Ectodysplasin signalling genes and phenotypic evolution in sculpins (Cottus)

Abstract

Despite their deeply conserved function among vertebrates, ectodysplasin (Eda) signalling genes are involved in microevolutionary change in humans and sticklebacks. If such a dual role is common, Eda signalling genes constitute hotspots for morphological evolution. Variation in sculpin (Cottus) skin prickling and body shape resembles patterns caused by variation in Eda signalling in sticklebacks. We mapped Eda signalling genes and performed quantitative trait locus mapping in crosses between Cottus rhenanus and Cottus perifretum. A genomic region containing the Eda receptor (Edar) was strongly associated with prickling and contributed to shape. The expression of Edar in developing prickles and skeletal elements in Cottus was confirmed by in situ hybridization. Coding sequence changes between Edar alleles in C. rhenanus and C. perifretum exceeded sequence differentiation in other vertebrates. However, it is likely that additional genetic elements besides coding changes affect the phenotypic variation. Although the phenotype in a natural hybrid lineage between C. rhenanus and C. perifretum resembles C. perifretum, the respective coding Edar alleles are not fully fixed (88.6%). Hence, our results support an involvement of Eda signalling in microevolutionary changes, but imply that the Edar gene is affected by multiple evolutionary processes that vary among freshwater sculpins.

1. Introduction

Sculpins (Cottidae) are promising targets for evolutionary analyses because their speciation rates are among the fastest in fishes [1,2]. A monophyletic group of Cottus has radiated in freshwater habitats of the Northern Hemisphere [3]. Closely related species and populations in Europe and North America display polymorphism in skin prickling (figure 1a) [4–7]. Although the variation in prickling can be gradual, we refer to intensely prickled forms as, ‘prickled’ and to a reduced prickling as, ‘non-prickled’ throughout this text. Prickling consists of calcified spicules embedded in the skin, and it is likely that prickles and fish scales are homologous. The presence of scales is the ancestral state in fishes, but reduced squamation already evolved among marine sculpins [8]. Hence, it is not clear whether the last common ancestor of Cottus already carried skin prickling. This inference is complicated by the fact that the function of prickling is not known. The phylogenetic relationships of freshwater Cottus do not reflect the differentiation into prickled and non-prickled species [3], implying a fast evolutionary turnover of skin prickling. These replicated cases of evolution in a phylogenetic context allow us to identify the genetic basis of evolutionary change [9]. This permits analyses of the connections between molecular and phenotypic evolution within a diverse radiation of fishes. Moreover, comparisons with more distantly related species can uncover patterns of parallel evolution.

Figure 1.

Prickling and ectodermal bones in sculpins. (a) The degree to which prickles cover the body differs between Cottus species. Ossifications encapsulating the lateral line pores are visible in both species (black triangles). Prickles were stained using alizarin red in C. perifretum (top) and C. rhenanus (bottom). (b) Skin prickling was scored as categories (0–4; left). The number of leaves from the sunflower graph (right) indicates the number of individuals in each category. C. rhenanus are mostly devoid of prickles, whereas invasive Cottus share intense prickling with C. perifretum.

Homology of prickles and scales implies that genes known to be involved in the development of fish scales should affect prickles. Five different genes from the ectodysplasin (Eda) signalling pathway have been shown to influence ectodermal structures [10–12] and are good candidates for causing variation in skin prickling. Particularly, Eda and the Eda receptor (Edar) affect the development of scales, skeletal and dental structures in zebrafish and medaka [13,14]. Eda signalling also controls the development of lateral plate polymorphisms between marine and freshwater sticklebacks [15]. Eda signalling in sticklebacks represents one of the best-studied examples in which the links between phenotypic and genetic traits have been demonstrated in an adaptive evolutionary context [16]. An involvement of Eda signalling in microevolutionary processes contrasts with the fact that functions of the involved genes are deeply conserved among vertebrates [11]. This suggests a dual role in maintaining conserved functions while facilitating rapid evolutionary change [12]. To date, rapid evolution involving Eda signalling genes has been described for sticklebacks and humans, and is also suspected to occur in other vertebrates [12,17]. If such patterns are indeed widespread, then Eda signalling genes could constitute hotspots of phenotypic evolution [17]. The purpose of this study is to test whether Eda signalling genes contribute to the variation of prickling phenotypes in Cottus because they resemble the lateral plate polymorphism in sticklebacks.

The exploration of candidate genes from the Eda pathway in Cottus is facilitated through comparisons with sticklebacks (Gasterosteus sp.). Recent phylogenetic studies have suggested that Scorpaeniformes, including Cottidae, are among the closest relatives of the Gasterosteiformes [18]. In agreement with this, stickleback and Cottus genomes have a well-conserved synteny [19]. Therefore, the transfer of genomic information from the sequenced genome of stickleback to Cottus is very promising and permits generalization of findings from the stickleback.

We focus on the European freshwater sculpins C. rhenanus and C. perifretum. They differ in a number of morphological traits, including skin prickling, body shape, number of vertebrae and the development of the lateral line system [4,5,20]. It is noteworthy that quantitative trait loci (QTL) that underlie lateral plate phenotypes also had effects on skeletal elements and shape in sticklebacks [21,22]. Accordingly, Eda signalling potentially affects several of the phenotypes that differ between C. rhenanus and C. perifretum, which can be analysed through genetic mapping [19]. Moreover, we have found a natural hybrid lineage (invasive Cottus) that has emerged from the two species and offers unique possibilities to study phenotypes genetically [4]. Parental genetic traits segregate in invasive Cottus, permitting analyses of genetic associations in natural populations. A conspicuous feature of invasive Cottus is that their prickling and shape resemble one ancestor (C. perifretum, prickled) more than the other one (C. rhenanus, non-prickled) [4]. The similarity to one parental species seems paradoxical given the common notion that hybrids are intermediate between parental taxa. Similarity of invasive Cottus and C. perifretum can be reconciled only with an admixed genome [23] if the alleles that cause the respective phenotypes in invasive Cottus originate also from C. perifretum.

We hypothesized that Cottus prickles represent an example of rapid Eda signalling evolution. Here, we used QTL analysis to identify genomic regions that affect prickling in F₂ crosses between pure C. perifretum and C. rhenanus. QTL mapping was complemented with analyses specifically targeting Eda signalling genes. Evidence that the best candidate gene according to our QTL analysis plays a role in the development of prickles and the differentiation of Cottus skeletons was sought using in situ hybridization. We also analysed signatures of molecular evolution by comparing Cottus sequences to sequences of closely and distantly related species, and we tested whether or not alleles in the invasive Cottus gene pool are indeed derived from C. perifretum.

2. Material and methods

(a). Study populations

This study analyses traits in two Cottus species and compares the results with their natural hybrid lineage as well as outgroup species. C. perifretum (prickled) was represented by two populations, Witte Nete (WN) and Laarse Beek (LB), and C. rhenanus (non-prickled) was represented by the populations Broel and Naaf. Stocks were collected and bred in aquaria for genetic mapping (table 1 and electronic supplementary material, table S1). Outgroup samples for evolutionary analyses were available from previous studies ([4,23], electronic supplementary material, table S2). See the electronic supplementary material.

Table 1.

QTL mapping results for prickling and body shape in F₂ crosses between C. perifretum (WN and LB) and C. rhenanus (Broel and Naaf). Traits include prickling, shape CVA scores, HL/EW (head length/eye width ratio), HL/BL (head length/body length ratio). Numbers of individuals and families used for each trait are given. QTL that exceed the significance threshold (LOD) in a permutation test (α < 0.05) are shown with their respective linkage group (LG), map position (cM), maximum LOD score, the most closely linked markers and the percent variance explained (PVE).

trait	Broel × WN						Naaf × LB
	N	LG	cM	LOD	marker	PVE	N	LG	cM	LOD	Marker	PVE
prickling	453 (6)	3	66	9.38	EDAR8930	9.09	195 (4)	3	84.5	21.96	ctg16139	11.55
								3	36.9	21.48	EDAR8930	10.69
shape CVA	274 (4)	n.a.	n.a.	n.a.	n.a.	n.a.	267 (4)	3	84.5	17.44	ctg16139	13.61
								3	36.9	8.09	EDAR8930	0.97
HL/EW		16	46.2	3.75	ctg04045	6.10		3	84.5	10.17	ctg16139	8.83
								3	36.9	4.53	EDAR8930	0.62
HL/BL		3	15.6	4.80	ctg01920	7.74		10	30.5	3.47	ctg01402	5.93

(b). Phenotype analysis

Individuals were raised to a minimal length of 3 cm. Prickles were stained with alizarin red and scored (figure 1). Body shape was analysed from a dorsal perspective (figure 2) using geometric morphometric approaches (canonical variate analysis, CVA) and traditional morphometric ratios. See the electronic supplementary material.

Figure 2.

Body shape differentiation in sculpins. (a) A set of nine anatomical landmarks was chosen to capture important shape differences between Cottus species from a dorsal view (top). The shape changes can be visualized as arrows that describe the direction and magnitude of shape changes when mean landmark configurations are transformed from C. perifretum to C. rhenanus (bottom). (b) Populations of the two Cottus species separate along two CV axes. The parental species separate along the first CV axis. (c) Traditional morphometric ratios obtained from distances between landmarks equally reveal differences in shape. Ratios between head length and body length or eye width distinguish C. rhenanus and C. perifretum.

(c). Cottus homologues of the Eda signalling pathway

Genome sequences of Cottus Eda signalling genes (Eda, Edar, Edaradd, Troy, Traf6 and Nemo) were assembled from Cottus genomic DNA and cDNA sequences. Eda signalling gene genomic scaffolds were screened to identify indel polymorphisms (electronic supplementary material, table S3) that we used for genetic mapping. See the electronic supplementary material.

(d). Quantitative trait loci analysis

Interspecific F₂ crosses between populations of C. perifretum (prickled) and C. rhenanus (non-prickled) were used for QTL mapping. Breeding pairs represented two independent types of crosses (Broel × WN and Naaf × LB). Within each cross type, the directions of the crosses (species affinity of father) varied among breeding pairs (electronic supplementary material, table S1). Genotypic data from several breeding pairs were combined for each cross type. ‘Family’ (offspring from one pair) was used as a covariate in the QTL analysis. The analysis was based on genotypes and a Cottus genetic map that were generated previously [19]. The existing data were complemented with genotypes for additional gene-specific markers and phenotypic values (electronic supplementary material, table S4). See the electronic supplementary material.

(e). Whole mount in situ hybridization

Prickled Cottus (C. perifretum and invasive Cottus) were bred to obtain eggs and developing larvae. Both eggs and larvae were fixed at several time points, and in situ hybridization experiments targeting the expression pattern of Edar were performed based on a protocol from Harris et al. [13]. See the electronic supplementary material.

(f). Population genetic analysis of the Cottus Edar gene

Ancestry-informative single nucleotide polymorphism (SNP) alleles for C. perifretum versus C. rhenanus (electronic supplementary material, table S2) were used to study the ancestry of Edar alleles in the invasive Cottus gene pool and the phenotypes associated with Edar alleles. Edar coding sequences of C. perifretum, C. rhenanus and other teleost fishes were used to infer allele origins (electronic supplementary material, table S2) and to test for signatures of long-term positive selection. See the electronic supplementary material.

3. Results

(a). Variation in Cottus morphology

Species-specific differences in prickling and body shape distinguish the parental species. Individuals representing pure ancestral species (C. rhenanus: Broel (n = 30), Naaf (n = 30); C. perifretum: LB (n = 40), WN (n = 32)) were studied to determine variations in body shape and the extent to which prickles cover the body. Prickling is significantly more pronounced in C. perifretum than in C. rhenanus (figure 1; Kruskal test p < 2.20 × 10⁻¹⁶). Significant differentiation in body shape can be observed in landmark-based morphometric analyses (figure 2a). Species separate along the first canonical variate axis (Wilk's lambda = 0.1371, p < 2.22 × 10⁻¹⁶), because C. rhenanus has a smaller, shorter head and a longer body than C. perifretum (figure 2b). Populations within species show weaker differentiation captured by CV axis 2 (figure 2b). Differences between species also manifest in traditional morphometric measurements (figure 2c). C. rhenanus individuals have a smaller head length/body length ratio than C. perifretum (t-test p-values: Broel × WN = 5.59 × 10⁻¹²; Naaf × LB = 1.39 × 10⁻¹⁰), which suggests C. perifretum have a longer head and a shorter body; C. rhenanus have a higher head length/eye width ratio than C. perifretum (t-test p-values: Broel × WN = 5.86 × 10⁻¹⁰; Naaf × LB = 4.36 × 10⁻¹¹). Phenotypes segregate in 775 F₂ individuals (electronic supplementary material, table S4), which enables QTL mapping.

(b). Mapping of candidate genes in Cottus

Cottus Eda signalling genes were identified and their map positions in Cottus were validated. Blast searches indicate that each of the isolated Cottus genes corresponded to a single Gasterosteus homologue of Eda, Edar, Edaradd, Nemo, Troy and Traf6. The genetic locations of Eda signalling genes (except for Traf6) were mapped to positions on Cottus linkage groups (LGs) that correspond to the locations of the stickleback homologues (electronic supplementary material, figure S1). While we did not find a useful genetic marker to directly map Cottus Traf6, this gene is presumably localized on Cottus LG5 based on transferring positional information from the stickleback genome [19].

(c). Quantitative trait loci analysis

The QTL analysis yielded a single QTL that affected prickling and body shape and contained the Edar gene in Cottus. All data for the QTL analysis, including individual based genotypic data and phenotypic traits are summarized in electronic supplementary material, table S4. The highly significant QTL for prickling was detected on LG3 for the consensus map (not shown) as well as in separated analyses of Broel × WN and Naaf × LB crosses (LOD = 9.38/21.48; table 1 and electronic supplementary material, figure S2). This corresponds to stickleback chromosome (Chr) XVI. Both types of crosses have the highest QTL peak around EDAR8930, which is located at the second intron of the Edar gene. The regression of prickling phenotype on genotypes at EDAR8930 shows that C. perifretum-derived alleles are associated with prickling in both Naaf × LB and Broel × WN crosses (electronic supplementary material, figure S3). Prickling of heterozygotes at EDAR8930 was intermediate and differed significantly from homozygous parental species genotypes (Kruskal test p-values: Broel × WN = 3.61 × 10⁻¹²; Naaf × LB < 2.2 × 10⁻¹⁶).

The Edar containing the QTL interval for Broel × WN crosses has a length of 13.76 cM and corresponds to stickleback Chr XVI from 12.127596 to 14.323519 Mb. The corresponding QTL interval for Naaf × LB crosses is much wider with an additional peak (ctg16139, LOD = 21.96) at the distal end (length of 54.15 cM; stickleback Chr XVI from 12.127596 to 17.802566 Mb). For ctg16139, prickling of heterozygotes was also intermediate and differs significantly from homozygous parental genotypes (Kruskal test p-value: Naaf × LB < 2.2 × 10⁻¹⁶). Besides Edar, these regions contain 85 and 217 genes, respectively. Differences in map intervals between the two Cottus maps may be due to different marker orders, which manifest when separate maps are reconstructed for different families in the Naaf × LB crosses. It is not possible to resolve this, however, because the number of individuals per family is too small to gain high confidence about results for relatively closely linked markers (not shown). The map reconstructed from combined Broel × WN crosses is free of conflicting signals. According to our analysis, none of the other Eda signalling genes or any other genomic region had a significant effect on prickling.

Several QTL contribute to differences in body shape (table 1 and electronic supplementary material, figure S4). Most relevantly, QTL on LG3 affected both prickling and shape in Naaf × LB crosses (table 1). The regression of body shape phenotypes on significant QTL marker genotypes shows association in Naaf × LB crosses and in Broel × WN crosses (electronic supplementary material, figure S5). This included QTL with effects on CV axis 1 scores as well as the head length/eye width ratio; markers included EDAR8930 (LOD = 8.09/4.53) and ctg16139 (LOD = 17.44/10.17). No significant QTL for CV axis 1 scores was detected for Broel × WN crosses. A QTL for head length/eye width ratio was detected on LG16 (ctg04045, LOD = 3.75) in Broel × WN crosses. The head length/body length ratio was affected by a QTL on LG3 (ctg01920, LOD = 4.80) in Broel × WN crosses, and by a different QTL on LG10 (ctg01402, LOD = 3.47) in Naaf × LB crosses. Hence, the results were only partially consistent between mapping families. None of the other Eda signalling genes shows a significant association with shape (electronic supplementary material, figure S1).

(d). Edar expression during Cottus development

The expression of Edar coincided with the development of relevant morphological features. Edar expression was first observed in invasive Cottus during the development of skeletal elements after hatching (0–15 days post hatching (dph) at 11°C). Tissues with Edar expression included gill rakers, teeth, branchial bones and fins rays (electronic supplementary material, figure S6). Edar expression could not be detected on the lateral body surface before 35 dph at 11°C. At 35 dph, clusters of cells express Edar. First prickles become visible as pointed emergences of the epidermis around 39 dph (figure 3a) and could be observed without staining under a stereomicroscope. Emerging prickles are marked by Edar expression at the distal tips in 39–51 dph fish (figure 3a). Prickles develop from the anterior to the posterior along the body. Fully developed prickles were first observed anteriorly. Concurrently, posterior regions showed clusters of cells expressing Edar. The number of prickles that develop initially (figure 3a) does not correspond to the final number observed in adult fish (figure 1a). Development of new prickles occurs at different stages, even in regions that already carry mature prickles. Edar expression was also observed in the developing head canal system (8–12 dph; electronic supplementary material, figure S6 g and S6 h) and around the pores of the lateral line system (figure 3b). Expression in the lateral line system coincided with the development of prickles. Pertaining to this, older fish stained with alizarin red revealed ossified structures surrounding the lateral line pores and encapsulating the underlying lateral line canal (figure 1a). Patterns of Edar expression as described here for invasive Cottus were also observed in C. perifretum from WN (not shown).

Figure 3.

In situ hybridization of Edar transcripts during the development 39–51 days post hatching (dph) of prickled invasive Cottus. (a) Edar expression can be detected at 39 (top), 45 (middle), 51 (bottom) days post hatching (dph). In general, prickles develop from anterior to posterior along the fish body. (b) Enlarged images of Edar expression on the side of the fish body (51 dph) reveal expression in patches of epithelial cell clusters (blue triangles) and in the tips of maturing prickles (green triangles). At the same time, Edar expression could also be observed at the pores of the developing lateral line (black triangles).

(e). Population genetic analysis

The origins of Edar alleles that differ between C. rhenanus and C. perifretum can be traced among distantly and closely related fishes. We assembled the Edar gene sequence from genomic DNA reads and validated the Edar exon boundaries using the transcriptome assembly of C. rhenanus. The 11 exons of Cottus Edar gene encode a 519 amino acid protein that shows extensive sequence identity to Edar sequences from other fish species and mammals (electronic supplementary material, figure S7). Five SNPs in the coding region (exons 3, 4, 6 and 9) distinguish samples representing C. rhenanus (non-prickled) and C. perifretum (prickled; table 2), all of which are non-synonymous except the one from exon 9. C. rhenanus and C. perifretum each carry two derived non-synonymous alleles respectively that are not shared between them (table 2). Only C. perifretum carries a derived neutral substitution. This leads to a striking departure from an otherwise deeply conserved amino acid sequence for both species. A comparison with distantly related teleosts revealed that the most widely encountered Edar amino acid sequence is conserved among different clades of Cottus that also vary in prickling (electronic supplementary material, table S2 and figure S7). Exceptions to this were found in C. perifretum from Zwaanebeek and in C. rhenanus from Duenn, all of which are heterozygous for adjacent alleles that are characteristic of both species. Moreover, a specimen of C. gobio (Salzbach) carries an Edar haplotype otherwise typical for C. rhenanus.

Table 2.

Diagnostic SNPs from the coding region of Edar in parental species and invasive Cottus. The first rows contain amino acids and codons typically observed in C. perifretum, C. rhenanus and other Teleostei. The ancestral alleles inferred through outgroup comparison are marked with asterisk. The following rows indicate frequencies of these alleles in invasive Cottus.

	prickling	Exon3	Exon4-a	Exon4-b	Exon6	Exon9
ancestral state Teleostei	n.a.	L	N	T	A	G
C. perifretum	prickled	TTC (F)	AAC (N)*	ACC (T)*	ACC (T)	GGT (G)*
C. rhenanus	non-prickled	CTC (L)*	AGC (S)	AAC (N)	GCC (A)*	GGG (G)*
invasive Cottus	C. perifretum homozygote	52	51	51	45	27
	heterozygote	13	13	13	18	28
	C. rhenanus homozygote	1	1	1	3	10
	C. perifretum ancestry	0.886	0.885	0.885	0.818	0.631

(f). Accelerated evolution of Cottus Edar sequences

Cottus Edar coding sequences have been subject to accelerated evolutionary change. Four of the Edar SNPs (excluding one from exon 9) lead to amino acid changes (table 2 and electronic supplementary material, figure S7). Three SNPs from Edar exons 3 and 4 are located in the ‘tumour necrosis factor receptor domain’ (TNFR domain) [11], and two SNPs from exon 4 are in the cysteine-rich repeat 3 [24]. A comparison with the UniProt database revealed no descriptions of mutant phenotypes affected by these amino acid changes in humans. Together, the species-specific SNPs constitute a conspicuous excess of coding changes between the two species of Cottus according to the best-fit substitution model. Model selection of the best branch model with four d_N/d_S ratios was done according to a Bayesian information criterion (BIC of best model: 14 442.6 versus model with 1 d_N/d_S: 14 457.6). The d_N/d_S between Edar copies from more distantly related fish species and other vertebrates vary around 0.059, with the exception of tilapia, which has evolved faster (d_N/d_S of 0.184). In contrast, the two Cottus species display a maximum-likelihood estimate of d_N/d_S far above one (reaching the maximum allowed value by the optimization algorithm). The K_a/K_s value between C. rhenanus and C. perifretum is 1.176. These high ratios are due to the presence of four independent non-synonymous changes (two on each branch) and only a single synonymous substitution, most likely on the C. perifretum branch.

(g). Ancestry of Edar alleles in invasive Cottus

Allele frequencies of the five Edar SNPs in the invasive Cottus gene pool (hybrid lineage) showed high frequencies of alleles (81.8–88.6%) derived from C. perifretum (prickled; table 2). Notably, one invasive Cottus individual (prickled) carrying two alleles of the C. rhenanus coding sequence was identified. There are no signs of recombination between ancestral alleles of the first three SNPs in invasive Cottus (electronic supplementary material, table S2). Such linkage of species-specific alleles is not evident for the SNPs on exon 6 and 9, and the single synonymous substitution in exon 9 has a markedly lower prevalence of the C. perifretum allele (63.1%) in the invasive Cottus gene pool.

4. Discussion

Our analysis supports the hypothesis that alleles of the Edar gene contribute to skin prickling and other phenotypes. Moreover, we detected unique signatures of coding sequence evolution in Cottus that are, however, most likely not directly involved in prickling differentiation between the two species studied here. Nonetheless, the results suggest that a pathway that underlies adaptive evolution in stickleback [16] is associated with intriguing evolutionary patterns in Cottus. Sculpins may thus represent an additional group of fishes in which Eda signalling is involved in rapid microevolutionary processes, despite the deep conservation of this pathway among vertebrates [11,12]. A single QTL region that is associated with skin prickling contains a considerable number of genes. The LOD scores peak around Edar, the most likely candidate to explain the prickling variation between the two Cottus species. Our analysis does not support a role of other Eda signalling genes although we specifically evaluated them using gene-specific markers. This study is limited in that functional evidence through genetic manipulations is not available. In the absence of such studies, we discuss the possible involvement of the Edar gene in generating phenotypic diversity in Cottus.

(a). Development of skin prickling

Skin prickling in Cottus and the lateral armour plates in sticklebacks are most likely homologous with scales, because both are composed of dermal bone material. Both structures share an anterior-to-posterior mode of development and the evolution of reduced patterns that are limited to the midline of the anterior part of the body [4,5,25]. This may reflect the sister group relationships between Scorpaeniformes and Gasterosteiformes [18]. In other bony fishes, a single scale appears as an initiator at the lateral line and additional scales are added successively in regular rows from posterior-to-anterior regions of the body [26]. Cottus may differ from other fishes in that additional prickles are added in between existing prickles as the fish grow, but it is unclear at what age the development of new prickles ceases. Eda signalling was shown to critically affect the development of scales in a wide range of fishes, including sticklebacks [13–15]. Colosimo et al. [15] have shown that the Eda gene is the key locus that affects lateral plate reduction in sticklebacks. In contrast to sticklebacks, Cottus phenotypes are primarily associated with the receptor Edar. While Edar plays only a minor role in stickleback [10], it is known to be involved in the development of scales in other fishes [13,14]. A possible role of Edar in Cottus prickle development was supported through in situ hybridization. It is likely that Edar is involved in the onset of the prickling development, as we observed clusters of cells that express Edar and differentiate into prickles. Although our analysis does not provide hints of other prickling QTL, it is likely, that the number of samples and markers used in this study were too low to detect additional genetic factors that contribute to the variation in prickling.

(b). Possible pleiotropic effects of Edar

Pleiotropic effects on body shape and scale derivatives are shared by stickleback and Cottus. We confirmed that Edar is expressed in Cottus during the development of bones and it is known that Eda signalling influences skull shape in zebrafish mutants [13]. Admittedly, there are other QTL affecting shape besides the region containing Edar and most likely many genes affecting shape [21,22]. However, given the focus of this study, loci (or a locus) with pleiotropic effects on shape and prickling are most relevant. Interestingly, QTL intervals that affect different skeletal traits including head shape were linked to Eda in sticklebacks [21,22]. These effects are shared by Cottus but are linked to a different QTL (carrying Eda in stickleback and Edar in Cottus). This underlines the possible importance of the Eda signalling pathway for skull shape.

Cottus Edar might also influence the development of the lateral line system that varies in Cottus species [5]. In line with this, we observed expression of Edar in the developing sensory canal system during the formation of skull elements. At a later stage, while prickles developed, Edar was expressed in confined areas around the lateral line pores. This expression was followed by the development of ossified structures that encapsulate parts of the lateral line canal and pores in fully developed Cottus. Patterns of neuromasts in the lateral line genetically mapped to the Eda region in stickleback and Eda transgenic sticklebacks have alterations to both ectopic plates and neuromast distribution [27]. Interactions between bony dermal structures and neuromast patterning during development have also been observed in zebrafish and medaka [28]. Although the latter studies reported Eda, it is plausible that Edar could play a role as both genes are part of the same pathway.

(c). Association of Edar alleles with Cottus species

SNPs in the Edar coding sequence that distinguish C. rhenanus and C. perifretum occur in other Cottus species, and are combined into different haplotypes. Hence, the evolution of C. rhenanus and C. perifretum alleles can be explained through recombination between ancestral haplotypes. It is likely that alleles are shared to some extent between the two study species owing to ancestral polymorphism and because gene flow occurs [4,23,29]. Likewise, gene flow between C. gobio and C. rhenanus occurred within the river Rhine basin [4] which might explain the presence of alleles typical of C. rhenanus in one specimen of C. gobio from Salzbach (tributary to the upper Rhine). Our data on invasive Cottus suggests that recombination among the SNPs in exons 3 and 4 is constrained, whereas alleles in exons 6 and 9 occur in varying combinations. The same may apply for individuals of C. perifretum from Zwaanebeek that are heterozygotes for allele states in exons 3 and 4 that are otherwise typical for C. rhenanus.

(d). Functions of Cottus Edar alleles

While it is possible that the coding changes contribute to the phenotypic difference between two Cottus species, our results suggest that additional genetic elements in the Edar gene play a role. We observed C. perifretum prickling and shape phenotypes in invasive Cottus carrying C. rhenanus Edar alleles. In heterozygous individuals, this could be explained through dominance of the C. perifretum allele. Yet, heterozygous genotypes in our F₂ crosses display intermediate degrees of prickling, which does not suggest complete dominance. Moreover, the prickling in one invasive Cottus with a homozygous C. rhenanus coding sequence can only be reconciled with our results if additional genetic elements that are linked with Edar play a decisive role in determining the phenotypes studied here. If true, such elements must have originated from C. perifretum and recombined with the coding sequence from C. rhenanus. Such genetic elements could be of a regulatory nature, as it was found for the Eda gene that causes polymorphism in sticklebacks [15,30]. Edar coding sequences typical of C. perifretum and C. rhenanus are not shared by prickled or non-prickled Cottus in the Cottopsis clade, the Uranidea clade and the Cottus clade [3], which implies that they are not involved in the recurrent evolution of prickling phenotypes. Future studies will have to validate whether prickling phenotypes in other Cottids are due to effects of Edar, the Eda signalling pathway or entirely different genes.

Low Edar d_N/d_S values in most fishes are consistent with the general pattern of deep conservation of function of this gene. Strikingly, C. perifretum and C. rhenanus each harbour two different and derived amino acid changes. These non-synonymous changes greatly exceed observed non-synonymous changes among more distantly related fish species. Likewise, the between species K_a/K_s value is larger than 1. Coding changes of this magnitude imply adaptive evolution. The positions of the coding sequence changes in Cottus Edar alleles provide some hints of possible functions. Three of the four non-synonymous changes are situated in the extracellular TNFR domain, which interacts with the Eda ligand [11,24]. While there are no known mutations that correspond to amino acid changes between C. rhenanus and C. perifretum, these changes are candidates that may affect the Eda–Edar interaction. It is still possible that they contribute to the variation in pricking and of other phenotypic traits not studied here. A possible alternative explanation for the accelerated evolution of extracellular domains is related to interactions with endogenous retroviruses. Retroviruses have been found to drive the accumulation of non-synonymous changes in extracellular domains of the transmembrane receptor XPR1 [31].

(e). Habitat adaptation through Eda signalling

Comparisons with the well-studied stickleback can suggest functions of phenotypes associated with Cottus Edar alleles. Eda signalling is thought to play a key role in the ecological adaption of sticklebacks, because alleles are tightly coupled with ecological conditions [9,16]. Likewise, prickling phenotypes are often coupled with ecological conditions: prickling varies across ecotones of Cottus hybrid zones between invasive Cottus and C. rhenanus in Germany [32]; between species that use different ecological niches such as C. asper and C. aleuticus in Alaska [6]; and between lineages of C. asper in different habitats [7]. Unfortunately, the precise adaptive functions of prickling and shape are not evident, and we are not aware of a general pattern of association of the phenotype with any particular ecological factor. Such inference is complicated owing to the possible pleiotropy of Eda signalling and the QTL carrying Edar. Although not genetically mapped here, Cottus also vary greatly in development of the lateral line system [5], which represents another potential trait affected by Edar. Another possibly related pattern of differentiation in scale phenotypes occurs in marine sculpins from the North American pacific coast. Phylogenetically independent changes in the number of scales and body size suggest that selection related to oxygen usage operated on these traits between intertidal and subtidal habitats [8]. This adds to the evolutionary ecological complexity that has to be considered in evolutionary studies on Eda signalling.

(f). Hybrid intermediacy in an admixed lineage

We have previously studied the genetic make-up of hybrid ‘invasive Cottus' within the River Rhine system [4]. The high similarity of invasive Cottus to one parent species, C. perifretum, is intriguing given that the admixed genome contains approximately 40% genetic material from C. rhenanus [23]. Among other traits, this is true for skin prickling, body shape and the sensory head canal that is frequently interrupted at the chin [4,5]. Our results imply that a single genetic region spanning the Edar gene could cause this pattern. In agreement, we find a prevalence of C. perifretum Edar alleles in the invasive Cottus gene pool. While it is tempting to suspect that the C. perifretum Edar allele has been subject to selection in invasive Cottus, there is a chance that ancestry excess at a single locus may have evolved by drift alone. Much of the variance in ancestry among different loci in the invasive gene pool can be explained by drift [23]. The fixation of adaptive alleles could also be slowed down if they are dominant, thus hiding effects of recessive alleles in heterozygous states. These considerations demonstrate that tests for genotypic selection alone may not suffice to test whether Edar alleles contribute to rapid adaptation. Further studies that address the functional relevance of traits affected by Eda signalling and the regulation of Edar in Cottus are needed. If a genetic architecture as found here for Cottus phenotypes is common in hybrid lineages, evolutionary change may quickly blur hybrid intermediacy and challenge recognition of hybrids and hybrid lineages.

Ethics

K. Lessenich (Berirksregierung Köln) has given permits to conduct the fieldwork. Cottus were bred with permission from S. Hauschildt (Veterinaeramt in Plön).

Data accessibility

The datasets supporting this article have been uploaded as part of the electronic supplementary material.

Authors' contributions

J.C. collected morphology and genotype data and performed all genetic analyses and in situ hybridization. A.N. analysed body shape. J.A. performed Illumina sequencing. F.S. assembled sequences. J.C. and A.N. conceived the study and wrote the manuscript. All authors have read and approved the final manuscript.

Competing interests

The authors declare no conflict of interest.

Funding

F.S. is supported through National Science Foundation awards (DBI-1350041 and IOS-1237880). The project was supported by a DFG grant (no. 762/2-1) and an ERC grant (Evolmapping) to A.N.