Parallel selection on ecologically relevant gene functions in the transcriptomes of highly diversifying salmonids

Background Salmonid fishes are characterised by a very high level of variation in trophic, ecological, physiological, and life history adaptations. Some salmonid taxa show exceptional potential for fast, within-lake diversification into morphologically and ecologically distinct variants, often in parallel; these are the lake-resident charr and whitefish (several species in the genera Salvelinus and Coregonus). To identify selection on genes and gene categories associated with such predictable diversifications, we analysed 2702 orthogroups (4.82 Mbp total; average 4.77 genes/orthogroup; average 1783 bp/orthogroup). We did so in two charr and two whitefish species and compared to five other salmonid lineages, which do not evolve in such ecologically predictable ways, and one non-salmonid outgroup. Results All selection analyses are based on Coregonus and Salvelinus compared to non-diversifying taxa. We found more orthogroups were affected by relaxed selection than intensified selection. Of those, 122 were under significant relaxed selection, with trends of an overrepresentation of serine family amino acid metabolism and transcriptional regulation, and significant enrichment of behaviour-associated gene functions. Seventy-eight orthogroups were under significant intensified selection and were enriched for signalling process and transcriptional regulation gene ontology terms and actin filament and lipid metabolism gene sets. Ninety-two orthogroups were under diversifying/positive selection. These were enriched for signal transduction, transmembrane transport, and pyruvate metabolism gene ontology terms and often contained genes involved in transcriptional regulation and development. Several orthogroups showed signs of multiple types of selection. For example, orthogroups under relaxed and diversifying selection contained genes such as ap1m2, involved in immunity and development, and slc6a8, playing an important role in muscle and brain creatine uptake. Orthogroups under intensified and diversifying selection were also found, such as genes syn3, with a role in neural processes, and ctsk, involved in bone remodelling. Conclusions Our approach pinpointed relevant genomic targets by distinguishing among different kinds of selection. We found that relaxed, intensified, and diversifying selection affect orthogroups and gene functions of ecological relevance in salmonids. Because they were found consistently and robustly across charr and whitefish and not other salmonid lineages, we propose these genes have a potential role in the replicated ecological diversifications.


Background
Identifying the molecular mechanisms underlying adaptive phenotypic divergence is a central challenge for evolutionary biology; a key first step is to detect genes under selection rather than reflecting background neutral evolutionary processes. Parallel or convergent evolution at the molecular level may, or may not, be associated with phenotypic parallelisms across species, but the idea remains compelling [1][2][3] and has been an important analytical framework to advance research in non-model systems [4][5][6]. Molecular parallelism or convergence can be inferred either from nucleotide sitespecific changes [5,[7][8][9] or at a higher level, in the sense of similar genes being targeted by similar selective forces [10][11][12][13].
Fishes have proven a powerful ecological and evolutionary group for comparisons of genes under selection and that are associated with ecological and evolutionary novelty. Sticklebacks, for example, have become a model group of repeated ecological adaptation of Holarctic marine and freshwater distribution [14][15][16]. In cichlid fishes, adaptive potential and highly malleable phenotypes are spread throughout the family. In some cases, it has been shown that relaxed selection [17][18][19][20] or positive selection [21,22] correlate with phenotypic diversification. However, ecological opportunity differs dramatically among cichlid lineages [23][24][25][26][27], which makes it difficult to pinpoint taxa in which adaptive potential is elevated due to a shared genetic toolset [3,6]. In contrast, freshwater lake-resident salmonids of different species and genera have similar ecological opportunity and commonly sympatric distributions across the northern hemisphere [28][29][30]. Furthermore, the freshwater habitats of northern fishes were all colonised on a similar postglacial timescale [30,31], unlike the dramatically different and complex colonisation histories of cichlids [4,[32][33][34].
Salmonid fishes are increasingly used as model organisms in evolutionary research, because of their ecological diversity, economic value, and replicated evolution of distinct ecomorphs and traits in some taxa, such as depth specialisation and alternative migratory tactics [35][36][37][38][39][40][41][42]. Two salmonid genera in particular, the whitefishes (Coregonus) and the charrs (Salvelinus), are not sister taxa but exhibit parallel (or convergent) adaptive tendencies in freshwaters across the northern hemisphere. They have repeatedly diverged into various within-lake ecomorphs along the depth axis over short evolutionary time spans that are unmatched in any other salmonid species [2,[43][44][45][46][47]. The evolutionary and molecular basis for why Coregonus and Salvelinus show such a high degree of ecomorphological adaptability while other salmonid species do not is, however, unknown [39,46,47].
Determining how single and combined effects of selection act at the molecular level is facilitated by new analytical tools [48][49][50]. These selective processes are associated with adaptive evolution in different ways and are most powerful when linked with known lineagespecific phenotype changes or phenotypic diversification [51,52]. Two selective processesrelaxed and intensified selectionare on opposite ends of the spectrum. Relaxed selection decreases the selective constraints of a gene and can lead to the accumulation of nonsynonymous substitutions and consequently changes in the amino acid sequence of a protein. By releasing a gene of selective constraints, relaxed selection can potentially foster phenotypic novelty, plasticity, and evolutionary innovation [4,19,49,53]. In contrast, intensified selection increases selective constraints but can also manifest as positive intensified selection leading to more differences at some sites of a gene [54,55]. Intensified selection implies changes to a gene have strong fitness consequences [49]. Additionally, lineage-specific episodic diversifying selection, or positive selection, will leave other signatures at the sequence level, such as more nonsynonymous changes than expected under neutrality at a subset of positions in a gene on some branches in the phylogenetic tree (i.e., branch-site model) [50,56]. While relaxed and intensified selection are antithetical, in either case diversifying (positive) selection can simultaneously act at a proportion of sites in a gene [57,58].
It has long been proposed that the propensity for ecological speciation in some salmonid lineages is associated with shared patterns of relaxed or diversifying selection on ecologically relevant genes and gene function terms [2,3,19,39,59,60]. We focus on the well-characterised and richly diversifying genera Coregonus and Salvelinus, which show repeated within-lake divergences into distinct ecomorphs across the northern hemisphere [46,47]. The Coregonus species assessed are lake resident at least since postglacial times and have high rates of within-lake adaptive divergence [45,46,61]. Salvelinus species are mostly freshwater residents and have undergone frequent adaptive divergence into ecomorphs along the depth axis, predominantly within lakes [38,47,62,63]. Representatives of all other major salmonid lineages, Oncorhynchus, Salmo, and Thymallus, were also included in the dataset; these are generally riverine or anadromous genera that do not extensively diversify within lakes [64,65]. By assessing consistency across two non-sister lineages, Coregonus and Salvelinus, our approach mitigates against false positives. The focus on orthogroups within and across species, rather than single genes, alleviates the problem of differing relaxation of selective constraint in duplicated compared to non-duplicated or rediploidised genes [66], which is particularly important in salmonids due to the whole-genome duplication (WGD) that their common ancestor experienced 80-103 Mya [67][68][69].
Here, we use a genome-wide comparative approach to test for shared evidence of selection at particular categories of genes, gene functions, and gene ontology terms in the two highly diversifying lineages, Coregonus and Salvelinus, relative to all other major salmonid lineages. We test a comprehensive suite of 2702 orthologous protein-coding gene sets (orthogroups) for signals of parallel relaxed, intensified, and diversifying/positive selection in Coregonus and Salvelinus (average of 4.77 genes per orthogroup; 4.82 Mbp in total; average of 1783 aligned bp per orthogroup). By distinguishing among different kinds of selection in replicate across two independent lineages, our approach can pinpoint the action of selective pressure more accurately. We find that different types of selection target different gene sets and functions in salmonids, with novel and established ecological relevance for repeated, parallel diversification potential.

Selection parameter distribution and number of orthogroups under selection
Shared molecular response to selection in two whitefish and two charr species was inferred relative to six background species (five salmonids and one pike, Fig. 1). The selection parameter k in whitefish and charr, ranging from 0 (very relaxed) to 50 (very intensified), had a median value of 0.992 across orthogroups and was significantly different from the neutral expectation of 1 (Wilcoxon signed-rank test: V = 1,995,900, p = 8.859E-06). Visually, there was an excess of orthogroups with k close to 0, indicating a high number of orthogroups under pronounced relaxed selection (Fig. 2). The number of orthogroups with k < 1 (1387; relaxed selection prevailing) was slightly higher than the number of orthogroups with k > 1 (1308; intensified selection prevailing), but not significantly so (Fisher's Exact Test, p = 0.288).
On 2702 orthogroups in the final dataset we conducted analyses of relaxed and intensified selection (in RELAX) and diversifying/positive selection (in aBSREL with branch-site model). We inferred 138 orthogroups to be under relaxed selection (k < 1, false discovery rate (FDR) < 0.10) in either Coregonus or Salvelinus, of which 122 were found in both Coregonus and Salvelinus. On the other hand, 105 orthogroups showed signals of intensified selection (k > 1, FDR < 0.10), of which 78 included both Coregonus and Salvelinus. The number of relaxed orthogroups in Coregonus and Salvelinus was significantly higher compared to the number of intensified orthogroups (one-sided Fisher's Exact Test, p = 0.035). Of the 2702 orthogroups, branch-site selection analyses inferred 111 orthogroups as being under significant diversifying/positive selection (FDR < 0.10), of which 92 included both Coregonus and Salvelinus. Thus, these orthogroups harbour a proportion of sites with significantly elevated dN/dS (= ω) values in at least one of Fig. 1 Maximum-likelihood phylogenetic tree of nine salmonid species (foreground taxa Coregonus and Salvelinus in red) and outgroup northern pike (Esox lucius; in blue). Node support values are bootstrap values from 1000 bootstrap replications. Branch lengths correspond to the number of substitutions per site. All pictures used here are under public domain. Coregonus and Salvelinus are two genera with exceptional ability for repeated, rapid diversification into ecomorphs within lakes that is unmatched in other salmonid taxa [2,[43][44][45][46][47] the foreground branches leading to Coregonus or Salvelinus taxa.
After averaging selection parameter values for each gene ontology (GO) term, 13 of 1478 GO terms showed significant deviations from the null expectation of k = 1 (Wilcoxon signed-rank tests: p < 0.05; Fig. 3). Eight of these had significantly elevated k values, indicating intensified selection. The other five had significantly lowered k values, which is evidence for relaxed selection. The GO term enrichment results agreed with the general shift of selection (distribution of k) in all orthogroups. The GO terms 'carbohydrate metabolic process' and 'obsolete acylcarrier-protein biosynthetic process', for example, were also present in the orthogroups under intensified selection. The 'ATPase activity' and 'proton transmembrane transport' GO terms were also found among orthogroups under relaxed selection. Other deviating GO terms were 'DNA repair' and 'protein deubiquitination', with evidence for intensified selection, and 'exocytosis' and 'protein dephosphorylation', with evidence for relaxed selection (Fig. 3).

Gene functions under relaxed selection
Blast2GO annotation and UniProt/Swiss-Prot literature research on the orthogroups under relaxed selection identified gene functions with potential relevance for the diversification process in charr and whitefish. Such functions included visual perception (e.g., 'peripherin-2-like'), gene and gene product regulation (e.g., 'E3 ubiquitin-protein ligase RNF128-like'), lipid metabolism (e.g., 'calcium-independent phospholipase A2-gamma-like'), muscle and heart growth (e.g., 'dual specificity protein phosphatase 6'), locomotion (e.g., 'serine/threonine-protein phosphatase PP1-beta catalytic subunit'), and immunity (e.g., 'adaptorrelated protein complex 1', 'natterin-3-like'), but also genes with a role for various nervous system processes We observed compelling trends of GO term enrichment (one-tailed Fisher's Exact Tests, uncorrected p < 0.05 but FDR > 0.10) in the orthogroups under relaxed, intensified, and diversifying selection that largely agree with the research literature on the genes contained in those orthogroups (Fig. 4, Table 1). We found the 122 orthogroups under significant relaxed selection in Coregonus and Salvelinus were enriched for a total of 11 GO terms associated with transcriptional regulation, serine family amino acid metabolism, lipid metabolism, and oxidoreductase activity, amongst others ( Table 1). The REVIGO redundancy analysis results showed transcriptional regulation, serine family amino acid metabolism, lipid metabolism, and acrosome reaction to be among the few non-redundant GO terms (frequency and significance plot of non-redundant GO terms: Fig. 4a, includes clustering by semantic similarity). Transcriptional regulation and serine family amino acid metabolism were the most frequent non-redundant GO terms. In total, six of 11 GO terms were found to be non-redundant.
Among the top ten enriched functions in relaxed orthogroups in both Coregonus and Salvelinus, behaviour and many neural function GO terms and KEGG pathways were found in gene set enrichment analyses (Fig. 5). This is in agreement with the neural process orthogroups and serine family amino acid metabolism GO terms obtained in the GO term enrichment analysis above. The behaviour gene set was the only gene set that was significantly enriched after FDR correction (Fig. 5). Other overrepresented functions included, for example, negative regulation of signalling, urogenital system development, the peroxisome pathway, vascular smooth muscle contraction, and the AGE-RAGE signalling pathway, which plays a major role in inflammation and infection.

Gene functions under intensified selection
The gene functions of the 78 orthogroups under intensified selection in both Coregonus and Salvelinus (results of relaxed and intensified selection analyses: Additional file 1) were found from literature search to be frequently involved in functions relevant for lipid and carbohydrate metabolism (e.g., 'acetyl-CoA carboxylase beta' and 'endoplasmic reticulum mannosyl-oligosaccharide 1,2-alphamannosidase-like', respectively) as well as neurological and bone development (e.g., neurological development: 'synapsin-3', bone development: 'cathepsin K precursor' and 'paired like homeodomain 1').
The orthogroups under intensified selection in Coregonus and Salvelinus were enriched for transcriptional regulation GO terms, but also for those associated with ubiquitine-related processes and steroid hormone receptor activity, amongst others. A total of 18 GO terms were overrepresented (Table 1). Transcriptional regulation and several signalling processes were the only highfrequency GO terms among the few non-redundant GO terms in the REVIGO analysis (frequency and significance plot of non-redundant GO terms: Fig. 4b, includes clustering by semantic similarity). In total, nine of 18  GO terms remained after the REVIGO redundancy analysis.
In the gene set enrichment analysis of all intensified selection orthogroups present in Coregonus and Salvelinus, the 'actin filament-based process' GO term, the 'spliceosome' and several signalling KEGG pathways were among the top enriched functions (Fig. 5). Other functions included 'cellular protein-containing complex assembly', 'fatty acid elongation', 'progesterone-mediated oocyte maturation', and 'steroid biosynthesis'. Overall, the gene enrichment results (Fig. 5) mostly agree with the GO term overrepresentations (Fig. 4b).

Gene functions under diversifying selection
A large number of the 92 orthogroups under diversifying selection in Coregonus and Salvelinus were found in literature search to contain genes involved in regulation of gene expression, signal transduction and transmembrane transporter genes, but also immunity-related genes and a gene of the FOX set of genes, 'FOX I1-ema', a tissue-specific splicing factor important in otic placode formation and jaw development in zebrafish [71] (orthogroups under diversifying selection: Additional file 2, includes associated GO terms).
Orthogroups under diversifying selection were enriched for GO terms associated with transmembrane transport, phospholipid metabolic processes, acetyl-CoA carboxylase activity, various lipid metabolic processes, regulation of Wnt signalling pathway, and RNA splicing, amongst others ( Fig. 4c, Table 1). A total of 47 GO terms were overrepresented, of which 23 remained after the REVIGO redundancy analysis. Pyruvate metabolism, several signal transduction processes, lipid metabolism, and transmembrane transport processes were shown to be amongst the non-redundant GO terms in the REVIGO analysis (frequency and significance plot of non-redundant GO terms: Fig. 4c, includes clustering by semantic similarity). Compared to the orthogroups under relaxed or intensified selection ( Fig. 4a,b), the orthogroups under diversifying selection included a higher number of rather dissimilar low-frequency GO terms, apart from a cluster of similar metabolic GO terms (Fig. 4c). Only one GO term, 'DNA binding', was underrepresented (p < 0.05); with zero occurrences in the orthogroups under diversifying selection in Coregonus and Salvelinus but 77 occurrences in all other orthogroups.

Overlap between selection types
We identified nine orthogroups that showed both signals of relaxed selection (RELAX) and diversifying selection (aBSREL) and 12 orthogroups that showed both signals of intensified selection (RELAX) and diversifying selection (aBSREL) (Fig. 6a, Table 2). The overlap between orthogroups under relaxed and diversifying selection was higher than expected by chance, but not significantly so (hypergeometric expectation: 3.6 vs observed 9; onetailed Fisher's Exact Test: p = 0.126). The overlap between orthogroups under intensified and diversifying selection was significantly higher than expected by chance (hypergeometric expectation: 2.1 vs observed 12; onetailed Fisher's Exact Test: p = 0.004).

Discussion
Our analyses of shared selection in the highly diversifying taxa Coregonus and Salvelinus identified genes and gene functions with deviating signatures of selection compared to five relatively less diversifying salmonid taxa and one non-salmonid species used as background.
We identified more orthogroups under relaxed selection   Table 2). We explored the underlying biological relevance and associated gene functions of these orthogroups under selection and observed trends of overrepresented GO terms in all three sets of orthogroups under selection (Fig. 4, Table 1). What is more, we identified 13 GO terms with k distributions deviating from the null expectation (Fig. 3).
In our functional analyses of the above selected orthogroups, genes in orthogroups under relaxed selection were found to be involved in processes such as transcriptional regulation, nervous system function, muscle and heart growth, lipid metabolism, and immunity, while orthogroups under intensified selection were found to be more often involved in signalling processes, actin filamentbased processes, RNA splicing, carbohydrate metabolism, transcriptional regulation as well as nervous system and bone development (Additional file 1). Orthogroups under relaxed selection were enriched for 'serine family amino acid metabolism' while those under intensified selection tended to be enriched more strongly for various signalling pathways (Fig. 4, Table 1). 'Regulation of transcription' was overrepresented in orthogroups under both relaxed and intensified selection. Orthogroups with signals of diversifying selection showed a clear trend towards enrichment for signal transduction, transmembrane transport, proteolysis, lipid metabolism, and pyruvate metabolism GO terms (Fig. 4, Table 1) and often contained genes involved in transcriptional regulation, development, lipid metabolism, and immunity (orthogroups under diversifying selection, Additional file 2, including associated GO terms). Therefore, our results imply different intensities of selection on different gene functions and processes.

Selection parameter distributionshift to relaxation?
Our results agree with the repeated finding of a relaxation of selective constraint in rapidly diverging taxa [18,19,72,73]. One possible cause behind this observation could be an increase of ecological opportunity due to the emergence of new habitat and a decrease in interspecies competition, for example in caves [72] or within postglacial lakes [74][75][76][77][78][79]. The adaptation to life in lakes, before or in the course of (repeated) postglacial colonisations, such as in charr and whitefish, could be such an ecological opportunity, entailing a release of selective pressures on a substantial number of genes.
From the above results we can also conclude that relaxation of selective constraints does not affect the whole set of protein-coding genes to the same degree. Purifying selection and selective constraint still affect a large number of genes with functions that ensure an organism's integrity, survival, and reproduction (see orthogroups under intensified selection) [80]. However, in charr and whitefish we observed a larger number of orthogroups Fig. 6 Overlapping orthogroups under selection and biological process GO terms. a Overlap between orthogroups under relaxed and diversifying or intensified and diversifying selection in both Coregonus and Salvelinus. b Biological process GO terms of orthogroups under both relaxed and diversifying selection. c Biological process GO terms of orthogroups under both intensified and diversifying selection. In b and c, rectangles are clustered into larger, coloured rectangles based on relationships among GO terms [70]  under relaxed selection that therefore escaped the effects of purifying selection. The observation of largely different GO terms in orthogroups under relaxed selection or intensified selection agrees with a large number of studies that demonstrated different selective pressures on different gene functions [81,82]. Various functional and structural constraints govern the substitution patterns within genes. For example, amino acid alterations in some groups of proteins can easily make them dysfunctional [81], while other kinds of proteins are under less constraint to vary [83] or are even beneficial to fitness when being different from the majority of the population (i.e., negative frequency-dependent or balanced selection) [84][85][86]. The observation of intensified selection on DNA repair in our dataset and the overrepresented GO terms related to oxidative stress and apoptotic signalling, for example, could highlight the importance of an effective response to environmental stress in rapidly

Gene functions under relaxed selection in Coregonus and Salvelinus
Many of the orthogroups under parallel relaxed selection, i.e. those that experienced a release of selective constraint in Coregonus and Salvelinus, have been linked to the divergence of these species along the benthic-limnetic axis within lakes found in earlier research [74,[87][88][89][90][91]. Examples include the 'peripherin-2-like' gene, which has been implicated in visual adaptation in S. namaycush in Lake Superior [92] and the 'natterin-3-like' gene, which plays a putative role as an immunopeptide and experienced pronounced expression divergence in Icelandic S. alpinus ecomorphs [93,94]. Based on our analyses, we propose these are linked to the evolution of novel phenotypes and the variation in feeding and swimming behaviour, energy storage, and the release of competition in postglacial lakes (behaviour, muscle and heart growth, locomotion, lipid metabolism, nervous system function) as well as a release or change of parasite and pathogen burden (immunity genes) in postglacial lakes [95]. This implies changes in selective pressure on these gene functions as a consequence of shifts in the ecology of highly diversifying salmonids.

Gene functions under intensified selection in Coregonus and Salvelinus
Genes under intensified selection are expected to have a strong role for survival and reproduction (i.e., fitness) of organisms [49]. We found the orthogroups under intensified selection showed a high number of processes associated with key metabolic pathways (mainly of carbohydrates and lipids), signal transduction, and regulation of gene expression and the immune system, indicating strong selective pressures on these processes and functions in Coregonus and Salvelinus. The occurrence of immunity genes under intensified selection in our results would rather imply changing parasite and pathogen regimes rather than a complete release of pathogen burden [96]. Interestingly, our results also indicated strong selective pressures on actin filament-based processes and RNA splicing as well as protein modification, which are functions with a crucial role in development [96][97][98][99]. Among these genes, some were found to be directly linked to differences in bone development [e.g., 'paired like homeodomain 1' and 'cathepsin K precursor') [100][101][102]. We speculate that these developmental process genes may be relevant in the extreme morphological diversity that can be found among closely related Coregonus species and particularly Salvelinus species [90,[100][101][102][103][104].

Gene functions under diversifying selection in Coregonus and Salvelinus
Diversifying (i.e., positive) selection on genes, which in this context means that selective pressure favours amino acid polymorphism in a particular gene, affected 92 orthogroups in Coregonus and Salvelinus relative to the non-diversifying taxa. The enrichment of lipid, glycerolipid, and pyruvate metabolism genes is a strong indication that these metabolic processes experienced diversification/positive selection in Coregonus and Salvelinus, which could imply a role in ecological diversification and habitat shifts across the depth axis in lakes. Lipid metabolism has been commonly found to be under diversifying selection in freshwater fishes, such as sticklebacks [105] and salmonids [89,106]. Also, diversifying/positive selection on transmembrane transport, for example, has been shown repeatedly in teleost fish adapting to the freshwater environment, such as sculpin and alewife, putatively because of its role in osmoregulation [107,108].

Overlap between selection types
A combination of approaches, such as aBSREL and RELAX as we apply here, has been shown to reliably infer genes under selection and is also able to appropriately distinguish between positive selection and relaxed selection, which are hard to disentangle using single conventional methods [109][110][111]. This is particularly important since it has been known for some time that the number of false positives in tests for positive or diversifying selection can be substantial [112][113][114][115]. Overlapping sets of orthogroups under selection inferred with different analytical approaches, as we resolve here, can therefore implicate consistent, strong, and specific signals of selection. For example, orthogroups with signals of relaxed and diversifying selection likely experienced a release of purifying selection, with the potential for evolutionary diversification [19,49,116]. One of the orthogroups with the most extreme signal of relaxed and diversifying selection was 'adaptor-related protein complex 1, mu 2 subunit (ap1m2)' (Table 2). Apart from a role in endothelial and intestinal immunity [117], ap1m2 plays a central role in the development of endoderm-derived organs (e.g., stomach and intestine) in zebrafish [118]. Given the differing parasite pressure [119,120] and the diversity of trophic niches in lake systems [121,122], genes under reduced selective constraint could play a crucial role in phenotypic diversity in the often lake-dwelling Salvelinus and Coregonus. Another gene showing a strong signal of relaxed and diversifying selection was 'solute carrier family 6 (neurotransmitter transporter), member 8 (slc6a8)', which regulates creatine uptake in tissues with high energy demands such as muscle and brain tissue [122][123][124]. Recent research on C. mykiss and C. maraena found two copies of this gene, which were either mainly expressed in muscle or kidney tissue, with the strongest expression overall in muscle [125]. Changes of selection regimes on these functions are expected, given that Salvelinus and Coregonus experienced a diversification in trophic and swimming behaviour, ecology, and life history (e.g., Salvelinus and Coregonus can be free swimming, lake bottom dwelling, or restricted to the littoral zone of lakes). Our analyses independently corroborate an important functional role of this gene in highly diversifying salmonids.
Orthogroups showing both intensified and diversifying selection likely constitute targets of "true" positive selection rather than relaxed selection misinterpreted as positive selection. Genes in these orthogroups are thus under stronger selective pressure and are potentially of higher importance to survival and overall fitness [49] in our analyses of Coregonus and Salvelinus. Two of the most strongly significant orthogroups under intensified and diversifying selection involved the genes 'synapsin-3 (syn3)' and 'cathepsin K precursor (ctsk)'. syn3 is involved in neurotransmitter secretion [126] and plays a crucial role in early neuronal differentiation and in neuronal progenitor cell development in fish [127] as well as in mammals [128]. ctsk, encoding a collagenase, has an important function in bone modelling and remodelling as well as bone homeostasis in vertebrates [129,130] and is differentially expressed in S. alpinus [102]. Both syn3 and ctsk are among the genes experiencing the highest selective pressure in Coregonus and Salvelinus as compared to other salmonid taxa, which is an indication of their importance for survival in these two genera and warrants further research.
Not only specific genes but also the gene functions associated with orthogroups under selection were found to have biological relevance. Changes in heart and muscle function, immunity genes, lipid metabolism, and transcriptional regulation have previously been implicated in diversification in salmonids [81,[131][132][133][134][135][136], cichlids [33], ants [83], and tonguefishes [84]. Therefore, our findings from molecular evolution agree with evidence from ecomorphological divergence of Coregonus and Salvelinus of immune response, feeding, metabolism, and locomotion [46,137,138].

Whole-genome duplication and potential effects on relaxed selection
Multiple lines of evidence suggest that whole-genome duplication (WGD) events give rise to vast amounts of genetic material that can subsequently experience elevated substitution rates and relaxation of selective constraints [139][140][141]. The lineage ancestral to salmonids experienced a WGD around 80-103 Mya [67][68][69]. It has been speculated that this may have contributed to new phenotypic innovations and to the ecological success of salmonids [67,[142][143][144][145], although a contribution to diversification rates is contentious [67,146]. While conventional approaches for the detection of positive and relaxed selection based on orthologues in salmonids would potentially be biased in the case of duplicated and non-rediploidised genes, approaches based on orthogroups, as we applied here, alleviate this issue by explicitly comparing selection shared in Coregonus and Salvelinus relative to all other salmonid lineages [58]. This approach circumvents the difficulty of inferring orthologues in taxa that experienced multiple duplications and unequal rediploidisation.
While genes duplicated during the salmonid WGD might be more susceptible to relaxed selection [49,147], charr and whitefish are in different subfamilies and not sister genera (Fig. 1) [41,148]. Therefore, based on evolutionary relationships, they should have rediploidisation patterns more similar to their sister genera than to each other and there is no reason to suspect unequal rediploidisation among genera to be the cause of the relaxed selection signals we identified in Coregonus and Salvelinus. It is more likely that these two genera experienced similar selective pressures due to common diversification patterns, environmental conditions, or other common ecological or life history traits. Our results from molecular evolution analyses reflect these shared ecological and evolutionary patterns. However, recent results on gene expression in duplicated genes of salmonids have shown that relaxed selection often occurs in downregulated gene duplicates [149], which might be associated with dosage balance effects [150,151]. Therefore, future research should ideally analyse selection in concert with gene expression and duplication.

Conclusions
Our findings suggest that there is some evidence for a parallel relaxation of selective constraint in the repeatedly diversifying salmonid lineages Coregonus and Salvelinus compared to all other salmonid lineages, which are known to be less diversifying. Genes in orthogroups under relaxed selection are involved in functions with a potential role in the rapid diversification and ecological adaptation that can be observed in wild populations of Coregonus and Salvelinus and experienced a release of selective pressure, including genes involved in behaviour, muscle function, and infection. On the other hand, orthogroups under intensified selection, such as various signal transduction and regulatory genes, are under stronger selective pressure and are consequently expected to have higher fitness effects in charr and whitefish compared to other salmonids. An independent analysis of orthogroups under diversifying selection showed an enrichment for signal transduction genes and GO terms of various metabolic processes, while genes involved in DNA binding were underrepresented. Orthogroups with both signals of relaxed and diversifying or intensified and diversifying selection can give further hints as to what selective processes govern the evolution of these genes and are also important candidates for gene sets under particularly weak (relaxed) selective pressure (while still experiencing molecular diversification), or strong (intensified) selective pressure. The discovered orthogroups under selection are a valuable resource for further investigations into the importance of certain genes for rapid diversification in salmonids and beyond.

Taxonomic dataset
All analyses were performed on RNAseq transcriptome datasets drawn from a search in NCBI and a literature search in Google Scholar (keywords: each salmonid genus and species, in combination with "transcriptome" and/or "RNAseq" or "RNA-seq") to identify all available studies. The final dataset represents one transcriptome per species of transcriptomes of all available taxa as of October 2016, with additional in-house data for Salmo salar, Salmo trutta, Salvelinus alpinus, and Coregonus lavaretus that was published at a later point [88]. We aimed to retain transcriptomes with a maximal number of overlapping transcripts across salmonids, which consisted of nine species representing all five major genera (Coregonus, Oncorhynchus, Salmo, Salvelinus, and Thymallus), and the closest relative as outgroup (Esox lucius) (Fig. 1, Table 3). The selected transcriptome assemblies were all derived from several tissues so as to obtain almost complete sets of transcripts (see references in Table 3). The topology for foreground branch definitions is based on the phylogenetic tree shown in Fig. 1, which was obtained using maximum-likelihood tree estimation based on a preliminary set of 1564 orthologues (derived from the same set of transcriptomes; orthologue alignments available upon request) and in agreement with the current understanding of salmonid phylogenetic relationships [41,148]. From the full dataset, ranging from 59.0 to 209.2 Mbp per transcriptome (Table 3), we obtained 2702 sequences of orthogroups (mean number of orthologues per orthogroup: 4.77; mean length 1783 bp; range 607-14,743 bp; total 4.82 Mbp). Using a Blast2GO [156] analysis (for details see below), we confirmed that the dataset was not enriched for any particular GO categories (Fisher's Exact Test: all p > 0.10) and therefore not a biased representation of transcripts.

Orthogroup and OG tree inference
The longest open reading frames (ORFs) in the transcriptome sequences of the ten species (excluding mitochondrial genes due to different genetic codes that could bias the selection analyses) and, additionally, protein-coding genes of the salmon genome (NCBI RefSeq assembly accession: GCF_000233375.1) [68] were determined using TransDecoder v2.1 (https://github. com/TransDecoder/TransDecoder). Next, the standard mode of OrthoFinder v2.2.3 [157], which uses a five-step algorithm that mitigates gene length bias and other biases introduced by other methods, was used to infer orthogroups from the TransDecoder ORFs. Based on the OrthoFinder orthogroup assignments, gene sequences were then extracted from the whole transcriptomes with custom Python scripts (Python v3.6.5) if they were also present in the protein-coding sequences of the Atlantic salmon genome v2 (NCBI RefSeq assembly accession: GCF_000233375.1) [68] to exclude spurious transcripts. To further minimise the number of spurious transcripts retained, only orthogroups with a higher or equal number (but not larger than four) of orthologous genes in the focal species compared to the salmon genome were included to avoid inclusion of collapsed orthologues (i.e., multiple genes that were merged into one during assembly, which seems to be a common issue in the salmonid transcriptomes published so far according to in-house results and A. Yurchenko and M. Carruthers, personal communication). Orthologues were then combined into orthogroups using custom Python scripts if they matched the same gene sets in the salmon reference genome. Duplicated orthogroups, inferred as multiple occurrences of the same orthologue sets, orthogroups with a number of sequences exceeding 80, which could include an inflated number of duplicates, or lower than seven, to avoid orthogroups unrepresentative of the whole dataset, and those with fewer than seven taxa were removed with custom Python scripts. The longest ORFs with a minimum length of 200 bp were then extracted from this orthogroup set with get_orfs_or_cdss.py (https://github.com/peterjc/ pico_galaxy/blob/master/tools/get_orfs_or_cdss/get_orfs_ or_cdss.py). A codon-based alignment was then produced using PRANK v.140603 [158]. The aligned orthogroups were then trimmed and filtered with trimAl (−resoverlap 1.0, −seqoverlap 0.38 -noallgaps) [159], optimising both alignment quality and the number of retained alignments. Using the final set of orthogroups as input for FastME [160], we then constructed phylogenies of each orthogroup separately for use in the selection analyses below. The number of retained genes per orthogroup was calculated using custom BASH (BASH v3.2) and Python scripts.

Selection analyses
Foreground branches for subsequent selection analyses were labelled using a custom R script (R v3.5.1) and the R package ape [161][162][163]. All selection analyses were then performed in HYPHY v2.3.13.20180521beta(MP) [48]. Orthogroup alignment FASTA file headers were shortened using custom Python scripts and FASTA and tree files were combined with custom BASH scripts. Using the above foreground definitions, RELAX [49], included in the HYPHY package, was run to infer orthogroups with significant signals of relaxed or intensified selection in the foreground branches relative to the background branches and to infer the selection strength parameter k (as defined in [49]). The software aBSREL (adaptive Branch-Site Random-Effects Likelihood inference) [50], also contained in the HYPHY package, was then used to obtain orthogroups with a proportion of sites under significant diversifying or positive selection in the foreground branches relative to the background branches. Relevant information from both the RELAX and aBSREL output was extracted with custom BASH scripts. Multiple testing was accounted for with false-discovery rate (FDR) correction [164] or Bonferroni correction [165,166] in R. We report on the FDR level of significance throughout the manuscript but also report the orthogroups under selection after Bonferroni correction in Additional file 1. Significant differences in the number of orthogroups under relaxed or intensified selection were quantified using Fisher's Exact Tests in R. A two-sided one-sample Wilcoxon signedrank test in R was used to infer whether the overall selection parameter (k) distribution differed significantly from the null expectation of 1. Selection parameter distributions were visualised using the R package ggplot2 [167]. Hypergeometric expectations of orthogroups under multiple types of selection were calculated in R.

Gene ontology analyses
GO terms were inferred using a complementary set of methods in Blast2GO v5.2.5 [156]. We used BLASTn (megablast) [168,169] using QBlast (NCBI Blast Server) and standard web blast against non-redundant nucleotide sequences and in a second step against protein sequences. A BLAST expectation value (e-value) threshold of 1.0E-3 was used. Only the top blast hit was retained per orthogroup BLAST search. BLAST descriptors were extracted for later annotations. BLASTn was run with a word size of 28, a low complexity filter, a high-scoring segment pair (HSP) length cutoff of 33, and an HSP-hit coverage of 0. GO mapping was performed with the Goa version 2018.12 database [170]. Orthogroups were remapped with InterProScan [171][172][173] using EMBL-EBI InterPro to improve mapping success. Annotations were then created using an annotation cut-off of 55, GO weight of 5, an e-value filter of 1.0E-6, an HSP-hit coverage cut-off of 0, and a hit filter of 500. Further annotations were obtained using merging of InterProScan GOs with existing annotations. Annotations were then augmented by ANNEX [174]. Remaining unannotated sequences were annotated using blast descriptions, applying a minimum similarity of 85 and validation of annotations. Gene functions were inferred based on Blas-t2GO and Uniprot/Swiss-Prot gene information for fish, mice, and humans (last accessed on 27/01/2019). GO term enrichment analyses were performed using Fisher's Exact Tests in Blast2GO on the whole dataset and all sets of orthogroups under selection (from RELAX and aBSREL with Bonferroni multiple correction, FDR correction, or no correction) with FDR multiple correction (FDR threshold of 0.10). GO terms were extracted from the Blast2GO output using custom R scripts. Overlap of sets of orthogroups under selection was determined with custom R scripts. Additionally, GO term enrichment was tested using custom R scripts to see trends in over-and underrepresentation (Blast2GO only outputs significantly over-and underrepresented GO terms after FDR correction). Violin plots of selection parameter k distributions, compared across GO terms, were plotted with the ggplot2 R package. The significance of deviations from the null expectation of k = 1 was tested for each GO term using two-sided onesample Wilcoxon signed-rank tests in R.
To summarise and visualise the obtained GO terms in the different sets of orthogroups under selection, we used the REVIGO online analysis tool with an allowed similarity score of GO terms of 0.9, the zebrafish reference GO term database available in REVIGO, and Sim-Rel as semantic similarity measure (last access: 01.10.2019) [70], which makes use of a semantic clustering algorithm and the p-values from the GO term enrichment analysis in R above. In this context, semantic similarity refers to the degree of relatedness between two entities/GO terms by the similarity in meaning of their annotations [175]. REVIGO clusters GO terms of sets of genes based on this algorithm in a twodimensional semantic similarity space. Additionally, REVIGO ranks the GO terms of genes according to their redundancy.

Gene set enrichment analysis
Gene set enrichment analysis (GSEA) was performed on gene symbols of zebrafish homologues of human genes derived from Blast2GO. Zebrafish homologues were obtained from http://www.informatics.jax.org/downloads/ reports/HOM_AllOrganism.rpt using a custom R script. The selection parameter (k) values from the above RELAX analysis were transformed into -1 to 1 scores in R, with negative values representing relaxed selection orthogroups and positive values representing intensified selection orthogroups. We then used WebGestalt [176][177][178][179] on the zebrafish gene symbols and associated transformed k scores to derive a) the top ten enriched biological process GO terms and b) the top ten enriched KEGG pathways in both the relaxed and intensified orthogroups. We performed 10,000 permutations for each GSEA run. All other parameters were kept at default settings (last use: 02.10.2019).