TL;DR
This paper introduces a novel method to quantify histone modifications by counting high-resolution peaks, creating gene profiles that can be linked to cellular phenotypes. They successfully apply this to model the complex differentiation of blood cells, demonstrating that multiple histone modifications collectively contribute to this process.
What Problem Does This Paper Address?
Understanding how cells differentiate, particularly into the diverse types of blood cells (hematopoiesis), is a fundamental challenge in biology. This process is regulated by intricate epigenetic mechanisms, including histone modifications. While it's known that these modifications influence gene expression, precisely quantifying and integrating signals from multiple modifications to understand complex biological processes like differentiation has been difficult. The paper aims to bridge this gap by developing a quantitative approach to profile histone modifications and use these profiles to test hypotheses about cellular differentiation.
Methods
The core of the paper's methodology involves several key steps:
- Peak Calling: Instead of just looking at the presence or absence of histone modifications, the authors focus on counting "high-resolution peaks" within genes. This is a more granular approach to quantifying the modification signal. They developed a specific algorithm to identify these small peaks.
- Normalization: To account for variations in sequencing depth, they explored two normalization methods: Counts Per Million (CPM) and Reads Per Kilobase per Million (RPKM). CPM focuses on the concentration of peaks, while RPKM considers their distribution across the gene.
- Cell Type Profiling: For each cell type, they aggregated data from multiple samples to create a robust histone modification profile. They also employed feature selection to retain only the most informative genes.
- Similarity Network Fusion (SNF): To integrate information from the six different histone modification marks, they used the SNF algorithm. This method combines multiple similarity networks (one for each modification) into a single, unified network that captures shared and strong signals across all modifications.
- Hypothesis Testing with Graph Cuts: To evaluate the myeloid/lymphoid differentiation hypothesis, they transformed the similarity networks into dissimilarity networks. They then applied graph-cut techniques to partition these networks, comparing the cost of a "hypothesis cut" (separating myeloid and lymphoid cells) with the "maximum cut" (the optimal partitioning of the graph). This allows them to quantify how well the data supports the proposed differentiation model.
Datasets
The study utilized data from the blueprint consortium and the International Human Epigenome Consortium (IHEC). Specifically, they used genome-wide profiles of six histone modifications (H3K4me1, H3K4me3, H3K27me3, H3K36me3, H3K27ac, and H3K9me3) from 24 distinct cell types involved in hematopoiesis. These cell types were carefully selected to ensure that data for all six histone modifications were available for each.
Key Insights and Ideas
- Peaks as a Quantitative Signal: The paper demonstrates that counting high-resolution histone modification peaks can serve as a meaningful quantitative measure, analogous to gene expression levels. This offers a new way to build profiles of epigenetic marks.
- Histone Modifications Behave Like Gene Expression: The distribution of histone modification peaks across genes mirrors that of gene expression, with only a fraction of genes showing significant activity. This suggests a shared regulatory logic.
- Collective Contribution of Modifications: All six histone modifications examined appear to contribute almost equally to the myeloid/lymphoid differentiation process. This highlights the combinatorial nature of epigenetic regulation.
- SNF for Integrated Epigenetic Analysis: Similarity Network Fusion proves to be an effective tool for integrating diverse epigenetic signals, providing a more comprehensive view of cellular states.
- Myeloid/Lymphoid Model is an Approximation: While the classical myeloid/lymphoid differentiation model is a reasonable approximation, the results suggest it doesn't fully capture the complexity of hematopoiesis, hinting at other regulatory mechanisms at play.
Key Results
- The developed peak-calling method successfully created gene profiles for histone modifications.
- Experiments on 24 cell types showed that the myeloid/lymphoid differentiation hypothesis is well-supported by the data, with high scores (around 0.9) from the hypothesis testing scheme for both individual modifications and the fused SNF network.
- The RPKM normalization generally resulted in slightly lower scores than CPM, suggesting that the concentration of histone modification peaks (CPM) might be more critical for triggering a phenotype than their uniform spread across a gene (RPKM).
- The analysis indicated that all histone modifications contribute similarly to the myeloid/lymphoid differentiation.
Limitations and Future Work
- Model Completeness: The authors acknowledge that while the myeloid/lymphoid model is a good approximation, it likely doesn't encompass the entire complexity of hematopoietic differentiation, suggesting other regulatory mechanisms are involved.
- Future Directions: The paper opens avenues for using differential analysis techniques to identify specific genes involved in particular phenotypes based on their histone modification profiles. The developed framework could also be applied to validate and compare hypotheses in other biological contexts.
Summary updated