Non-additive effects of schizophrenia risk genes reflect convergent downstream function

Abstract

Genetic studies of schizophrenia (SCZ) reveal a complex polygenic risk architecture comprised of hundreds of risk variants, the majority of which are common in the population at-large and confer only modest increases in disorder risk. Precisely how genetic variants with individually small predicted effects on gene expression combine to yield substantial clinical impacts in aggregate is unclear. Towards this, we previously reported that the combinatorial perturbation of four SCZ risk genes ("eGenes", whose expression is regulated by common variants) resulted in gene expression changes that were not predicted by individual perturbations, being most non-additive among genes associated with synaptic function and SCZ risk. Now, across fifteen SCZ eGenes, we demonstrate that non-additive effects are greatest within groups of functionally similar eGenes. Individual eGene perturbations reveal common downstream transcriptomic effects ("convergence"), while combinatorial eGene perturbations result in changes that are smaller than predicted by summing individual eGene effects ("sub-additive effects"). Unexpectedly, these convergent and sub-additive downstream transcriptomic effects overlap and constitute a large proportion of the genome-wide polygenic risk score, suggesting that functional redundancy of eGenes may be a major mechanism underlying non-additivity. Single eGene perturbations likewise fail to predict the magnitude or directionality of cellular phenotypes resulting from combinatorial perturbations. Overall, our results indicate that polygenic risk cannot be extrapolated from experiments testing one risk gene at a time and must instead be empirically measured. By unravelling the interactions between complex risk variants, it may be possible to improve the clinical utility of polygenic risk scores through more powerful prediction of symptom onset, clinical trajectory, and treatment response, or to identify novel targets for therapeutic intervention.

Keywords: CRISPR; human induced pluripotent stem cells; neurons; psychiatric genomics; schizophrenia.

Figure 1.. Prioritization and manipulation of synaptic, regulatory, and multi-function brain eGenes regulated by SCZ.

A. Schematic of SCZ eGene identification and prioritization. Fine-mapped GWAS loci were co-localized with post-mortem brain eQTLs using COLOC, identifying 25 candidate SCZ eGenes. Transcriptomic imputation using prediXcan identified a further ~250 significant genes brain-specific genetically regulated gene expression (GREX) predicted to be associated with SCZ. 22/25 eGenes identified using COLOC overlapped with the prediXcan gene list. For further study, 15 of these eGenes were separated into three functional groups based on gene ontology annotations. B. Predicted GREX levels in dorsolateral prefrontal cortex (DLPFC) calculated for the fifteen eGenes in a Swedish SCZ cohort. Aberrant expression of eGenes is predicted to impact SCZ case-control status in a dose-dependent manner. C. Summary schematic of experimental set up interrogating functions of eGenes. Individual and joint perturbation of eGenes using CRISPRa and RNAi in hiPSC-derived glutamatergic neuron cultures was followed by neuronal phenotyping using three modalities: RNA-seq, high content imaging and multi-electrode array recording.

Figure 2.. Perturbation of SCZ eGenes results in differential expression of genes relating to brain disorders and synaptic function.

A. Gene set enrichment analysis (GSEA) performed across a collection of 698 manually curated gene-sets with a neural theme revealed enrichments of gene-sets related to brain disorders and synaptic functions across 12/15 eGene perturbations. B. Across brain disorder-related gene-sets, 5/5 Synaptic eGene perturbations showed strong enrichment of other genes relating to schizophrenia risk, including both common variant-linked genes and copy number variant (CNV) genes associated with schizophrenia. C. Across presynaptic function gene-sets, 10/15 eGene perturbations showed enrichment of genes relating to synaptic vesicle localization and transport. D. Across postsynaptic function gene-sets, 8/15 eGene perturbations showed enrichment of genes relating to components of glutamatergic neurotransmission. AUD = alcohol use disorder, BD = bipolar disorder, FMRP = Fragile X Mental Retardation Protein, ID = intellectual disability, PTSD = post traumatic stress disorder, SCZ = schizophrenia, DEGs = differentially expressed genes.

Figure 3.. Perturbation of SCZ eGenes within functional categories results in non-additive effects on transcription impacting expression of genes linked to brain disorders and synaptic function.

A. Schematic of differential expression analysis. Individual eGene perturbations, the implementation of the expected additive model based on the latter and the measured combinatorial perturbation permitting the detection of interactive effects through comparison with the additive model. B. Combinatorial perturbation of eGenes within, but not across functional pathways resulted in non-additive effects on expression across 16.8% (Synaptic eGenes) and 20.2% (Regulatory eGenes) of the transcriptome. C. Following joint perturbation of the Synaptic eGene set, the majority (>95%) of non-additive genes showed significantly less differential expression in the measured combinatorial perturbation relative to the expected additive model (“less downregulated” and “less upgregulated” categories). D. GSEA of non-additive genes in the Synaptic eGene set demonstrated significant enrichment for genes relating to brain disorders and synaptic function. SCZ = schizophrenia, CNV = copy number variant, FMRP = Fragile X Mental Retardation Protein, FDR = false discovery rate. See Box 1 for example schematics and explanation for each sub-category of non-additive effect.

Figure 4.. Convergence accounts for non-additive effects within functional pathways.

A-E. Meta-analysis of differentially expressed genes (DEGs) elicited by individual eGene perturbations for each five-gene grouping using METAL to identify DEGs that showed altered expression consistently in the same direction across all five eGene perturbation conditions for each set of eGenes. A. Convergence across individual eGene perturbations is correlated with the degree of non-additive effect seen in the corresponding joint perturbation condition. Pearson’s r² = 0.6569, p=0.0147. B. For each joint eGene perturbation group, non-additive impacts on transcription were compared with genes showing significant convergence across individual perturbations for the same eGene set. C. Evidence of convergence was found in 1070 genes across the synaptic eGene perturbations, 761 of which also exhibited non-additive effects in the additive-combinatorial comparison for the same set. D. Evidence of convergence was found in 1070 genes across the regulatory eGene perturbations, 1000 of which also exhibited non-additive effects in the additive-combinatorial comparison for the same set. E. No significant non-additive effects and only minimal convergence could be seen in eGene perturbations across functional pathways. F. GSEA of convergent genes in the Synaptic eGene set demonstrated significant enrichment for genes relating to brain disorders and presynaptic functions. G. Bayesian biclustering identified significant convergence of co-expression networks unique to synaptic pathways that replicated in over 25% of iterations. Major node genes mediating convergent networks of synaptic eGenes included OR10G6, an olfactory receptor gene, and MIR495, which has previously been implicated in Alzheimer’s disease. H. GSEA of co-expressed network genes in the Synaptic eGene set demonstrated significant enrichment for genes relating to brain inflammation and schizophrenia risk. GSEA = gene-set enrichment analysis; SCZ = schizophrenia; CNV = copy number variant; PPI = protein-protein interaction; EAE = Experimental Autoimmune Encephalomyelitis; FDR = false discovery rate.

Figure 5.. Within and across-pathway synergistic genes are associated with SCZ risk.

Left: phenotypic variance explained (R²) by genome-wide (purple), synaptic (magenta), regulatory (blue) and random (green) pathway PRS. The size of the dot represents number of genes included in each pathway/gene set. The P next to each dot represents the empirical “competitive” P-value calculated by PRSet to evaluate pathway enrichment. Right: phenotypic variance explained (R²) normalized by the number of genes within each pathway.

Figure 6.. Combinatorial perturbation of SCZ eGenes within functional categories results in impaired neurite outgrowth, synaptic expression and neuronal hyperactivity.

A. Combinatorial perturbation of all five synaptic eGenes in D7 iGLUTs resulted in significant reduction of neurite outgrowth relative to a combinatorial scramble gRNA + shRNA control. B. Combinatorial perturbation of all five synaptic eGenes in D21 iGLUTs resulted in significant reduction of Syn1+ puncta density relative to a combinatorial scramble gRNA + shRNA control. C. LOESS plots; combinatorial perturbation of eGenes within but not across functional categories results in transient neuronal hyperactivity in MEA recordings of D28–42 iGLUTs. D. Summary heatmap of neurite outgrowth, puncta density and neuronal activity data for within and across function eGene perturbations. N = minimum of 2 independent experiments across 2 donor lines with 10-12 technical replicates per condition. One-way ANOVA with post-hoc Bonferonni multiple comparisons test. * = p<0.05; ** = p<0.01; *** = p<0.001; **** = p<0.0001.