A genome-scale single cell CRISPRi map oftransgene regulation across human pluripotent stem cell lines

Population-scale resources of genetic, molecular, and cellular information form the basis for understanding human genomes, charting the heritable basis of disease, and tracing the effects of mutations. Pooled perturbation assays applied to cellular models, probing the effect of many perturbations coupled with an scRNA-seq readout (Perturb-seq), are especially potent references for interpreting disease-linked mutations or gene expression changes. However, the utility of existing maps has been limited by the comprehensiveness of perturbations possible, and the relevance of their cell line context. Here, we present the first genome-scale CRISPR interference (CRISPRi) perturbation map with single-cell RNA sequencing readout across many human genetic backgrounds in human pluripotent cells. T o do so, we establish large-scale CRISPRi screening in human induced pluripotent stem cells from healthy donors, using over 20,000 guide RNAs to target 7,226 genes across 34 cell lines from 26 genetic backgrounds, and gather expression data from nearly 2 million cells. We comprehensively map trans expression changes induced by the target knockdowns, which complement co-expression patterns in unperturbed cells and facilitate the functional annotation of target genes to biological processes and complexes. Consistency of targeting protein complex members point to protein complexes as a nexus for aggregating transcriptional variation, revealing novel interaction partners. We characterise variation in perturbation effects across donors, with expression quantitative trait loci linked to higher genetic modulation of perturbation effects but overall low replication of trans effects due to knockdown of their corresponding cis regulators. This study pioneers population-scale CRISPR perturbations with single cell readouts that will fuel foundation models for the future of effective modulation of cellular disease phenotypes. 1 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint

Cellular models allow interrogating disease phenotypes and basic processes in controlled experiments. Undifferentiated induced pluripotent stem cells (iPSCs) are an established system for modelling human development and disease 1–3. These cells are generated by transforming easy-to-acquire cell types, such as human fibroblasts, into an embryonic-like state, where cells have the capacity to differentiate into the three germ layers. Given the cell-type specificity of many diseases and their inherent ability to self-renew and differentiate, iPSCs represent a powerful tool for studying human variation in cell types that are otherwise difficult to obtain. Consequently, many efforts have been made in identifying the effects of common variation on molecular phenotypes in these cells, e.g. expression quantitative trait locus (eQTL) of common, rare and structural variants 1–3. While this identified thousands of loci altering expression levels of nearby genes in cis, the regulatory landscape underlying the downstream consequences on cellular pathways and function are to a large extent still poorly understood. In particular, existing studies on trans effects, in particular in human pluripotent cells, are underpowered due to insufficient genomic resources and small effect sizes of genetic variation in the natural population 3. T o complement natural genetic variation, CRISPR has recently emerged as a powerful tool for gene editing, silencing or activation in a targeted and cost-efficient manner. By combining pooled CRISPR-based screening with single-cell gene expression as a read-out, we are now able to study molecular consequences of genetic perturbations on candidate genes 4–8, identify their downstream targets and infer their biological function and relation to disease. Where non-interventional single-cell expression studies can identify co-expressed genes, inducing a genomic perturbation provides directionality on the nature of a regulatory relationship. T o date, CRISPR-based screens have mainly been used to elucidate gene function9 and to identify regulatory networks 5,10. However, resources that map perturbation responses at a genome-scale remain scarce and despite their relevance for understanding human disease and development comprehensive screens have not yet been conducted in iPSCs. Further, while it is well-known that genetic background matters both for mutation impact on disease risk 11, as well as its effect in a functional assay 1,3,12, no efforts have been made to account for different genetic backgrounds in existing studies so far 13–15 . Here, we present results from a genome-scale CRISPRi screen conducted in iPSCs derived from tens of individuals with scRNA-seq readout. By combining two distinct experimental designs, leveraging the number of perturbations in our dataset, as well as a diverse set of genetic backgrounds, we show that this population genomic perturbation approach with single cell phenotyping can be used to recover known regulatory networks, study cell type-specific biology and compare effects of synthetic dosage change and natural genetic variation. This work represents the second genome-scale CRISPRi screen with scRNA-seq readout and is the only study that accounts for the role of genetic background on gene function, establishing a necessary and fundamental building block for conducting population-scale genetic analyses using genome perturbation techniques. 2 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint

Measuring knockdown effects across multiple donors at scale First, we set out to chart the landscape of genetic perturbation effects in multiple individuals by performing gene knockdowns using CRISPRi coupled with a single-cell RNA-seq readout (Figure 1A, Methods). T o do so, we targeted 7,226 genes in 34 high-quality iPS cell lines derived from 26 distinct healthy donors. The targets were selected to cover genes whose knockdown in iPSCs or cancer cell lines 16 exhibited growth defects (2,264 genes and 4,594 genes, respectively, of which 1,725 genes exhibited fitness effects in both) or are highly expressed in iPSCs (2,093 additional genes; Supplementary Table ST1-1, Figure S1A, Methods). The targeting library included three guide RNAs (gRNAs) per gene from the Dolcetto library 17, as well as 40 non-targeting guides. Following quality control, demultiplexing of cell lines from genotypes, and gRNA assignment (Methods) we obtained 219,206 cells assigned to a source cell line and a targeting guide (median of 8 cells per gRNA and 25 cells per gene; Methods, Figure S1B, Supplementary Table ST1-2), as well as 499,998 control cells assigned to a donor, but either no guide or a non-targeting gRNA. We considered target genes with a minimum coverage of 10 cells (6,673/7,226, 92%) for further analysis based on power calculations (Figure S1C, Methods), and computed log-fold changes that quantify the average perturbation effects across all cell lines for each target (6,673 targets x 6,471 expressed genes; Figure 1D, Methods). We next quantified the sources of global variation in the gene expression profiles. As expected for a large-scale single cell RNA sequencing study, technical factors, cell quality and cell line were major drivers of expression heterogeneity between cells (Figure 1B-C, S1D). Transcriptional changes induced by CRISPRi knockdown were relatively small compared to those global sources of heterogeneity (Figure 1C), suggesting that CRISPRi targeting induces mostly subtle transcriptional changes rather than complete cell state shifts within the considered time period of 3-6 days post-infection (DPI). The quality of CRISPR targeting can be evaluated based on the on-target expression changes, as well as its effect on downstream genes beyond the target gene (“trans effect”). The mean expression of a targeted gene was significantly lower (Benjamini-adjusted p-value < 0.1) compared to control cells for 2,900 of the 4,874 (60%) knockdowns targeting an expressed gene (median log-fold change = -0.42 across all significant on-target trans effects), confirming an overall efficient target down-regulation in our system. In contrast, only 0.02% of all target-expressed gene combinations showed evidence of downregulation in trans beyond the target (8,833 out of 43,176,109 tested target-expressed gene combinations, Figure S2-1A). T o assess the quality of our measured trans effects, we considered a set of well characterised transcription factors with known regulatory relationships18 across diverse cell types, states, and transitions. All of the 17 significant trans effects in our data that overlapped these annotations (Figure S2-1B) were concordant with the effect direction expected from the function of the transcription factor. Specifically, knockdown of known activators (16 of 17 significant trans effects) resulted in reduced expression of the downstream genes, while knockdown of the known repressor BACH1 caused up-regulation of its downstream target, HMOX1. This remarkable concordance (p = 0.02, binomial test), highlights the potential for validating and charting gene regulatory effects from our data. 3 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Figure 1 | Quantifying knockdown effects in cells from multiple donors at scale. A) Experimental design of a genome-scale CRISPRi screen with single-cell RNA-seq read-out from 34 human iPSC cell lines. B) UMAP of the expression profiles of cells assigned to a donor and targeting guide after correcting for technical covariates, colored by cell line. C) Variance decomposition, displaying relative effects of cell line, genomic background and target gene perturbation on gene expression. D) Overview of the analysis strategy to understand individual trans effects (orange), E) similarity of effects across all genes (green), F) and similarity of perturbation responses across all targets (purple). Global transcriptional changes caused by gene-dosage reduction Signatures of perturbation effects can reflect regulatory mechanisms in three ways: (i) a trans effect of a target gene on downstream genes reflects links in a regulatory network (Figure 1D), (ii) correlation of trans effects for two target genes indicates similar function of the targets (Figure 1E), and (iii) correlation of gene responses in trans indicates common control by shared regulators (Figure 1F ). We next elucidated the factors that determine gene expression control in trans, shared trans effects (co-regulating targets), and shared regulators (co-regulated expressed genes). Perturbation effects are largest for targets affecting growth and transcription and acting on highly expressed and variable downstream genes 4 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Our resource charts the impact of perturbing genes expressed or essential in iPSCs. Their knockdown resulted in a range of transcriptional change, with some targeted genes inducing hundreds of trans effects (Figure 2A). About half of the target genes (51%; 3,374/6,674) had at least one significant trans effect upon knockdown, with a mean of 2.57 per target gene. The genes with the highest number of significant trans effects have roles in a variety of fundamental biological functions including pluripotency maintenance, transcription and splicing. For example, members of the RNA polymerase associated factor (PAF) complex PAF1, CTR9 and RTF1, had 962, 847 and 594 significant trans effects, while pluripotency maintenance complex members CNOT1, CNOT3 and myc-associated factor MAX had 473, 281 and 144, respectively. These genes, which have previously been shown play important roles in the maintenance of pluripotency 19,20, reflect the requirements for the stability of the core transcriptional circuitry in human iPSCs on the timescale of several days. Next, we asked which properties of the targeted gene predict the number of significant trans effects they have. We included different biological features, as well as the number of assigned cells, in a regularized generalized linear model (Methods), identifying essentiality of the target as the most important feature, which is in line with findings in other cell types 21 . This link was weaker for growth effects at earlier time-points (three and five days post-infection), as knockdowns with faster-acting growth defects could suffer from survival bias, with their fitness effects setting in before data were collected (Figure 2B). Among additional features, the number of Gene Ontology annotations, in particular being associated with gene expression 22, as well as evolutionary conservation and protein complex membership of the target gene were also associated with a higher number of significant trans effects, an observation consistent with the hypothesis that constrained genes are required for diverse important functions 23,24. We then considered genes that responded in trans to the knockdown of one or multiple target genes (here-on referred to as regulators). Over half (67%, 4,365/6,471) of the expressed genes had at least one regulator, with an average of 2.6 (Figure S2-1C). Controlling for differences in expression level, the number of associated hallmark biological pathways25 and high expression heritability 1 were the strongest predictors of having more regulators, while evolutionary conservation was the most informative predictor for having fewer (Figure S2-1D). These observations are consistent with previous hypotheses that genes with essential roles are more likely to be conserved across species and robust to perturbation of upstream regulators within one context, while genes with context-dependent expression levels are more responsive to regulation 23,24,26 . Next, we asked how the uncovered trans effects induced by CRISPRi knockdown could yield additional insights into the downstream regulatory consequences of natural genetic variants. Compared to identifying 512 significant trans effects by testing the expression of 2,044 genes across 26,877 SNPs in a collection of 1,367 iPSC lines 3 (total of 230,010,344 tests; 2.2e-04%), our screen identified a total of 1,328 (3.5%) significant trans effects for the 5,127 genes with known cis eQTLs among the set of target genes. Out of the trans eQTLs identified previously, 67 pairs of cis and trans genes were also quantified in our data. However, we only found shared signal for one trans-eQTL, where the expression of SLC3A2 was reduced upon knockdown of SLC7A8. The other trans eQTLs mapped in iPSCs did not yield a strong evidence of enrichment of significant expression change due to CRISPRi (Figure S2-1E). The discrepancy between these two sets of trans effects could be explained 5 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint by stronger effect sizes in the cis gene in our data (Figure S2-2F ). Given the matching experimental cell line context, and the more severe expression perturbation in cis due to CRISPRi, this lack of replication of signal from a large natural population suggests that complementary assays such as CRISPRi or CRISPR activation can help to verify regulatory insights from existing trans eQTL studies and avoid false negative and positive signal. Similarity of trans effects reveals protein complexes as the nexus for integrating gene expression changes Similarity of perturbation effect has proven to be one of the strongest lines of evidence for a functional link between two genes 9,27,28. We therefore computed the correlation of gene expression changes upon knockdown for every pair of target genes across all expressed genes (Figure 1E). A rich tapestry of functional relationships emerged, that well recapitulated broad functional roles of the targets (Figure 2C). In addition to genomic proximity, which has been shown to be a strong predictor of off-target behavior 29,30, protein complex co-membership was the strongest predictor of similar trans effects (Figure 2D). Indeed, knocking down individual members of a diverse set of protein complexes, such as the integrator complex, EIF3 complex and 26S proteasome, as well as complexes that have previously associated with the maintenance of pluripotency and self-renewal such as the eukaryotic gene expression regulator CCR4-NOT complex 31 and histone modification complexes NuA4/Tip60-HAT 32, RNA N6-methyladenosine methyltransferase 33 and INO80 34, induced similar expression changes (Figure S2-2A-B). Further, target pairs that were part of common co-essentiality modules defined by the Cancer Dependency Map 27 also had more similar perturbation effects (Figure 2C). While these modules largely overlap with protein complexes, they additionally link complexes involved in similar functions. For example, knocking down members of RNA polymerase pre-initiation complexes TFIID and TFIIH had similar impact as knocking down members of the mediator complex and members of the RNA polymerase core complex itself (Figure S2-2C). This demonstrates how large-scale perturbation screening can link genes across hierarchies spanning from regulatory links to complexes and processes. Consequently, we hypothesised that we could gain insight into the functions of poorly characterised genes by comparing their trans effects with those of better-studied protein complex members. We therefore aggregated cells with knockdowns from the same protein complex to obtain an average per-complex trans effect and computed its correlation to the trans effect of each target gene. We then prioritised 24 candidate gene-complex pairs based on correlation strength and shared known function or cellular compartment 35, and predicted pairwise interaction structures between each target and all complex members using AlphaFold2-multimer36 to assess the biophysical plausibility of the interaction (Supplementary Table ST2-1). Complementary information from known complex crystal structures and other biological evidence can help determine if interactions are likely to occur. 6 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Figure 2 | Global transcriptional changes caused by gene-dosage reduction. A) Histogram of the number of significant trans effects per target. T arget genes with the highest numbers are labelled. B) Model coefficients (x-axis) for predicting number of trans effects based on properties of the target gene (y-axis, DPI = days post infection). C) Heatmap of targets by trans effect similarity for the 280 knockdowns (with at least two co-regulating targets and at least one trans effect). D) Model coefficients (y-axis) for predicting similarity between trans effects based on functional relationship of two target genes (x-axis) E) Scatter plot of the trans effects induced by 7 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint knockdown of genes in the TOMM40 complex (x-axis) and PHB (y-axis). F) T op view of an α-helix on PHB (green) predicted to interact with the β-barrel of the TOMM40 (blue) complex in a similar manner to known complex members, via a shared interface with TOMM737 (yellow). G) Heatmap of correlation between expressed genes based on perturbation responses (lower-triangle) and natural single-cell co-expression (upper-triangle) for 313 expressed genes with the most co-regulated expressed genes (> 3 co-regulated genes). H) The regulatory behavior of hypoxia regulators HIF1A, ARNT and VHL on the glycolysis pathway. I) Regulators of commonly used iPSC marker genes. Knockdowns are indicated by green nodes while downstream genes are indicated in purple. Down-regulation upon knockdown is indicated in blue while up-regulation is indicated by a red arrow. Our candidate interactions form plausible complexes based on pDockQ 38 scores significantly more frequently than random protein pairs (p < 10 -15, Kolmogorow-Smirnow test) or known non-interacting pairs (p = 5 x 10 -13), while following a similar distribution to known interactions from CORUM (p = 0.97) 39. Eight of the 24 prioritized target-complex pairs had at least one plausible predicted interaction based on both pDockQ score and the visual inspection of the structures predicted by AlphaFold2, of which seven had a coherent interaction with the complex. This includes rediscovering the known structure for the interaction between DDX39B and THOC2 in the TREX complex, and finding plausible interfaces in between SMC3 and MED16 from the mediator complex 40,41 as well as DDX41 and the tri-SNP complex 42–44 via the homologous LSM2, LSM5 and LSM7 proteins45, cases where there is known to be an interaction but the structure of it is unknown. Beyond recovering known interactions, we identified four novel interactions with predicted plausible structures: PHB with the TOMM40 complex (Figure 2E-F ), RAB10 with the Paf complex (Figure S-2D-E) and ELP3 and CTU1 with the EIF2B complex (Figure S-2-2F-G ). These discoveries highlight the richness of functional data produced by genome-scale perturbation screens. Genes with similar perturbation responses reflect activated cellular pathways and replicate naturally co-expressed gene clusters Similarity between expression changes of two genes in trans across different perturbations indicates shared regulation (Figure 1F ). T o analyse this effect in our data, we computed correlations between perturbation responses to all targeted genes for every pair of expressed genes and observed high values between members of various cellular stress pathways (Figure 2G ). For example, downregulation of the master transcriptional regulator HIF1A and its binding partner ARNT caused down-expression of glycolysis genes LDHA, GAPDH and ALDOA, while knockdown of HIF1A degradation gene VHL resulted in the up-regulation of these genes, thus recovering the regulative relationship between hypoxia and the glycolysis pathways 46,47 (Figure 2H, S2-3A). Similarly, targeting genes in the mevalonate arm of the cholesterol biosynthesis pathway 48 resulted in the up-regulation of all other genes in the pathway (Figure S2-3B-C) indicating a shared feedback mechanism regulating their expression. We next quantified which annotations best predict similarity of perturbation responses for all gene pairs and found co-expression of single-cell wild-type gene expression 49 to be most informative (Pearson’s R = 0.41; Figure S2-3D-E, Figure 2G ). In addition to cellular stress pathways and common protein complex membership (Figure S2-3D), which share signal with single-cell co-expression modules, this signal was driven by co-regulated expression modules, such as that linked to pluripotency maintenance (Figure 2I; Figure S2-3F ). We 8 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint observed co-regulation of a set of five commonly used iPSC marker genes, CD24, TERF1, TDGF1, L1TD1 and DNMT3B50–56, which are also typically co-expressed in wild-type iPSCs49. The knockdowns that drove the shared signal were master pluripotency regulator OCT4 and knockdowns with predicted off-target effects (Supplementary Table ST2-2, Methods) and components of the mediator and Paf complexes, which resulted in their down-regulation, and NuA4 complex member ING3 gene, CNOT1 and the MAX transcription factor, which resulted in the up-regulation of these genes, all of which have previously been attributed to the maintenance of stem cell pluripotency and development 20,32,57,58. The finding of expression co-variation between individuals in natural populations to mirror common perturbation responses is of practical importance, as it allows designing efficient screening campaigns based on observational data. For example, savings on sequencing costs can be achieved by choosing a small set of genes for targeted capture, informed by correlation patterns in single cell co-expression, reasonably expecting to recover the main perturbation responses thanks to the shared signal. The influence of genetic background on gene perturbation effects Gene perturbation effects can sometimes be suppressed or exacerbated by genetic

and small molecules 59. We were therefore interested in understanding how often the trans regulatory effects that we observe differ in magnitude across healthy individuals, as this would indicate a mechanism for modulation. T o understand the malleability of gene perturbation impact by natural genetic variation, we capitalized on the diversity of donors of the induced pluripotent stem cell lines, and asked to what extent the variation of trans effects of a gene knockdown across cells could be attributed to genetic factors. We explored a targeted panel of 1,355 guides targeting 444 genes with 20 non-targeting guides, across 20 cell lines from 10 donors (2 lines per donor), focusing on genes more likely to have variable function across different cell lines (Figure 3A, Methods, Supplementary Table 3-1). After quality control, we recovered 1,161,864 cells of which we assigned 635,022 (55%) to 444 targets across 19 cell lines, collecting a median of 74 cells per target gene per line (Figure S3-1A-B, Supplementary Table 3-2). The knockdowns were again effective, with significant down-regulation of over 90% target genes in 15 of the 19 lines (Figure S3-1C). For every target gene, we computed trans effects on all 6,517 expressed genes (mean log-normalized expression > 0.1) across cell lines and for individual cell lines, following the analysis of the genome-scale screen (Methods). Across lines, the trans effects replicated the results from the genome-scale screen (Pearson’s R=0.73 across all trans effects that were significant in either screen), confirming that our earlier results were robust and reproducible (Figure S3-1D). When comparing perturbation effects in individual cell lines (Methods), we observed patterns consistent between the lines, with similar covariance structure between target genes (Figure S3-2A-B) and highly correlated response gene signatures (Figure S3-2C). T o identify differences between genetic backgrounds, we considered only the 10 cell lines for the 5 donors where both cell lines from the donor indicated high response to CRISPR perturbation (Figure S3-1C) and focused on 68,321 trans effects (2.4% of all possible ones) that were significant in at least one cell line and could not be attributed to any off-target or cross-mapping effects (line-specific |LFC| > 0.1, Benjamini-Hochberg adjusted p-value < 0.1, 9 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Methods). We found 2,941 trans effects (4.3% of tested) for which the donor component explained a significant fraction of variance (Benjamini-Hochberg adjusted p-value < 0.1), which we call heritable effects (Methods). The knockdowns with the largest excess of heritable trans effects included pluripotency maintenance genes such as master pluripotency regulator OCT4 and CNOT3; splicing genes SNRNP70, SRSF1 and ZMAT2; as well as FA2H and CACNA1A, two genes whose misexpression has previously been attributed to the rare diseases hereditary spastic paraplegia and hereditary cerebellar ataxia, respectively 3 . Conversely, trans effects due to knockdowns of PAF1 complex members CTR9, RTF1 and CDC73 and EIF3 complex members EIF3C and EIF3B were heritable less often than expected (Figure 3B, Figure S3-3A), indicating lack of segregating variation in the molecular impact of these key structures. The interpretation of a heritable trans effect depends on the expression of the downstream gene in the unperturbed state (Figure 3C). If the gene expression was heritable in control cells (consistently variable across donors, Benjamini-Hochberg adjusted p-value < 0.1; 246/2,941 cases), heritable trans effects can emerge due to near-complete repression, or other outcomes that remove the variation between donors (“loss in heritability”, 35/246), such as for C9orf135 expression changes due to OCT4 knockdown (Figure S3-3B). Usually, this was marked by down-regulation of the expressed gene (28 out of 35, 80%, Figure 3D), though there were also several instances where a knockdown increased the expression of a gene in a single donor, removing expression differences between donors. An example for this is the up-regulation of mitochondrial cytochrome b (MT-CYB) in kolf_2 and kolf_3 cells upon knocking down various components of the large mitochondrial ribosome such as MRPL55 (Figure S3-3C). Alternatively, expression heritability can be preserved where transcriptional change amplified effects of known genetic regulators like cis eQTLs or was small compared to the natural variation between donors (“maintenance of heritability”, 211/246) (Figure S3-3D). In a second scenario, we considered heritability upon knockdown of the target genes to be potentially gained, if we could not find evidence for heritable expression in the control cells (2,695/2,941 cases, p-value > 0.9). Indeed, 251 of these cases showed significant evidence for heritable gene expression after the knockdown, and the majority (187/251; 75%) of them were a result of up-regulation in the expressed gene in a single donor (Figure 3E). Such trans effects would often be overlooked when testing across all lines where the dilution of signal results in insignificant trans effects for most of these cases (103/187, 55%). This gain in heritability can potentially be explained by genetic differences in the action of responsible genomic regulators, among other mechanistic reasons. We found that 3 of 7 previously mapped trans eQTL hotspot genes 3 that we knocked down were linked to a heritable trans effect showing a gain in heritability in genes not previously linked to the eQTL: CREB3L2 on MICU2, ZNF208 on SEH1L and ZNF611 on MOB1A, (Figure 3F ; S3-3F-G). Most heritable trans effects could not be directly linked to heritable expression before or after knockdown at the chosen significance levels. For example, the expression of PRKD3 upon OCT4 knockdown is heritable, and associated to a previously mapped cis eQTL at chr2:37614653 (Figure 3G ). However, PRKD3 expression heritability in our wild-type control cells could not be determined (p-value = 0.68) from the small number of donors. Larger screens with more donors will be required to categorize such heritable trans effects reliably. 10 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Figure 3 | Genetic background influences perturbation response. A) Analysis of variation in trans effects by considering a panel of 444 targets across 10 donors, with 2 cell lines per donor. B) Knockdowns with the most and least heritable trans effects, controlling for the number of trans effects tested. C) Breakdown of heritable, perturbation-induced, transcriptional change (out of 3,051). D) Mean transcriptional fold-change of trans effects resulting in loss of heritability of expression. E) Mean transcriptional fold-change of trans effects resulting in gain in heritability of expression. F) Expression change of MICU2 due to knockdown of eQTL hotspot CREB3L2. G) Expression change of PRKD3 due to knockdown of OCT4, separated by allele of a cis-eQTL of PRKD3. T owards design, implementation, and analysis of genome- and population-scale single cell CRISPR screens The decrease of single cell library preparation and sequencing costs is making perturbation screening across a population, using a rich gene expression readout a reality. We have performed the first genome-scale screen with high-dimensional read-out across multiple individuals, and found that nearly all aspects of experimental design have an impact on the outcomes, which is important to consider in future studies. 11 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint In the targeted screen, guide identity had the largest effect on the observed trans effects among the genetic and experimental factors we tested (Figure S4A-B). While effects of different guides for the same target were mostly consistent (median Pearson’s R = 0.61 across all genes with at least 25 trans effects), unintended effects have been shown manifest from targeting similar sequences in other regions of the genome 60, in particular where such regions lie in close proximity of a TSS of another gene 9,61. Of 2,067 potential off-target trans effects, 112 exhibited significant repression of the predicted off-target gene and, on average, exhibited 24-fold greater variation due to guide than other trans effects (p-value = 1.3e-05, two-tailed t-test) (Methods, Supplementary Table ST4-1). Even stricter criteria might be needed to rule out all possible off-target effects. Perfect complementarity as short as 9bp of the guide seed sequence and promoter have been observed to result in off-target activity 61 and such observations were replicated by those guides fulfilling this criteria on the expression levels of OCT4 (Figure S2-3F). In addition to guide identity, CRISPRi efficacy explained a large portion of variation in the observed trans effects (Figure S4B). The four lines where expression changes of the targeted genes that were not significant for over 50% of the library saw muted global transcriptional changes (Figure S3-1C, S3-2C). Both the CRISPRi on-target efficacy and the gRNA assignment rate were highly predicted by dCas9 expression in the cell line (Figure S4C), the latter likely due to the protective effect of the enzyme against degradation. These cell line effects indicate that successful screening requires a high dose of the Cas enzyme, and accounting for its efficacy in analysis, even after experimental selection for highly performing lines. The experimental design required considerations on the number of cells per knockdown, approach to pooling cells from individuals, and timing of the assay were key to ensure that the resulting data were well-powered to draw reliable conclusions. First, depending on the sought effect size, anywhere from 10 to 200 cells may be needed to be profiled, as evaluated from a downsampling experiment (Figure S1C). Second, in an additional experiment knocking down 161 genes (Supplementary Table ST4-2) in 24 cell lines, and pooling prior to transfection of dCas9-KRAB-MeCP2, we observed that 6 out of 24 lines accounted for 98.4% of all the cells (Figure S4D), indicating technical challenge in ensuring balance across cell lines. Finally, in another experiment of 483 guides knocking down 161 genes with strong growth effects (Supplementary Table ST4-2) measured 14 days post-infection resulted in only 34% of cells successfully assigned to a guide, and no targets with significant downregulation (Figure S4E). In comparison, gRNAs could be successfully assigned for 52% of cells whose knockdown targeted a gene with weak growth effects, and 59% of the targeted genes had significant expression changes (Figure S4E). By sequencing earlier on days 3, 4, 5 or 6 post-infection and pooling cell lines late, we were able to recover sufficient numbers of cells in our later experiments (Figure S1B) and consistent target down-regulation (Figure S2-1A), highlighting the importance of carefully selecting the sequencing time-point when measuring impact of genes with fitness defects. 12 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint

We have presented the first study in healthy humans to combine natural genetic variation with engineered perturbations and single cell sequencing readout. The impact of perturbations on gene expression ranges from limited impact to changing expression of thousands of downstream genes and highlights an overall concordance of observations from different research approaches, but also some differences. The main directions of variation in gene expression changes after knockdown are similar to those observed without perturbations in single iPS cells from a larger cohort 49 (Figure 2G ). Many of these co-expressed gene modules can be explained by a coordinated response to a signal, such as stress factors. When transcription factor perturbations had an effect in our system, they were always consistent with the annotated role of the factor as an activator or repressor. Lead cis eQTL SNPs explained substantial variation in heritable downstream perturbation effects as well. However, trans eQTLs, perhaps the most direct comparison to our study setup, mostly did not replicate, indicating either false positive mapping results for eQTLs, winner’s curse to diminish their effect sizes, false negative perturbation effects here, or discrepancies in experimental context. Cellular model resources with trans regulation maps as we have established here, power the analyses of cis eQTLs and expression-altering disease mutations by imputing their trans effects in silico and prioritising candidate genes. Similar effects of perturbations indicated membership in the same protein complex or co-essentiality module. This allowed us to detect most complex members for several pieces of core cellular machinery. In addition, we could identify likely protein complex interaction partners by combining consistency of perturbation effect with AlphaFold2-enabled prediction that forecasts a confident interaction event. Similar screens in other cellular contexts where different complexes impact on gene expression could be used to identify their interaction partners or regulators. Overall, protein complexes emerge as a nexus for integrating transcriptional changes and modulating downstream effects (Figure 4). Differences in gene expression, arising from factors like transcriptional stochasticity, cell cycle stages, epigenetic states, or genetic background, do not always lead to changes in protein abundance due to buffering mechanisms at the protein level 12. For instance, excess uncomplexed proteins resulting from increased mRNA expression may be degraded due to exposure of otherwise hidden hydrophobic residues. Similarly, when rate-limiting components of protein complexes are degraded, the entire complex cannot form, leading to shared phenotypic impacts. These buffering mechanisms ensure that genetic effects on gene expression do not always manifest as changes at the protein level, thereby stabilizing cellular functions as reflected in the overall high degree of consistency in trans effects across cell lines. T ogether, our results indicate that protein complexes play a key role in mitigating the impact of gene expression variability, thus preserving the integrity of downstream processes. 13 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Figure 4 | Robustness of knockdown effects to inter-individual differences. A venerable question in genetics is the extent of impact of modifiers, mutations that do not have large independent effects themselves, but modulate others. This is important both to be able to predict mutation impact, but also to identify the effects that can be modified at all, such that other ways to modulate cells, such as small molecules, could also be used. Many such alleles have been mapped in yeast, other model organisms, and human disease genes62,63. Our experimental design across multiple donors allowed us to identify perturbation effects that vary between individuals. Such heritable changes are small on average compared to technical impact of batches or sequencing coverage, as well as relative to biological covariates such as cell cycle state or CRISPR reagent efficacy. Cis eQTLs were frequently a likely natural cause for these effects, with the lead genetic variant associated to gene expression also influencing the response to perturbation. The technological advances and cost reductions support scaling of genetic screening with single cell sequencing readouts. We have identified several factors that contribute substantially to the success of such campaigns, including strategy of pooling lines, sequencing time point, efficacy of engineering, reflected both by Cas9 activity, as well as on-target perturbation effect, gRNA recovery, and gRNA off-target effects. For CRISPRi specifically, off-targets are more promiscuous, with sequence matches in the 10nt of the PAM-proximal region sufficient to cause substantial unintended effects 61. Large scale screening campaigns should map the impact of these variables separately before the start, in conditions that precisely match the scale-up phase later, to avoid complications. 14 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Our study marks the beginning of genome-wide CRISPR screening with population-scale single cell RNA sequencing readout. This approach can be applied in cell models, organoids, and primary cells, where single screens have already provided new insights into human cell workings. Expanding these screens across many individuals and including more diverse genetic backgrounds will advance our understanding of disease causes and untangle genetic effects. While requiring careful technical optimization, such data are essential to building foundation models of individual-specific perturbation responses to genetic and small molecule changes, and ultimately predict and control cell behaviour.

Grant support. C.F ., E.M.P , Y .Z., K.C., L.C., S.U., A.D., J.S., M.E.S., D.M., E.G.N., A.B., S.L., Y .G., L.P ., O.S., B.V. were supported by Wellcome (220540/Z/20/A). M.E.S was supported by Wellcome (220442/Z/20/Z). B.V., J.M.B were supported by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – 540147573. Author contributions. Designed project: L.P ., C.F ., B.V., L.C., Y .Z., E.M.P ., S.U. Performed experiments: E.M.P ., Y .Z., K.A.X.C., S.C., A.B. Analysed and interpreted data: C.F ., B.V., A.D., M.J.B, J.M.B., J.S., M.E.S., D.M., E.G.N., S.L., Y .G., B.F ., D.B. Supervised study: L.P ., O.S., B.V. Wrote paper: C.F ., B.V., L.P ., O.S., A.D. Declaration of interests O.S. is a paid advisor of Insitro. Inc. A.B. has been a founder and consultant for EnsoCell since August 2023. L.P . receives remuneration and stock options from ExpressionEdits. The other authors have no other interests to declare. Resource availability Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Leopold Parts ([email protected]).

availability This study did not generate new unique reagents. Data and code availability - Raw data for the genome-scale and targeted screen are available from SRA under the accession number ERP165335. 15 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint - Processed data, identified trans effects and heritability scores as well as additional data summaries to reproduce the findings from this manuscript are available from Figshare: https://figshare.com/s/14edeeab56eb8a885df3 - Code used for processing and analysis of the data is available from GitHub: https://github.com/claudiafeng123/crispri_scrnaseq_hipsci. - An App for interactive exploration of the data is available here: https://www.sanger.ac.uk/tool/crispri-scrna-seq-hipsci/ Supplemental information ● Supplementary Figures S1 - SM ● Supplementary T able ST1-1 : Selected genes for CRISPR knockdown in the genome-scale screen ● Supplementary T able ST1-2: Number of assigned cells per knockdown in the genome-scale screen ● Supplementary T able ST2-1: Plausibility of physical interactions between genes and complexes with similar transcriptional change due to knockdown ● Supplementary T able ST2-2: off-targets of OCT4 ● Supplementary T able 3-1: Lines used for the targeted screen ● Supplementary T able 3-2: Selected genes for CRISPR knockdown for the targeted screen ● Supplementary T able 3-3: Heritable trans effects ● Supplementary T able ST4-1: Possible off-target effects based on less than 3 nucleotide mismatches to a Fantom 5 promoter ● Supplementary T able ST4-2: Genes selected for CRISPR knockdown in the pilot study ● Supplementary T able STM-1: Primers used ● Supplementary T able STM-2: Sequences used 16 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint

Experimental design Gene selection for the genome-scale panel Guides were selected for the genome-scale panel based on wild-type expression levels in iPSCs, as well as those that demonstrated growth effects. Guides were then separated into those with targeting genes with fitness effects (fitness genes), which were measured at 3, 4 and 5 DPI, and those without (non-fitness genes), which were measured at 6 DPI. 40 non-targeting control guides from Dolcetto A were used for both panels. T o identify genes with fitness effects, we conducted an essentiality screen using the Dolcetto library17. We used 57,050 guides targeting 18,899 genes from the Dolcetto A library and 57,011 guides targeting 18,897 genes from the Dolcetto B library to knock down a total of 18,940 genes in fiaj_1 cells and harvested cells at 3, 4, 5, 9 and 10 days post-infection. At each time-point, fitness effects were quantified by calculating the log 2-fold change of normalised cell counts compared to that of the read counts in the plasmid library 64 and genes were considered to have fitness effects if the median fitness effect at day 10 across all guides was less than -1. The three guides with the lowest log 2 fold change at day 10 post-transfection were then chosen for screening. If fewer than 3 guides were available across both Dolcetto A and B libraries, all available guides were chosen. In total, this part of our library consisted of 6,784 guides targeting 2,264 targeted fitness genes. Additional genes, without a fitness effect iPSCs were selected based on fitness effects in cancer cell lines, as well expression level in iPSCs. Fitness effects in cancer cell lines were assessed based on the CERES scores of all 1,376 lines in the DepMap consortium 16. Gene expression in iPSCs was measured in a pilot screen on fiaj_1 cells and expression values per cell were normalised by total sum scaling with a scale factor of 100,000 and log-transformation with a pseudo-count of 1 65. Additional targets were considered if they were either highly expressed in iPSCs (normalised expression > 0.1), or had strong fitness effects in cancer cell lines and were expressed in iPSCs (genes with a CERES score > 0.22 or 0.01) or had variable fitness effects (genes with a CERES score standard deviation across lines > 0.15 and a normalised expression > 0.01). In addition, the 50 genes with highest CERES score standard deviation but expression < 0.01 were selected. For each gene, 3 guides were selected from the Dolcetto A library, complementing with guides from the Dolcetto B library if less than three guides were available in Dolcetto A. In total, 4,962 genes and 14,883 guides were selected. Gene selection for the targeted panel Genes were selected for the targeted panel for measuring genetic background effects based on their effect size observed in the genome-scale screens. Only target genes with at least 20 cells per gene, log-normalized expression greater than 0.1 and minimum correlation of significant trans effects across timepoints and guides larger than 0.5 were considered for selection. Of these, all genes with more than 12 significant trans effects were selected for the panel (n=110 genes, 106 of which were fitness genes). For comparison, we added 97 17 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint genes with 5-12 significant trans effects (all of which were fitness genes) and 203 genes with fewer than 5 significant trans effects (141 of which were fitness genes). In addition, we considered 7 genes that have been linked to eQTLs with many trans effects3, 38 genes associated with monogenic diseases 3, 17 genes with high variance of CERES scores across lines16, 15 genes with a high expression heritability 1 and ARID1A, EZH2 and BCOR. Guides were chosen as in the genome-scale screen apart from 5 outlier guides, for which the target gene log-fold change was higher than the upper bound of a 95% confidence interval in a linear regression of guide-level versus target-level log-fold changes and a suitable replacement guide with a similar growth-effect ten days after transfection could be found in the Dolcetto libraries. This resulted in a total of 1,355 guides targeting 444 genes and 20 non-targeting control guides for the targeted panel. Experimental protocol Molecular cloning The libraries were cloned into the lentiviral expression library pKLV2-U6gRNA5(BbsI)-ccdb-PGKpuroBFP-W (Addgene 67974) 66. Briefly, the guide libraries were ordered from Twist Biosciences as 215-mer oligo pool. The pool was composed of several sub-pools to allow for the selective amplification of gRNAs that were amplified with subpool specific primers (Supplementary Table STM-1). BBSI-digested amplicons encoding gRNAs were inserted into the BBSI-digested vector by Gibson assembly (NEB Gibson Assembly Master Mix) according to manufacturer’s specifications, and transformed by electroporation (NEB 10-beta Electrocompetent E. coli C3020K). Bacterial cells were cultured overnight in liquid culture and plasmid DNA was extracted. The plasmid libraries were pooled together in equimolar ratios to achieve the desired final libraries. For the construction of the pB-CAGGS-dCas9-KRAB-MeCP2-BSD-mScarlet plasmid, the pB-CAGGS-dCas9-KRAB-MeCP2 (Addgene 110824) vector was digested with NotI (NEB) and EcorV (NEB). The EF1α promoter and blasticidin resistance gene was amplified by PCR using primers #1009 and #1010 (Supplementary Table STM-1). The SNV40 polyA signal was amplified by PCR using primers #1013 and #1014 (Supplementary Table STM-1). The mScarlet sequence was amplified by PCR from plasmid pmScarlet_C1 (Addgene 85042) using primers #1016 and #1012 (Supplementary Table STM-1). All products were purified with Monarch DNA Cleanup Columns (NEB). T2A sequence was ordered as a gBlock from IDT . A Gibson assembly with 4 fragments is incubated at 50ºC for 30 minutes and transformed by electroporation. Cell culture Human iPSCs were cultured on Vitronectin XF (StemCell T echnologies, 07180)-coated plates and mT eSR Plus medium (StemCell T echnologies). The medium was changed every other day throughout expansion and all experiments. Cell lines were cultured at 37°C, 5% CO2. dCas+ cell line generation and activity validation For the generation of dCas9-KRAB-MeCP2 iPS cell lines, 3 x 10 5 wild type cells were seeded into 12-well plates with ROCKi containing media. For the transfection of one line, 18 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint 600ng of pB-CAGGS-dCas9-KRAB-MeCP2-BSD-mScarlet, 300 ng of mPBase 67 and 100ng of a reporter plasmid encoding for GFP were mixed with 50ul of Opti-Mem in one tube and 50ul of Opti-Mem was mixed with 2 ul of Lipofectamine Stem (Invitrogen) in another tube. After 5 minutes of incubation at room temperature, the contents of the tubes were mixed together and incubated for another 10-30 minutes at room temperature. During incubation, the media in the wells was refreshed and 0.5ml media was added. After incubation, 100ul of the complexes were added to the wells. 24h after transfection, 1ml of media was added to cells. 48h after transfection, blasticidin (TOKU-E) selection was started using a concentration of 2µg/ml. The cells were cultured in selection for 2 weeks. T o validate the dCas9-KRAB-MeCP2 activity of the cells, an adopted method of the previously published Cas9 validation system was used 64. Briefly, cells were transfected with a plasmid that encodes for BFP and GFP and either a mock gRNA or a gRNA targeting GFP TSS. 1 x 10 5 cells were seeded into 24-well plates. Cells were transfected with either the mock or silencing construct using Lipofectamine Stem 24h later. BFP and GFP expression were measured three days after transfection at FACS. dCas9-KRAB-MeCP2 activity was calculated based on the median expression of GFP in BFP positive cells. Two replicate measurements were made for all cell lines for both conditions. Lentivirus production and determination of lentiviral titer Supernatants containing lentiviral particles were produced by transient transfection of 293FT cells using Lipofectamine LTX (Invitrogen). 5.4 μg of a lentiviral vector, 5.4 μg of psPax2 (Addgene 12260), 1.2 μg of pMD2.G (Addgene 12259) and 12 μl of PLUS reagent were added to 3 ml of OPTI-MEM and incubated for 5 min at room temperature. 36 μl of the LTX reagent was then added to this mixture and further incubated for 30 min at room temperature. The transfection complex was added to 80%-confluent 293FT cells in a 10cm dish containing 10 ml of culture medium. After 48 h viral supernatant was harvested and fresh medium was added. After 24h the lentiviral supernatant was collected and mixed with the first supernatant which was then stored at -80°C. For gRNA library lentiviral titration on dCas9-KRAB-MeCP2 expressing iPSCs, iPSCs were harvested by Accutase (Stemcell T echnologies) as single cells. iPSCs (3.6x10 5/well in 6-well plate) were infected with at least five serial dilutions of lentiviral supernatant supplemented with 10µM Rock inhibitor Y-27632 (Stemcell T echnologies). Uninfected cells were used as negative control. The transduced cell mixture was cultured in 6-well plates in 2ml/well. 24h post transduction, the medium was refreshed with mT eSR Plus without Rock inhibitor. After three days of cell culture the cells were harvested for FACS analysis and the level of BFP expression was measured. Virus titer was estimated and scaled up accordingly for subsequent screens. Screening and sequencing Cells were transduced with the lentivirus aiming for an MOI of 0.2. The cells were seeded at a density of 2.0 x 10 5 to 4.5 x 10 5 depending on the day of harvest. Media was refreshed 24h after transduction. Cells were harvested either on day 3, 4, 5 or 6 after transduction. On collection day, cells were harvested with accutase, spun down and resuspended in eBioscience Fixable Viability Dye eFluor 780 (Invitrogen) that was diluted 5000-fold in eBioscience™ Flow Cytometry Staining Buffer (Invitrogen). Cells were stained for at least 5 19 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint minutes and then filtered with Scienceware® Flowmi™ Cell Strainer (SP Belart). Cells were then sorted based on dead/alive-staining, BFP and mScarlet expression on MA900 Multi-Application Cell Sorter (Sony), The BD Influx™ (BD Biosciences) or MoFlo XDP Cell Sorter (Beckman Coulter). An equal number of cells were sorted for all the lines. 12 lines and 8 lines were pooled together for the genes with and without fitness effects, respectively, and 1.65 x 10 4 cells were loaded in a 10X inlet. Chromium Next GEM Single Cell 5' Kit v2 (10X Genomics) was used for transcriptome capture, with a modified protocol where we added an extra primer to the GEM generation mix to capture gRNAs 21. Computational analysis Unless otherwise stated, all analyses were performed in R (version 4.3.1) 68 and Seurat (version 5.0.3). Read alignment using CellRanger Reads were aligned with CellRanger 69 (version 6.0.1), processing each inlet separately. Alignment was conducted using default parameters, using genome build GRch38 as a reference, an adding additional sequences for BFP , mScarlet, BSD and dCas9-KRAB-MeCP2 (Supplementary Table STM-2). The sgRNAs were aligned to libraries for the fitness genes and other genes, respectively. For one inlet, the minimum threshold for the GEX/Cite-Seq cell barcode overlap was lowered from 0.1 to 0.01. De-multiplexing of cells based on natural genetic variation Individual cells were assigned to the source cell lines by de-multiplexing using natural sequence variants, as each pool consisted of lines from different individuals. We first used cellSNP 0.1.7 70 to call genotypes from the bam files containing the 10x read sequences for all cells passing the CellRanger filters. We used bcftools 71 (version 1.10.2) to subset a list of candidate SNPs 72 to only lines present in each inlet and filtered for a min. allele frequency threshold of 0.01 and minimum aggregated count of 20. This output was used in Vireo (version 0.2.1) 73 to de-multiplex the cells into the number of lines present in each pool using genotype data for each donor provided by the HipSci consortium 1, modified variant coordinates from GRCh37 to the genome build GRch38 using CrossMap 74. In total, 69% and 72% of cells were uniquely assigned to one cell line in the genome-wide and targeted screens, respectively. Doublets and unassigned cells were removed for further analysis. Quality control and filtering High quality cells were retained based on three criteria: number of RNA UMI counts per cell, number of unique features per cell and percentage of mitochondrial RNA 75. The number RNA UMI counts and unique features per cell were either bimodal or trimodal for each inlet and we removed cells that were in the lowest mode of number of features and UMI counts using inlet-specific thresholds between 1,926 and 39,260 UMIs per cell (average of 14,186 RNA UMIs across inlets), 2,000 features per cell and a percentage of mitochondrial genes above 10%. After filtering, we assigned cell cycle scores for each cell using Seurat’s CellCycleScoring function with cell cycle marker genes retrieved as cc.genes.updated.201065. 20 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Guide assignment T o establish an optimal guide assignment strategy, known to impact power and discoveries 76, we considered a pilot data-set knocking down 161 genes with weak fitness effects in 24 iPSC lines (Supplementary Table ST4-1). We employed five different tools and evaluated the quality of each assignment by considering the number of knockdowns with significant on-target down-regulation and the median number of cells per guide (Figure SMA-B). Based on this, we considered the relative UMI abundance of the most abundant guide with respect to the total number of guide UMIs in a cell for guide assignment in all further analyses. Cells were assigned to a guide if the relative frequency of the most abundant guide was in the upper mode across cells within a cutoff window 0.5 and 1 (median threshold across inlets was 0.75, minimum 0.5 and maximum 0.88) and had a minimum of 3 UMIs in the cell. All other cells were considered unassigned. The percent assigned cells varied across inlets and library sizes, ranging from 9% to 78% in the genome-wide experiments and 37% to 69% in the targeted experiments, with an average of 30% and 55% of cells assigned to a single guide, respectively. Data integration and variance component analysis For all cells passing quality control we normalised the data by total sum scaling with a scale factor of 10,000 and log-transformation with a pseudo-count of 1 and combined these results using Seurat’s merge function, keeping all genes with a minimum normalised expression of 0.1 in all inlets, resulting in a total of 6,471 expressed genes in the genome-wide screen and 6,517 expressed genes in the targeted screen. T o produce UMAP plots, we extracted highly variable features using FindVariableFeatures, performed PCA using RunPCA and calculated a UMAP embedding on the top 20 PCs. T o quantify the contribution of the different variables on the transcriptome heterogeneity, we used a linear mixed model on the expression of the 2,000 most highly variable genes in a variance component analysis including donor / cell line, batch / inlet, cell cycle phase, sequencing time point and target gene as random effects and percentage of mitochondrial genes and total number of UMIs per cell as fixed effects. T o remove technical and batch differences as well as line-specific effects, the corresponding variables were regressed out from the normalised expression data using ScaleData with vars.to.regress set to the respective variables and the PCA and UMAP were re-calculated on the residuals of the model. Quantifying perturbation effects T o quantify perturbation effects in the genome-wide screen, we defined all unassigned cells as well as cells assigned to a non-targeting guide as control cells. Based on the gene expression measurements of all control cells, we used a linear model to estimate the effects of cell line, inlet, percent of mitochondrial genes, cell cycle scores and total number of UMIs per cell on the gene expression. T o assess the effect of a perturbation within an assigned cell, we calculated the expected expression of each gene based on the linear model and compared this to the observed expression, yielding a perturbation effect profile for each cell defined as the difference of the expected and observed expression. T o assess overall perturbation effects per guide, per target or per target x line pairing, we averaged these effects across all cells assigned to a guide, target or target x line pairing, respectively, with significance evaluated based on a z-test using the residuals variance of the control fit. For the genome-wide screen, targets were considered for analysis if they had 21 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint a minimum of 10 cells assigned to it, individual guides if they had a minimum of 5 cells. This left a total of 14,982 guides and 6,673 targets to be considered, for which trans effects on 6,471 genes were calculated, giving a total of 96,948,522 and 43,180,983 trans effects across guides and targets, respectively. For the targeted screen, effects for each of the 444 knockdowns were computed across all lines, as well as separately for each of the lines with where there were least 10 assigned cells (total of 8,204 trans effects). As in the genome-wide screen, only the transcriptomic changes for the genes whose log-normalized mean expression was greater than 0.1 were computed (6,517 expressed genes). In total, effects were computed across 53,465,468 target, line, expressed gene triplets. For cis effects by natural genetic variants, we used the same procedure to estimate cis effects size on the known cis gene,using all control cells in the model and replacing the cell line covariate with the number of alternate alleles as a proxy for the genotype of each donor. Cis effect sizes for every eQTL were determined as the model coefficient. Trans effects were considered significant if the p-value after Benjamini-Hochberg correction across all targets, tested genes and lines (if applicable) was below 0.1. Power estimation based on down-sampling experiments We estimated the variance of the estimated transcriptional change due to knockdown and impact of the number of assigned cells using a bootstrap procedure. For this, we considered 118 targets of varying effect sizes (11 < # of differentially expressed genes < 551) where the full data set had at least 1,000 assigned cells. For each target, we subsampled all cells with replacement to obtain a simulated dataset of 5, 10, 25, 50, 100, 250, 500 and 1000 cells. Transcriptomic changes for all expressed genes were then computed separately on each of these data sets exactly as on the full data set (see Quantifying perturbation effects) This was done 25 times per knockdown, resulting in a total of 25 separate estimates of transcriptional effect for 8 different sample sizes for each of the 118 targets. Identifying co-regulated and co-regulating genes T o quantify the similarity of targets (and expressed genes) based on their perturbation effects (perturbation response) we calculated Pearson’s correlation between targets (expressed genes) based on the log-fold changes for all 6,471 expressed genes (6,673 well-powered targets). A total of 22,261,128 target-target and 20,933,685 expressed gene-expressed gene pairs were considered for analysis. Two targets (expressed genes) were considered to be co-regulating (co-regulated) if the absolute Pearson’s correlation between their trans perturbation (response) was greater than 0.2. Quantifying heritability of perturbation effects across donors Heritability was estimated in the targeted screen, considering the 5 donors where both cell lines from each donor demonstrated strong response CRISPR perturbation. We considered every target-expressed gene pair where we observed a significant trans effect in at least one line (68,321 target x expressed gene pairs). For each of these pairs, the perturbation effect profile for every assigned cell obtained from Quantifying Perturbation Effects was fitted with a linear mixed model using normalized target gene expression as a fixed effect and guide, cell line and donor as random effects. Donor effect was quantified by computing the likelihood ratio between the full model and the model without donor as a random effect. T o 22 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint assess significance we created a permutation scheme to obtain an empirical null distribution of the donor effects. For this, donor labels were permuted across cell lines such that two cell lines from one donor were assigned to different donors after the permutation. Thereby we retain the cell line structure in the data but permute the donor structure. This yielded a total of 544 permutations for the 5 pairs of cell lines. Permutations for every target-gene pair were computed until 10 null values greater than or equal to the true value were observed (stronger observations), or until 10 4 null values were computed. The empirical p-value was estimated to be ptd = (max(10, # Stronger Observations + 1)/(min(# of permutations, 10 4) + 1). Effects where the Benjamini-Hochberg adjusted p-value < 0.1 were considered to be significant. All models were fitted using the lmer function from the R package lme4 (v1.1-35.1) 77. Log-likelihood was computed using the logLik function from the R stats package. Identification of off-target effects A knockdown was considered an o ff-target effect based on two criteria: - Genomic proximity: guide sequence could be mapped to within 1kbp of a transcription start site of another gene in the Fantom 5 database 78,79 - Sequence similarity: any sequence with fewer than 3nt mismatches could be mapped to within 1kbp of a transcription start site of another gene in the Fantom 5 database78,79. T o identify guides with potential off-target effects on OCT4, we additionally considered any guide in our library whose first 9nt of their seed sequence could be mapped to a region within 2kbp of a transcription start site of OCT4 in the Fantom 5 database. Functional annotations Functional annotations were used throughout analysis, such as for predicting number of trans effects and number of regulating target genes, similarity of transcriptional e ffects upon target downregulation and co-perturbation. T o do this, we made use of the following annotations: Conservation scores. Conservation scores were obtained for each target from the Bioconductor package phastCons100way.UCSC.hg38 (version 3.7.1) 23. Wildtype expression, correlation and variance of expression. Wild-type gene expression, variance and co-expression was calculated based on undi fferentiated iPSCs 49. Values were computed based on the log-normalised expression values after regressing out effects due to donor and technical covariates (percent of mitochondrial genes, total number of UMIs per cell and number of genes expressed). T wo genes were considered co-expressed on the single-cell level if their absolute Pearson correlation was above 0.3. Essentiality. Essentiality was quanti fied as described in Gene selection for the genome-scale panel based on an iPSC cell line and the DepMap consortium 16. 23 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Expression heritability. Heritability of wild-type expression of iPSCs was obtained from the HipSci consortium 1. Protein-protein interactions. Known gene interactions were obtained from the OmniPath database80 using the import_all_interactions function from OmnipathR (version 3.10.1) 81. A pair of genes was considered to be interacting if they formed an interaction pair in the OmniPath database (undirected). Protein complexes. Known protein complexes were obtained from the Omnipath database80 using the import_omnipath_complexes(resources = c('CORUM', 'hu.MAP')) function from OmnipathR (version 3.10.1) 81. Pairs of genes were considered to be a protein complex pair if they were both members of at least one common protein complex. Transcription factor regulation. T ranscription factor-target gene interactions were obtained from the DoRothEA database 82 using the function get_dorothea in the decoupleR package (version 2.8.0) 83, using all pairs with con fidence level A or B. Hallmark gene sets. Knockdown and target genes were annotated by their membership in release 7.5.1 of the mSigDB hallmark gene sets 25. T wo genes were considered to be a hallmark gene set pair if they were both members of at least one gene set. Co-essentiality. Co-essentiality modules were taken from Wainberg et. al. 27. T wo genes were considered to be a co-essentiality module pair if they were both members of at least one co-essentiality module. GO T erms. GO term annotations were obtained from the authors of the gPro filer database84. Pairs of genes were considered to be an enriched gPro filer pair if they had at least 1 GO annotation in common. Expression Quantitative Loci in Human iPSCs. Putative trans-eQTLs were obtained from 3 on February 27, 2023. A pair was considered a cis eQTL-trans eQTL pair if the target gene and identified trans gene in our data corresponded to a cis eQTL gene and its trans effect. Unless otherwise stated, prediction was done by fitting an elastic net regression model with alpha = 0.1 using of the glmnet package (version 4.1-8) 85. When predicting the discrete, positive values (i.e. the number of trans effects and regulating knockdowns), a Poisson regression was used. When predicting binary response (i.e. similarity of perturbation response or perturbation pro files), a logistic regression was used. T o account for di fferences in statistical power we controlled for the number of cells by adding this as a term in the model. 24 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Protein Complex Prediction We identified candidate complex interactions based on the correlation between candidate target genes downstream effects and those of known complex members (Pearson’s R > 0.2). Candidates were then categorised into known and novel interactions based on literature review and prioritised according to shared function and cellular compartment with the complex. AlphaFold-Multimer 86 (version 2.3) was used to model pairwise interactions between each target gene and all members of the candidate complex. pDockQ 38,87 scores were calculated for each interaction as well as a random background sample of protein pairs and a set of known protein interactions. We identified plausible target-complex interactions with a combination of manual examination of predicted structures and pDockQ scores. We then aligned the top predicted pairwise target-complex member interactions with known complex structures using PyMol 88 (version 2.5 Open-Source). MOFA Multi-modal factor analysis was used to compare trans effects across cell lines using MOFA2 (version 1.12.1) 89,90. For this, log-fold change values for 445 target genes, expression of 6,517 genes, dCas9-KRAB-MeCP2, BSD and mScarlet and 19 cell lines were z-transformed and input to MOFA with default parameters, with each cell line as a separate view in the model and the number of factors set to 8. 25 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Figure S1 | Coverage and variance in the genome-scale screen. A) Venn diagram of genes selected as targets. B) Number of lines (y-axis) for different numbers of cells recovered per line after genotyping (x-axis) with gRNA assigned (blue) and not (grey). C) Estimated absolute error of expression log-fold changes for varying number of assigned cells (top) (relative to estimates from all genes, with a minimum of 1,000; Methods), histogram of cells per target gene (bottom). D) Number of genes (y-axis; log10 scale) with increasing amounts of variance explained (colors) by cell cycle and technical artefacts (x-axis). 26 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Figure S2-1 | Molecular signatures of the trans effects of gene knockdown. A) Down-regulation of target genes due to CRISPRi. Expression of targeted (red markers) and other genes (grey markers) Red dots show target gene expression values in control cells (x-axis) and assigned cells (y-axis), grey dots show expression values of expressed genes beyond the target. B) Quantile-quantile plot of p-values of trans effect of transcription factors in the DoRoTHeA database. Orange points indicate the trans effects of transcription factors and their known targets while gray points indicate trans effects the same transcription factors with non-targets. Number of downstream genes (y-axis) with different numbers of regulators (x-axis). Labels: six genes with most upstream regulators. C) Histogram of the number of regulators per expressed gene. Expressed genes with the highest numbers of regulators are labelled. D) Absolute model coefficients (x-axis) for predicting the number of regulators based on properties of the expressed gene (y-axis). Blue: negative coefficients (fewer regulators); red: positive coefficients (more regulators). E) Volcano plot of the log-fold change (x-axis) and log-scale significances (Benjamini Hochberg adjusted p-values, y-axis) for trans effects of (target, expressed gene) pairs with a known eQTLs acting in cis on the target and trans on the expressed gene. Dashed line: p=0.1. Labels: two pairs with corrected p-value less than 1. F) Comparison of effect size of natural variation in expression attributed to a cis eQTLs and CRISPRi. Red: cis eQTLs with at least one significant trans effect. Gray: cis eQTLs without any significant trans effects. 27 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Figure S2-2 | Plausible protein interactions predicted from similar trans effects of knockdowns. Similarity of trans effects between target genes (x- and y-axis) involved in the A) regulation of transcription, B) translation and post-translational processing and C) transcription. Heatmap color: Pearson’s R. Annotation color: covariates of biological processes involved (see legend). D) A plausible quadramer formed between RAB10, WDR61, CDC73 and PAF1. RAB10 clashes with CTR9 when we consider the larger structure of Paf 91. E) Predicted binding structure between RAB10 and the Paf complex. F) Predicted binding structure of CTU1 with EIF2B3 at a buried interface 92 (6O81) of the EIF2B complex. G) Predicted binding structure of ELP3 with EIF2B3 and EIF2B5 of the EIF2B complex. 28 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Figure S2-3 | Co-regulated modules identified by perturbation response similarity | A) Heatmap of trans effects (log-fold change, color) of genes in the glycolysis pathway (x-axis) due to knockdown of hypoxia pathway regulators ARNT, HIF1A and VHL (y-axis). B) Joint up-regulation of cholesterol biosynthesis pathway members due to down-regulation of a pathway member. Purple nodes: Genes in the cholesterol biosynthesis pathway Red edges: up-regulation of arrow target upon knockdown of arrow source. C) As A), but change (color) of cholesterol biosynthesis gene expression (x-axis) upon knockdown of genes in cholesterol biosynthesis gene expression (y-axis). D) Predicting correlation between co-perturbation profiles of downstream effects. Coefficient (y-axis) for different covariates (x-axis) in a generalized linear model trained to predict correlation of downstream gene log-fold change vectors for pairs of targets. E) Correlation of gene expression values across single cells in wild-type iPSCs (x-axis) against correlation in response to perturbations of different targets in CRISPRi screening (y-axis). F) As A) and C) but of trans effects on iPSC marker genes CD24, DNMT3B, L1TD1, TDGF1 and TERF1 (x-axis) due to different target genes (y-axis). Single star: predicted off-target activity on OCT4 (Methods). Double star: inconsistent transcriptional change between guides for the same gene (maximum correlation between guides < 0.1). 29 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Figure S3-1 | Recovery of the targeted screen. A) UMAP representation (x- and y-axis) of technical covariate corrected expression of all assigned cells (markers) in the targeted screen, Colors: cell lines. B) Number of assigned cells (y-axis) per knockdown per line (x-axis). Dashed line: median number of cells per knockdown per line. C) Number of knockdowns with at least 10 assigned cells, plotted per line. Blue: number of knockdowns with significant (Benjamini-Hochberg adjusted p-value < 0.1, t-test) on-target down-regulation in a line. Red: additional number of knockdowns with insignificant on-target down-regulation. D) Concordance of all trans effects that were significant in either the genome-scale or target screen. Log-fold change across all cells in the targeted screen (y-axis) compared to genome-scale screen (x-axis) for 288,089 (target, downstream gene) pairs. A point represents a target-expressed gene pair. 30 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Figure S3-2 | Global effect of transcriptional change due CRISPRi perturbation across cell lines. A) Percentage of log-fold change variance explained (color) by different MOFA factors (x-axis) in different cell lines (y-axis). B) MOFA weights of a knockdown (markers) for different factors (x- and y- axis), for different factor combinations (panels). C) Correlation (color) of MOFA factors (panels), between cell lines (x- and y-axis). 31 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Figure S3-3 | Genetic background influences transcriptional response due to knockdown. A) Number of heritable trans effects vs. # of trans effects tested per gene. Genes in red indicate knockdowns with more heritable trans effects than expected, blue genes indicate knockdowns with fewer heritable trans effects than expected given the number of significant trans effects across lines. B) An example of loss of heritability. C9orf135 expression change due to knockdown of OCT4. C9orf135 expression (y-axis; log-normalized) in individual cells (markers) from different cell lines (x-axis, colors) with OCT4 knockdown. Colored dash: mean expression in knockdown in cell line. Grey dash: mean expression in control cells in cell line. Colored arrow: median expression change in line in response to knockdown. C) As B), but expression change of MRPL55 due to knockdown of MRPL55. D) Expression change of PCSK9 due to knockdown of MED7. E) Expression change of SEH1L due to knockdown of the trans eQTL hotspot ZNF208. F) Expression change of MOB1A due to knockdown of the trans eQTL hotspot ZNF611. 32 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Figure S4 | Variation of transcriptional change due CRISPRi perturbation. A) Strategy for quantifying sources of variation in transcriptional response due to knockdown. B) Percentage of variance explained in transcriptional response due to knockdown due to CRISPRi efficacy (on-target expression), guide, donor and cell line. C) dCas9-KRAB-MeCP2 activity vs. fraction of assigned cells. D) Number of cells recovered per cell line after 14 days of selection (y-axis) for different cell lines (x-axis) in a pilot experiment with early pooling of lines. E) Repression log-fold change (x-axis) and log-scale p-value (y-axis) of target gene (markers) in a pilot experiment with early pooling of lines and late sequencing time point (14 days post-infection). 33 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Figure SM | Comparison of guide assignment strategies. We utilized a Poisson-Gaussian model 21, a Gaussian-Gaussian model 69 adopted from previous work, as well as assigning cells to a guide if and only if more than a fixed threshold (> 5) of guide UMIs were detected (fixed threshold), if and only if the fraction of a guide compared to all UMIs in a given cell was greater than a given threshold (ratio) and a modified version of the Gaussian-Gaussian model where assignments were further filtered so that assignments based on fewer than 5 guide UMIs were disregarded (filtered Gaussian-Gaussian). T o evaluate the quality of each guide assignment method, we considered A) the number of assigned cells per guide and B) on-target repression. 34 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint

1. Kilpinen, H., Goncalves, A., Leha, A., Afzal, V., Alasoo, K., Ashford, S., Bala, S., Bensaddek, D., Casale, F .P ., Culley, O.J., et al. (2017). Common genetic variation drives molecular heterogeneity in human iPSCs. Nature 546, 370–375. 2. Jakubosky, D., D’Antonio, M., Bonder, M.J., Smail, C., Donovan, M.K.R., Young Greenwald, W.W., Matsui, H., D’Antonio-Chronowska, A., Stegle, O., Smith, E.N., et al. (2020). Properties of structural variants and short tandem repeats associated with gene expression and complex traits. Nat. Commun. 11, 1–15. 3. Bonder, M.J., Smail, C., Gloudemans, M.J., Frésard, L., Jakubosky, D., D’Antonio, M., Li, X., Ferraro, N.M., Carcamo-Orive, I., Mirauta, B., et al. (2021). Identification of rare and common regulatory variants in pluripotent cells using population-scale transcriptomics. Nat. Genet. 53, 313–321. 4. Dixit, A., Parnas, O., Li, B., Chen, J., Fulco, C.P ., Jerby-Arnon, L., Marjanovic, N.D., Dionne, D., Burks, T ., Raychowdhury, R., et al. (2016). Perturb-Seq: Dissecting Molecular Circuits with Scalable Single-Cell RNA Profiling of Pooled Genetic Screens. Cell 167, 1853–1866.e17. 5. Adamson, B., Norman, T .M., Jost, M., Cho, M.Y ., Nuñez, J.K., Chen, Y ., Villalta, J.E., Gilbert, L.A., Horlbeck, M.A., Hein, M.Y ., et al. (2016). A Multiplexed Single-Cell CRISPR Screening Platform Enables Systematic Dissection of the Unfolded Protein Response. Cell 167, 1867–1882.e21. 6. Datlinger, P ., Rendeiro, A.F ., Schmidl, C., Krausgruber, T ., Traxler, P ., Klughammer, J., Schuster, L.C., Kuchler, A., Alpar, D., and Bock, C. (2017). Pooled CRISPR screening with single-cell transcriptome readout. Nat. Methods 14, 297–301. 7. Xie, S., Duan, J., Li, B., Zhou, P ., and Hon, G.C. (2017). Multiplexed Engineering and Analysis of Combinatorial Enhancer Activity in Single Cells. Mol. Cell 66, 285–299.e5. 8. Jaitin, D.A., Weiner, A., Yofe, I., Lara-Astiaso, D., Keren-Shaul, H., David, E., Salame, T .M., T anay, A., van Oudenaarden, A., and Amit, I. (2016). Dissecting Immune Circuits by Linking CRISPR-Pooled Screens with Single-Cell RNA-Seq. Cell 167, 1883–1896.e15. 9. Replogle, J.M., Saunders, R.A., Pogson, A.N., Hussmann, J.A., Lenail, A., Guna, A., Mascibroda, L., Wagner, E.J., Adelman, K., Lithwick-Yanai, G., et al. (2022). Mapping information-rich genotype-phenotype landscapes with genome-scale Perturb-seq. Cell 185, 2559–2575.e28. 10. Yao, D., Binan, L., Bezney, J., Simonton, B., Freedman, J., Frangieh, C.J., Dey, K., Geiger-Schuller, K., Eraslan, B., Gusev, A., et al. (2023). Scalable genetic screening for regulatory circuits using compressed Perturb-seq. Nat. Biotechnol. https://doi.org/10.1038/s41587-023-01964-9. 11. Niemi, M.E.K., Martin, H.C., Rice, D.L., Gallone, G., Gordon, S., Kelemen, M., McAloney, K., McRae, J., Radford, E.J., Yu, S., et al. (2018). Common genetic variants contribute to risk of rare severe neurodevelopmental disorders. Nature 562, 268–271. 12. Mirauta, B.A., Seaton, D.D., Bensaddek, D., Brenes, A., Bonder, M.J., Kilpinen, H., HipSci Consortium, Stegle, O., and Lamond, A.I. (2020). Population-scale proteome variation in human induced pluripotent stem cells. Elife 9. 35 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint https://doi.org/10.7554/eLife.57390. 13. Morris, J.A., Caragine, C., Daniloski, Z., Domingo, J., Barry, T ., Lu, L., Davis, K., Ziosi, M., Glinos, D.A., Hao, S., et al. (2023). Discovery of target genes and pathways at GWAS loci by pooled single-cell CRISPR screens. Science 380, eadh7699. 14. Tian, R., Gachechiladze, M.A., Ludwig, C.H., Laurie, M.T ., Hong, J.Y ., Nathaniel, D., Prabhu, A.V., Fernandopulle, M.S., Patel, R., Abshari, M., et al. (2019). CRISPR Interference-Based Platform for Multimodal Genetic Screens in Human iPSC-Derived Neurons. Neuron 104, 239–255.e12. 15. Gasperini, M., Hill, A.J., McFaline-Figueroa, J.L., Martin, B., Kim, S., Zhang, M.D., Jackson, D., Leith, A., Schreiber, J., Noble, W.S., et al. (2019). A Genome-wide Framework for Mapping Gene Regulation via Cellular Genetic Screens. Cell 176, 1516. 16. Meyers, R.M., Bryan, J.G., McFarland, J.M., Weir, B.A., Sizemore, A.E., Xu, H., Dharia, N.V., Montgomery, P .G., Cowley, G.S., Pantel, S., et al. (2017). Computational correction of copy number effect improves specificity of CRISPR-Cas9 essentiality screens in cancer cells. Nat. Genet. 49, 1779–1784. 17. Sanson, K.R., Hanna, R.E., Hegde, M., Donovan, K.F ., Strand, C., Sullender, M.E., Vaimberg, E.W., Goodale, A., Root, D.E., Piccioni, F ., et al. (2018). Optimized libraries for CRISPR-Cas9 genetic screens with multiple modalities. Nat. Commun. 9, 5416. 18. Holland, C.H., Szalai, B., and Saez-Rodriguez, J. (2020). Transfer of regulatory knowledge from human to mouse for functional genomics analysis. Biochim. Biophys. Acta Gene Regul. Mech. 1863, 194431. 19. Zheng, X., Dumitru, R., Lackford, B.L., Freudenberg, J.M., Singh, A.P ., Archer, T .K., Jothi, R., and Hu, G. (2012). Cnot1, Cnot2, and Cnot3 maintain mouse and human ESC identity and inhibit extraembryonic differentiation. Stem Cells 30, 910–922. 20. Park, J., Park, S., and Lee, J.-S. (2023). Role of the Paf1 complex in the maintenance of stem cell pluripotency and development. FEBS J. 290, 951–961. 21. Replogle, J.M., Norman, T .M., Xu, A., Hussmann, J.A., Chen, J., Cogan, J.Z., Meer, E.J., T erry, J.M., Riordan, D.P ., Srinivas, N., et al. (2020). Combinatorial single-cell CRISPR screens by direct guide RNA capture and targeted sequencing. Nat. Biotechnol. 38, 954–961. 22. Ashburner, M., Ball, C.A., Blake, J.A., Botstein, D., Butler, H., Cherry, J.M., Davis, A.P ., Dolinski, K., Dwight, S.S., Eppig, J.T ., et al. (2000). Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat. Genet. 25, 25–29. 23. Siepel, A., Bejerano, G., Pedersen, J.S., Hinrichs, A.S., Hou, M., Rosenbloom, K., Clawson, H., Spieth, J., Hillier, L.W., Richards, S., et al. (2005). Evolutionarily conserved elements in vertebrate, insect, worm, and yeast genomes. Genome Res. 15, 1034–1050. 24. Bergmiller, T ., Ackermann, M., and Silander, O.K. (2012). Patterns of evolutionary conservation of essential genes correlate with their compensability. PLoS Genet. 8, e1002803. 25. Liberzon, A., Birger, C., Thorvaldsdóttir, H., Ghandi, M., Mesirov, J.P ., and T amayo, P . (2015). The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell Syst 1, 417–425. 36 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint 26. Luo, H., Gao, F ., and Lin, Y . (2015). Evolutionary conservation analysis between the essential and nonessential genes in bacterial genomes. Sci. Rep. 5, 13210. 27. Wainberg, M., Kamber, R.A., Balsubramani, A., Meyers, R.M., Sinnott-Armstrong, N., Hornburg, D., Jiang, L., Chan, J., Jian, R., Gu, M., et al. (2021). A genome-wide atlas of co-essential modules assigns function to uncharacterized genes. Nat. Genet. 53, 638–649. 28. Costanzo, M., VanderSluis, B., Koch, E.N., Baryshnikova, A., Pons, C., T an, G., Wang, W., Usaj, M., Hanchard, J., Lee, S.D., et al. (2016). A global genetic interaction network maps a wiring diagram of cellular function. Science 353. https://doi.org/10.1126/science.aaf1420. 29. T sai, S.Q., and Joung, J.K. (2016). Defining and improving the genome-wide specificities of CRISPR-Cas9 nucleases. Nat. Rev. Genet. 17, 300–312. 30. Doench, J.G. (2018). Am I ready for CRISPR? A user’s guide to genetic screens. Nat. Rev. Genet. 19, 67–80. 31. Zukeran, A., T akahashi, A., T akaoka, S., Mohamed, H.M.A., Suzuki, T ., Ikematsu, S., and Yamamoto, T . (2016). The CCR4-NOT deadenylase activity contributes to generation of induced pluripotent stem cells. Biochem. Biophys. Res. Commun. 474, 233–239. 32. Fazzio, T .G., Huff, J.T ., and Panning, B. (2008). Chromatin regulation Tip(60)s the balance in embryonic stem cell self-renewal. Cell Cycle 7, 3302–3306. 33. Huang, W., Chen, T .-Q., Fang, K., Zeng, Z.-C., Ye, H., and Chen, Y .-Q. (2021). N6-methyladenosine methyltransferases: functions, regulation, and clinical potential. J. Hematol. Oncol. 14, 117. 34. Poli, J., Gasser, S.M., and Papamichos-Chronakis, M. (2017). The INO80 remodeller in transcription, replication and repair. Philos. Trans. R. Soc. Lond. B Biol. Sci. 372. https://doi.org/10.1098/rstb.2016.0290. 35. UniProt Consortium (2015). UniProt: a hub for protein information. Nucleic Acids Res. 43, D204–D212. 36. Jumper, J., Evans, R., Pritzel, A., Green, T ., Figurnov, M., Ronneberger, O., Tunyasuvunakool, K., Bates, R., Žídek, A., Potapenko, A., et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589. 37. Fusaro, G., Dasgupta, P ., Rastogi, S., Joshi, B., and Chellappan, S. (2003). Prohibitin induces the transcriptional activity of p53 and is exported from the nucleus upon apoptotic signaling. J. Biol. Chem. 278, 47853–47861. 38. Basu, S., and Wallner, B. (2016). DockQ: A Quality Measure for Protein-Protein Docking Models. PLoS One 11, e0161879. 39. T sitsiridis, G., Steinkamp, R., Giurgiu, M., Brauner, B., Fobo, G., Frishman, G., Montrone, C., and Ruepp, A. (2023). CORUM: the comprehensive resource of mammalian protein complexes-2022. Nucleic Acids Res. 51, D539–D545. 40. Kagey, M.H., Newman, J.J., Bilodeau, S., Zhan, Y ., Orlando, D.A., van Berkum, N.L., Ebmeier, C.C., Goossens, J., Rahl, P .B., Levine, S.S., et al. (2010). Mediator and cohesin connect gene expression and chromatin architecture. Nature 467, 430–435. 37 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint 41. Ramasamy, S., Aljahani, A., Karpinska, M.A., Cao, T .B.N., Velychko, T ., Cruz, J.N., Lidschreiber, M., and Oudelaar, A.M. (2023). The Mediator complex regulates enhancer-promoter interactions. Nat. Struct. Mol. Biol. 30, 991–1000. 42. Qin, K., Jian, D., Xue, Y ., Cheng, Y ., Zhang, P ., Wei, Y ., Zhang, J., Xiong, H., Zhang, Y ., and Yuan, X. (2021). DDX41 regulates the expression and alternative splicing of genes involved in tumorigenesis and immune response. Oncol. Rep. 45, 1213–1225. 43. Andreou, A.Z. (2021). DDX41: a multifunctional DEAD-box protein involved in pre-mRNA splicing and innate immunity. Biol. Chem. 402, 645–651. 44. Shinriki, S., Hirayama, M., Nagamachi, A., Yokoyama, A., Kawamura, T ., Kanai, A., Kawai, H., Iwakiri, J., Liu, R., Maeshiro, M., et al. (2022). DDX41 coordinates RNA splicing and transcriptional elongation to prevent DNA replication stress in hematopoietic cells. Leukemia 36, 2605–2620. 45. Dybkov, O., Preußner, M., El Ayoubi, L., Feng, V.-Y ., Harnisch, C., Merz, K., Leupold, P ., Yudichev, P ., Agafonov, D.E., Will, C.L., et al. (2023). Regulation of 3’ splice site selection after step 1 of splicing by spliceosomal C* proteins. Sci Adv 9, eadf1785. 46. Lee, J.W., Ko, J., Ju, C., and Eltzschig, H.K. (2019). Hypoxia signaling in human diseases and therapeutic targets. Exp. Mol. Med. 51, 1–13. 47. Haase, V.H. (2009). The VHL tumor suppressor: master regulator of HIF . Curr. Pharm. Des. 15, 3895–3903. 48. Mullen, P .J., Yu, R., Longo, J., Archer, M.C., and Penn, L.Z. (2016). The interplay between cell signalling and the mevalonate pathway in cancer. Nat. Rev. Cancer 16, 718–731. 49. Cuomo, A.S.E., Seaton, D.D., McCarthy, D.J., Martinez, I., Bonder, M.J., Garcia-Bernardo, J., Amatya, S., Madrigal, P ., Isaacson, A., Buettner, F ., et al. (2020). Single-cell RNA-sequencing of differentiating iPS cells reveals dynamic genetic effects on gene expression. Nat. Commun. 11, 810. 50. Novak, G., Kyriakis, D., Grzyb, K., Bernini, M., Rodius, S., Dittmar, G., Finkbeiner, S., and Skupin, A. (2022). Single-cell transcriptomics of human iPSC differentiation dynamics reveal a core molecular network of Parkinson’s disease. Commun Biol 5, 49. 51. Shakiba, N., White, C.A., Lipsitz, Y .Y ., Yachie-Kinoshita, A., T onge, P .D., Hussein, S.M.I., Puri, M.C., Elbaz, J., Morrissey-Scoot, J., Li, M., et al. (2015). CD24 tracks divergent pluripotent states in mouse and human cells. Nat. Commun. 6, 7329. 52. Liu, Q., Wang, G., Lyu, Y ., Bai, M., Jiapaer, Z., Jia, W., Han, T ., Weng, R., Yang, Y ., Yu, Y ., et al. (2018). The miR-590/Acvr2a/T erf1 Axis Regulates T elomere Elongation and Pluripotency of Mouse iPSCs. Stem Cell Reports 11, 88–101. 53. Hough, S.R., Laslett, A.L., Grimmond, S.B., Kolle, G., and Pera, M.F . (2009). A continuum of cell states spans pluripotency and lineage commitment in human embryonic stem cells. PLoS One 4, e7708. 54. Emani, M.R., Närvä, E., Stubb, A., Chakroborty, D., Viitala, M., Rokka, A., Rahkonen, N., Moulder, R., Denessiouk, K., Trokovic, R., et al. (2015). The L1TD1 protein interactome reveals the importance of post-transcriptional regulation in human pluripotency. Stem Cell Reports 4, 519–528. 38 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint 55. Iwabuchi, K.A., Yamakawa, T ., Sato, Y ., Ichisaka, T ., T akahashi, K., Okita, K., and Yamanaka, S. (2011). ECAT11/L1td1 is enriched in ESCs and rapidly activated during iPSC generation, but it is dispensable for the maintenance and induction of pluripotency. PLoS One 6, e20461. 56. Wongtrakoongate, P ., Li, J., and Andrews, P .W. (2014). DNMT3B inhibits the re-expression of genes associated with induced pluripotency. Exp. Cell Res. 321, 231–239. 57. T akahashi, K., and Yamanaka, S. (2006). Induction of pluripotent stem cells from mouse embryonic and adult fibroblast cultures by defined factors. Cell 126, 663–676. 58. Li, M., and Belmonte, J.C.I. (2017). Ground rules of the pluripotency gene regulatory network. Nat. Rev. Genet. 18, 180–191. 59. Riordan, J.D., and Nadeau, J.H. (2017). From Peas to Disease: Modifier Genes, Network Resilience, and the Genetics of Health. Am. J. Hum. Genet. 101, 177–191. 60. Doench, J.G., Fusi, N., Sullender, M., Hegde, M., Vaimberg, E.W., Donovan, K.F ., Smith, I., T othova, Z., Wilen, C., Orchard, R., et al. (2016). Optimized sgRNA design to maximize activity and minimize off-target effects of CRISPR-Cas9. Nat. Biotechnol. 34, 184–191. 61. Rohatgi, N., Fortin, J.P ., Lau, T ., Ying, Y ., Zhang, Y ., Costa, M., and Reja, R. (2024). Seed sequences mediate off-target activity in the CRISPR-interference (CRISPRi) system. bioRxiv, 2024.04.10.588881. https://doi.org/10.1101/2024.04.10.588881. 62. Parts, L., Batté, A., Lopes, M., Yuen, M.W., Laver, M., Luis, B.S., Yue, J., Pons, C., Eray, E., Aloy, P ., et al. (2021). Natural variants suppress mutations in hundreds of essential genes. Mol. Syst. Biol. https://doi.org/10.15252/msb.202010138. 63. Ünlü, B., Pons, C., Ho, U.L., Batté, A., Aloy, P ., and van Leeuwen, J. (2023). Global analysis of suppressor mutations that rescue human genetic defects. Genome Med. 15. https://doi.org/10.1186/s13073-023-01232-0. 64. Usluer, S., Hallast, P ., Crepaldi, L., Zhou, Y ., Urgo, K., Dincer, C., Su, J., Noell, G., Alasoo, K., El Garwany, O., et al. (2023). Optimized whole-genome CRISPR interference screens identify ARID1A-dependent growth regulators in human induced pluripotent stem cells. Stem Cell Reports 18, 1061–1074. 65. Butler, A., Hoffman, P ., Smibert, P ., Papalexi, E., and Satija, R. (2018). Integrating single-cell transcriptomic data across different conditions, technologies, and species. Nat. Biotechnol. 36, 411–420. 66. Tzelepis, K., Koike-Yusa, H., De Braekeleer, E., Li, Y ., Metzakopian, E., Dovey, O.M., Mupo, A., Grinkevich, V., Li, M., Mazan, M., et al. (2016). A CRISPR Dropout Screen Identifies Genetic Vulnerabilities and Therapeutic T argets in Acute Myeloid Leukemia. Cell Rep. 17, 1193–1205. 67. Yusa, K., Zhou, L., Li, M.A., Bradley, A., and Craig, N.L. (2011). A hyperactive piggyBac transposase for mammalian applications. Proc. Natl. Acad. Sci. U. S. A. 108, 1531–1536. 68. Ripley, B.D. (2001). The R project in statistical computing. MSOR Connect. 1, 23–25. 69. Zheng, G.X.Y ., T erry, J.M., Belgrader, P ., Ryvkin, P ., Bent, Z.W., Wilson, R., Ziraldo, 39 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint S.B., Wheeler, T .D., McDermott, G.P ., Zhu, J., et al. (2017). Massively parallel digital transcriptional profiling of single cells. Nat. Commun. 8, 14049. 70. Huang, X., and Huang, Y . (2021). Cellsnp-lite: an efficient tool for genotyping single cells. Bioinformatics. https://doi.org/10.1093/bioinformatics/btab358. 71. Danecek, P ., Bonfield, J.K., Liddle, J., Marshall, J., Ohan, V., Pollard, M.O., Whitwham, A., Keane, T ., McCarthy, S.A., Davies, R.M., et al. (2021). Twelve years of SAMtools and BCFtools. Gigascience 10. https://doi.org/10.1093/gigascience/giab008. 72. CellSNP - browse /SNPlist at SourceForge.Net https://sourceforge.net/projects/cellsnp/files/SNPlist/. 73. Huang, Y ., McCarthy, D.J., and Stegle, O. (2019). Vireo: Bayesian demultiplexing of pooled single-cell RNA-seq data without genotype reference. Genome Biol. 20, 273. 74. Zhao, H., Sun, Z., Wang, J., Huang, H., Kocher, J.-P ., and Wang, L. (2014). CrossMap: a versatile tool for coordinate conversion between genome assemblies. Bioinformatics 30, 1006–1007. 75. Luecken, M.D., and Theis, F .J. (2019). Current best practices in single‐ cell RNA‐seq analysis: a tutorial. Mol. Syst. Biol. 15, e8746. 76. Braunger, J.M., and Velten, B. (2024). Guide assignment in single-cell CRISPR screens using crispat. bioRxiv, 2024.05.06.592692. https://doi.org/10.1101/2024.05.06.592692. 77. Bates, D., Mächler, M., Bolker, B., and Walker, S. (2015). Fitting Linear Mixed-Effects Models Using lme4. J. Stat. Softw. 67, 1–48. 78. Abugessaisa, I., Noguchi, S., Carninci, P ., and Kasukawa, T . (2017). The FANTOM5 Computation Ecosystem: Genomic Information Hub for Promoters and Active Enhancers. Methods Mol. Biol. 1611, 199–217. 79. Hodgkins, A., Farne, A., Perera, S., Grego, T ., Parry-Smith, D.J., Skarnes, W.C., and Iyer, V. (2015). WGE: a CRISPR database for genome engineering. Bioinformatics 31, 3078–3080. 80. Lab, S.-R. OmniPath :: Intra- & intercellular signaling knowledge. https://omnipathdb.org/. 81. Türei, D., Korcsmáros, T ., and Saez-Rodriguez, J. (2016). OmniPath: guidelines and gateway for literature-curated signaling pathway resources. Nat. Methods. 82. Garcia-Alonso, L., Holland, C.H., Ibrahim, M.M., Turei, D., and Saez-Rodriguez, J. (2019). Benchmark and integration of resources for the estimation of human transcription factor activities. Genome Res. 29, 1363–1375. 83. Badia-i-Mompel, P ., Vélez Santiago, J., Braunger, J., Geiss, C., Dimitrov, D., Müller-Dott, S., T aus, P ., Dugourd, A., Holland, C.H., Ramirez Flores, R.O., et al. (2022). decoupleR: ensemble of computational methods to infer biological activities from omics data. Bioinformatics Advances 2, vbac016. 84. Raudvere, U., Kolberg, L., Kuzmin, I., Arak, T ., Adler, P ., Peterson, H., and Vilo, J. (2019). g:Profiler: a web server for functional enrichment analysis and conversions of gene lists (2019 update). Nucleic Acids Res. 47, W191–W198. 85. Friedman, J., Hastie, T ., and Tibshirani, R. (2010). Regularization Paths for Generalized 40 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint Linear Models via Coordinate Descent. J. Stat. Softw. 33, 1–22. 86. Evans, R., O’Neill, M., Pritzel, A., Antropova, N., Senior, A., Green, T ., Žídek, A., Bates, R., Blackwell, S., Yim, J., et al. (2022). Protein complex prediction with AlphaFold-Multimer. bioRxiv, 2021.10.04.463034. https://doi.org/10.1101/2021.10.04.463034. 87. Bryant, P ., Pozzati, G., and Elofsson, A. (2022). Improved prediction of protein-protein interactions using AlphaFold2. Nat. Commun. 13, 1265. 88. PyMOL http://www.pymol.org/pymol. 89. Argelaguet, R., Velten, B., Arnol, D., Dietrich, S., Zenz, T ., Marioni, J.C., Buettner, F ., Huber, W., and Stegle, O. (2018). Multi-Omics Factor Analysis-a framework for unsupervised integration of multi-omics data sets. Mol. Syst. Biol. 14, e8124. 90. Argelaguet, R., Arnol, D., Bredikhin, D., Deloro, Y ., Velten, B., Marioni, J.C., and Stegle, O. (2020). MOFA+: a statistical framework for comprehensive integration of multi-modal single-cell data. Genome Biol. 21, 1–17. 91. Kokic, G., Wagner, F .R., Chernev, A., Urlaub, H., and Cramer, P . (2021). Structural basis of human transcription–DNA repair coupling. Nature 598, 368–372. 92. Kenner, L.R., Anand, A.A., Nguyen, H.C., Myasnikov, A.G., Klose, C.J., McGeever, L.A., T sai, J.C., Miller-Vedam, L.E., Walter, P ., and Frost, A. (2019). eIF2B-catalyzed nucleotide exchange and phosphoregulation by the integrated stress response. Science 364, 491–495. 41 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 28, 2024. ; https://doi.org/10.1101/2024.11.28.625833doi: bioRxiv preprint