Sample and data collection

This study involves sampling and killing of wild fishes, and for such studies a scientific fish fieldwork from the Directorate of Fisheries in Iceland (www.fiskistofa.is) is needed. Arnar Pálsson, an author on this study was in charge and present for all sampling efforts and, for the period of study, had such scientific fish fieldwork permits (#0460/2021-2.0 and #0042/2014-2.13). Sexually mature arctic charr from Lake Þingvallavatn were collected (Fall 2020/21) using composite nets laid overnight (fish died in net) and seine nets (fish were killed by single blow to the head). Fish were randomly sampled, except we aimed for similar representation of morphs and sexes (LB: 49 (21%), SB: 82 (34%), PL: 61 (25%) and PI-charr: 48 (20%), S1 Appendix). Each fish was measured (weight, fork length (FL)), and photographed from left lateral view (Canon EOS 77D, EFS 18–135 mm lens), and sex, sexual maturity, and stomach contents were evaluated by dissection. Morph classification was based on sampling location and date (morphs differ in spawning times and location), external morphology and stomach contents [44,49,50]. Otoliths were extracted from each fish and age determined under 4X magnification in a stereomicroscope. Six jaw bones (dentary, articular-angular, quadrate, premaxilla, maxilla, supramaxilla) were extracted using maceration protocols [50] and photographed (Canon EOS 77D, 50 mm Microlens). The bones from the left side of the head were arranged stereotypically: the dentary and maxilla were photographed and landmarked in medial view, the articular-angular, quadrate and supramaxilla in lateral view, and the premaxilla in fronto-lateral view. To capture external morphology for morph classification, 37 2D landmarks (see S2 Appendix) were digitized for each individual using tpsdig2 (http://www.sbmorphometrics.org/soft-dataacq.html). To study variation in internal feeding structures, we registered 2D landmarks on the six jaw bones with tpsdig2 (Fig 1D–1I and S3 Appendix). All landmarking (external and internal) was conducted by one person (GOJ). To test repeatability of the bone landmarks 168 specimens were landmarked twice. Two landmarks were excluded after the repeatability analysis, LM 12 for premaxilla (Fig 1G and S3 Appendix) and LM 15 for the articular-angular (Fig 1E and S3 Appendix) because they were added after the first trial. All images of fish and bones are available on Figshare+ (doi:10.25452/figshare.plus.25118825). Landmark data, flatfiles and R-scripts are available on Guithub (https://github.com/GudbjorgOskJ/CharrBonesTVV).

Statistical analyses

Landmark data were analysed using the R package geomorph version 4.0.0 [51,52]. Landmark coordinates were aligned using Generalized Procrustes Analysis (GPA) via the function gpagen. Aligned shape variables were then used for all subsequent analyses. Outliers were identified using plotOutliers and removed if the bone was damaged. We tested the repeatability of the bone landmark sets in two ways, using Procrustes ANOVA, using procD.lm, to characterize measurement error (differences between replicates), as well as two-block partial least squares (PLS) analysis with the function two.b.pls to analyze the degree of association between the two replicates (trials). Centroid size (CS) was calculated separately for each of the six bones (CS_bone; indicates use of CS of a bone, and applied to each bone separately), head (CS_head) and whole-body (CS_body). As size is an important predictor for shape in fishes [53,54] and the morphs are known to differ in adult size and proportions [37], absolute size and the relationship between CS_body and CS_head as well as their relationship with each bone (CS_bone) were examined. We tested for differences in size and allometry of size measures between morphs using Procrustes ANCOVA using the function procD.lm. As salmonid morphology is influenced by sex and age [55,56], we investigated the influence of these variables on morph mean and allometry for size and shape. We then did pairwise tests of groups using the pairwise function to test for significant differences between groups. External morphology (whole-body and head shape) was then examined using Procrustes ANOVA to characterize morph differences.

We tested for bone shape differences among morphs and differences in allometric relationships of bone shape and bone size across morphs using Procrustes ANCOVA, using procD.lm, with CS_bone, morph and their interaction as independent variables. Null (shape = CS_bone), reduced (shape = CS_bone + morph) and full models (shape = CS_bone x morph) were compared to determine best fit. When the full model had the best fit, differences in allometric trajectories were investigated using pairwise with CS_bone as a covariate. We tested for significant differences in both vector angles (vector correlation) and vector lengths (rate of shape change per unit covariate), considering significance among both tests as robust support for biological differences. For all tests α = 0.008 was chosen, using the Bonferroni method as a conservative correction for multiple comparisons (six bone datasets). Additionally, we considered comparisons with Z scores <3 as poorly supported, and interpreted results below that as not biologically meaningful and chose the reduced model for further interpretation. Differences in allometry among morphs were investigated using plotAllometry and visualized using predicted values from regression analyses. Principal Component Analysis (PCA, using plotTangentSpace) was used to visually represent shape variation among individuals and morphs. For bones that did not show significant differences in allometry among morphs, residuals from a regression of shape on size (size-corrected shape variables) via procD.lm were used for visualization. Shape variation along PC axes was summarized with deformation grids and vector plots from shape.predictor.

To analyze integration among bones, pairs of bones were compared using two-block partial least squares (PLS) analysis with the function two.b.pls. We compared elements within and among bones of the upper and lower jaws. The effect sizes were compared to determine if there was significant variation in levels of integration across different bone pairs or morphs using compare.pls. For all analyses we used randomized residuals in 1000 permutation procedures to assess statistical significance and estimate the effect sizes (RRPP implemented in geomorph, [57]. For integration tests α = 0.003 was chosen as a conservative threshold (Bonferroni correction for 15 pairwise comparisons among bones). Figures were generated using R and formatted using Inkscape.

Differences in body size, sex and age of sympatric morphs

To capture the morphological diversity in the four sympatric morphs from Lake Þingvallavatn we studied 240 sexually mature individuals. As was expected, morphs differed in size (fork length FL, F = 350.94, p << 0.001), log weight (F = 381.28, p << 0.001) and age (F = 56.75, p << 0.001, S4–S6 Appendices). SB- and PL-charr were smaller at sexual maturity than the LB- and PI-charr. The weight was influenced by sex (p < 0.001), and there was a significant sex by morph interaction (p = 0.009). The external whole-body (p = 0.001) and head (p = 0.001) shape differed by morph, with pelagic morphs having a more fusiform body shape and a more pointed snout (S7 Appendix). There was significant sexual dimorphism in whole-body shape (R² = 0.04, p = 0.001, S8 Appendix), with males having a more humped back compared to females (not a craniofacial trait). These results corroborate morphological descriptions of the morphs [44] and known sexual dimorphism [56]. There was weak sexual dimorphism (R² = 0.01, p = 0.007) for shape of head and all six bones (R² ≤ 0.02, S8 Appendix) that will not be discussed further. Age was strongly correlated with FL (R² = 0.7), but always explained a lower proportion of variation than FL (for external shape, bone size and shape variation). Therefore, FL was used as a covariate in allometry analyses of bone size relative to body size.

Allometry of bone size and shape

Examination showed that all six bone landmarks sets were highly repeatable (S9, S10 and S11 Appendix), the association between replicates was high (r-PLS ~ 0.84–0.99) and variation due to measurement error was low relative to variation among individuals (R² = 0.002–0.005, vs. R² = 0.89–0.94, S11 Appendix). Thus, the landmarks developed here for shape of these six bones in salmonids appear to be robust.

Predictably, the size of all bones (CS_bone) increased with body length (log_e FL, R² ≥ 0.94, p = 0.001, Fig 2 and S6 and S12 Appendices). Morph and morph by size interaction, while significant, accounted for a small fraction of the variation (R² ≤ 0.04, S6 Appendix). Additional testing revealed small differences in bone size allometry between pairs of morphs (S13 and S14 Appendices). These differences mostly reflected differences between larger (PI-charr) and smaller (SB-, PL-charr) morphs. PI-charr have significantly larger dentary bones than all other morphs (p = 0.001), as well as significantly larger articular-angular, maxilla, and supramaxilla bones than both LB- and PL-charr (p < 0.002). PL-charr also have a significantly smaller quadrate, premaxilla and supramaxilla compared to SB-charr (p = 0.001). As CS_bone was highly correlated with body size (CS_body and log_e FL) and a common allometric relationship was found among morphs, we used CS_bone as size measures in subsequent analyses.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Size and shape allometry of dentary, articular-angular, premaxilla and maxilla (from top to bottom, inset pictures of each bone).

Left panels: relationships between bone size (y) and fork length (x) by morph; Right panels: relationships between bone shape (y) and bone size (x). Inset are the associated shape changes related to each component, grey outlines the mean shape, and black the two extremes for each PC. Shown are values for individuals (open circles) and the predicted values (filled) of regressions for both bone size (CS_bone) vs body size (log_e FL) or bone-shape versus bone-size. Log_e, natural log transformation.

https://doi.org/10.1371/journal.pone.0300359.g002

For all bones, shape was significantly affected by size (CS_bone), the allometric relationship explained 20–30% of shape variation for dentary, articular-angular, quadrate, and maxilla (R² = 0.2–0.3, Table 1, for premaxilla and supramaxilla, bone size only explained 10–15%). Note, for the dentary and articular-angular morph explained a similar amount of the shape variation as size (Table 1 and S15 Appendix). The allometry of all bones varied by morph (p = 0.001, Table 1), except for the quadrate, however the effects were small for maxilla and supramaxilla. Further comparisons of vector correlation angles and lengths indicated that allometric relationships vary only markedly by morph in the dentary (S16–S18 Appendices). For this bone, while the common allometry and morph effects on shape were much larger than the interaction, there was significant support for allometric differences among several morph pairs (S16 Appendix). Specifically, the PI-charr differed significantly in allometric trajectory orientation from all other morphs (p ≥ 0.003), converging on a similar shape as LB-charr at larger size. While the allometric trajectories of LB- and SB-charr had similar lengths, the rate of shape change was higher in LB- compared to SB-charr (p = 0.001). Therefore, we present below the mean shape differences among morphs from the full model (CS_bone X morph) for the dentary but for all other bones results from the reduced model (CS_bone + morph), as both common allometry and morph affected shape variation significantly (p = 0.001). All visualizations, unless otherwise noted, have been corrected for size to remove the effects of allometry on shape. We focus first on the lower jaw bones and then the upper jaw.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. The influence of size (bone centroid size) and morph on shape variation in 6 jaw bones.

Analysed with Procrustes ANCOVA, testing for main and interaction terms. For all bones, except quadrate, was the full model with interaction the best fit.

https://doi.org/10.1371/journal.pone.0300359.t001

Benthic and pelagic divergence in shape of specific lower jaw bones

For the dentary and articular-angular, the morph effect on shape variation was similar in magnitude as the common allometric effect (20–30% of shape variation, Table 1). For the other bones the size (allometry) was considerably stronger.

Most dramatically, pairwise tests revealed differences in dentary shape between all morph pairs across the benthic-pelagic axis, except among the two smaller morphs (SB- and PL-charr, p = 0.119, Table 2). Also, the comparison PI-SB-charr was weakly supported (Z < 3). No shape differences were within morphotype (LB-SB-charr, PL-PI-charr). The dentary shape differences by morph were visualized using PCA (Fig 3), and due to variation in allometry among morphs, size effects were not removed. The morphs separated along a benthic-pelagic axis on PC1 (50% of the total variation) with pelagic morphs, on average, having an elongated dentary, increased length of the dental palate, more acute angle between ridges, and vertically inclined lingual palate relative to benthic morphs, which aligns with the hypothesis of shorter lower jaw in benthic charr. PC1 represents elongation of the dentary bone, with positive values representing a more elongate shape, with extension of the dental palate containing the teeth, and negative values representing a compact shape with reduction in the length of the dental palate. Positive values were also associated with a more acute angle between the dorsal and central ridges, reducing the size of the Meckelian notch, as well as a more vertical inclination of the lingual palate. PC2 explained 13% of the total variation, and mainly split the morphs by size, the larger morphs (LB-, PI-charr) having positive values and taller bones. The associated shape changes mainly affected the height of the whole bone, with negative PC values associated with thinner bones overall, reduced height of the mandibular symphysis, coronoid process, and ventral shelf. Together, PC1 and PC2 separated all morphs clearly in shape space, but they overlap (possibly due to overlapping sizes ranges and/or more size variation within LB-charr).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3.

Strong morph effects on shape variation in the lower jaw bones, dentary (top) and articular angular (bottom). For the dentary, size effects were not removed, but effects of size were removed for articular angular (and other bones, see below). PC-plots of specimens and deformation grids show how shape differences by morphs (grey outlines the mean shape and black lines the two extremes for each PC). None of the shape warps are magnified. Each dot represents an individual and the ellipses 95% CI for the distribution by morph (large dot represents the mean for each morph in these dimensions). PC1 and 2 explain 50% and 13% of the variation in dentary, respectively. For the articular angular (bottom) PC1 and 2 explain 37% and 13%.

https://doi.org/10.1371/journal.pone.0300359.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Morph differences in bone shape estimated by pairwise distances in mean shape, from the best fit ANCOVA models for each bone (Table 1).

https://doi.org/10.1371/journal.pone.0300359.t002

Pairwise tests for articular-angular shape also revealed differences between all morph pairs across the benthic-pelagic axis. Biologically meaningful differences were not found within morphotypes (Z<3, see Table 2). The morphs aligned along a benthic-pelagic axis on PC1 which explained 37% of the variation. The two pelagic morphs were largely overlapping (positive values), with elongated anterior process of the angular “wing”, shorter coronoid process, and narrower quadrate facet. The benthic morphs had more shape variation (especially the LB-charr, as in dentary) and more compact bone shape and wider quadrate facet overall, with LB-charr individuals being most compact (Fig 3). PC1 represents changes in relative sizes of the anterior process of the angular ‘wing’, retroarticular, and quadrate facet. Individuals with positive PC1 values show anterior extension of the anterior process of the angular “wing”, increase in the size of the anterior process of the retroarticular but dorso-ventral compression of the retroarticular, and narrow quadrate facet. Negative values associate with a more compact shape overall, particularly with restrictions to the length of the anterior process of the angular "wing”, however the coronoid process is extended anteriorly, a more square retroarticular, and wider quadrate facet. PC2 explains 13% of the variation and mainly represents differences in the ventral process of the retroarticular, with individuals with negative PC values with a shorter process. The morphs did not separate along PC2.

In sum, the morph differences in the articular-angular and in particular the dentary were large. In contrast, for the third lower jaw bone the quadrate morph had weak effects on shape (R² = 0.02, Z < 3, Table 1). PCA and pairwise tests for mean quadrate shape differences did not reveal shape divergence among morphs (Table 2 and S19 Appendix).

Weaker divergence in shape of upper jaw bones

For the upper jaw bones the magnitude of the morph and size effects on shape were similar (~10–15%, Table 1) but notably lower than in the dentary and articular-angular. Morph had strongest effect on the premaxilla and weakest (nearly none) on the supramaxilla. The analyses were done on size corrected data, assuming common allometry.

Pairwise tests for mean premaxilla shape differences among morphs indicated significant differences in all morph pairs across the benthic-pelagic axis (p ≤ 0.001, Table 2). No differences were found within morphotype. The morphs separated along a benthic-pelagic axis on PC1 (25% of the variation) which captured morphological shape differences of the posterior and anterior limbs. The PL- and PI-charr had positive PC1 values with elongated and narrow limbs, while SB- and LB-charr overlapped with more negative PC1 values, having compact and wide limbs (Fig 4). PC2 explained 14% of the variation, representing variation due to differences in the curvature of ascending limb but did not separate morphs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Pronounced divergence in premaxilla of the upper jaw and subtle changes in maxilla shape.

Size corrected PC-plots of specimens and deformation grids showing how the morphs differ in mean (top) premaxilla and (bottom) maxilla shape (grey outlines the mean shape and black lines the extremes for each PC). None of the warps (are magnified, representation of PC plots and warps as in Fig 3. For premaxilla (top) PC1 and PC2 explain 25% and 14% of the variation respectively. For maxilla (bottom) PC2 and PC3 explain 23% and 10% respectively (PC1 was biased by sampling error, and not depicted).

https://doi.org/10.1371/journal.pone.0300359.g004

For maxilla shape, pairwise tests of mean shape between morphs revealed significant separation between all morphs (p = 0.001), except the SB-LB-charr and LB-PI-charr (Table 2). Therefore, maxilla diverged neither exclusively along the benthic-pelagic axis nor the size gradient. The morphs separated on the benthic-pelagic axis on PC3 (10% of variation). This component represents the height of the maxilla (positive: tall and negative: short) aligning with our hypothesis, but the distributions overlapped substantially (Fig 4). PC1 explained 33% of the total variation, but it represents technical variation (either due to rotation or tilting of the bone during photographing, that was randomly distributed with respect to morph). PC2 (23% of total variation), representing biological variation due to differences in the curvature of the maxillary head and angle of the maxillary neck, and separated the PL slightly from other morphs. In combination, PC2 and PC3 indicate that PL-charr has reduced morphological variation and appears to have a thinner maxillary body and caudal lobe than the other morphs.

Pairwise comparisons of mean supramaxilla shape differences among morphs indicated SB deviated from all other morphs (p = 0.001, Table 2). PC1 and PC2 explained 62% of variation, but no combination of PCs revealed clearly how SB-charr shape deviated (S19 Appendix). We note the morphological variation in SB appeared reduced relative to other morphs.

Covariation in shape change among bones

Lastly, we tested for covariation in shape among bones. Within the upper jaw and within the lower jaw, shape covaried significantly among all bones (p = 0.001). The degree of covariation was greatest for the dentary and articular-angular (r-PLS = 0.92; Z = 7.08, S20 Appendix) and much higher compared to the other pairs (r-PLS = 0.54–0.69). Among the jaws, all covariation among bones was significant (p = 0.001) and the degree of covariation was moderate (r-PLS ~ 0.54–0.78, S20 Appendix). We did not find significant differences in effect size among pairs in the lower jaw, however in the upper jaw integration between the premaxilla and supramaxilla was found to be lower (p = 0.006) than all other comparisons, and integration between the maxilla and supramaxilla was found to be higher (p = 0.002) (S20 Appendix). In addition, analyses did not reveal significant differences in effect size between morphs for any of the bone pairs (comparisons limited to bones that showed significant covariation or biological importance as described above, quadrate and supramaxilla excluded).

Bone shape variation and functional divergence

We found strong divergence in the lower jaw along the benthic-pelagic axis. The characteristic subterminal mouth of the benthic morphs associates with a shortening of the dentary and to lesser extent articular-angular (relative to the quadrate facet) aligns with our hypothesis (3). Additionally, the anterior region of the dentary around the mandibular symphysis is angled ventrally relative to pelagic charr (Fig 3). This may aid in creating a more optimal angle to grip attached prey or perform scraping behaviours on a surface to dislodge prey, such as snails. The benthic morphs may use a combination of biting and ramming behaviours, as suction forces are likely weak. Taller dentary and articular-angular bones could be adaptations for feeding on benthic prey, increasing size of the mouth cavity and efficiency of consuming benthic organisms [69]. The coronoid processes of the dentary and articular-angular are enlarged in benthic charr, which could associate with a larger mandibular portion of the adductor mandibulae and a larger aponeurosis maxillaris [70], thus allowing for greater bite forces in benthic morphs which could aid in dislodging attached prey items [63,71,72]. Relatively more elongated lower jaw was expected and observed in pelagic morphs. Consistently, previous study on lower jaw bone shape in juveniles of O. mykiss found the anadromous population (with presumed pelagic lifestyle) had a more elongated and streamlined lower jaw than the resident population [17]. The authors postulated that elongated, streamlined lower jaws could be adaptations to the pelagic ocean environment. It would be interesting to see if this holds in more salmonid populations, and also if this is accompanied by notable allometric changes in the dentary as we found here.

In the upper jaw, we found premaxilla shape to diverge along the benthic-pelagic axis more strongly than the maxilla. PL-charr had smaller premaxilla overall (importantly relative to SB which is of same size), with shorter ascending limbs and narrow anterior and posterior limbs. Reductions in the height of the ascending process and elongation of the anterior-posterior axis of the premaxilla has been found in other pelagic fishes [64,73,74] that use ram feeding in the water column [75] while suction feeders with highly protrusible jaws have long ascending limbs [76]. Moreover, shortening and steeper angles of the ascending limb of the premaxilla can also be associated with increased biting forces [77] and possibly linked to biting behaviours. The ascending limb of the premaxilla has been found to extend more dorsally in pelagic fish [78] and in highly protrusible jaws [79], however, in our study the ascending limb was proportionally longer in the benthic morphs. As the premaxilla, and the skull as a whole, is relatively rigid in salmonids [sensu 70], the biomechanics of similar morphological changes in fishes with protrusible jaws could be incomparable. For the maxilla, the third shape PC associated with the benthic-pelagic axis. The maxilla of fishes is generally involved in the protrusion of the upper jaw maxillary apparatus, while moveable in salmonids such function may not be present [70] and the proposed movement of these components to aid in salmonid feeding [80] has been criticized [70]. The shape change seen in the bone matches hypothesis 2, based on observations done on external photographs in pelagic charr [49].

Neither the quadrate nor supramaxilla diverged along a benthic-pelagic axis (but showed some weak morph differences). While shape variation in them could reflect biomechanical differences due to size [81,82], they did not vary along a morph size gradient. In salmonids, the function of the supramaxilla (if any) is unknown. The morph differences in supramaxilla may reflect correlations with linked bones (like the maxilla), as the high level of integration between these bones indicates. In contrast, the quadrate is a critical linkage in the suspensorium, its main function to allow rotation of the lower jaw. Since changes to the maximum angle of opening potentially needed to capture different prey would likely only require subtle morphological changes in this bone, it is unlikely we could detect these differences given our methods, or divergent selection may not have acted strongly on quadrate shape.

While adult morphology should reflect long-term prey consumption (e.g., higher pelagic prey consumption associates with stronger pelagic features), variation in the mechanisms of capturing and processing of different prey items within and between sympatric charr morphs, and potential impacts on morphological variation, is poorly understood. To our knowledge biomechanical studies of salmonids have not yet been done on fishes that specialize on such divergent prey as snails and zooplankton. Feeding was compared in PL- and SB-charr, revealing reduced efficiency of zooplankton feeding (pelagic) in SB-charr juveniles due to lower attack rates [40], increased handling time [83], increased ejection of prey [83] and poorer retention in the branchial apparatus likely due to shorter gill rakers [40]. Both morphs likely capture zooplankton using a similar strategy, i.e., suction [40], however may exhibit differences in capture efficiency as a result of morphological differences associated with ecological divergence [40]. Prior studies are limited to biomechanics of piscivorous feeding in brook trout (Salvelinus fontinalis) [84] or theoretical models of suspensorium and lower jaw movement [85]. Salmonid jaw elements are likely to be largely immobile due to rigid attachments of the components [70]. Therefore, the use and efficiency of suction is likely limited compared to pelagic fishes with highly mobile jaws [86], and perhaps not highly differentiated among arctic charr morphs.

Benthic-pelagic divergence can manifests in lower jaw length relative to head length [65], as seen in subterminal mouths of benthic arctic charr [37,44,87]. Shelled freshwater snail (Lymnaea peregra) is the main prey of the derived SB- and LB-charr in Þingvallavatn [37,40]. In general, hypertrophied pharyngeal jaws [88–92], molariform pharyngeal teeth [93,94], and decreased gill raker length and density [40,44,95,96] are all morphological adaptations widely employed across fishes that consume large quantities of hard-shelled benthic prey. Also, mechanisms for capturing attached prey vary widely among species [63,71,72]. Use of strong suction, gripping and biting, lateral movements, and highly protrusible jaws are commonly utilized traits [63,71,72]. However, details of the intersection of these biomechanical and behavioural components for feeding on divergent prey resources in Þingvallavatn morphs is currently unexplored. In many groups of fishes, strong benthic-pelagic ecological divergence has led to modification in morphology and behaviour to allow for optimal feeding performance [16,58,59,63,65,74,76,78,97–100]. Clearly, investigation of the behavioural and biomechanical strategies individuals and morphs use to both capture and process prey, and bone shape in the same individuals, is needed to reveal how arctic charr specialize on such divergent prey types.

While we expected adaptions to divergent prey resources to be expressed in shape divergence of cranial elements among morphs, other morphological modifications can occur that do not impact bone shape as studied here. Internal bone size and architecture has been shown to diverge along the benthic-pelagic axis, with benthic feeders exhibiting more robust hollow cranial bones with complex internal structure, similar to mammalian long-bones, and planktivores having thin solid bones [101]. Bones of benthic scrapers also have higher rates of mineralization [102] and are more resistant to bending [101]. We found benthic morphs (LB- and SB-charr) to have thicker and more robust bones overall (relative to PL- and PI-charr), but variation in their internal structure and mineralization is unknown. The differences in adult bone size and shape found here may associate with differentiation in muscle mass and surface area of attachment sites, and as skeletal and musculature may contribute independently to biomechanical diversity [103], adaptive changes to soft tissues may both be more critical in foraging divergence (i.e. proportionally larger impact on biomechanics of feeding related to speed, force, and mode of prey capture) and be more apparent than associated changes in bone shape in these morphs.

Influence of extrinsic and intrinsic factors on bone variation within and among morphs

The observed shape variation in the jaw bones within and between morphs could be influenced by differences in their ecology and food preferences [see above, 40,41], genetic composition of the morphs [36,104] and/or interactions of specific genes and environmental factors.

Adaptive plasticity may contribute to ecological and morphological divergence among sympatric and allopatric charr [21,105,106]. We studied wild caught individuals of the four morphs that mostly operate in different habitats and feed on different prey items that vary seasonally in abundance and nutrition, all factors that may influence their development and adult morphology [40,41,45]. Because the morphs differ in the location, timing and synchronicity of spawning [43], their juveniles may receive different maternal provisions and encounter different juvenile habitats. Arctic charr shows developmental plasticity in body shape, size and behaviour in response to benthic vs. pelagic diet treatment in experimental settings [105,107,108]. Behavioural traits related to habitat choice, prey capture and processing can influence variation in bone traits [97]. For example, in sheepshead wrasse (Archosargus probatocephalus), individuals consuming a higher proportion of hard-shelled prey had larger adductor mandibulae and jaw bones but also employed biting mode of feeding to dislodge prey as well as buccal manipulations to crush shells and separate hard and soft parts before ingestion [109]. Thus, some of the morphological differences between (and/or within) morphs revealed here could reflect this developmental plasticity. Note however, not all environmentally induced variation in bone shape (plastic response) may be adaptive [110].

Genetic differences by groups can be assessed with experiments and population genetics. Common garden rearing of offspring of Þingvallavatn morphs indicate genetic influence on many aspects of juvenile and adult morphology [46,49,56,67,111], though maternal and/or transgenerational parental effects could also play a role [112]. Three of the morphs (PL-, SB-, and LB-charr) are genetically separable [36] and genomic data revealed substantial allele frequency differences between them, on nearly all linkage groups [35,39]. We saw the strongest morph differences in the dentary and premaxilla bones of the lower and upper jaws, respectively, that may reflect alleles influencing these bones specifically. Models indicate these three morphs have been separate for thousands of generations, with limited gene flow between them [35,36]. Curiously, the PI-charr is genetically more heterogeneous, most individuals are genetically similar to PL-charr, but others have either LB-charr or mixed PL-/LB-charr ancestry [39]. One hypothesis is that PI-charr result from ontogenetic niche shifts of juvenile PL-charr that learn to catch sticklebacks, abandoning zooplankton feeding, which may allow individuals to attain larger size and develop piscivorous characteristics [44]. Another suggestion [39] is that juvenile LB-charr and/or LB/PI hybrids also shift to piscivory. For most bones studied here, the PI-charr tended to align with PL-charr (consistent with ancestry), except in the maxilla where PI clustered closer to LB-charr. This heterogeneity of phenotypes in PI-charr could either reflect variation in genes, or the impact of piscivory on development of specific bones. Piscivory and associated size increase may place high functional demands on bone shape [113] and enhance divergence in shape and even suppress morphological variation among piscivorous charr of different genetic origins. As mapping can identify loci, genes or specific polymorphisms impacting variation in shape [114–116], QTL or association studies of bone shape variation in PI-, LB- and PL-charr could identify alleles and pathways of relevance for bone shape variation in salmonids.

Most likely, interactions between particular alleles and specific environmental factors, acting at specific times or over broader developmental windows [117], produce the observed morph differences in bone shape. Considering a hypothetical case, SB-charr may have alleles that influence the development of the dentary and articular-angular (lower jaw bones), that will induce the subterminal morphology–in individuals that eat snails as juveniles. Tests of such hypotheses require joint study of genetic and environmental influences on specific aspects of how bones develop and acquire their shape and functions. To distinguish between genetic and transgenerational environmental effects, mapping, crosses between morphs and multigenerational studies are needed. One could test the sensitivity of shape-alleles to environmental factors, like specific juvenile food types (e.g., benthic or pelagic). Analyses of gene expression can also reveal developmental pathways that respond differently to environment by morph. Already RNA-, methylation- and miRNA-sequencing of embryos of these morphs and aquaculture charr have demonstrated differences by morphs [118,119]. For instance, in PL-, LB- and SB-charr reared in a common garden ~2000 genes were differentially expressed by morphs during development (from formation of the gill-arches to establishment of the viscerocranium primordia) [104]. This proves the morphs differ in genes influencing multiple developmental systems that could induce these bone shape differences. Specifically, as common garden experiments revealed ossification of skull elements in juveniles starts earlier in SB- compared to PL-charr [111], it would be interesting to explore effects of genes and environmental factors on ossification and bone shape. Most probably interactions of specific environmental attributes and genetic composition cause the observed morph differences in bone shape, for instance via differences in patterns of growth or differentiation in the bone primordia or linked tissues.

Future avenues in research of bone shape variation in salmonids

We studied variation in six dentary bones in sexually mature fish of derived morphs from one lake. This study could be expanded in several ways. First, size, age and sex can influence traits [37,120]. We studied sexually mature individuals, and average age and size varied quite substantially by morph. It would be interesting to study the emergence of bone shape variation during ontogeny in these morphs, by sampling age classes and genotyping individuals (to assign them to morph). For bone shape, size (fork length) had stronger effect on shape than age. Also, despite significant difference among morphs in allometries (age-size relationship as well as between size measures of head, body, bones), these effects were rather small, and we concluded bone size was a robust proxy for overall size (and was used to account for size variation). But with more samples it might be possible to disentangle and better estimate these effects. The same applies to potential sexual dimorphism or morph specific sex-dimorphism in bone size or shape allometry. Secondly, as four derived morphs were studied, the phenotypes could not be polarized with respect to an outgroup. Studies of anadromous charr could identify ancestral vs. derived character states. Arctic charr in Iceland have repeatedly evolved along a benthic-pelagic axis and subsequently experienced modification to allow for capture and processing of diverse prey items (e.g., benthic macroinvertebrates, fish, snails) [31,32,35,87]. It would be interesting to study the degree of parallelism in bone shape variation and integration in more sympatric benthic-pelagic morphs or the miniaturized SB-charr. Thirdly, the salmonid head contains several dozens of bones and the cranial case. Ingimarsson [47] found a signal (but weak) of morph differences in linear measures of several other bones, but geometric morphometrics on these and other bones (in 2D or 3D from CT-scans) may reveal this more systemically and highlight potentially adaptive changes [54,121,122]. Analyses of associated variation in soft tissues (muscles and ligaments), cartilage and ossification would also be interesting. We found rather high integration in shape and size variation among bones within and between the upper and lower jaws. However, covariation of shape and univariate traits (like teeth numbers) or variation in cranial kinesis, how the bones lie together in the skull and how they move together [84], was not analysed here. The feeding structures of fishes are highly functionally integrated [3,86]. Study on the seven dolly varden trout (Salvelinus malma) morphs in Lake Kronotskoe indicated morph specific combinations of morphological traits and differences in benthic-pelagic ecologies manifest in mouth position and jaw length divergence early in ontogeny [24]. Three cranial bone sets had with coordinated shifts in ossification: jaws (premaxilla, maxilla and dentary), intraoral tongue‐bite apparatus (vomer and lingual bone, gill arch elements) and skull dermal bones (supraethmoid, frontal and preopercle), most likely reflecting differences in early ossification [24]. Studies on additional internal skull elements and their integration are needed to understand the functional capacity of different jaw elements and how they relate to the remarkable feeding divergence in arctic charr.