Phylomorphometrics reveal ecomorphological convergence in pea crab carapace shapes (Brachyura, Pinnotheridae)

Abstract Most members of the speciose pea crab family (Decapoda: Brachyura: Pinnotheridae) are characterized by their symbioses with marine invertebrates in various host phyla. The ecology of pea crabs is, however, understudied, and the degree of host dependency of most species is still unclear. With the exception of one lineage of ectosymbiotic echinoid‐associated crabs, species within the subfamily Pinnotherinae are endosymbionts, living within the body cavities of mollusks, ascidians, echinoderms, and brachiopods. By contrast, most members of the two other subfamilies are considered to have an ectosymbiotic lifestyle, sharing burrows and tubes with various types of worms and burrowing crustaceans (inquilism). The body shapes within the family are extremely variable, mainly in the width and length of the carapace. The variation of carapace shapes in the family, focusing on pinnotherines, is mapped using landmark‐based morphometrics. Mean carapace shapes of species groups (based on their host preference) are statistically compared. In addition, a phylomorphometric approach is used to study three different convergence events (across subfamilies; between three genera; and within one genus), and link these events with the associated hosts.

It is worth noting that also in these subfamilies, the host specificity is understudied and some species now considered to be free-living might have unknown host associations (McDermott, 2009).
Pinnotherids have evolved a wide range of ecomorphological adaptations that could be linked to their presumed host choice (described for the subfamily Pinnotherinae in de Gier & Becker, 2020).
These include: (A) several types of setae on the walking legs and claws used for feeding, swimming, or camouflaging; (B) asymmetry and widening of the walking legs' segments for feeding purposes and/or grip within the host (or host tube/burrow); and (C) various ornamentations, setaetion and colouration patterns, shape differences, and a variation of thickness of the carapaces in order to fit inside their hosts or to blend with their hosts' colouration (de Gier & Becker, 2020, and references therein for examples of adapted species). In addition, in sexual dimorphic species, females have evolved an enlarged pleon to carry eggs, making them almost immobile and very vulnerable to predation if they ever leave their host (Baeza, 2015). This is likely a consequence of living hidden within a host (de Gier & Becker, 2020). Although the variation in the abovementioned characters is most diverse in the speciose pea crab family, similar adaptations to endo-or ectosymbiotic lifestyles can be found in other brachyuran ("true" crab) taxa (Castro, 2015;Serène, 1961).
Most of the ecomorphological adaptations mentioned above have only briefly been described in taxonomic and phylogenetic literature (Campos, 1996a(Campos, , 2016Manning, 1993;, and were not directly studied with respect to the crabs' host choice. In a large-scale study, Laughlin (1981) mentioned that various pinnixine pea crabs (mainly including species then attributed to Pinnixa White, 1846) have a much wider carapace than the studied pinnotherines (in this case, species of Pinnotheres Bosc, 1801), and linked this character to their biology. More recently, Hultgren et al. (2022) analyzed the relationship between the host choice of a wide range of pea crabs and their carapace size ratios, considering also their phylogenetic positions. In this way, convergence in carapace shapes could be studied. They did this by testing the aspect ratios of 149 species, 59 of which had known phylogenetic positions (see Palacios Theil et al., 2016).
The present study elaborates on the analyses by Hultgren et al. (2022), by using additional morphometrics to investigate the relationship between the adult female carapace shape, and the host choice, focusing on all currently included pinnotherine members. A phylomorphospace approach will be used, including both symbiotic, as well as free-living outgroup species from the two other subfamilies. This projection of the phylogeny should reveal clusters and convergence patterns in the data (Stayton, 2015), indicating that the colonization of similar host phyla has led to analogous carapace shapes in the evolution of pea crabs.

| Selection of illustrations
Similarly to the methods described by Hultgren et al. (2022), published illustrations of 181 pea crab species (in particular Pinnotherinae; see below) were collected through an extensive literature search. Available dorsal views of adult female carapaces (independently of the number of pereiopods depicted in the illustration) were selected for the analyses due to their often obligatory symbiosis, being restricted to remain inside their host. By contrast, males often leave their host, and juveniles have been found to switch hosts multiple times before reaching their adult stages (de Gier & Becker, 2020). In addition, one rarely figured species, and two species without previously published illustrations were photographed using a Leica M165c stereo microscope with a Leica From all the currently recognized species of Pinnotherinae (excluding two genera discussed below), 168 species could be included in the study. Thirty-five species were excluded due to the absence of illustrations of the dorsal view of adult female crabs (Appendix S1).
To cover the morphological variation of the non-pinnotherine pea crabs, outgroup illustrations were selected of ten species from the other pea crab subfamilies. Nine were selected from Pinnixinae and one from Pinnixulalinae. A representative of both the genera Sakaina Serène, 1964 andParapinnixa Holmes, 1895 were also added but considered as outgroups, although they are still within the subfamily Pinnotherinae (WoRMS Editorial Board, 2022), as the phylogeny of Palacios Theil et al. (2016) suggests that this placement is rather questionable. The subfamily classification of pea crabs seems to be unstable and in need of further research, and therefore the subfamily status for these species will be annotated as "Pinnotherinae?". In addition, one species with a tentative placement basal to the three | 3 of 14 de GIER subfamilies is included as an outgroup: Tetrias fischerii (A. Milne-Edwards, 1867). This species has no subfamily status (WoRMS Editorial Board, 2022). Five of the outgroup species were chosen based on their recorded, or presumed endosymbiotic lifestyle: Tetrias fischerii is thought to be associated with bivalves (Milne-Edwards, 1873), the pinnixines Pinnixa barnharti Rathbun, 1918 and Pinnixa tumida Stimpson, 1858 are thought to be internal symbionts of holothurians (Dai & Yang, 1991;Zmarzly, 1992), and adults of the pinnixines Scleroplax faba (Dana, 1851) and Scleroplax littoralis (Holmes, 1895) are commonly found in bivalve hosts (Zmarzly, 1992).
The latter two are suggested to be morphotypes of the same species Zmarzly, 1992) but are treated as separate species in the analyses. Scleroplax faba was reported in various other host types (gastropods, ascidians, and holothurians) in juvenile specimens (Zmarzly, 1992).

| Landmark selection and morphometrics
Collector bias is a common problem in morphometric studies when selecting landmark data (e.g., Percival et al., 2019), as is the use of nonhomologous and inconsistent datapoints (e.g., nonuniform orientations of specimens; Collins & Gazley, 2017). These problems could be evaded due to the uniformity in the orientations of illustrations used in taxonomic pea crab publications: only uniform (dorsal) orientations with visible ocular carapace ridges (cavities for the eyes) were used (with the exception of anteriorly ornamented species).
Landmark (LM) selection was done to digitize the right half of the pea crabs' carapace shape, with the inclusion of three landmarks (LM 1, 3, 22), one semi-landmark (LM 2), and 18 sliding semi-landmarks (curve) (LM 4 to 21) along the lateral and caudal margin of the carapace (see Figure 2). Because of the lack of homologous anatomical features on the lateral curvature of pea crab carapaces, sliders were used to capture the shape variation. Landmark data were gathered in tpsDig2 (v. 2.31) (Rohlf, 2017) and analyzed using R v. 4.2.1 and Rstudio v. 2022.07.0 (R Core Team, 2022;RStudio Team, 2022), using the packages geomorph v. 4.0.4 and ggplot2 v. 3.3.6 (Adams et al., 2022;Baken et al., 2021;Wickham, 2016). A generalized Procrustes analysis was performed to scale, transform and rotate all images, or a subset of the images for morphospace analyses. In this way, the scale was set to "uniform", also to take into account potential uniform swelling due to preservation in ethanol. A Procrustes pairwise (M)ANOVA with a residual randomization permutation procedure (1000 permutations, RRPP;  was performed to find significant differences in the mean shape data based on the host associations of the specimens. The Procrustes pairwise ANOVA test compares the mean shapes of two specified groups by checking the relative distance between these two shapes (Goodall, 1991). In this way, it explains if the differences between the groups are large enough (e.g., significant) in comparison to the variation within the groups.
The full dataset, including outgroup species, as well as a subset only including the "true" Pinnotherinae (i.e., excluding the members of Parapinnixa and Sakaina), was tested and compared. All species were labeled considering their host association: bivalve-, gastropod-, holothurian-, ascidian-, echinoid-, and brachiopod-associated, and tube/burrow-dwelling. The echinoderm associates were separated based on the external or internal nature of their symbiosis (ecto-or endosymbiotic). In addition, two outgroup species are labeled as "free-living", although they could have been dislodged from their in the dataset. These 21 species are often rarely caught, poorly described, or have a very questionable host association. Two of these species (Hospitotheres powelli Manning, 1993 andPinnotheres taichungae Sakai, 2000) were previously recorded as burrow-dwelling or free-living. It has been argued that they may have been dislodged from their host or that their host was destroyed during collection (de Gier & Becker, 2020;McDermott, 2009). Despite this, the choice was made to include them in the dataset, in order to test whether there is a predictive value in the analysis (see Discussion).

| Phylomorphospace analyses
In order to include the available phylogenetic information of 33 spe-  Revell, 2012). Three convergence events were highlighted in the phylomorphospace using ggplot2. The phylogeny reconstruction is treated as an overlay on the presented morphospace. The branches were also used to statistically test three potential convergence events in the phylomorphospace plot.
Similarity-based measures (C 1 to C 4 , and corresponding p-values) were calculated using the R package convevol v. 1.3 (Stayton, 2018) as described by Stayton (2015). Examples of their uses are presented by Serb et al. (2017), Zelditch et al. (2017), Stange et al. (2018), and Grossnickle et al. (2020). A custom R-script (Zelditch et al., 2017;Zelditch, pers. comm.) was used to run 1000 replicates to check the results from the convevol package. For the calculations of the C-values, PC-values were used from PC1 to PC3 (84.5% of the explained data).
A phylogenetically informed ANOVA (Phylogenetic Generalized Least Squares; PGLS) was performed to investigate the impact of host choice on the shape variation in the data while controlling for the independence of the residuals from the phylogeny (Adams & Collyer, 2018;Mundry, 2014). This was done for 33 species, using the procD.pgls() command in geomorph, with Pagel's lambda (λ) (Pagel, 1999) set at 1.0 (a high phylogenetic signal-Brownian motion model). For comparison, a regular Procrustes ANOVA/regression, without the implementation of a phylogenetic framework, was performed for the 33 species (similar to the pairwise test explained above). In both analyses, a similar RRPP approach was used as mentioned above (1000 permutations).

| Morphospaces and mean shapes
The morphometric analyses of the scored landmarks revealed the overall variation in carapace shapes (for a morphospace plot with numbers indicating the species, see Appendix S2 (list) and S3 (figure)). When including the outgroup (non-pinnotherinae) species, the first two of 43 principal components (PCs) explain 74.4% of the variation in the data ( Figure 3). PC1 to PC11 together explain 99% of the data, meaning that the rest of the 32 PCs explain less than 1% of the data. Along these first two axes, the mean shapes change mainly in width and length: along PC1, the carapace changes from an elongated and rounder shape (PC1 min ) to a widened, angular shape, with more defined ocular cavities in dorsal view (PC1 max ). Along PC2, the widest point of the carapace seems to slightly shift from the anterolateral side (PC2 min ) to the posterolateral side (PC2 max ), meaning that the shape in the middle would have the widest point in the middle of the carapace. In addition, the rostrum seems to be much wider and more defined in specimens from the upper side of the plot (PC2 max ) ( Figure 3). This also means that a perfectly round species would approximately be found in the center of this plot (0,0).
There is a clear separation between the ingroups and the outgroups, except for one outgroup species: Tetrias fischerii. Outgroup species with a similar host association to ingroup members (namely endosymbionts of holothurians or bivalves; see Appendix S2, S3) | 5 of 14 de GIER are closer to the ingroup than to the rest of the outgroup species, which are burrow-and tube-dwelling ( Figure 3). The significant difference between the point cloud of this host type and the rest of the host categories was confirmed by the pairwise Procrustes ANOVA (p < .01; Table 1).
All species of the ingroup are covered by a vast cloud of bivalveassociated points (Figure 3; Appendix S2, S3). Although overlapping, significant results were found in the data by the pairwise ANOVA, taking all 43 PCs into account (Table 1). The five ascidian-associated species group on the left side of the plot. Ascidian-associated species were found to be shaped significantly different from gastropodassociated (p = .010) and externally echinoid-associated species (p = .027). In addition, a nearly significant difference was found between the mean shape of the ascidian-associated species and the bivalve (p = .056), and internal holothurian associates (p = .069).
Bivalve-associated species were significantly different from gastropod associates (p = .044), internal associates of holothurians (p = .047), and external echinoid associates (p = .007). Between these last two groups, a significant result was found (p = .006). Lastly, internal holothurian associates were significantly differently shaped than gastropod associates (p = .015). The actual morphological differences between the carapace shapes are explained in detail below.
When excluding the outgroup from the analyses, the bivalveassociated convex hull overlaps all but five specimens with a known host association (Figure 4). These species (the external echinoid associate Dissodactylus latus Griffith, 1987, the holothurian-associated Holothuriophilus trapeziformis Nauck, 1880, and the three gastropodassociated Mesotheres unguifalcula (Glassel, 1936), Orthotheres bayou Ho, 2016, andOrthotheres turboe Sakai, 1969) have a broader body shape than the rest of the ingroup (i.e., a higher AR carapace aspect ratio; Hultgren et al., 2022). Running the analysis with only pinnotherine members influences the p-values of the Procrustes pairwise ANOVA to be lower in all significant values from the analysis with the outgroup species included (Table 1), and the first two F I G U R E 3 Morphospace plot showing the total variation of dorsal carapace shapes of both the in-and outgroups. Warps show extreme shape variation along the first and second PCs. Colors of points and convex hulls correspond to host association type, and shapes give an indication if the species have an unknown, free-living, or generalist symbiotic lifestyle. Diamonds show the non-pinnotherine outgroups. Illustrated species correspond to linked datapoints in morphospace: top, Fabia tellinae Cobb, 1973(after Campos, 1996b; right, Glassella floridana (Rathbun, 1918) (after ; bottom, Durckheimia lochi Ahyong & Brown, 2003(after Ahyong & Brown, 2003; left, Austrotheres pregenzeri Ahyong, 2018(after Ahyong, 2018) (setae in illustrations omitted; crabs not to scale).  Lastly, the rostrum of ascidian-associated species is slightly broader and more pronounced.

| Phylomorphospace approach
A pruned phylogeny tree was projected on the morphospace The three above-mentioned convergence events were statistically tested for their significance (Table 2), indicated by their similarity-based measures (C-values) and significance.
A PGLS analysis was performed to test the dependency of the morphometric data, taking the phylogenetic history of the F I G U R E 4 Morphospace plot showing the total variation of dorsal carapace shapes of the ingroup. Mean shapes of major host-associated groups are plotted (black) and compared with the mean shape of the entire ingroup (gray/white). Colors of points and convex hulls correspond to host association type, and shapes give an indication if the species have an unknown or generalist symbiotic lifestyle. Note that the y-axis (PC2) is flipped compared with Figure 3, and the aspect ratio is reduced to 0.5 for comparison with other plots.
de GIER included species into account. The PGLS resulted in an insignificant p-value (p = .823; R 2 = .151). Thus, the landmark (shape) data and the placement within the morphospace of the 33 included species are not associated with a host group, once phylogenetic nonindependence is taken into consideration ( Figure 5). The comparative Procrustes ANOVA test, excluding the phylogenetic F I G U R E 5 Phylomorphospace plot and the projected ultrametric phylogeny reconstruction. Specimens not included in the phylogeny are omitted in the morphospace for better readability. Colors of points and convex hulls correspond to the species' host association. PCs and corresponding shape changes along these axes same as in Figure 3. Three convergence events are highlighted with arrows in both the phylomorphospace (I to III) and the tree, of which three species (A to C) are illustrated next to the tree: A, Dissodactylus latus Griffith, 1987(after Griffith, 1987

| Shape differences
Although the pairwise Procrustes ANOVA shows significant differences between the mean carapace shapes of the host-associated groups (Table 1), the differences appear to be very inconspicuous ( Figure 4). The mean shape of all bivalve-associated pinnotherine species included in this study was very similar to that of the entire set of the analyzed ingroup species. Therefore, assigning a particular pea crab carapace shape to a specific host group seems impos-   Hultgren et al. (2022) already suggest that two bivalve-associated species (Scleroplax faba and S. littoralis; both included in the present analyses as outgroup) have evolved from having a wide carapace shape (as can be seen in the tube-and burrow-dwelling outgroup species) to having a relatively round carapace. The mean AR (carapace aspect ratio) of these two species was found to be significantly higher than the pinnotherine bivalve associates, but the green-  (Table 2; following Stayton, 2015). Although this is a rather low value due to the distance between the three species, under a Brownian motion model, this result is significant (p < .001), indicative of "true" convergence between these members of the in-and outgroup. Similarly, the two holothurian-associated outgroup species (Pinnixa barnharti Rathbun, 1918 and Pinnixa tumida Stimpson, 1858) seem to have undergone a similar presumed convergence event, shifting away from the rest of the outgroups towards the ingroup (Figure 3). This possible convergence might be related to the endosymbiotic host choice of these two species, as is also discussed for S. faba and S. littoralis by de Gier and Becker (2020) and Hultgren et al. (2022). However, whether these two Pinnixa species are phylogenetically related is unknown, although P. barnharti is found in Californian waters (Zmarzly, 1992), whereas P. tumida is known from Japan and China (Dai & Yang, 1991), which might suggest they are not very closely related. DNA analyses are needed to investigate TA B L E 2 Similarity-based measures of converge for three presumed convergence events in pea crab species combinations (1000 replicates, PC1 to PC3; 84.5% of the data explained) Note: Due to Zaops ostreum (Say, 1817) being the ingroup species with the shortest overall distance to the two Scleroplax species, this species was chosen to represent the ingroup in this (III) calculation. p-Values indicating the probability that the degree of convergence exceeds what would be expected from a randomly evolving lineage are in bold if significant (p < .05). Gier & Becker, 2020). A potential correlation between the shape of the host bivalve and the carapace shape of the symbiont cannot be tested with the current datasets, but more symbionts of elongated bivalves can be found among the pinnotherine (some of which take wide and/or otherwise aberrant shapes: e.g., Raytheres (Campos, 2004), Serenotheres Ahyong & Ng, 2005, and Visayeres Ahyong & Ng, 2007(Ahyong & Ng, 2007Campos, 2002;Ng & Meyer, 2016)).

| Convergence events and host specificity
The C 1 value of this event shows an average of 84.5% convergence, with a highly significant probability (p < .001; Table 2).
There are several other species with a wide carapace, of which no phylogenetic information was available (see Figure 4). convergence measure analysis for these two species, a C 1 value of 81.6% was found, with a significant p-value of .004 (Table 2).

| Ancestral reconstructions and ecomorphological trends
Besides analyzing convergence events, the phylomorphospace approach allows for a close examination of the evolution of shape in the deeper branches of the phylogeny (e.g., Ford et al., 2016). The currently presented phylogeny reconstruction "starts" with a somewhat widened outgroup species, Tetrias fischerii, and the much-more widened species Parapinnixa cortesi ( Figure 5). This first bivalveassociated species is plotted between the large cloud of ingroup species, and the main tube-and burrow-dwelling outgroups (including its currently designated sister species P. cortesi). In

| Future perspectives
A problem posed by the currently presented data was a large number of species with unknown host, presumed free-living, or with generalist host associations. Not all of these species will truly be free-living and might have been dislodged or wandering from their host organism when sampled (McDermott, 2009). Using the present morphospace plots, the association type (endo-or ectosymbiotic) might be speculated by looking at the data clouds (e.g., the outgroup species Pinnixulala heardi Felder & Palacios Theil, 2020b, whose carapace shape is perfectly in line with the other tube-and burrow-dwelling outgroups, as was already speculated by Felder and Palacios Theil (2020b); Figure 3). Within the ingroup, "predicting" a specific host type for species without a known association may be more difficult, and unexpected host associations might influence the shape of the convex hulls in the morphospaces and consequently influence the p-values of the pairwise (M)ANOVA (Table 1) using the currently presented methods for years, studying the evolution of shape as a result of ecological factors, mainly in vertebrates (Claverie & Wainwright, 2014;Curth et al., 2017;Dugo-Cota et al., 2019;Kulemeyer et al., 2009;Sherratt et al., 2019), and less so in invertebrates (Bush et al., 2006;Malcicka et al., 2017).
This is due to the sampling limitation of selecting homologous 2D, or 3D landmarks in all samples (Zelditch et al., 2012). Firm homologous structures like skeletons seem to be easier to compare, but soft-bodied invertebrate taxa pose a problem in this respect.
In the current study, the phylomorphospace approach shows already multiple presumed convergent evolutionary pathways in a limited phylogenetic framework. The presented data suggest that host-switching events could have had an important role in the evolution of carapace shapes of non-pinnotherinae pea crabs, moving from a tube/burrow-dwelling biology to a strictly endosymbiotic lifestyle within bivalves, in adult crabs. Within the pinnotherinae, however, host switches between phyla seem to have had almost no effect on the evolution of carapace shape. This suggests that a shift in lifestyle from ecto-to endosymbiotic could be the driver for carapace (and overall body) shape diversification, rather than betweenphyla host switches. This might also be the case in other symbiotic crustacean taxa with similar evolutionary switches in their lifestyles.
For example, various palaemonid shrimp lineages have evolved from having a free-living lifestyle to a life in symbiosis with an invertebrate host (e.g., Frolová et al., 2022). In addition, some lineages have had multiple between-phyla host switches, some resulting in a shift from ecto-to endosymbiosis (Chow et al., 2021;Horká et al., 2016).
These evolutionary pathways resulted in a wide range of morphological adaptations, ranging from changes in the morphology of the walking legs, eyes, and overall carapace shape (Dobson et al., 2014;Fransen, 1994Fransen, , 2002. Although not studied in detail, symbiotic amphipods from the family Leucothoidae (inhabiting coral rubble, but also bivalve, ascidian, and sponge hosts) might also have diversified in a similar matter (e.g., White, 2011). Lastly, the extremely specious copepod order Harpacticoida has had multiple shifts from a freeliving life to a commensal or parasitic ecto-or endosymbiosis. A wide range of vertebrate and invertebrate hosts are utilized by these copepods, which is possibly the driver for their body shape diversification (e.g., Huys, 2016).

ACK N OWLED G M ENTS
The

FU N D I N G I N FO R M ATI O N
This project was funded by Naturalis Biodiversity Center (Leiden, The Netherlands).

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that supports the findings of this study are available in the supplementary material of this article.