Beyond Cis-Regulation: How Coexpression Networks Reveal the Trans-Regulatory Architecture of Schizophrenia Risk
The field of psychiatric genetics has reached an inflection point. While genome-wide association studies (GWAS) have identified hundreds of loci associated with schizophrenia, translating these statistical associations into mechanistic insights remains challenging. Most risk variants reside in non-coding regions, presumably affecting gene regulation, yet traditional transcriptome-wide association studies (TWAS) have largely confined themselves to cis-regulatory effects, variants within ±1 Mb of their target genes. This approach, while tractable, captures only a fraction of expression heritability and overlooks the coordinated transcriptional programs that likely underlie complex psychiatric phenotypes.
A new study from the Lieber Institute for Brain Development, led by Rossi, Pergola, and colleagues, fundamentally challenges this cis-centric paradigm. By developing coexpression-informed frameworks that explicitly model trans-regulatory effects, the research team has identified 766 genes associated with schizophrenia in the Psychiatric Genomics Consortium wave 3 (PGC3) dataset, with 641 representing novel associations not captured by conventional TWAS approaches. This work demonstrates that incorporating distal regulatory mechanisms substantially expands our ability to map genetically regulated expression to disease risk.
Methodological Innovation: INGENE and MODULE
The study introduces two complementary frameworks for capturing trans-regulatory effects. INGENE (Imputed Network Gene-Expression trans-eQTL) models how cis-regulated coexpression partners collectively influence a target gene’s expression. Essentially, if gene A is predicted by its local variants and consistently coexpresses with gene B, INGENE uses A’s genetically predicted expression as a predictor for B. This approach leverages the principle that cis-mediated trans-effects, where a variant affects a nearby gene that subsequently influences distant genes, represent a substantial component of trans-regulation.
MODULE (MODule quantitative trait Loci Eigengene) takes a different tack, identifying SNPs associated with the first principal component (eigengene) of gene coexpression modules. Rather than propagating individual partner effects, MODULE captures variants that coordinately regulate entire functional gene sets. Importantly, both frameworks exclude cis-SNPs (±1 Mb) from the target gene to ensure that detected signals represent genuine trans-effects rather than residual local regulation.
The researchers trained these models using RNA-seq data from six post-mortem brain regions in the LIBD collection, then validated predictions in independent GTEx and CommonMind Consortium datasets. This cross-dataset replication strategy, requiring positive correlation between predictions from different training cohorts, substantially reduced false positives while maintaining biological signal. The approach yielded 18,403 INGENE-predictable genes and 16,721 MODULE-predictable genes across regions, compared to ~26,000 for standard elastic-net (CIS) and EpiXcan models.
Critically, the study demonstrates that cis and trans components are complementary rather than redundant. While cis-models explained higher per-gene variance (mean adjusted R² of 7-9% in GTEx), trans-models substantially expanded gene coverage. Integration of all four predictors (CIS, EpiXcan, INGENE, MODULE) through maximum likelihood estimation significantly improved prediction accuracy for 18,744 genes, with trans components providing significant incremental explanatory power beyond cis-only models.
Key Findings: Trans-Regulation and Disease Architecture
Applying this integrated framework to 102,613 PGC3 participants revealed several important insights into schizophrenia’s genetic architecture. First, trans-predicted genes showed stronger enrichment for disease-associated variants than cis-predicted genes. MODULE-derived SNP weights correlated more strongly with PGC3 effect sizes (r = 0.28-0.42) than CIS or EpiXcan weights (r = 0.14-0.18), suggesting that network-level regulatory variants may be preferentially enriched for pathogenic effects.
Second, the 766 identified genes converged on biologically interpretable pathways with regional specificity. In the dorsolateral prefrontal cortex, associations enriched for AMPA receptor trafficking, clathrin-mediated endocytosis, and synaptic vesicle dynamics—processes central to excitatory neurotransmission and synaptic plasticity. Across multiple regions, antigen processing and MHC class I pathways showed consistent upregulation, supporting neuroimmune hypotheses of schizophrenia pathogenesis while also implicating MHC molecules’ non-canonical roles in synaptic pruning and receptor trafficking.
Third, cell-type enrichment analysis revealed that disease-associated genes preferentially localized to excitatory neurons (particularly CA3 pyramidal and dentate gyrus-like populations) and GABAergic interneurons, with regional variation in directionality. The DLPFC showed upregulation in excitatory populations and downregulation in inhibitory neurons, consistent with excitation-inhibition imbalance models.
Fourth, approximately 72% of significant associations mapped outside genome-wide significant GWAS loci, with trans-only predictions particularly enriched in these distal regions. This suggests that coexpression-informed approaches can identify risk genes whose regulatory variants fall below GWAS detection thresholds or exert effects through complex, multi-gene mechanisms not captured by proximity-based gene mapping.
Advancing the Field: Implications for Psychiatric Genetics
This work addresses several longstanding challenges in psychiatric genetics. The “missing heritability” problem the gap between twin-based heritability estimates and variance explained by identified variants, partly reflects our incomplete understanding of regulatory architecture. By demonstrating that trans-effects substantially contribute to genetically regulated expression and disease risk, this study provides a framework for capturing previously inaccessible components of expression heritability.
The findings also inform ongoing debates about omnigenic versus core gene models of complex traits. The observation that trans-predicted genes cluster within coexpression networks enriched for schizophrenia-associated variants supports a model where peripheral genes contribute to disease risk through their network connections to core pathogenic processes. The enrichment of transcription factors among genes whose cis-variants serve as MODULE trans-eQTLs suggests potential hierarchical regulatory structures worthy of further investigation.
Methodologically, the cross-dataset validation strategy employed here offers a template for improving reproducibility in eQTL-based studies. The requirement that trans-predictions replicate across independent training cohorts substantially reduced false positives while maintaining biological signal, as evidenced by the 88% overlap of MODULE predictions between CMC and GTEx despite these datasets’ modest sample sizes.
Future Directions: From Association to Mechanism
Future Directions: From Association to Mechanism
Several research avenues emerge from this work. First, integrating single-cell eQTL data could refine cell-type-specific regulatory networks and resolve whether observed trans-effects reflect within-cell-type regulation or cross-cell-type signaling. Recent large-scale single-nucleus studies in human brain provide the necessary resolution for such analyses.
Second, incorporating temporal dynamics—particularly developmental trajectories—could illuminate when and how risk variants exert their effects. The observation that schizophrenia-associated genes show age-dependent coexpression patterns suggests that trans-regulatory effects may be developmentally regulated, with implications for critical period interventions.
Third, experimental validation of predicted trans-regulatory relationships through CRISPR perturbation screens or induced pluripotent stem cell models could establish causality and identify druggable nodes within disease-relevant networks. The identification of specific transcription factors enriched among trans-regulatory variants provides concrete starting points for such experiments.
Fourth, extending these approaches to other psychiatric and neurological disorders could reveal shared versus disorder-specific regulatory architectures. The authors demonstrate preliminary enrichment for bipolar disorder and major depression, suggesting broader applicability.
Finally, integrating coexpression-informed TWAS with other functional genomic modalities—chromatin accessibility, histone modifications, 3D genome organization—could provide a more complete picture of how non-coding variants influence gene regulation across spatial and temporal dimensions.
Conclusion: Embracing Regulatory Complexity
This study represents a significant advance in our ability to map genetic risk to molecular mechanisms in psychiatric disorders. By moving beyond the cis-regulatory lamppost and embracing the complexity of trans-regulation through coexpression networks, the authors have substantially expanded the catalog of schizophrenia-associated genes while providing mechanistic hypotheses about their coordinated function.
The identification of 641 novel associations is impressive, but perhaps more important is the demonstration that trans-regulatory effects are both detectable and biologically meaningful in the context of complex disease. As the field moves toward precision psychiatry, understanding how genetic variants orchestrate coordinated transcriptional programs across brain regions and cell types will be essential for developing targeted interventions.
The frameworks developed here—INGENE and MODULE—provide a roadmap for incorporating network-level regulation into genetic studies of brain disorders. As sample sizes grow and multi-omic datasets become more comprehensive, these approaches will likely reveal increasingly refined pictures of the regulatory architectures underlying psychiatric risk. The future of psychiatric genetics lies not in finding.
We’ve created an accompanying article that focuses on the big picture and real-world impact of this research, without the technical details.
Dr. Giulio Pergola thinks of his schizophrenia research in the context of true crime. He’s the detective. The motive is a person’s genetic risk for mental illness. The weapon is how the brain works, or doesn’t work, in some cases.