Insight into the population structure of hardhead silverside, Atherinomorus stipes (Teleostei: Atherinidae), in Belize and the Florida Keys using nd2

Abstract Little is known about the natural history, biology, and population genetic structure of the Hardhead Silverside, Atherinomorus stipes, a small schooling fish found around islands throughout the Caribbean. Our field observations of A. stipes in the cays of Belize and the Florida Keys found that populations tend to be in close association with the shoreline in mangrove habitats. Due to this potential island‐based population structuring, A. stipes represents an ideal system to examine questions about gene flow and isolation by distance at different geographic scales. For this study, the mitochondrial gene nd2 was amplified from 394 individuals collected from seven different Belizean Cays (N = 175) and eight different Floridian Keys (N = 219). Results show surprisingly high haplotype diversity both within and between island‐groups, as well as a high prevalence of unique haplotypes within each island population. The results are consistent with models that require gene flow among populations as well as in situ evolution of rare haplotypes. There was no evidence for an isolation by distance model. The nd2 gene tree consists of two well‐supported monophyletic groups: a Belizean‐type clade and a Floridian‐type clade, indicating potential species‐level differentiation.

Although it is extremely difficult to identify all of the factors that impact the dispersal of a species, an insight into the history, evolution, and phylogeography allows for a greater understanding of the complex population dynamics that drive the genetic structuring of populations (Tipton, Gignoux-Wolfsohn, Stonebraker, & Chernoff, 2011). This scientific perspective is essential for proper conservation management of marine taxa, especially for organisms that inhabit vulnerable environments undergoing rapid disturbance, fragmentation, and destruction.
The aim of this study is to analyze the phylogeography of the Hardhead Silverside, Atherinomorus stipes (Müller & Troschel, 1848;Figure 1a), within and between island-groups found off the coast of Belize and along the Florida Keys. Atherinomorus stipes belongs to the family Atherinidae, commonly known as true silversides, within the order Atheriniformes, which consist primarily of ecologically important surface foragers found throughout temperate and tropical regions (Bloom, Unmack, Gosztonyi, Piller, & Lovejoy, 2012). This species, a planktivore that feeds equally in sea grass and mangrove habitats (Vaslet et al., 2015), is one of the most abundant fishes found in close association with mangrove communities throughout the Caribbean (Vaslet, Bouchon-Navaro, Charrier, Louis, & Bouchon, 2010;. Despite their ubiquity, A. stipes has received very little scientific attention. Their phylogenetic placement, both within the family Atherinidae and within the Atheriniformes, has been debated for many years and remains unresolved (Dyer & Chernoff, 1996;Aarn & Ivantsoff, 1997;Sparks & Smith, 2004;Bloom et al., 2012;Near et al., 2012;Betancur-R et al., 2013;Sasaki & Kimura, 2014;Campanella et al., 2015).
We examine the population structure of A. stipes in two island-groups: the Belizean Cays and the Florida Keys. The close association of A. stipes to mangrove shores is ideal for examining population structuring within and between island-groups because of the potential for the restriction of gene flow due to habitat heterogeneity, fragmentation of suitable habitat, and distance among islands. We sequence the mitochondrial gene nd2 in order to address the following questions: (i) are the Belize Cays and Florida Keys island-groups genetically homogeneous and (ii) are populations within each island-group homogeneous? We predict that populations of A. stipes exhibit genetic structuring both within and between the island-groups. We also predict that the isolation by distance (IBD) model will explain genetic divergence in relation to geographic distance.

| Sample collection
The range of A. stipes has been described as Caribbean-wide, but an explicit habitat characterization is lacking (Chernoff, 2002;Vaslet, Bouchon-Navaro, Charrier, et al., 2010;. Based on observations made in the field, we note that the ideal habitat for A. stipes was within close proximity to mangrove roots in combination with a sandy bottom substrate and the presence of turtle grass (Figure 1). Individual islands in Belize and Florida were chosen for sampling based on the presence of this described habitat, their geographic distance to other sampled islands, and the accessibility for collection (Figures 2 and 3); the latitude and longitude of each sampling location is listed in Table A1. Approximately 100 μl of extracted DNA was produced per sample.
Final DNA concentration for all samples was determined on a Thermo Scientific NanoDrop™ ND-2000 1-position spectrophotometer.
The mitochondrial gene nd2 was amplified and analyzed for this study. Mitochondrial genes are useful in analyzing the matrilineal relationships between populations and closely related species due to their high variability within species (Avise, 2000). A ~1,200 base-pair (bp) fragment of nd2 was amplified using the GLN and ASN primers obtained from Kocher et al., 1995. PCR parameters followed the protocol of Tipton et al. (2011). Seven μl of PCR product mixed with 1 μl of Gel Loading Dye was run at 100 V for 30 min in a 1.5% agarose gel with 5 μl of SYBRsafe (Invitrogen). Samples with a visible band ~1,200 bp in length were purified in each primer direction following the Exo-AP PCR product purification protocol described by the DNA Analysis Facility on Science Hill at Yale University for Standard Service Sequencing; all samples were shipped and sequenced at this facility.
The forward and reverse sequences were aligned using ClustalW multiple alignment in BioEdit v7.1.7 (Hall, 1999)  F I G U R E 3 Haplotype frequency within Floridian populations. Sampling locations are indicated by the blue dot and accompanied by the site code and a pie graph of their haplotypes. Pie graphs display the frequency of haplotypes of nd2 within each population. Each pie graph is color coded to display the haplotypes found in that population. The blue portions of each graph represent the universal haplotype, and the red portion of each pie graph indicates the proportion of population-specific haplotypes. Other colors are haplotypes shared among some populations. Code designations can be found in Tables 1 and A1 Miya et al., 2003). The curated sequences were deposited in GenBank under accession numbers MF924405-MF924566.

| Phylogenetic analyses
The model of best fit of sequence evolution for all haplotypes was GTR + I based on AIC and BIC indices (jModelTest v2.1.4; Guindon & Gascuel, 2003;Darriba, Taboada, Doallo, & Posada, 2012 (Clement, Posada, & Crandall, 2000) was used to create a statistical parsimony haplotype network with a connection limit set to 95% for both the Floridian and Belizean island-groups.
Analyses of molecular variance (AMOVA) was conducted in Arlequin 3.5.1.2 (Excoffier & Lischer, 2010 were also generated in Arlequin. When genetic structuring is observed, the role of the IBD model of evolution can be assessed (Wright, 1943). An analysis of IBD allows for an evaluation of both dispersal and the amount of gene flow that is occurring between populations (Puebla, Bermingham, & Guichard, The table displays the sample size, total number of haplotypes, the total number of population-specific haplotypes, the haplotype diversity (Hd), the nucleotide diversity (π), and the average number of mutations between haplotypes within each island population sample. The sample sizes reflect successfully sequenced individuals minus four highly differentiated Florida fish.
2009). Evidence for the IBD model of evolution within each islandgroup was tested by performing a linear regression analysis, using both log and standardized scale transformations in the R Stats Package (R Core Team 2000) and a Mantel permutation test in Arlequin using 10,000 randomized replicates to calculate statistical significance (Mantel, 1967  and the value of π was 0.00194 (0.00148 ≤ π ≤ 0.00304; Table 1).

| Genetic diversity and differentiation
Overall, Hd, π, and the average number of mutations between haplotypes were greater within the Belizean island-group than within the Floridian island-group.
A total of 58 haplotypes were observed within the Belizean islandgroup and 104 haplotypes within the Floridian island-group. Among the 58 haplotypes identified in the Belizean island-group, eight haplotypes were shared by more than one population as follows: (i) three haplotypes (1, 5, and 14) were "universal," defined as being shared among all populations; (ii) haplotype 2 was shared among five populations; (iii) haplotype 31 was found in three populations; and (iv) and three haplotypes (25, 26, and 30) were found in two populations ( Figure 2). Only one individual was captured at location B-TC, and it possessed one of the "universal" haplotypes (haplotype 5). The remaining 50 haplotypes were classified as "unique," which we defined as being a populationspecific haplotype 1 (Figure 2). The total number of haplotypes within each population ranged from 11 to 16, with six to 12 of these haplotypes deemed "unique" to a population (Table 1). Approximately 50% of individuals within each population expressed one of the three universal haplotypes. Approximately 20-30% of individuals within each population exhibited a "unique" haplotype ( Figure 2).
Among the 104 total haplotypes identified in Florida, nine were distributed as follows: (i) haplotype 62 was "universal"; (ii) haplotype 65 was shared among six island populations; (iii) haplotype 66 was observed among four island populations; and (iv) six haplotypes (haplotypes 71, 77, 82, 93, 113, and 124) were found in two populations. The remaining 95 haplotypes were "unique" (Figure 3; Table 1). The total number of haplotypes observed in each population ranged from eight to 18. The population from the Atlantic side of Marathon Key (F-MK) consisted of only nine individuals that possessed eight haplotypes. Of these, six were "unique" to F-MK. In all other sampled populations, which consisted of approximately 30 individuals each, the fewest number of total haplotypes observed was 11, and the number of "unique" haplotypes ranged from eight to 16 (Table 1). The "universal" haplotype was found in 22-52% of individuals within all populations ( Figure 3). The percentage of individuals with "unique" haplotypes within each population ranged between 35% and 67% ( Figure 3).
All tests of neutrality produced negative values for all populations (Table A2). Values of Tajima's D for all Florida populations, with exception of F-MK, were statistically significant (p < .05), while none of the Belizean populations were found to be statistically significant (  Table A2).
The variation between the Belizean and Floridian island-groups was found to be highly significant (p < .00001; Table 2). Additionally, a global AMOVA comparing the populations within Belize indicated that the among-groups variance was also highly significant (p < .00001; Table 2). However, the sources of variation attributable to the "among populations within groups" and "within populations" categories were not significant (Table 2). Within Florida, the variation among and within populations was highly significant (p < .00001 and p = .02444, respectively; Table 2). There were no significant differences (p > .078) between populations on the Gulf of Mexico side of the Florida Keys versus those populations on the Atlantic side (Table 2).

| Relationships among haplotypes
The Within the Floridian-type/Barbadian-type clade, haplotypes from the Florida Keys and haplotypes from Barbados formed two separate, well-supported monophyletic groups ( Figure 4). The genetic divergence between the Floridian-type and Barbadian-type haplotypes was approximately 2.3%. The Belizean-type haplotypes formed a well-supported monophyletic group sister to the Floridian-type/Barbadian-type clade ( Figure 4). Surprisingly, four fish sampled from Floridian populations had haplotypes that were highly divergent from all other Floridian fish and were nested within the Belizean-type haplotype clade (Figure 4).
The statistical parsimony haplotype networks for Belizean-type haplotypes and Floridian-type haplotypes differed greatly in their structures. The Belizean-type haplotype network exhibited a more complex structure with four major "universal" haplotypes found in varying frequencies (N = 14-52; Figure 5a). These four "universal" haplotypes were separated by one or two base-pair substitutions from the majority of minor haplotypes, with the maximum number of steps from a major a haplotype being seven (Figure 5a). There was no  Figure 5a). These three haplotypes differed by four-to-six base-pair substitutions from two of the major Belizean haplotypes (Figure 5a).
In contrast, the Floridian-type haplotype network exhibited a classic "starburst" pattern ( Figure 5b). Starbursts consist of a single common haplotype with numerous minor haplotypes that are one or two base pairs removed from this common haplotype (Shields & Gust, 1995;Grant & Bowen, 1998;Avise, 2000). The common haplotype, termed "universal" in this study, found in Florida was observed in 85 individuals among all Florida populations in nearly equal proportion (Figures 3 and 5b). The vast majority of minor haplotypes, which were predominately classified as "unique," were only observed at a frequency of one or two total individuals.

| DISCUSSION
Haplotypes of nd2 from A. stipes were highly divergent (ca. 4.5%) between the Belize Cays and Florida Keys island-groups. The two specimens from Barbados, the type locality of A. stipes, formed the sister group to the Floridian clade ( Figure 4). The divergence between the sister lineages was 2.3%. Although the clades of silversides from

| Potential speciation
The most striking result of this study was the surprisingly large degree of divergence between the major haplotypes of nd2 in Belize and the Florida Keys. These island-group clades were found to be distinct and highly significant (AMOVA; p < .00001) with a genetic divergence of 4.5% (Figure 4). Additionally, the Florida and Barbados populations differed by 2.3%. These levels of intraspecific divergence are considered to be relatively large and could indicate that A. stipes from within each of these island-groups represent independent evolutionary lineages (Gomes, Pessali, Sales, Pompeu, & Carvalho, 2015).
Evolutionary forces acting on isolated gene pools can result in rapid genetic differentiation and potential speciation (Barraclough, 1998;Puebla et al., 2009;O'Leary et al., 2016). The varying degrees of differentiation among the haplotypes from Belize, Florida Keys, and Barbados suggest that there was restricted gene flow among certain island-groups. Although the type locality of A. stipes (Müller & Troschel, 1848) is Barbados, confidence in the application of this species name requires a comprehensive analysis across its geographic range.
While the use of a single mitochondrial gene to analyze population structure is limited due to potential discordance with other gene trees, it represents an important starting point for examining the phylogeographic patterns of this species (Degnan & Rosenberg, 2009).
The identification of species boundaries based upon morphology can underestimate biodiversity throughout the marine realm (Knowlton, 2000). The evolution of independent genetic lineages that create cryptic biodiversity has important conservation implications because current management practices may not protect each discrete, genetic stock (Beheregaray & Sunnucks, 2001).
The haplotypes of Belizean populations exhibited a complex pattern that included several connected starbursts ( Figure 5). This pattern is similar to the haplotype network observed in several populations of A. endrachtensis within isolated marine lakes in Palau (Gotoh et al., 2011). The nonsignificant values of Tajima's D, Fu's F S , Fu and Li's D*, and Fu and Li's F* (Table A2) for most of the Belizean populations suggest that nd2 was evolving neutrally and did not depart from the genetic drift mutation equilibrium (Tajima et al., 1998;Fu, 1997;Fu & Li, 1993;Gotoh et al., 2011).
A recent bottleneck event or demographic crash results in significant, negative values of Tajima's D and Fu's F S due to the excess of rare alleles that arise within the population during its recovery and subsequent expansion (Depaulis et al., 2003). Because Tajima's D and Fu's F S have greater statistical power for detecting more recent events, these statistics may not illuminate older demographic crashes (Depaulis et al., 2003). This could explain the observed discrepancy between the neutrality tests for the Belizean and Floridian island-groups.
Additionally, the degree of habitat disturbance or destruction can markedly affect the genetic structure of fish populations, resulting in deviations from neutral evolutionary processes (Shulman & Bermingham, 1995;Fauvelot et al., 2003;Gonzalez et al., 2016).
The mangrove habitat found along the Florida Keys has been highly disturbed by coastal development. The degraded and discontinuous state of these habitats may serve to fragment Floridian populations. This is in stark contrast to the majority of observed mangrove habitats within the protected waters off the coast of Belize.

| Population-level differences and gene flow within island-groups
Within each island-group, populations differed significantly from each other and had a high percentage of "unique" haplotypes (Figures 2 and   3). "Unique" haplotypes comprised between 20% to almost 70% of the total haplotypes within each island population (Figures 2 and 3).
Nonetheless, each population contained relatively high percentages of common haplotypes.
There are several ways that the pattern of genetic similarities To what extent has gene flow been occurring among islands?
Is the gene flow historic or contemporary? Despite the genetic heterogeneity among populations, the data do not suggest that populations have been completely isolated. The results are consistent with the inference that there has been gene flow among populations because of the relative frequencies of the common haplotypes. The common haplotypes are most parsimoniously interpreted as older in origin than the rare, "unique" haplotypes. The common haplotypes may represent ancestral or founder haplotypes (Templeton & Sing, 1993;Crandall, 1996;Avise, 2000;Gotoh et al., 2011;Tipton et al., 2011). The life history characteristics of A. stipes may serve to fragment this species and reduce gene flow (discussed below). Fragmented populations tend to evolve more rapidly due to higher levels of genetic drift (Barraclough, 1998;Puebla et al., 2009;O'Leary et al., 2016), and thereby explain the very high percentages of "unique" haplotypes within populations.

| Isolation by distance
Isolation by distance predicts that genetic similarity is inversely proportional to geographic distance among populations. Evidence for IBD in marine systems is relatively rare and can vary widely due to the spatial scales of the sampling locations, the population density, and the dispersal capabilities of marine species (Puebla et al., 2009) (Meirmans, 2012). However, a Mantel test may result in an erroneously significant p-value because it cannot discriminate among a host of alternative spatially structured models, such as IBD and geographic clustering (Meirmans, 2012;Guillot & Rousset, 2013). Because of this, and because the linear regression analysis was nonsignificant, there is insufficient evidence to conclude that the Belizean populations fit the IBD model.

| Life history traits
The dispersal capabilities of this species, whether by active swimming or drifting in ocean currents, directly impact gene flow within and among island-groups. The dispersive ability of A. stipes, however, has never been explicitly studied, although it is likely that it exhibits life history traits that are homologous to closely related atherinid species (Takemura, Sado, Maekawa, & Kimura, 2004;Francisco et al., 2009;Gotoh et al., 2011;Mazlan et al., 2012). These atherinid species display various ecological characteristics that result in reduced dispersal capabilities, including adhesive demersal eggs that attach to vegetation, a short larval stage with well-developed larvae, and a strict association with coastal environments (Takemura et al., 2004;Francisco et al., 2009;Gotoh et al., 2011). The eggs of A. stipes from Belize and Florida in our samples had filaments (B Chernoff, pers. obs.).
The pelagic larval state of marine fishes can be rather difficult to track due to their small size and the spatial scale of distribution (Mora & Sale, 2003;Cowen, Gawarkiewicz, Pineda, Thorrold, & Werner, 2007). However, as Shulman and Bermingham (1995) concluded, ocean current patterns and length of the pelagic larval phase may have the greatest influence on marine fish dispersal and population connectivity. Although little is known about the dispersal abilities of A. stipes, the ocean currents flow to the north and east from Belize (Shulman & Bermingham, 1995) and may help explain the presence of Belizean haplotypes in four individuals from the southern Florida Keys.
Additionally, high levels of larval dispersal among populations may diminish signals of IBD at smaller spatial scales over time (D'Aloia et al., 2014). The lack of barriers to larval dispersal may explain why geographic distance did not correlate with the amount of genetic differentiation at the island-group scale. Despite this difficulty, it is important to determine the dispersal potential of A. stipes in order to elucidate the biological mechanisms that facilitate gene flow.

| CONCLUSION AND FUTURE WORK
The Belizean and Floridian island-groups were found to be highly divergent from each other. Within island-groups, populations exhibited high numbers of haplotypes and differed significantly from one another though there was no or insufficient evidence for IBD.
Populations within both island-groups were characterized by high percentages of shared haplotypes and high percentages of rare but "unique" haplotypes. Two potential hypotheses were discussed that treat the common haplotypes as evidence of gene flow among popu- The close association of A. stipes to mangrove habitats throughout the Caribbean presents an ideal opportunity to examine the potential influences of habitat fragmentation on intraspecific genetic structuring within and among island-groups. Mangrove ecosystems are critical habitats because they serve as nurseries, feeding grounds, and shelters for many marine organisms (Laegdsgaard & Johnson, 2001). The destruction of mangrove communities has occurred at a staggering rate, as approximately one-third of the world's mangroves have disappeared over the last 60 years (Alongi, 2002;Hamilton & Casey, 2016).
This was caused by a variety of anthropogenic factors, such as aquaculture, agriculture, industrial and residential development, forestry uses, and recreational planning (Ellison & Farnsworth, 1996;Valiela, Bowen, & York, 2001). Regardless of whether the mangroves are removed, local disturbances that increase sedimentation and water turbidity negatively affect foraging species, such as A. stipes, which rely heavily on visual cues to search for food in the water column (Thresher, 1983;Vaslet, Bouchon-Navaro, Charrier, et al., 2010;. Furthermore, the alteration and elimination of mangrove habitats may contribute to population isolation and extinction.
As this study only analyzed a single mitochondrial gene, there were limitations on the conclusions that we were able to make.

CONFLICT OF INTEREST
None declared.

AUTHORS CONTRIBUTION
All authors equally contributed to the preparation of this manuscript. 1 We acknowledge that if sample sizes were to get very large, the probability of finding a population-specific or "unique" haplotype in another population increases in proportion to the degree of gene flow among populations.
Insight into the population structure of hardhead silverside,