Introduction

Body colouration strongly dictates how species interact within their respective environments because body colouration is a trait under strong, selective pressures. In environments where the visual background is influenced by a combination of factors, social signalling, effectiveness of camouflage, reproductive success, and fitness for a given species can vary substantially because of its body colouration (Gray and McKinnon 2007; Kekäläinen et al. 2010; Lehtonen et al. 2023). Factors contributing to the complexity of the visual background in marine environments include turbidity, depth, type of substrate, presence of prey, predators, and other species such as corals, seaweeds, and seagrass beds, which serve as hideouts (Marshall et al. 2019; Salis et al. 2019; dos Anjos et al. 2022). Together, these elements create a dynamic backdrop against which a species must adapt. However, one clear aspect of this interplay of environmental variables is the consistent light attenuation along the depth gradient, which serves as a stable environmental axis that can exert substantial influence on the diversification of body colouration of species.

The genus Sebastes Cuvier 1829 is a diverse group encompassing over 110 species that are colloquially called rockfishes (Hyde and Vetter 2007; Ingram and Kai 2014). These species can be found inhabiting different marine environments, from shallow coastal waters to profound depths of over 1000 m. Many Sebastes species, roughly 70 of them, are found in the Northeast Pacific Ocean, whereas the northwestern Pacific Ocean is home to approximately 30 different species (Hyde and Vetter 2007; Ingram and Kai 2014). Within the genus, recently diverged sister species show wide overlap in latitudinal range, but less overlap in depth distribution (Ingram 2011). Extensive research and phylogenetic comparative analyses indicate that the depth at which rockfishes reside is a primary driver of diversification, leading to the development of significant differences in body shape, eye size, otolith shape and size, acoustic and visual sensitivity, and lifespan (Stefánsson et al. 2009; Ingram and Shurin 2009; Sivasundar and Palumbi 2010; Ingram 2011, 2015; Shum et al. 2014; Ingram and Kai 2014; Tuset et al. 2015, 2016; Heras and Aguilar 2019; Kolora et al. 2021; Treaster et al. 2023). The combination of certain values in these biological traits suggests the occurrence of correlational selection, in which multiple traits are simultaneously influenced by a single ecological constraint (Lande and Arnold 1983). In several groups, this type of selection is usually accompanied by colouration differences (McKinnon and Pierotti 2010), offering valuable insights into the process of speciation in parapatry and sympatry (Gray and McKinnon 2007; Forsman et al. 2008; McLean and Stuart-Fox 2014).

Rockfishes undergoing speciation can differ in body colourations in concordance with asymmetric depth distributions within their habitats (Hyde et al. 2008; Venerus et al. 2013; Artamonova et al. 2013; Zhuang et al. 2015, 2020; Deville et al. 2023a), and adaptive changes in visual perception (Sivasundar and Palumbi 2010; Shum et al. 2014; Deville et al. 2023b). Colour differences between fishes can be attributed to differences in the synthesis pathways of chromatophores and pigments, and the metabolism of acquired pigments through diet (e.g. Yamanome et al. 2007; Kumbar and Ganesh 2021; Fang et al. 2022). Both processes can be physiologically and morphologically regulated in response to the amount of light present in the visual surroundings of a species (Gamble et al. 2003; Pandolfi et al. 2003; Lehtonen et al. 2023). As a result, these environmentally influenced regulations in colouration may substantially contribute to the colour diversification of rockfishes, which dwell in various visual environments and experience rapid diversification.

The stunning diversity of colours across rockfishes varies from striking and vibrant hues of red, orange, and yellow to the more understated and cryptic shades of black, brown, and grey (Orr et al. 2000). Some species exhibit consistent colourations throughout their bodies, while others have unique colouration patterns characterised by the dominance of bright colours with irregular dark patches or spots, or vice versa, species with dominance of dark colours accompanied with bright patches or spots (Figs. 1 and 2). Ecomorphological comparisons of closely related species and ecotypes indicate that brightly coloured rockfishes tend to migrate into deeper areas during mating periods and consequently have slower growth rates than dark-coloured species (Venerus et al. 2013; Zhuang et al. 2015, 2020; Deville et al. 2023a). The depth segregation of rockfishes based on colourations may be possibly dictated by the need for photoprotective pigmentation in shallower waters. This is because the pigments responsible for bright colourations do not absorb UV irradiation as those producing dark colourations (Armstrong et al. 2000; Mizusawa et al. 2018). However, the depth segregation of bright- and dark-coloured rockfishes has not yet been comprehensively tested using quantitative comparative analyses and information from most species of the genus. The absence of these analyses may be attributed to the lack of appropriate image-analysis techniques and the limited availability of high-quality photographs. Recently developed methods that consider both the amount and similarity of colours (e.g. Van Belleghem et al. 2018; Weller and Westneat 2019) now allow us to quantify and compare the colouration of these species from digital photographs available in public repositories. In this study, we employed phylogenetic comparative analyses to evaluate the colour variation in Sebastes rockfishes considering (1) depth distribution of species, (2) life-history traits (growth rate), and biological characteristics that are also influenced by depth (e.g. eye width, body depth, etc., Ingram 2011; Ingram and Kai 2014), and (3) closely related species, particularly species pairs and triads categorised as species complexes (e.g. Roques et al. 1999; Kai and Nakabo 2004, 2008, 2013; Narum et al. 2004; Hyde and Vetter 2007; Hyde et al. 2008; Wilson et al. 2008; Muto et al. 2019; Nishimoto et al. 2019). We discuss how colourations, life-history traits, and morphological features of Sebastes rockfishes have been influenced by the respective habitats in which they live.

Fig. 1
figure 1

Colour categories in three rockfishes (S. aleutianus, S. babcocki, and S. nebulosus). Eight colour categories for these rockfishes were extracted from bodies excluding eyes, and dorsal, anal, and pelvic fins (shown within blue). Colour categories are represented in Hex codes and percentages of pixels

Full size image
Fig. 2
figure 2

Mean four colour categories for 99 Sebastes rockfishes. Stacked barplots display colour categories in descending order from branches to tips of the phylogenetic tree. The three dotted circles around the barplots indicate the 30, 60, and 90% cumulative presence of the mean four colour categories. Photos of 16 species were included as references. Branches were coloured based on the geographical distribution of species found in FishBase. The phylogenetic tree with stacked barplots was plotted using the phytools package (Revell 2024) 

Full size image

Materials and methods

Data collection

A total of 377 photos from 100 of the 111 described Sebastes species were retrieved from iNaturalist, zukan, anglers, Flickr, Mexican-fish, WDFW (Washington Department of Fish and Wildlife), FishPix, and other online repositories (Supplementary Information Table S1). In the first-four repositories, the species identity of the photos is assigned and discussed by several contributors. In contrast, in Mexican-fish and WDFW, species identity is independently assigned by the repositories themselves, often uploading photos of reference individuals for each fish species. In the case of FishPix, species identity of photos is supported by the Kanagawa Prefectural Museum of Natural History and the National Museum of Nature and Science of Japan. For each species, photos with individuals displaying the colouration described in FishBase (Froese and Pauly 2023) and taxonomic descriptions were selected (Supplementary Information Table S2), leaving out descriptions with uncommon colourations (e.g. Muto et al. 2013a, 2016; Myoung et al. 2022). Photos of fresh specimens in lateral view and with clear illumination were chosen to reduce bias when detecting colours. At least three photos were compiled for 91 species, two for seven species, and one for two species. The average number of photos per species was 3.7. The dorsal, anal, and pectoral fins, and eyes were manually excluded from photos (Fig. 1, within blue). These fins were excluded because they were not extended in many photos, or they were partially covered by the pectoral fin or objects. Eyes were excluded because some individuals exhibited exophthalmia, causing abnormal whitish colourations that were not observed in other fishes. In species with fin colours significantly differing from body colours, the colours from the excluded fins are accounted for the pectoral and caudal fins. The backgrounds of the photos were removed, and photos were saved in PNG format for colour detection.

Data on the depth distribution ranges and growth rates of 97 and 77 species, respectively, were retrieved from FishBase, individual publications, and online repositories (Supplementary Information Table S3). The mean growth rate was considered for species with multiple entries in FishBase, whereas the combined growth rates of males and females were selected from individual publications. In seven species without growth-rate information, these rates were estimated from the maximum age using the formula of Froese (2022). For these estimations, the maximum age was attained and reported to be approximately 99% of the maximum body size and not 95%, as suggested by the author. A maximum body size of 99% was considered because, at 95%, the growth rates were underestimated by a factor of two in species with reported growth rates. In addition, data on the body depth, eye and pectoral fin widths, and lower jaw and gill raker lengths were retrieved for 91 species. These variables were selected because their variations are correlated with the depth distribution of species (Ingram 2011; Ingram and Kai 2014). Information on these morphological traits was compiled from Ingram and Kai (2014) and log-transformed for further analyses.

For phylogenetic reconstruction, we retrieved sequences of nine genetic regions (12S ribosomal RNA [12S], 16S ribosomal RNA [16S], cytochrome oxidase b [CytB], cytochrome oxidase c subunit I [COXI], mitochondrial control region [D-loop], proline transfer RNA [tRNA Pro], threonine transfer RNA [tRNA Thr], internal transcribed spacer 1 [ITS1], and the recombination activating 2 [RAG2]) from 108 species and four outgroups: Helicolenus avius, Hozukius emblemarius, Sebasticus marmoratus, and Sebastolobus alascanus. The genetic regions and outgroups were chosen from the following Hyde and Vetter (2007) and Ingram and Kai (2014). We included new sequences for S. cheni and S. ventricosus (16S, CytB, COXI, D-loop, tRNA Pro, tRNA Thr, and RAG2) (Supplementary Information Table S3). For species with multiple available sequences for one genetic region, the longest sequence was selected. Only S. diaconus and S. varispinis were left out because of the lack of DNA sequences.

Colour detection

Analyses were performed using R v.4.1.2 (R Core Team 2021). The getHistList function of the colordistance package (Weller and Westneat 2019) was used to decompose 10,000 pixels in each photo to values of red, green, and blue (RGB values) and binarize them into eight colour categories. The default binning method (histogram binning) was used because the colour categories have the same extensions for each photo, making comparisons of categories across photos simple. This method can also retain the details of photos, such as small colours, instead of collapsing them into a larger colour category (Weller and Westneat 2019). The output for each photo was a data frame containing centres of the eight colour categories in the RGB space, and the proportion of pixels contained within them. Colour categories of each photo were transformed into colour Hex codes for their use in plots.

Phylogenetic reconstruction

Each genetic region was separately aligned and then alignments were processed with Gblocks 0.9.1 (Castresana 2000) to discard poorly aligned positions while allowing the inclusion of gaps to increase the number of phylogenetically informative variable sites. The processed alignments were concatenated to obtain a final alignment of 5119 bp. Then, time-calibrated phylogenetic reconstruction was performed using BEAST v.2.6 (Bouckaert et al. 2019) following the instructions of Ingram and Kai (2014). For this reconstruction, the best substitution model from BEAST model test (Bouckaert and Drummond 2017) under all reversible model search, the Yule process to build the prior tree, and a fast relaxed Lognormal clock (Zhang and Drummond 2020), were used. The analysis was run using a Markov Chain Monte Carlo simulation of 150,000,000 steps, storing 10,000 trees from the posterior distribution. The maximum clade credibility tree (MCC tree) was calculated considering a 10% of burn-in with the programme TreeAnnotator included in BEAST v.2.6.

Colour variation across species

The RGB values and percentages of the eight colour categories extracted from each photo were used to estimate a matrix of pairwise Earth mover’s distances (EMD) with the getColorDistanceMatrix function. EMD were selected because they account for both similarity and amount of the colour categories to produce distance measurements for any pair of images (Weller and Westneat 2019). A total of 29 species pairs composed of sister species were identified based on previous phylogenetic trees (Hyde and Vetter 2007; Ingram and Kai 2014; Ingram 2015) and our estimated MCC tree (Fig. 2 and Supplementary Information Fig. S1). In 17 of these species pairs, the next closely related species (hereafter third species) was added to form species triads containing three species with sympatric occurrence (Supplementary Information Fig. S1). The matrix of pairwise EMD was partitioned into intraspecific EMD (hereafter “within” set), EMD between sister species (“sister” set), EMD of each species within a pair with the third species of a triad (“triad” set), and the rest of EMD between species (“other” set). To evaluate the differences between these four sets of EMD, box plots and an ANOVA test were constructed and performed, respectively.

For 98 species with more than one photo, the RGB values and percentages of the resulting eight colour categories were averaged to obtain consensus colour categories. For this purpose, data frames within each species were arranged by placing colour categories in descending order of presence. In 20 species, the descending orders of percentages from visually similar colour categories were consistent amongst individuals, while in other 78 species, the colour categories were manually arranged before estimating the consensus colour categories. By arranging them, we avoided homogenization and creation of highly dissimilar colour categories within each species. The eight consensus colour categories from each species were transformed into RGB integers using the formula (red × 2562) + (green × 256) + blue (Love et al. 2006).

A second matrix of EMD was estimated using the consensus colour categories of the species. This matrix of EMD was reduced using multidimensional scaling (hereafter PCoA, since it is also known as Principal Coordinate Analysis) (cmdscale function) to represent variation of colour categories amongst species. PCoA was used because this method projects high-dimensional data into a few dimensions while preserving the distances between observations. Correlations between the three axes of major variation (PCoA1, PCoA2, and PCoA3) and the RGB integers and percentages of the consensus colour categories of all species were assessed.

Phylogenetic comparative analyses

A phylomorphospace was constructed by projecting the MCC tree into a space of morphological traits defined by the PCoA axes (Sidlauskas 2008). Phylogenetic signals of the PCoA axes were estimated by considering the Pagel’s λ estimator (Pagel 1999). λ is a scaling parameter for the correlations between species with respect to the expected correlation under Brownian evolution. Then, phylogenetic generalized least squares (PGLS) were fitted to assess whether the PCoA axes were correlated with square-rooted values of the maximum depth distribution and growth rate of species while considering their phylogenetic relatedness. In addition, PGLS models were used to evaluate whether the PCoA axes are correlated with log-transformed values of body depth, eye and pectoral fin widths, and lower jaw and gill raker lengths. The possible influence of the phylogenetic signal on each interaction of variables was assessed by comparing a model with the estimated λ to a model with no phylogenetic signal (λ = 0).

Groups of species pairs and triads were used to assess the relationship between EMD and dissimilarities in depth distribution and phylogenetic distances while excluding earlier speciation events. Dissimilarity in depth distribution was estimated as the ratio of the differences in the minimum and maximum limits of the depth distribution of two species to the combined depth distribution of both species. Phylogenetic distances between species were estimated using the cophenetic.phylo function (ape package, Paradis and Schliep 2019). As each species triad contained three species (Supplementary Information Fig. S1), EMD, dissimilarities in depth distribution, and phylogenetic distances from each pair of species were added to obtain unique values for each triad. The tips of the phylogenies used for these analyses represent the comparisons between species pairs and triads; thus, the tips of the MCC tree were pruned to represent one member of each species pair and the third species of each species triad, respectively (Supplementary Fig. S1). PGLS models were carried out using these variables and the pruned MCC trees, as detailed in previous analyses.

Results

Within species, similar colour categories were observed across photos, despite differences in illumination and quality (Supplementary Information Fig. S2). More than 93% of the pixels were represented by four colour categories in all photos (Fig. 1 and Supplementary Information Fig. S3). The colour categories with the highest presence (hereinafter dominant colour categories) contained more than 50% of the pixels in 239 photos (Supplementary Information Fig. S3). Significant differences were found between the four sets of EMD (P value < 0.0001) except when the “sister” and “triad” sets were compared (P value > 0.50). The “within” set had a lower mean (0.1833 ± 0.007) than the “sister” set (0.2765 ± 0.0122), the “triad” set (0.2882 ± 0.0106), and the “other” set (0.3228 ± 0.0009) (Fig. 3). This indicates that it is possible to distinguish intraspecific and interspecific differences within the genus, allowing the analysis of colour variation between closely related species.

Fig. 3
figure 3

Earth mover’s distances (EMD) within species, between sister species, between each species of a pair with the next closely related species (Third species), and between other pairs of species (Other distances)

Full size image

The dominant colour categories harboured more than 50% of the pixels in 65 species (Fig. 2). In the Northwest Pacific, 14 species of 29 exhibited dominant dark-colour categories, while the Northeast Pacific harbours 34 species (out of 57) with a dominance of bright colour categories (Fig. 2). The first three PCoA axes accounted for 51.34% of the variation. PCoA1 was negatively correlated with the RGB integers of the first and second most abundant colour categories (Table 1). Species with dominant colour categories related to black had positive values on the PCoA1 axis, while species with dominant colour categories in the ranges of red and orange had negative values on the same axis (Fig. 4). PCoA2 was positively and negatively correlated with the percentages of the first and second most abundant colour categories, respectively (Table 1). Species with large PCoA2 values exhibited high dominance in the first colour category (e.g. S. miniatus, S. phillipsi, S. ventricosus, and S. viviparus), whilst species with low PCoA2 values displayed two colour categories with similar percentages as observed in species with large, coloured patches (e.g. S. elongatus, S. owstoni, and S. serranoides) (Fig. 4 and Supplementary Information Fig S4). PCoA3 was not analysed or discussed, as it only accounted for 4.32% of the explained variation and its correlations with the RGB integers, and percentages of pixels were already accounted for PCoA1 and PCoA2 (Table 1).

Table 1 Correlations of PCoA1, PCoA2, and PCoA3 on RGB integers and percentages from the four most abundant colour categories in all species
Full size table
Fig. 4
figure 4

Phylomorphospace of 100 species. Sky-blue circle indicates the confidence ellipse at 90% level. The phylomorphospace was constructed from the time-calibrated maximum clade credibility tree. Species were coloured based on the colour category with the highest presence. Symbols indicate geographic distribution of species. NAT North Atlantic, NEP Northeast Pacific, NWP Northwest Pacific, and TP Trans-Pacific. The relationship between S. diaconus and S. mystinus was represented as a dotted red line because of the lack of DNA sequences from S. diaconus. Some species with large values in extremes of the PCoA1 and PCoA2 axes were labelled. The scatter plot of explained variation for the first ten PCoA axes is displayed

Full size image

The phylomorphospace showed convergences of dominant colour categories amongst distantly related species and some closely related species (Fig. 4). Only PCoA1 had a significant phylogenetic signal (λ = 0.6698, P value < 0.0001) under Brownian motion expectations. In contrast, the phylogenetic signal of PCoA2 indicated that variation within this axis is not significantly determined by the intrinsic phylogenetic relatedness of species (λ = 0.2007, P value > 0.15). Both the maximum depth distribution and growth rate were significant predictors of PCoA1 (maximum depth distribution: P value < 0.002, growth rate: P value < 0.023) in contrast to PCoA2, which was only correlated with growth rate (P value < 0.004) (Table 2). Scatter plots of PCoA1 on the maximum depth distribution and growth rate showed that species with dominance of bright colour categories are found in deeper waters and have lower growth rates than species with dominance of dark-colour categories, irrespective of their geographical distribution (Fig. 5). The scatter plots of PCoA2 on growth rate showed a decreasing trend between these variables (Fig. 5). PGLS models using information of species pairs only indicated a positive correlation between EMD and depth range dissimilarity (slope = 0.2023, P value < 0.05) (Fig. 6). PGLS models including species triads did not show any significant correlation between EMD and phylogenetic distance or depth range dissimilarity.

Table 2 Parameters of the phylogenetic generalized lest squares (PGLS) analyses modelling each of the main axes of colour variation (PCoA1 and PCoA2) on maximum depth distribution, growth rate, and five body measurements
Full size table
Fig. 5
figure 5

Phylogenetic generalized least squares (PGLS) models of PCoA1 and PCoA2 on square-rooted maximum depth distribution and growth rate. Species were coloured using their respective dominant colour categories. Symbols indicate geographical distribution of species. NAT North Atlantic, NEP Northeast Pacific, NWP Northwest Pacific, and TP Trans-Pacific. The only nonsignificant correlation displayed here is between PCoA2 and square-rooted maximum depth distribution. The regression lines and confidence intervals estimated using the PGLS models are represented by dotted lines and sky-blue areas, respectively

Full size image
Fig. 6
figure 6

Phylogenetic generalized least squares (PGLS) model of the Earth mover’s distance (EMD) on depth distribution dissimilarity when comparing species pairs. The regression line and confidence interval estimated using the PGLS model are represented by a dotted line and a sky-blue area, respectively

Full size image

Both PCoA1 and PCoA2 were not correlated with body depth. PCoA1 was significantly correlated with both eye and pectoral fin widths (P values < 0.001), while PCoA2 was significantly correlated with lower jaw length (P value < 0.006) and eye width (P value < 0.02) (Table 2, Fig. 7). Gill raker length was significantly correlated with PCoA1 (P value < 0.001), but the correlation was lost after phylogenetic correction. The scatter plots of both PCoA1 and PCoA2 on the morphological traits indicated that species with dominance of bright colour categories had wider eyes and narrower pectoral fins than species with dominance of dark-colour categories (Fig. 7). Meanwhile species with uniform body colourations are endowed with wider eyes and larger lower jaws than species with large, coloured patches.

Fig. 7
figure 7

Phylogenetic generalized least squares (PGLS) models of PCoA1 and PCoA2 on morphological traits. Morphological traits were log-transformed for these analyses. Species were coloured based on the colour with the highest presence. Only models with significant correlations are displayed. The dotted lines and the sky-blue areas represent the regression lines and confidence intervals estimated using the PGLS models, respectively

Full size image

Discussion

Our results confirmed that rockfishes with dominance of bright colour categories such as red, orange, and yellow, tend to inhabit deeper waters and have slower growth rates than their congeners with dominance of black, grey, and brown colour categories. Dark-coloured rockfishes are not found at depths deeper than 500 m, where only rockfishes with dominance of red categories are found (Fig. 5). The segregation of bright and dark-coloured rockfishes along depth hints at a photoprotective strategy to mitigate the potential down-regulation of immune responses induced by the UV irradiation in the shallow layers of water (Lawrence et al. 2020). This strategy is concordant with the contrasting photoprotective attributes of the chromatophores in fishes and their regulation under different levels of UV exposition (Mueller and Neuhauss 2014; Mizusawa et al. 2018). Melanophores, which produce dark colourations such as black, grey, and brown, can absorb UV radiations because of the melanin pigment, mitigating potential negative effects on rockfishes with dominance of dark-colour categories. In contrast, erythrophores and xanthophores, which contain carotenoids and coloured pteridines responsible for the bright colourations, do not exhibit the same photoprotective properties as melanophores. Carotenoids only have absorbances in the visible regions of the light spectra (400–520 nm) (blue–green) (Armstrong et al. 2000), while the photoprotective properties of coloured pteridines have not been conclusively proven yet. Hence, rockfishes with dominance of bright colours inhabit deeper waters to likely lessen their exposition to UV radiation in shallow waters. Besides melanophores, leucophores harbour colourless pteridine compounds with maximum absorbances at 320–330 nm (UV range) (Armstrong et al. 2000). These chromatophores are additional photoprotective components in species such as S. babcocki, S. cheni, S. owstoni, S. rubrivinctus, S. schlegelii, S. vulpes, and S. zonatus, in which pale-colour categories were found at percentages between 30 to 50% (Fig. 2).

Despite their variable depth distributions as adults, rockfishes often recruit in shallow environments with low-relief rocks, macrophytes, and seagrass beds (Love et al. 1991). However, recruitment in shallow environments comes with a trade-off for bright-coloured rockfishes because the long wavelengths of their body colours (red, orange, etc.) are still reflected (Marshall et al. 2019), making them more visible and susceptible to predation, while reducing their foraging success (Zhuang et al. 2020; Deville et al. 2023a). Conversely, in deeper environments, species with dominance of red- and orange-colour categories appear visually black due to the increasing light attenuation at deeper waters (Marshall et al. 2019), which does not require individuals to perform escape responses as often as in shallow waters. Since they spend more time in deeper environments with lower temperatures and oxygen, their metabolic and growth rates decrease (Cailliet et al. 2001; Black et al. 2021). The ecological consequences of rockfishes’ colourations align with expectations under environment-contingent natural selection, where colour polymorphism experiences varying degrees of fitness in response to different light environments (Gray and McKinnon 2007; Seehausen et al. 2008).

Brightly coloured rockfishes are endowed with morphological adaptations such as wider eyes, and narrower pectoral fins, which facilitate capturing more photons for the detection of prey and/ or predators in dim-light environments, and perform faster predatory and escape movements, respectively (Fulton et al. 2005; de Busserolles et al. 2020). Larger lower jaws come into play in species with high dominance of the first colour category (e.g. S. miniatus, S. phillipsi, etc.) (Tables 1 and 2, Figs. 2 and 7), allowing them to produce stronger bite forces than species with coloured irregular patches (Westneat 2003; de Busserolles et al. 2020). Meanwhile, the presence of irregular dark patches or spots in brightly coloured rockfishes (e.g. S. chlorostictus, S. elongatus, S. eos, S. rosenblatti, and S. umbrosus) (Fig. 2) can help them to obscure their shape, camouflaging them in the background and hampering the detection of their body contours in shallower environments (Cuthill et al. 2005; Stevens and Merilaita 2008). Other morphological adaptations in bright-coloured rockfishes include larger dorsal spines than in dark-coloured species, which allow the former to perform more effective defensive responses (Zhuang et al. 2020; Deville et al. 2023a). However, they were not assessed here due to the lack of data for most species.

Along the gradient of depth, the separation of bright and dark-coloured rockfishes in concordance with differences in growth rates and morphological traits (Table 2, Fig. 7), indicated that colouration differences are included in the correlational selection process across Sebastes rockfishes. This correlational selection has important implications for the diversification of rockfishes since combinations of certain colourations with specific values in life-history traits and morphological features may have led to changes in the genome structure and developmental pathways (Forsman et al. 2008; McKinnon and Pierotti 2010; Kolora et al. 2021). Bright-coloured rockfishes that dwell in deeper waters are endowed with other adaptations related to body colouration to live in their environments. Voronin et al. (2021) have pointed out that rockfishes in deeper waters have higher levels of specific lipids in their muscles that help them to maintain buoyancy while providing the necessary energy for performing vertical migrations through the water column. These lipid adaptations are possible through the differential expression of genes and pathways that control lipid metabolism and synthesis (Kolora et al. 2021). The presence of red- and orange-coloured rockfishes in the deeper ocean layers is related to these genetic variations because the metabolism, transport, and accumulation of carotenoids are tightly connected with the lipid metabolism (Yang et al. 2021). Moreover, the correlational selection includes positively selective genes responsible for DNA repair in bright-coloured rockfishes with large lifespans (e.g. S. aleutianus, S. aurora, and S. ruberrimus) (Heras et al. 2015; Kolora et al. 2021). These positively selected genes may come into play whenever these rockfishes undertake vertical migrations, mitigating the potential negative effects of UV radiation exposure in shallow environments.

Although genomic evidence regarding the development of colourations in rockfishes is not presented here, comparative studies in fishes displaying similar colourations anticipate that genes associated with carotenoid transport, deposition, and xanthophores formation may be upregulated in species with high dominance of red categories (e.g. S. aurora, S. mentella, S. polyspinis, S. variegatus, etc.) (Fig. 2), while genes responsible for melanin synthesis should be highly expressed in species with dominance of black categories (e.g. S. ijimae, S. ventricosus, S. vulpes, etc.) (Fang et al. 2022). Furthermore, the fact that juveniles of bright- and dark-coloured rockfishes feed on prey items containing carotenoids (Love et al. 1991) supports the idea that their differentiation in dominant colourations is due to the differential absorption of carotenoids from their shared prey, which can only occur through the differential expression of genes involved in carotenoid metabolism or xanthophores synthesis.

The analyses of species pairs confirmed that colour differences are initially correlated with dissimilarity in depth distributions. Most species pairs displayed only slight differences in their dominant colour categories (e.g. S. flammeus and S. iracundus, S. ijimae and S. zonatus, etc.) (Fig. 6), reflecting the moderate level of phylogenetic signal in the PCoA1. Only a pair of sister species (S. rubrivinctus and S. serriceps) displayed the divergence of reddish and blackish colours in the dominant categories (Fig. 2), which suggests that the main divergence in rockfishes with dominance of dark and bright colour categories in shallower and deeper species, respectively, took place over 3 million years ago (i.e.phylogenetic distance > 6) (Fig. 6). This divergence is coincident with the evolution of fundamental ecological morphotypes, including those associated with the differentiation between shallow reef and deep reef species (Gunderson and Vetter 2006; Hyde and Vetter 2007). Relatively recent divergences between rockfishes with dominance of dark and bright colourations have been reported when comparing ecotypes within two species inhabiting the northwestern Pacific (S. vulpes and S. zonatus) (Muto et al. 2019) and within the only two species reported in the southern hemisphere (S. capensis and S. oculatus) (Venerus et al. 2013). Of them, the divergence between dark and bright ecotypes within S. oculatus hypothetically occurred less than 350, 000 years ago, with the dark-coloured ecotype found in shallower environments than the bright-coloured ecotype (Venerus et al. 2013).

The analyses of species triads showed that colour and phylogenetic distances were not correlated. This finding disagrees with expectations where genetic stochastic mechanisms such as gene drift and mutation, lead to the establishment of defined colour differences between diverging populations in scenarios of sympatric speciation (Forsman et al. 2008). The random arrangement of dissimilarities in depth distribution and colour distances indicates, once again, that the asymmetric depth distributions of rockfishes are solely an initial condition for their speciation (Ingram 2011). Significant genetic divergences in fast-evolving markers suggest that colour distinctions between closely related species would function as prezygotic isolating barriers (Hyde and Vetter 2007). However, the lack of increase in colour differences from species pairs to triads (Fig. 3) also suggest that in the case of rockfishes, colouration may not be an important trait for identifying conspecifics at depth since there is limited available light. Hence, colour differences may only be effective prezygotic isolating barriers when combined with other traits like body size, and the utilisation of acoustic, chemical, and visual communication systems, and within specific scenarios of assortative mating (Schumer et al. 2017; Deville et al. 2023b). Moreover, considering the reports of hybridization within the genus (Valentin et al. 2002; Muto et al. 2013b; Buonaccorsi et al. 2005; Burford 2009; Saha et al. 2017; Keller et al. 2022; Deville et al. 2023b), the colour differences between closely related species are not increasing (Fig. 3) because of heterozygote advantage (Forsman et al. 2008) or because the divergence between closely related species is progressing independently of their colour differences (Stankowski and Ravinet 2021).

The exploration of colour differences estimated here was straightforward. However, it is important to acknowledge that the colour differences estimated here do not account for the spatial distribution of colour categories, particularly regarding the comparison of “homologous” pixels across species. To gain a more comprehensive understanding of these variations, it is essential to perform analyses of colour distribution, especially when dealing with closely related species that share similar dominant colour categories but differ in colouration patterns (Kronforst and Papa 2015; Weller and Westneat 2019). Taking for instance, the case of the species pair S. kiyomatsui and S. scythropus, which exhibit a similar reddish/pinkish dominant colour category, yet these species can be distinguished by unique coloured patches (Kai and Nakabo 2004) that had the highest values of presence after the most dominant colour category in each respective species (Fig. 2). However, in the context of this study, the quantification of colours using the colordistance package was chosen as the preferred approach over spatial methods like patternize (Van Belleghem et al. 2018) because of the remarkable diversity of body colours within Sebastes. By these means, it was ensured a more general exploration of colour variations and distinctions amongst species.

Conclusions

Our analyses confirmed the expectations that bright-coloured rockfishes inhabit deeper waters and have slower growth rates than dark-coloured rockfishes. The presence of bright-coloured rockfishes in deeper waters may represent a photoprotective strategy against the potential negative effects of UV radiation and the higher predation risk and predatory difficulty for these species in shallower environments. However, these rockfishes are endowed with morphological adaptations to deal with the challenging conditions in these environments. The separation of body colourations along with differences in growth rate and morphological traits confirmed that the correlational selection occurring in rockfishes includes body colouration. This selection is expected to lead to changes in genome structure and development pathways that include lipid and pigments metabolisms, and chromatophores development. The analyses of species pairs and triads indicated that depth is an initial driver of colour differentiation and that colour differences do not progressively increase with genetic divergences within triads. With these analyses, it was observed that the main divergence between bright- and dark-coloured rockfishes was related to the divergence of fundamental ecological morphotypes. Overall, this study emphasizes the intricate relationship between genetics, adaptation, metabolism, and the striking diversity of body colourations observed in rockfishes.