Architecture of the human regulatory network derived from ENCODE data

Mark Gerstein, Anshul Kundaje, Manoj Hariharan, Stephen G. Landt, Koon‐Kiu Yan, Chao Cheng, Xinmeng Jasmine Mu, Ekta Khurana, Joel Rozowsky, Roger P. Alexander, Renqiang Min, Pedro Alves, Alexej Abyzov, Nick Addleman, Nitin Bhardwaj, Alan P. Boyle, Philip Cayting, Alexandra Charos, David Ziyou Chen, Yong Cheng, Declan Clarke, Catharine Eastman, Ghia Euskirchen, Seth Frietze, Yao Fu, Jason Gertz, Fabian Grubert, Arif Harmanci, Preti Jain, Maya Kasowski, Phil Lacroute, Jing Leng, Lian Jin, Hannah Monahan, Henriette O’Geen, Zhengqing Ouyang, E. Christopher Partridge, Dorrelyn Patacsil, Florencia Pauli, Debasish Raha, Lucı́a Ramı́rez, Timothy E. Reddy, Brian D. Reed, Minyi Shi, Teri Slifer, Jing Wang, Linfeng Wu, Xinqiong Yang, Kevin Y. Yip, Gili Zilberman-Schapira, Serafim Batzoglou, Arend Sidow, Peggy Farnham, R Myers, Sherman M. Weissman, M SnyderNature 2012 · DOI

Abstract

Transcription factors bind in a combinatorial fashion to specify the on-and-off states of genes; the ensemble of these binding events forms a regulatory network, constituting the wiring diagram for a cell. To examine the principles of the human transcriptional regulatory network, we determined the genomic binding information of 119 transcription-related factors in over 450 distinct experiments. We found the combinatorial, co-association of transcription factors to be highly context specific: distinct combinations of factors bind at specific genomic locations. In particular, there are significant differences in the binding proximal and distal to genes. We organized all the transcription factor binding into a hierarchy and integrated it with other genomic information (for example, microRNA regulation), forming a dense meta-network. Factors at different levels have different properties; for instance, top-level transcription factors more strongly influence expression and middle-level ones co-regulate targets to mitigate information-flow bottlenecks. Moreover, these co-regulations give rise to many enriched network motifs (for example, noise-buffering feed-forward loops). Finally, more connected network components are under stronger selection and exhibit a greater degree of allele-specific activity. The regulatory information obtained in this study will be crucial for interpreting personal genome sequences and understanding basic principles of human biology and disease.

Introduction: The Logic of Human Gene Regulation

While transcription factors (TFs) have long been recognized as the primary "switches" of gene expression, the underlying logic of the human regulatory circuit has remained elusive. In simpler organisms like E. coli, these networks are well-mapped, but the human genome’s scale and complexity present a significant combinatorial challenge. TFs do not act in isolation; they form complex teams that vary based on genomic context. This study seeks to move beyond a simple catalog of binding sites to provide a systems-level understanding of how these interactions are structured, how information flows through the network, and how evolutionary pressure maintains its integrity.

Methodology: Mapping the Meta-Network

The researchers utilized ChIP-seq data to identify the genomic binding sites of 119 transcription-related factors across 450+ experiments, primarily focusing on five human cell lines (notably K562 and GM12878). To analyze this data, several novel frameworks were employed:

  • Hierarchy Height Metric: TFs were ranked using a normalized metric, $h = (O - I) / (O + I)$, where $O$ represents out-degree (factors regulated) and $I$ represents in-degree (factors regulating the TF). This allowed the network to be organized into three tiers: Executive, Middle-level, and Foreman.
  • Combinatorial Modeling: Machine learning was used to identify "biclusters"—specific groups of TFs that consistently bind together in particular genomic contexts.
  • Meta-network Integration: The TF network was integrated with data on microRNAs, protein-protein interactions, and phosphorylation to create a multi-layered view of cellular regulation.

The Regulatory Hierarchy: Executives, Managers, and Foremen

The human regulatory network is organized into a sophisticated three-tier hierarchy. "Executive" TFs at the top possess high out-degrees, regulating a vast array of downstream targets; these factors tend to have general functions and correlate strongly with overall gene expression levels. At the bottom, "Foremen" exhibit high in-degrees and are more specialized, often showing tissue-specific activity.

Interestingly, the middle tier acts as a functional bottleneck for information flow. To mitigate this, middle-level TFs show the highest levels of collaboration and co-regulation. Furthermore, the hierarchy is not strictly top-down; distal regulatory elements (enhancers) often create feedback loops where lower-level factors can influence their superiors.

Context-Specific Combinatorial Binding

A key finding of the study is that TF co-association is highly dependent on genomic context. TFs choose different partners depending on whether they are binding at a gene's promoter (proximal) or at a distant enhancer (distal).

Transcription factor co-association and context-specific binding patterns.

As shown in the analysis of co-association contexts, specific clusters of factors (biclusters) emerge that are unique to either proximal or distal regions. For example, certain factors like GATA1 and FOS exhibit distinct partner preferences that shift based on their distance from the transcription start site. This context-specificity suggests that the "wiring" of the genome is far more dynamic than previously thought, with enhancers providing a layer of regulation distinct from that of promoters.

Evolutionary Constraints and Network Motifs

The network's topology is deeply reflected in the human genome's evolution. Highly connected "hubs" and top-level executives are under significantly stronger negative selection than peripheral nodes, showing lower non-synonymous SNP density. This confirms that the "executives" of the regulatory network are evolutionarily indispensable.

Functional analysis also revealed a high enrichment of the Feed-Forward Loop (FFL) motif. In this three-node pattern, Factor A regulates Factor B, and both regulate Target C. This structure is believed to act as a noise-filtering mechanism, ensuring that gene expression only changes in response to persistent signals rather than transient fluctuations.

Finally, the study identified a link between network connectivity and allele-specific activity. TFs with higher "allelicity"—those sensitive to DNA sequence variations between parental alleles—tend to have more target genes, suggesting that the most connected components of the network are also the most sensitive to individual genetic variation.

Conclusion and Future Directions

This comprehensive map of the human regulatory network provides a vital framework for interpreting personal genomes and understanding disease. However, limitations remain. The current data is largely derived from five cell lines, and the network likely undergoes significant "rewiring" in different tissues or disease states. Future work will need to transition from this static map to a dynamic understanding of how these hierarchies respond to external stimuli and environmental stress in real-time.

Summary updated

Abstract

Transcription factors bind in a combinatorial fashion to specify the on-and-off states of genes; the ensemble of these binding events forms a regulatory network, constituting the wiring diagram for a cell. To examine the principles of the human transcriptional regulatory network, we determined the genomic binding information of 119 transcription-related factors in over 450 distinct experiments. We found the combinatorial, co-association of transcription factors to be highly context specific: distinct combinations of factors bind at specific genomic locations. In particular, there are significant differences in the binding proximal and distal to genes. We organized all the transcription factor binding into a hierarchy and integrated it with other genomic information (for example, microRNA regulation), forming a dense meta-network. Factors at different levels have different properties; for instance, top-level transcription factors more strongly influence expression and middle-level ones co-regulate targets to mitigate information-flow bottlenecks. Moreover, these co-regulations give rise to many enriched network motifs (for example, noise-buffering feed-forward loops). Finally, more connected network components are under stronger selection and exhibit a greater degree of allele-specific activity (that is, differential binding to the two parental alleles). The regulatory information obtained in this study will be crucial for interpreting personal genome sequences and understanding basic principles of human biology and disease.
BibTeX
@inproceedings{gerstein2012architecture,
  author = {Mark Gerstein and Anshul Kundaje and Manoj Hariharan and Stephen G. Landt and Koon‐Kiu Yan and Chao Cheng and Xinmeng Jasmine Mu and Ekta Khurana and Joel Rozowsky and Roger P. Alexander and Renqiang Min and Pedro Alves and Alexej Abyzov and Nick Addleman and Nitin Bhardwaj and Alan P. Boyle and Philip Cayting and Alexandra Charos and David Ziyou Chen and Yong Cheng and Declan Clarke and Catharine Eastman and Ghia Euskirchen and Seth Frietze and Yao Fu and Jason Gertz and Fabian Grubert and Arif Harmanci and Preti Jain and Maya Kasowski and Phil Lacroute and Jing Leng and Lian Jin and Hannah Monahan and Henriette O’Geen and Zhengqing Ouyang and E. Christopher Partridge and Dorrelyn Patacsil and Florencia Pauli and Debasish Raha and Lucı́a Ramı́rez and Timothy E. Reddy and Brian D. Reed and Minyi Shi and Teri Slifer and Jing Wang and Linfeng Wu and Xinqiong Yang and Kevin Y. Yip and Gili Zilberman-Schapira and Serafim Batzoglou and Arend Sidow and Peggy Farnham and R Myers and Sherman M. Weissman and M Snyder},
  title = {{Architecture of the human regulatory network derived from ENCODE data}},
  booktitle = {Nature},
  year = {2012},
  doi = {10.1038/nature11245},
}
Figure