Recursive motif analyses identify brain epigenetic transcription regulatory modules

Sharmi BanerjeeOpenReview 2024 · OpenReview

TL;DR

This paper introduces a novel computational pipeline to identify "epigenetic transcription regulatory modules" (ETRMs) by analyzing differentially methylated regions in the mouse brain. By combining methylation data clustering with recursive motif analysis, the approach uncovers how transcription factors interact with DNA methylation to influence gene expression, offering a new way to understand complex epigenetic regulation.

What Problem Does This Paper Address?

DNA methylation is a crucial epigenetic mechanism that controls gene expression by altering DNA structure and influencing the binding of proteins, particularly transcription factors (TFs). While large-scale methylome data is increasingly available, tools to interpret how these methylation changes relate to TF activity and gene regulation are limited. Specifically, understanding the interplay between dynamic DNA methylation and TF binding in complex biological processes like brain development has been challenging. This paper aims to bridge this gap by developing a computational method to systematically identify regulatory modules that link differentially methylated regions to specific TFs.

Methods

The core of this work is a two-step computational pipeline designed to identify ETRMs:

  1. Clustering Differentially Methylated Regions (DMRs): The pipeline first uses the Weighted Correlation Network Analysis (WGCNA) algorithm to group DMRs that exhibit similar methylation patterns across various samples. This step is crucial for reducing data complexity and focusing on regions likely co-regulated by common mechanisms.
  2. Recursive Motif Identification: For each cluster of co-methylated DMRs, a novel recursive motif search algorithm is employed. This algorithm iteratively identifies the most significantly enriched TF binding motif ("key TF") within a cluster. Then, it recursively searches for motifs in regions surrounding the "key TF" motif, effectively identifying TFs that likely form regulatory modules with the key TF. This process continues until no significant motifs are found, allowing for the discovery of intricate TF networks.

A key methodological enhancement is the creation of a comprehensive motif database. This database augments the standard HOMER database with additional datasets containing motifs specifically related to methylated DNA (from MeDReaders) and motifs identified through methylation-sensitive SELEX experiments (categorized as 'MethylPlus' and 'MethylMinus' based on methylation's effect on binding).

Datasets

The study utilized the following datasets:

  • Methylome Datasets:
    • Frontal cortices from TET1 knockout (TET1KO) and TET2 knockout (TET2KO) mouse brains. TET1KO data was generated using reduced representation bisulfite sequencing, while TET2KO data used whole-genome bisulfite sequencing.
  • 66 methylomes for mouse forebrain, midbrain, and hindbrain regions during development, as well as sorted mouse brain cells (astrocytes, oligodendrocytes, excitatory neurons, PV neurons, and VIP neurons).
  • ChIP-seq Datasets: Brain ChIP-seq data for various transcription factors were used to validate the predicted ETRMs.

Key Insights and Ideas

  • ETRMs as Functional Units: The concept of ETRMs, defined as sets of TFs with binding sites adjacent to a "key TF" within co-methylated regions, provides a framework for understanding how TFs cooperate in response to DNA methylation changes.
  • DNA Binding Domain Significance: The study suggests that the DNA binding domain of a TF plays a significant role in determining whether it interacts with methylated DNA sites. TFs with Homeobox and C2H2 Zinc Finger domains were frequently associated with methylated DNA, while HMG-box and ETS domains appeared in various categories.
  • TET1 vs. TET2 Specificity: The distinct sets of differentially methylated sites identified in TET1KO and TET2KO brains highlight the specific roles of these demethylation enzymes in brain development. TET1 loss appeared to affect promoter methylation more significantly than TET2 loss.
  • Co-methylated Regions Host Similar TF Families: A notable observation was that genomic loci within the same co-methylated cluster often contained motifs for TFs belonging to the same family (e.g., Zinc Finger, MADS, HMG), suggesting coordinated regulation within these families.
  • Recursive Motif Search Uncovers Hidden Networks: The recursive approach was crucial for identifying TF motifs that might be missed in a single, non-iterative analysis, revealing subtle interactions and regulatory modules.

Key Results

  • Identification of Distinct ETRMs: The pipeline successfully identified distinct ETRMs for TET1KO and TET2KO mouse brains. For TET1KO, TFs like Egr1, Zic3, and Zeb1 were implicated, while Lhx family members were associated with TET2 function.
  • Quantitative Differences in DMSs: TET1KO brains showed a significantly higher number of differentially methylated sites (DMSs) (42,558) compared to TET2KO brains (12,900), with only a small overlap (643).
  • Validation with ChIP-seq: The predicted ETRMs were supported by ChIP-seq data, showing enrichment of binding sites for the predicted TFs within the corresponding ETRM regions. For example, AP-1 ETRMs were enriched for FOS and JUN binding, and Atoh-ETRM for Atoh1 binding.
  • Complex TF Interactions: The study revealed that ETRMs can involve TFs with different DNA binding domains (e.g., NF1, Tgfi2, Ascl1) or TFs from the same family (e.g., MEF2, SOX).

Limitations and Future Work

  • Motif Database Limitations: The HOMER and JASPAR databases, while extensive, do not contain motifs for all transcription factors. Some TFs bind DNA based on structure rather than sequence motifs, which are not captured by this approach.
  • ChIP-seq Data Availability: The availability of high-quality ChIP-seq data for specific experimental conditions and developmental stages remains a limitation for comprehensive validation.
  • 3D Chromatin Structure: The current model does not integrate information about 3D chromatin folding, which could reveal further interactions between ETRMs.
  • Future Directions: The authors suggest integrating chromatin configuration data into ETRM modeling. They also propose that the developed pipeline can be applied to genomic regions beyond differentially methylated loci to better understand the causes and consequences of methylation alterations in various biological contexts.

Summary updated

Abstract

DNA methylation is an epigenetic modification modulating the structure of DNA molecule and the interactions with its binding proteins. Accumulating large-scale methylation data motivates the development of analytic tools to facilitate methylome data mining. One critical phenomenon associated with dynamic DNA methylation is the altered DNA binding affinity of transcription factors, which plays key roles in gene expression regulation. In this study, we conceived an algorithm to predict epigenetic regulatory modules through recursive motif analyses on differentially methylated loci. A two-step procedure was implemented to first group differentially methylated loci into clusters according to their correlations in methylation profiles and then to repeatedly identify the transcription factor binding motifs significantly enriched in each cluster. We applied this tool on methylome datasets generated for mouse brains which have a lack of DNA demethylation enzymes TET1 or TET2. Compared with wild type control, the differentially methylated CpG sites identified in TET1 knockout mouse brains differed significantly from those determined for TET2 knockout. Transcription factors with zinc finger DNA binding domains including Egr1, Zic3, and Zeb1 were predicted to be associated with TET1 mediated brain methylome programming, while Lhx family members with Homeobox domains were predicted to be associated with TET2 function. Interestingly, genomic loci from a co-methylated cluster often host motifs for transcription factors sharing the same DNA binding domains. Altogether, our study provided a systematic approach for epigenetic regulatory module identification and will help throw light on the interplay of DNA methylation and transcription factors.
BibTeX
@misc{banerjee2024recursive,
  author = {Sharmi Banerjee},
  title = {{Recursive motif analyses identify brain epigenetic transcription regulatory modules}},
  year = {2024},
  url = {https://openreview.net/forum?id=S9Tib5XuLv},
  howpublished = {OpenReview},
}
Figure