IMPACT: Genomic Annotation of Cell-State-Specific Regulatory Elements Inferred from the Epigenome of Bound Transcription Factors

Abstract

Despite significant progress in annotating the genome with experimental methods, much of the regulatory noncoding genome remains poorly defined. Here we assert that regulatory elements may be characterized by leveraging local epigenomic signatures where specific transcription factors (TFs) are bound. To link these two features, we introduce IMPACT, a genome annotation strategy that identifies regulatory elements defined by cell-state-specific TF binding profiles, learned from 515 chromatin and sequence annotations. We validate IMPACT using multiple compelling applications. First, IMPACT distinguishes between bound and unbound TF motif sites with high accuracy (average AUPRC 0.81, SE 0.07; across 8 tested TFs) and outperforms state-of-the-art TF binding prediction methods, MocapG, MocapS, and Virtual ChIP-seq. Second, in eight tested cell types, RNA polymerase II IMPACT annotations capture more cis-eQTL variation than sequence-based annotations, such as promoters and TSS windows (25% average increase in enrichment). Third, integration with rheumatoid arthritis (RA) summary statistics from European (N = 38,242) and East Asian (N = 22,515) populations revealed that the top 5% of CD4⁺ Treg IMPACT regulatory elements capture 85.7% of RA h2, the most comprehensive explanation for RA h2 to date. In comparison, the average RA h2 captured by compared CD4⁺ T histone marks is 42.3% and by CD4⁺ T specifically expressed gene sets is 36.4%. Lastly, we find that IMPACT may be used in many different cell types to identify complex trait associated regulatory elements.

Keywords: CD4(+); T cells; arthritis; autoimmune; binding; epigenomics; heritability; polygenic; rheumatoid; transcription.

Figure 1

IMPACT: A Genome Annotation Strategy to Identify Cell-State-Specific Regulatory Elements IMPACT learns a chromatin profile of cell-state-specific regulation, distinguishing master TF (red) regulatory elements (TF-bound motif sites, blue) from inactive regulatory elements (unbound motif sites, purple). Here, cell-state-specific open chromatin and cell-state-specific H3K4me1 are strong predictors of cell-state-specific regulatory elements. Cell-state-nonspecific open chromatin and nonspecific H3K4me1 are less informative, marking all types of regulatory elements, while H3K9me3 strongly implicates inactive regulatory elements. IMPACT should re-identify regulatory elements marked by master TF binding (peak 1) and those with similar chromatin profiles, presumably sites of related cell-state-specific processes (peak 2). IMPACT should not predict regulation at cell-state-nonspecific elements (peak 3), such as promoters of housekeeping genes.

Figure 2

IMPACT Outperforms State-of-the-Art TF Binding Prediction Methods (A) IMPACT outperforms MocapG, MocapS, and Virtual ChIP-seq in predicting cell-state-specific TF binding across 8 TFs, illustrated by AUPRCs on the same training and testing data across 50 trials, with the exception of the MocapS model for FOXP3. (B) Prediction of Pol II binding in 6 cell types reveals that IMPACT outperforms Virtual ChIP-seq.

Figure 3

IMPACT Genome-wide Regulatory Tracks (A) Cell-state-specific regulatory element IMPACT predictions for canonical target genes of T-BET, GATA3, STAT3, and FOXP3. (B) Highly weighted features of Th1, Th2, Th17, and Treg IMPACT annotations.

Figure 4

Pol II IMPACT Captures cis-eQTL Causal Variation Better than Sequence-Based Annotations across Eight Cell and Tissue Types Enrichment of cis-eQTL chi-square association values with Pol II IMPACT annotations, created for peripheral blood (A), fibroblasts (B), stomach (C), liver (D), left ventricle heart (E), sigmoid colon (F), pancreas (G), and CD4⁺ T cells (H), highlighting top performing IMPACT annotation compared to enrichments of sequence-based functional annotations. Values in parentheses after annotation name are the average annotation value across all common variants, e.g., the effective size of the annotation. ^∗ denotes permutation p < 0.05, ^∗∗ permutation p < 0.01, ^∗∗∗ permutation p < 0.001. Intervals at the top of each bar represent the 95% confidence interval of the enrichment estimate.

Figure 5

CD4⁺ T Cell-State IMPACT Annotations Are Strongly Enriched for RA Heritability (A) Enrichment of RA h2 in CD4⁺ T IMPACT for EUR and EAS populations. Values below cell states are the average annotation value across all common (MAF ≥ 0.05) SNPs, e.g., the effective size of the annotation. (B) Standardized annotation effect size ($\tau$^∗) of each annotation separately conditioned on annotations from the baseline-LD model. For (A) and (B), ^∗∗∗p < 0.001. (C) Proportion of total causal RA h2 explained by the top 5% of SNPs in each IMPACT annotation. For all panels, intervals at the top of each bar represent the 95% confidence interval.

Figure 6

CD4⁺ Treg IMPACT Annotation Significantly Captures RA Heritability Conditional on Strongly Enriched CD4⁺ T Cell Regulatory Annotations (A) RA h2 enrichment of the CD4⁺ Treg IMPACT annotation and compared T cell functional annotations. Values below cell states represent the effective size of the annotation. From left to right, we compare Treg IMPACT to genome-wide FOXP3 motif sites, FOXP3 ChIP-seq, the “Averaged Tracks” annotation, which assigns each SNP a value proportional to the number of overlapping IMPACT epigenomic features, the top five cell-type-specific histone modification annotations, in terms of independent $\tau$^∗, the top five cell-type-specifically expressed gene sets (Web Resources), in terms of independent $\tau$^∗, and T cell super enhancers. (B) CD4⁺ Treg IMPACT annotation standardized effect size ($\tau$^∗, teal) conditional on other T cell-related functional annotations (coral).$\tau$^∗ for independent analyses are denoted by the top of each black bar, as a reference for the conditional analyses, denoted by the top of each colored bar. ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗p < 0.001. For both panels, intervals at the top of each bar represent the 95% confidence interval.

Figure 7

IMPACT Cell-State-Specific Regulatory Element Annotation Effect Sizes across 42 Polygenic Traits (A) Signed log₁₀ p values of $\tau$^∗ for 42 traits across 13 cell-state-specific IMPACT annotations, capturing h2 in distinct sets of complex traits, shown by significantly positive $\tau$^∗. Each IMPACT annotation is described by its target cell state and key TF used for training in parentheses. (B) Signed log₁₀ p values of $\tau$^∗ for 42 traits across annotations representing the TF ChIP-seq used to train the corresponding IMPACT annotations. ChIP-seq annotations are described by the cell state in which the particular TF (in parentheses) was assayed. Color shown only if p value of $\tau$^∗ < 0.025 after multiple hypothesis correction.

Figure 8

IMPACT A Priori Identifies Variants with Measured Functionality (A) Enrichment of posterior probabilities of putatively causal RA SNPs in the top 1% of SNPs with CD4⁺ Treg regulatory element probabilities highlights the BACH2, ANKRD55, CTLA4/CD28, IRF5, and TNFAIP3 loci. (B and C) IMPACT regulatory element probabilities (black) at putatively causal SNPs with experimentally validated differential enhancer activity (bolded) and other 90% credible set SNPs (unbolded) at two RA-associated loci, CTLA4/CD28 and TNFAIP3.