Integrated QTL mapping and CRISPR screening in pooled iPSC-derived microglia reveals genetic drivers of neurodegenerative risk

Integrated QTL mapping and CRISPR screening in pooled iPSC-derived microglia reveals genetic drivers of neurodegenerative risk | bioRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-M677548'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search New Results Integrated QTL mapping and CRISPR screening in pooled iPSC-derived microglia reveals genetic drivers of neurodegenerative risk Marta Perez-Alcantara , Sam Washer , Yixi Chen , Juliette Steer , Daianna Gonzalez-Padilla , Joe McWilliam , David Willé , Nikos Panousis , Peep Kolberg , Elena Navarro Guerrero , Kaur Alasoo , Hazel Hall-Roberts , Julie Williams , Sally A. Cowley , View ORCID Profile Gosia Trynka , Andrew Bassett doi: https://doi.org/10.1101/2025.08.18.670767 Marta Perez-Alcantara 1 Wellcome Sanger Institute, Wellcome Genome Campus , Hinxton, CB10 1SA, UK 2 Open Targets, Wellcome Genome Campus , Hinxton, CB10 1SA, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Sam Washer 1 Wellcome Sanger Institute, Wellcome Genome Campus , Hinxton, CB10 1SA, UK 2 Open Targets, Wellcome Genome Campus , Hinxton, CB10 1SA, UK 3 James and Lillian Martin Centre for Stem Cell Research, Sir William Dunn School of Pathology, University of Oxford , South Parks Road, Oxford, OX1 3RE, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Yixi Chen 1 Wellcome Sanger Institute, Wellcome Genome Campus , Hinxton, CB10 1SA, UK 2 Open Targets, Wellcome Genome Campus , Hinxton, CB10 1SA, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Juliette Steer 1 Wellcome Sanger Institute, Wellcome Genome Campus , Hinxton, CB10 1SA, UK 2 Open Targets, Wellcome Genome Campus , Hinxton, CB10 1SA, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Daianna Gonzalez-Padilla 1 Wellcome Sanger Institute, Wellcome Genome Campus , Hinxton, CB10 1SA, UK 2 Open Targets, Wellcome Genome Campus , Hinxton, CB10 1SA, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Joe McWilliam 1 Wellcome Sanger Institute, Wellcome Genome Campus , Hinxton, CB10 1SA, UK 2 Open Targets, Wellcome Genome Campus , Hinxton, CB10 1SA, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site David Willé 5 GSK Research, GSK Medicines Research Centre , Gunnels Wood Road, Stevenage, SG1 2NY, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Nikos Panousis 5 GSK Research, GSK Medicines Research Centre , Gunnels Wood Road, Stevenage, SG1 2NY, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Peep Kolberg 2 Open Targets, Wellcome Genome Campus , Hinxton, CB10 1SA, UK 6 Institute of Computer Science, University of Tartu , Tartu, 51009, Estonia Find this author on Google Scholar Find this author on PubMed Search for this author on this site Elena Navarro Guerrero 7 Target Discovery Institute, University of Oxford , Oxford, OX3 7FZ, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Kaur Alasoo 2 Open Targets, Wellcome Genome Campus , Hinxton, CB10 1SA, UK 6 Institute of Computer Science, University of Tartu , Tartu, 51009, Estonia Find this author on Google Scholar Find this author on PubMed Search for this author on this site Hazel Hall-Roberts 4 UK Dementia Research Institute at Cardiff University , Maindy Road, Cardiff CF24 4HQ, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Julie Williams 4 UK Dementia Research Institute at Cardiff University , Maindy Road, Cardiff CF24 4HQ, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Sally A. Cowley 3 James and Lillian Martin Centre for Stem Cell Research, Sir William Dunn School of Pathology, University of Oxford , South Parks Road, Oxford, OX1 3RE, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Gosia Trynka 1 Wellcome Sanger Institute, Wellcome Genome Campus , Hinxton, CB10 1SA, UK 2 Open Targets, Wellcome Genome Campus , Hinxton, CB10 1SA, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Gosia Trynka For correspondence: ab42{at}sanger.ac.uk Andrew Bassett 1 Wellcome Sanger Institute, Wellcome Genome Campus , Hinxton, CB10 1SA, UK 2 Open Targets, Wellcome Genome Campus , Hinxton, CB10 1SA, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site For correspondence: ab42{at}sanger.ac.uk Abstract Full Text Info/History Metrics Supplementary material Preview PDF Abstract Mounting evidence implicates microglia in neurodegeneration, but linking disease-associated genetic variants to target genes and cellular phenotypes is hindered by the inaccessibility of these cells. We differentiated 261 human iPSC lines into microglia-like cells (iMGL) in pools with phenotypic (differentiation, phagocytosis and migration) and single-cell transcriptomic readouts. Burden analysis of deleterious variants detected 36 genes influencing microglial phenotypes. Expression quantitative trait locus (eQTL) analysis found 7,121 eGenes, and 79 colocalizations across four neurodegenerative disease GWAS, half of which had limited prior evidence of causality. Integration of eQTL and phenotypic associations highlighted the role of disease-relevant variants including LRRK2 and TREM2 acting via microglial phagocytosis. A coupled CRISPR screen identified a role of TREM2 in phagocytosis and highlighted the importance of cellular state in directionality of phenotype. By contextualizing variant effects within disease-relevant microglial states, we provide a comprehensive framework for interpreting the function of risk loci in neurodegenerative disorders. Introduction Genome-wide association studies (GWAS) have identified thousands of genetic variants linked to complex traits and diseases, but translating these into biological insight remains challenging. Most variants are non-coding and likely influence disease by regulating gene expression, often in specific cell types, states, or environments. While expression quantitative trait locus (eQTL) mapping helps link regulatory variants to their target genes, it does not reveal the downstream cellular consequences. Similarly, population-scale mapping of genetic effects on cellular traits links genetic variation to functional outcomes, however, such studies have been limited by practical challenges: disease-relevant cell types are often difficult to access, and detecting modest genetic effects on complex cellular traits typically requires large sample sizes. Human induced pluripotent stem cells (hiPSCs) offer a scalable platform to model human biology in vitro , enabling access to otherwise inaccessible cell types, like those of the brain. They are also suitable for exploring how natural genetic variation influences gene expression and cellular traits independent of uncontrolled environmental effects, and provide a foundation for targeted functional follow-up using CRISPR perturbations, enabling stronger causal inference for variant-to-gene-to-phenotype relationships 1 – 6 . Pools of iPSCs enable the use of natural genetic variation while enhancing experimental scalability and reducing differentiation variability and batch effects. Recent studies have leveraged large numbers of donors to investigate how genetic variation influences iPSC-derived neuronal responses to stress and viral susceptibility 6 , demonstrating links between genetic background, gene regulation, and cellular phenotypes, identifying hundreds of eQTLs that colocalize with disease-associated loci from GWAS. Despite these advances, pooled iPSC-based approaches present several technical challenges. Both somatic and germline genetic variants can affect differentiation efficiency, resulting in imbalanced representation of donors within a pool. This introduces variability that can confound downstream analyses. Additionally, independent differentiation pools are still prone to batch effects, complicating data integration across experiments 6 . Improving our understanding of the factors that drive efficient and consistent differentiation is essential for scaling these models and maximizing their utility across the research community. Microglia have been implicated in neurodegeneration through their involvement in neuroinflammation, synaptic pruning, and clearance of pathological proteins. GWAS have identified hundreds of loci altering risk of neurodegenerative disease, and disease-associated variants, especially in Alzheimer’s disease (AD), are enriched in microglial-specific chromatin regulatory regions and genes 7 – 10 . Rare variants have further implicated microglia as a key cell type, e.g. the R47H mutation in microglia-specific TREM2 gene increases risk of AD 11 . However, the precise impact of the common variants linked to neurodegenerative disease on gene expression and microglial phenotypes mostly remains poorly understood, in part, due to the inaccessibility of the primary human cellular material. Human iPSC-derived microglia-like cells (iMGL) have emerged as valuable cell models to study molecular mechanisms dysregulated by disease variants. Here, we develop an experimental framework of pooled iMGL from 261 lines to map the impact of genetic variation on microglial molecular and cellular phenotypes. By combining single-cell transcriptomics in naive and stimulated states with functional assays measuring differentiation, phagocytosis, and migration, we capture a wide spectrum of microglial responses. We identify both common and rare variant effects through integration of genotype information, single-cell QTL mapping, genome-wide and gene-based association testing, and independent functional identification and validation of key regulators using CRISPR screens. Our suite of assays contextualises variant effects within microglial states relevant to disease enabling functional interpretation of risk variants implicated in neurodegeneration. Results Single-cell eQTL mapping in resting and stimulated iMGL We differentiated 16 pools totalling 261 iPSC lines (247 HipSci 12 from healthy donors and 14 IPMAR 13 lines including patients with early or late onset AD; Supplementary Table 1 and 2), with 16-72 lines per pool, to iMGL in naive (untreated), LPS, and IFN𝛄 treated conditions, to mimic inflammatory immune responses ( Methods , Figure 1a ) 14 . After quality control ( Methods ) we retained ∼2 million cells and integrated the pools using donors shared across all pools to remove batch effects (Supplementary Figure 1b). We analysed microglial subtypes present in our population using marker gene expression and label transfer 15 ( Figure 1b-d ; Supplementary Figure 1b-d). This showed that untreated cells presented a signature similar to disease-associated microglia (DAM, expressing CD9 , LGALS3 and PLA2G7 , 85% of cells), a small percentage of homeostatic microglia (HM, expressing P2RY12 and CX3CR1 , 2%) and a proliferative cluster (proliferating; expressing MKI67 , ASPM , UBE2C , RRM2 and DLGAP5 16 , 11%). Download figure Open in new tab Figure 1: Transcriptomic profile of iPSC-derived microglia. a) Experimental summary. b) Markers of microglial subtypes expressed in the three treatments. c) Aggregated marker scores from AddModuleScore of DAM, CRM and IFN markers of microglial subtypes overlaid over UMAP of merged treatments. d) Abundance of each annotated cluster per treatment. e) Pathway activity analysis with PROGENy on the pairwise differential expression results. f) Number of significant eGenes. g) Proportion of pairwise significant shared eQTL effects. g) eQTL in an eGene with diverging effects in IFN𝛄 and LPS. DAM: Disease- associated microglia, CRM: Cytokine-response microglia, HM: Homeostatic microglia, IFN: interferon response microglia. LPS treatment induced a signature of cytokine-response microglia (expressing CCL2 , CCL3 , CCL4 and IL1B ) in 69% of cells), with 10% of cells retaining a DAM signature, consistent with the microglial LPS response reported in the literature in mice and human 17 . Conversely, IFN𝛄-treated cells had a smaller proliferative cluster (1%), and overall showed increased expression of HLA genes and interferon response (IFN markers, including IFIT1 , IFIT3 and ISG15 ), as expected 18 . To maximize cell number, and thus power in downstream analyses, we focused only on proliferative and non-proliferative cell clusters. Differential gene expression (DE) analysis ( Methods ) identified 1726-2053 significant (FDR-adjusted p-value 1) DE genes per comparison (IFN𝛄 or LPS against the untreated baseline, and IFN𝛄 vs LPS, Supplementary Figure 2b, Supplementary Table 3). Gene set enrichment analysis (GSEA) of mSigDb Hallmark pathways (Supplementary Figure 2c) confirmed the increased expression of IFN𝛄 and inflammatory response genes in both the IFN𝛄 and LPS-treated samples. Pathway activity analysis ( Figure 1e , Methods ) highlighted a significantly increased JAK-STAT pathway activity in IFN𝛄, and to a lesser degree LPS, compared to untreated microglia. FOXA2 was upregulated in the untreated samples compared to both LPS and IFN𝛄 (Supplementary Figure 2d), consistent with its protective role against inflammatory phenotypes in microglia 19 . This indicated that our iMGL recapitulated the inflammatory transcriptional response. We next investigated how genetic variation influences gene expression in resting and stimulated iMGL. Following genotype and gene expression quality control (see Methods ), we retained 6,405,518 variants and 188-189 lines for analysis. This resulted in 7,121 unique significant eGenes (FDR <5%; Figure 1f , Supplementary Table 4) and 12,230 associated variants across the three treatments, split fairly evenly across conditions. To assess the consistency of genetic effects across conditions, we compared each significant eQTL-eGene pair across the three treatments ( Methods ). This revealed a high degree of pairwise eQTL sharing across treatments (83-85%; Figure 1g ). When considering only significant eGenes (FDR < 0.05), 74.6% were shared across all three treatments (Supplementary Figure 3b, Supplementary Table 5). Most of these shared eGenes had similar effect sizes (differences in beta < 0.5), while only 0.3% of eGenes showed the same direction of effect but significantly different magnitude, and 0.3% exhibited opposite directions in at least two conditions such as the diverging effects between LPS and IFN𝛄 for PARP1 and CD55 ( Figure 1h ). Only 11.2% of eGenes were significantly shared by any two treatments, while 14.2% were treatment-specific. The treatment-specific eGenes highlighted regulation of genes in processes specifically induced by treatment responses, e.g. STAT1 and PPARG, two genes in this category, are central in the enriched network of nitrogen metabolism in IFN𝛄-stimulation (STRINGdb FDR = 3 x 10 - 5 , Supplementary Figure 3c-d), which regulates the production of the nitric oxide compounds and macrophage polarization in response to inflammatory signals 20 . A regression-based framework for donor identity estimation in pooled differentiation assays While pooling iPSCs from multiple donors improves scalability, a major challenge lies in the variable response of individual donor lines to cell culture conditions. Some donors differentiate rapidly or proliferate efficiently, leading to imbalanced representation in the final cell pool. This variation can distort assay readouts and confound downstream association analyses if not properly accounted for. To address this, it is critical to monitor donor composition during differentiation to accurately measure donor-specific effects. We developed a method, poodleR (POoled dOnor Deconvolution by LEast square Regression), to estimate donor proportions from low-depth whole genome sequencing (WGS). This approach leverages known donor genotypes and minor allele frequencies across sequenced variants in the pool, applying constrained least squares regression to deconvolute the contribution of each donor with high accuracy, even at low coverage ( Figure 2a ). Download figure Open in new tab Figure 2: Deconvolution method and statistical analysis of differentiation efficiency associations. a) Schematic representation of the deconvolution method. b) Relative error at representative line proportions. c) Schematic representation of the microglia differentiation steps, including each step where line proportion was measured. d) Correlation between WGS deconvolution method and vireo estimates. e) Line proportions at every differentiation stage for a representative pool. f) Genes with significant (Bonferroni-adjusted p- value < 0.05) burden of deleterious variants on microglial differentiation efficiency for iPSC to macrophage precursor, macrophage precursor aging, and macrophage precursor to microglia stages. g) Violin plots of differentiation efficiency across deleterious variant burden for selected significant genes. h) Effect of aggregate burden within all significant genes (of same directionality) on differentiation efficiency for iPSC to macrophage precursor, and macrophage precursor to microglia. i) Area Under the Receiver Operating Characteristic curve (AUROC) for the three aggregate burden measures from significant results. To benchmark the accuracy of poodleR , we tested it on in silico pools with equal donor contributions (ten donors per pool). Even at a low mean sequencing depth (0.25×), the absolute error remained below 0.05 (Supplementary Figure 4a). We then evaluated performance under more realistic conditions using in silico pools of 19 donors with unequal proportions ranging from 0.0005 to 0.39. In this setting, absolute errors ranged from 0.005 to 0.03, while relative error at higher donor proportions remained low (∼5%, Figure 2b , Supplementary Figure 4b; see Methods ). As expected, relative error increased for donors with smaller proportions, reflecting the reduced number of reads overlapping informative variants. Despite this, poodleR achieved comparable accuracy to existing methods, such as census-seq, with both showing ∼94% accuracy at donor proportions of 2% (Supplementary Figure 5B). We also observed some donor-specific differences in accuracy, with certain donors consistently showing slightly higher or lower estimation precision (Supplementary Figure 5C) 21 , 22 . This may be explained by the degree of genetic similarity between donors in a pool, which can affect the ability to resolve individual contributions. By leveraging single-cell RNA sequencing (scRNA-seq) at the microglia stage, we were able to directly benchmark the accuracy of our WGS-based donor deconvolution method, poodleR , against established single-cell deconvolution approaches. Specifically, we used Vireo to assign cell line identities from scRNA-seq and then calculated global donor proportions for comparison. The results demonstrated the high accuracy of poodleR , with a Pearson’s correlation of R = 0.99 across all donor proportions in matched samples ( Figure 2d ). Even for donors contributing less than 5% to the pool, the correlations remained strong ( R = 0.87), underscoring the robustness of the method across a wide range of donor abundances. Genetic effects underpin donor composition during pooled iMGL differentiation The iMGL differentiation protocol (∼42 days) has key transitions at day 14 and day 28, marking commitment to the macrophage precursor and microglial stages. We used poodleR to monitor donor abundance dynamics at different differentiation stages in 16 pools. DNA was collected for WGS at the iPSC, macrophage precursor (at different days in culture, termed “young”, “intermediate” and “old”), and microglia stages, with the latter also being treated with LPS or IFN𝛄 ( Figure 2c ). Despite all pools presenting approximately equal cell line proportions at the iPSC stage, line abundance varied substantially during iMGL differentiation ( Figure 2e ). Competition differed throughout the process, such that donor lines taking over during iPSC growth (Supplementary Figure 5b) did not correlate with those outcompeting at the later stages ( Figure 2e ). Competition in culture is common to all pooled iPSC-derived cell studies, reflecting the variation of proliferation, survival and differentiation capacity of the donor cell lines (referred to from now on as “differentiation efficiency”, see Methods ). Several lines were shared across pools to control for batch effects: two lines (hegp_3 and aowh_2) were present in all pools, with several others repeated in a variable number of pools. We also included seven pairs of hiPSC clones from the same donor to measure clonal variability (Supplementary Table 1). We observed consistent changes in donor abundance for the same cell lines repeated across different pools (average interquartile ranges for repeated lines < 1, consistently smaller than that of a non- shared line; Supplementary Figure 5c), as well as across replicates of different iPSC clones from the same donor (data not shown). This indicates that genetic effects underpin differentiation efficiency. We therefore tested the effects of genetic variants on changes in donor proportions across key differentiation transitions: iPSC to macrophage precursor, young to aged precursors, and precursor to microglia, but did not detect any genome-wide significant variants (Supplementary Fig. 6d). However, there were three signals at p-value A within 20 kb of OCIAD1 and OCIAD2 (involved in the JAK-STAT and Notch pathway regulating cell cycle progression and differentiation 23 ), and another at chr7:138680001G>A within an intron of SVOPL . The third signal chr4:135084336C>T was found in the differentiation from iPSC to macrophage precursors within the lncRNA ENSG00000248434. Given the limited power to detect individual common variant effects we next evaluated the burden of deleterious exonic variants per gene. Using whole exome sequencing (WES), we tested all genes containing at least one deleterious (7,757 genes), missense non-deleterious (1,012 genes), or synonymous variants (9,080 genes) ( Figure 2d ; Methods ). We identified three genes with significant deleterious variant burden during the transition from iPSC to macrophage precursor (Bonferroni corrected p = 2 × 10⁻⁴): BCOR , MLLT4 , and LRRC55 (beta = –0.76, p = 2 × 10⁻⁵; beta = –0.76, p = 6 × 10⁻⁶; and beta = 0.47, p = 8 × 10⁻⁵, respectively; Figure 2f , Supplementary Table 6). Notably, deleterious BCOR variants have previously been linked to decreased dopaminergic neuron differentiation and enhanced iPSC survival 3 , 24 . This association was also evident when considering missense non-deleterious variants (Supplementary Figure 7a). For the transition from young to aged macrophage precursors, we identified a single significant gene, NEK11 (beta = 0.66, Figure 2g ), which encodes a kinase involved in the G2/M DNA damage checkpoint and is upregulated in cells with an arrested cell cycle 25 . The precursor-to-iMGL transition yielded 18 gene associations ( Figure 2f , Supplementary Table 6), most in the direction of enhancing microglial differentiation. One example is INPP5D (beta = 0.15, Figure 2g ), an AD-associated gene that regulates the inflammasome in microglia, and is also associated with monocyte and eosinophil counts 8 , 26 . We found that the cumulative burden of deleterious variants can be used to partially predict lines that may disproportionately expand or disappear in pooled cultures ( Figure 2h–i ; AUC = 0.59–0.79). This aggregate score was constructed by summing the burdens of deleterious variants across all genes that were significant in the burden tests and showed consistent directionality of effect for each transition. The global burden of deleterious variants across all genes and lines showed a modest yet significant overall effect on precursor to iMGL transition (beta = 0.0008, p = 8 × 10⁻⁵; Supplementary Figure 7b). Finally, aggregating the significant genes across all differentiation transitions revealed significant enrichment for essential genes ( Methods ) and genes required in macrophage survival 27 (p = 0.02 and 0.04 for DepMap and Covarrubias et al. ’s sets, respectively; Supplementary Figure 7c). Genetic mediators of phagocytosis and migration in iMGL Microglia mediate key neurodegenerative processes by sensing environmental cues and clearing protein aggregates, dead cells and synapses. Two essential functions related to these are migration and phagocytosis. Therefore, identifying genetic variants and genes that regulate these functions informs mechanisms underlying cell phenotypes and disease. We used a dual-fluorescent reporter assay to quantify phagocytic capacity, where iMGLs engulf GFP- and mCherry-labeled dead SH-SY5Y cells (pmChGIP SH-SY5Y). GFP fluorescence is quenched rapidly in the acidic endolysosomal compartment. We sorted two cell populations (mCherry⁺/⁻) by FACS and measured donor abundance by WGS ( Figure 3a ). This assay measures at the same time phagocytosis, acidification and degradation of reporter cells, which we will describe subsequently as the “phagocytosis” phenotype. Migration was assayed using transwells: microglia were placed on top, and DNA from cells migrating to the bottom or those remaining at the top was sequenced ( Fig. 3a ). The lower compartment contained chemoattractant (C5a) or media alone, to measure chemotaxis versus random migration. Download figure Open in new tab Figure 3: Genetic associations with cellular phenotype. a) Schematics of the assays and statistical testing strategy. b) Significant results for the burden tests of deleterious variants with phagocytosis and migration, per treatment. c) Comparison of the phagocytosis GWAS p-values at LRRK2 and the p-value for the aggregated LRRK2 gene signal in the eQTL-informed TWMR. Red line signals genome-wide significance for GWAS (5x10 - 8 ) and the blue line the significance for TWMR (10 - 6 ). Gene-annotated horizontal lines with larger dots indicate significance for TWMR. Inserted scatterplot shows phagocytosis phenotype (y axis) vs gene expression, with blue linear regression line. d) Miami plots of TWMR results (p-values per gene) integrating phagocytosis or migration GWAS and eQTL results, per treatment. Red lines signal genome- wide significance for TWMR (∼5x10 - 5 , above). Only significant genes with F-statistic above 10 are highlighted. e) Colocalization plot for the phagocytosis GWAS and the eQTLs in the LPS-treated sample at LRRK2 locus. No genome-wide significant associations were detected for either phenotype, likely due to limited power from small sample size and phenotypic variability (Supplementary Figs. 9–10), consistent with previous reports on cellular trait GWAS in in vitro models 28 , 29 . However, when testing for gene burden of exonic deleterious variants we identified 11 genes ( IL37 , PHLDB3 , CCR5 , HOXD4 , OR5P2 , FERMT1 , GPATCH11 , MUM1L1 , MFAP5 , PIK3C2A , TNFSF13 ) associated with phagocytosis and three ( IFI44L , CCDC88C , CXorf36 ) with migration ( Figure 3b , Supplementary Fig. 11a, Supplementary Table 7, Methods ). For example, CCR5 , a chemokine receptor highly expressed by microglia, was linked to phagocytosis in IFNγ-treated and untreated iMGLs (Supplementary Fig. 11b), in line with the role of CCR5 in mediating microglial recruitment and activation at sites of CNS injury or neurodegeneration, in synaptic plasticity, and upregulation in AD- associated microglia 30 . Additionally, IL37 was associated with enhanced phagocytosis in untreated microglia ( Fig. 3b ), aligning with prior findings in monocytes and macrophages 31 . Given that phenotype-associated variants are enriched for expression quantitative trait loci (eQTLs), 32 , 33 , 34 we leveraged our microglia-specific eQTL map to perform a transcriptome-wide Mendelian Randomization (TWMR) analysis 35 ( Methods ). Unlike variant-level tests, TWMR models joint causal effects of multiple eGenes at a locus, accounts for correlated regulatory architecture, mitigates horizontal pleiotropy, and is well-suited to underpowered GWAS. Applied to phagocytosis and migration across three treatment conditions, TWMR identified 32 significant gene–trait associations (Bonferroni p < 0.05 and F- statistic above 10; Fig. 3c–d , Supplementary Table 8). Among the 19 phagocytosis genes, we identified DYNC2LI1 (LPS-treated), a component of the dynein motor complex which is essential for various intracellular transport processes 36 , and PEX2 (IFNγ-treated) which is involved in peroxisome-mediated lipid and reactive oxygen species modulation, a process dysregulated in disease associated microglia 37 , 38 . Notably, LRRK2 showed a strong causal link with phagocytosis in LPS-treated cells (α = 0.67, p = 4×10⁻ 14 ; Fig. 3c ) consistent with the role of the G2019S activating variant on phagocytic activity in iMGL 39 . This association, driven by the lead LRRK2 eQTL, is markedly stronger than conventional association testing ( Figure 3d ), which does not leverage gene expression information. We further validated this effect using colocalization (H4=0.91, Figure 3e , Methods ). LRRK2 is implicated in the pathobiology of PD and AD, and these results further establish a role of common genetic variants in regulating the expression of LRRK2 and phagocytic activity in iMGL. Amongst the 16 migration-associated genes we observed VRK1, VANGL1, SPTBN1, EDN1, EPDR1, BBS7, and GCLC , genes involved in cell polarity and motility. 40 – 48 Importantly, we also uncovered disease-relevant genes, including MTHFSD (phagocytosis) and SORL1 (migration), implicated in Amyotrophic Lateral Sclerosis (ALS) and AD GWAS, respectively. SORL1 has been recently implicated in lysosomal regulation, a process closely intertwined with chemotaxis 49 . We also identified a phagocytic role for CST3 , a cysteine protease that regulates amyloid-beta deposition and is linked to AD risk. 50 , 51 We further identified microglial immune regulators such as PRMT6, EDN1 , and LY86 , shown to promote proinflammatory phenotypes in microglia and macrophages. 52 , 53 , 54 , 55 Together, this approach, combining GWAS of cellular phenotypes with eQTLs, reveals new biologically and clinically relevant gene-phenotype relationships. iMGL transcriptomic and phenotypic regulation reveals mechanisms of neurodegenerative risk Building on transcriptomic profiling and genetic analyses of migration and phagocytosis in iMGLs, we examined how these regulatory mechanisms intersect with neurodegenerative disease risk variants. First, using differentially expressed genes between resting and stimulated states revealed significant enrichment of AD 7 and PD candidate genes across all treatments (GSEA, Benjamini-Hochberg [BH]-adjusted p < 0.05; Fig. 4a ), indicating that IFNγ- and LPS-activated iMGLs recapitulate disease-relevant transcriptional signatures. Download figure Open in new tab Figure 4: Linking iMGL biology to neurodegenerative disease. a) Enrichment of DEGs across treatments in AD and PD candidate genes. Color denotes adjusted p-value of enrichment using GSEA. Stars denote adjusted significance of * < 0.05, **<0.01,***<0.001 b) Enrichment of DEGs across AD-PRS in hallmark pathways within each treatment. Color denotes normalised enrichment score from GSEA. Stars denote adjusted significance of * < 0.05, **<0.01,***<0.001 c) Number of independent GWAS loci tested for colocalization. d) Number and percentage of colocalizations per GWAS dataset. e) Degree of sharing of colocalized eGenes across treatments. f) Colocalization plot of PILRA eGene with AD GWAS in the LPS- treated iMGL. g) Colocalization plot of TREM2 eGene with AD GWAS in the untreated iMGL. PP = posterior probability. Next, we explored if we could capture transcriptional changes induced by the cumulative effects of disease risk alleles using polygenic risk scores (PRS). We focused on AD, given its evidence for microglia involvement in disease pathology 7 . We computed PRS for each donor ( Methods , Supplementary Fig. 12), including the 14 IPMAR donors with extreme PRS, to assess gene expression differences across treatments. PRS-based differential expression analysis (PRS- DEA, Methods ) identified 143 significant genes in untreated, 56 in IFNγ-, and 21 in LPS-treated iMGLs (FDR 75% remained significant after permutation testing (Supplementary Fig. 13e). We observed that genes identified through PRS-DEA converged on specific pathways (Supplementary Fig. 14a; Fig. 4b ; Methods ). High AD PRS was associated with reduced expression of oxidative phosphorylation genes across all conditions, consistent with AD-linked mitochondrial dysfunction 56 . Inflammatory response pathways were upregulated in both IFNγ- and LPS- treated microglia with high PRS, with a stronger signal in the LPS condition, including significant activation of STAT1 targets, driven by increased expression of SPI1, STAT1, and RFX family targets. These results support a model in which increased AD risk involves heightened microglial inflammatory responses 57 . Furthermore, genes upregulated at high PRS within LPS were significantly enriched for AD heritability using partitioned LDSC (Supplementary Figure 14b, Methods). We next assessed colocalization between iMGL eQTL and GWAS loci from AD, PD, ALS, and multiple sclerosis (MS) ( Figure 4c ). Among 77 AD independent loci from Bellenguez et al. (2022) we found 27 unique colocalizing eGenes at 22 loci (posterior probability of colocalization > 70%, Supplementary Table 10): ALKBH5 , ASPHD1 , BIN1 , CASS4 , CIAO2A , CTSH , DOC2A , FAM131B , GPR141, GSTK1 , IDUA , KAT8 , LLGL1 ,, MS4A6A , NSF , PILRA, PILRB, PLEKHA1 , PTK2B, RABEP1 , RASA1, SORL1 , TREM2 , TRIM37, TRPM7, YPEL3 , and ZYX . From Nalls et al. (2019) PD GWAS (85 loci) we found seven unique colocalizing eGenes ( LRRK2, TMEM163, NUP42, NSF, SPNS1, GPNMB, and STX4 ) at six loci. Finally, for MS (Patsopoulos et al. (2019), 145 signals) there were 41 unique eGene colocalizations at 29 loci and for ALS (van Rheenen et al. (2021), 15 signals) four unique eGene colocalizing at three loci ( C9orf72 , SCFD1 , MOB3B , and ERGIC1 ). There were 79 unique eGenes colocalizing within any of the GWAS loci ( Figure 4d-e , Supplementary Table 10- 11). AD had the highest colocalization rate (29%), followed by MS and ALS (20%); PD showed ∼4-fold fewer colocalizations (7%) despite more loci being tested (85 PD vs 77 AD loci). The larger number of colocalizations for AD highlights shared causal mechanisms underpinning AD and gene expression regulation in iMGL. Using Open Targets 58 , we assessed how many of the 79 colocalizing eGenes were previously prioritized as likely causal. We found that 37 (47%) had low prioritization scores (L2G < 0.2), including 15 of 27 AD colocalizations (56%), with limited additional genetic evidence from the literature (Supplementary table 11). This suggests that previous AD gene prioritization efforts may have missed key microglia-specific regulatory contexts. We observed 16 of the 79 colocalized genes were unique to the LPS condition, while 15 were unique to IFNγ ( Figure 4e ), thus identifying loci with complex, context specific gene regulatory effects, manifesting only upon a specific stimulation. For example, the AD-associated locus 16:30010081, colocalized with a different eGene depending on the condition: DOC2A in the untreated state, ASPHD1 in IFNγ-treated cells, and YPEL3 in LPS-treated cells. Our results recapitulated eight out of 14 AD colocalizations, three out of four PD colocalizations, and the single ALS colocalization ( SCFD1 eGene) previously reported in primary microglia 59 , 60 (Supplementary Table 10), highlighting the similarity between iMGL and primary cells and reinforcing iMGLs as a scalable, disease-relevant model of brain immune function. Our data enabled nomination of new candidate causal genes at disease loci in stimulated iMGL, e.g. at an AD locus, variants in high LD colocalized with expression of both PILRA and PILRB (PPH4 > 0.70; Fig. 4f , Supplementary Figure 15). These genes encode transmembrane receptors with distinct roles in inflammatory signaling and phagocytosis. The common AD-protective PILRA chr7:100406823 C-allele is in LD with PILRA G78R coding variant, which reduces ligand binding, including decreased entry of HSV-1 61 , supporting the hypothesis that recurrent infections may increase AD risk. PILRA KO microglia present altered immune signaling, including reduced cytokine production and increased chemotaxis 62 . PILRB, in contrast, encodes an activating receptor that signals through TYROBP/DAP12, promoting inflammatory responses, chemotaxis towards amyloid beta and phagocytosis, and increased neuronal damage 63 , 64 . While PILRA is often cited as the causal gene at this locus based on fine- mapping, rare coding variants 61 , and previously reported colocalization, our data provide colocalization evidence for both PILRA and PILRB (Supplementary Figure 15). In IFNγ-treated cells, the lead eQTL variant for PILRB (chr7:100334426C>T, rs7384878, beta = 0.58) is also the lead AD GWAS variant (beta = 0.92, PPH4=0.99). This variant is in high LD (r 2 =0.85) with the lead PILRA eQTL variant chr7:100482234A>T (rs6971558, beta = 0.22) in the same condition. In LPS-treated microglia, another eQTL variant (chr7:100373690T>C, rs2405442), also in strong LD with the GWAS lead variant (r 2 =0.97), serves as the lead eQTL variant for PILRA (beta = 0.27) and PILRB (beta = 0.99), further supporting the involvement of both genes. To our knowledge, this is the first evidence of colocalization for both PILRA and PILRB in microglia. In untreated microglia we identified TREM2, which encodes a receptor for amyloid beta and apolipoprotein particles, and upon ligand binding, activates microglia via the TYROBP/DAP12 adaptor, similar to PILRB, triggering migration, cytokine production, and phagocytosis. Rare missense mutations in TREM2 have been implicated in AD and neurodegenerative disorders 65 , 66 . We found a colocalization (PPH4 = 0.90, Figure 4g ) driven by a common eQTL variant, where the disease protective allele increases the gene expression (chr6:41191794 C>G, rs3800342; FDR=0.02, beta = 0.12; AD beta = -0.059; MAF = 0.36). This variant lies in the 3’ of TREM2 with putative regulatory activity. Fine mapping of the TREM2 locus in the Bellenguez et al . AD GWAS identified three independent rare risk variants, chr6:41181270 A>G (rs60755019, MAF =0.004, OR = 1.55), chr6:41161514 C>T (rs75932628, R47H, MAF = 0.003, OR = 2.39) and chr6:41161469 C>T (rs143332484, R62H, MAF = 0.013, OR = 1.41). The latter two are on the same haplotype (D′ = 1) as our common lead eQTL variant (Supplementary Figure 16a–c). This pattern is consistent such that the rare non- risk alleles are always on the AD-protective eQTL allele background, suggesting that increased TREM2 expression may reduce AD risk, providing a mechanistic link between common regulatory variation and rare coding risk alleles. Of the 79 colocalizing eGenes, SORL1 (AD) and LRRK2 (PD). were also identified in our TWMR analysis, providing a link from GWAS variant association, through eQTL to cellular phenotype. LRRK2 is an established PD gene, and our TWMR results indicate its role in regulating phagocytosis. In LPS-treated iMGLs, the lead colocalizing variant between phagocytosis and LRRK2 expression (chr12:40189297 T>G; phagocytosis LPS beta = –0.387, MAF = 0.44, Figure 3e ) is in perfect LD (D′ = 1, r 2 =1) with the lead LRRK2 eQTL in the same condition (chr12:40178345 T>C, eQTL in LPS beta = –0.59, MAF=0.44). In contrast, the lead LRRK2 eQTL in IFNγ-treated cells that colocalizes with PD-associated variants (chr12:40220632 C>T; eQTL-IFNγ beta = 0.39; PD GWAS beta = 0.14, MAF=0.13, Supplementary Figure 17) is only in moderate LD with the aforementioned phagocytosis-associated variant (chr12:40189297 T>G, D′ = 0.87, r 2 =0.15, MAF = 0.44). Despite this, the former variant is also significantly associated with changes in gene expression in LPS despite not being the lead signal in the region (nominal p-value = 6x10 - 13 , eQTL-LPS beta = 0.86), following the same direction of increased gene expression associating with increased phagocytosis. These observations suggest that LRRK2 expression and its phenotypic consequences are modulated by treatment-specific regulatory variants. The more common haplotype, tagged by the T alleles at chr12:40189297 and chr12:40178345, respectively, is associated with increased LRRK2 expression and enhanced phagocytosis in LPS-treated cells. However, this haplotype most often also carries the major C allele at chr12:40220632, which reduces LRRK2 expression in IFNγ-treated cells and is associated with reduced PD risk. Because the lead variants in LPS are relatively common (MAF = 0.44), LD estimates with the rarer IFNγ-specific variant may be less precise. Together, these findings underscore the complexity of regulatory architecture at the LRRK2 locus and highlight the importance of accounting for context-specific gene regulation when interpreting disease associations. Further functional validation will be needed to fully disentangle the molecular mechanisms linking LRRK2 to microglial function and PD risk. Finally, to explore how disease-linked genes might influence microglial behavior, we correlated gene expression with phagocytosis and migration phenotypes for genes within AD and PD GWAS loci ( Methods ). Expression of 47 AD- and 19 PD- candidate genes from GWAS associate significantly with phagocytosis (FDR < 0.05), and 49 AD- and 18 PD-associated genes with migration (Supplementary Table 12). Among these, LRRK2 expression increases with phagocytic activity (log₂FC = 0.30), consistent with the direction of its eQTL effect, and also with migration (log₂FC = 0.51). Conversely, TREM2 expression is negatively associated with both phenotypes (log₂FC = –0.11 and –0.28, respectively; Supplementary Fig. 18a). Looking at global expression patterns, we observed an inverse relationship between several regulatory pathways and TFs compared to those seen along the PRS differential expression axis (Supplementary Fig. 18b-c, Figure 4b ), including increased oxidative phosphorylation gene expression associated with higher phagocytic activity. Additionally, to assess the collective phenotypic impact of disease-associated eQTLs, we constructed microglia-specific PRS from regions with strong eQTL-AD GWAS colocalization (PPH4 > 70%), computed per treatment ( Methods ). We found a significant inverse association between the polygenic component of the AD eQTL-PRS and phagocytosis in LPS-treated microglia (p = 0.04, beta = –0.14; Supplementary Fig. 19), suggesting that increased genetic risk is associated with reduced phagocytic activity under inflammatory conditions. These findings underscore the utility of pooled iMGL assays in uncovering disease-relevant regulatory mechanisms and suggest that neurodegenerative risk genes modulate microglial gene expression and function. To build on this, we performed targeted CRISPR knockouts of genes implicated in Alzheimer’s disease. Targeted KO CRISPR screening identifies modulators of phagocytic uptake of dead neurons in iMGL To better understand the causal link between AD-associated GWAS genes and microglia function we performed a pooled CRISPR KO screen for phagocytosis in iMGL. Cells were transduced with a dual-guide CRISPR library (NeuroKO) targeting 251 genes consisting of the top 2-3 candidate genes at each AD GWAS locus and controls (Supplementary Table 13). Following 14 days of differentiation, iMGL were incubated with dead/fixed pmChGIP SH-SY5Y to assess phagocytosis. iMGL were then fixed and sorted into four bins based on mCherry intensity, reflecting different levels of phagocytosis, rates of degradation or export 14 ( Figure 6a and 6b ). Download figure Open in new tab Figure 6: KO Screening in iMGL to identify regulators of phagocytosis of dead neurons a) Outline of the screening protocol, a full method available at Washer et al., (2025) 70 b) Representative images of iMGL (turquoise) phagocytosing the dual reporter cargo (yellow), only once the dual reporter SH-SY5Y is internalised to the phagolysosome does the signal change from yellow to single red within 10 minutes of internalisation c) Rank plot of the genetic targets where KO results in increased phagocytosis (red), or decreased phagocytosis (blue). d) Boxplots showing the distribution of the FDR significant hits ( HAVCR2, TREM2, ITGAX ) dgRNA across the four sorted populations in the three repeated screens confirm enrichment. e) The log2FC of the three dgRNA in the three FDR significant hits between the highest and lowest phagocytosing populations, aggregated from the three screens using MAGeCK. f) TREM2 KO was validated using arrayed CRISPR screening in four different genetic backgrounds with three replicates. TREM2 expression was significantly reduced and g) phagocytosis was significantly increased in iMGL transduced with TREM2 sgRNA and not Intergenic sgRNA. Asterisks indicate p-values from Wilcoxon test (* p<0.05, ** p<0.01, *** p<0.001). h) Representative western blots for WT iMGL, VPX only transduced iMGL, TREM2 sgRNA transduced iMGL, Intergenic sgRNA transduced iMGL. The TREM2 and Intergenic were selected with puromycin to reduce the background. i) TREM2 targeting ASO resulted in significant KD of TREM2 at IC90 and j) a significant increase in phagocytosis in iMGL, compared to cells treated with matched scramble controls. Asterisks indicate p-values from Anova (* p<0.05, ** p<0.01, *** p<0.001). k) The TREM2 KO line BIONi010-C-17 was differentiated to iMGL or macrophages (iMac) and showed opposite effects on phagocytosis in the different cell types. p-values from Wilcoxon test, *** p<0.001. We identified three genes at FDR p<0.05 with HAVCR2 KO decreasing phagocytosis, and ITGAX and TREM2 KO increasing phagocytic signal. Additionally, we observed a suggestive signal for several biologically relevant genes reducing phagocytosis ( RAB7A, HMGCR, SORT1, INPP5D and LRP1; Figure 6c ), and increasing phagocytosis ( VAMP4, ATG14, PLCG2, CTSH, and PRKN ; Figure 6c ). Examining the normalised dgRNA counts across the four sorted bins showed clear directional trends for enrichment, indicating this is not a sampling bias ( Figure 6d /e). We further examined the correlation of genes shared between the NeuroKO screen and an independent targeted screen using a single sgRNA approach with 3x sgRNAs per gene and these showed a strong correlation and direction of effect (Pearson’s R=0.86, p=9.1e - 5 ), despite being different guides, vectors, and independent repeats (Supplementary figure 20), highlighting detection of robust phenotypic effects. Interestingly, the identification of TREM2 KO resulting in an increased mCherry signal (log2FC = 0.389, FDR p-value 0.00125) was consistent in direction with the eQTL-phagocytosis effect. This result was somewhat unexpected given previous reports that TREM2 knockout reduces phagocytosis in macrophages and microglia 67 – 69 . To confirm the validity of our finding, we evaluated its consistency across multiple orthogonal approaches, including arrayed CRISPR, antisense oligonucleotide treatment (ASO), and iPSC knockout lines. Divergent phagocytic activity in TREM2 knockout in iMGL and iMacrophage lineages To confirm the TREM2 finding we undertook an arrayed CRISPR approach utilising four different iPSC genetic backgrounds 71 – 74 . We transduced them with lentiviral pools of three sgRNA all-in-one lentivectors targeting either TREM2 or intergenic regions (as cutting controls). We harvested protein for TREM2 western blotting and performed the phagocytosis assay as previously described. We identified approximately 50% knockout of TREM2 at the protein level in cells treated with TREM2 targeting guides compared to those targeted with intergenic controls (Wilcoxon, p-value <0.01, N=9 (3 lines, 3 repeats); Figure 6f , h). We identified an increase in the proportion of mCherry positive iMGL in cells treated with TREM2 targeting sgRNA across all four genetic backgrounds, confirming that TREM2 knockout increases phagocytosis in iMGL (Wilcoxon p-value=0.0029, N=12 (4 lines, 3 repeats); Figure 6g ). To confirm that the observed phenotype was not a result of lentiviral KO of TREM2, we used TREM2- targeting ASO 73 . At IC90 we observed a strong decrease in TREM2 protein expression with TREM2 targeting ASO but not in the scrambled controls (ANOVA, p-value=0.0017) ( Figure 6i ) and a significant increase in phagocytosis (ANOVA, p-value=0.004) ( Figure 6j ). A second ASO targeting TREM2 showed the same phenotype at IC90 (Supplementary Figure 21). We further validated this finding using the clonal TREM2 KO iPSC line (BIONi010- C-17) and isogenic control (BIONi010-C) differentiated to both iMGL and macrophage-like cells (iMacs) in parallel. When differentiated to iMGL, there was an increase in phagocytic signal with TREM2 KO, confirming our results from lentiviral or ASO treatments. Interestingly, when the same precursors were differentiated into iMacs there was reduced phagocytosis with TREM2 KO (Wilcoxon p-value=0.00058, N=7 ( Figure 6k )), consistent with previous literature 67 – 69 . These results demonstrate that modulation of TREM2 leads to different cellular phenotype outcomes, depending on the cell state and cell type. Taken together, our results demonstrate that the impact of TREM2 modulation on phagocytosis is highly cell state–dependent, with robust knockdown and knockout increasing phagocytosis in iMGL but reducing it in iMacs, and that only strong reductions in TREM2 expression are sufficient to elicit a phenotypic response. This highlights the importance of cellular context in interpreting the role of TREM2 in neurodegenerative disease, where microglia are thought to mediate key aspects of pathology, including impaired debris clearance and chronic inflammation. Discussion Establishing causal links between genetic variants identified by GWAS, their effector genes and the cellular mechanisms they perturb remains a central challenge in complex traits. For neurodegenerative diseases, an additional challenge resides in inaccessibility of the primary tissue, such as microglia. Context-specificity of gene regulation further complicates interpretation, as non- coding regulatory variant effects often manifest only under defined cellular states, stimuli, or developmental stages. Our study addresses these challenges and further emphasizes the critical role of context-specific effects, apparent in our eQTL and CRISPR results. We demonstrate that pooled iPSC-derived iMGL provide a powerful and scalable model for dissecting genotype–phenotype relationships. Despite inherent variability in differentiation efficiencies, we demonstrate we can control for such effects using a careful experimental design including shared and repeat donors, and pooling of genetically matched donors. Combined with our low-depth WGS- based statistical method for donor deconvolution (poodleR), and quantitative modeling of donor effects we can robustly infer genetic associations across molecular and cellular traits. We benchmarked the performance of poodleR against single-cell inferred donor deconvolution and demonstrated its robust performance, contributing a new modelling tool to this expanding community. In line with previous studies of similar sample size, we were underpowered to detect significant effects of common variants on cellular phenotypes (differentiation efficiency, phagocytosis, and migration). Nevertheless, we detected effects at a suggestive p-value < 10 - 7 , indicating that these variants could contribute to shaping cellular phenotypes, but larger sample sizes are necessary to reach sufficient statistical power. However, stronger effects driven by rare deleterious variant analyses identified multiple gene burden hits associated with differentiation, phagocytosis, and migration. We identified 22 significant genes linked to cell expansion at various stages of differentiation, and observed that the cumulative burden of deleterious variants in these genes can partially predict disproportionate donor expansion. This highlights the potential of using it as a pre-screening tool to identify pools of donor lines with matched differentiation outcomes, which will improve the efficiency of this approach in future. Furthermore, we identified enrichment of deleterious variants in 14 genes associated with phagocytosis and with migration, capturing established phenotypic gene effects as well as highlighting novel genes in relevant biological processes. These results demonstrate that in vitro pooled iPSC models are effective for identifying genetic regulators of cellular phenotypes. As these approaches scale to larger numbers of donors, they will gain power to resolve both rare, high-impact loss-of-function variant effects, and more subtle contributions from common variants, enabling systematic dissection of the genetic architecture underlying cell biology. Integrating transcriptomics into our approach added a critical molecular layer that enhanced variant interpretation at multiple levels, linking genetic variation to gene expression, identifying mediators of cellular phenotypes, and strengthening connections to neurodegenerative disease associations. First, differential gene expression analysis across the spectrum of AD polygenic risk and different conditions (resting, LPS and IFNγ) revealed that high PRS is consistently associated with oxidative phosphorylation gene expression across conditions, aligning with established links between AD and mitochondrial dysfunction 56 . In parallel, inflammatory response pathways were upregulated in both IFNγ- and LPS-treated microglia with high PRS. These findings support a model in which elevated genetic risk for AD drives exaggerated microglial inflammatory responses 57 and suggest that disease-associated alleles converge on gene programs that impair key cellular functions, emphasising the need of further functional phenotyping to elucidate the mechanisms linking genetic risk to microglial dysfunction. Second, we leveraged transcriptomic data to identify genes that mediate microglial cell phenotypes (phagocytosis and migration), and to inform the interpretation of disease associations. By incorporating transcriptome-wide Mendelian randomization (TWMR) and our matched eQTL results into our analysis of cell phenotype GWAS, we identified 34 significant gene–trait associations linked to phagocytosis or migration. These analyses revealed several biologically relevant genes, including known AD-associated loci such as LRRK2 , which we now connect to phagocytosis, and SORL1 , which we link to migration, providing functional context to genetic risk loci through direct effects on microglial behaviour. Third, our study provided insights into the regulation of gene expression in both resting and stimulated microglial states. We identified thousands of eGenes, with approximately ∼75% shared across all treatment conditions. These eGenes showed significant colocalizations with neurodegenerative disease risk loci. For AD, 29% of tested loci showed evidence of colocalization, while colocalization rates were lower for ALS, MS, and particularly PD. Notably, LRRK2 and TREM2 , canonical PD and AD risk genes, respectively, showed condition-specific colocalization patterns, underscoring the context-dependent architecture of gene regulation at these loci. In LRRK2 , we observed allelic heterogeneity across conditions, with distinct regulatory variants contributing to phagocytosis and disease risk. Furthermore, nearly half of the colocalizations identified here were not previously reported in the Open Targets locus2gene analysis, reinforcing the importance of using both cell-type– and cell-state–relevant models to detect regulatory variants mediating disease risk. These findings align with previous work showing that genes with dynamic, temporal regulation are more likely to colocalize with GWAS loci than those with static, baseline eQTLs 75 . To functionally validate the role of candidate genes in regulating microglial phenotypes, we performed pooled CRISPR knockout screens to quantify phagocytosis and autophagic flux across putative AD-relevant genes identified by GWAS. Several known regulators emerged as significant hits ( TREM2 , PLCG2 , VAMP4 ), alongside previously uncharacterized candidates such as HAVCR2 , a gene within a recently identified AD GWAS locus. Notably, TREM2 knockout consistently increased phagocytic signal in iMicroglia, in contrast to prior reports in macrophage or microglial cells. TREM2 is a well-established phagocytic receptor, yet studies have reported conflicting effects upon its downregulation. A review by Jay et al. 66 suggested that TREM2 ’s role in phagocytosis is AD-stage dependent, where reduced expression impairs phagocytosis in early pathology but may enhance it in later stages. In advanced AD, microglia often adopt a disease-associated microglia (DAM) transcriptional state and accumulate around amyloid plaques. TREM2 knockout in this context has been shown to decrease DAM marker expression and promote a return to a more homeostatic phenotype 15 . Our transcriptomic profiling revealed that untreated iMicroglia in our system closely resemble DAM-like states, which may explain the observed increase in phagocytic activity following TREM2 knockout and the colocalization of TREM2 eQTLs with AD risk loci. Conversely, in iPSC-derived macrophages (iMacs), TREM2 knockout showed a trend toward reduced phagocytosis, consistent with previous studies. These findings suggest that basal cell state profoundly influences the phenotypic consequences of gene perturbation and must be carefully considered in functional genomic studies of disease-relevant traits. In summary, this multifaceted approach enabled us to connect genetic variants, both common and rare, to effector genes and gene programs, downstream cellular phenotypes, and potential disease mechanisms. Transcriptomic and phenotypic analyses in iMGL offer a powerful framework for interrogating microglial dysregulation in neurodegenerative disease. The stimulation-specific eQTLs and colocalizations we identified highlight the importance of immune context: placing cells under relevant inflammatory conditions revealed regulatory effects that would otherwise remain undetected, providing insight into how genetically driven disease risk is mediated through dynamic gene regulation. More broadly, our study demonstrates the feasibility and value of integrating genetic association data, single-cell transcriptomics, and pooled functional genomics to investigate genotype–phenotype relationships at scale. While we focused on iPSC-derived microglia, the experimental and computational framework is readily applicable to other iPSC-derived or primary cell types across a range of diseases. This includes applications in inflammatory, metabolic, cardiovascular, and psychiatric disorders, where context-specific regulation and cellular function are critical to pathogenesis. Future efforts will benefit from increased donor diversity, expanded perturbation conditions, and broader phenotype panels. Together, these approaches will be essential for systematically resolving the molecular mechanisms underlying complex traits and for translating genetic risk into mechanistic insight across diverse cellular systems. Methods Human iPSC maintenance 261 human iPSCs lines from 254 European donors were obtained from the HipSci project ( http://www.hipsci.org , Cambridgeshire 1 NRES REC Reference 09/H0304/77, HMDMC 14/013, for a list of lines used, see Supplementary Table 1), IPMAR project 13 or commercially from Bioneer: TREM2 knockout (BIONi010- C-17) and isogenic control (BIONi010-C). iPSCs were thawed onto tissue culture treated 6-well plates (Corning, 3516), coated with 10 μg/mL vitronectin (VTN-N) (Life Technologies, A14700) or 10 μg/ml Vitronectin-XF (StemCell Technologies, 07180) using complete Essential 8 (E8) medium (Thermo Fisher, A1517001) supplemented with 10% CloneR™ (StemCell Technologies, 05888). After thawing, cells were expanded in E8 medium for at least 2 passages using Gentle Cell Dissociation Reagent (StemCell Technologies, 100-0485) for cell dissociation. Multiple lines were synchronized by adjusting the splitting ratio of each line, aiming for 60-85% of confluence at each passage and on the pooling day. To enable pooling of more lines, mini-pools of iPSCs lines were generated, cryopreserved and subsequently thawed for final pooling. Briefly, iPSC cultures were dissociated into single cell suspension with Accutase (Millipore, SCR005) and resuspended in E8 medium containing 10 μM Rock inhibitor Y-27632 (StemCell Technologies, 72305). Cell counting was performed using Countess II (Thermo Fisher) and equal amounts of each iPSC line were pooled in E8 medium containing 10 μM Rock inhibitor Y-27632. The cell suspension was seeded at 40,000 cells/cm 2 on vitronectin and cryopreserved after 3 days in Cell Freezing Medium consisting of 90% Knockout Serum Replacement (Life Technologies, 10828-028) and 10% DMSO (Sigma, D2650). The mini-pools were subsequently thawed and passaged once before final pooling. iPSC pooling and microglial differentiation Microglial differentiation via Embryoid Body (EB) formation was performed according to Washer et al. (2022) 14 . Briefly, iPSC cultures were dissociated into single cell suspensions using Accutase and resuspended in E8 medium containing 10 μM Y-27632. Cell counts were performed using Countess II (Thermo Fisher) and equal proportions of each iPSC line were pooled in E8 medium supplemented with 10 μM Y-27632, 50 ng/mL BMP-4 (Peprotech, 120-05ET), 20 ng/mL SCF (Peprotech, 300-07), and 50 ng/mL VEGF (Peprotech, 100-20A). The pooled suspension was seeded at 10,000 cells in 100 µL per well in Round Bottom Ultra-Low Attachment 96-well plates (Corning, 7007). Plates were centrifuged at 300 x g for 3 min and incubated at 37°C 5% CO2 for 3 days. Each pool contained between 16 to 72 donors. On EB Day 3, 50 µL spent media was removed from each well and 100 µL fresh media (E8 medium supplemented with 10 μM Y-27632, 50 ng/mL BMP-4, 20 ng/mL SCF and 50 ng/mL VEGF) was added. An extra centrifugation at 300 x g for 3 min was performed if mini-satellite EBs were observed around the main EB. On EB Day 6, every 42 EBs were transferred to one T75 flask (Corning, 430641U) coated with 0.1% gelatin (Sigma, G1890) and cultured in Factory Medium: X- VIVO (Lonza, BE02-060F), supplemented with 2mM Glutamax (Life Technologies, 35050-061), 1:100 Pen/Strep (Life Technologies, 15140-122), 55μM 2-mercaptoethanol (Life Technologies, 31350-010), 100 ng/mL M-CSF (Peprotech, 300-25), and 25 ng/mL IL-3 (Cell Guidance Systems, GFH80). Factory flasks were maintained at 50mL volume with 50%-80% media changes every 3-4 days. On Factory Day 35 – 57, the floating precursors in spent media were harvested for a 14-day microglial differentiation. Cells were sieved through 40 µm cell strainers (Falcon, 352340), centrifuged and pooled together. Cell counts were performed manually using disposable Neubauer-Improved haemocytometers (NanoEnTek, DHC-N01-50) and cells resuspended in ITMG media: Advanced DMEM/F12 (Life Technologies, 12634-010) supplemented with 2mM Glutamax, 1:100 Pen/Strep, 100 ng/mL IL-34 (Peprotech, 200-34), 50 ng/mL TGFβ1 (Peprotech, 100-21C), 25 ng/mL M-CSF, 10 ng/mL GM-CSF (Peprotech, 300-03- 20). Cell suspensions were seeded at 500,000 cells per 6-well (Greiner Bio-One, 657160) for single-cell RNAseq, 3,906,250 cells per T75 flask (Greiner Bio-One, 658170) for migration assay, or 1,126,400 cells per 6-well for phagocytosis assay. Media was topped up to double the volume on day 3 or 4, and subsequent half media change was performed on day 7 and day 10 or 11. iMacrophage differentiation The floating precursors were harvested for a 7 day macrophage differentiation (Van Wilgenburg et al 2013). Precursors were resuspended in iMac media consisting of X-VIVO 15 (Lonza), supplemented with 2mM GlutaMAX, 100 ng/mL M-CSF. Precursors were incubated at 37℃ and 5% CO2 and differentiated for 7 days with a half media change on day 3 or 4. IFN γ and LPS stimulation On day 11 of the 14-day microglial differentiation, cells were treated for 72 hrs with freshly prepared 20 ng/mL IFNγ (R&D Systems, 285-IF) or 100 ng/mL LPS (Sigma, L5668) diluted in ITMG media. Sample preparation for single cell RNAseq On microglial differentiation day 14, the cells were washed twice with DPBS (-/-) (Life Technologies, 14190-094) dissociated with 5 mM EDTA (Life Technologies, 15575-038) supplemented with 4 mg/mL lidocaine (Sigma L5647). The cells were incubated at 37°C for up to 30 min before adding 0.04% BSA (Sigma A9543 or A8806) in DPBS to stop the dissociation. The cells were dissociated using a P1000 or 5mL stripette and collected into a centrifuge tube. After centrifugation, the cells were resuspended in 0.04% BSA and sieved through a 30 μm cell strainer (Miltenyi Biotec, 130-098-458) twice. After centrifugation, the cells were resuspended in 0.04% BSA and manually counted using Neubauer-Improved C- Chip disposable haemocytometer. The concentrations were adjusted to 449,612 cells/ml to load 17,400 viable cells per lane on Chromium Next GEM Single Cell 5ʹ Kit v2 (10x Genomics, PN-1000263) or 1,491,021 cells/ml to load 115,405 viable cells per lane on Chromium Next GEM Single Cell 5’ HT Kit v2 (10x Genomics, PN- 1000356). Subsequent GEM handling and library preparations were performed according to the manufacturer’s instructions. Libraries were QC’ed using Qubit dsDNA HS assay (Life Technologies, Q32854), BioAnalyser (Agilent, 2100) or TapeStation (Agilent, 4150) and/or KAPA library Quant Kit for Illumina (Roche, KK4824). Multiplexed libraries were sequenced using HiSeq or NovaSeq 6000 (Illumina) with read length 28-10-10-90, 26-10-10-90, 100 paired-end or 150 paired-end, aiming for a depth of at least 20,000 reads per cell. Migration assay Microglia were seeded from precursors of Factory Day 39 – 43 at a density of 3,906,250 cells per T75 flask (Greiner Bio-One, 658175). On microglial day 14, the cells were dissociated with 5 mM EDTA (Life Technologies, 15575-038) supplemented with 4 mg/mL lidocaine (Sigma L5647). AdvDMEM/F12 was added to stop the dissociation and the cell suspensions were sieved through 30 µm cell strainers (Miltenyi Biotec, 130-098-458). The centrifuged cell pellets were resuspended in ITMG media and cell count were performed manually using Neubauer-Improved C-Chip disposable haemocytometer. For each 6-well insert (Sarstedt, 83.3930.500), 2.7mL suspension containing 825,000 cells was seeded to the inner well. After cells were settled for 15 min at room temperature, 3.168 mL ITMG media with or without 3 nM C5a (R&D Systems, 2037-C5-025/CF) was added underneath each 6-well. After incubation at 37°C for 6 hrs, genomic DNA was harvested and sister wells were fixed in 4% PFA for imaging. Genomic DNA from upper and lower sides of the transwells were harvested using FLOQSwabs (COPAN, 520CS01) and cell pellets were collected from the washes of each side of the transwells. Corresponding swabs and pellets were combined for genomic DNA extraction using Puregene Cell Kit (QIAGEN, 158043) according to manufacturer’s instructions. To quantify the migration rate, whole well imaging was performed on EVOS Fl Auto (Life Technologies) after fixation in 4% PFA and staining with NucBlue™ Live ReadyProbes™ Reagent (Life Technologies, R37605). After imaging all cells, the unmigrated portion on top of the transwells was removed with swab and the migrated portion at the bottom of the transwells was imaged. The number of nuclei in all-cells images was quantified using Cellpose (version 2.0.5) with -- diameter 5 --flow_threshold 0 --cellprob_threshold -8. The number of nuclei in migrated-cells images was quantified using Cellprofiler (version 4.2.1) with the threshold strategy set as Global, threshold method set as Robust background, lower bounds on threshold set to 0.015, and upper bounds on threshold set to 1.0. The migration rate was calculated as the number of nuclei in migrated-cells divided by the number of nuclei in all-cells . The percentage of chemotaxis was calculated by subtracting the cell migration rate in -C5a condition from the cell migration rate in +C5a condition. Generation of mCherry-eGFP pmChGIP SH-SH-SY5Y, maintenance & fixation mCherry-eGFP fusion SH-SY5Y were generated as previously described in Washer et al. 2022, and are henceforth referred to as pmChGIP SH-SY5Y, clone F4 was used for all experiments. pmChGIP SH-SY5Y were maintained in DMEM/F12 (ThermoFisher, 11320033) supplemented with 2mM GlutaMAX (ThermoFisher, 35050061) and 10% FBS (Gibco, 10500-064) at 37°C 5% CO2. Cells were maintained until reaching confluency before washing with DPBS (-/-) to remove dead cells and lifted with TryPLE Express (Life Technologies, 12604- 013) and harvested into 50mL tubes. Flasks were washed with suspension containing harvested pmChGIP SH-SY5Y. Cells were centrifuged at 400g for 5 min at room temperature before removing supernatant and resuspending with 1 mL DPBS (-/-). Cells were then adjusted to 8x10 6 cells/mL in DPBS (-/-) before adding an equal volume of 4% PFA (1:1 dilution, final concentration 2%) mixed vigorously with a stripette and incubated at room temperature for 10 min, with gentle vortexing every 2 min. The 2% PFA/cell mix was diluted to 1% by adding an equal volume of DPBS (-/-) before centrifuging at 1200g for 5 min at room temperature. The supernatant was removed and the pellet resuspended 5mL DPBS (-/-) before repeating the centrifugation and resuspension a further two times (total three). The fixed pmChGIP-SHSY5Y were then counted and stored at 4°C overnight. Prior to addition to iMGL, the pmChGIP-SH-SY5Y were resuspended in ITMG media. Dual colour phagocytosis assay Microglia were seeded from precursors of Factory Day 46 – 57 at a density of 1,126,400 cells per 6-well (Greiner Bio-One, 657160). On microglial day 14, fixed SH-SY5Y cells were added to microglia in ITMG media at a ratio of 2:1 SH- SY5Y:microglia. After incubation at 37°C for 2 hrs, wells were washed with HBSS (+/+, Life Technologies, 14025-050) to remove cargos and with DPBS (-/-) before lifted with TryPLE express (Life Technologies, 12604-013) and collected into 0.04% BSA. The cell suspension was fixed in 4% PFA and washed with DPBS twice before resuspended in 0.04% BSA for sorting on Mo-Flo XDP (Beckman Coulter). Genomic DNA extraction was extracted from fixed cells using Puregene Cell Kit according to manufacturer’s instructions and quantified using Qubit dsDNA HS assay. Genomic DNA library prep for phenotypic assays Genomic DNA extracted from migration and phenotypic assays was quantified using Qubit dsDNA HS assay. Library prep was performed using NEBNext® Ultra™ II FS DNA Library Prep Kit for Illumina (NEB, E7805) with NEBNext® Multiplex Oligos for Illumina® (Dual Index Primers Set 1) (NEB, E7600) following manufacturer’s instructions. Libraries were QC’ed using Qubit dsDNA HS assay, BioAnalyser or TapeStation and/or KAPA quantification qPCR. Multiplexed libraries were sequenced using NovaSeq 6000 (Illumina). Cloning of Cas9-guide all-in-one vector Three guide pairs per target (plus intergenic controls) were cloned into pLentiCRISPRv2 (Addgene https://www.addgene.org/52961 ) with a modified guide scaffold (Addgene https://www.addgene.org/50946/ ) in an arrayed format. Briefly, forward primers containing the 5’ guide, homology arms to plasmid backbone and dual guide cassette 76 and reverse primers containing the 3’ guide, homology arms to plasmid backbone and dual guide cassette were synthesized by IDT in an arrayed format. Sequence of forward primers: tatcttgtggaaaggacgaaacaccGNNNNNNNNNNNNNNNNNNNgtttcagagctaga aatagcaagttg Sequence of reverse primers: gctgtttccagcatagctcttaaacNNNNNNNNNNNNNNNNNNNCTGCATTGGCCGGG AATTGAAC The dual cassette flanked by dual guides and homology arms was PCR amplified with the above primers and a template plasmid containing the dual guide cassette, using Q5 High-Fidelity 2X Master Mix (NEB, M0492) with an annealing temperature of 64°C. Sequence of dual guide cassette: gtttcagagctagaaatagcaagttgaaataaggctagtccgttatcaacttgaaaaagtggcaccga gtcggtgcGCAGAGGCATTGGTGGTTCAGTGGTAGAATTCTCGCCTCCCACGCGGGA GaCCCGGGTTCAATTCCCGGCCAATGCAg The gel extracted (NEB, T1020) or column purified (NEB, T1030) PCR was cloned into Bsmb1-V2 (NEB, R0739) digested pLentiCRISPR v3.0 backbone using Gibson Assembly® Master Mix (NEB, E2611) at a ratio of 3:1 insert:backbone. The Gibson product was transformed into DH5-α cells (NEB, 2987U) and clones were miniprep using Qiaprep 96 plus miniprep kit (QIAGEN, 27291). The sequence of each clone was verified by sequencing with universal U6 primer. Post arrayed cloning, library was pooled and the representation was quantified by MiSeq (Illumina) before expansion using electrocompetent cells (Lucigen, 60242-1) to ensure sufficient coverage and maxiprep using ZymoPURE™ II Plasmid Maxiprep Kit (Zymo, D4203). For Miseq verification, a nested PCR was performed using the following primers and KAPA HiFi HotStart ReadyMix (Roche, KK2602). PCR1 forward primer: ACACTCTTTCCCTACACGACGCTCTTCCGATCTCTTGTGGAAAGGACGAAACA PCR1 reverse primer: TCGGCATTCCTGCTGAACCGCTCTTCCGATCTttcccactcctttcaagacc 12 cycles of 98°C 20s, 60°C 15s and 72°C 15s PCR1 was column purified (NEB, T1030) and 1/100 was used as template for PCR2. PCR2 forward primer: AATGATACGGCGACCACCGAGATCTACACTATAGCCTACACTCTTTCCCTACACGAC GCTCTTCCGATCT PCR2 reverse primer: AATGATACGGCGACCACCGAGATCTACACTATAGCCTACACTCTTTCCCTACACGAC GCTCTTCCGATCT 14 cycles of 98°C 20s, 70°C 15s, 72°C 15s Production of lentiviral particles Lenti-X HEK293T (Takarabio, 632180) were maintained in HEK media (DMEM (Corning, 15-013-cv), 10% FBS, 2mM GlutaMAX) at 37°C 5% CO2 and split when at 70-80% confluency using TrypLE Express following standard practice. Prior to seeding for lentiviral production, plates were coated with 0.1% Gelatin/dH2O (Sigma, G1393) for 1 hour at 37°C. For large scale production Lenti- X HEK293T were seeded in 15cm dishes at 14x10 6 cells/dish in 30mL HEK media, for small scale production in 6wp. 18-24 hours post seeding the confluency was checked (approx 70%) before transfecting with 7.5μg pSIV3+ or lentiviral library, 18.5μg psPAX2 (Addgene, 12260), and 4μg pMD2.G (Addgene, 8454) in 7.5mL OptiMEM (Invitrogen, 51985026) with lipofectamine LTX (Invitrogen, 15338100) per 15cm plate, topped up to 15mL with OptiMEM. 24 hours post transfection, the transfection mix was replaced with HEK media. At 48 hours post transfection viron containing media was harvested into 50mL tubes and replaced with fresh HEK media. Viron media was centrifuged at 500g for 5 min to pellet any HEK cells, before filtration through 0.45μm filters. Filtered viral supernatant was split between 50mL tubes and concentrated through centrifugation at 20000g for 3 hours at 4°C. Supernatant was removed and the viral pellets resuspended on ice for 1 hour in 1.5% BSA/DPBS. Resuspended virons were combined and stored at 4°C. The process was repeated at 72 hours post transduction before combining both the 48 hour harvest and 72 hour harvest and split into single use aliquots and stored at -80°C. A full method can be found in Washer et al. 2025. Total protein extraction and western blotting At d14 ITMG media was removed and either 50μL (96wp) or 100μL (48wp) RIPA (Thermofisher, 89900) containing protease inhibitor (Roche, 11836170001) was added to each well and the plate incubated on ice for 10 min. Lysates were transferred to a 96 well PCR plate, sealed, and vortexed for 20 seconds. The plates were centrifuged at 5000 rpm for 5 min to pellet any cell debris and lysates transferred to a clean 96 well plate and stored at -80°C. Total protein was calculated by BCA assay following manufacturers instructions (Thermo Scientific, 23225). 10-40μg of protein was then diluted in water, NuPAGE reducing agent (Invitrogen, NP0009) and NuPAGE LDS sample buffer (Invitrogen NP0007), boiled at 95°C for 5 min and loaded onto a 4-15% Mini- PROTEAN TGX precast gel (BioRad, 4561086) with 4μL Spectra Multicolor High Range Protein Ladder (Thermo-scientific, 26625). Samples were run at 100V for 75 min. Protein was transferred to a PVDF membrane (activated in methanol for 2 min) at 100V for 75 min. Membranes were then blocked for 1 hour at RT with 5% Milk PBS-Tween (PBST). Membranes were washed with PBST before staining with primary antibody made up in 1% Milk PBST overnight at 4°C. Primary antibody was removed and membranes washed with PBST before incubating with the HRP conjugated secondary antibody (prepared in 1% Milk PBST) for 1 hour at room temperature. Membranes were then washed and stained with enhanced chemiluminescence substrate for 5m before imaging. For data analysis images underwent densitometry in ImageJ. Target protein levels were normalised to respective housekeeping controls before being normalized to the untreated control to give “Relative Protein Expression (to Control)”. All data analysis was performed in R. CRISPR KO screening for phagocytosis KOLF2.1S hiPSC were differentiated to iMGL as described above. The library contains 783 dual-guide all-in-one CRISPR vectors targeting 251 AD GWAS hits and intergenic controls, henceforth known as NeuroKO. Genes were chosen from prioritisation in Schwartzentruber et al . 5 , and OpenTargets Locus2Gene score 58 for the additional loci discovered in Bellenguez et al. 8 Guide RNA pairs were designed using anchor guides chosen using the prioritisation scheme from MinLib 77 and paired with guides targeting the same exon using a tRNA-based dual guide system 78 . In order to maintain high coverage of the library, between 1.1 x 10 8 and 1.23 x 10 8 precursors were transduced at a multiplicity of infection between 1.2 - 1.5 along with a viral accessory protein (VPX), which has been shown to improve transduction efficiency in this model 76 . We performed 3 replicates of the screen, defined as 3 separate transductions and harvests of the precursors. After differentiation to iMGL for 14 days (14d) in ITMG we then incubated the iMGL with pmChGIP SH-SY5Y at a ratio of 2 pmChGIP SH-SY5Y to 1 iMGL and phagocytosis was undertaken for 6 hours. Non-phagocytosed cargo were washed, iMGL lifted, and fixed with 2% PFA, before sorting on the mCherry signal in four bins: the top 20% (top T20), 55-75% (upper middle UM), 25-45% (lower middle LM) and bottom 20% (lowest L20). In total between 0.92 x 10 6 and 1.7 x 10 6 cells were sorted into each bin and a total of between 4 x 10 6 - 6 x 10 6 total cells sorted across the four bins in the three screens. Following DNA extraction and assuming 6pg of DNA per cell, the coverage of screen across the three repeats were estimated to be between 263 and 454 per bin. dgRNA sequences were amplified and indexed by PCR and sequenced by Illumuna NovaSeq. Following deconvolution, dgRNA sequences were quantified by MAGeCK 79 . Mapping efficiency and guide representation by gini index was examined to confirm the quality of the data using MAGeCKFlute 80 and are shown in (Supplementary Fig. 22). Following this, the guide overrepresentation was calculated between the L20 and T20 populations, representing lowest phagocytosing and highest phagocytosing iMGL to provide a list of hits. Full methods are available at Washer et al. (2025) 70 . Arrayed Lentiviral Screening AOWH_2, HEGP_3, and KOLF2.1S lines from the HipSci repository ( www.hipsci.org ) were used for all CRISPR screening validation. 3 sgRNA lentiviruses to TREM2 and Intergenic regions were cloned into pLentiCRISPR v3.0 as described above and individual viruses were generated as previously described. Individual virus titres were calculated by resazurin survival following puromycin selection on KOLF2.1S iMGL. Following successful titring the 3 TREM2 sgRNA were pooled, and the 3 INT sgRNA were pooled and stored at - 80°C. AOWH_2, HEGP_3, and KOLF2.1S were transduced as follows. PreMacs were harvested from respective factories (age 32d, 36d, 42d) and adjusted to a final concentration of 4.84x10 5 cells/mL in ITMG. Polybrene was added to a final concentration of 4μg/mL and inverted to mix. 200μL of cells were aliquoted into 48wp (Greiner, 677180) to give 9.68x10 4 cells/well (WT). VPX was added to the remaining cell suspension at the determined concentration for full KD of SAMHD1, cells were inverted and 200μL of cells were aliquoted to give a VPX control (VPX). The remaining cell suspensions (containing polybrene and VPX) were split between two 15mL tubes and either TREM2 lentiviral pools (TREM2) or INT lentiviral pools (INT) were added, cells were inverted and 200μL of cells were aliquoted. Cells were incubated at 37°C 5% CO2. 24hr post transduction, all media was removed and replaced with 200μL fresh ITMG. On d4/5 post-transduction 200μL fresh ITMG was added to each well, a subset of wells were selected with puromycin (at final concentration 1μg/mL) to enrich for transduced cells and to check transduction efficiency (TREM2 - Puro, INT - Puro). Two further 50% media changes were undertaken on d7 and d10 post transduction with assays completed on d14. Protein was harvested as previously described and underwent western blotting for TREM2 and Vinculin to confirm levels of TREM2 KD. The remaining wells were fed the dual fluorescent reporter pmChGIP SH-SY5Y line for 6 hours before washing, lifting, fixing with 2% PFA, and undergoing flow cytometry as previously described. TREM2 ASO Knockdown An overview of the ASO experiments is provided in Supplementary Fig. 23. TREM2-171, TREM2-192, Scramble 1, and Scramble 2 ASO sequences from Vandermeulen et al. 81 were ordered with modifications from IDT and resuspended to 100μM in DPBS (-/-) and stored at -20°C. To obtain IC50 and IC90 curves, KOLF2.1S preMics were seeded in 96 well plates (Greiner 655180) at 3x10 4 cells/well in 96 well plates in 100μL ITMG. 72hr post seeding 50μL of ITMG was replaced with 50μL fresh ITMG. 7 days post seeding 50μL ITMG was removed and individual TREM2 ASO were added serially in triplicate to give final concentrations (20μM, 5μM, 1.25μM, 312nM, 78.1nM, 19.5nM, 4.88nM, 1.22nM, 305pM, 76.3pM), in 100μL ITMG. Another 50% media change occurred on d10 before harvesting total protein on d14. IC50 and IC90 were calculated in R using the drm() function from the drc package (version 3.0-1) (Supplementary Fig. 21) 82 . For validation experiments the above protocol was repeated at the respective IC50 and IC90 doses for TREM2-171 and TREM2-192, with the scramble controls matching the respective IC doses for each TREM2 targeting ASO. AOWH_2, HEGP_3, and KOLF2.1S were used as follows. PreMics were harvested from respective factories (age 32d, 36d, 42d) and adjusted to a final concentration of 4.84x10 5 cells/mL in ITMG before plating 200μL in 48wp (Greiner, 677180). At d3-4 post seeding, a 50% media change was undertaken. At d7 ASO were diluted at 2x concentration in ITMG before removing 100μL from the cells and adding 100μL 2x ASO in ITMG (final concentration 1x). At d10 a 50% media change was undertaken and assays completed at d14. Protein was harvested as previously described and underwent western blotting for TREM2 and Vinculin to confirm levels of TREM2 KD. The remaining wells were fed the dual fluorescent reporter pmChGIP SH-SY5Y line for 6 hours before washing, lifting, fixing with 2% PFA, and undergoing flow cytometry as previously described. TREM2 protein levels and phagocytosis are shown as relative to vehicle control. Genotype imputation and quality control Raw genotyping array files were obtained per line from the HipSci project ( http://www.hipsci.org ) and from IPMAR project 13 , and imputed using the genimpute pipeline from the eQTL Catalogue. Briefly, it converts genome coordinates from GRCh37 to GRCh38 with CrossMap v.0.4.1 83 , aligns them with 1000 Genomes Project High Coverage 84 reference panel with Genotype Harmonizer v.1.4.20 85 , excludes variants with Hardy-Weinberg p-value 0.05 and minor allele frequency < 0.01 with bcftools 86 , performs genotype pre-phasing with Eagle v.2.4.1 86 , 87 , and imputation with Minimac4 88 . Post-imputation QC excludes variants with imputation r 2 0.01, and multiplies genotype dosage of male samples on the Non-PAR region of the X chromosome by two. Sex was imputed for donors that were missing this information using bcftools +guess-ploidy on region X:2781480-155701381. Donor proportion deconvolution by whole genome sequencing We used whole genome sequencing (WGS) to measure the proportion of the different donors within each sample. This method (named poodleR ) requires that donors are genotyped, and then variants subset to those that present non- identical genotypes across all donors per individual pool. WGS files had duplicated reads removed with Picard’s MarkDuplicates 89 , and alleles were counted at the non-identical variants detailed before using Bam-readcount 90 . Alternative allele frequencies were estimated (only for the most abundant minor allele in the case of multiallelic variants, and only for point mutations) for all predefined variants covered by at least one WGS read. Finally, at each variant i the alternative allele frequency b is the sum across the total number of donors in the pool ( k ) of the product of each donor’s ( j ) minor allele dosage a and proportion w : Extending this for all donors and positions requires the genotype matrix A coded as minor allele dosages: Donor proportions were calculated by least-square regression: With constraints: and [0,1] for all i . Accuracy estimation of donor proportion deconvolution We simulated WGS pools with known donor proportions sequenced at a variety of mean depths to estimate the performance of the method. For that, we first estimated the mean depth and number of reads from individual donor WGS sequencing files, and calculated the total number of reads needed to achieve the target mean depth. Then, we calculated the number of reads needed per donor per mean depth to achieve a target donor proportion in a simulated pool, and sampled those reads randomly from individual donor WGS sequencing files before merging them for all donors into a simulated pool of reads. We finally counted the occurrence of each minor allele over the total number of reads, and estimated the donor proportions. We did this for a range of proportions per mean depth: equal proportions for ten donors, equal proportions (0.2) for five donors where other five were zero (ten donors in total), and at a range of proportions from 0.0005 to 0.39 from 19 donors to simulate a real-life scenario. We then estimated the absolute error (Supplementary Figure 1a) comparing the real proportions we sampled to those estimated by the method. The method’s maximum absolute error is 3% from a mean depth of 0.25X onwards, across the range of unequal proportions for a representative pool of 19 donors. The other two scenarios present a lower absolute error. Relative error was calculated as abs((real proportion - estimated proportion) / real proportion)*100 (Supplementary Figure 1b). Relative accuracy was measured as 100 - relative error. We observed that at higher (over 10%) real proportions the estimated proportions were always underestimated. Under 10% real proportions, these were always overestimated. This was taken into account to adjust the real donor proportions from our pools, as described in the next section. Estimation of differentiation efficiency, migration, and phagocytosis cellular phenotypes WGS data from 14–15 pooled experiments, comprising 243 donor iPSC lines (19– 24 per pool), was generated for the cellular phenotypes (differentiation efficiency, migration or phagocytosis). Donor proportions within pools were estimated using poodleR as described above. Then the cellular phenotypes were calculated as follows. First, the raw donor proportions were adjusted upwards or downwards, by adding or subtracting the absolute error estimated at the closest simulated proportion (based on the simulations described in the previous section). For the differentiation efficiency phenotype, log-fractions were calculated per line, pool and replicate as log(proportion at later stage /proportion at earlier stage) for macrophage precursors vs iPSC, old vs young macrophage precursors, and microglia vs macrophage precursors, the latter while matching the ages of the macrophage precursors from which they originated (see Supplementary Table 2). Then log-fractions were scaled per replicate. For the phagocytosis phenotype log-fractions were calculated per line, pool and replicate as log(proportion mCherry+ /proportion mCherry-), then scaling per replicate. Of note, these donor proportions were deconvoluted including the genotype of the SH5Y5Y line that is added to the pools during the phagocytosis assay. For the migration phenotype log-fractions were calculated as log(proportion bottom side of transwell / proportion top side of transwell) per line, pool and replicate in the C5a+ and C5a- conditions separately, then the C5a+ fractions were normalised to C5a- fractions per replicate, and finally they were scaled per replicate. Functional variant effect annotation of WES Whole exome sequencing (WES) files for 226 available lines present in our study were filtered as in Rouhani et al. 24 and only high-quality variants (with PASS filter) were retained. Functional variant effects were annotated using the variant effect predictor 91 (VEP, v. 111) for each line. We defined gene coordinates as those present in the Ensembl gene annotation (GRCh37, v. 102). After variant annotation, variant files were merged with bcftools v. 1.19 and coordinates lifted over with Picard v. 2.26.2. Variants were then classified as deleterious (69,223 variants across 16,348 genes) if they were annotated as frameshift, stop gained, transcript ablation, splice acceptor or splice donor; or if they were annotated as missense, start lost and protein altering and had CADD Phred scores (v. 1.7) over 15. All other missense variants were considered missense non-deleterious (simply “missense” – 62,022 variants across 15,523 genes). Test of deleterious and missense variant effects on cellular phenotypes Gene-level effects of missense and deleterious variants were tested for each phenotype (averaging the scaled phenotype per line across pools and replicates) using the sequence kernel association optimal test (SKAT-O) 92 , combining SKAT and burden tests. Empirical p-values were calculated by resampling the residuals 1000 times from the null model (using bootstrapping) while adjusting for covariates: sex, average of the minimum line proportion per replicate, and the first two genotype principal components (and the scaled differentiation from macrophage precursor to microglia in the case of the phagocytosis and migration phenotypes, to control for possible effects of differentiation on these two). Since SKAT-O does not provide a beta estimate on the phenotype, we calculated it through burden tests only for SKAT-O significant genes, by performing linear regression aggregating the variants per gene as done in Puigdevall et al. 3 and fitting a linear mixed model with the following formula using lmerTest in R: Scaled phenotype (phagocytosis or migration) ∼ deleterious burden + sex + minimum line proportion in replicate + scaled differentiation + genotype PC1 + genotype PC2 + (1|pool) For the scaled differentiation efficiency phenotype (which reflects the proliferation, survival and differentiation capacity of each line), we fitted the same formula without including scaled differentiation as a covariate, for the effects on iPSC to macrophage precursors, and young to aging macrophage precursors: Scaled differentiation efficiency ∼ deleterious burden + sex + minimum line proportion in replicate + genotype PC1 + genotype PC2 + (1|pool) For the effects on macrophage precursors to microglia the test was performed in the same way including an additional (1|treatment) term. P-values were corrected by Bonferroni adjustment for multiple testing. Significant genes after Bonferroni correction for every comparison had their deleterious burden scores aggregated together, separated by the directionality (positive or negative) of the beta, and tested using the same model above. AUC was calculated by building a binary classification of “dropouts” for lines with scaled differentiation efficiency more than 2 standard deviations below the mean, or of “takeovers” in the opposite direction, and then testing the aggregated burden of deleterious variants in significant genes in the relevant direction as predictors using roc() and auc() in the pROC package 93 . Significant differentiation genes as tested by SKAT-O were tested for enrichment in DepMap essential genes and macrophage survival genes from Covarrubias et al . 27 by Fisher’s exact test (Supplementary Table 14). Single cell RNA-seq processing and line identity deconvolution FASTQ files were processed with 10x Genomics Cell Ranger v.6.0.1 94 using default parameters. In preparation for line identity deconvolution in scRNA-seq, the reference genotype files were subset to exons using vcftools v.0.1.16 95 . Then variants were subset to those that had non-identical genotypes across all lines per individual pool. Single cell RNA-seq deconvolution was performed from the cellranger BAM outputs with Vireo v.0.5.8 96 and cellSNP-lite v.1.2.2 97 using default parameters. 3,404,020 cells were processed in Seurat v.5 98 (Supplementary Table 15), filtering out low quality cells (those with more than 10% mitochondrial genes, those that presented a number of UMI counts under the 5th percentile, doublets and cells with unassigned line identity; Supplementary Figure 1a). This left in total 2,341,497 cells remaining across all pools and treatments. Each pool per treatment was log normalised using NormalizeData() in Seurat, and the 2,000 most variable genes were selected for integration, which was performed on a “sketched” subset of 5000 cells, sampling cells without replacement while incorporating rare cell types with SketchData(). For the subset of cells, variable genes were selected again and their expression scaled, before dimensionality reduction and clustering using PCA, finding shared nearest neighbors (SNN), and performing modularity optimization based clustering with the Louvain algorithm. Finally, we performed dimensionality reduction by Uniform Manifold Approximation Projection (UMAP) using default parameters, providing the foundation for the integration across all pools using Harmony with IntegrateLayers() . The remaining cells were then projected onto the pre- integrated transcriptomic manifold using ProjectIntegration() and ProjectData() . The integrated UMAPs (per treatment and merged) were used for cell subtype annotation with individual marker genes and differential expression gene sets from Mancuso et al. (2024) 15 by using Seurat’s AddModuleScore() . Expression heatmaps and markers’ expression dotplots and UMAPs were plotted with SCpubr 99 and scCustomize 100 Polygenic risk score calculations Polygenic risk scores (PRS) were calculated with PRSice2. First, we excluded the APOE regions from the genotype file from the HipSci and IPMAR lines. Then PRS was calculated using the AD GWAS summary statistics (from Bellenguez et al. 8 ) and the LD information from the European ancestry (EUR) subset from 1000 Genomes, using a p-value threshold of 0.1 and leaving other options as default. We then calculated the PRS for the EUR subset from 1000 Genomes, using the same variants from the HipSci and IPMAR lines genotype file. We adjusted the raw scores for 5 genotype PCs for both datasets, and scaled the HipSci and IPMAR lines’ scores to those of 1000 Genomes EUR. We then added the scores for the APOE alleles with values B(APOE e2)= -0.47, B(APOE e3)=0, B(APOE e4)=1.12 as in Leonenko et al. 101 and scaled them to those of 1000G EUR. Finally we added the scaled polygenic and APOE scores to obtain the full PRS. The distribution of PRS in the HipSci, IPMAR, and 1000G EUR lines are shown in Supplementary Figure 12a-c. For microglia-specific PRS, regions 250Kb around the lead colocalizing loci between eQTLs and AD GWAS were taken (per treatment), as used as an input for PRSice2, both for data from our 261 donors and for that of 1000 Genomes EUR used in the scaling. All other steps to build the scores were identical. The distribution of microglial-specific PRS is shown in Supplementary Figure 12e. Differential expression analysis across treatments Raw UMI counts were aggregated by line, pool, and treatment replicate for the non-proliferative clusters. We then removed samples with < 100 cells (leaving 194 lines for analysis) and retained genes that were present in at least 30% of the donors with at least 1 count per million (CPM) (Supplementary figure 2a). We then transformed the data to log2-counts per million and calculated sample-level weights from the mean-variance relationship using voom() from limma 102 , while adjusting for pool random effects with duplicateCorrelation() . Finally a linear model was fit for the full dataset including donor effects and log10(number of cells) as covariates. The final fit was as follows: Sex and age effects were not fit as they were confounded with donor, and age was not confidently recorded in several cases. We then tested every treatment contrast (LPS vs untreated, IFN𝛄 vs untreated, and IFN𝛄 vs LPS) before moderated t-statistics were calculated with eBayes() . False discovery rate (FDR)-adjusted p-values were calculated per gene using the Benjamini- Hochberg procedure. Log2 fold changes in expression take the second element in the contrast as the baseline. The final numbers of differentially expressed genes are: View this table: View inline View popup Differential gene expression across PRS Raw UMI counts were aggregated by line, pool, and treatment replicate as in the differential expression analysis across treatments. Genes and lines were QC’d, counts were transformed and pool was accounted for in the same way. We fitted the following linear mixed model per treatment: We did not fit age for the reasons mentioned before, and because it is confounded with high PRS for those IPMAR donors with AD diagnosis. We then tested PRS effects, for which each gene’s coefficients are interpreted as log2 fold changes per unit increase in PRS. This resulted in the following number of significant (FDR-adjusted p-value 0) DE genes: View this table: View inline View popup Download powerpoint Empirical p-values were calculated for each gene per treatment by shuffling each line’s PRS 100 times, sampling the t-statistics per gene and calculating p- values by estimating how often the t-statistic from the real set of PRS was more extreme than that of the permuted set. Log2 fold changes in expression are referred to each unit change in PRS. Differential gene expression across phenotypes UMI counts were aggregated and lines filtered as in the previous section. Afterwards, we averaged across replicates the scaled phenotypes (migration and phagocytosis) and the minimum line proportion of the fraction to account for higher variance of the fractions from lines that present in smaller proportions in the pool. Finally the differential expression analysis was performed per treatment with limma , accounting for pool effects as in the previous section and fitting the following regression model: Genes with FDR-adjusted p-value 0 were considered significantly differentially expressed. Log2 fold changes in expression are referred to each unit change in the scaled phenotype. Enrichment analysis in differential expression and eQTL results Signed t-statistics from differential expression results were centered to ensure a normal distribution before enrichment analysis. For transcription factor activity inference, an univariate linear model was fitted per treatment using the decoupleR package and the CollecTRI network of weighted TF-target relationships. PROGENy pathways were fitted in the same way for the differentially expressed genes between treatments. Reactome hallmark pathways and candidate gene sets were tested with Gene Set Enrichment Analysis using the msigdbr and fgsea packages in R, and BH-adjusted p-values are reported. AD and PD candidate genes were defined by the prioritised gene sets from Schwartzentruber et al. 7 for AD, and from the top 2 genes from the Open Targets locus2gene score per associated locus for PD. Filtering genotypes and lines for eQTL and phenotype association analysis Genotype quality control included filtering out variants with imputation INFO < 0.7, HWE 10% of samples, and MAF < 5% in our samples. We also excluded lines with over 10% missing genotypes, which resulted in 6,405,518 variants used for eQTL and association analyses. We then filtered the 1000G genotype file in the same way and performed joint principal component analysis. The second line in each pair of clones (letw_5, lizq_3, zaie_1, romx_2, seru_7, qonc_2, sebn_4) and those lines that did not map close to European populations (boqx_2, garx_2, sojd_3, yoch_6) were excluded from the analysis (Supplementary Figure 12d). This left 250 donors for downstream analysis. eQTL mapping Single cell RNA-seq counts were aggregated into pseudobulks per donor, and grouping together all non-proliferative clusters. This decision was driven by two factors: first, the lack of clearly defined subtypes within each treatment condition; and second, the observation that the number of cells comprising a given cluster strongly correlates with the number of detectable differentially expressed genes and eQTL-associated genes (eGenes) 75 . We then performed several quality control filters for the pseudobulks: at the gene level, retaining genes that were present in at least 30% of the donors with at least 1 count per million (CPM), and log-normalising and scaling gene counts; and at the line level, removing those lines with < 100 cells per pseudobulk and removing the second individual in a pair of clones, and those lines for which we didn’t have genotype information or that mapped outside of European populations when projected within the PCA of 1000 Genomes data. This left 188 (IFN𝛄 and LPS) and 189 (untreated) lines for analysis in the Non-proliferative group. We included two genotype PCs and 5-120 expression PCs as covariates in the linear regression model. eQTL analysis was performed with tensorQTL v.1.0.9 104 including variants in the cis neighbourhood of each expressed gene (250kb up and down the transcription start site, 500kb total). To correct for the number of tests performed per gene we used tensorQTL in the map_cis mode, which permutes gene expression values a thousand times to obtain empirical false discovery rates (FDR) per gene-variant tested. We then used q-value correction on the genes tested to obtain a final set of significant (FDR<0.05) eGenes. The final set of expression PCs used maximized the number of significant eGenes: 63 for the untreated sample, 73 for IFN𝛄, and 84 for LPS. The degree of eQTL sharing was calculated by multivariate adaptive shrinkage with mashR , subsetting the results per treatment to the lead eQTL-eGene pairs, gathering these pairs from other treatments if not present, and estimating the lfsr and posterior means. eQTL were deemed shared if they had lfsr < 0.05 and if they had posterior means (betas) within 0.5 and the same sign. Genome-wide association analysis of cellular phenotypes Genotypes and lines were processed as detailed for eQTL analyses. To avoid inflation of false positives due to repeated line measurements and to limit the effect of the more variable phenotype fractions of less abundant donors, samples per line, pool and replicate, were filtered to those in which the smallest proportion of the pair in the fraction reached at least 1% abundance, and then the scaled fractions were averaged across replicates and pools. This left 133, 137, and 140 lines left for testing in the untreated, IFN𝞬, and LPS treatments, respectively, in the phagocytosis phenotype. In the migration phenotype, there were 147, 92, and 147 lines left for testing in the untreated, IFN𝞬, and LPS treatments, respectively. In the differentiation phenotype, there were 209, 229, and 230 lines left for testing in the untreated, IFN𝞬, and LPS treatments, respectively. Association tests were performed by fitting a linear regression model with the following formula: Mean phenotype across pools ∼ Alternative allele dosage + sex + number of pool replicates + genotypePC1-5 Colocalization analysis of eQTL and GWAS External disease GWAS and eQTL datasets were subset to common significant loci, those where a significant eQTL variant and a significant GWAS variant were within 500kb of each other. After subsetting to shared variants between the two datasets, colocalization was performed using coloc under the single causal variant assumption for regions with at least 50 shared variants. Locus plots were made using locuszoomr , taking LD information from 1000 Genomes GRCh38 high coverage via LDlink . Colocalizations involving eQTLs and association results from our iMGL phenotypes were performed in the same way but subsetting to common significant eQTL variants. Fine mapping of TREM2 AD GWAS locus We used susie_rss() from susieR for fine mapping of selected GWAS loci, taking LD information 1000 Genomes GRCh38 high coverage via LDlink and using the harmonised GWAS summary statistics. Non-default arguments for susie_rss() were: n=aggregate number of cases and controls in GWAS; L=1; coverage = 0.60; min_abs_corr = 0.0. Testing for enrichment in disease heritability near gene sets The enrichment of expressed gene sets for common variants associated with disease (GWAS variants) was tested using stratified LDSC 105 as implemented with CELLECT 106 . Because CELLECT uses as input a continuous representation of cell type expression between zero and one, we re-scaled the t-statistics of differential expression analyses min/max method from 0 (least significant) to 1 (most significant and largest effects), after sorting by either positive or negative sets, and taking those genes with opposite sign as zero. eGene lists were created in the same way by sorting and scaling their combined measure of local false sign rate (lfsr) and posterior mean (pm) estimates (-log10lsfr x pm) from mashR . Finally, CELLECT - LDSC leverages GWAS variants’ betas and LD structure and the top ranked genes to give a disease enrichment p-value for each treatment. Transcriptome-wide Mendelian Randomization analysis To estimate the causal effect of gene expression on microglial phenotypes we employed the TWMR method 35 , which estimates the effects of multiple exposures (eGenes) on each outcome (phenotype GWAS result) employing multiple instruments (independent eQTL variants). For each eGene (“focus gene”), per treatment, we selected the most significant eQTL variant and the top colocalizing variant (if present). We then selected eGenes for which any of those variants is an eQTL, and expanded the search for each eGene to other significant eQTL variants (nominal p-value < 10 - 8 ). We then pruned the variants to only those in low LD (R 2 <0.2, prioritizing those from the focus gene), and eGenes to those that were not highly correlated (R 2 <=0.4 across all variants). We then extracted the betas from the phenotype GWAS and calculated the causal effect of the gene expression on the phenotype solving the following: Where E is the n × k matrix of betas of n SNPs on k eGenes (as a single cis-eQTL often affects several nearby eGenes in a correlated way). G is the vector of phenotype GWAS beta values (of length n, for all variants tested). C is the pairwise LD matrix between the n variants, where LD is the R 2 taken from the EUR cohort from TOPMed 107 . Despite using overlapping samples for exposures and outcomes, strong instrument selection mitigates the inflation of TWMR estimates 108 , and thus we selected genes with F parameter values > 10. Association of PRS with microglial phenotype Full PRS and microglia-specific PRS were related to microglial phenotypes by linear regression. Phenotypes were processed as for the genome-wide association, and a regression model was fitted for all remaining donors with the following formula: Mean phenotype across pools ∼ PRS + sex + number of pool replicates + mean of minimum donor proportion across pools Where PRS can be the full PRS including the polygenic component and the APOE component, or the last two on their own. Genotype PCs were not included as covariates as PRS were already adjusted for these. Funding This work was funded by OpenTargets (OTAR2065) Code and data availability The raw flow cytometry, western blotting, processed sequencing data and code to replicate the analyses done in this publication is available on Zenodo (10.5281/zenodo.16684932). The donor deconvolution R package poodleR is available on GitHub. All raw sequencing data is available under study EGAS00001004854 (scRNA- seq) and EGAS00001005143 (DNA-seq), comprising scRNA-seq and WGS data from all pools. Donor and line genotypes are available at www.hipsci.org and the genotype of SH5Y5Y is available at https://systemsbiology.uni.lu/shsy5y/ . Contributions A.B., G.T. and S.A.C. conceived the study. S.W., Y.C., J.S. and J.M. performed the laboratory experiments. M.P-A., D.G-P. and P.K. performed the bioinformatics data analysis. S.W. performed and analysed the CRISPR screen. G.T., A.B., S.A.C., K.A. N.P. and D.W. supervised the analysis. M.P-A., S.W., G.T., A.B., D.G-P. and J.C. wrote the manuscript. All authors interpreted the results and provided critical comments on the manuscript. Conflicts of interest A.B. is a founder of and consultant for Ensocell therapeutics. Acknowledgements DRICU iPSC lines were generated pursuant to funding received by Cardiff from the UK Dementia Research Institute (award number UK DRI-3201) through UK DRI Ltd, principally funded by the Medical Research Council, and The Moondance Foundation), and are provided by Cardiff to the Recipient Organisation with the approval of the UK DRI IPMAR Research Group. The authors wish to acknowledge the support of the Cytometry Core Facility and Scientific Operations sequencing facilities at Wellcome Sanger Institute and members of the Bassett and Trynka labs for helpful discussions. Funder Information Declared Open Targets, https://ror.org/000bp7q73 , OTAR2065 Wellcome Trust, https://ror.org/029chgv08 , 220540/Z/20/A Footnotes ↵ * Joint first authors References 1. ↵ Alasoo , K. et al. Shared genetic effects on chromatin and gene expression indicate a role for enhancer priming in immune response . Nature Genetics 50 , 424 – 431 ( 2018 ). OpenUrl CrossRef PubMed 2. ↵ Panousis , N. I. et al. Gene expression QTL mapping in stimulated iPSC- derived macrophages provides insights into common complex diseases . bioRxiv 2023.05.29.542425 ( 2023 ) doi: 10.1101/2023.05.29.542425 . OpenUrl Abstract / FREE Full Text 3. ↵ Puigdevall , P. , Jerber , J. , Danecek , P. , Castellano , S. & Kilpinen , H . Somatic mutations alter the differentiation outcomes of iPSC-derived neurons . Cell Genom 3 , 100280 ( 2023 ). 4. ↵ Antón-Bolaños , N. et al. Brain Chimeroids reveal individual susceptibility to neurotoxic triggers . Nature 631 , 142 – 149 ( 2024 ). OpenUrl CrossRef PubMed 5. ↵ Schwartzentruber , J. et al. Molecular and functional variation in iPSC-derived sensory neurons . Nat Genet 50 , 54 – 61 ( 2018 ). OpenUrl CrossRef PubMed 6. ↵ Farbehi , N. et al. Integrating population genetics, stem cell biology and cellular genomics to study complex human diseases . Nat Genet 56 , 758 – 766 ( 2024 ). OpenUrl CrossRef PubMed 7. ↵ Schwartzentruber , J. et al. Genome-wide meta-analysis, fine-mapping and integrative prioritization implicate new Alzheimer’s disease risk genes . Nat Genet 53 , 392 – 402 ( 2021 ). OpenUrl CrossRef PubMed 8. ↵ Bellenguez , C. et al. New insights into the genetic etiology of Alzheimer’s disease and related dementias . Nat Genet 54 , 412 – 436 ( 2022 ). OpenUrl CrossRef PubMed 9. Jonsson , T. et al. Variant of TREM2 associated with the risk of Alzheimer’s disease . N Engl J Med 368 , 107 – 116 ( 2013 ). OpenUrl CrossRef PubMed Web of Science 10. ↵ Chen , X. et al. Microglia-mediated T cell infiltration drives neurodegeneration in tauopathy . Nature 615 , 668 – 677 ( 2023 ). OpenUrl CrossRef PubMed 11. ↵ Li , Y. et al. TREM2: Potential therapeutic targeting of microglia for Alzheimer’s disease . Biomed Pharmacother 165 , 115218 ( 2023 ). OpenUrl PubMed 12. ↵ Kilpinen , H. et al. Common genetic variation drives molecular heterogeneity in human iPSCs . Nature 546 , 370 – 375 ( 2017 ). OpenUrl CrossRef PubMed 13. ↵ Emily Maguireo , Jincy Winston , Sarah H Ellwood , Rachel O’Donoghue , Bethany Shaw , Atahualpa Castillo Morales , Samuel Keat , Alexandra Evans , Rachel Marshall , Lauren Luckcuck , Laura Brown , Elisa Salis , Ganna Leonenko Nicola Denning , EADB consortium & Nicholas D Allen , Valentina Escott-Price , Caleb Webber , Philip R Taylor , Rebecca Sims , Sally A Cowley , Julie Williams , Sarah M Carpanini , Hazel Hall-Roberts . Modeling common Alzheimer’s disease with high and low polygenic risk in human iPSC: A large-scale research resource . Stem Cell Reports 102570 , ( 2025 ). 14. ↵ Washer , S. J. et al. Single-cell transcriptomics defines an improved, validated monoculture protocol for differentiation of human iPSC to microglia . Sci Rep 12 , 19454 ( 2022 ). OpenUrl CrossRef PubMed 15. ↵ Mancuso , R. et al. Xenografted human microglia display diverse transcriptomic states in response to Alzheimer’s disease-related amyloid-β pathology . Nat Neurosci 27 , 886 – 900 ( 2024 ). OpenUrl CrossRef PubMed 16. ↵ Tirosh , I. et al. Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq . Science 352 , 189 – 196 ( 2016 ). OpenUrl Abstract / FREE Full Text 17. ↵ Lively , S. & Schlichter , L. C . Microglia Responses to Pro-inflammatory Stimuli (LPS, IFNγ+TNFα) and Reprogramming by Resolving Cytokines (IL- 4, IL-10) . Front Cell Neurosci 12, 215 ( 2018 ). 18. ↵ Conserved and cell type-specific transcriptional responses to IFN-γ in the ventral midbrain . Brain, Behavior, and Immunity 111, 277 – 291 ( 2023 ). 19. ↵ Oh , S.-M. et al. Combined Nurr1 and Foxa2 roles in the therapy of Parkinson’s disease . EMBO Mol Med 7 , 510 – 525 ( 2015 ). OpenUrl Abstract / FREE Full Text 20. ↵ Chen , S. et al. Macrophages in immunoregulation and therapeutics . Signal Transduct Target Ther 8 , 207 ( 2023 ). 21. ↵ Mitchell , J. M. et al. Mapping genetic effects on cellular phenotypes with ‘cell villages’ . bioRxiv 2020.06.29.174383 ( 2020 ) doi: 10.1101/2020.06.29.174383 . OpenUrl Abstract / FREE Full Text 22. ↵ Wells , M. F. et al. Natural variation in gene expression and viral susceptibility revealed by neural progenitor cell villages . Cell Stem Cell 30 , 312 – 332 .e13 ( 2023 ). OpenUrl CrossRef PubMed 23. ↵ Sinha , S. , Bheemsetty , V. A. & Inamdar , M. S . A double helical motif in OCIAD2 is essential for its localization, interactions and STAT3 activation . Sci Rep 8 , 7362 ( 2018 ). OpenUrl CrossRef PubMed 24. ↵ Rouhani , F. J. et al. Substantial somatic genomic variation and selection for BCOR mutations in human induced pluripotent stem cells . Nature Genetics 54 , 1406 – 1416 ( 2022 ). OpenUrl CrossRef PubMed 25. ↵ Sabir , S. R. , Sahota , N. K. , Jones , G. D. D. & Fry , A. M . Loss of Nek11 Prevents G2/M Arrest and Promotes Cell Death in HCT116 Colorectal Cancer Cells Exposed to Therapeutic DNA Damaging Agents . PLoS One 10 , e0140975 ( 2015 ). OpenUrl PubMed 26. ↵ Chen , M.-H. et al. Trans-ethnic and Ancestry-Specific Blood-Cell Genetics in 746,667 Individuals from 5 Global Populations . Cell 182 , 1198 – 1213 .e14 ( 2020 ). OpenUrl CrossRef PubMed 27. ↵ Covarrubias , S. et al. High-Throughput CRISPR Screening Identifies Genes Involved in Macrophage Viability and Inflammatory Pathways . Cell Rep 33 , 108541 ( 2020 ). 28. ↵ Tegtmeyer , M. et al. High-dimensional phenotyping to define the genetic basis of cellular morphology . Nature Communications 15 , 1 – 12 ( 2024 ). OpenUrl PubMed 29. ↵ Glastonbury , C. A. et al. Machine Learning based histology phenotyping to investigate the epidemiologic and genetic basis of adipocyte morphology and cardiometabolic traits . PLOS Computational Biology 16 , e1008044 ( 2020 ). OpenUrl 30. ↵ Ma , W. et al. The intricate role of CCL5/CCR5 axis in Alzheimer disease . J Neuropathol Exp Neurol 82 , 894 – 900 ( 2023 ). OpenUrl CrossRef PubMed 31. ↵ Zhao , Y. et al. Interleukin-37 reduces inflammation and impairs phagocytosis of platelets in immune thrombocytopenia (ITP) . Cytokine 125 , 154853 ( 2020 ). 32. ↵ Fehrmann , R. S. N. et al. Trans-eQTLs reveal that independent genetic variants associated with a complex phenotype converge on intermediate genes, with a major role for the HLA . PLoS Genet 7 , e1002197 ( 2011 ). OpenUrl CrossRef PubMed 33. ↵ Gusev , A. et al. Integrative approaches for large-scale transcriptome-wide association studies . Nature Genetics 48 , 245 – 252 ( 2016 ). OpenUrl CrossRef PubMed 34. ↵ Lappalainen , T. et al. Transcriptome and genome sequencing uncovers functional variation in humans . Nature 501 , 506 – 511 ( 2013 ). OpenUrl CrossRef PubMed Web of Science 35. ↵ Porcu , E. et al. Mendelian randomization integrating GWAS and eQTL data reveals genetic determinants of complex and clinical traits . Nat Commun 10 , 3300 ( 2019 ). OpenUrl CrossRef PubMed 36. ↵ Xiang , X. & Qiu , R . Cargo-Mediated Activation of Cytoplasmic Dynein . Front Cell Dev Biol 8 , 598952 ( 2020 ). 37. ↵ Ding , L. et al. Peroxisomal β-oxidation acts as a sensor for intracellular fatty acids and regulates lipolysis . Nat Metab 3 , 1648 – 1661 ( 2021 ). OpenUrl PubMed 38. ↵ The key roles of reactive oxygen species in microglial inflammatory activation: Regulation by endogenous antioxidant system and exogenous sulfur-containing compounds . European Journal of Pharmacology 956 , 175966 ( 2023 ). OpenUrl CrossRef PubMed 39. ↵ Ohtonen , S. et al. Human iPSC-derived microglia carrying the LRRK2- G2019S mutation show a Parkinson’s disease related transcriptional profile and function . Sci Rep 13 , 22118 ( 2023 ). OpenUrl PubMed 40. ↵ Chen , R. & Zhang , Y . EPDR1 correlates with immune cell infiltration in hepatocellular carcinoma and can be used as a prognostic biomarker . J Cell Mol Med 24 , 12107 – 12118 ( 2020 ). OpenUrl PubMed 41. Lu , C. & Zhou , J . Analysis of the causal relationship between five chosen factors and early-onset Alzheimer’s disease: A Mendelian randomization study . J Alzheimers Dis 103 , 1135 – 1149 ( 2025 ). OpenUrl PubMed 42. Dreyer , C. A. , VanderVorst , K. & Carraway , K. L ., 3rd. Vangl as a Master Scaffold for Wnt/Planar Cell Polarity Signaling in Development and Disease . Front Cell Dev Biol 10 , 887100 ( 2022 ). OpenUrl PubMed 43. Sun , X. et al. Inhibition of VRK1 suppresses proliferation and migration of vascular smooth muscle cells and intima hyperplasia after injury via mTORC1/β-catenin axis . BMB Rep 55 , 244 – 249 ( 2022 ). OpenUrl PubMed 44. The evolving spectrum of LAMA2 related congenital muscular dystrophy (MDC1)-Case series and review of literature . Brain Disorders 17 , 100181 ( 2025 ). OpenUrl 45. Lu , J.-W. et al. Overexpression of endothelin 1 triggers hepatocarcinogenesis in zebrafish and promotes cell proliferation and migration through the AKT pathway . PLoS One 9 , e85318 ( 2014 ). OpenUrl CrossRef PubMed 46. Dai , L.-P. et al. SPTBN1 attenuates rheumatoid arthritis synovial cell proliferation, invasion, migration and inflammatory response by binding to PIK3R2 . Immun Inflamm Dis 10 , e724 ( 2022 ). OpenUrl 47. Chiuso , F. et al. Ubiquitylation of BBSome is required for ciliary assembly and signaling . EMBO Rep 24 , e55571 ( 2023 ). OpenUrl CrossRef PubMed 48. ↵ Yang , W. , Chen , W. , Ding , M. , Guo , H. & Ji , C . 326P GCLC inhibits lung metastasis of renal cell carcinoma by regulating glutathione metabolic pathway . Ann. Oncol . 35 , S1528 ( 2024 ). 49. ↵ Mishra , S. et al. The Alzheimer’s Disease Gene SORL1 Regulates Lysosome Function in Human Microglia . Glia 73 , ( 2025 ). 50. ↵ Mathews , P. M. & Levy , E . Cystatin C in aging and in Alzheimer’s disease . Ageing Res Rev 32 , 38 – 50 ( 2016 ). OpenUrl CrossRef PubMed 51. ↵ Wang , Z. et al. Deep post-GWAS analysis identifies potential risk genes and risk variants for Alzheimer’s disease, providing new insights into its disease mechanisms . Sci Rep 11 , 20511 ( 2021 ). 52. ↵ PRMT6 deficiency or inhibition alleviates neuropathic pain by decreasing glycolysis and inflammation in microglia. Brain , Behavior, and Immunity 118 , 101 – 114 ( 2024 ). OpenUrl 53. ↵ Palmer , J. C. , Barker , R. , Kehoe , P. G. & Love , S . Endothelin-1 is elevated in Alzheimer’s disease and upregulated by amyloid-β . J Alzheimers Dis 29 , 853 – 861 ( 2012 ). OpenUrl PubMed 54. ↵ Abdul , Y. , Jamil , S. , He , L. , Li , W. & Ergul , A . Endothelin-1 (ET-1) promotes a proinflammatory microglia phenotype in diabetic conditions . Can J Physiol Pharmacol 98 , 596 – 603 ( 2020 ). OpenUrl PubMed 55. ↵ de Vries , D. H. et al. Integrating GWAS with bulk and single-cell RNA- sequencing reveals a role for LY86 in the anti-Candida host response . PLOS Pathogens 16 , e1008408 ( 2020 ). OpenUrl PubMed 56. ↵ Zhao , Y. & Xu , H . Microglial lactate metabolism as a potential therapeutic target for Alzheimer’s disease . Molecular Neurodegeneration 17 , 1 – 3 ( 2022 ). OpenUrl PubMed 57. ↵ Cai , Y. , Liu , J. , Wang , B. , Sun , M. & Yang , H . Microglia in the Neuroinflammatory Pathogenesis of Alzheimer’s Disease and Related Therapeutic Targets . Front. Immunol . 13 , 856376 ( 2022 ). 58. ↵ Buniello , A. et al. Open Targets Platform: facilitating therapeutic hypotheses building in drug discovery . Nucleic Acids Res 53 , D1467 – D1475 ( 2025 ). OpenUrl CrossRef PubMed 59. ↵ Fujita , M. et al. Cell subtype-specific effects of genetic variation in the Alzheimer’s disease brain . Nat Genet 56 , 605 – 614 ( 2024 ). OpenUrl CrossRef PubMed 60. ↵ Young , A. M. H. et al. A map of transcriptional heterogeneity and regulatory variation in human microglia . Nature Genetics 53 , 861 – 868 ( 2021 ). OpenUrl CrossRef PubMed 61. ↵ Rathore , N. et al. Paired Immunoglobulin-like Type 2 Receptor Alpha G78R variant alters ligand binding and confers protection to Alzheimer’s disease . PLOS Genetics 14 , e1007427 ( 2018 ). OpenUrl 62. ↵ Weerakkody , T. et al. PILRA regulates microglial immunometabolism to reduce disease pathology as a candidate therapeutic target for Alzheimer’s disease . Research square (preprint ) ( 2025 ) doi: 10.21203/rs.3.rs-3954863/v2 . OpenUrl CrossRef 63. ↵ Audrain , M. et al. Reactive or transgenic increase in microglial TYROBP reveals a TREM2-independent TYROBP-APOE link in wild-type and Alzheimer’s-related mice . Alzheimers Dement 17 , 149 – 163 ( 2021 ). OpenUrl CrossRef PubMed 64. ↵ Reus , L. M. et al. Connecting dementia risk loci to the CSF proteome identifies pathophysiological leads for dementia . Brain 147 , 3522 – 3533 ( 2024 ). OpenUrl PubMed 65. ↵ Filipello , F. et al. Defects in lysosomal function and lipid metabolism in human microglia harboring a TREM2 loss of function mutation . Acta Neuropathol 145 , 749 – 772 ( 2023 ). OpenUrl CrossRef PubMed 66. ↵ Jay , T. R. , von Saucken , V. E. & Landreth , G. E . TREM2 in Neurodegenerative Diseases . Mol Neurodegener 12 , 56 ( 2017 ). 67. ↵ Hall-Roberts , H. et al. TREM2 Alzheimer’s variant R47H causes similar transcriptional dysregulation to knockout, yet only subtle functional phenotypes in human iPSC-derived macrophages . Alzheimers Res Ther 12 , 151 ( 2020 ). 68. Claes , C. et al. Human stem cell-derived monocytes and microglia-like cells reveal impaired amyloid plaque clearance upon heterozygous or homozygous loss of TREM2 . Alzheimers Dement 15 , 453 – 464 ( 2019 ). OpenUrl PubMed 69. ↵ Reich , M. et al. Alzheimer’s Risk Gene TREM2 Determines Functional Properties of New Type of Human iPSC-Derived Microglia . Front Immunol 11 , 617860 ( 2020 ). OpenUrl PubMed 70. ↵ Washer , S. et al. Protocol for pooled FACS based CRISPR knockout screening in human iPSC-derived microglia . STAR Protocols (accepted ) ( 2025 ). 71. ↵ Shi , Y. & Holtzman , D. M . Interplay between innate immunity and Alzheimer disease: APOE and TREM2 in the spotlight . Nat Rev Immunol 18 , 759 – 772 ( 2018 ). OpenUrl CrossRef PubMed 72. Mai , Z. et al. Molecular recognition of the interaction between ApoE and the TREM2 protein . Transl Neurosci 13 , 93 – 103 ( 2022 ). OpenUrl CrossRef PubMed 73. ↵ Kober , D. L. et al. Functional insights from biophysical study of TREM2 interactions with apoE and Aβ . Alzheimers Dement ( 2020 ) doi: 10.1002/alz.12194 . OpenUrl CrossRef 74. ↵ Krasemann , S. et al. The TREM2-APOE Pathway Drives the Transcriptional Phenotype of Dysfunctional Microglia in Neurodegenerative Diseases . Immunity 47 , 566 – 581 .e9 ( 2017 ). OpenUrl CrossRef PubMed 75. ↵ Soskic , B. et al. Immune disease risk variants regulate gene expression dynamics during CD4 T cell activation . Nat Genet 54 , 817 – 826 ( 2022 ). OpenUrl CrossRef PubMed 76. ↵ Navarro-Guerrero , E. et al. Genome-wide CRISPR/Cas9-knockout in human induced Pluripotent Stem Cell (iPSC)-derived macrophages . Sci Rep 11 , 4245 ( 2021 ). OpenUrl CrossRef PubMed 77. ↵ Gonçalves , E. et al. Minimal genome-wide human CRISPR-Cas9 library . Genome Biol 22 , 40 ( 2021 ). 78. ↵ Thomas Burgold , Emre Karakoc , Emanuel Gonçalves , Inigo Barrio- Hernandez , Lisa Dwane , Romina Oliveira Silva , Emily Souster , Mamta Sharma , Alexandra Beck , Gene Ching Chiek Koh, Lykourgos-Panagiotis Zalmas, Mathew J Garnett , Andrew Bassett. Genetic interaction library screening with a next-generation dual guide CRISPR system. biorXiv ( 2025 ) doi: 10.1101/2024.03.28.587052 . OpenUrl Abstract / FREE Full Text 79. ↵ Li , W. et al. MAGeCK enables robust identification of essential genes from genome-scale CRISPR/Cas9 knockout screens . Genome Biol 15 , 554 ( 2014 ). 80. ↵ Wang , B. et al. Integrative analysis of pooled CRISPR genetic screens using MAGeCKFlute . Nat Protoc 14 , 756 – 780 ( 2019 ). OpenUrl CrossRef PubMed 81. ↵ Vandermeulen , L. et al. Regulation of human microglial gene expression and function via RNAase-H active antisense oligonucleotides in vivo in Alzheimer’s disease . Mol Neurodegener 19 , 37 ( 2024 ). 82. ↵ Ritz , C. , Jensen , S. M. , Gerhard , D. & Streibig , J. C . Dose-Response Analysis Using R . ( CRC Press , 2019 ). 83. ↵ Zhao , H. et al. CrossMap: a versatile tool for coordinate conversion between genome assemblies . Bioinformatics 30 , 1006 – 1007 ( 2014 ). OpenUrl CrossRef PubMed Web of Science 84. ↵ High-coverage whole-genome sequencing of the expanded 1000 Genomes Project cohort including 602 trios . Cell 185 , 3426 – 3440 .e19 ( 2022 ). OpenUrl CrossRef PubMed 85. ↵ Deelen , P. et al. Genotype harmonizer: automatic strand alignment and format conversion for genotype data integration . BMC Res Notes 7 , 901 ( 2014 ). 86. ↵ Danecek , P. et al. Twelve years of SAMtools and BCFtools . Gigascience 10 , ( 2021 ). 87. ↵ Loh , P.-R. , Palamara , P. F. & Price , A. L . Fast and accurate long-range phasing in a UK Biobank cohort . Nat Genet 48 , 811 – 816 ( 2016 ). OpenUrl CrossRef PubMed 88. ↵ Mosca , M. J. & Cho , H . Reconstruction of private genomes through reference-based genotype imputation . Genome Biol 24 , 271 ( 2023 ). 89. ↵ Picard . https://broadinstitute.github.io/picard/ . 90. ↵ Khanna , A. et al. Bam-readcount - rapid generation of basepair-resolution sequence metrics . Journal of Open Source Software 7 , 3722 ( 2022 ). OpenUrl CrossRef 91. ↵ McLaren , W. et al. The Ensembl Variant Effect Predictor . Genome Biol 17 , 122 ( 2016 ). 92. ↵ Lee , S. , Wu , M. C. & Lin , X . Optimal tests for rare variant effects in sequencing association studies . Biostatistics 13 , 762 – 775 ( 2012 ). OpenUrl CrossRef PubMed Web of Science 93. ↵ Robin , X. et al. pROC: an open-source package for R and S+ to analyze and compare ROC curves . BMC Bioinformatics 12 , 77 ( 2011 ). 94. ↵ Zheng , G. X. Y. et al. Massively parallel digital transcriptional profiling of single cells . Nat Commun 8 , 14049 ( 2017 ). 95. ↵ Danecek , P. et al. The variant call format and VCFtools . Bioinformatics 27 , 2156 – 2158 ( 2011 ). OpenUrl CrossRef PubMed Web of Science 96. ↵ Huang , Y. , McCarthy , D. J. & Stegle , O . Vireo: Bayesian demultiplexing of pooled single-cell RNA-seq data without genotype reference . Genome Biology 20 , 1 – 12 ( 2019 ). OpenUrl CrossRef PubMed 97. ↵ Huang , X. & Huang , Y . Cellsnp-lite: an efficient tool for genotyping single cells . Bioinformatics 37 , 4569 – 4571 ( 2021 ). OpenUrl CrossRef PubMed 98. ↵ Hao , Y. et al. Dictionary learning for integrative, multimodal and scalable single-cell analysis . Nat Biotechnol 42 , 293 – 304 ( 2024 ). OpenUrl CrossRef PubMed 99. ↵ Blanco-Carmona , E . Generating publication ready visualizations for Single Cell transcriptomics using SCpubr . bioRxiv 2022.02.28.482303 ( 2022 ) doi: 10.1101/2022.02.28.482303 . OpenUrl Abstract / FREE Full Text 100. ↵ samuel-marsh/scCustomize: Version 0.6.0. doi: 10.5281/zenodo.5706431 . OpenUrl CrossRef 101. ↵ Leonenko , G. et al. Identifying individuals with high risk of Alzheimer’s disease using polygenic risk scores . Nature Communications 12 , 1 – 10 ( 2021 ). OpenUrl CrossRef PubMed 102. ↵ Ritchie , M. E. et al. limma powers differential expression analyses for RNA- sequencing and microarray studies . Nucleic Acids Res 43 , e47 ( 2015 ). OpenUrl CrossRef PubMed 103. STRING: functional protein association networks . https://string-db.org/ . 104. ↵ Van Allen Gad Getz , A. T.-W. F. A. N. J. H. S. G. S. A. J. K. K. A. E. M. Scaling computational genomics to millions of individuals with GPUs . Genome Biol . 20 , ( 2019 ). 105. ↵ Finucane , H. K. et al. Partitioning heritability by functional annotation using genome-wide association summary statistics . Nature Genetics 47 , 1228 – 1235 ( 2015 ). OpenUrl CrossRef PubMed 106. ↵ Timshel , P. N. , Thompson , J. J. & Pers , T. H . Genetic mapping of etiologic brain cell types for obesity . Elife 9 , ( 2020 ). 107. ↵ Taliun , D. et al. Sequencing of 53,831 diverse genomes from the NHLBI TOPMed Program . Nature 590 , 290 – 299 ( 2021 ). OpenUrl CrossRef PubMed 108. ↵ Jiang , T. , Gill , D. , Butterworth , A. S. & Burgess , S . An empirical investigation into the impact of winner’s curse on estimates from Mendelian randomization . Int J Epidemiol 52 , 1209 – 1219 ( 2023 ). OpenUrl CrossRef PubMed View the discussion thread. Back to top Previous Next Posted August 19, 2025. Download PDF Supplementary Material Email Thank you for your interest in spreading the word about bioRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Integrated QTL mapping and CRISPR screening in pooled iPSC-derived microglia reveals genetic drivers of neurodegenerative risk Message Subject (Your Name) has forwarded a page to you from bioRxiv Message Body (Your Name) thought you would like to see this page from the bioRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Integrated QTL mapping and CRISPR screening in pooled iPSC-derived microglia reveals genetic drivers of neurodegenerative risk Marta Perez-Alcantara , Sam Washer , Yixi Chen , Juliette Steer , Daianna Gonzalez-Padilla , Joe McWilliam , David Willé , Nikos Panousis , Peep Kolberg , Elena Navarro Guerrero , Kaur Alasoo , Hazel Hall-Roberts , Julie Williams , Sally A. Cowley , Gosia Trynka , Andrew Bassett bioRxiv 2025.08.18.670767; doi: https://doi.org/10.1101/2025.08.18.670767 Share This Article: Copy Citation Tools Integrated QTL mapping and CRISPR screening in pooled iPSC-derived microglia reveals genetic drivers of neurodegenerative risk Marta Perez-Alcantara , Sam Washer , Yixi Chen , Juliette Steer , Daianna Gonzalez-Padilla , Joe McWilliam , David Willé , Nikos Panousis , Peep Kolberg , Elena Navarro Guerrero , Kaur Alasoo , Hazel Hall-Roberts , Julie Williams , Sally A. Cowley , Gosia Trynka , Andrew Bassett bioRxiv 2025.08.18.670767; doi: https://doi.org/10.1101/2025.08.18.670767 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Genomics Subject Areas All Articles Animal Behavior and Cognition (7616) Biochemistry (17625) Bioengineering (13852) Bioinformatics (41825) Biophysics (21397) Cancer Biology (18524) Cell Biology (25417) Clinical Trials (138) Developmental Biology (13350) Ecology (19858) Epidemiology (2067) Evolutionary Biology (24277) Genetics (15581) Genomics (22459) Immunology (17698) Microbiology (40278) Molecular Biology (17134) Neuroscience (88400) Paleontology (666) Pathology (2823) Pharmacology and Toxicology (4812) Physiology (7632) Plant Biology (15106) Scientific Communication and Education (2042) Synthetic Biology (4281) Systems Biology (9807) Zoology (2266)