Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
Filters applied. Clear all
. 2011 Aug 9;108(32):13353-8.
doi: 10.1073/pnas.1103105108. Epub 2011 Jul 26.

Cooperative Transcription Factor Associations Discovered Using Regulatory Variation

Affiliations
Free PMC article

Cooperative Transcription Factor Associations Discovered Using Regulatory Variation

Konrad J Karczewski et al. Proc Natl Acad Sci U S A. .
Free PMC article

Abstract

Regulation of gene expression at the transcriptional level is achieved by complex interactions of transcription factors operating at their target genes. Dissecting the specific combination of factors that bind each target is a significant challenge. Here, we describe in detail the Allele Binding Cooperativity test, which uses variation in transcription factor binding among individuals to discover combinations of factors and their targets. We developed the ALPHABIT (a large-scale process to hunt for allele binding interacting transcription factors) pipeline, which includes statistical analysis of binding sites followed by experimental validation, and demonstrate that this method predicts transcription factors that associate with NFκB. Our method successfully identifies factors that have been known to work with NFκB (E2A, STAT1, IRF2), but whose global coassociation and sites of cooperative action were not known. In addition, we identify a unique coassociation (EBF1) that had not been reported previously. We present a general approach for discovering combinatorial models of regulation and advance our understanding of the genetic basis of variation in transcription factor binding.

Conflict of interest statement

The authors declare no conflict of interest.

Figures

Fig. 1.
Fig. 1.
Overview of the ALPHABIT pipeline. (A) In a model of cooperativity, the binding of one factor depends on the binding of another. For example, when a STAT1 motif is present, both STAT1 and NFκB are bound. Loss of the STAT1 motif decreases binding of NFκB, despite the presence of an NFκB motif. (B) Associated factor discovery process. Variable motifs are searched for in variable NFκB binding peaks and the difference in “motif score” (i.e., match to consensus) is correlated with difference in NFκB binding. Significant predictions are validated by ChIP-Seq and subject to subsequent analysis.
Fig. 2.
Fig. 2.
Output of ALPHABIT analysis. The ALPHABIT pipeline identifies variable motifs that are predictive of NFκB binding. (A) Covariance of SNPs in an EBF motif with binding of NFκB at a single locus. Colors correspond to populations (red: YRI; blue: CHB/JPT; purple: CEU), as in ref. . Binding of EBF in this region is also validated in GM12878 by ChIP-Seq (orange). The EBF motif is shown below with variable residues highlighted (first genotype in dashed lines, second in solid lines). (B) Analysis across all binding sites shows correlation between differences in EBF motifs (quantified as “motif scores”) and differences in NFκB binding signal.
Fig. 5.
Fig. 5.
Coexpression of predicted coassociated factors. Expression of three of the predicted coassociated TFs (IRF2, STAT1, and EBF) exhibit correlation (0.398, 0.379, 0.341, respectively) with NFκB expression in a large, independent dataset (9,395 gene-expression experiments from the GEO). ZNF143, CTCF, and E2A exhibit a much lower correlation (0.102, 0.128, and 0.137, respectively).
Fig. 3.
Fig. 3.
Distance dependence of cooperative interactions: The correlation between motif PWM match and NFκB binding strength drops when analyzing larger windows around the NFκB binding peak for motifs in validated binding sites. The number of sites was too low for distance dependence analysis of E2A motifs. As expected, proximal motifs are more predictive of NFκB binding than distal motifs.
Fig. 4.
Fig. 4.
Correlation of coassociated factors with NFκB binding. NFκB binding ratios (signal to background) can be predicted by fitting linear models of combinations of the control factors (A) and the “full” model (the control factors combined with four coassociated factors, EBF, STAT1, E2A, IRF2) (B). Although all signals are expected to be correlated because of open chromatin regions, a model fit using the coassociated factors (STAT1, EBF, E2A, and IRF2; r = 0.703) (B) is significantly more predictive of NFκB binding than one fit using the negative controls (ZNF143 and CTCF; r = 0.444) (A).

Similar articles

See all similar articles

Cited by 22 articles

See all "Cited by" articles

Publication types

LinkOut - more resources

Feedback