Predicting the DNA binding specificity of mutated transcription factors using family-level biophysically interpretable machine learning.

Sequence-specific interactions of transcription factors (TFs) with genomic DNA underlie many cellular processes. High-throughput in vitro binding assays coupled with computational analysis have made it possible to accurately define such sequence recognition in a biophysically interpretable yet mechanism-agonistic way for individual TFs. The fact that such sequence-to-affinity models are now available for hundreds of TFs provides new avenues for predicting how the DNA binding specificity of a TF changes when its protein sequence is mutated. To this end, we developed an analytical framework based on a tetrahedron embedding that can be applied at the level of a given structural TF family. Using bHLH as a test case, we demonstrate that we can systematically map dependencies between the protein sequence of a TF and base preference within the DNA binding site. We also develop a regression approach to predict the quantitative energetic impact of mutations in the DNA binding domain of a TF on its DNA binding specificity, and perform SELEX-seq assays on mutated TFs to experimentally validate our results. Our results point to the feasibility of predicting the functional impact of disease mutations and allelic variation in the cell-wide TF repertoire by leveraging high-quality functional information across sets of homologous wild-type proteins.

Identifying genetic regulatory variants that affect transcription factor activity.

Genetic variants affecting gene expression levels in humans have been mapped in the Genotype-Tissue Expression (GTEx) project. Trans-acting variants impacting many genes simultaneously through a shared transcription factor (TF) are of particular interest. Here, we developed a generalized linear model (GLM) to estimate protein-level TF activity levels in an individual-specific manner from GTEx RNA sequencing (RNA-seq) profiles. It uses observed differential gene expression after TF perturbation as a predictor and, by analyzing differential expression within pairs of neighboring genes, controls for the confounding effect of variation in chromatin state along the genome. We inferred genotype-specific activities for 55 TFs across 49 tissues. Subsequently performing genome-wide association analysis on this virtual trait revealed TF activity quantitative trait loci (aQTLs) that, as a set, are enriched for functional features. Altogether, the set of tools we introduce here highlights the potential of genetic association studies for cellular endophenotypes based on a network-based multi-omics approach. The transparent peer review record is available.

Optimisation of TP53 reporters by systematic dissection of synthetic TP53 response elements.

TP53 is a transcription factor that controls multiple cellular processes, including cell cycle arrest, DNA repair and apoptosis. The relation between TP53 binding site architecture and transcriptional output is still not fully understood. Here, we systematically examined in three different cell lines the effects of binding site affinity and copy number on TP53-dependent transcriptional output, and also probed the impact of spacer length and sequence between adjacent binding sites, and of core promoter identity. Paradoxically, we found that high-affinity TP53 binding sites are less potent than medium-affinity sites. TP53 achieves supra-additive transcriptional activation through optimally spaced adjacent binding sites, suggesting a cooperative mechanism. Optimally spaced adjacent binding sites have a ∼10-bp periodicity, suggesting a role for spatial orientation along the DNA double helix. We leveraged these insights to construct a log-linear model that explains activity from sequence features, and to identify new highly active and sensitive TP53 reporters.

Defining the fine structure of promoter activity on a genome-wide scale with CISSECTOR.

Classic promoter mutagenesis strategies can be used to study how proximal promoter regions regulate the expression of particular genes of interest. This is a laborious process, in which the smallest sub-region of the promoter still capable of recapitulating expression in an ectopic setting is first identified, followed by targeted mutation of putative transcription factor binding sites. Massively parallel reporter assays such as survey of regulatory elements (SuRE) provide an alternative way to study millions of promoter fragments in parallel. Here we show how a generalized linear model (GLM) can be used to transform genome-scale SuRE data into a high-resolution genomic track that quantifies the contribution of local sequence to promoter activity. This coefficient track helps identify regulatory elements and can be used to predict promoter activity of any sub-region in the genome. It thus allows in silico dissection of any promoter in the human genome to be performed. We developed a web application, available at cissector.nki.nl, that lets researchers easily perform this analysis as a starting point for their research into any promoter of interest.

Emerging questions in transcriptional regulation.

What new questions can we ask about transcriptional regulation given recent developments in large-scale approaches?

Benchmarking DNA binding affinity models using allele-specific transcription factor binding data.

Transcription factors (TFs) bind to DNA in a highly sequence-specific manner. This specificity can manifest itself in vivo at heterozygous loci as a difference in TF occupancy between the two alleles. When applied on a genomic scale, functional genomic assays such as ChIP-seq typically lack the statistical power to detect allele-specific binding (ASB) at the level of individual variants. To address this, we propose a framework for benchmarking sequence-to-affinity models for TF binding in terms of their ability to predict allelic imbalances in ChIP-seq counts. We show that a likelihood function based on an over-dispersed binomial distribution can aggregate evidence for allelic preference across the genome without requiring statistical significance for individual variants. This allows us to systematically compare predictive performance when multiple binding models for the same TF are available. We introduce PyProBound, an easily extensible reimplementation of the ProBound biophysically interpretable machine learning framework. Configuring PyProBound to explicitly account for a confounding sequence-specific bias in DNA fragmentation rate yields improved TF binding models when training on ChIP-seq data. We also show how our likelihood function can be leveraged to perform de novo motif discovery on the raw allele-aware ChIP-seq counts.

Transcription factor paralogs orchestrate alternative gene regulatory networks by context-dependent cooperation with multiple cofactors.

In eukaryotes, members of transcription factor families often exhibit similar DNA binding properties in vitro, yet orchestrate paralog-specific gene regulatory networks in vivo. The serially homologous first (T1) and third (T3) thoracic legs of Drosophila, which are specified by the Hox proteins Scr and Ubx, respectively, offer a unique opportunity to address this paradox in vivo. Genome-wide analyses using epitope-tagged alleles of both Hox loci in the T1 and T3 leg imaginal discs, the precursors to the adult legs and ventral body regions, show that ~8% of Hox binding is paralog-specific. Binding specificity is mediated by interactions with distinct cofactors in different domains: the Hox cofactor Exd acts in the proximal domain and is necessary for Scr to bind many of its paralog-specific targets, while in the distal leg domain, the homeodomain protein Distal-less (Dll) enhances Scr binding to a different subset of loci. These findings reveal how Hox paralogs, and perhaps paralogs of other transcription factor families, orchestrate alternative downstream gene regulatory networks with the help of multiple, context-specific cofactors.

Prediction of protein-ligand binding affinity from sequencing data with interpretable machine learning.

Protein-ligand interactions are increasingly profiled at high throughput using affinity selection and massively parallel sequencing. However, these assays do not provide the biophysical parameters that most rigorously quantify molecular interactions. Here we describe a flexible machine learning method, called ProBound, that accurately defines sequence recognition in terms of equilibrium binding constants or kinetic rates. This is achieved using a multi-layered maximum-likelihood framework that models both the molecular interactions and the data generation process. We show that ProBound quantifies transcription factor (TF) behavior with models that predict binding affinity over a range exceeding that of previous resources; captures the impact of DNA modifications and conformational flexibility of multi-TF complexes; and infers specificity directly from in vivo data such as ChIP-seq without peak calling. When coupled with an assay called KD-seq, it determines the absolute affinity of protein-ligand interactions. We also apply ProBound to profile the kinetics of kinase-substrate interactions. ProBound opens new avenues for decoding biological networks and rationally engineering protein-ligand interactions.

Transcription factor regulation of eQTL activity across individuals and tissues.

Tens of thousands of genetic variants associated with gene expression (cis-eQTLs) have been discovered in the human population. These eQTLs are active in various tissues and contexts, but the molecular mechanisms of eQTL variability are poorly understood, hindering our understanding of genetic regulation across biological contexts. Since many eQTLs are believed to act by altering transcription factor (TF) binding affinity, we hypothesized that analyzing eQTL effect size as a function of TF level may allow discovery of mechanisms of eQTL variability. Using GTEx Consortium eQTL data from 49 tissues, we analyzed the interaction between eQTL effect size and TF level across tissues and across individuals within specific tissues and generated a list of 10,098 TF-eQTL interactions across 2,136 genes that are supported by at least two lines of evidence. These TF-eQTLs were enriched for various TF binding measures, supporting with orthogonal evidence that these eQTLs are regulated by the implicated TFs. We also found that our TF-eQTLs tend to overlap genes with gene-by-environment regulatory effects and to colocalize with GWAS loci, implying that our approach can help to elucidate mechanisms of context-specificity and trait associations. Finally, we highlight an interesting example of IKZF1 TF regulation of an APBB1IP gene eQTL that colocalizes with a GWAS signal for blood cell traits. Together, our findings provide candidate TF mechanisms for a large number of eQTLs and offer a generalizable approach for researchers to discover TF regulators of genetic variant effects in additional QTL datasets.

Systematic in vitro profiling of off-target affinity, cleavage and efficiency for CRISPR enzymes.

CRISPR RNA-guided endonucleases (RGEs) cut or direct activities to specific genomic loci, yet each has off-target activities that are often unpredictable. We developed a pair of simple in vitro assays to systematically measure the DNA-binding specificity (Spec-seq), catalytic activity specificity (SEAM-seq) and cleavage efficiency of RGEs. By separately quantifying binding and cleavage specificity, Spec/SEAM-seq provides detailed mechanistic insight into off-target activity. Feature-based models generated from Spec/SEAM-seq data for SpCas9 were consistent with previous reports of its in vitro and in vivo specificity, validating the approach. Spec/SEAM-seq is also useful for profiling less-well characterized RGEs. Application to an engineered SpCas9, HiFi-SpCas9, indicated that its enhanced target discrimination can be attributed to cleavage rather than binding specificity. The ortholog ScCas9, on the other hand, derives specificity from binding to an extended PAM. The decreased off-target activity of AsCas12a (Cpf1) appears to be primarily driven by DNA-binding specificity. Finally, we performed the first characterization of CasX specificity, revealing an all-or-nothing mechanism where mismatches can be bound, but not cleaved. Together, these applications establish Spec/SEAM-seq as an accessible method to rapidly and reliably evaluate the specificity of RGEs, Cas::gRNA pairs, and gain insight into the mechanism and thermodynamics of target discrimination.

Context-Dependent Gene Regulation by Homeodomain Transcription Factor Complexes Revealed by Shape-Readout Deficient Proteins.

Eukaryotic transcription factors (TFs) form complexes with various partner proteins to recognize their genomic target sites. Yet, how the DNA sequence determines which TF complex forms at any given site is poorly understood. Here, we demonstrate that high-throughput in vitro DNA binding assays coupled with unbiased computational analysis provide unprecedented insight into how different DNA sequences select distinct compositions and configurations of homeodomain TF complexes. Using inferred knowledge about minor groove width readout, we design targeted protein mutations that destabilize homeodomain binding both in vitro and in vivo in a complex-specific manner. By performing parallel systematic evolution of ligands by exponential enrichment sequencing (SELEX-seq), chromatin immunoprecipitation sequencing (ChIP-seq), RNA sequencing (RNA-seq), and Hi-C assays, we not only classify the majority of in vivo binding events in terms of complex composition but also infer complex-specific functions by perturbing the gene regulatory network controlled by a single complex.

Toward a Mechanistic Understanding of DNA Methylation Readout by Transcription Factors.

Epigenetic DNA modification impacts gene expression, but the underlying molecular mechanisms are only partly understood. Adding a methyl group to a cytosine base locally modifies the structural features of DNA in multiple ways, which may change the interaction with DNA-binding transcription factors (TFs) and trigger a cascade of downstream molecular events. Cells can be probed using various functional genomics assays, but it is difficult to disentangle the confounded effects of DNA modification on TF binding, chromatin accessibility, intranuclear variation in local TF concentration, and rate of transcription. Here we discuss how high-throughput in vitro profiling of protein-DNA interactions has enabled comprehensive characterization and quantification of the methylation sensitivity of TFs. Despite the limited structural data for DNA containing methylated cytosine, automated analysis of structural information in the Protein Data Bank (PDB) shows how 5-methylcytosine (5mC) can be recognized in various ways by amino acid side chains. We discuss how a context-dependent effect of methylation on DNA groove geometry can affect DNA binding by homeodomain proteins and how principled modeling of ChIP-seq data can overcome the confounding that makes the interpretation of in vivo data challenging. The emerging picture is that epigenetic modifications affect TF binding in a highly context-specific manner, with a direction and effect size that depend critically on their position within the TF binding site and the amino acid sequence of the TF. With this improved mechanistic knowledge, we have come closer to understanding how cells use DNA modification to acquire, retain, and change their identity.

In vivo Hox binding specificity revealed by systematic changes to a single cis regulatory module.

Hox proteins belong to a family of transcription factors with similar DNA binding specificities that control animal differentiation along the antero-posterior body axis. Hox proteins are expressed in partially overlapping regions where each one is responsible for the formation of particular organs and structures through the regulation of specific direct downstream targets. Thus, explaining how each Hox protein can selectively control its direct targets from those of another Hox protein is fundamental to understand animal development. Here we analyse a cis regulatory module directly regulated by seven different Drosophila Hox proteins and uncover how different Hox class proteins differentially control its expression. We find that regulation by one or another Hox protein depends on the combination of three modes: Hox-cofactor dependent DNA-binding specificity; Hox-monomer binding sites; and interaction with positive and negative Hox-collaborator proteins. Additionally, we find that similar regulation can be achieved by Amphioxus orthologs, suggesting these three mechanisms are conserved from insects to chordates.

Low-Affinity Binding Sites and the Transcription Factor Specificity Paradox in Eukaryotes.

Eukaryotic transcription factors (TFs) from the same structural family tend to bind similar DNA sequences, despite the ability of these TFs to execute distinct functions in vivo. The cell partly resolves this specificity paradox through combinatorial strategies and the use of low-affinity binding sites, which are better able to distinguish between similar TFs. However, because these sites have low affinity, it is challenging to understand how TFs recognize them in vivo. Here, we summarize recent findings and technological advancements that allow for the quantification and mechanistic interpretation of TF recognition across a wide range of affinities. We propose a model that integrates insights from the fields of genetics and cell biology to provide further conceptual understanding of TF binding specificity. We argue that in eukaryotes, target specificity is driven by an inhomogeneous 3D nuclear distribution of TFs and by variation in DNA binding affinity such that locally elevated TF concentration allows low-affinity binding sites to be functional.

High-throughput identification of human SNPs affecting regulatory element activity.

Most of the millions of SNPs in the human genome are non-coding, and many overlap with putative regulatory elements. Genome-wide association studies (GWAS) have linked many of these SNPs to human traits or to gene expression levels, but rarely with sufficient resolution to identify the causal SNPs. Functional screens based on reporter assays have previously been of insufficient throughput to test the vast space of SNPs for possible effects on regulatory element activity. Here we leveraged the throughput and resolution of the survey of regulatory elements (SuRE) reporter technology to survey the effect of 5.9 million SNPs, including 57% of the known common SNPs, on enhancer and promoter activity. We identified more than 30,000 SNPs that alter the activity of putative regulatory elements, partially in a cell-type-specific manner. Integration of this dataset with GWAS results may help to pinpoint SNPs that underlie human traits.

Accurate and sensitive quantification of protein-DNA binding affinity.

Transcription factors (TFs) control gene expression by binding to genomic DNA in a sequence-specific manner. Mutations in TF binding sites are increasingly found to be associated with human disease, yet we currently lack robust methods to predict these sites. Here, we developed a versatile maximum likelihood framework named No Read Left Behind (NRLB) that infers a biophysical model of protein-DNA recognition across the full affinity range from a library of in vitro selected DNA binding sites. NRLB predicts human Max homodimer binding in near-perfect agreement with existing low-throughput measurements. It can capture the specificity of the p53 tetramer and distinguish multiple binding modes within a single sample. Additionally, we confirm that newly identified low-affinity enhancer binding sites are functional in vivo, and that their contribution to gene expression matches their predicted affinity. Our results establish a powerful paradigm for identifying protein binding sites and interpreting gene regulatory sequences in eukaryotic genomes.

A unified approach for quantifying and interpreting DNA shape readout by transcription factors.

Transcription factors (TFs) interpret DNA sequence by probing the chemical and structural properties of the nucleotide polymer. DNA shape is thought to enable a parsimonious representation of dependencies between nucleotide positions. Here, we propose a unified mathematical representation of the DNA sequence dependence of shape and TF binding, respectively, which simplifies and enhances analysis of shape readout. First, we demonstrate that linear models based on mononucleotide features alone account for 60-70% of the variance in minor groove width, roll, helix twist, and propeller twist. This explains why simple scoring matrices that ignore all dependencies between nucleotide positions can partially account for DNA shape readout by a TF Adding dinucleotide features as sequence-to-shape predictors to our model, we can almost perfectly explain the shape parameters. Building on this observation, we developed a post hoc analysis method that can be used to analyze any mechanism-agnostic protein-DNA binding model in terms of shape readout. Our insights provide an alternative strategy for using DNA shape information to enhance our understanding of how cis-regulatory codes are interpreted by the cellular machinery.

Systematic prediction of DNA shape changes due to CpG methylation explains epigenetic effects on protein-DNA binding.

BACKGROUND: DNA shape analysis has demonstrated the potential to reveal structure-based mechanisms of protein-DNA binding. However, information about the influence of chemical modification of DNA is limited. Cytosine methylation, the most frequent modification, represents the addition of a methyl group at the major groove edge of the cytosine base. In mammalian genomes, cytosine methylation most frequently occurs at CpG dinucleotides. In addition to changing the chemical signature of C/G base pairs, cytosine methylation can affect DNA structure. Since the original discovery of DNA methylation, major efforts have been made to understand its effect from a sequence perspective. Compared to unmethylated DNA, however, little structural information is available for methylated DNA, due to the limited number of experimentally determined structures. To achieve a better mechanistic understanding of the effect of CpG methylation on local DNA structure, we developed a high-throughput method, methyl-DNAshape, for predicting the effect of cytosine methylation on DNA shape. RESULTS: Using our new method, we found that CpG methylation significantly altered local DNA shape. Four DNA shape features-helix twist, minor groove width, propeller twist, and roll-were considered in this analysis. Distinct distributions of effect size were observed for different features. Roll and propeller twist were the DNA shape features most strongly affected by CpG methylation with an effect size depending on the local sequence context. Methylation-induced changes in DNA shape were predictive of the measured rate of cleavage by DNase I and suggest a possible mechanism for some of the methylation sensitivities that were recently observed for human Pbx-Hox complexes. CONCLUSIONS: CpG methylation is an important epigenetic mark in the mammalian genome. Understanding its role in protein-DNA recognition can further our knowledge of gene regulation. Our high-throughput methyl-DNAshape method can be used to predict the effect of cytosine methylation on DNA shape and its subsequent influence on protein-DNA interactions. This approach overcomes the limited availability of experimental DNA structures that contain 5-methylcytosine.

SelexGLM differentiates androgen and glucocorticoid receptor DNA-binding preference over an extended binding site.

The DNA-binding interfaces of the androgen (AR) and glucocorticoid (GR) receptors are virtually identical, yet these transcription factors share only about a third of their genomic binding sites and regulate similarly distinct sets of target genes. To address this paradox, we determined the intrinsic specificities of the AR and GR DNA-binding domains using a refined version of SELEX-seq. We developed an algorithm, SelexGLM, that quantifies binding specificity over a large (31-bp) binding site by iteratively fitting a feature-based generalized linear model to SELEX probe counts. This analysis revealed that the DNA-binding preferences of AR and GR homodimers differ significantly, both within and outside the 15-bp core binding site. The relative preference between the two factors can be tuned over a wide range by changing the DNA sequence, with AR more sensitive to sequence changes than GR. The specificity of AR extends to the regions flanking the core 15-bp site, where isothermal calorimetry measurements reveal that affinity is augmented by enthalpy-driven readout of poly(A) sequences associated with narrowed minor groove width. We conclude that the increased specificity of AR is correlated with more enthalpy-driven binding than GR. The binding models help explain differences in AR and GR genomic binding and provide a biophysical rationale for how promiscuous binding by GR allows functional substitution for AR in some castration-resistant prostate cancers.

Network-based approaches that exploit inferred transcription factor activity to analyze the impact of genetic variation on gene expression.

Over the past decade, a number of methods have emerged for inferring protein-level transcription factor activities in individual samples based on prior information about the structure of the gene regulatory network. We discuss how this has enabled new methods for dissecting trans-acting mechanisms that underpin genetic variation in gene expression.