Periodic Reporting for period 1 - CHROMISE (Identifying Variable Chromatin Modules using single-cell epigenomics)
Berichtszeitraum: 2022-09-01 bis 2024-08-31
A thorough understanding of the genetic contribution to complex traits or disease susceptibility is of great biomedical importance, and may allow to prioritize regions for targeted editing of genetic variants that predispose to or are directly causal for onset of disease. However, only a few studies have so far been able to mechanistically disentangle how regulatory variants contribute to variability in human inter-individual disease susceptibility, as most trait-associated variants appear to fall in non-coding, likely regulatory regions of the genome. Classical views postulated that regulatory variation affects the interaction of transcription factors (TFs) with DNA, which locally affects gene expression and chromatin modifications. However, only a small part of inter-individual variable TF binding can be explained by sequence differences in the respective binding sites. Conversely, DNA regions exhibit a high level of local molecular coordination, which are referred to as variable chromatin modules (VCMs) (other names include cis-regulatory domains (CRDs), chromatin nanodomains or microdomains). VCMs comprise coordinated TF binding and modification of histones in confined genomic locations, which collectively define the activity of the entire locus, i.e. the whole VCM. Within these VCMs, genetic variation can impact chromatin accessibility independent of gene expression, and genetic changes in only certain regulatory elements (REs) can control the activity profile of all other molecular phenotypes. Here we have benchmarked and outlined the best practices to infer VCMs from high-throughput epigenomics data. We further show that single-cell epigenomics (via assay for transposase-accessible chromatin with sequencing (ATAC-seq)) may provide a new means to map VCMs from limited input material. Finally, we provide evidence that VCMs are dynamic during cell state changes, even when the underlying regulatory elements may not change in their epigenetic profile.
We first benchmarked and assessed three main available computational methods for mapping VCMs using publicly available data for lymphoblast cell lines. We identified important criteria for mapping VCMs, including sample size, signal-to-noise ratio, and genomic distance and (empirical) p-value thresholds. We finally concluded that of the tested approaches, Clomics was the most appropriate tool, and provide recommendations for parameter values that should be used. We also applied the resulting workflow on public data for several types of immune cells and identified wide variations in terms of VCM localization and embedded genes between the tested cell types, with the highest similarity between more closely-related cell types.
We then obtained adipose stem and progenitor cells (ASCs) from fat biopsies from twenty-two humans. We mixed them together and profiled these in a single scATAC-seq experiment, followed by demultiplexing of the cells to the individuals based on genotype information. We used the resulting data for aggregating individuals to pseudo-bulk data and used this as input for mapping VCMs. We used the information learned from the benchmarking exercises to map a substantial number of VCMs using this approach. We also generated bulk ATAC-seq for the individuals as a reference, and found that mapping VCMs from single-cell data yields a lower but still decent number of VCMs compared to bulk (~50%). This is likely owing to the more sparse nature of single-cell data and may be solved by deeper sequencing or including more cells per human. As the single cell based assays also identified VCMs on a range of positive control regions such as the RHD and GSTT1 gene, we concluded that single-cell epigenomes may be a means to infer regulatory hierarchies including VCMs.
We also used the hASCs to differentiate these to mature adipocytes. We readily observed stark phenotypes variation between the humans in terms of how well the cell lines differentiated into mature adipocytes, despite the fact that prior to differentiation these cells seemed molecularly comparable. To generate a comprehensive and expanded dataset, we mixed ASCs from 22 donors and used these prior to differentiation (t0) or differentiated these together to either mature adipocytes (t14 adipo) or osteoblasts (t14 osteo). We mixed all donors onto a single 10x Chromium chip per time point for single cell ATAC-seq and demultiplexed the individual donors based on the genotypes (that we obtained using SNP arrays). These assays allowed us to map VCMs during cell state transition and provide a comprehensive overview of their stability and dynamics. Finally, we obtained a range of genetic variants (also referred to as quantitative trait loci (QTLs)) associated with chromatin modules.
We then obtained adipose stem and progenitor cells (ASCs) from fat biopsies from twenty-two humans. We mixed them together and profiled these in a single scATAC-seq experiment, followed by demultiplexing of the cells to the individuals based on genotype information. We used the resulting data for aggregating individuals to pseudo-bulk data and used this as input for mapping VCMs. We used the information learned from the benchmarking exercises to map a substantial number of VCMs using this approach. We also generated bulk ATAC-seq for the individuals as a reference, and found that mapping VCMs from single-cell data yields a lower but still decent number of VCMs compared to bulk (~50%). This is likely owing to the more sparse nature of single-cell data and may be solved by deeper sequencing or including more cells per human. As the single cell based assays also identified VCMs on a range of positive control regions such as the RHD and GSTT1 gene, we concluded that single-cell epigenomes may be a means to infer regulatory hierarchies including VCMs.
We also used the hASCs to differentiate these to mature adipocytes. We readily observed stark phenotypes variation between the humans in terms of how well the cell lines differentiated into mature adipocytes, despite the fact that prior to differentiation these cells seemed molecularly comparable. To generate a comprehensive and expanded dataset, we mixed ASCs from 22 donors and used these prior to differentiation (t0) or differentiated these together to either mature adipocytes (t14 adipo) or osteoblasts (t14 osteo). We mixed all donors onto a single 10x Chromium chip per time point for single cell ATAC-seq and demultiplexed the individual donors based on the genotypes (that we obtained using SNP arrays). These assays allowed us to map VCMs during cell state transition and provide a comprehensive overview of their stability and dynamics. Finally, we obtained a range of genetic variants (also referred to as quantitative trait loci (QTLs)) associated with chromatin modules.
The notion that regulatory elements in the genome work together to orchestrate gene expression has been emerging in the recent decade, but comprehensive mapping of such collaborative units with high resolution is frequently costly and technically challenging. The work presented here provides an orthogonal approach to do this in an efficient and high throughput manner, and shows that it is possible to do this based on single cell data. As single cell datasets are generated in increasing number, the obtained framework will be beneficial for (re)analysis of such datasets. We thus foresee that the obtained frameworks may aid in future studies on how variants in disease-associated genetic loci (for example obtained through GWAS) could be selected for targeted genetic therapies. This is relevant as with the rise of technologies such as base and prime editing, making genetic changes directly in human cells becomes possible and is readily performed in the clinical setting.