Habitat transitions alter the adaptive landscape and shape phenotypic evolution in needlefishes (Belonidae)

Abstract Habitat occupancy can have a profound influence on macroevolutionary dynamics, and a switch in major habitat type may alter the evolutionary trajectory of a lineage. In this study, we investigate how evolutionary transitions between marine and freshwater habitats affect macroevolutionary adaptive landscapes, using needlefishes (Belonidae) as a model system. We examined the evolution of body shape and size in marine and freshwater needlefishes and tested for phenotypic change in response to transitions between habitats. Using micro‐computed tomographic (µCT) scanning and geometric morphometrics, we quantified body shape, size, and vertebral counts of 31 belonid species. We then examined the pattern and tempo of body shape and size evolution using phylogenetic comparative methods. Our results show that transitions from marine to freshwater habitats have altered the adaptive landscape for needlefishes and expanded morphospace relative to marine taxa. We provide further evidence that freshwater taxa attain reduced sizes either through dwarfism (as inferred from axial skeletal reduction) or through developmental truncation (as inferred from axial skeletal loss). We propose that transitions to freshwater habitats produce morphological novelty in response to novel prey resources and changes in locomotor demands. We find that repeated invasions of different habitats have prompted predictable changes in morphology.

Many fish clades are restricted to either marine or freshwater habitats. However, other fish groups exhibit greater lability of habitat occupancy, with evolutionary reconstructions suggesting multiple independent transitions between marine and freshwater habitats. For example, pufferfishes (Santini et al., 2013;Yamanoue et al., 2011), drums (Lo et al., 2015), herring, longfin herrings, and anchovies (Bloom & Lovejoy, 2012;Bloom & Lovejoy, 2014), sculpins and other cottoid fishes (Buser et al., 2019), stingrays, and needlefishes (Bloom & Lovejoy, 2017) include both marine species and freshwater species distributed across multiple continents. These trans-marine/freshwater clades provide optimal study systems for understanding how habitat shifts alter the adaptive landscape and drive the evolution of ecological novelty and morphological disparity (Davis, Unmack, Pusey, Pearson, & Morgan, 2014).
Needlefishes (Belonidae) are typically elongate piscivorous mesopredators that swim just below the water's surface. They are distributed globally in subtropical and tropical marine, brackish, and freshwater environments, and fossil evidence suggests these fishes have been persistent predators in these waters for 8-10 million years (de Sant'Anna, Collette, & Godfrey, 2013). Several species occur exclusively in freshwater rivers of South America, Central America, and Southeast Asia. They exhibit considerable body size variation, ranging in length from the 5.0 cm freshwater Belonion apodion (Collette, 1966) to pelagic marine species that reach up to 2.0 m, such as Tylosurus crocodilus (Péron & Lesueur, 1821) and Ablennes hians (Valenciennes, 1846) (Collette, 2003). The repeated invasions of freshwater by marine beloniformes on multiple continents, their variation in body size and shape, and putative ecological novelty in riverine habitats (Collette, 1966;Goulding & Carvalho, 1983;Lovejoy & De Araújo, 2000) make them an excellent study system for examining morphological diversification associated with habitat transitions.
Here, we investigated how habitat transitions have affected morphological diversification in needlefishes. We analyzed body shape and size, including functional features such as fin placement, body tapering, and skull shape, and used micro-computed tomography scanning to assess axial skeleton morphology. Our objectives were fourfold: (a) to describe the primary axes of body shape and size variation in needlefishes, (b) to test for differences in morphological diversity between marine and freshwater taxa, (c) to test for differences in rates and patterns of morphological evolution between marine and freshwater taxa, and (d) to determine whether evolutionary transitions between marine and freshwaters alter macroevolutionary adaptive landscapes. Our study demonstrates that needlefishes have experienced divergent selective regimes as a result of habitat transitions.

| Image acquisition and geometric morphometric analyses of body shape
We photographed whole specimens against a foam background using insect pins or held beneath a glass panel to minimize artifacts from warping and twisting. Images of whole needlefishes were landmarked ( Figure 2) using the program tpsDig2 (v. 2.31; Rohlf, 2004

| Computed tomography scanning and axial skeleton meristics
Macroevolutionary changes in body size in fishes frequently involve alterations to the number, spacing, or size of vertebral elements, particularly in slender elongate fishes (Ward & Mehta, 2010

| Phylogeny
For all comparative analyses, we used the time-calibrated phylogeny for Beloniformes from Bloom and Lovejoy (2017), which is based on a multigene dataset (cytb, rag1, rag2, tmo-4c4) of 3,318 base pairs for 104 species and represents the most densely sampled phylogeny for this group. We used the drop.tip function to trim the phylogeny to include only the species in our morphological dataset (n = 31).
We then inferred the evolutionary history of habitat using stochastic character mapping (Bollback, 2006;Huelsenbeck, Nielsen, & Bollback, 2003) with the make.simmap function (phytools; Revell, 2012; Figure 1) for 1,000 trees. The evolutionary history of habitat was reconstructed on 1,000 random trees using the posterior distribution from Bloom and Lovejoy (2017) to account for phylogenetic uncertainty. To assess the best model for the transition matrix, we fitted the following models: (a) a model allowing for equivalent rates of transition for both freshwater and marine lineages ("ER") and (b) a model allowing these rates to vary ("ARD" or "all rates different") using the function ace in the package ape (v. 5.3; Paradis, Claude, & Strimmer, 2004). We then compared the two models (ER vs. ARD) using a likelihood-ratio test and found that the ER was the F I G U R E 1 Trimmed phylogenetic tree from Bloom and Lovejoy (2017)

| Phylomorphospace and adaptive optima analyses
We examined whether marine and freshwater clades overlap in trait space or whether lineages are exploring alternative regions of morphospace. We used the broken stick method to determine the number of informative principal component axes to retain for analyses (screeplot.cca function in the package vegan). We then visualized a belonid morphospace by plotting these remaining PC axes and projected the phylogeny onto species values to form a phylomorphospace (Sidlauskas, 2008), as implemented in phytools (Revell, 2012). Convex hulls were fit to marine and freshwater taxa, separately based on the method of Eddy (1977), using the chull function. We used compare.evol.rates function (from package geomorph; Adams et al., 2016) to determine whether rate shifts in the evolution of body shape are associated with habitat transitions. We iterated this process 5,000 times using phylogenetic simulation, whereby simulated tip data are obtained under Brownian motion using a common evolutionary rate pattern for all species on the phylogeny (Denton & Adams, 2015). From Adams, Collyer, Otarola-Castillo, and Sherratt (2014) "From the data the net rate of shape evolution for each group in the multi-dimensional space is calculated, and a ratio of rates is obtained." Since we only compared between two groups (marine and freshwater), the ratio of the maximum to minimum rate was not used as a test statistic.
We tested three evolutionary models in the package OUwie (v.  OUwie uses complex OU models that cannot always be reliably detected when the statistical power is low (Boettiger, Coop, & Ralph, 2012), and low power can lead to complex OU models being incorrectly favored over models of Brownian evolution (Cooper, Thomas, & FitzJohn, 2016;Ho & Ané, 2014). To determine whether we had significant power to accurately detect the complex models, we performed 1,000 OUwie simulations for max body size and mean vertebral count using the function OUwie.sim. The simulated datasets were performed with the parameter estimates for the best-fit model of each morphological character in our empirical dataset ( Table 2).
The simulated data were then run through all three models in OUwie to determine whether the simulated model could be accurately recovered with our sample size.

| RE SULTS
Belonids have undergone transitions from marine to freshwater habitats six times, with no reversals to the marine environment ( Figure 1).
Correspondingly, we found biased directionality in habitat transitions; transitions from marine to freshwater habitats were almost twice as likely as freshwater to marine (7.205 vs. 4.191 changes).
Geometric morphometric analyses of body shape found differences in body shape among the sampled marine and freshwater taxa.
In most cases, convex hulls for freshwater taxa encompassed greater regions of morphospace than those of marine taxa (Figures 2   and 3 Note: Emboldened rows represent best-fit model based on lowest AICc score. θ fw is the estimated trait optima for freshwater species, θ mar is the estimated trait optima for marine species, α fw is the estimated pull toward the optimal trait value for freshwater species, α mar is the estimated pull toward the optimal trait value for marine species, σ 2 fw is the estimated rate parameter freshwater species, and σ 2 mar is the estimated rate parameter for marine species. Collette, 1966), the Southeast Asian freshwater needlefish Xenentodon cancila (Hamilton, 1822), and Neotropical freshwater Potamorrhaphis (P. eigenmanni Miranda Ribeiro, 1915 and P. petersi Collette, 1974). These freshwater fishes appear to be exploring novel regions of morphospace, and all are notably smaller freshwater taxa.  Results of our simulations for body size show that our dataset has enough statistical power to clearly separate the different multipeak OU models from the models of Brownian motion and the single-peak OU model ( Figure 6). Furthermore, our simulations show that we could accurately recover the estimated theta in most of our simulations. Nevertheless, we can clearly discriminate between single-peak and multi-peak models, as well as recover the correct placement of the optimal trait values ( Figure 6), allowing us to conclude that marine lineages evolved toward a larger body size than freshwater lineages ( Figure 5).

B. dibranchodon
The OUwie analysis on mean vertebral count supported multiple models: the OU model with a single adaptive optimum, a Brownian motion model, and a multi-peak OU model. The failure to differentiate among these models was likely due to lack of statistical power in our dataset; therefore, we do not discuss the results of the model test for mean vertebral count.

| Body shape and habitat
Ecological transitions among habitats clearly shape the adaptive landscape and result in both novel and repeated bauplans, outcomes that support both contingent and deterministic evolution (Blount, Lenski, & Losos, 2018). Freshwater needlefish lineages have both retained ancestral, marine bauplans and evolved radical departures from these same bauplans (e.g., Belonion). As a result, marine and freshwater taxa exhibit overlapping, yet staggered morphospace occupation Sigma Squared (Figures 2 and 3). In addition, considering all phylomorphospace configurations in Figures 2 and 3, freshwater taxa are more morphologically diverse than marine taxa. This demonstrates that habitat transitions have promoted diversification of body shapes and size, as well as faster rates of shape and size evolution in belonids overall perhaps due to release of ecological limits on clade diversification in novel habitats (Betancur-R et al., 2012;Bloom & Egan, 2018).
What are the evolutionary patterns in morphological change associated with habitat transitions? Across a myriad examples of habitat transitions, from marine Antarctic shallows, to tropical reefs and non-reefs, or within African Rift lakes, fishes have evolved along a bentho-pelagic axis, with deeper, laterally compressed bodies associated with complex benthic habitats and more fusiform shapes associated with open water (Hulsey et al., 2013;Rutschmann et al., 2012;Tavera et al., 2018). In contrast, we find that needlefishes in marine and freshwater exhibit niche conservatism because they have not deviated from epipelagic or limnetic habitats, typically cruising just below the water's surface (Goulding & Carvalho, 1983). Instead, we suggest that microhabitat and locomotory demands for either precise maneuvering (most freshwater taxa) or sustained swimming (many marine taxa) are key determinants of body shape evolution in needlefishes and have directed phenotypic novelty (Figures 2 and 3).
Interestingly, phenotypic novelty in freshwater needlefishes evolved independently in different geographic areas. For exam-

ple, freshwater lineages including South American Potamorrhaphis
and Belonion, as well as Southeast Asian Xenentodon, invaded novel regions of morphospace relative to marine taxa and likely in response to open niches in continental rivers (Foster, 1973;Goulding & Carvalho, 1983), as indicated by the phylomorphospace ( Figure 2). Both lineages exhibit an overall shortening of the body relative to marine taxa, while also having rounded or squared caudal fins (Collette, 1966;Foster, 1973), which likely facilitate maneuvering in the highly structured habitats in which they occur, that is, smaller rivers, streams, and wetland habitats.  have pectoral fins that are shifted ventrally relative to marine needlefishes and have larger eyes relative to body size ( Figure 2

| Vertebrae evolution
In contrast to our body size data, we did not detect differences in mean vertebral count (Figure 4) or adaptive optima (Figure 7) between marine and freshwater belonids. This is surprising given the documented correlation between vertebral counts and body size in fishes (Lindsey, 1975;Ward & Brainerd, 2007). Ward and Brainerd (2007) surveyed seven actinopterygian clades and showed that variability in cranial elongation was negligible compared to variability in the axial skeleton for explaining body length. However, Ward and Mehta (2010) reported that body length is often positively correlated with head length. In the case of belonids, we found that changes in absolute body length in belonids can stem from either elongation or truncation of their needle-like jaws, or similar changes to the axial skeleton. The evolutionary and developmental plasticity of skull morphology observed in beloniformes might make the crania more amenable to modification, while changes to the number of axial skeletal elements (centra) appear more static.
An interesting case of potential contrasting mechanisms for the has an elongated lower jaw but short upper jaw (observed in many subadult needlefish species; Lovejoy et al., 2004), and has lost or reduced both axial and appendicular skeletal elements (Collette, 1966).
Overall, these findings suggest strong selection for reduced body sizes in Neotropical freshwater taxa (Weitzman & Vari, 1988) and multiple means by which that selection can effect changes .

ACK N OWLED G M ENTS
We thank T. Buser, C. Harris-Weaver, and J. Wakileh who were instrumental for analyses and data collection.

CO N FLI C T O F I NTE R E S T
The authors declare no conflicts of interest. writing -review and editing (equal).

DATA AVA I L A B I L I T Y S TAT E M E N T
Sequence data are deposited in GenBank and supplemented by data from Bloom and Lovejoy (2017)