Abstract The central nervous system (CNS) represents a uniquely immune-privileged environment, with inflammatory responses involving several resident CNS-specific cell types. While stereotyped cellular and transcriptional responses recur across varied diseases, relevant signaling pathways and regulatory networks are not fully understood. Here, we investigate multi-modal inflammatory gene networks at large scale by developing a high-throughput RNA-seq screening and analysis workflow. As proof-of-concept, we investigate genetically heterogeneous mice from a large-scale chemical mutagenesis screen to identify novel functionally relevant variants in six genes previously linked to human CNS disorders: Nrros, Ctsd, Smpd1, Idua, Nlrp1a, and Inpp5d. We leverage the readily interpretable data from our large-scale study to demarcate distinct inflammatory states arising from each mutation. In all, our work provides a validated analysis framework for identifying discrete gene expression modules that are engaged divergently across disease contexts, which can be used to discover novel regulators of CNS neuroimmune homeostasis. Supplementary Information The online version contains supplementary material available at 10.1186/s12974-025-03556-7. Keywords: Microglia, ENU-mutagenesis, RNA-seq, Ctsd, Nlrp1a, Inpp5d Highlights Brain bulk-tissue RNA-seq database from a large-scale ENU-mutagenized mouse cohort. Variants in Nrros, Ctsd, Smpd1, Idua, Nlrp1a, and Inpp5d alter microglia homeostasis. An Nlrp1a gain-of-function mutation causes widespread astrocyte and microglia activation. Cross-comparison of six disease and inflammatory states facilitates interpretation of bulk tissue transcriptomes. Supplementary Information The online version contains supplementary material available at 10.1186/s12974-025-03556-7. Introduction In vivo genetic perturbations are a cornerstone for the exploration of complex tissue-level biology and investigation of multifaceted disease processes. Spontaneous mutations such as shiverer, lurcher, reeler, Wld^S, and stargazer cause overt phenotypes in mouse and have provided critical insights that inform our understanding of nervous system development, homeostasis, and disease [[64]1–[65]5]. Furthermore, mutagenesis screens in mice using the chemical mutagen N-Ethyl-N-nitrosourea (ENU) have helped define many immune regulators [[66]6, [67]7], including Tlr4 contributions to pathogen sensing [[68]8] and Gsdmd or Ninj1 to pyroptosis or plasma membrane rupture, respectively [[69]9, [70]10]. However, a major hurdle to understanding inflammatory mechanisms in the central nervous system (CNS) is an incomplete understanding of tissue-specific regulatory cues and to what degree these are distinct from other tissues. High-content transcriptomic datasets have identified stereotyped brain inflammation signatures including disease-associated microglia (DAM), interferon (IFN) responses, reactive astrocytes, disease-associated oligodendrocytes, or infiltration of peripheral immune cells. However, different disorders exhibit variations in character and intensity of these responses [[71]11, [72]12]. Although meta-analyses have been informative, many uncertainties remain around the regulation of inflammatory modules and our ability to make generalized statements pertaining to relationships among them [[73]11, [74]13–[75]15]. As an example, expression of Microglia Signature Genes (MSGs) distinguishes CNS-resident microglia from other tissue macrophages [[76]16, [77]17]. Microglia lose MSG expression in vitro, impeding study of upstream regulators and downstream functions [[78]18, [79]19]. In vivo investigations have identified Nrros, Itgb8, and their roles in TGF-β-induced maturation as key signals for expression of MSGs and suppression of border-associated macrophage (BAM) signatures, which emerge within CNS meninges, choroid plexus, or perivascular niches [[80]20–[81]22]. However, despite commonalities in MSG loss, the transcriptional changes resulting from loss of TGF-β signaling differ from early development or disease states [[82]23, [83]24]. Relatedly, TREM2 is required for MSG loss and DAM signature induction after apoptotic cell exposure or during neurodegeneration but is dispensable for similar changes in developing white matter [[84]24–[85]28]. To advance our understanding of CNS-specific immunoregulation, here we combine high-throughput RNA-seq technologies with a heterogeneous ENU-mutagenized mouse colony to develop a screen for CNS homeostasis and inflammation phenotypes. We illustrate the effectiveness of our approach by identifying novel variants in Nrros, Ctsd, Smpd1, and Idua, which are all linked to neurological disorders in human and trigger distinct microglial phenotypes in mouse. We additionally uncover a striking reactive astrocyte phenotype resulting from a Nlrp1a gain-of-function mutation, identifying CNS roles for NLRP1A inflammasomes. Using a reverse-genetics strategy, we identify CNS neutrophilia and white matter microglia activation phenotypes in Inpp5d loss-of-function mice. While there is phenotypic overlap for all mutants identified, we apply a broad multimodal analysis framework to reveal unique fingerprints for each, highlighting key characteristics of each inflammatory response. In all, this work provides an optimized screening platform for large-scale CNS phenotyping and demonstrates its utility for disentangling inter-related CNS inflammatory syndromes. Results A screening and analysis framework for rapid detection of aberrant brain states To broadly characterize neurological phenotypes of ENU-mutagenized mice, we established a workflow for transcriptional profiling and histological validation of many samples in parallel (Fig. [86]1A). Spurred by the advent of single-cell approaches, several emerging techniques employ cDNA barcoding, early pooling, and 3’-end sequencing to enhance throughput and lower cost for either single-cell or bulk RNA-seq [[87]29–[88]33]. We sought to apply these methodological advances to screen hundreds of ENU pedigrees (12 mice from each pedigree), collecting one hemibrain from each mouse for RNA extraction and banking the other hemibrain in fixative for future immunostaining. Fig. 1. [89]Fig. 1 [90]Open in a new tab A screening framework for high-throughput identification of inflamed brains. A) Schematic of the screening approach. Created with biorender.com. B) Normalized mean expression of eight genes comprising the Microglia Signature gene set for wild-type (black) or Nrros^−/− (blue) bulk brain RNA-seq datasets generated using each library prep method (Tru-seq, 3’UPX, BRB-seq, Tag-seq, or SMART-Seq mRNA 3’DE-seq). The z’ score, which captures the sensitivity to detect the Nrros^−/− phenotype, is provided for each method. Data is plotted as mean ± SEM for n = 6–12. C) Microglia Signature gene set score expression for Nrros^−/− mice (blue) or mice from pedigree 42274 (red) compared to Nrros^+/+ mice (black) or other mice screened from an initial batch of 330 pedigrees (gray). Three mice from pedigree 42274 had low gene set scores (affected, triangles), compared to nine mice from the same pedigree with typical scores (unaffected, red circles). Data points represent individual mice. D) Gene expression heatmap for genes comprising the Microglia Signature gene set for pedigree 42274 affected (triangles) and a subset of unaffected (circles) mice. E) Gene expression heatmap for differentially-expressed genes (FDR < 0.1) between affected Nrros^C247Y/C247Y mice (triangles) and unaffected mice from the same pedigree (circles). F) Representative images with higher-resolution inset for TMEM119 immunostaining of unaffected (top) or affected Nrros^C247Y/C247Y (bottom) sagittal brain sections. Left panels scalebar: 1 mm. Right panels scalebar: 100 μm. Deconvolution approaches have been used to infer cell type composition of whole-tissue bulk RNA-seq datasets using marker gene signatures of the composite cell types as a guide [[91]34]. In an analogous fashion, we endeavored to define gene sets that could be used to score microglia transcriptional states from bulk brain RNA-seq data. To this end, we selected eight MSGs whose expression levels in bulk brain tissue are strongly altered by Nrros deletion, are highly specific to microglia over other CNS cell types, are highly specific to microglia over other tissue resident macrophages, and exhibit high expression in bulk brain RNA (Supplementary Table 1) [[92]16, [93]19, [94]21]. Furthermore, we confirmed the high expression level and microglia-enrichment for each microglia signature gene using a published dataset (Figure [95]S1) [[96]35]. These genes are also commonly downregulated in DAM populations, for example in neurodegenerative disease models [[97]13, [98]15, [99]26]. We used this gene set to assess the sensitivity of different sequencing methods to detect MSG abnormalities of Nrros^−/− mice. Because microglia are a relatively low-abundance CNS cell type, their transcripts are not plentiful in bulk brain RNA samples. Lower-abundance transcripts exhibit higher repeat-measurement variability, with the lower-limit of variance described by the Poisson distribution (Figure [100]S2) [[101]36]. Nonetheless, we found that MSG abnormalities from Nrros^−/− brain RNA samples remained easy to detect after downsampling Tru-seq RNA-seq datasets to four million reads per sample. The sensitivity for detecting MSG changes is summarized using a z’ factor (z’=0.55), which gauges assay signal-to-noise and indicates suitability of assays for large-scale screening [[102]37] (Fig. [103]1B). Additionally, we explored scalable, low-cost methods for RNA-seq library preparation. Targeting a sequencing depth of one to four million reads per sample, we tested four methods for high-throughput RNA-seq library preparation: UPX 3’ seq, BRB-seq, Tag-seq, and SMART-Seq mRNA 3’DE [[104]29, [105]32]. Each method readily detected MSG abnormalities of Nrros^−/− mice (Fig. [106]1B), although the inter-sample variability deviated from a theoretical Poisson distribution in a manner suggestive of RNA/cDNA library sub-sampling prior to sequencing (Figure [107]S2). In our implementation, the SMART-Seq mRNA 3’DE protocol exhibited good signal-to-noise separation (z’ = 0.33) (Fig. [108]1B), indicating strong separation between affected and unaffected distributions even at 3 standard deviations and suggesting that false-positives or false-negatives resulting from signal overlap would be few even on the scale of thousands of samples. Nrros^C247Y exhibits microglia signature abnormalities To test whether our workflow is sensitive for de novo identification of microglia phenotypes in a highly heterogeneous sample set, we next sequenced brains from 331 pedigrees of ENU-mutagenized mice. Nrros^−/− brain RNA samples were randomly incorporated into screening sets, and these samples were readily distinguishable based on MSG scores (Fig. [109]1C). Additionally, the 331 st pedigree (pedigree 42274) had three mice with exceptionally low MSG scores, which resembled the Nrros^−/− samples and were multiple standard deviations outside of the distribution of unaffected mice (Fig. [110]1C). To identify the causal variant responsible for the phenotype of the mice with a low MSG score in pedigree 42274, we performed exome-sequencing and identified a single variant, Nrros^C247Y, which was homozygous in all three affected mice but absent in unaffected siblings. Cysteine 247 lies in the leucine-rich repeat domain, and the tyrosine substitution is predicted to be damaging by mouse-adapted Polyphen2 [[111]38]. RNA samples from Nrros^C247Y/C247Y brains and a subset of unaffected brains from the same pedigree were re-processed with the same protocol and read depth as above to generate an independent technical replicate, confirming reduced MSG expression (Fig. [112]1D). Differential-expression analysis of Nrros^C247Y/C247Y versus unaffected mice identified significantly lower expression of several MSGs and higher expression for markers of border-associated macrophages such as Ms4a7, Ccl12, and Pf4, which are known to be elevated in Nrros^−/− mice (Fig. [113]1E) [[114]21]. Interestingly, antigen presentation markers Cd74 and H2-Ab1, which are specific to a subset of BAMs, are also upregulated [[115]39]. Other changes (e.g. downregulation of Cldn11) likely reflect white matter or other tissue changes due to the use of whole-tissue RNA for analysis. To validate the microglia-signature phenotype of Nrros^C247Y/C247Y mice, we sectioned fixed hemibrains that had been set aside during initial sample collection (Fig. [116]1A). Affected mice exhibited a stark loss of TMEM119 staining (Fig. [117]1F), consistent with previously reported staining profiles of Nrros^−/− mice [[118]20, [119]21]. These findings illustrate the suitability of our screening protocol for detecting microglia-related phenotypes at scale. Extensive microglial activation in Ctsd^T263A mice We next asked whether our screening approach could identify additional phenotypes beyond MSG abnormalities. We defined a signature of tissue macrophages using genes whose expression correlates with Csf1r, a macrophage marker gene, across a large collection of bulk tissue RNA-seq datasets that had been previously assembled [[120]40]. Csf1r-coexpressed genes were filtered to remove genes enriched in microglia over other macrophage populations, and to remove low-abundance transcripts with high observed variance (Supplementary Table 1) [[121]40, [122]41]. We also defined a separate set of DAM markers based on upregulation in whole-brain RNA-seq datasets and DAM subcluster specificity in single-cell RNA-seq datasets (Supplementary Table 1) [[123]26, [124]28, [125]42]. We extended the screen to encompass 1,596 additional pedigrees and identified several outlier samples with both high macrophage scores and high DAM gene set scores (Fig. [126]2A). We first focused on pedigree 55142, which included three mice with high DAM scores and elevated expression of a subset of macrophage markers (Fig. [127]2A-C). Additional outliers represent other affected pedigrees characterized in this work or affected pedigrees whose causal genes are still being determined; additionally, several pedigrees bearing only a single affected mouse were found to result from hydrocephalus upon inspection of brain tissue. Fig. 2. [128]Fig. 2 [129]Open in a new tab Identification of a severe microglial phenotype for Ctsd^T263A/T263A mice. A) Gene set scores for pedigree 55142 affected mice (Ctsd^T263A/T263A mice, red triangles) or unaffected mice (red circles) versus other mice from the screen (gray circles). B, C) Gene expression heatmap for genes belonging to the DAM gene set. (B) or Macrophage gene set. (C) for affected (Ctsd^T263A/T263A, red triangles) or unaffected (red circles) mice from pedigree 55142. D) Gene expression heatmap for differentially-expressed genes (Ctsd^T263A/T263A vs. unaffected FDR < 0.01). Symbols and color scale are same as in B. E, F) Representative images at low magnification (top) or high magnification (bottom), or quantification in thalamus. (F) for microglia lysosome/activation marker CD68 in coronal brain sections. Top panel scalebar: 1 mm. Bottom panel scalebar: 100 μm. Data points represent individual mice. Data is plotted as mean ± SEM, student’s t-test (two-sided), ****P < 0.0001 for F. To identify the causal variant responsible for pedigree 55142, we sequenced exomes of three affected and three unaffected mice. All affected mice (and no unaffected mice) carried a set of two homozygous mutations in close proximity on chromosome 7, resulting in coding mutations Ctsd^T263A and Ifitm2^H56Q, both predicted to be damaging or deleterious mutations [[130]38, [131]43]. We reasoned that these mutations co-segregated in the affected mice due to linkage disequilibrium. Because Ctsd mutations cause a lysosomal storage disease in human and regional glial activation in mouse and sheep [[132]44–[133]48] whereas Ifitm2 mutations have not been linked to either [[134]49], Ctsd^T263A is likely sufficient to cause the observed phenotype. Genotyping of all 12 mice from pedigree 55142 confirmed that the Ctsd^T263A/T263A genotype is specific to affected mice, with all 9 unaffected mice bearing wild-type Ctsd or are heterozygous for Ctsd^T263A. To validate the RNA-seq result, fixed hemibrains that had been set aside from the original samples were sectioned and stained for the microglia activation marker CD68. Extreme and highly region-specific elevation of CD68 staining was observed for Ctsd^T263A/T263A mice, with the strongest staining observed in thalamus and deeper layers of the cortex (Fig. [135]2E, F). Loss-of-function variants of human CTSD lead to neuronal ceroid lipufuscinosis-10 (CLN10) [[136]45, [137]46], with rare hypomorphic variants resulting in delayed juvenile-onset disease with prolonged survival into early adulthood [[138]50]. Previously developed loss-of-function mouse lines are also non-viable past weaning [[139]48, [140]51], whereas three Ctsd^T263A/T263A mice survived to 8 weeks of age, providing a valuable model to study developmental progression related to juvenile-onset CLN10. To characterize the Ctsd^T263A/T263A phenotype more deeply, we leveraged the high information content of the RNA-seq dataset and performed differential expression analysis. Despite the relatively low sample size and sequencing depth, we were able to observe a robust spectrum of gene expression changes (Fig. [141]2D). Pathway enrichment analysis highlighted upregulation of Reactome pathways related to innate immune cells (Fcer1g, Itgb2, Trem2, Tyrobp, Cd68, C1qc), cytokine signaling (Cxcl10, Ccl5, Ccl4, and Timp1), and interferon signaling (Ifit1, Ifit3, Psmb8, Gbp2, and Usp18). Other upregulated marker genes belonging to the DAM gene set (Lgals3 and Gpnmb) or characteristic of astrocyte and oligodendrocyte alterations (Serpina3n, C4b, and Gfap) indicate a complex CNS inflammatory environment. These results illustrate the utility of our screening platform for identification of CNS inflammatory phenotypes distinct from those used for assay optimization and identify an adult-viable mouse model for studying CLN10. Nuanced activation phenotypes in Smpd1^C248S and Idua^E130V mice Using the macrophage and DAM gene set scores, we identified several mice (from pedigree 55279) exhibiting mild elevation of macrophage and DAM signature scores (Fig. [142]3A). The changes in expression for individual marker genes was small relative to overall variance (Fig. [143]3B, C), unlike the clearer signal observed for Ctsd^T263A (Fig. [144]2B, C). We performed exome sequencing on five affected mice, identifying a Smpd1^C248S variant present in all of them. Genotyping analysis of the full pedigree identified nine mice with a homozygous Smpd1^C248S/C248S substitution, and three heterozygous or wild-type mice that were unaffected. Fig. 3. [145]Fig. 3 [146]Open in a new tab Identification of a mild brain microglial phenotype for Smpd1^C248S/C248S mice. A) Gene set scores for pedigree 55279 affected mice (Smpd1^C248S/C248S mice, red triangles) or unaffected mice (red circles) versus other mice from the screen (gray circles). B, C) Gene expression heatmap for genes belonging to the DAM gene set (B) or Macrophage gene set (C) for affected (Smpd1^C248S/C248S, red triangles) or unaffected (red circles) mice from pedigree 55279. D) Gene expression heatmap for differentially-expressed genes (Smpd1^C248S/C248S vs. unaffected FDR < 0.05). Symbols and color scale are same as in B. E, F) Representative images at low magnification (top) or high magnification (bottom), or quantification in cortex (F) for microglia lysosome/activation marker CD68 in coronal brain sections. Top panel scalebar: 1 mm. Bottom panel scalebar: 100 μm. Data points represent individual mice. Data is plotted as mean ± SEM, student’s t-test (two-sided), ****P < 0.0001 for F. To investigate whether the RNA-seq signals correspond to bona fide alterations of microglia, hemibrains were sectioned and stained for CD68. Strong elevation was clear in all four tested Smpd1^C248S/C248S mice, with an evenly-distributed staining pattern distinct from Ctsd^T263A/T263A mice (Fig. [147]3E, F). Smpd1 knockout mice likewise exhibit foamy macrophage accumulation in multiple tissues, including microglia bearing phospholipid inclusions in the retina [[148]52, [149]53]. Because Smpd1 variants cause lysosomal storage disease in human [[150]54] and CNS degeneration with hyperactivated microglia in mouse [[151]52, [152]53, [153]55], the Smpd1^C248S variant is likely responsible for the observed phenotype. Differential expression analysis of Smpd1^C248S/C248S brain RNA versus the same pedigree’s unaffected mice revealed upregulated inflammatory markers Gpnmb (DAM) and Gfap (reactive astrocytes), but downregulated markers of mature oligodendrocytes (Mbp, Mog, Mal, Opalin, Cldn11, Tshb, Qdpr, Ppp1r17) (Fig. [154]3D). These results are indicative of disrupted oligodendrocyte homeostasis accompanying inflammation in Smpd1^C248S/C248S mice. We next assessed whether our screening approach could identify mice with abnormal microglia but lacking strong MSG or DAM signals. One mouse (pedigree 52131) exhibited a high macrophage score, without other inflammatory signatures (Figure [155]S3A-C). Two mice (including the one exhibiting the high macrophage score) were selected for histological analysis based on hierarchical clustering relatedness. Highly elevated CD68 staining was readily apparent in these two mice, with relatively even distribution throughout the brain reminiscent of the Smpd1^C248S/C248S (Figure [156]S3E, F). Exome sequencing of the two affected and three unaffected mice identified Idua^E130V as the only candidate variant. Genotyping analysis confirmed that the two affected mice were homozygous for Idua^E130V, whereas unaffected mice were heterozygous or bore wild-type Idua. Idua encodes a glycosaminoglycan hydrolase, deficiencies of which lead to mucopolysaccharidosis and abnormal brain development. Differential expression analysis of the two Idua^E130V/E130V mice versus the same pedigree’s unaffected mice identified a set of consistent upregulated genes (Figure [157]S3D). However, no Reactome pathway showed significant enrichment, and inflammatory markers were conspicuously unaltered. Together, Smpd1^C248S/C248S and Idua^E130V/E130V pedigrees illustrate the sensitivity of our screening approach to call subtle transcriptional abnormalities, particularly when aided by histological validation. Furthermore, we find that Smpd1^C248S/C248S and Idua^E130V/E130V mutations result in similar histological abnormalities related to microglia lysosome accumulation, despite distinct transcriptional signatures. An Nlrp1a gain-of-function mutation drives widespread reactive astrocytes We next explored whether our screen could identify mice with reactive astrocyte abnormalities. We curated a set of markers with enriched expression in reactive astrocytes, opting to include some markers also upregulated by disease-associated oligodendrocytes to provide a larger gene set that would provide higher sensitivity upon averaging [[158]11, [159]12, [160]26, [161]27, [162]56] (Supplementary Table 1). Although we observed reactive astrocyte score elevation accompanying DAM score elevation for Ctsd^T263A/T263A samples, two mice (pedigree 55208) exhibited high reactive astrocyte scores accompanied by relatively mild DAM scores (Fig. [163]4A-C). Exome-sequencing of two affected and four unaffected mice identified variants common to both affected mice, including Nlrp1a^D573G. The ENU-derived mutation in Nlrp1a caused an Asp to Gly amino acid substitution at position 573 that was predicted to be likely damaging or deleterious [[164]38, [165]43]. Genotyping analysis found that both affected mice were homozygous for the Nlrp1a^D573G allele, whereas unaffected mice bore wild-type Nlrp1a or were heterozygous. Immunostaining analysis revealed striking microglial hypertrophy and IBA1 staining elevation in Nlrp1a^D573G/D573G mice (Fig. [166]4E, F). Differential expression and pathway enrichment analysis identified upregulation of Reactome pathway markers for neutrophils (S100a8, S100a9, Lrg1, and Lcn2), interferon response (Ifitm3, Ifitm1, Gbp2, and Gbp4), antigen presentation (B2m, Psmb8, H2-k1, H2-aa and H2-d1), and the classical complement cascade (C1qb, C1qc, C4b, and C3) (Fig. [167]4D). Fig. 4. [168]Fig. 4 [169]Open in a new tab Identification of a distinct reactive astrocyte phenotype for Nlrp1a^D573G/D573G mice. A) Gene set scores for pedigree 55208 affected mice (Nlrp1a^D573G/D573G mice, red squares) or unaffected mice (red circles) versus other mice from the screen (gray circles). B, C) Gene expression heatmap for genes belonging to the DAM gene set (B) or Macrophage gene set (C) for affected (Nlrp1a^D573G/D573G, red triangles) or unaffected (red circles) mice from pedigree 55208. D) Gene expression heatmap for differentially-expressed genes (Nlrp1a^D573G/D573G vs. unaffected FDR < 0.001). Symbols and color scale are same as in B. E, F) Representative images at low magnification (top) or high magnification (bottom), or quantification in cortex (F) for microglia marker IBA1 in coronal brain sections. Top panel scalebar: 1 mm. Bottom panel scalebar: 100 μm. Data points represent individual mice. Data is plotted as mean ± SEM, student’s t-test (two-sided), **P < 0.01 for F. Germline gain-of-function mutations in Nlrp1a and NLRP1, the human orthologue of murine Nlrp1a, cause lethal systemic inflammation [[170]57] and a spectrum of inflammatory disorders [[171]58–[172]61], respectively. To test whether Nlrp1a^D573G leads to constitutive inflammasome activation and resulting lytic cell death akin to other previously-reported NLR family gain-of-function variants [[173]62], we turned to a keratinocyte cell model that is broadly used for probing human NLRP1 gain of function mutations [[174]58–[175]61]. Stable cell lines were generated, bearing doxycycline-inducible transgenes encoding for either GFP, wild type NLRP1, a previously characterized hyperactive NLRP1^A66V 53–56 or NLRP1^D748G, with D748 representing the equivalent position to murine D573. All constructs were doxycycline-inducible and achieved comparable expression levels (Fig. [176]5A). Unlike wild-type NLRP1, the induction of NLRP1^D748G expression resulted in significant lytic cell death, as measured by membrane-impermeant propidium iodide (PI) uptake (Fig. [177]5B). This effect was comparable to the reported hyperactive NLRP1^A66V missense mutation, indicating NLRP1^D748G was a gain-of-function mutation, albeit with delayed kinetics. Fig. 5. [178]Fig. 5 [179]Open in a new tab NLRP1^D748G and Nlrp1a^D573G are recessive gain-of-function mutations and lead to inflammasome activation and neuroinflammation. A) Western blot of Ker-CT reconstituted with respective constructs illustrating expression of wild-type (WT) or mutant versions of Homo sapiens NLRP1 (hsNLRP1) or GFP negative control, with or without 18 h of doxycycline induction (dox). B) Cell death measurement of Ker-CT reconstituted with respective constructs in the presence of doxycycline over time as propidium-iodide (PI) positive cell counts plotted as a percentage of maximal cell death achieved with Triton X-100 treatment at the beginning of the experiment. C) Schematic diagram showing the design of the AAVs used to infect primary mouse microglia. Created with BioRender.com. D) Quantification of Nlrp1a mRNA expression level after AAV incubation for 5 or 6 days as indicated in mouse primary microglia measured by RT-qPCR. Target gene expression was first normalized to the housekeeping gene Gapdh and then normalized relative to the expression in microglia with no AAV treatment. n = 4, from two different litters of mice. Data are represented by mean ± SEM. E) Representative images showing the PI + microglia 12 h after the addition of PI dye and 0.5 mM Val-boroPro in the wells without AAV, or infected with GFP, Nlrp1a ^WT or Nlrp1a ^D573G AAV, as indicated. Top, phase images with red fluorescence; bottom, red fluorescence only. Scalebar: 100 mm. F) In the presence of Val-boroPro, quantification of the PI + cells for the conditions indicated at each time point is shown in time course traces (left) and the quantification of the PI + cells at 12 h after PI dye addition is shown in bar graph (right). All points were normalized to percentage of PI + cells of the no AAV treated microglia at 12 h. n = 4, from two different litters of mice. Data are represented by mean ± SEM. Time course traces, two-way ANOVA, ***, p < 0.001. Bar graphs, one-way ANOVA with Tukey’s multiple comparisons test, *, p < 0.05, **, p < 0.01. G, H) Representative images at low magnification (top) or high magnification (bottom), or quantification in whole brain (H) for microglia marker IBA1 in coronal brain sections. Top panel scalebar: 1 mm. Bottom panel scalebar: 100 μm. I, J) Representative images at low magnification (top) or high magnification (bottom), or quantification in whole brain (J) for astrocyte marker GFAP in coronal brain sections. Top panel scalebar: 1 mm. Bottom panel scalebar: 100 μm. K, L) Representative images at low magnification (left) or high magnification (right), or quantification in thalamus or fimbria (L) for neutrophil marker MPO in coronal brain sections. Yellow arrowheads indicate neutrophils. Large panels scalebar: 1 mm. Small panels scalebar: 100 μm. Data points represent individual mice. Data is plotted as mean ± SEM, student’s t-test (two-sided), *P < 0.05 for H, J, L. Single-cell RNA-seq datasets indicate that Nlrp1a is not highly expressed by murine astrocytes, but instead primarily expressed by microglia in the mouse brain [[180]26]. We thus reasoned that microglial inflammasome activation may be upstream of the reactive astrocyte phenotype. To explore this hypothesis, we tested the effects of AAV-driven overexpression of Nlrp1a or Nlrp1a^D573G in primary microglia cultures with or without NLRP1 agonist Val-boroPro [[181]63] (VB; Fig. [182]5C–F, Figure [183]S4). We found strong overexpression of both Nlrp1a constructs at relevant time points for the experiment (Fig. [184]5D), with Nlrp1a^D573G inducing cell death or sensitizing cells to VB-induced cell death (Fig. [185]5E, F). Thus, Nlrp1a^D573G represents a gain-of-function mutation sufficient to enhance inflammasome activation in microglia. We next re-animated the Nlrp1a^D573G pedigree using cryo-preserved samples. Nlrp1a^D573G mice were viable as heterozygotes, but fewer homozygous mice survived to weaning. In all, sixteen Nlrp1a^+/+, thirty-eight Nlrp1a^D573G/+ and only four Nlrp1a^D573G/D573G homozygous mice were recovered from a series of heterozygous crosses. Recovered Nlrp1a^D573G/D573G homozygous mice had increased IBA1 microglial (Fig. [186]5G, H) and GFAP astrocytic staining (Fig. [187]5I, J) throughout the brain, as well as elevated neutrophil marker MPO in the fimbria and thalamus (Fig. [188]5K, L). These findings corroborate the results from our initial screen in which Nlrp1a^D573G pathway markers for DAM, reactive astrocytes, and neutrophils were enriched. These findings validate our screening approach for identification of reactive astrocyte states and provide contrast with the inflammatory syndromes associated with Nlrp1a^D573G versus Nrros^C247Y, Ctsd^T263A, Smpd1^C248S, or Idua^E130V. Furthermore, we find that Nlrp1a^D537G is a gain-of function variant that results in widespread astrocyte activation, likely downstream of inflammasome hyperactivation in Nlrp1a-expressing microglia and/or neutrophils. White matter inflammation in Inpp5d loss-of-function mice We next leveraged previously collected exome sequencing data ([189]https://databases.apf.edu.au/mutations/) to identify ENU-mutagenized pedigrees bearing mutations in a specific gene of interest, Inpp5d. The human orthologue INPP5D has been implicated in Alzheimer’s Disease through GWAS [[190]64], and Inpp5d loss-of-function mice exhibit altered microglia-plaque interactions [[191]65–[192]67]. However, Inpp5d loss-of-function also leads to peripheral inflammatory syndromes driven by myeloid cell hyperaccumulation in various tissues [[193]68], and conditional deletion from microglia has recently been reported to elevate phagocytosis and complement signaling associated with over-pruning of synapses developmentally [[194]69]. To better understand the impact of Inpp5d loss-of-function on CNS homeostasis, we reanimated a line bearing the Inpp5d^W8X loss-of-function allele for analysis with our screening pipeline (Fig. [195]6A). Fig. 6. [196]Fig. 6 [197]Open in a new tab Reverse-genetics application identifies neutrophil and striatum phenotypes from Inpp5d disruption. A) Schematic of the reverse-genetics approach. Created with biorender.com. B, C) DAM (B) or Neutrophil (C) gene set scores for littermates bearing wild-type, heterozygous, or homozygous Inpp5d^W8X alleles. D) Gene expression heatmap for differentially-expressed genes (Inpp5d^W8X/W8X vs. unaffected FDR < 0.05). Each column represents one mouse’s bulk brain RNA-seq sample, with the genotype specified by the color at the top of each column. E) Representative images (E) at low magnification (top) or high magnification in white matter (bottom) for microglia marker IBA1 in coronal brain sections. Top panel scalebar: 1 mm. Bottom panel scalebar: 100 μm. F) Quantification of IBA1 area coverage in cortex (ctx, top) or fimbria white matter (fim, bottom). G) Quantification of CD68 area coverage in fimbria white matter Data points represent individual mice. Data is plotted as mean ± SEM, one-way ANOVA with Tukey’s post-hoc test, *P < 0.05, **P < 0.01 for B, C, F, G. Because myeloid infiltration into peripheral organs occurs with Inpp5d knockout [[198]68], we defined Peripheral Myeloid gene set using markers highly enriched in neutrophils or monocytes within myeloid RNA-seq or brain single-cell RNA-seq datasets (Supplementary Table 1) [[199]16, [200]26]. Inpp5d^W8X/W8X homozygous mice exhibited elevated DAM and Peripheral Myeloid scores (Fig. [201]6B, C), but heterozygotes showed no change. Differential-expression and pathway analysis identified Reactome pathways associated with neutrophil degranulation (S100a8, S100a9, Lcn2, and Lrg1) and complement signaling (C1qb and C1qc) (Fig. [202]6D). Histological examination identified a mild IBA1 clustering phenotype that did not affect gray matter regions such as cortex but was evident in white matter tracts such as the fimbria (Fig. [203]6E-G). These results illustrate a reverse-genetics application of our platform resulting in identification of white matter inflammation associated with neutrophil gene expression signature changes consequent to Inpp5d disruption. Multimodal analysis of heterogenous brain inflammation syndromes Many of the diverse phenotypes highlighted from our screen exhibit partially overlapping sets of differentially-expressed genes. We next explored to what extent distinct gene expression modules could be used to highlight similarities and dissect differences across phenotypes. To more directly compare the inflammatory state associated with each genetic variant, we plotted gene set scores side-by-side for each pedigree (Fig. [204]7A), incorporating additional gene set scores describing the abundance of border-associated macrophage signatures (BAM) [[205]21], IFN-response signatures [[206]13, [207]70], and mature oligodendrocyte signatures [[208]26, [209]71] (Supplementary Table 1). Each mouse line presented a different response profile. Ctsd^T263A/T263A mice exhibited strong DAM and IFN responses associated with reactive astrocyte markers and elevated macrophage content, but without indication of peripheral monocyte/neutrophil infiltration. Nlrp1a^D573G/D573G mice strongly induce reactive astrocyte and peripheral myeloid markers, with relatively mild elevation of IFN-response and DAM markers compared to Ctsd^T263A/T263A. Smpd1^C248S/C248S exhibited moderate DAM and macrophage score elevation, but no reactive astrocyte signature. Idua^E130V/E130V mice exhibited only mild elevation of macrophage marker genes without DAM or reactive astrocyte signature elevations, despite striking changes in CD68 immunoreactivity in tissue sections. Nrros^C247Y/C247Y mice exhibited elevated BAM signature with reduced MSG scores, consistent with prior reports of Nrros^−/− [[210]21]. Inpp5d^W8X/W8X mice exhibited strong monocyte/neutrophil signature and reactive astrocyte scores without detectable deviation on other metrics. Fig. 7. [211]Fig. 7 [212]Open in a new tab Multimodal transcriptional phenotyping across pedigrees of interest. A) Gene set scores for mutant affected (striped bars) versus unaffected (solid bars) mice from pedigrees producing Ctsd^T263A/T263A (darkest color), Nlrp1a^D573G/D573G, Smpd1^C248S/C248S, Idua^E130V/E130V, Nrros^C247Y/C247Y, or Inpp5d^W8X/W8X (lightest color) mice. Gene set scores are for DAM (orange), Macrophage (red), Reactive Astrocyte/Oligodendrocyte (green), Microglia Signature (blue), Peripheral Myeloid (purple) BAM (brown), IFN-response (yellow), or Mature Oligodendrocyte (magenta). Data is plotted as mean ± SEM, two-way ANOVA with Tukey’s post-hoc test, **P < 0.01. B-F) Differentially-expressed genes (FDR < 0.001 in at least one comparison) plotted as log[2] fold-change (LFC) in expression associated with Ctsd^T263A/T263A (x-axis) compared to each-gene’s y-axis LFC in expression associated with Nlrp1a^D573G/D573G (B), Smpd1^C248S/C248S (C), Idua^E130V/E130V (D), Nrros^C247Y/C247Y (E), or Inpp5d^W8X/W8X (F). Gene name labels are provided for a subset of genes of interest, with genes belonging to gene set scores colored as in A. A simple linear regression was performed to determine the r^2 value. Mature oligodendrocyte marker expression was also differentially affected across these six pedigrees (Fig. [213]7A). Ctsd^T263A/T263A, Smpd1^C248S/C248S, and Inpp5d^W8X/W8X mice exhibit decreased mature oligodendrocyte markers, whereas Idua^E130V/E130V exhibited elevated mature oligodendrocyte scores. These findings suggest alterations in mature oligodendrocyte density and/or myelination that is not strictly coupled to any other set of inflammatory markers assessed. To more precisely define similarities and differences in regulation of individual marker genes, we next compared each pedigree’s expression changes directly to those of Ctsd^T263A/T263A (Fig. [214]7B-F). We plotted the fold-change in expression for Ctsd^T263A/T263A (x-axis) versus a comparator mutant such as Nlrp1a^D573G/D573G (y-axis, Fig. [215]7B), including only genes that had very high confidence (FDR < 0.01) for differential-expression for at least one of the two mutants. This analysis highlighted an overall correlation between Ctsd^T263A/T263A and Nlrp1a^D573G/D573G gene expression changes, with upregulation of astrocyte/oligodendrocyte response marker Serpina3n and upregulation of IFN-response markers Ifitm3 and Oasl2 (Fig. [216]7B). However, stark differences between the phenotypes are also readily appreciated, with neutrophil makers S100a8 and Lrg1 more strongly upregulated in Nlrp1a^D573G/D573G brains, and DAM markers Lgals3 and Gpnmb upregulated much more strongly with Ctsd^T263A/T263A (Fig. [217]7B). Smpd1^C248S/C248S and Ctsd^T263A/T263A exhibit more similarities, with DAM markers exhibiting upregulation and mature oligodendrocyte markers Plp1, Mal, and Opalin exhibiting downregulation in both mutants (Fig. [218]7C). Idua^E130V/E130V showed weaker expression changes overall compared to Ctsd^T263A/T263A (Fig. [219]7D), and Nrros^C247Y/C247Y was associated with strong changes in MSG markers Selplg or Hexb, in addition to BAM marker Ms4a7 (Fig. [220]7E). Finally, Inpp5d^W8X/W8X exhibited strong upregulation of neutrophil markers Lrg1 and S100a8, but not other markers of glial activation. These comparisons also highlight behavior for other marker genes of interest such as Lcn2, which is highly upregulated by reactive astrocytes, but is also highly expressed by neutrophils (Fig. [221]7F). In all, individual gene analysis supports the validity of gene set score comparisons, which capture the predominating transcriptional changes that differentiate the mouse lines analyzed. These results identify inflammatory heterogeneity across diseased mouse brains, and provide a framework for categorization of inflammatory states based on bulk tissue RNA-seq analysis. Discussion We describe a powerful approach for multimodal phenotyping of genetically heterogenous brain samples, which we utilize to discover new variants causing varied inflammatory states. By capturing a diverse set of inflamed brains in a single dataset, we illustrate the effectiveness of our curated gene set scores for differentiating mutant phenotypes and provide histological validation to confirm key findings. Although higher informational content can be derived from single-cell or spatial transcriptomic approaches, the higher throughput afforded by our approach enabled screening in vivo at a sufficiently large scale to identify novel variants. Illustrating the broad potential utility of our approach and dataset, we discover several specific variants that alter brain immune homeostasis. We identify an adult-viable mouse model related to CLN10 disease that exhibits strong upregulation of IFN-response genes, consistent with recent reports of interferon-regulating cGAS/STING activation playing a central role in Ctsd-knockout neuronal cells [[222]72]. Additionally, we find an Nlrp1a gain-of-function mutation that promotes widespread glial activation, with evidence of myeloid cell infiltration and IFN signaling in the brain. Structural studies investigating an auto-inhibited human NLRP1 [[223]73] found that D748 (the human equivalent position to D573 in mouse Nlrp1a) is located at the center of the interface between the NACHT and LRR domains. Given that NACHT-LRR domain interactions facilitate autoinhibition in NLRC4 and gain-of-function NLRP3 variants also occur at this interface [[224]74, [225]75], the spontaneous inflammasome activation we observe for Nlrp1a^D573G may likewise result from a loss of autoinhibitory regulation. Although we identified Nlrp1a^D573G based on the striking astrocyte phenotype, Nlrp1a expression patterns suggest that astrocytes are affected indirectly, for example via release of microglial interleukins, which can potently stimulate astrocyte reactivity [[226]56]. In human, NLRP1 gain-of-function variants are known to cause autoinflammatory disorders [[227]58–[228]61, [229]76], including recessive NACHT domain variants associated with multiple sclerosis [[230]77]. We also leverage the ENU-mutagenized mouse colony to investigate Inpp5d loss of function mice. Inspired by human genome-wide association signals, Inpp5d is being explored as a target for Alzheimer’s Disease, and deletion of microglial Inpp5d alters microglia interactions with amyloid plaques mouse models of disease [[231]65–[232]67]. We find a CNS neutrophil phenotype and white matter inflammation phenotype, indicating that systemic inhibition of Inpp5d may cause undesirable CNS effects akin to the autoinflammatory conditions observed in peripheral tissues of Inpp5d knockout mice [[233]68]. However, Inpp5d heterozygous knockout is sufficient to partially affect microglia-plaque relationships [[234]66], and we do not see a linear gene dosage effect on homeostatic dysregulation in heterozygous loss-of-function mice, suggesting potential for a therapeutic window of desired activity with limited undesirable effects. These findings highlight the potential utility of reverse-genetics mining of existing databases to identify novel function-altering variants and for exploring in vivo impact of candidate therapeutic targets. The high information-content provided by RNA-seq can be used to assess many phenotypes simultaneously. Although our assay conditions were optimized with microglia abnormalities in mind, we also identify reactive astrocyte hits using the same dataset. By carefully defining gene set scores, additional CNS phenotypes could be explored, and the same approach could be applied to other tissues or other heterogeneous mouse colonies with minimal modification [[235]78]. However, several technical limitations should be considered. First, variance is high for expression measurements of any single gene, with variability decreasing in proportion to higher transcript abundance; averaging across multiple genes and focusing on more abundant transcripts increases power. Second, the overall variance of a gene or gene set measurement must be considered against realistic effect magnitudes; for instance, microglia density can vary wildly with experimental manipulation without acute toxicity [[236]79, [237]80], but two-fold elevation for an abundant cell type such as astrocytes would be unprecedented. Finally, each gene added to a gene set introduces risk that the desired signals from the cell type or cell state of interest may become diluted by expression from other irrelevant populations. Thus, desired signals may be masked due to the genes chosen for investigation being non-specific for the phenotype of interest. Although our initial investigations have focused on genes related to prior discoveries from human syndromes, additional candidate variants with less established relationships to CNS immune regulation will be of interest for future study. In all, our findings provide rich insights into diverse inflamed brain states arising from a diverse range of causal variants, and provide an experimental and analytical platform that can be rapidly integrated into future studies. Methods Mice and exome sequencing All animals used in this study were cared for and used in accordance with protocols approved by the Australian National University Animal Experimentation Ethics Committee (A2018/07 and A2021/22) and the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes. Founder C57BL/6NCrl males were treated with 3 × 100 mg/kg ENU (Sigma) and mutations bred to homozygosity in three generations as described [[238]81]. Generation 3 mice were euthanized by cervical dislocation, usually between 8 and 12 weeks old, and an ear notch for DNA and brain samples were collected. Typically, 12 siblings from each pedigree were screened. DNA extraction, exome sequencing, and variant calling were performed as previously described [[239]81]. RNA extraction One hemisphere was collected in 1.6 mL of RNAlater (ThermoFisher) and stored at 4°C for up to 5 days. To homogenize the tissue, RNAlater was removed, and a 5 mm stainless steel bead (Qiagen) and 1.6 mL of Qiazol (Qiagen) was added. The tissue was homogenized using a Tissue Lyser II (Qiagen) for at least 30 minutes until a homogenous solution was confirmed. 400 µL of the homogenate solution was transferred into a new tube and allowed to sit for 5 minutes before adding 100 µL of chloroform (Sigma). The homogenate-chloroform mixture was shaken vigorously, allowed to stand for 2 minutes before centrifugation for 5 minutes at 4°C and 1000 x g. The top clear layer was removed and stored at -80 °C. For RNA extraction, 120 µL was transferred into a QIAgen S-block and processed on the QIAcube using the RNeasy 96 QIAcube HT Kit. RNA sequencing Several RNA-seq library methods were tested following published or commercial protocols including BRB-seq [29], Tag-seq [32], QIAseq UPX 3’ Transcriptome Kits (3’UPX, Qiagen), TruSeq RNA Library Prep Kit v2 (Truseq, Illumina). However, with the exception of Fig. [240]1B and FigS1, all cDNA fragment Libraries were constructed using the SMART-Seq mRNA 3’DE Kit (Takara Bio) following manufacturer’s recommendations, with minor modifications. Total RNA samples were processed in batches of 1152, distributed across twelve 96-well plates. For each sample, 4 ng total RNA was used as input for cDNA synthesis, then amplified for 8 cycles, with each sample in a row receiving a unique in-line index. Amplified cDNAs within a row were then combined into a 12-plex pool, cleaned up with Ampure XP beads (Beckman Coulter) and quantified using the Qubit HS DNA assay kit (Thermo Fisher Scientific). For each of the 96 cDNA pools generated from 12 plates, 400 pg were carried into Nextera XT library preparations (Illumina Inc.), each assigned with one of the 96 unique combinations of Forward and Reverse HT Index Primers included in the SMART-Seq mRNA 3’DE kit. After Ampure XP clean up and Qubit quantification, the 96 pools were combined in an equimolar manner to generate a 1152-plex Illumina library. This library was concentrated with Ampure XP beads, quantified with Qubit and assessed on the TapeStation with the High Sensitivity D1000 assay (Agilent Technologies). Library pooling was normalized using an initial measurement on an Illumina MiSeq with a v2 Nano flow cell, and concentrations were re-adjusted when necessary prior to sequencing on a NovaSeq 6000 S2 100 cycle kit (Illumina Inc.), with a read length configuration 96-8-8-26. Raw files were basecalled and demultiplexed using bcl2fastq2 v2.20.0 (Illumina Inc.), generating data for each of the 96 12-plex pools. Further Demultiplexing based on in-line indexes was performed using the SMART-Seq mRNA 3’DE Demultiplexer software (Takara bio). RNA-seq data analysis RNA-seq reads were aligned to the mouse reference genome GRCm38 using GSNAP [[241]82]. Only uniquely mapping reads were carried forward to downstream analysis. Sample quality was further assessed on a per-plate basis based on total read counts, percentage of reads mapping to the reference genome, and percentage of reads mapping to exons. Reads predominantly mapped to transcript 3’ ends, as expected from the library preparation method. The mean read Depth was 1.84 million exonic reads mapped per sample, with a standard Deviation of 0.67 million reads. Samples with less than 350,000 mapped exonic reads were discarded due to anticipated variance in gene expression level estimates. For a subset of samples with low initial read depth, library preparation and sequencing were repeated and data included in the dataset in sufficient reads were observed. For each sample, counts were normalized per million total exonic reads (CPM). CPM values for each gene were normalized to that gene’s mean expression level across the full dataset, then the mean of these normalized values was calculated for each gene set. For gene set scores with low baseline expression (DAM, BAM, Peripheral Myeloid, Reactive Astrocyte, and IFN-response scores), a stabilizing offset of + 1 CPM was added to each gene’s expression value prior to normalization to mitigate high variance of scarcely detectable genes. Hits were typically selected if gene set score signal differed from the mean by 4 standard deviations or more. A subset of hits was re-sequenced to provide technical replicate data and increase confidence in the signals. Gene set score selection For gene set selection, initial lists were generated by selecting genes with highest fold-changes from various differential-expression analyses of relevant datasets [[242]12, [243]16, [244]21, [245]42, [246]56]. Initial lists that gave high-variance scores across unaffected mice were refined further. For the Macrophage gene set, EGAD coexpression analysis was performed to identify marker genes whose expression correlated most highly with Csf1r across a large collection of mouse bulk tissue RNA-seq datasets [[247]40, [248]41], and genes with higher expression by microglia over other tissue macrophage populations were eliminated [[249]16]. For the Mature Oligodendrocyte gene set, genes were selected based on high cell-type specificity in a large brain scRNA-seq dataset analysis [[250]71]. For the IFN-response gene set, genes were taken from a previously-defined gene module [[251]13], then filtered to eliminate genes exhibiting low fold-induction relative to other module genes, either for affected mice in our dataset or previously published viral-infection datasets [[252]70]. All gene sets were further filtered to remove genes with a high observed coefficient of variance baselines (typically CV > 0.6) across unaffected mice from our dataset. Gene sets were checked using existing datasets; for Figure [253]S1, a pseudobulk data from a published single-cell RNA-seq dataset were used (wild-type mice only) [[254]35]. Differential expression analysis and visualizations Differential expression analysis was performed using voom + limma [[255]83] comparing affected mice to unaffected mice from the same pedigree. For heatmaps, varying FDR cutoffs were used to generate sets of 20 to 50 highest-confidence differentially expressed genes. The same set of genes used for heatmaps were exported for pathway analysis with Enrichr [[256]84] and Reactome [[257]85] pathways. Only pathways with adjusted p-value < 0.05 are reported in the text. Tissue sectioning Hemibrains were fixed in 4% PFA (ProSciTech) and coronally sectioned at 30 μm in a solid matrix block (MultiBrain processing by Neuroscience Associates). Sectioned brain sheets were stored at − 20 °C in cryoprotectant solution (30% glycerol, 30% ethylene glycol in PBS) until use. Chromogenic staining, imaging, and analysis For Figs. [258]1F, [259]2E, [260]3E and [261]4E, chromogenic stains for CD68 and TMEM119 were conducted at Neuroscience Associates as previously described [[262]42]. Neutral red counterstain was included for CD68 stains. For Fig. [263]5G and I and S3E, IBA1, GFAP, and CD68 was performed was as previously described [[264]86]. All stained brain sections were scanned with a NanoZoomer-XR slide scanner (Hamamatsu Photonics) and quantified using Fiji software v1.54f (ImageJ). The quantification for the percent area coverage by a stain was calculated by tracing regions of interests (i.e. cortex overlaying the hippocampus; thalamus) and assigning a single threshold per unique stain. All histological studies utilized two to three brain sections per mouse, which were then averaged for an individual biological replica unless otherwise noted. Fluorescent staining, imaging, and analysis Fluorescent staining was performed as previously described [[265]86]. Brain sections were scanned with the VS200 slide scanner (Olympus) and quantified using QuPath v.0.5.1. The quantification for the percent area coverage by a stain was calculated by tracing regions of interests (i.e. cortex overlaying the hippocampus; thalamus) and assigning a single threshold per unique stain. All histological studies utilized one to two brain sections per mouse, which were then averaged for an individual biological replica unless otherwise noted. Genotyping Genotyping was performed using allele specific primers optimized for KASP assay (LGC Biosearch Technologies). From primer sequences see Supplementary Table 2. In vitro investigation of Nlrp1a variants AAVs were generated by VectorBuilder. Names and IDs of AAVs are listed below: Vector ID Vector Name VB240308-1371kwt pAAV[Exp]-SFFV > EGFP: WPRE3 VB240308-1362kmg pAAV[Exp]-SFFV > mNlrp1a[[266]NM_001004142.2]: WPRE3 VB240308-1363skb pAAV[Exp]-SFFV>{mNlrp1 D573G}:WPRE3 [267]Open in a new tab Primary microglia were cultured from P0-P2 pups as previously described [[268]35]. Briefly, mouse brains were dissected out and the brain tissue was disrupted by trituration using a 10 mL serological pipette in cold DMEM media. The homogenate was spun at 300 g for 5 min. The pellet was resuspended in DMEM media and filtered through a 70 μm cell strainer. Dissociated cells were cultured in 175 cm^2 flasks with media containing DMEM, 10%FBS and 1% penicillin/streptomycin. Flasks were rinsed with PBS and new media was added after 24 h. After 11–12 days, microglia were shaken off (125 rpm for 2 hr), collected and plated at 40k/well in the presence of 20 ng/mL CSF-1 (R&D systems, 416-ML-010/CF). One day later, microglia were infected with AAV at MOI of 400,000. 5 days post infection, Propidium Iodide (PI, ThermoFisher, P3566, 1:2000) was added to the culture with or without Val-boroPro (APExBIO, B3941) and images were taken every 2 hours using Incucyte Live-Cell-Analysis system, capturing both phase and red fluorescence images. Dead cell (PI + cell) puncta were quantified using the built-in analysis modules. The percentage of PI positive cells were calculated by normalizing the number of PI + cells in the experimental wells to the number of PI + cells in the wells that were treated with 5 mM Val-boroPro or 0.5% Triton X-100 to induce all cell death in the presence of PI dye. Red fluorescence images were created using ImageJ from raw images exported from Incucyte. Phase images were directly from Incucyte. To measure Nlrp1a mRNA level in microglia, RNA was extracted using RNAeasy mini kit (Qiagen) from microglia cell lysates and RT-qPCR was performed using qScript XLT One-Step RT-qPCR ToughMix, Low ROXTM (QuantaBio). TaqMan assay probes were obtained from ThermoFisher Scientific: Mm03047263_m1 (Nlrp1a) and Mm99999915_g1 (Gapdh housekeeping control). Cell culture Human Ker-CT keratinocytes (ATCC Cat# CRL-4048) were cultivated in a 1:2 mixture of Ham’s F12 (Thermo Fisher Scientific Cat# 11765054) and DMEM (high-glucose, no glutamine, no calcium; Thermo Fisher Scientific Cat# 21068028) supplemented with 1% nonessential amino-acids (Thermo Fisher Scientific Cat# 11140050), 0.5% EpiLife defined growth supplement (Thermo Fisher Scientific Cat# S0125), 25 mg/mL bovine pituitary extract (Thermo Fisher Scientific Cat# 13028014), 20 ng/mL epidermal growth factor (PeproTech Cat# AF-100-15), 10 mM HEPES (Sigma-Aldrich Cat# H0887), 2 mM GlutaMAX (Thermo Fisher Scientific Cat# 35050061), 0.1 mM CaCl[2] (PromoCell Cat# C-34006), and 1% penicillin–streptomycin (Thermo Fisher Scientific Cat# 15140122). Cell lines were initially tested to be free of mycoplasma contamination. Generation of stable cell lines For this study, Ker-CT cells were generated that are deficient for NLRP1 and lack Puromycin resistance, which had been used for immortalization of those cells [[269]87]. Knock-outs were generated as described [[270]88] and following sites were targeted: NLRP1: 5′-ACTCCAAGTAACAGGCCAGG-3′/5′- ATTCCTGACGTTTCATCCAG-3′), Puromycin: 5′-ACGGTGGCCAGGAACCACGC-3′/5′-GCCTTCCATCTGTTGCTGCG-3′). For reconstitution experiments, stable cell Lines were generated by transfection using Lipofectamine 3000 according to the supplier’s protocol with a combination of a plasmid coding for piggyBac transposase and a plasmid harboring transposon-specific inverted terminal repeats (ITRs) sequences flanking a cassette coding for doxycycline-inducible GFP, NLRP1^wt, NLRP1^A66V, or NLRP1^D748G and constitutive Puromycin expression for selection. 48 h after transfection, cells were selected using 1 µg/mL puromycin (Thermo Fisher Scientific Cat# A1113803) and stable pools generated were used for the experiments. Immunoblotting For western blot experiments, 0.35 × 10^6 Ker-CT cells were plated per well of a 12 well plate and were induced the following day with 1 µg/mL doxycycline (Clontech Cat# 631311) for 18 h in the presence of 20 µM Z-VAD-FMK (Promega Cat# G7232) or were left untreated in the presence of 20 µM Z-VAD-FMK. 18 h after induction, cells were lysed in 400 µL 1x Bolt LDS sample buffer (Thermo Fisher Scientific Cat# B0007) containing 1x NuPAGE sample reducing agent (Thermo Fisher Scientific Cat# NP0004) and samples were boiled for 5 min at 95 °C. Subsequently, samples were subjected to denaturing and reducing bis-tris SDS–PAGE (Thermo Fisher Scientific Cat# NW04125BOX). Following, samples were blotted onto 0.2 μm PVDF membranes (Bio-Rad, # 1620177), blocked in 5% milk in TBST, and incubated with indicated primary and corresponding secondary antibodies (see Methods). Chemiluminescent signals were recorded using a FluorChem R System. Live-cell imaging For Live cell imaging and cell Death experiments using an Incucyte system, 0.25 × 10^5 cells were plated per well of a 96 well plate (Corning Cat# 3903). 24 h later cells were either left untreated or gene expression was induced with 1 µg/mL doxycycline in the presence of 1 µg/mL PI (Thermo Fisher Scientific Cat# P1304MP). For the lysis control, control wells of the respective genotypes were treated at the start of the experiment with 1% Triton-X100 (final concentration). Cells were imaged immediately after treatments every hour for 24 h and PI uptake measured. Antibody Source Identifier Rabbit polyclonal anti-Iba1 Wako Cat# 019-19741 Rat monoclonal anti-CD68 Bio-Rad Cat# MCA1957 Rabbit anti-Tmem119 Abcam Cat# ab209064 Goat polyclonal anti-rabbit biotin VWR Cat# 76484-260 Goat polyclonal anti-rat biotin VWR Cat# 10637-566 Goat polyclonal anti-mouse MPO Biotechne Cat# AF3667 CD68, Iba1, TMEM119 Neuroscience Associates [271]https://www.neuroscienceassociates.com/technologies/staining/ Mouse monoclonal anti-NLRP1 BioLegend Cat# 9F9B12 Rabbit polyclonal anti-GFP Cell Signaling Technology Cat# 2555 Rabbit monoclonal anti- β-actin Cell Signaling Technology Cat# 5125 Horse anti-mouse IgG-HRP Cell Signaling Technology Cat# 7076 Goat anti-rabbit IgG-HRP Cell Signaling Technology Cat# 7074 [272]Open in a new tab Supplementary Information [273]12974_2025_3556_MOESM1_ESM.xlsx^ (11.8KB, xlsx) Supplementary Material 1. Supplementary Table 1 – Genes comprising each gene set score [274]12974_2025_3556_MOESM2_ESM.xlsx^ (9.7KB, xlsx) Supplementary Material 2. Supplementary Table 2 – Genotyping primer sequences [275]12974_2025_3556_MOESM3_ESM.png^ (351.2KB, png) Supplementary Material 3. Supplementary Figure 1: Validation of microglia-signature gene expression in an independent single-cell RNA-seq dataset. Pseudobulk analysis of genes used for the microglia gene set score across 12 cellular clusters based on wildtype samples from a published single-cell RNA-seq dataset [85] A) Gene set score averaged across 8 microglia signature genes (log2 expression . B) Expression of each individual microglia signature genes. MSG: Microglia Signature Genes. PVM: Perivascular macrophages. BMEC: Brain microvascular endothelial cells. NPC: Neural progenitor cells. OPC: Oligodendrocyte precursor cell. Reelin: Reelin-positive interneurons. VLMC: Vascular leptomeningeal cells. VSMC: Vascular smooth muscle cells. [276]12974_2025_3556_MOESM4_ESM.png^ (129.9KB, png) Supplementary Material 4. Supplementary Figure 2: Illustration of the observed variance in expression level (coefficient of variation) for individual genes across repeat sequencing experiments, plotted as a function of gene expression level (counts per million, or CPM), with both axes on a log scale. Samples sequenced with different methods (Tru-seq, BRB-seq, or SMART-Seq mRNA 3’DE seq) exhibit different profiles. Red dashed line represents the theoretical Poisson distribution for each experiment’s mean read depth. [277]12974_2025_3556_MOESM5_ESM.png^ (935.9KB, png) Supplementary Material 5. Supplementary Figure 3: Identification of dysregulated microglia in Idua^E130V/E130V mice. A) Gene set scores for pedigree 52131 affected mice (Idua^E130V/E130V mice, red triangles) or unaffected mice (red circles) versus other mice from the screen (gray circles). B, C) Gene expression heatmap for genes belonging to the DAM gene set (B) or Macrophage gene set (C) for affected (Idua^E130V/E130V, red triangles) or unaffected (red circles) mice from pedigree 52131. D) Gene expression heatmap for differentially-expressed genes (Idua^E130V/E130V vs unaffected). Symbols and color scale are same as in B.E, F) Representative images at low magnification (top) or high magnification (bottom), or quantification in cortex (F) for microglia lysosome/activation marker CD68 in coronal brain sections. Top panel scalebar: 1 mm. Bottom panel scalebar: 100 μm.Data points represent individual mice. Data is plotted as mean ± SEM, student’s t-test (two-sided), ****P < 0.0001 for F. [278]12974_2025_3556_MOESM6_ESM.jpeg^ (322KB, jpeg) Supplementary Material 6. Supplementary Figure 4: AAV-overexpression of mNlrp1a D573G in microglia monoculture. A) Representative images showing the PI + microglia 12 hours after the addition of PI dye in the wells without AAV, or infected with GFP, Nlrp1a WT or Nlrp1a D573G AAV, as indicated. Top, phase images with red fluorescence; bottom, red fluorescence only. Scalebar: 100 mm. B) In the absence of Val-boroPro, quantification of the PI + cells for the conditions indicated at each time point is shown in time course traces (left) and the quantification of the PI + cells at 12 hours after PI dye addition is shown in bar graph (right). All points were normalized to percentage of PI + cells of the no AAV treated microglia at 12 hr. n = 4, from two different litters of mice. Acknowledgements