Bioinformatics

RNA-seq analysis was performed as described [29]. Reads were processed with Trim Galore, mapped with the subread aligner and reads were assigned the ENSEMBLE genome using featureCounts [30], using standard settings. The samples had on average 25.3 million RNA reads assigned to genes, and the lowest and highest numbers were 21.6 and 29.6 million reads, respectively. Differential testing was performed using the edgeR function glmlRT [31], with a model of the form ∼ group + participant, where group encoded the four combinations of untrained/trained and Basal/240 min.

RRBS analysis was performed as described [32]. Reads were preprocessed with Trim Galore. Mapping and methylation level estimation were performed with Bismark [26], using standard settings. The average coverage was 8.8 million reads and varied from 2.7 to 17.5 million reads. Differential methylation was found using BiSeq [28], with a model formulated as ∼ timestamp + participant, where timestamp encoded the four time-point sequences. The trained and untrained states were analyzed separately.

In order to discriminate adipocyte-specific reads from those due to immune cell infiltration induced by acute exercise in our RRBS and RNA-seq datasets, we used available epigenetic and transcriptomic data profiling macrophages.

Macrophage methylation data (produced by the BLUEPRINT consortium, EGAD00001000914 and EGAD00001001139) were downloaded from http://dcc.blueprint-epigenome.eu/#/datasets after data access agreement. Macrophage expression profiles were downloaded from Gene Expression Omnibus (accession number {"type":"entrez-geo","attrs":{"text":"GSE36952","term_id":"36952"}}GSE36952). All datasets were processed using the same settings as the data generated in this study. Macrophage methylation percentages were calculated for the genomic clusters previously defined on white adipose tissue methylation data. For each participant, macrophage infiltration after training (20, 60 and 240 min for methylation and 240 min for RNA) was calculated by minimizing:

subject to the constraint:

where v_t, v_basal and v_macrophage are the fractions of methylation or the log2 counts per million (logCPM) for expression at time point t, at the basal state or in macrophages, respectively and x is the fraction of macrophage infiltration. Fraction of infiltration was calculated separately for the trained and the untrained state. The formula was solved using quadratic programming as implemented in the R package quadprog. Residual differences were calculated as:

For RNA-seq data, e was calculated on logCPM, while for methylation data, CpGs were aggregated to an overall error of the cluster by taking the median. The distribution of residuals was observed to be symmetrically centered on zero. Individual clusters or genes whose change could be attributed to macrophage infiltration were identified by comparing them to the set of all clusters or genes using a Mann–Whitney U test.