TL;DR
This paper introduces a new model to systematically determine the genomic distance over which transcription factors (TFs) influence gene expression. It reveals two distinct classes of TFs – short-range and long-range – with different regulatory properties, and shows that these ranges are further modulated by the genomic context, specifically topologically associating domains (TADs) and their chromatin states. This work provides a more nuanced framework for understanding gene regulation and interpreting genetic variants associated with diseases.
What Problem Does This Paper Address?
Identifying which genes a transcription factor (TF) regulates is crucial for understanding gene expression and cellular function. However, TF binding sites are often located far from the genes they regulate, making it challenging to establish direct links. Traditional methods often rely on arbitrary distance thresholds or proximity to gene promoters, which can be inaccurate. While advancements like Hi-C have revealed genome organization into topologically associating domains (TADs) that can facilitate enhancer-promoter interactions, it's still unclear how TFs operate within these structures and how their regulatory reach varies. This paper aims to systematically define the "regulatory range" of TFs and understand the factors that influence it, moving beyond simple proximity-based assignments.
Methods
The core of this study is the Regulatory Potential (RP) model. This model quantifies the regulatory effect of a TF on a gene by summing the contributions of all its binding sites, weighted by their distance to the gene's transcription start site (TSS). The influence of each binding site decays exponentially with distance, with a "decay distance" parameter (Δ) representing the distance at which the effect is halved.
The authors developed two main approaches to infer this optimal decay distance (Δ*):
- TF Perturbation Data: By analyzing gene expression changes after knocking down or overexpressing a TF, they identified differentially expressed (DE) genes. They then found the Δ* that best separated these DE genes from non-DE genes based on the RP model's predictions.
- Gene Expression Cohorts: When perturbation data was scarce, they used correlations between TF gene expression and other genes' expression across various cell lines and tissues. They inferred Δ* as the value that maximized the concordance between TF binding site predictions (RP model) and gene expression correlations.
They also employed:
- H3K27ac ChIP-seq: To assess active promoter and enhancer regions.
- Hi-C data: To define TADs and analyze chromatin conformation.
- eQTL data: To study the relationship between genetic variants, gene expression, and genomic distance.
- GWAS data: To investigate the role of TFs in disease association.
Datasets
The study utilized several large-scale datasets:
- TF ChIP-seq data: From the Cistrome DB, comprising thousands of TF binding profiles across various cell types.
- Gene expression data:
- Cancer Cell Line Encyclopedia (CCLE): Thousands of gene expression profiles from cancer cell lines.
- Genotype-Tissue Expression (GTEx) project: Gene expression data from multiple human tissues.
- Other gene expression cohorts (20-24).
- Histone modification ChIP-seq data: Specifically H3K27ac, from 1545 human samples.
- Hi-C data: To define topologically associating domains (TADs).
- eQTL data: From GTEx, linking genetic variants to gene expression.
- GWAS Catalog: For disease-associated single-nucleotide polymorphisms (SNPs).
Key Insights and Ideas
- Two Classes of TFs: The most striking finding is the clear division of TFs into "short-range" (regulating genes within ~1 kb of binding sites) and "long-range" (extending regulatory influence up to ~100 kb or more). This is a fundamental new categorization of TF behavior.
- Context Matters: A TF's regulatory range isn't fixed. It's significantly influenced by the genomic context, particularly the type of TAD (A-type: active, B-type: repressed) and its chromatin state.
- Long-Range TFs in Repressed Regions: Long-range TFs show even greater regulatory reach in B-type (repressed) TADs compared to A-type (active) TADs, suggesting they play a key role in bridging distant regulatory elements in less active genomic environments.
- Pioneer Factor Properties: Long-range TFs are more likely to exhibit "pioneer-like" properties, meaning they can bind to closed chromatin and facilitate the binding of other TFs, which is crucial for establishing gene expression programs.
- Auto-regulation in Long-Range TFs: Long-range TFs often show auto-regulatory behavior, binding to the regulatory regions of their own genes, which may help maintain stable lineage-specific expression.
- Chromatin State Modulates Range: The activity level of a TAD (e.g., high vs. low H3K27ac) directly impacts a TF's regulatory decay distance, with more active TADs leading to shorter regulatory ranges. This suggests a dynamic interplay between chromatin state and TF function.
- Improved GWAS Interpretation: The TAD-wise analysis, considering relative TF abundance within TADs, offers a more powerful way to interpret non-coding GWAS variants compared to simple SNP-TF peak overlaps.
Key Results
- TF Classification: The study identified two distinct modes for TF regulatory decay distances (Δ*), with modal values around 1 kb for short-range TFs and >10 kb for long-range TFs.
- TAD Preference: Short-range TFs tend to bind more in A-type TADs, while long-range TFs show a preference for B-type TADs.
- Regulatory Distance Differences: Long-range TFs have significantly longer regulatory decay distances in B-type TADs (ΔB > ΔA).
- Chromatin State Effect: Regulatory decay distances become shorter in TADs that are more active (higher H3K27ac). For example, RELA's regulatory distance was shorter in lung-active TADs when analyzed with lung expression data.
- GWAS SNP Association: The TAD-wise TF abundance approach significantly improved the ranking of known disease-associated TFs compared to simple SNP-TF peak overlaps.
- Disease Clusters: The analysis revealed co-occurring diseases linked to specific TF clusters, such as autoimmune diseases and chronic lymphocytic leukemia, or type-2 diabetes and breast cancer.
Limitations and Future Work
- Data Coverage: While extensive, the study acknowledges that the Cistrome DB and CCLE/GTEx cohorts do not cover all TFs in all cellular contexts.
- Model Simplifications: The RP model, while powerful, does not encompass the full complexity of gene regulation, such as the independence of enhancer contributions or the precise mechanisms of interaction.
- Hi-C Limitations: Current Hi-C technology and data processing might not fully capture all regulatory interactions, leading to a discrepancy between observed interactions and inferred regulatory distances.
- Mechanistic Understanding: The precise mechanisms by which chromatin states influence regulatory decay distances and how non-looping mechanisms contribute to long-range regulation require further investigation.
- Future Directions: The authors suggest further work to understand the mechanistic basis of TF-specific and context-dependent gene regulatory effects, and to develop more sophisticated hierarchical probabilistic frameworks for GWAS interpretation.
Summary updated