Comparative regulomics supports pervasive selection on gene dosage following whole genome duplication

Abstract

Background: Whole genome duplication (WGD) events have played a major role in eukaryotic genome evolution, but the consequence of these extreme events in adaptive genome evolution is still not well understood. To address this knowledge gap, we used a comparative phylogenetic model and transcriptomic data from seven species to infer selection on gene expression in duplicated genes (ohnologs) following the salmonid WGD 80-100 million years ago.

Results: We find rare cases of tissue-specific expression evolution but pervasive expression evolution affecting many tissues, reflecting strong selection on maintenance of genome stability following genome doubling. Ohnolog expression levels have evolved mostly asymmetrically, by diverting one ohnolog copy down a path towards lower expression and possible pseudogenization. Loss of expression in one ohnolog is significantly associated with transposable element insertions in promoters and likely driven by selection on gene dosage including selection on stoichiometric balance. We also find symmetric expression shifts, and these are associated with genes under strong evolutionary constraints such as ribosome subunit genes. This possibly reflects selection operating to achieve a gene dose reduction while avoiding accumulation of "toxic mutations". Mechanistically, ohnolog regulatory divergence is dictated by the number of bound transcription factors in promoters, with transposable elements being one likely source of novel binding sites driving tissue-specific gains in expression.

Conclusions: Our results imply pervasive adaptive expression evolution following WGD to overcome the immediate challenges posed by genome doubling and to exploit the long-term genetic opportunities for novel phenotype evolution.

Fig. 1

Expression level evolution following WGD. a Phylogenetic tree of the species included in the study, with the estimated time of the salmonid-specific whole genome duplication (Ss4R) indicated. b Conceptual illustration of the expression level evolution tests. c Proportion of complete singleton (top) and ohnolog (bottom) gene trees with significant shifts in expression level in a salmonid ancestor. d, e Heatmaps show tissue expression, from an independent tissue atlas in Atlantic salmon, of ohnolog pairs where one copy has shifted up (d) or down (e) in liver. Barplots show the distribution of the number of tissues where the shifted copy has lower or higher expression than the conserved copy. Only ohnologs from complete orthogroups (panel c) are included in the heatmap. Each ohnolog pair (row) is scaled so that red signifies the highest expression across the two copies and blue the lowest. The color bar indicates the number of tissues that are experiencing a shift in expression in the same direction as that of liver (down (d), up (e)) between the shifted and conserved copy. f Proportion of partial gene trees (i.e., trees with some gene loss) with significant shifts in expression level in a salmonid ancestor. The shadings indicate that we report here up/down shifts for the complete salmonid clade and the partial salmonid clade separately, which is in contrast to panel c where both salmonid clades are complete and therefore indistinguishable. g Cumulative proportion of dN/dS for ohnologs with one copy shifted down, versus their conserved counterpart. Results are shown for all ohnologs with one copy shifted down (down-shift) and for the subset that is down-shifted in all tissues in the tissue atlas (down-shift all tissues affected). h Cumulative proportion of TE content in promoters of ohnologs with one copy shifted down

Fig. 2

Symmetry of regulatory divergence. a Ohnolog expression evolution categories and expression evolution asymmetry for ohnologs in each evolutionary category. The expression asymmetry is calculated as the absolute value of the mean difference between ohnolog pair expression levels in all salmonid species. One sided Wilcoxon test p-values are reported for significant asymmetry differences between symmetric and asymmetric regulatory categories. b KEGG pathways significantly enriched (p < 0.05) in different expression evolution categories. Larger circles indicate a higher proportion of genes in the pathway with the shift. c Expression asymmetry between salmonid ohnolog pairs in selected pathways, calculated by taking the absolute value of the mean difference in expression between ohnolog pairs in all salmonid samples. d Correlation between expression asymmetry (see (c) for details) and the dN/dS of the ortholog in the pike sister lineage. e Predicted bound TFBS from TF-footprinting in promoters of ohnologs in the five expression evolution categories as well as those ohnologs with no significant shift in expression levels. For each ohnolog pair in each category, copies are grouped based on the lowest (to the left) and highest (to the right) p-value in the OU-test for expression level shift. p-values from significant paired Wilcoxon tests are indicated above boxplots: *** < 1e−03, **** < 1e−04, ***** = 0

Fig. 3

Transcription factor binding site evolution. a The number of liver-specific TFs (56 in total) with at least one bTFBS in the promoters of the 30 ohnologs with one liver-specific up-shifted copy (Up) or one conserved copy (Cons). b Tissue expression of the 30 ohnolog pairs where one copy has evolved a liver-specific gain in expression (color bar: up-shifted copies are red and conserved copies are gray) and 22 liver-specific TFs predicted to bind at least one-third of the targets (purple). TFs are named according to their motif(s) in JASPAR. Liver-specific genes are defined as having liver expression levels in the 90% quantile and tau-scores > 0.6. Each gene (row) is scaled so that red signifies the highest expression across the tissues and blue the lowest. c Regulatory network reconstructed for the ohnologs and selected TFs from b using footprinting data. Ohnologs are represented by circles sized by their regulatory complexity (in-degree) and colored according to their evolutionary expression shift with red signifying up-shift and blue down-shift. TFs are represented by diamonds with the nine most up-shift-biased TFs shown. A directed gray edge means that the TF has at least one bTFBS in the promoter of the gene. A dotted undirected green edge connects ohnologs