Transcription factor expression is the main determinant of variability in gene co-activity

Abstract

Many genes are co-expressed and form genomic domains of coordinated gene activity. However, the regulatory determinants of domain co-activity remain unclear. Here, we leverage human individual variation in gene expression to characterize the co-regulatory processes underlying domain co-activity and systematically quantify their effect sizes. We employ transcriptional decomposition to extract from RNA expression data an expression component related to co-activity revealed by genomic positioning. This strategy reveals close to 1,500 co-activity domains, covering most expressed genes, of which the large majority are invariable across individuals. Focusing specifically on domains with high variability in co-activity reveals that contained genes have a higher sharing of eQTLs, a higher variability in enhancer interactions, and an enrichment of binding by variably expressed transcription factors compared to genes within non-variable domains. Through careful quantification of the relative contributions of regulatory processes underlying co-activity, we find transcription factor expression levels to be the main determinant of gene co-activity. Our results indicate that distal trans effects contribute more than local genetic variation to individual variation in co-activity domains.

Non-coding variants impactcis-regulatory coordination in a cell type-specific manner

“…This observation is also consistent with the notion that cmQTLs frequently disrupt the binding sites of cell-type specific TFs, which then could impact histone modification deposition and cooperation between CREs within a CM in a cell type-specific manner. This finding complements recent observations showing that genetic variants impact CRE interactions relevant for establishing gene expression through direct perturbation of TF binding to enhancers [85] and that TFs are the main drivers of gene expression cooperativity [86].…”

Model-X knockoffs reveal data-dependent limits on regulatory network identification

“…In light of these findings, existing results based on conditional independence alone should not be taken to reflect direct causal effects or direct binding, unless the underlying causal sufficiency assumption has been tested on the relevant datasets. Failure of causal sufficiency casts doubt on recent TRN work making explicit causal interpretations of conditional independence structure (Buschur, Chikina, & Benos, 2020;van Duin, Krautz, Rennie, & Andersson, 2022;Mohan, London, Fazel, Witten, & Lee, 2014;Qiu et al, 2020;Wang, Solus, Yang, & Uhler, 2017).…”

Transcription factor expression is the main determinant of variability in gene co‐activity

“…Co‐activity scores for included individuals can be accessed at Zenodo ( https://doi.org/10.5281/zenodo.7180322 ; Data ref: van Duin et al , 2022 ).…”

“…Co‐activity scores and ABC predicted interactions are available at Zenodo ( https://doi.org/10.5281/zenodo.7180322 ; Data ref: van Duin et al , 2022 ). Code for core analyses of this manuscript is available at https://github.com/anderssonlab/van_Duin_et_al_2023 .…”