Functional implications of polygenic risk for schizophrenia in human neurons

We predicted that eGenes linked to schizophrenia would share substantial downstream transcriptomic changes with a common direction of effect (termed “convergence”). Although convergence has been described in the context of loss-of-function autism spectrum disorder risk genes, these rare mutations almost never co-occur in the same individual. The convergent impact of common variants which are frequently inherited together, and the impacts of which are apparent only in aggregate remain unknown. We targeted twenty-one schizophrenia eGenes in iGLUTs using pooled and arrayed CRISPR-based approaches, significantly perturbing seventeen (CALN1, CLCN3, DOC2A, FES, FURIN, GATAD2A, NAGA, PCCB, PLCL1, THOC7, TMEM219, SF3B1, SNAP91, SNCA, UBE2Q2L, ZNF823, ZNF804A), and resolving convergent impacts robust to experimental and donor effects. To test if convergence influenced the outcome when eGenes were inherited in combination (i.e. if eGene effects sum linearly according to the additive model), we compared manipulation of eGenes one at a time and in groups defined by annotated functions at the synapse (“synaptic”: SNAP91, CLCN3, PLCL1, DOC2A, SNCA), or regulating transcription (“regulatory”: ZNF823, INO80E, SF3B1, THOC7, GATAD2A), or with un-related non-synaptic, non-regulatory biology (“multi-function”: CALN1, CUL9, TMEM219, PCCB, FURIN), and random combinations thereof. Altogether, with broad relevance across complex polygenic disease our work begins to experimentally determine answers to the long-standing question of how risk variants interact in human neurons. Overall design: Pooled CRISPR screening combined single-cell RNA sequencing readouts and direct detection of sgRNAs. Two independently designed, constructed, and validated pooled CRISPRa libraries were transduced into iGLUTs from two donors in independent experiments at unique developmental time-points (DIV7 or DIV21). To identify groups of genes with similar expression patterns across eGene perturbations we define “convergent networks” as relationships between genes that are co-regulated by shared biological mechanisms. To study the strength and composition of convergent networks, we define “network convergence” as the sum of the network connectivity score (i.e., networks with fewer nodes and more interconnectedness have increased convergence). We endeavored to identify the biological factors (e.g., number of eGenes, functional similarity of eGenes, and eGene co-expression) that influenced network convergence. eGene number tested the number of eGenes used to generate a convergent network. Functional similarity (i.e., the degree of shared biological functions amongst eGenes) was calculated two ways: Gene Ontology semantic similarity scoring (within biological pathway, cellular component, and molecular function), and synaptic/signaling score (proportion of eGenes with annotated function as either “signaling” for pooled or “synaptic” for arrayed) We next manipulated eGenes in combination to approximate the polygenic nature of schizophrenia and test if convergence between eGenes influences observed effects. Given that genes implicated in synaptic biology and epigenetic/transcriptional regulation are enriched for the schizophrenia risk we sought to generate three groups of eGenes, linked to synaptic biology, gene regulation, or neither (Fig. 1A, arrayed experiment). Unconstrained by the unidirectionality of pooled CRISPR screens, we did not restrict our list to eGenes with a single direction of effect. From the 18 coding genes prioritized by the intersection of transcriptomic imputation and colocalization, eGenes were separated into discrete functional categories based on gene ontology annotations. Our final gene list included five synaptic genes (SNAP91, CLCN3, PLCL1, DOC2A, SNCA), five regulatory genes (ZNF823, INO80E, SF3B1, THOC7, GATAD2A), and five genes with non-synaptic, non-regulatory functions, termed “multi-function” (CALN1, CUL9, TMEM219, PCCB, FURIN). We then applied an arrayed design (i.e., distinct conditions in each well) to manipulate schizophrenia eGenes alone and in combination, allowing us to capture cell autonomous and non-cell autonomous effects in a manner not possible in the pooled design. Three to five vectors per gene were tested in 7-day-old (D7) iGLUTs, identifying the single vector that best achieved the level of significant perturbation predicted by eQTL analyses as confirmed by qPCR. Each eGene was perturbed in 21-day-old (D21) iGLUTs for 72 hours, individually and jointly, including appropriate vector and scrambled controls, from two neurotypical donors with average polygenic risk scores (one experimental batch per donor). Three groups of five random genes, one group of ten random genes, and one group of all fifteen genes were also included. Significant (p<0.05) changes in eGene expression in iGLUTs were confirmed by RNAseq in 13/15 eGenes (SNAP91, CLCN3, PLCL1, DOC2A, SNCA, ZNF823, SF3B1, THOC7, GATAD2A, CALN1, TMEM219, PCCB, FURIN) (SI Fig. 1G, SI Fig. 19A); we validated the magnitude and direction of experimental eGene perturbation relative to the dosage effects of the top predicted causal SNPs (e.g. eQTL effect size) and predicted eGene expression changes.