We examined post-mortem frozen brain samples and matching blood from 45 patients with intermediate-onset sporadic AD and 44 control individuals who died of other causes, including 27 donors age and sex-matched donors with the AD cohort (Figure 1A; Figure 1—figure supplement 1A and Supplementary file 1). APOE risk allele frequency for patients and controls was comparable to published series (Murrell et al., 2006; Sando et al., 2008; Lumsden et al., 2020; Figure 1—figure supplement 1A), and analysis of germline mutations did not identify deleterious variants in the 140 genes associated with neurological diseases. Myeloid/microglia, neurons, and glia/stromal cells were purified by flow cytometry using antibodies against PU.1 and NeuN (Evrony et al., 2012; Figure 1B, Figure 1—figure supplement 1B and C). Single nuclei (sn)RNA-seq was performed on PU.1⁺ nuclei from one control and three AD patients to evaluate microglia enrichment following PU.1⁺ purification, and a cell-type annotation analysis indicated that ~94% of PU.1⁺ nuclei correspond to microglia (Figure 1C and Figure 1—figure supplement 1D–1H). Cortex samples were obtained from all donors but hippocampus samples were mostly obtained from AD patients (Figure 1A; Supplementary file 1). A total of 744 DNA samples from blood, PU.1⁺ nuclei, NeuN⁺ nuclei, and Double Negative nuclei (glia/stromal cells) from patients and controls (Figure 1A) were submitted to targeted hybridization/capture and deep-DNA targeted sequencing (TDS, Figure 1D, see Materials and methods), at mean coverage of ~1100 x (Figure 1—figure supplement 1I), for a panel of 716 genes (3.43 Mb, referred to below as BRAIN-PACT) which included genes reported to carry somatic variants in clonal proliferative disorders (n=576 genes) Cheng et al., 2015; Durham et al., 2019 or that have been reported to be associated with neurodegenerativediseases (n=140) (Bras et al., 2012; Renton et al., 2014; Karch et al., 2014; Karch and Goate, 2015; Turner et al., 2013; Ferrari et al., 2015; Kouri et al., 2015; Scholz and Bras, 2015; Nalls et al., 2014; Supplementary file 2, see Materials and methods).

After QC and filtering of germline variants, variant calling using ShearwaterML, and a curated Mutect1 analysis identified 826 somatic synonymous and non-synonymous single-nucleotide-variations (SNVs), at an allelic frequency >0.3% (mean 1.3%) in the 744 samples, corresponding to an overall variant burden of 0.3 mut/Mb (Figure 1E). Sixty-six/826 SNV were present in more than one sample (Supplementary file 3). Droplet digital-PCR performed on pre-amplification DNA for ~10% of the 760 unique SNV was positive in 90% of cases (Figure 1E; Supplementary file 3). After annotation using the OncoKB Chakravarty et al., 2017 and ClinVar (Landrum et al., 2014) databases for disease-associated or causative variants (Figure 2D and F; Supplementary file 3), 96 unique SNV were classified as Pathogenic (P)-SNV. 40% of these P-SNV were tested by droplet digital-PCR and confirmed in 95% of cases (Figure 1E and Supplementary file 3). Positive and negative results in matching brain samples from individual donors were confirmed in 100% of samples at a mean depth of ~5000 x (range 648–23.000 x) (Supplementary file 3). A venn-diagram analysis of SNVs detected in PU.1⁺, NeuN⁺, DN, and blood samples indicated that most (>90%) SNV and P-SNV were cell-type or tissue-specific, with ~5% of SNV and ~8% of P-SNV shared between the blood and brain of individual donors (Figure 1F; Supplementary file 3). These data indicate that targeted deep-sequencing of purified nuclei allows to detect of clonal mosaic variants with high sensitivity and specificity. In addition, ‘bar-coding’ of clonal variants across tissues suggests that infiltrating myeloid cells of peripheral origin account for ~5% of microglia somatic diversity, and therefore that blood clones have a detectable but minor contribution to microglia, consistent with its local maintenance and proliferation (Askew et al., 2017; Réu et al., 2017).

Somatic clonal diversity of the different cell types, as evaluated by the SNV/megabase burden was higher in blood (1 mut/Mb) and PU.1⁺ nuclei (0.5 mut/Mb) than for DN and neurons (0.18 mut/Mb) (Figure 2—figure supplement 1A). The SNV/mb burden of blood and PU.1⁺ nuclei increased as a function of age (Figure 2A and Supplementary file 3) as previously reported for proliferating cells (Martincorena et al., 2015; Martincorena et al., 2018; Jaiswal et al., 2014; Genovese et al., 2014). Interestingly, the SNV/mb burden of blood cells from age-matched controls was higher than for AD patients (Figure 2B). In contrast, there was no difference in SNV/mb burden between PU.1⁺, NEUN⁺, and DN samples from AD patients and age-matched controls (Figure 2B), and between PU.1⁺ nuclei from the cortex, hippocampus, and brainstem/cerebellum samples (Figure 2C). These data altogether indicate that the clonal diversity of microglia and blood both increase with age, and that the clonal diversity of blood cells is lower in AD than in age-matched controls who died of other causes including cancer and cardiovascular diseases (see Materials and methods). This is consistent with recent studies showing that clonal hematopoiesis is associated with a higher risk of several diseases related to ageing such as cardiovascular diseases, but is inversely associated with the risk of AD (Jaiswal and Ebert, 2019; Bouzid et al., 2023).

In contrast to the global SNV burden, increased P-SNV burden was correlated not only with age (Figure 2D), but also with the disease status (AD) (Figure 2E–H). Within the control group, the SNV and P-SNV burden was higher in the blood of controls treated for cancer (Figure 2—figure supplement 1B). The P-SNV burden per Mb was selectively and highly enriched in PU.1⁺ samples from AD patients in comparison to age-matched controls (p=0.0003, Figure 2E). Analysis of PU1⁺ P-SNV/Mb burden per brains region indicated that the P-SNV/Mb burden was similar between brain regions within each group (Figure 2F and Figure 2—figure supplement 1C), and therefore attributable to AD status rather than sampling bias. Analysis of mutational load per donor confirmed that microglial clones carrying P-SNV were enriched in the brain of AD patients in comparison to age-matched controls (Figure 2G). Despite the relatively modest cohort size, a logistic regression analysis confirmed the association between the presence of P-SNVs in PU.1⁺ nuclei and AD after adjusting for sex and age (OR = 7; p=0.0035, Figure 2G and Figure 2—figure supplement 1D). A mixed-effects linear regression model analysis also showed an excess of P-SNVs in AD independently of the effect of age (p=0.0215) (Figure 2H and Figure 2—figure supplement 1E).

In addition, genes targeted by P-SNV were all expressed in microglia (Figure 2—figure supplement 1F and Supplementary file 3) and the analysis of P-SNV/Mb mutational load restricted to genes that are not expressed in microglia did not show an enrichment of candidate pathogenic variants in AD patients (Figure 2—figure supplement 1G and Supplementary file 3). Altogether, these results show an association between microglia clones carrying P-SNV and AD in this series.

Pathways analysis of genes carrying P-SNV in microglia from AD patients, against the background of the 716 genes sequenced, showed that the most significant pathways enriched were the receptor tyrosine kinase/MAP-Kinase pathways (Reactome, GO, and canonical pathways, Figure 3A and Supplementary file 4), corresponding to pathogenic/oncogenic variants in 6 of the 15 genes of the classical MAPK pathway (Rauen, 2013) (CBL, BRAF, RIT1, NF1, PTPN11, KRAS), TEK, and the KEGG Chronic Myeloid Leukemia (CML) pathway, which includes the former plus SMAD5 and TP53 (Figure 3B, C, Figure 2—figure supplement 2). Mutational load for MAPK genes was significantly higher in AD patients in comparison to age-matched control (Figure 3C). Other enriched pathways, albeit less significant, included genes involved in DNA repair and chromatin binding/methyltransferase activity (Figure 3B; Supplementary file 4). No pathway was enriched in age-matched controls. Of note, we did not observe microglia P-SNVs within genes reported to be associated with neurological disorders (Supplementary file 2) in patients (Supplementary file 3). P-SNV targeting genes of the classical RTK/MAPK pathway (Figure 3C) were detected in the PU.1⁺ samples from ~25% of the AD patients tested (p=0.0145 vs age-matched controls, Figure 3D, Figure 2—figure supplement 2). Strikingly, half of these patients (six patients, 13% of AD patients in this series) carried recurrent P-SNV in the RING domain of CBL (Schnittger et al., 2012; Fernandes et al., 2010; Sargin et al., 2007; Dunbar et al., 2008; Bernard et al., 2014; Loh et al., 2009; Grand et al., 2009; Klampfl et al., 2013; Niemeyer et al., 2010; Javadi et al., 2013; Ogawa, 2019; Figure 3B–E). Two additional patients presented with P-SNV in the Switch II domain of RIT1 (Gómez-Seguí et al., 2013; Figure 3B–F). Microglia from the three other patients carried activating KRAS (p.A59G), PTPN11 (p.T73I), and TEK (p.R1099*) oncogenic variants previously described in cancer and sporadic venous malformations (Kim et al., 2016; Niihori et al., 2005; Soblet et al., 2013; Figure 3B and D). In addition, a 12^th patient carried a gain of function (GOF) U2AF1 (p.S34F) variant (Okeyo-Owuor et al., 2015), which is not a ‘classical MAPK gene’ but activates the MAPK pathway in myeloid malignancies (Smith et al., 2019; Figure 3B, Figure 2—figure supplement 2). Two patients carried two different MAPK activating variants: microglia from one patient carried an activating BRAF (p.L505H) variant Choi et al., 2014 in addition to loss of function (LOF) variant CBL (p.C416S), and another patient carried the NF1 (p.L2442*) LOF variant Heim et al., 1995; Bollag et al., 1996 in addition to the activating RIT1 (p.M90I) variant (Figure 3D, Figure 2—figure supplement 2). Five patients also carried additional P-SNV targeting genes involved in DNA repair with tumor suppressor function Song et al., 2019; Zannini et al., 2014, including the loss of function variants in ATR (c.6318A>G) Fang et al., 2004 and SMC1A (p.X285_splice) (Figure 3D, Figure 2—figure supplement 2), and in DNA/histone methylation including TET2 (p.Q1627*) (Schnittger et al., 2012; Haferlach et al., 2014), IDH2 (p.R140Q) (Ward et al., 2010), and PBRM1 (c.996–7T>A) (Brownlee et al., 2014; Figure 3D, Figure 2—figure supplement 2). Finally, two patients carried oncogenic variants in genes from the KEGG Chronic Myeloid Leukemia (CML) pathway, SMAD3 (p.R373C) Ku et al., 2007, and TP53 (pX261_splice) Bougeard et al., 2015 (Figure 3D and Figure 2—figure supplement 2). The detection of multiple oncogenic variants in the same patients is reminiscent of the features observed in myeloproliferative disorders described outside the brain (Ogawa, 2019; Haferlach et al., 2014).

CBL is an E3 ubiquitin-protein ligase that negatively regulates RTK signaling via MAPK (Liyasova et al., 2015). CBL somatic and germ-line LOF variants such as R420Q have been previously associated with tumoral diseases including clonal myeloproliferative disorders Schnittger et al., 2012; Fernandes et al., 2010; Sargin et al., 2007; Dunbar et al., 2008; Bernard et al., 2014; Loh et al., 2009; Grand et al., 2009; Klampfl et al., 2013; Niemeyer et al., 2010; Javadi et al., 2013; Ogawa, 2019 and RASopathies (Brand et al., 2014), respectively. We confirmed that CBL RING-domain variants found in AD patients increased MAPK phosphorylation in response to EGF upon expression of HA-tagged WT or mutant alleles in HEK293T cells (Figure 3E). RIT1 is a RAS GTPase, and somatic or germ-line GOF variants such as RIT1 F82L and RIT1 M90I, also enhance MAPK signaling in malignancies Gómez-Seguí et al., 2013 and RASopathies (Meyer Zum Büschenfelde et al., 2018; Aoki et al., 2013), respectively. As in the case of CBL variants, the 2 RIT1 variants found in AD patients increased MAPK phosphorylation in response to FBS in HEK293T cells expressing these mutant alleles (Figure 3F). These data altogether indicate that a subset of AD patients (12/45, ~27% of this series) present with microglial clones carrying one or several oncogenic variants that activate the RTK/MAPK pathway, and are characterized by recurrent oncogenic variants in CBL and RIT1.

The allelic frequencies at which MAPK activating variants are detected in brain samples from AD patients range from ~1–6% in microglia (Figure 3G), which correspond to mutant clones representing 2 to 12% of all microglia in these samples, assuming heterozygosity. This range of allelic frequency is frequently observed for the MAPK-activating BRAF^V600E variant in microglia isolated from brain samples of six patients diagnosed with BRAF^V600E+ histiocytosis, a rare clonal myeloid disorder associated with neurodegeneration (Mass et al., 2017; Boyd et al., 2020; Bhatia et al., 2020; Diamond et al., 2016; Héritier et al., 2018; Figure 3G and Supplementary file 5). These data suggested that the size of the mutant microglial clones in AD patients was compatible with a role in a neuro-inflammatory/neurodegeneration process.

Pathogenic variants that did not involve the MAPK pathway included LOF variants in the DNA repair gene CHEK2 including CHEK2 c.319+1 G>A Cybulski et al., 2011 and CHEK2 R346H (Figure 2—figure supplement 2 and Figure 3—figure supplement 1A), Mediator Complex gene MED12 (Graham and Schwartz, 2013), Histone methyltransferases SETD2 Yang et al., 2016 and KMT2C/MLL3, the DNA methyltransferase DNMT3A (Haferlach et al., 2014; Walter et al., 2011), DNA demethylating enzymes TET2 and the Polycomb proteins ASXL1 (Haferlach et al., 2014). Of note, TET2, DNMT3, and KMT2C variants when present, were frequently detectable in the patients’ matching blood at low allelic frequency (Figure 2—figure supplement 2). TET2, DNMT3, and KMT2C are frequently mutated in clonal hematopoiesis (Jaiswal et al., 2014; Genovese et al., 2014), suggesting that in contrast to other variants, the presence of TET2, DNMT3, and KMT2C/MLL3 in the brain of patients may reflect the entry of blood clones in the brain.

In half of the AD patients, no microglia pathogenic variants were identified. Targeted deep sequencing (TDS) cannot identify variants located outside of the BRAIN-PACT panel, such as other potential additional variants that would activate the MAPK pathway. Therefore, we performed whole exome sequencing (WES) of PU.1⁺ nuclei at an average depth ~400 x, in selected samples from 48 donors, including samples from most of the patients negative for pathogenic variants by TDS (n=17 out of 22), a selection of patients with variants identified by TDS (n=16 out of 23), and 15 controls, followed by a curated Mutect analysis. Only 6/15 (40%) of the pathogenic SNVs previously identified by TDS and confirmed by ddPCR were detectable by WES in these samples (Figure 3H), indicating a lower sensitivity of WES. Nevertheless, after annotation by four modeling predictors Polyphen, SIFT, CADD/MSC, and FATHMM-XF (Xi et al., 2004; Shihab et al., 2015; Adzhubei et al., 2010; Ng and Henikoff, 2001; Kircher et al., 2014; Itan et al., 2016) additional SNVs predicted to be deleterious with high confidence were identified in 8/22 patients without pathogenic variants identified by TDS (Figure 2—figure supplement 2 and Supplementary file 6). Interestingly, four of the predicted deleterious variants identified by WES targeted genes that regulate the MAPK pathway ARHGAP9, ARHGEF26, CHD8, and DIXDC1 (Figure 2—figure supplement 2 and Supplementary file 6).

CBL variants increased ERK phosphorylation upon lentiviral transduction in BV2 murine microglial cells (Blasi et al., 1990; Henn et al., 2009; Figure 3—figure supplement 1B). However, as this line was immortalized by v-Raf, which might interfere with the study of the MAPK pathway, we also stably expressed WT and variant CBL, RIT1, KRAS, PTPN11 alleles in SV-U19–5 transformed mouse ‘MAC’ lines (Yu et al., 2008; Xiong et al., 2011) (see Methods and Figure 4—figure supplement 1A and B). MAC lines expressing CBL, RIT1, KRAS, and PTPN11 variants presented with increased ERK phosphorylation and/or increontrols, as measured by Western immunoblotting and EdU incorporation (Figure 4A and Figure 4—figure supplement 1A and B). In addition, Hallmark and KEGG pathway analysis of RNAseq data from control and mutant lines showed increased RAS, TNF, IL6, and JAK STAT signaling, complement, inflammatory responses, and mTOR pathway activation signatures in mutants (Figure 4B and Supplementary file 7). These data indicated that microglia variants from patient’s activate murine microglial cells and growth factor-dependent macrophages with proliferative and inflammatory responses in vitro. However, overexpression of mutant alleles in mouse cell lines does not necessarily recapitulate or predict the effects of a heterozygous genetic variant in physiological conditions. Thus, we investigated the role of CBL^C404Y allele in heterozygous human primary microglia-like cells.

We used prime editing Anzalone et al., 2019 of human induced pluripotent stem cells (hiPSCs, see Methods) to generate isogenic hiPSCs clones heterozygous for the patients’ variants (Figure 4C and Figure 4—figure supplement 1). We focused our analysis on CBL^404C/Y mutant lines because CBL was mutated in six patients and two of them carried the same CBL c.1211G>A p.C404Y variant (Figure 2—figure supplement 2). Microglia-like cells were differentiated from two independent hiPSC-derived CBL^404C/Y lines and their isogenic CBL^404C/C controls (Figure 4C and Figure 4—figure supplement 1C). CBL^404C/Y and isogenic CBL^404C/C microglia-like cells expressed similar amount of CBL total mRNA and protein, and CBL^404C/Y cells expressed wt and mutant mRNA in similar amounts, as expected assuming bi-allelic expression of CBL (Figure 4—figure supplement 1D-5F). CBL^404C/Y cells presented with a phenotype comparable to isogenic CBL^404C/C microglia-like cells for expression of IBA1, CSF1R, NGFR, EGFR, CD11b, MRC1, CD36, CD11c, Tim4, CD45, and MHC Class II (Figure 4—figure supplement 1G). Their viability was also comparable to control (Supplementary file 5H). However, CBL^404C/Y cells were larger and presented with more lamellipodia, resulting in an amoeboid morphology less frequently observed in isogenic controls (Figure 4D and E), and their proliferation rate was slightly increased, as measured by EdU incorporation (Figure 4F). Moreover CBL^404C/Y cells cultured in CSF1-supplemented medium also presented with a higher basal pERK level than control when restimulated with CSF1 (Figure 4—figure supplement 1I), and ERK phosphorylation after stimulation of starved microglia-like cells with CSF-1 was increased by ~ twofold in comparison to isogenic WT (Figure 4G). Altogether, these results showed that heterozygosity for a CBL^C404Y allele is sufficient to activate human microglia-like cells increasing their proliferation and ERK activation.

Gene Set Enrichment Analyses (GSEA) of RNAseq comparing CBL^404C/Y and isogenic CBL^404C/C microglia-like cells showed upregulation of Glycolysis, Oxidative Phosphorylation, and mTORC1 signatures, indicating increased metabolism and energy consumption by the mutant cells (Figure 5A and Supplementary file 8). In addition, as observed in MAC lines, CBL^404C/Y cells upregulated complement, TNF, and JAK STAT signaling and inflammatory signatures (Figure 5A; Supplementary file 8; Ghosh et al., 2018). Increased production of TNF, IL-6, IFN-γ, IL-1β, C3, and complement Factor H (CFH) by CBL^404C/Y cells was confirmed by ELISA (Figure 5B). In addition, CBL^404C/Y microglia-like cells also presented with signatures from the KEGG database associated with neurodegenerative disorders (Figure 5A; Supplementary file 8), and for the recently published human microglia AD scRNA-seq signature, obtained by analysis of 24 sporadic AD patients and 24 controls (Mathys et al., 2019; Figure 5C). These data indicated that heterozygosity for the CBL^C404Y allele is sufficient to drive expression of a neuroinflammatory/AD signature in a human microglia-like cell type, characterized by increased metabolism and the production of neurotoxic cytokines known to interfere with normal brain homeostasis.

Analysis of the snRNA-seq data from five samples of purified microglia nuclei from four donors control, AD without and with pathogenic variants (Figure 1C, Figure 1—figure supplement 1D–H and Supplementary file 9) using unsupervised Louvain clustering and GSEA showed that microglia samples from patients carrying variants were enriched for the signatures observed in the MAC lines and CBL^404C/Y cells (Figure 5D and E and Figure 5—figure supplement 1). In particular, microglia cluster 2 and 2B, were most enriched for the inflammatory, TNF, mTOR, and oxidative phosphorylation and glycolysis signatures in patients carrying variants (AD52, AD53) but not the controls (C11, AD34) (Figure 5D and E and Figure 5—figure supplement 1). Despite the small size of the mutant clones and the low sensitivity of scRNA-seq to detect rare allelic variants, KRAS A59G variant reads were detected in cluster 2/2B from patient AD52.

Altogether, the above results support the hypothesis that patients’ microglial clones carrying pathogenic mutations are associated with a metabolic and neuroinflammatory signature that includes the production of neurotoxic cytokines in vitro and in vivo.

The study was conducted according to the Declaration of Helsinki. Human tissues were obtained with patient-informed consent and used under approval by the Institutional Review Boards from Memorial Sloan Kettering Cancer Center (IRB protocols #X19-027). Snap-frozen human brain and matched blood were provided by the Netherlands Brain Bank (NBB), the Human Brain Collection Core (HBCC, NIH), Hospital Sant Joan de Déu, and the Rapid Autopsy Program (MSKCC, IRB #15–021). Samples were neuropathologically evaluated and classified by the collaborating institutions as AD (Dubois et al., 2007; Braak and Braak, 1991; Braak and Braak, 1995; McKhann et al., 1984; McKhann et al., 2011) or non-dementia controls. The mean age of AD patients is 65 y old (55.5% female, 44.5% male). The mean age of all controls is 54 y old (60% female, 40% male), and the mean age of AD age-matched controls was 70 y old (60% female, 40% male). The overall mean of the post-mortem delay interval was 9.8 hr. Patients did not present with germline pathogenic PSEN1/2/3 or APP AD’s associated variants. For additional information on donor’s brain regions, sex, age, cause of death, Apoe status, Braak status see Supplementary file 1. To avoid possible contamination of sequencing data with mutations associated with donor’s tumoral disease in the group of non-dementia controls, we refrained from selecting cases with blood malignancies or with brain tumors. Samples from histiocytosis patients were collected under GENE HISTIO study (approved by CNIL and CPP Ile-de France) from Pitié-Salpêtrière Hospital and Hospital Trousseau and from Memorial Sloan Kettering Cancer Center.

All samples were handled and processed under Air Clean PCR Workstation. An average of 400 mg of frozen brain tissues were homogenized with a sterile Dounce tissue grinder using a sterile non-ionic surfactant-based buffer to isolate cell nuclei (‘homogenization buffer:’ 250 mM Sucrose, 25 mM KCL, 5 mM MgCl2, 10 mM Tris buffer pH 8.0, 0.1% (v/v) Triton X-100, 3 μM DAPI, Nuclease Free Water). Homogenate was filtered in a 40 μm cell strainer and centrifuged 800 g 8 min 4 °C. To clean-up the homogenate, we performed a iodixanol density gradient centrifugation as follows: pellet was gently mixed 1:1 with iodixanol medium at 50% (50% Iodixanol, 250 mM Sucrose, 150 mM KCL, 30 mM MgCl2, 60 mM Tris buffer pH 8.0, Nuclease Free Water) and homogenization buffer. This solution layered to a new tube containing equal volume of iodixanol medium at 29% and centrifuged 13.500 g for 20 min at 4 °C. Nuclei pellet was gently resuspended in 200 μl of FACS buffer (0.5% BSA, 2 mM EDTA) and incubated on ice for 10 min. After centrifugation 800 g 5 min 4 °C, sample was incubated with anti-NeuN (neuronal marker, 1:500, Anti-NeuN-PE, clone A60 Milli-Mark) for 40 min. After centrifugation 800 g 5 min 4 °C, sample was washed with 1 X Permeabilization buffer (Foxp3 /Transcription Factor Staining Buffer Set, eBioscience) and centrifuged 1300 g for 5 min, without breaks to improve nuclei recovery. Staining with anti-Pu.1 antibody in 1 X Permeabilization buffer (myeloid marker 1:50, Pu.1-AlexaFluor 647, 9G7 Cell Signaling) was performed for 40 min. After a wash with FACS buffer samples were prepared for FACS. Nuclei were FACS-sorted in a BD FACS Aria with a 100 μm nozzle and a sheath pressure 20 psi, operating at ~1000 events per second. Nuclei were sorted into 1.5 ml certified RNAse, DNAse DNA, ATP, and Endotoxins tubes containing 100 μl of sterile PBS. For detailes on sorted samples see Supplementary file 1. Sorting purity was >95%. Sorting strategy is depicted in Figure 1—figure supplement 1. Of note, the Double-negative gate is restricted to prevent cross-contamination between cell types. Nuclei suspensions were centrifuged 20 min at 6000 g and processed immediately for gDNA extraction with QIAamp DNA Micro Kit (Qiagen) following manufacture instructions. DNA from whole-blood samples was extracted with QIAamp DNA Micro Kit (Qiagen) following manufacture instructions. Flow cytometry data was collected using DiVa 8.0.1 Software. Subsequent analysis was performed with FlowJo_10.6.2. For sorting strategy, see Figure 1—figure supplement 1.

DNA samples were submitted to the Integrated Genomics Operation (IGO) at MSKCC for quality and quantity analysis, library preparation and sequencing. DNA quality mas measured with Tapestation 2200. All samples had a DNA Integrity Number (DIN) >6. After PicoGreen quantification,~200 ng of genomic DNA were used for library construction using the KAPA Hyper Prep Kit (Kapa Biosystems KK8504) with eight cycles of PCR. After sample barcoding, 2.5 ng-1µg of each library were pooled and captured by hybridization with baits specific to either the HEME-PACT (Integrated Mutation Profiling of Actionable Cancer Targets related to Hematological Malignancies) assay, designed to capture all protein-coding exons and select introns of 576 (2.88Mb) commonly implicated oncogenes, tumor suppressor genes Cheng et al., 2015 and/or HEME/BRAIN-PACT (716 genes, 3.44 Mb, Supplementary file 2) an expanded panel that included additional custom targets related to neurological diseases including, Alzheimer’s Disease, Parkinson’s Disease, Amyotrophic Lateral Sclerosis (ALS), and others (Supplementary file 1; Bras et al., 2012; Renton et al., 2014; Karch et al., 2014; Karch and Goate, 2015; Turner et al., 2013; Ferrari et al., 2015; Kouri et al., 2015; Scholz and Bras, 2015; Nalls et al., 2014). To simplify, in the manuscript, the combined panel is referred to as ‘BRAIN-PACT.’ In Supplementary file 3, ‘Heme-only’ or ‘Brain-only’ is indicated in the cases for which only one or the other panels were used. Capture pools were sequenced on the HiSeq 4000, using the HiSeq 3000/4000 SBS Kit (Illumina) for PE100 reads. Samples were sequenced to a mean depth of coverage of 1106 x (Control samples: 1071 x, AD samples 1100 x). For detailed information on the sample quality control checks used to avoid potential sample and/or barcode mix-ups and contamination from external DNA, see Cheng et al., 2015.

The data processing pipeline for detecting variants in Illumina HiSeq data is as follows. First, the FASTQ files are processed to remove any adapter sequences at the end of the reads using cutadapt (v1.6). The files are then mapped using the BWA mapper (bwa mem v0.7.12). After mapping the SAM files are sorted and read group tags are added using the PICARD tools. After sorting in coordinate order the BAM’s are processed with PICARD MarkDuplicates. The marked BAM files are then processed using the GATK toolkit (v 3.2) according to best practices for tumor normal pairs. They are first realigned using ABRA (v 0.92) and then the base quality values are recalibrated with the BaseQRecalibrator. Somatic variants are then called in the processed BAMs using MuTect (v1.1.7) for SNV and ShearwaterML (Martincorena et al., 2015; Martincorena and Campbell, 2015; Martincorena et al., 2018). muTect (v1.1.7): to identify somatic variants and eliminate germline variants, we run the pipeline as follows: PU.1, DN and Blood samples against matching-NeuN samples, and NeuN samples against matching-PU.1. In addition, we ran all samples against a Frozen-Pool of 10 random genomes. We selected Single Nucleotide Variations (SNVs) [Missense, Nonsense, Splice Site, Splice Regions] that were supported by at least four or more mutant reads and with coverage of 50 x or more. Fill-out file for each project (~27 samples per sequencing pool), were used to exclude by manual curation, variants with high background noise. This resulted in 428 variants (Missense, Nonsense, Splice_site, Splice_Region).

ShearwaterML, was used to look for low allelic frequency somatic mutations as it has been shown to efficiently call variants present in a small fraction of cells with true positives being ~90%. Briefly, the basis of this algorithm is that is uses a collection of deep-sequenced samples to learn for each site a base-specific error model, by fitting a beta-binomial distribution to each site combining the error rates across all normal samples both the mean error rate at the site and the variation across samples, and comparing the observed variant rate in the sample of interest against this background model using a likelihood-ratio test. For detailed description of this algorithm please refer to Martincorena et al., 2015; Martincorena et al., 2018. In our data set, for each cell type (NeuN, DN, PU.1) we used as ‘normal’ a combination of the other cell types, i.e., PU.1 vs NeuN +DN, DN vs NeuN +PU.1, NEUN vs PU.1+DN, Blood vs NeuN +DN. Since all samples were processed and sequenced using the same protocol, we expect the background error to be even across samples. More than 400 samples were used as background leading to an average background coverage >400.000 x. Resulting variants for each cell type were filtered out as germline if they were present in more than 20% of all reads across samples. Additionally, variants with coverage of less than 50 x and more than 35% variant allelic frequency (VAF) were removed from downstream analysis. p-values were corrected for multiple testing using Benjamini & Hochberg’s False Discovery Rate (FDR) Reiner et al., 2003 and a q-value of cutoff of 0.01 was used to call somatic variants. Variants were required to have a least one supporting read in each strand. Somatic variants within 10 bp of an indel were filtered out as they typically reflect mapping errors. We selected Single Nucleotide Variations (SNVs) [Intronic, Intergenic, Missense, Nonsense, Splice Site, Splice Regions] that were supported by at least 4 or more mutant reads and annotated them using VEP. Finally, to reduce the risk of SNP contamination, we excluded variants with a MAF (minor allelic frequency) cutoff of 0.01 using the gnomeAD database. This resulted in 509 SNVs.

We compared the final mutant calls from Muetct1 and ShearwaterML and found that 30% of the events (111 variants) that were called by MuTect1 were also called by ShearwaterML. Overall a total of 826 variants (Supplementary file 3) were found, with a mean coverage at the mutant site of 668.3 X (10% percentile: 276 X, 90% percentile: 1181 X) and a mean of 29.1 mutant reads (10% percentile: 4, 90% percentile: 52), with 84% of mutated supported by at least five mutant reads (Supplementary file 3). The median allelic frequency was ~1.34% (Supplementary file 3). Negative results for matching brain negative samples were confirmed in 100% of samples at a mean depth of ~5000 x (range 648–23.000 x) (Supplementary file 3), confirming nuclei sorting purity of >95% for PU.1⁺, DN, and NEUN⁺ populations.

We performed validation of ~11% of unique variants (69/760) by droplet-digital PCR (ddPCR) on pre-amplified DNA or on libraries (in the cases where DNA was not sufficient). Around 15% (15/69) of the variants analyzed by ddPCR were called by ShearwaterML, ~44% (34/69) were called by Mutect1 and 40% (24/69) by both ShearwaterML +Mutect1. Altogether we confirmed 62/69 of variants tested (~90%). In addition, 61 assays (from variants detected in PU.1+nuclei) were tested in paired cell types isolated from the same brain region. Assays were also run in matching blood when available. The mean depth of ddPCR was ~5000 x and mutant counts of three or more were considered positive. VAF obtained by ddPCR correlated with original VAF by sequencing (R2 0.93, p<0.0001). For KRAS_G12D: Bio-Rad validated assay (Unique Assay ID: dHsaMDV2510596) and MTOR_Arg1616His_c.4847G>A: Bio-Rad validated assay (Unique Assay ID: dHsaMDV2510596) were used. The remaining assays were designed and ordered through Bio-Rad. For setting-up the right conditions for newly designed assays, cycling conditions were tested to ensure optimal annealing/extension temperature as well as optimal separation of positive from empty droplets. All reactions were performed on a QX200 ddPCR system (Bio-Rad catalog # 1864001). When possible, each sample was evaluated in technical duplicates or quartets. Reactions contained 10 ng gDNA, primers and probes, and digital PCR Supermix for probes (no dUTP). Reactions were partitioned into a median of ~31,000 droplets per well using the QX200 droplet generator. Emulsified PCRs were run on a 96-well thermal cycler using cycling conditions identified during the optimization step (95 °C 10’; 40–50 cycles of 94 °C 30’ and 52–56°C 1’; 98 °C 10’; 4 °C hold). Plates were read and analyzed with the QuantaSoft software to assess the number of droplets positive for mutant DNA, wild-type DNA, both, or neither. ddPCR results are listed in Supplementary file 3.

To classify somatic variants according to their pathogenicity we did as follows: Variants were classified as ‘pathogenic (P-SNV)’ if reported as ‘pathogenic/likely pathogenic’ by ClinVar Landrum et al., 2014 and/or ‘oncogenic/predicted oncogenic/likely oncogenic’ by OncoKb (Chakravarty et al., 2017; Supplementary file 3). These two databases report pathogenicity in cancer and other diseases, based on supporting evidence from curated literature (see corresponding citations in Supplementary file 3). We considered classical-MAPK-pathway genes those reported to be mutated in RASopathies: BRAF, CBL, KRAS, MAP2K1, NF1, PTPN11, SOS1, RIT1, SHOC2, NRAS, RAF1, RASA1, HRAS, MAP2K2, SPRED1 (Rauen, 2013; Tidyman and Rauen, 2016 ; Supplementary file 3).

We defined mutational load or mutational burden as the number of synonymous and non-synonymous somatic single-nucleotide-variant (SNV) per megabase of genome examined (Zehir et al., 2017). Overall, a total of 826 SNVs were detected resulting in 0.3 mutations/Mb sequenced. As detailed in the manuscript, the mutational load varies considerably across cell types and patients. To quantify mutational load we took into consideration the panel used for sequencing each sample: HEME-PACT (2.88 Mb) or the extended panel BRAIN-PACT (3.44 Mb) (see Supplementary file 2). Therefore, the number of mutations was normalized by the number of Mb sequenced for that specific sample. In the cases where we calculated mutational load per patient, we averaged the mutational load of each sample from that patient for a given cell type (i.e. if for one patient, 2 PU.1 samples were sequenced, one from hippocampus and one from superior parietal cortex (with BRAIN-PACT) then the mutational load for PU.1 for that patient is the mean of the mutational load of the 2 PU.1 samples analyzed). For the quantification of ‘pathogenic’ variants, the same analysis is performed, quantifying only variants that are reported as pathogenic by ClinVar and/or OncoKb. Statistical significance was analyzed with GraphPad Prism (v9) and R (3.6.3). Non-parametric tests were used when data did not follow a normal distribution (Normality test: D’Agostino-Pearson and Shapiro-Wilk test). For normally distributed data, unpaired t-test was used to compare two groups and one-way, nested one-way or two-way analyses of variance (ANOVA) were used for comparing more than two groups, as indicated in the Figure legends. For data that did not have a normal distribution, the tests performed were unpaired two-tailed Mann-Whitney U test and Kruskal–Wallis test and Dunn’s test for multiple comparisons. Pearson and Spearman were used for correlation analysis. In Figure 2G, we used multivariate logistic regression analysis to test if there was an association between Alzheimer’s disease and the presence of pathogenic variants in PU.1⁺ nuclei. We used Alzheimer’s disease as a dependent variable, and age, sex, and the presence of pathogenic variant/s (Yes/No) as co-variates. In all the statistical tests, significance was considered at p<0.05. For Venn Diagram plots, we used (Bardou et al., 2014).

To evaluate the correlation between P-SNV burden and disease status (non-dementia controls and AD) after adjusting for other factors such as individual donor and age, we performed linear mixed-effects regression modeling using the nlme package in R. In the subsequent analysis, we also tested another framework implemented in the lme4 package and confirmed similar findings. We estimated the linear mixed model via maximum likelihood method with nlminb optimizer. The model included individual donor as a random effect. We also tested if inclusion of other co-variates (i.e. sex, anatomical location of the brain, and source biobank of brain samples) as random effects inproved the overall model fitting via likelihood ratio test. However, none of these co-variates improved the model fitting (p>0.99). Thus, we used a relatively simple model that incorporated disease status and age as fixed effects and donor as random effect. The total explanatory power of the final model is substantial (conditional R^2=0.48). To assess the significance of age or disease status in predicting P-SNV burden, we constructed another model that does not incorporate the variable as fixed effect, and compared the two models via likelihood ratio test. The variable was considered significantly associated with P-SNV burden when the p-value was below 0.05 and the AIC increased after removing the variable.

Pathway enrichment analysis of genes target of variants was performed using Metascape Zhou et al., 2019 and the following ontology sources: KEGG Pathway (Huang et al., 2009a; Huang et al., 2009b), GO Molecular function (Ashburner et al., 2000; The Gene Ontology, 2019), Reactome Gene Sets Ashburner et al., 2000; The Gene Ontology, 2019, and Canonical Pathways (Schaefer et al., 2009). The list of 716 genes from the targeted panel were used as the enrichment background. Terms with a p-value <0.05, a minimum count of 3, and an enrichment factor >1.5 (the enrichment factor is the ratio between the observed counts and the counts expected by chance) are shown. p-values are calculated based on the cumulative hypergeometric distribution Zar, 2010.

To evaluate the expression levels of the genes identified in this study as target of somatic variants, we consulted a publicly available database (https://www.proteinatlas.org/), and also plotted their expression as determined by RNAseq in two studies (Galatro et al. GSE99074, Galatro et al., 2017, and Gosselin et al., 2017) (Supplementary Fig S3 and Figure S2). For data from Galatro et al. (GSE99074, Galatro et al., 2017), normalized gene expression data and associated clinical information of isolated human microglia (N=39) and whole brain (N=16) from healthy controls were downloaded from GEO. For data from Gosselin et al., 2017 , raw gene expression data and associated clinical information of isolated microglia (N=3) and whole brain (N=1) from healthy controls were extracted from the original dataset. Raw counts were normalized using the DESeq2 package in R (Love et al., 2014).

For snRNA-seq studies, we only selected samples with a RIN score in whole tissue of six or more. All samples were handled and processed under Air Clean PCR Workstation. About 250–400 mg of frozen brain tissues were homogenized with a sterile Dounce tissue grinder using a sterile homogenization buffer to isolate cell nuclei (250 mM Sucrose, 25 mM KCL, 5 mM MgCl2, 10 mM Tris buffer pH 8.0, 0.1% (v/v) Triton X-100, 3 μM DAPI, Nuclease Free Water and 20 U/ml of Superase-In RNase inhibitor, and 40 U/ml RNasin ribonuclease inhibitor). Homogenate was filtered in a 40 μm cell strainer and centrifuged 800 g 8 min 4 °C. To clean-up the homogenate, we performed a iodixanol density gradient centrifugation as follows: pellet was gently mixed 1:1 with iodixanol medium at 50% (50% Iodixanol, 250 mM Sucrose, 150 mM KCL, 30 mM MgCl2, 60 mM Tris buffer pH 8.0, Nuclease Free Water) and homogenization buffer. This solution layered to a new tube containing equal volume of iodixanol medium at 29% and centrifuged 13.500 g for 20 min at 4 °C. Nuclei pellet was resuspended in FACS buffer with RNAse inhibitors (0.5% BSA, 2 mM EDTA, Superase-In RNase inhibitor and 40 U/ml RNasin ribonuclease inhibitor) and centrifuged 800 g 5 min, 4 °C. Nuclei pellet was fixed with 90% ice-cold methanol and incubated for 10 min on ice, followed by a centrifugation at 1300 g (without brakes, which improves with nuclei recovery after fixation). The pellet was resuspended in permeabilization buffer (6% BSA, Superase-In RNase inhibitor 20 U/mL, RNasin ribonuclease inhibitor 40 U/mL and 0.05% Triton) followed by a centrifugation at 1300 g. Sample was incubated with anti-Pu.1 antibody (microglia marker 1:50, Pu.1-AlexaFluor 647, 9G7 Cell Signaling) in permeabilization buffer. After a wash with FACS buffer sample were ready for sorting. Nuclei are FACS-sorted in a BD FACS Aria with a 100 μm nozzle and a sheath pressure 20 psi, operating at ~1000 events per second. Nuclei were sorted into 1.5 ml certified RNAse, DNAse DNA, ATP, and Endotoxins tubes containing 100 μl of sterile PBS. For each population, we sorted >10⁵ nuclei into FACS buffer.

The single-nuclei RNA-seq of FACS-sorted nuclei suspensions was performed on Chromium instrument (10 X genomics) following the user guide manual (Reagent Kit 3’ v3.1). Each sample, containing approximately 10,000 nuclei at a final dilution of ~1000 cells/µl was loaded onto the cartridge following the manual. The individual transcriptomes of encapsulated cells were barcoded during RT step and resulting cDNA purified with DynaBeads followed by amplification per manual guidelines. Next, PCR-amplified product was fragmented, A-tailed, purified with 1.2 X SPRI beads, ligated to the sequencing adapters, and indexed by PCR. The indexed DNA libraries were double-size purified (0.6–0.8 X) with SPRI beads and sequenced on Illumina NovaSeq S4 platform (R1 – 26 cycles, i7 – 8 cycles, R2 – 70 cycles or higher). Sequencing depth was ~200 million reads per sample on average. FASQ files were processed using SEQC pipeline Azizi et al., 2018 for quality control, mapping to GRCH38 reference genome, and log2 transformation of the data with the default SEQC parameters to obtain the gene-cell count matrix.

Seurat v4.0.3 with default parameters was used to perform sctransform (SCT) normalization, integration and Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction. The FindClusters function was used for cell clustering. To improve clustering, all samples were analyzed in an integrated analysis, based on canonical correlation analysis (CCA). Cell types were annotated using the top 500 DEGs of each cell type in a human cortex database. Data can be accessed at https://weillcornellmed.shinyapps.io/Human_brain/. The removal of doublets using DoubletFinder and cells with high mitochondrial content (>10% mitochondrial RNA) yielded between 6437 and 9241 nuclei per patient and sample. Microglia represented 94 ± 3% of total cells. Unique Molecular Identifiers (UMIs) per nucleus and gene count per nucleus were comparable between donors. Integrated_snn at resolution 0.2 outlined 16 microglia clusters. Except for cluster 13 consisting of 97% of cells from the healthy control Control 11_AG, all donors and samples were represented in every cluster. One cluster contained few cells (0.84% of total microglia, for an average of 6.20 ± 1.60% for other clusters) and was marked by a low number of cluster-enriched genes and was excluded from further analyses. For pathway enrichment analysis, genes were pre-ranked using differential expression analysis in SCANPY Wolf et al., 2018 with Wilcoxon rank-sum method. Statistical analysis were performed using the fgseaMultilevel function in fgsea R package Korotkevich et al., 2019 for HALLMARK and KEGG pathways. Gene sets with p-value <0.05 and adjusted p-value <0.25 were selected and visualized using ggpubr and ggplot2 Hadley, 2016 R package. For the variant analysis of AD52_HIP harboring a KRAS^A59G (c.176C>G) clone, Integrative Genomics Viewer (IGV) software was used to display sequencing reads at KRAS c.176C (exon 3; GRCh38 chr12:25,227,348). Cells within each cluster were identified based on the 16-digit barcodes from SEQC-aligned reads. Barcodes were converted to the 10 X Genomics format and used to sample reads from each cluster within the original BAM file. BAM subsets for each cluster were read with IGV and reads with identical UMIs were filtered out to account for amplification bias.

Remaining libraries from a selected group of PU.1 and NEUN samples sequenced with BRAIN-PACT (see above) were sequenced by Whole-Exome-Sequencing (WES). Matching NEUN samples were sequenced to extract the germline variants. Around 100 ng of library were captured by hybridization using the xGen Exome Research Panel v2.0 (IDT) according to the manufacturer’s protocol. PCR amplification of the post-capture libraries was carried out for 12 cycles. Samples were run on a NovaSeq 6000 in a PE100 run, using the NovaSeq 6000 S4 Reagent Kit (200 Cycles) (Illumina). Samples were covered to an average of 419 X. The data processing pipeline for detecting variants in Novaseq data is as follows. First the FASTQ files are processed to remove any adapter sequences at the end of the reads using cutadapt (v1.6). The files are then mapped using the BWA mapper (bwa mem v0.7.12). After mapping the SAM files are sorted and read group tags are added using the PICARD tools. After sorting in coordinate order the BAM’s are processed with PICARD MarkDuplicates. The marked BAM files are then processed using the GATK toolkit (v 3.2) according to the best practices for tumor normal pairs. They are first realigned using ABRA (v 0.92) and then the base quality values are recalibrated with the BaseQRecalibrator. Somatic variants are then called in the processed BAMs using muTect (v1.1.7) for SNV and the Haplotype caller from GATK with a custom post-processing script to call somatic indels. The full pipeline is available here https://github.com/soccin/BIC-variants_pipeline (copy archived at Socci, 2022) and the post processing code is at https://github.com/soccin/Variant-PostProcess, (copy archived at Socci, 2025). We selected Single Nucleotide Variants (SNVs) [Missense, Nonsense, Splice Site, Splice Regions] that were supported by at least 8 or more mutant reads, variant allelic frequency above 5% and with coverage of 50 x. Annotation was performed using VEP. Finally, to reduce the risk of SNP contamination, we excluded variants with a MAF (minor allelic frequency) cutoff of 0.01 using the genomeAD database. Variants were classified as ‘candidate pathogenic’ when SNV is predicted to affect the protein as determined by PolyPhen-2 (possibly and probably damaging) and SIFT (deleterious) and CADD-MSC (high) and FATHMM-XF Functional Analysis through Hidden Markov Models (pathogenic) (Rogers et al., 2018; Supplementary file 5).

For HEK293 cells, lysates were resolved by SDS-PAGE, transferred to PVDF membranes, and probed with the appropriate antibodies. Horseradish peroxidase-conjugated secondary antibodies (GE Healthcare) and Western blotting substrate (Thermo) were used for detection. For anti-Cdc42 immunoprecipitation experiments, cell lysates (1 mg total protein) were incubated overnight with 1 µg of anti-Cdc42 antibody (Santa Cruz) and 25 µL of protein A agarose (Roche) at 4°C. Anti-Flag immunoprecipitations were done with anti-Flag M2 affinity resin (Sigma). The beads were washed three times with lysis buffer, then eluted with SDS-PAGE buffer, and resolved by SDS-PAGE. The proteins were transferred to PVDF membrane for Western blot analysis. Antibodies used are Phospho-p44/42 MAPK (pErk 1/2) (Thr202/Tyr204) is from Cell Signaling #4370, total p44/42 MAPK (Erk1/2) is from Cell Signaling #9102, HA tag from Millipore # 05–904, Flag antibody is from Sigma (#A8592), pCHEK2 (T383) antibody is from Abcam, #ab59408, Cdc42 antibody is from Santa Cruz (#sc87) and Anti-γ-Tubulin antibody (Sigma T6557). For Immunoprecipitation Kinase assay in HEK293T cells, cell lysates (1 mg protein) were incubated overnight with 30 µL of anti-Flag M2 affinity resin on a rotator at 4°C, then washed three times with Tris-buffered saline (TBS). A portion of each sample was eluted with SDS-PAGE sample buffer and analyzed by anti-Flag Western blotting. The remaining sample was used for a radioactive kinase assay. The immunoprecipitated proteins were incubated with 25 µL of reaction buffer (30 mM Tris, pH 7.5, 20 mM MgCl₂, 1 mg/mL BSA, 400 µM ATP), 650 µM CHKtide peptide (KKKVRSGLYRSPSMPENLNRPR, SignalChem), and 50–100 cpm/pmol of [γ Zar, 2010-P] ATP at 30 °C for 15 min. The reactions were quenched using 45 µL of 10% trichloroacetic acid. The samples were centrifuged and 30 µL of the reaction was spotted onto Whatman P81 cellulose phosphate paper. After washing with 0.5% phosphoric acid, incorporation of radioactive phosphate into the peptide was measured by scintillation counting. For MAC lines and hiPSC-derived cells, cell lysates obtained with RIPA buffer +1:1000 Halt Protease and Phosphatase Inhibitor Cocktail (Thermo Fisher Scientific) were sonicated three times for 30 s at 4 °C (Bioruptor, Diagenode). Protein quantification of supernatant was done with Precision Red Advanced Protein Assay (Cytoskeleton). Proteins were boiled for 5 min at 95 °C in NuPAGE LDS sample buffer (Invitrogen) and separated in NuPAGE 4–12% Bis-Tris Protein Gel (Invitrogen) in NuPAGE MES SDS Running Buffer (Invitrogen). Electrophoretic transfer to a nitrocellulose membrane (Thermo Fisher Scientific) was done in NuPAGE Transfer Buffer (Invitrogen). Blocking was performed for 60 min in TBS-T +5% nonfat milk (Cell Signaling) and incubated with primary antibodies at 4°C: rabbit anti-p44/42 MAPK (ERK1/2) (Cell Signaling; 1:1000); rabbit anti-P-p44/42 MAPK (Cell Signaling, 1:1000); rabbit anti-c-CBL (Cell signaling, 1:1000); rabbit anti-RIT1 (Abcam, 1:1000); mouse anti-KRAS (clone 3B10-2F2, Sigma, 1 μg/mL); mouse anti-Actin (clone MAB1501, Sigma, 1:10,000). Primary antibodies were detected using the secondary anti-rabbit IgG HRP-linked (Cell Signaling, 1:1000) or the anti-mouse IgG HRP-linked (Cell Signaling, 1:1000) were used to detect primary antibodies, with SuperSignal West Femto Chemiluminescent Substrate (Thermo Fisher Scientific) using a ChemiDoc MP Imaging System (Bio-Rad). pERK/ERK ratios were measured with ImageJ software.

Genomic DNA was extracted using QIAamp DNA Micro Kit (50) (Qiagen), following the manufacturer’s instructions. Total RNA was extracted using RNeasy Mini kit (Qiagen), following the manufacturer’s instructions. cDNA was generated by reverse transcription using Invitrogen SuperScript IV Reverse Transcriptase (Invitrogen) with oligo(dT) primers. The TaqMan gene expression assays used were c-CBL FAM (Hs01011446_m1), CBLb FAM (Hs00180288_m1) and GAPDH VIC (Hs02786624_g1) (Thermo Fisher Scientific). RT-qPCR was performed using Applied Biosystems TaqMan Fast Advanced Master Mix (Thermo Fisher Scientific) and a QuantStudio 6 Flex Real-Time PCR System (Thermo Fisher Scientific). The results were normalized to GAPDH. For droplet PCR analyses, assays specific for the detection of I383M, C384Y, C404Y and C416S in CBL and F82L and M90I in RIT1, A59G in KRAS, and corresponding WT sequences (listed below) were obtained from Bio-Rad. Cycling conditions were tested to ensure optimal annealing/extension temperature as well as optimal separation of positive from empty droplets. Optimization was done with a known positive control. After PicoGreen quantification, 2.6–9 ng gDNA or cDNA were combined with locus-specific primers, FAM- and HEX-labeled probes, HaeIII, and digital PCR Supermix for probes (no dUTP). All reactions were performed on a QX200 ddPCR system (Bio-Rad catalog # 1864001) and each sample was evaluated in technical duplicates. Reactions were partitioned into a median of ~19,000 droplets per well using the QX200 droplet generator. Emulsified PCRs were run on a 96-well thermal cycler using cycling conditions identified during the optimization step (95 °C 10’; 40 cycles of 94 °C 30’ and 52–55°C 1’; 98 °C 10’; 4 °C hold). Plates were read and analyzed with the QuantaSoft software to assess the number of droplets positive for mutant or wild-type DNA.

Flow cytometry analyses for surface antigens CSF1-R, CD11b, MRC1, α5β3, CD11c, Tim4, HLA-DR, CD45, CD14, NGFR, EGFR, CD36, and SIRPα were performed using PE-conjugated anti-CD115 (CSF1-R) (clone 9-4D2, BD Pharmingen), PE/Cy7-conjugated anti-CD11b (clone ICRF44, Biolegend), Alexa Fluor 488-conjugated anti-CD206 (MRC1) (clone 19.2, Thermo Fisher Scientific), PE-conjugated anti-integrin α5β3 (clone 23C6, R&D systems), PE/Cy5-conjugated anti-CD11c (Clone B-ly6, BD Pharmigen), APC-conjugated anti-Tim4 (Clone 9F4, BioLegend), PE/Cy7-conjugated anti-HLA-DR (clone G46-6, BD Pharmigen), BV650-conjugated anti-CD45 (clone HI30, BD Horizon), APC/Cy7-conjugated anti-CD14 (clone M5E2, Biolegend), PE-conjugated anti-NGFR (clone ME20.4, eBioscience), Alexa Fluor 647-conjugated anti-EGFR (clone EGFR.1, BD Pharmigen), APC/Cy7-conjugated anti-CD36 (clone 5-271, BioLegend), and APC-conjugated anti-CD172a (SIRPα) (Clone: 15 414, Thermo Fisher Scientific) antibodies. Iba1 expression was detected following fixation and permeabilization of macrophages using BD Cytofix/Cytoperm solution (BD Pharmingen). Cells were marked with Zombie Violet Viability (Biolegend). After incubation with FcR Blocking Reagent (Miltenyi Biotec), cells were stained with Alexa Fluor 555-conjugated anti-Iba1 antibody (clone E4O4W, Cell Signaling). Flow cytometry was performed using a BD Biosciences LSR Fortessa flow cytometer with Diva software. Data were analyzed using FlowJo (BD Biosciences LLC).

For hiPSC-derived cells, cell suspension was filtered through a 100 µm nylon mesh (Corning) and marked with Zombie Violet Viability (Biolegend). After incubation with FcR Blocking Reagent (Miltenyi Biotec), surface receptors were labeled with PE/Cy7-conjugated anti-CD11b (clone ICRF44, Biolegend), Alexa Fluor 488-conjugated anti-CD206 (MRC1) (clone 19.2, Thermo Fisher Scientific), BV650-conjugated anti-CD45 (clone HI30, BD Horizon), APC/Cy7-conjugated anti-CD14 (clone M5E2, Biolegend) prior to EdU detection. For proliferation studies in the mouse macrophage cell lines, macrophages were incubated with 10 µM EdU (Thermo Fisher Scientific) for 2 hr at 37 °C and collected by cell scraping and marked with Zombie Violet Viability (Biolegend) prior to EdU detection. EdU detection was performed using the Click-iT Plus EdU Alexa Fluor 647 Flow Cytometry Assay Kit (Thermo Fisher Scientific), following the manufacturer’s instructions. hiPSC-derived macrophages were analyzed using a BD Biosciences Aria III cell sorter and macrophages were identified as CD11b⁺CD45⁺CD14⁺MRC1⁺. The macrophage cell lines were analyzed using a BD Biosciences LSR Fortessa flow cytometer. Data were analyzed using FlowJo 10.6 (BD Biosciences LLC).

Supernatants of iPSC-derived microglia-like cells were analyzed for human inflammatory cytokines IL-6, TNFα, IL-1β, IFNγ and for the complement C3 and complement Factor H by Enzyme-linked immunosorbent assay (ELISA) at Eve Technologies (Calgary, AB).

Three biological replicates were processed for each condition/cell line. In view of RNA sequencing, phase separation in cells lysed in 1 mL TRIzol Reagent (Thermo Fisher Scientific) was induced with 200 µL chloroform and RNA was extracted from the aqueous phase using the miRNeasy Mini Kit (Qiagen) on the QIAcube Connect (Qiagen) according to the manufacturer’s protocol with 350 µL input, or using the MagMAX mirVana Total RNA Isolation Kit (Thermo Fisher Scientific) on the KingFisher Flex Magnetic Particle Processor (Thermo Fisher Scientific) according to the manufacturer’s protocol with 350 µL input. Samples were eluted in 30 µL RNase-free water. After RiboGreen quantification and quality control by Agilent BioAnalyzer, 231–500 ng of total RNA with RIN values of 9.4–10 underwent polyA selection and TruSeq library preparation according to instructions provided by Illumina (TruSeq Stranded mRNA LT Kit, Illumina), with 8 cycles of PCR. Samples were barcoded and run on a NovaSeq 6000 in a PE100 run, using the NovaSeq 6000 S4 Reagent Kit (200 Cycles) (Illumina). An average of 90 million paired reads was generated per sample. Ribosomal reads represented 0–1.6% of the total reads generated and the percent of mRNA bases averaged 79%.

FastQ files of 2×100 bp paired-end reads were quality checked using FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/, 2012). Samples with high quality reads (Phred score ≥ 30) were aligned to the Mus musculus genome (GRCm38.80) for the MAC lines or Homo sapiens (assembly GRCh38.p14) for the IPSCs lines using STAR aligner. For the MAC lines, we computed the expression count matrix from the mapped reads using HTSeq (www-huber.embl.de/users/anders/HTSeq) and one of several possible gene model databases. The raw count matrix generated by HTSeq are then be processed using the R/Bioconductor package DESeq (www-huber.embl.de/users/anders/DESeq) which is used to both normalize the full dataset and analyze differential expression between sample groups. For the ISPCs line dataset, gene quantification was performed using feature counts from the Subread package in R. Gene expression levels were normalized and log2 transformed using the Trimmed Mean of M-values (TMM) method and differential expression analysis was performed using the edgeR package in R. For hiPSC derived cells, GSEA (Hallmark, KEGG, GO, REACTOME) were performed using the fgsea package in R on a pre-ranked list (formula: sign(ogFC) * -log10(PValue)) on all expressed genes in the dataset. For the MAC cell lines dataset, GSEA was performed using gsea4.3.2 for KEGG and HALLMARK canonical pathways in MSigDB v 7.5.1. Significant genesets were selected based on an FDR ≤ 0.25.

For lists of differentially expressed genes, genes were selected with controlled False Positive Rate (B&H method) at 5% (FDR ≤ 0.05). Genes were considered upregulated/downregulated for log2 fold change >1.5 or <−1.5.

Statistical methods are detailed in the corresponding sections above (Quantification of mutational load and statistics, Bulk RNAseq analysis, SnRNA-seq analysis) and in the Fig. legends. p-values of 0.05 and adj. p-values (FDR) of 0.25 are considered significant unless otherwise specified.

All code used in this study has been previously published as referenced in the method section above.