_476693) antibodies at 4 °C overnight. After three washes with milk TBST, the membrane was incubated with the appropriate secondary antibody (rabbit anti-HRP (1:5000, Jackson ImmunoResearch Cat# 111-036-003, RRID:AB_2337942); mouse anti-HRP (1:5000, Santa Cruz Cat# sc-516102, RRID:AB_2687626)). Finally, the membrane was incubated in an enhanced chemiluminescent solution (Thermo Fisher Cat# 34095) and exposed using the Universal Hood III (BioRad Cat# 731BR00882). Quantification of blots was performed with ImageJ (version 1.53).

Cytokine/chemokine analysis

Total protein was extracted from the cortex and hippocampus of 2-month-old mice using the RIPA buffer and the multiplexing analysis was performed using the Luminex 200 system (Luminex, Austin, TX, USA) by Eve Technologies Corp. (Calgary, Alberta). Forty-five markers were simultaneously measured in the samples using Eve Technologies’ Mouse Cytokine 45-Plex Discovery Assay which consists of two separate kits; one 32-plex and one 13-plex (MilliporeSigma, Burlington, Massachusetts, USA). The assay was performed according to the manufacturer’s protocol. Assay sensitivities of these markers range from 0.3 to 30.6 pg/mL for the 45-plex.

Analysis of CA1 neurons and dentate gyrus

The Thy1-GFP-M allele was introduced in the ATRX floxed line to achieve sparse neuronal labeling. Fixed brains were sectioned at 100–150 µm thickness on a Vibratome Series 1000. Sections were mounted and imaged on a laser scanning confocal microscope (Leica Stellaris 5). Z-stacks of 0.5 µm (spine analysis) or 1.5 µm (dendrite analysis) thickness were obtained of the hippocampal CA1 region. Projections were traced and analyzed in a blinded manner using Neurolucida 360 Studio (MBF Bioscience, V. 2023.1.1) and Neurolucida Explorer (MBF Bioscience, V. 2022.2.1), respectively. For dendrite analysis, 8 neurons were imaged in 5 mice per genotype. A two-way ANOVA was performed in GraphPad Prism software. For spine analysis, five separate spine segments were measured in six neurons per mouse (n = 5 mice per genotype). Each spine was analyzed using Neurolucida Explorer with consistent detection settings. The Student T test (unpaired, two-tailed) was performed using GraphPad Prism. For DG analysis, brain sections were imaged at 10× magnification using an inverted Leica microscope (CTR 6500) using Volocity Software (version 7.0.0). Brain sections (n = 5–6) were quantified per brain using ImageJ for 5 mice per genotype. Ki67, DCX, and aCASP3 staining densities were normalized to the length of the inner blades of the DG.

Slice preparation for electrophysiology

Mice were deeply anesthetized using sodium pentobarbital (100 mg/kg intraperitoneally) and transcardially perfused with cold (2–4 °C) NMDG-HEPES solution (92 mM NMDG, 93 mM HCl, 2.5 mM KCl, 1.2 mM NaH₂PO₄, 30 mM NaHCO₃, 20 mM HEPES, 25 mM Glucose, 5 mM sodium ascorbate, 2 mM Thiourea, 3 mM sodium pyruvate, 10 mM MgCl₂, 0.5 mM CaCl₂ (300–310 mOsm), saturated with 95% O2/5%CO₂). Brains were quickly removed and placed in a cold NMDG-HEPES solution for slicing. Coronal sections (350 μm thick) containing the hippocampus were cut using a vibratome (VT-1200, Leica Biosystems). Slices were incubated at 34 °C for 15 min in NMDG-HEPES solution saturated with 95% O2/5%CO₂. Slices were then transferred to artificial cerebrospinal fluid (aCSF) (126 mM NaCl, 2.5 mM KCl, 26 mM NaHCO₃, 2.5 mM CaCl₂, 1.5 mM MgCl₂, 1.25 mM NaH₂PO₄, and 10 mM d-glucose (295–300 mOsm)), saturated with 95% O₂/5%CO₂ and maintained at room temperature until recording.

Electrophysiology measurements

Slices were transferred to a recording chamber superfused with aCSF at a flow rate of 1.5–2.0 mL/min and maintained at 27–30 °C. CA1 neurons were visualized using an upright microscope with infrared differential interference contrast optics (BX 51WI, Olympus). Borosilicate glass recording pipettes (BF120-69-15, Sutter Instruments) were pulled in a Flaming/Brown Micropipette Puller (P-1000, Sutter Instruments) with a resistance between 3 and 5 MΩ. Pipettes were filled with an internal solution (116 mM K-gluconate, 8 mM KCl, 12 mM Na-gluconate, 10 mM HEPES, 2 mM MgCl₂, 4 mM K2ATP, 0.3 mM Na3GTP, and 1 mM K2-EGTA (283–289 mOsm, pH 7.2–7.4)). Spike firing was measured in the current clamp from a holding potential of −80 mV using a step protocol from −160 to 460 pA in 40 pA increments. Glutamatergic sEPSCs were isolated by adding picrotoxin (100 μM) to the aCSF while holding the postsynaptic neuron at −80 mV in a voltage clamp. Access resistance was monitored throughout the recording and cells were discarded if the value exceeded 20 MΩ.

Data collection and analysis of electrophysiological measurements

Whole cell patch clamp recordings were obtained using a Multiclamp 700B amplifier (Molecular Devices, California, USA), low-pas filtered at 1 kHz and digitized at a sampling rate of 20 kHz using Digidata 1440A (Molecular Devices). Data was recorded on a PC using pClamp 10.6 (Molecular Devices) and analyzed using MiniAnalysis (Synaptosoft, Georgia, USA) for EPSCs, Clampfit (Molecular Devices) for membrane potential and a custom Python code (adapted from a white paper from Allen Cell Types Database: https://github.com/AllenInstitute/ipfx) for cell firing. Briefly, the slope (dV/dt) was measured by taking the difference in voltage between two time steps and dividing it by the resolution of acquisition. The time of the threshold crossing was detected by finding the time point where dV/dt was ≤5% of the maximum dV/dt of the rising phase. The following criteria were used to detect action potentials during current injection steps: (1) the duration from threshold to the peak is ≤5 ms, (2) amplitude is ≥2 mV and absolute peak ≥−30 mV, and (3) action potential trough (minimum membrane potential in the interval between the peaks of two consecutive action potentials) is ≤–22 mV. For sEPSC analysis, baseline data was taken at least 5 min after breaking through into whole-cell mode and a 0.5- or 1-min bin was used for analysis. To achieve an accurate measure of the amplitude, individual sEPSCs were visually screened in MiniAnalysis and events below 5 pA were not included.

Open field test

The mice were acclimated to the room for 30 min prior to testing. Then the mice were placed in an open arena (length 20 cm, width 20 cm, height 30 cm) for 2 h and locomotor activity was measured in 5 min intervals as previously described [17]. Locomotor activity was automatically recorded (AccuScan Instrument), and distance traveled, and time spent in the center were reported.

Y maze test

The mice were acclimated to the room for 30 min prior to testing. The mice were placed in the center of a symmetrical three-armed Y maze as described previously [17] and their activity was recorded for 5 min using AnyMaze. The order and number of entries into each arm were reported. Spontaneous alternations were counted when a mouse entered all three arms in a row without visiting a previous arm.

Light dark box

The mice were acclimated to the room for 30 min prior to testing and were then placed in a light-dark box and allowed to explore the arena for 10 min. Their activity was automatically recorded (AccuScan Instrument), and the percent time spent in light was reported.

Elevated plus maze

The mice were acclimated to the room for 30 min prior to testing and then placed in the center of the elevated plus maze (Med Associate) and their activity was recorded for 5 min using AnyMaze. Time spent in the open, closed, or center of the elevated plus maze was reported. The center of the mouse body was used as an indicator to determine its presence in an open, closed, or center.

Novel object recognition

The mice were habituated for two consecutive days in an empty arena (40 cm × 40 cm) for 5 min as described previously [17]. The next day, mice were trained by exposing them to two identical objects (A) for 10 min and their activity was recorded using AnyMaze. After 1.5 h of training, one of the objects was replaced with a novel object (B), and mice were exposed to both old (A) and novel (B) objects to test their short-term recognition memory. Similarly, mice were exposed to objects (A) and (B) 24 h after training to test their long-term recognition memory. Time spent with objects was recorded when mice sniff or touch the objects, but did not lean against and/or climb on the object.

Morris water maze

The Morris water maze test was conducted as described previously [17], where the mice were placed in a 1.5 m diameter pool with 25 °C water. Spatial cues were displayed around the pool and the platform was submerged 1 cm below the water’s surface. The mice were acclimated to the room for 30 min prior to testing and then trained to find the platform in four trials (90 s) a day for 4 consecutive days with a 15 min intertrial period. Their activity was recorded with AnyMaze. If the mice did not find the platform within the 90 s, they were gently guided onto the platform. Short or long-term spatial memory was tested by removing the platform on day 5 or day 12, respectively. The latency to find the platform over 4 days of training, and the percent time spent in each quadrant of the maze was recorded for the probe tests.

Contextual fear conditioning

The mice were acclimated to the room for 30 min prior to testing and then placed in a 20 cm × 10 cm enclosure with a metal grid floor connected to a shock generator. One wall of the enclosure was marked with a stripe pattern and mouse activity was recorded using AnyMaze. For training, mice were placed in the enclosure and allowed to freely explore it. After 150 s, a shock was administered (2 mA, 180 V, 2 s), and the mice were returned to their home cage after 30 s. After 24 h, contextual fear memory was tested by placing the mice in the enclosure for 6 min. Freezing time was reported in 30 s intervals, where freezing was defined as immobility lasting more than 0.5 s.

Statistical analysis

The Student T test (unpaired, two-tailed) or one-way ANOVA for experiments with one variable or two-way repeated-measures ANOVA with post hoc test for experiments with two variables were performed using GraphPad Prism software (GraphPad Software Inc, California, USA). All results are depicted as mean ± SEM unless indicated otherwise. P-values of less than 0.05 were considered to indicate significance. All statistical details are outlined in the figure legends, except for behavior tests, where statistics are provided in S12 Table.

S1 Fig. Morphology, density, and CD68 immunostaining in control and Atrx miKO microglia.

(A) Weight of 2- and 3-month-old control and ATRX miKO mice (2 months, CTL n = 11, ATRX miKO n = 7, p = 0.080; 3 months, CTL n = 6, ATRX miKO n = 6, p = 0.679 Student T test). (B) RT-qPCR of Cre and Sun1GFP transcripts in 2-month-old control and ATRX miKO mice (n = 4 each genotype, Sun1GFP p = 0.871; Cre p = 0.804, Student T test). Results were normalized to beta-actin transcript levels. (C) Quantification of immunofluorescence staining of Sun1GFP and IBA1 reveals >95% Cre expression in control and ATRX miKO mice across different brain regions (n = 3 each genotype, cortex p = 0.423, hippocampus p = 0.351, cerebellum p = 0.119, striatum p = 0.327). (D) Representative images of Ai14+ cells with round, medium, and elongated soma. (E, F) Quantification of soma area and aspect ratio in the cortex at 2- and 3-months of age (n = 3 each genotype, soma area 2 months p = 0.0003, 3 months p = 0.016; aspect ratio 2 months p = 0.029, 3 months p = 0.007, Student T test). (G, H) Quantification of nuclear area and aspect ratio in the cortex at 2- and 3-months (n = 3 each genotype, nuclear area 2 months p = 0.001, 3 months p = 0.034; aspect ratio 2 months p = 0.0937, 3 months p = 0.005, Student T test). (I) Quantification of elongated soma of microglia in hippocampal CA2, CA3, and DG of 2- and 3-months-old mice (n = 3 each genotype, CA2 elongated p = 0.027; CA3 elongated p = 0.002; DG elongated p = 0.018, Student T test). (J) Quantification of CD68 foci per Ai14-labeled microglia in hippocampal CA2, CA3, and DG of 2-, 3-, and 6-month-old mice (n = 3 each genotype, 2 months CA2 p = 0.861, 3 months CA2 p = 0.006, 6 months CA2 p = 0.193; 2 months CA3 p = 0.931, 3 months CA3 p = 0.023; 6 months CA3 p = 0.067; 2 months DG p = 0.243, 3 months DG p = 0.002, 6 months DG p = 0.935, Student T test). (K) Microglia density in hippocampal CA2/3 and DG of 2-, 3-, and 6-month-old control and ATRX miKO mice (n = 3 each genotype, CA2/3 2 months p = 0.975, 3 months p = 0.018, 6 months p = 0.451; DG 2 months p = 0.831, 3 months p = 0.007, 6 months p = 0.415, Student T test). In all panels, control data is shown in black and ATRX miKO data is shown in red. The data underlying this figure can be found in the S1 Data file.

https://doi.org/10.1371/journal.pbio.3002659.s001

(TIF)

S2 Fig. Chromatin accessibility in microglia lacking ATRX.

(A) Genomic distribution of DARs called by csaw and MACS. (B) Overlap of DARs called by csaw and MACS. (C) Heatmaps represent an association between gene expression and chromatin accessibility for cell cycle and DNA repair pathway genes. The z-score was computed from the RNA expression or ATAC-seq signals. (D) Macro view of UCSC tracks for example LINE, LTR, and SINE from Fig 4G.

https://doi.org/10.1371/journal.pbio.3002659.s002

(TIF)

S3 Fig. Comparison of TE subfamily expression in different cell types lacking ATRX.

Scatterplots of transposable element family expression differences between control and ATRX KO in (A) microglia, (B) oligodendrocyte precursor cells (OPCs), and (C) astrocytes. Significantly upregulated TE subfamiles are indicated in red and downregulated TE subfamiles are indicated in blue. TE subfamilies not displaying significant change between control and KO cells are shown in gray.

https://doi.org/10.1371/journal.pbio.3002659.s003

(TIF)

S4 Fig. RIG1 and cGAS levels in different brain regions.

(A) Expression of genes in the RIG1 and cGAS pathways. Variance stabilizing transformation (VST) counts from RNA-seq data are shown. n = 3 each genotype. (B) western blot analysis of RIG1 and cGAS in control and Atrx miKO cortex and hippocampus at 3 months. Quantification is shown in graphs on the right (n = 3 each genotype). (B) western blot analysis of RIG1 and cGAS in control and Atrx miKO cerebellum at 2 and 3 months. Error bars represent ±SEM. n = 3 each genotype. Unpaired Student T test. The data underlying this figure can be found in the S1 Data file.

https://doi.org/10.1371/journal.pbio.3002659.s004

(TIF)

S5 Fig. Effects of ATRX microglial deficiency on the dentate gyrus.

Quantification of Ki67+ proliferating cells (A) and DCX+ differentiating cells (B) in dentate gyrus of control and ATRX miKO mice (n = 5 control and n = 6 KO mice). (C) Immunofluorescence staining of dentate gyrus for activated caspase 3 reveals no difference in apoptosis between control and ATRX miKO mice (n = 3 each genotype). The data underlying this figure can be found in the S1 Data file.

https://doi.org/10.1371/journal.pbio.3002659.s005

(TIF)

S6 Fig. Supplemental data related to behavior tests.

(A) The number of entries in the open and closed arms or in the center of the elevated plus maze over 5 min (open arm entries p = 0.710, closed arm entries p = 0.986, center arm entries p = 0.577, Student T test). Error bars represent ± SEM. (B) Distance traveled and (C) swimming speed over 4 days of training (4 trials/day) in the Morris water maze task (distance traveled F(1, 33) = 0.5162, p = 0.477; swimming speed F(1, 132) = 0.1696, p = 0.681, two-way ANOVA). (D) Discrimination index at 1.5 h (t = 0.5848, df = 22, p = 0.5646) and 24 h (t = 1.696, df = 21, p = 0.1048) probe test of the novel object recognition test. The data underlying this figure can be found in the S1 Data file.

https://doi.org/10.1371/journal.pbio.3002659.s006

(TIF)

S3 Table. List of differentially accessible regions (DARs).

DARs identified in microglia nuclei sorted from control and ATRX miKO cortex and hippocampus and subjected to the assay for transposase-accessible chromatin followed by sequencing (ATAC-seq).

https://doi.org/10.1371/journal.pbio.3002659.s009

(XLSX)

S4 Table. List of top TFs with altered binding scores.

ATAC-seq data from sorted control and ATRX-null microglia nuclei was analyzed to identify the top represented transcription factors binding sites displaying differential binding scores (top 2%) using TOBIAS.

https://doi.org/10.1371/journal.pbio.3002659.s010

(XLSX)

S6 Table. List of altered TE transcripts in ATRX-null microglia.

Differentially expressed retroelements were identified from RNA-sequencing data from control and ATRX-null microglia. RetroTector analysis was applied to categorize these elements as full-length ERVs or solo LTRs.

https://doi.org/10.1371/journal.pbio.3002659.s012

(XLSX)