In vitro binding energies capture Klf4 occupancy across the human genome

TL;DR

This study quantitatively characterizes how the human transcription factor Klf4 binds to DNA sequences, revealing that a physics-based model accurately predicts its genome-wide binding patterns based on in vitro binding energy measurements.

Contribution

It introduces a combined linear and Ising model to predict TF binding energies from in vitro data, accurately capturing genome-wide occupancy without extra fitting.

Findings

In vitro binding energies predict Klf4 occupancy genome-wide

A physics-based model captures non-linear sequence dependence

Model accurately predicts binding patterns on long DNA molecules

Abstract

Transcription factors (TFs) regulate gene expression by binding to specific genomic loci determined by DNA sequence. Their sequence specificity is commonly summarized by a consensus binding motif. However, eukaryotic genomes contain billions of low-affinity DNA sequences to which TFs associate with a sequence-dependent binding energy. We currently lack insight into how the genomic sequence defines this spectrum of binding energies and the resulting pattern of TF localization. Here, we set out to obtain a quantitative understanding of sequence-dependent TF binding to both motif and non-motif sequences. We achieve this by first pursuing accurate measurements of physical binding energies of the human TF Klf4 to a library of short DNA sequences in a fluorescence-anisotropy-based bulk competitive binding assay. Second, we show that the highly non-linear sequence dependence of Klf4 binding energies can be captured by combining a linear model of binding energies with an Ising model of the coupled recognition of nucleotides by a TF. We find that this statistical mechanics model parametrized by our in vitro measurements captures Klf4 binding patterns on individual long DNA molecules stretched in the optical tweezer, and is predictive for Klf4 occupancy across the entire human genome without additional fit parameters.