A biogeographical profile of the sand cockroach Arenivaga floridensis and its bearing on origin hypotheses for Florida scrub biota

Abstract Florida scrub is a xeric ecosystem associated with the peninsula's sand ridges, whose intermittent Pliocene–Pleistocene isolation is considered key to scrub endemism. One scrub origin hypothesis posits endemics were sourced by the Pliocene dispersal of arid‐adapted taxa from southwestern North America; a second invokes Pleistocene migration within eastern North America. Only one study to date has explicitly tested these competing hypotheses, supporting an eastern origin for certain scrub angiosperms. For further perspective, we conducted a genetic analysis of an endemic arthropod, the Florida sand cockroach (Arenivaga floridensis), with two aims: (1) to reconstruct the peninsular colonization and residence history of A. floridensis and (2) determine whether its biogeographic profile favors either origin hypothesis. We sequenced the cox2 mitochondrial gene for 237 specimens (65 populations) as well as additional loci (cox1, nuclear H3) for a subset of Florida roaches and congeners. Using Network and Bayesian inference methods, we identified three major lineages whose genetic differentiation and phylogeographical structure correspond with late Pliocene peninsula insularization, indicating Arenivaga was present and broadly distributed in Florida at that time. Stem and crown divergence estimates (6.36 Ma; 2.78 Ma) between A. floridensis and western sister taxa span a period of extensive dispersal by western biota along an arid Gulf Coast corridor. These phylogeographical and phylogenetic results yield a biogeographic profile consistent with the western origin hypothesis. Moreover, age estimates for the roach's peninsular residence complement those of several other endemics, favoring a Pliocene (or earlier) inception of the scrub ecosystem. We argue that eastern versus western hypotheses are not mutually exclusive; rather, a composite history of colonization involving disparate biotas better explains the diverse endemism of Florida scrub.


| INTRODUCTION
Florida scrub is a fragmented xeric ecosystem largely confined to peninsular Florida (Menges, 1999), where it is partitioned across a series of relict beach ridges that formed sequentially during the Miocene, Pliocene, and Pleistocene epochs (Scott, 1997; Figure 1). Older ridges occupy the central peninsula, whereas the youngest corresponds roughly with present shorelines (Opdyke, Spangler, Smith, Jones, & Lindquist, 1984). Following inception, most ridges experienced rounds of inundation and isolation associated with sea level fluctuation (Webb, 1990). Today, these ridges retain scrub and related sandhill ecosystems, which were considered to have been widespread on the peninsula in the late Pleistocene (Myers, 1990) but experienced significant contraction to the ridges proper under more mesic conditions of the Holocene (Watts & Hansen, 1994).
Florida's sand ridges are characterized by quartzipsamment soils, providing porosity and drainage necessary to support the xeromorphic plant community that defines scrub-a shrubland composed of small evergreen oaks (Quercus chapmanii Sargent, Q. geminata Small, Q. inopina Ashe, Q. myrtifolia Willdenow) interspersed with Florida rosemary (Ceratiola ericoides Michaux) and maintained by low-frequency, high-intensity fires (Menges, 1999). Scrub is inherently patchy, spatially and temporally, and its historical flux, with attendant opportunities for isolation, is considered to have promoted speciation within this ecosystem. Indeed, Florida scrub is distinguished by high levels of endemism, which includes some 40 species of plants (Christman & Judd, 1990), four vertebrates (Moler, 1992;Rodgers, Kale, & Smith, 1996), and over 50 arthropods (Deyrup, 1989). Scrub endemics vary widely in overall distribution, with some species being confined locally within a single ridge and others occupying multiple ridges.
The origin of Florida scrub biota has been variously ascribed to Pliocene or Pleistocene epochs (Hubbell, 1961;Kurz, 1942;Neill, 1957), involving a "combination of both historical and edaphic factors" (Huck et al., 1989). One long-held hypothesis invoking eastward range expansions of arid-adapted biota into peninsular Florida (Myers, 1990) is based on a rich Pliocene fossil record representing numerous extralimital species with western or tropical affinities (Meylan, 1982;Morgan & Emslie, 2010). Western taxa purportedly dispersed along a recurrent Gulf Coast corridor-a broad belt of thorn scrub and savanna established by the increased aridity and lowered sea levels associated with Pliocene glaciation (Morgan & Emslie, 2010). Conversely, interglacial periods reduced corridor width, generated more mesic climatic conditions, and initiated Gulf Coast vicariance with regard to aridadapted species. Sundered from their western lineages, Florida populations underwent allopatric speciation in xeric peninsular settings.
An alternative scrub origin hypothesis, detailed in Germain-Aubrey et al. (2014), invokes southward dispersal events within eastern North America during Pleistocene glacial advances. Under this scenario, eastern taxa colonized the peninsula and, following local adaptation to xeric ridge environments, experienced ecological/spatial isolation sufficient for speciation (Swenson & Howard, 2005;Watts, 1975).
To test the two competing hypotheses, Germain-Aubrey et al.
(2014) generated molecular phylogenies for four angiosperm genera, focusing on the topological placement and sister taxon relationships of respective scrub endemics. Their results were ambiguous for one plant, the scrub plum (Prunus geniculata Harper), but they identified eastern origins for the remaining three species: Lewton's milkwort (Polygala lewtonii Small), scrub holly (Ilex opaca var. arenicola (Ashe) Ashe), and silk bay (Persea humilis Nash). However, age estimates for all four angiosperms dated to the Pliocene (or Miocene), a timeframe inconsistent with a Pleistocene colonization predicated by the eastern hypothesis. Additional reports have offered distributional data supporting eastern biogeographical sources for certain scrub taxa (Huck et al., 1989) and western sources for others (Hubbell, 1961;Zona & Judd, 1986), which suggests that eastern versus western origin hypotheses need not be mutually exclusive. With some 90 endemic species yet to be examined, a prevailing colonization pattern for Florida scrub has yet to be vetted.
Here, we examine the biogeography of a scrub arthropod, the Florida sand cockroach, Arenivaga floridensis Caudell (family Corydiidae). This species is the eastern representative of a genus otherwise distributed in arid settings from central Texas westward into California and Mexico (Hopkins, 2014a). As is characteristic of the genus, A. floridensis is fossorial and sexually dimorphic (females are wingless; Figure 1). Inhabiting scrub as well as adjacent sandhills communities, the species requires patches of open sand (Deyrup, 1994) and demonstrates a strong preference for loose substrate beneath light leaf litter of sand live oaks, Quercus geminata (Lamb, Justice, & Justice, 2006). Males occasionally engage in low, erratic flight at dusk, but juveniles and females appear to be completely fossorial (Deyrup, 1994). With populations documented from 11 peninsular ridges, A. floridensis ranks as Florida scrub's most geographically widespread faunal endemic (Lamb, Justice, & Justice, 2006).
The overall geographic distribution of Arenivaga appears consistent with expectations of the western origin hypothesis, underscoring F I G U R E 1 Map showing the Florida peninsula's sand ridge system, with sampled ridges depicted in color, and images of male and female (wingless) Arenivaga floridensis, illustrating species sexual dimorphism the potential of A. floridensis to provide additional biogeographical perspective on scrub colonization and endemism. Support for the western origin hypothesis (with concomitant rejection of the eastern origin hypothesis) would require A. floridensis to meet the following predictions: (1) Pliocene colonization of the Florida peninsula, (2) a derived topological placement within a phylogeny of the genus (its basal placement could potentially negate a western origin of Arenivaga), and (3) a nodal divergence estimate for A. floridensis and its sister species approximating a late Miocene-early Pliocene timeframe. To distinguish between hypotheses, we present a detailed intraspecific phylogeography for A. floridensis in conjunction with a phylogeny for the species and selected congeners. We compare spatiotemporal patterns of our results with the aforementioned predictions to assess support or refutation of the western origin hypothesis.

| Sampling and sequencing regimens
To examine genetic variation in Arenivaga floridensis, we pursued a dense intra-and inter-ridge sampling survey that yielded 237 roaches representing 65 localities throughout the species' range (Appendix).
Most specimens were captured by sifting sand samples through a two-tier wire-mesh (7.0 and 3.0 mm 2 ) sieve, which retains all but the smallest nymphs. Roaches were preserved in 95% ethanol, and a rear leg of each was processed for genomic DNA using Qiagen's DNeasy kit. We selected the mitochondrial gene cytochrome oxidase II (cox2) as our initial genetic marker to assess population divergence and phylogeographical structure, using primers and amplification parameters listed in Table 1. Amplicons were cleaned using exoSAP-IT (USB Corp.) prior to assay on an Applied Biosystems 3130 capillary sequencer. Resulting sequences, edited and assembled in SEQUENCER 4.9 (GeneCodes, Ann Arbor, MI), were aligned in CLUSTALX ver. 2.0 (Larkin et al., 2007) and translated to ensure correct reading frames.
We combined cox2 data for the western species and 32 A. floridensis (representing all major and minor ridges in our sample) with sequence data for two additional loci, mitochondrial cytochrome oxidase I (cox1) and nuclear histone 3 (H3), amplified as detailed in Table 1. Arenivaga erratica served as the outgroup for the cox2 dataset; Eupolyphaga sinensis (Walker) and Ergaula capucina (Brunner von Wattenwyl), representing additional corydiid genera, were outgroup taxa for multilocus analyses.

| Phylogeographical and phylogenetic analysis
To examine phylogeographical structure in A. floridensis, we constructed cox2 haplotype networks using the TCS algorithm (Clement, Posada, & Crandall, 2000) implemented in POPART (Leigh & Bryant, 2015). We also generated phylogenetic networks in SPLITSTREE v4 (Huson & Bryant, 2006) using the NeighborNet approach with default parameters. We used Bayesian inference (BI) analysis to estimate phylogenetic relationships within A. floridensis (cox2 dataset) and, subsequently, among species (multilocus dataset). Best-fit codon partitioning schemes and nucleotide substitution models were selected using the Bayesian in- sampling trees every 1,000 generations. The first 25% of the posterior distribution were discarded as burn-in. Likelihood values for postanalysis trees and parameters were evaluated for convergence using the MRBAYES "sump" command and the program TRACER v. 1.6 (http://evolve.zoo.ox.ac.uk/software.html?id=tracer). We used the MRBAYES "sumt" command to generate a majorityrule consensus tree and calculate posterior probabilities (PP) for consensus nodes. with the first 10% of each run discarded as burn-in. Convergence of parameters was accessed using TRACER, and trees were summed using TREEANNOTATOR.
In light of potential issues with concatenated gene data, we also produced a time-calibrated species tree using StarBEAST2 (Ogilvie, Bouckaert, & Drummond, 2016) in BEAST2 v2.4.5 (Bouckaert et al., 2014). Site models, clock models, and topologies were unlinked between the mitochondrial and nuclear loci; ploidy assignments for cox1 and cox2 were designated haploid, and H3 was designated diploid. The population model was set as "Analytical Population Size Integration." Topologies and divergence times were estimated as in BEAST analysis above, using the cox2 calibration rate.

| Haplotype distribution, phylogeography, and phylogenetics
We obtained a 684-bp fragment-effectively, the entire cox2 genefor all 237 specimens; sequence comparisons revealed 91 unique haplotypes (Appendix). Haplotype distribution was extremely localized, with 88 of the 91 haplotypes (97%) being observed at single localities. The three haplotypes (Fp2, SB1, Vi) representing more than one locality were limited to proximate sites 2-12 km apart. No haplotypes were shared between ridges despite the fact that certain inter-ridge populations were within 5 km of each other. Such localized spatial distribution, in part a function of scrub patchiness, may also reflect the species' limited vagility: Arenivaga floridensis is restricted to friable, sandy soils (Deyrup, 1994), and the wingless females may impose additional dispersal constraints.
The BI consensus tree for cox 2 identified three strongly supported lineages (PP > 0.99), with evident phylogeographical structure observed within and across lineages (Figures 2 and 3). The largest lineage, genetically and geographically, is designated the Atlantic-Central/Lake Wales clade. These three major lineages were also identified in both SPLITSTREE ( Figure 3) and TCS haplotype networks (not illustrated), the latter providing finer details of phylogeographic structure less pertinent to the predictions being tested.
The three lineages were again recovered (all PP = 1.0) in the concatenated BI analysis but with higher nodal support for the Atlantic-

| A western origin for Arenivaga floridensis
Arenivaga comprises a desert-dwelling clade of roaches distributed largely within southwestern North America. The genus is now recognized as being far more species rich (48 vs. 9 spp) than traditionally perceived, with its highest levels of diversity centered in northern Mexico (Hopkins, 2014a (Foster, Lunt, & Parrish, 2010) that may have elicited earlier southward migrations, which would provide greater temporal consistency with observed genetic differentiation between scrub endemics and their sister taxa. We have provided the first molecular phylogenetic evidence used to test and confirm the western origin hypothesis for a Florida scrub endemic and have compiled additional molecular phylogenies identifying both eastern and western progenitors to scrub species. These cases reveal more examples of eastern origin than western, and proximity alone would seem to favor eastern contribution potential over intermittent dispersal from the southwest. Still, we hesitate to endorse a prevailing eastern assembly when the evolutionary histories for most scrub species have yet to be determined. It is clear now, however, that arguments for a single regional source leading to the inception of this ecosystem can be rightly dismissed; rather, a composite biogeographical history involving disparate biotas better explains Florida scrub origins and endemism.

ACKNOWLEDGMENTS
We thank Mark Deyrup, Michael Evans, Michael Justice, Jackson Mosely, and Paul Skelley for their assistance in the field. Kenny Krysko graciously provided the inset map in Figure 2

CONFLICT OF INTEREST
None declared.

AUTHOR CONTRIBUTIONS
TL conceived the project; TJ, PM, TL, and JB collected specimens; TJ and TL generated/analyzed sequence data; MB and JB conducted network and phylogenetic analyses; HH provided material and sequence data for the western Arenivaga; TL wrote the manuscript; MB generated figures and contributed to the text.