DMRforPairs: identifying differentially methylated regions between unique samples using array based methylation profiles

Abstract

Background: Array based methylation profiling is a cost-effective solution to study the association between genome methylation and human disease & development. Available tools to analyze the Illumina Infinium HumanMethylation450 BeadChip focus on comparing methylation levels per locus. Other tools combine multiple probes into a range, identifying differential methylated regions (DMRs). These tools all require groups of samples to compare. However, comparison of unique, individual samples is essential in situations where larger sample sizes are not possible.

Results: DMRforPairs was designed to compare regional methylation status between unique samples. It identifies probe dense genomic regions and quantifies/tests their (difference in) methylation level between the samples. As a proof of concept, DMRforPairs is applied to public data from four human cell lines: two lymphoblastoid cell lines from healthy individuals and the cancer cell lines A431 and MCF7 (including 2 technical replicates each). DMRforPairs identified an increasing number of DMRs related to the sample phenotype when biological similarity of the samples decreased. DMRs identified by DMRforPairs were related to the biological origin of the cell lines.

Conclusion: To our knowledge, DMRforPairs is the first tool to identify and visualize relevant and significant differentially methylated regions between unique samples.

Figure 1

Flowchart and overview of DMRforPairs results of the Illumina data. (A) The subsequent steps of recoding probe classes, identifying regions, quantifying and testing methylation differences and exporting the results are described in detail in the main text. Briefly, 473,151 probes remained after quality control. Subsequently, probes not associated to any of the 11 classes or not included in any of the regions are discarded. 145,537 probes (35%) were included in 29,404 potential regions of interest. Finally, these were assessed for methylation differences. (B) Number of regions identified in the various pairwise analyses. * = relevant indicates regions with |ΔM| > 1.4, ** = significant indicates relevant + p_adjusted ≤ 0.05. “repl.” indicates technical replicates. (C,D) The density plots illustrate the distribution of all/the relevant/the significant regions with regard to the number of probes in each region. Only the comparisons of the two cancer cell lines (C) and the pair of lymphoblastoid cell lines (D) are depicted as the technical replicates yielded no significant DMRs.

Figure 2

Tuning of the of d_min and n_min parameters. (A) Number of regions identified and (B) fraction of all probes included in these regions using different settings of d_min and n_min. d_min denotes the maximal distance in bp allowed between two adjacent probes to be accepted in the same region. n_min denotes the minimal number of probes in a region (per sample). All runs of the algorithm were done using the 415,712 probes annotated to at least one Illumina class grouped according to gene/transcription start site/CpG island (recode parameter = 1). These benchmark statistics can be generated using the tune_parameters function in the algorithm (optional parallelization).

Figure 3

DMRforPairs output. (A) One row of the HTML table describing one DMR. Thumbnail, genomic annotation and descriptive statistics regarding (the difference between) the samples are presented as well as links to figures/tables illustrating the methylation patterns in the samples in detail. Direct links to the genomic region in two genome browsers are also provided (Ensembl & UCSC). Region IDs are generated on the fly by the regionfinder function and are specific to a dataset and to a set of DMRforPairs parameters. They are therefore not interchangeable between datasets/experiments and serve mainly as identifiers during exploration of the dataset. (B) Methylation level per probe (M and β values) plotted against its genomic position. These plots are generated for all relevant and significant regions. (C) Annotated visualization of DMRs (β values) ±10 kb. Black box indicates the DMR. Transcripts overlapping/near the region are retrieved from Ensembl. These plots are optionally generated for all relevant and significant regions. (D) Additional statistics (STATS link in table) as provided for each region.