Evidence for Miocene overwater colonization in Caribbean Cyrtognatha spiders

Island systems provide excellent arenas to test evolutionary hypotheses pertaining to gene flow and diversification of dispersal-limited organisms. Here we focus on an orbweaver spider genus Cyrtognatha (Tetragnathidae) from the Caribbean, with the aims to reconstruct its evolutionary history, describe its biogeographic history in the archipelago, and to estimate the timing and route of Caribbean colonization. Specifically, we test if Cyrtognatha biogeographic history is consistent with an ancient vicariant scenario (the GAARlandia landbridge hypothesis) or overwater dispersal. We reconstructed a species level phylogeny based on one mitochondrial (CO1) and one nuclear (28S) marker. We then used this topology to constrain a time-calibrated mtDNA phylogeny, for subsequent biogeographical analyses of over 100 originally sampled Cyrtognatha individuals. Our results suggest a monophyletic radiation of Caribbean Cyrtognatha, containing 11 to 14 species that are exclusively single island endemics. Our analyses refute vicariance and instead support an overwater colonization to the Caribbean in mid-Miocene. Having colonized Hispaniola first, Cyrtognatha subsequently dispersed to, and diversified on, the other islands of the Greater, and Lesser Antilles.


Introduction 27
Island biogeography is concerned with colonization and diversification of organisms on islands, 28 including empirical tests of evolutionary hypotheses pertaining to gene flow in dispersal-limited 29 organisms 1,2 . Islands are geographically widespread and diverse, and vary in shapes and sizes, age 30 and geologic origins, and show different degrees of isolation 3 . Darwin already recognized that this 31 combination of attributes makes islands appealing objects of scientific study 4 . Modern biogeography 32 recognizes the interplay among island histories, the specifics of their geography, and various 33 attributes of organisms that inhabit them 5 . 34 Amongst island systems some of the best studied in terms of biogeographic research are Hawaii 6-8 , 35 Galapagos 9-11 , Azores 12-14 , Canary 15-17 and Solomon 18-20 islands, as well as large continental fragments 36 such as Madagascar 21-23 and New Zealand 24-26 . However, the Caribbean island system 27-30 is the 37 single most 'published' island system in biogeography literature (Google Scholar title hits 237  38 compared with 195 for the second, New Zealand). The Caribbean Basin, also known as West Indies, 39 lies in the tropical zone between South and North American continents, and to the east of the Gulf of 40 Mexico. Combining over 700 islands, the Caribbean is considered among the world's biodiversity 41 hotspots [31][32][33] . In its most broad categorization, the Caribbean comprises three regions: (1) Greater 42 Antilles with the largest islands of Cuba, Hispaniola (Dominican Republic and Haiti), Puerto Rico and 43 Jamaica representing 90% of all land in Caribbean Sea; (2) Lesser Antilles with numerous smaller, 44 mostly volcanic, islands and (3) the Lucayan platform archipelago (the Bahamas). 45 The Greater Antillean islands of Cuba, Hispaniola, and Puerto Rico, but not Jamaica, are parts of the 46 old proto-Antillean arc that began its formation in the distant past, over 130 million years ago (MYA). 47 Through Caribbean plate tectonics, the proto-Antillean arc drifted eastward until settling at its 48 current location around 58 MYA 34,35 . Researchers disagree on the timing of the proto-Antillean arc 49 connection with South or North America in the Cretaceous or even on the existence of such a 50 connection. However, that distant past may have had little biological relevance for current biotas due 51 to a catastrophic effect of the bolide that crashed into Yucatan around 65 MYA which arthropods 52 would likely not have survived [36][37][38] , the primitively segmented spiders, family Liphistiidae and the mygalomorph  87  trapdoor spiders, likely do not balloon and have highly sedentary lifestyle imposing strict limits on  88 their dispersal potential. As a consequence, bodies of sea water or even rivers represent barriers that 89 limit their gene flow, which leads to micro-allopatric speciation [75][76][77][78][79][80]

Species delimitation 154
Because the current taxonomy of Cyrtognatha based on morphology is highly incomplete 82 , we 155 undertook species delimitation using CO1 data. To determine the molecular taxonomic operational 156 units (MOTUs), we used four different species delimitation methods, each with its online application: 157 PTP (Poisson tree process) 96 , mPTP (multi-rate Poisson tree process) 97 , GMYC (generalized mixed yule 158 coalescent) 98 and ABGD (automatic barcode gap discovery) 99 . We ran these species delimitation 159 analyses using the default settings, with the input tree for GYMC from BEAST2 100 and the input trees 160 for PTP and mPTP from MEGA 6.0 101 . 161

Phylogenetic analyses 162
We used MrBayes 102 to reconstruct an all-terminal phylogeny for a complete set of our original 163 Cyrtognatha material and outgroups using CO1 (Table 1). For Bayesian analysis we used the 164 Generalised time-reversible model with gamma distribution and invariant sites (GTR+G+I) as 165 suggested by AIC and BIC criterion in jModelTest2 103 . We ran two independent runs, each with four 166 MCMC chains, for 100 million generations, with a sampling frequency of 1000 and relative burn-in 167 set to 25%. The starting tree was random. 168 For a species level phylogeny, we then selected two individuals per MOTU and added 28S sequence 169 data for two partitions and analyzed this concatenated dataset under a Bayesian framework. As 170 above, jModelTest2 suggested GTR+G+I as the appropriate model, this time for both partitions. 171 These analyses had 28 terminals including outgroups (Table 1). The settings in MrBayes were as 172 above, but the number of MCMC generations was set to 30 million. Due to high mutation rates in 173 noncoding parts of nuclear genes like 28S, insertions and deletions accumulate through 174 evolution 104,105 , resulting in numerous gaps in a sequence alignment. We treated gaps as missing data 175 but also ran additional analyses applying simple gap coding with FastGap 106 . 176

Molecular dating analyses 177
We used BEAST2 100 for time calibrated phylogeny reconstruction (chronogram). We used a single 178 CO1 sequence per MOTU and trimmed the sequences to approximately equal lengths. We then 179 modified the xml file in BEAUti 100 to run three different analyses. The first analysis was run using 180 GTR+G+I as suggested by jModelTest2. The second analysis employed the package and model 181 bModelTest 107 . The third analysis used the package and RBS model 108 . All parameters were set to be 182 estimated by BEAST. We calibrated a relaxed log normal clock following Bidegaray-Batista and 183 Arnedo 109 with the ucld.mean set as normal distribution with mean value of 0.0112 and standard 184 deviation of 0.001, and the ucld.stdev set to exponential distribution with the mean of 0.666. We ran 185 an additional analysis using a fossil calibration point on the basal node of Caribbean MrBayes and BEAST analyses were run on CIPRES portal 112 . 197

Ancestral area estimation 198
We used BioGeoBEARS 113 in R version 3.5.0 114 to estimate ancestral range of Cyrtognatha in the 199 Caribbean. We used a BEAST produced ultrametric tree from the above described molecular dating 200 analysis as an input. We removed the outgroup Tetragnatha elongata and conducted the analyses 201 with the 13 Cyrtognatha MOTUs from six areas (Hispaniola, Jamaica, Puerto Rico, Cuba, Lesser 202 Antilles and Panama). We estimated the ancestral range of species with all models implemented in 203 BioGeoBEARS: DEC (+J), DIVALIKE (+J) and BAYAREALIKE (+J). We used log-likelihoods (LnL) with 204 Akaike information criterion (AIC) and sample-size corrected AIC (AICc) scores to test each model's 205 suitability for our data. All of our Cyrtognatha's MOTUs are single island endemics, therefore we 206 were able to reduce the parameter "max_range_size" to two 29,115,116 . 207 208

209
We collected 103 Cyrtognatha individuals from Cuba, Jamaica, Dominican Republic/Hispaniola, 210 Puerto Rico and Lesser Antilles (Fig. 1, Table 1). We confirmed that all individuals are morphologically 211 Cyrtognatha. Most species are undescribed but we identified two known species: C. espanola 212 (Bryant, 1945)  inter-and intraspecific branching patterns, offering us from 1 to 39 MOTUs. 223 The two gene and the CO1 phylogenies yielded nearly identical networks for the Caribbean taxa, but 224 strikingly different phylogenies due to different root placement (Fig. 2, Supplementary Fig. S1). 225 Therefore, we ran an additional analysis constraining the root of the mtDNA phylogeny to reflect that 226 of the two gene phylogeny, with otherwise the same settings (Fig. 1, Supplementary Fig. S2 Caribbean taxa. However, this relationship is not recapitulated in the concatenated, species level, 234 phylogeny (Fig. 2, Supplementary Fig. S3). The concatenated phylogeny also supports monophyly of 235 the Caribbean taxa but recovers the clade of C. espanola and C. SP12 from Hispaniola as sister to all 236 other Caribbean Cyrtognatha. The species level phylogeny is generally better supported, with the 237 exception of a clade uniting species from Lesser Antilles, Cuba, Hispaniola and Puerto Rico. In both 238 Bayesian analyses the chains successfully converged and ESS as well as PRSF values of summarized 239 MCMC runs parameters were appropriate 102 . 240 Chronograms produced by BEAST, using either exclusively CO1 mutation rate or incorporating the 241 additional fossil for time calibration, exhibited very similar time estimates (Fig. 3, Supplementary Fig.  242 S5). We decided to proceed with the mutation rate-only calibrated phylogeny for further analyses 243 because it is less likely to contain known potential biases when calibrating with scarcely available 244 fossils and geological information 117,118 . The molecular dating analyses based on the three different 245 models in BEAST largely agreed on node ages with less than 1 million years variation. bModelTest method of model selection see 107 ). The BEAST chronogram using bModelTest (Fig. 3,  254 Supplementary Fig. S4) yielded the best supported results, amongst the above mentioned 255 approaches, based on ESS values, and was therefore used in subsequent biogeographical analyses. 256 This chronogram (Fig. 3) Table S1). The estimation of ancestral states suggests that the most recent common 264 ancestor of all Caribbean Cyrtognatha in our dataset most likely (62%) resided on Hispaniola (Fig. 4). 265 Moreover, all the Greater Antillean island clades as well as the Lesser Antillean clade most likely 266 originated from Hispaniola with the following probability: Jamaican clade (47%), Lesser Antillean 267 clade (40%), Cuban clade (63%) and Puerto Rican clade (89%) (Fig. 4, Supplementary Table S1). 268 269

Discussion 270
We reconstruct the first Cyrtognatha phylogeny using molecular data from over 100 individuals of 271 this rarely collected group. Our results support Cyrtognatha as a relatively young clade, having 272 diverged from a common ancestor with its sister genus Tetragnatha, in early-to mid-Miocene, and 273 colonized the Caribbean in mid-Miocene. As we discuss below, these estimated ages, combined with 274 the phylogenetic patterns, refute vicariant explanations of their Caribbean origin, including the 275 GAARlandia hypothesis. Instead, the patterns suggest colonization of Hispaniola, and subsequent 276 dispersal to other islands. 277 The all-terminal phylogeny (Fig. 1)  clearly not reflected in our BEAST chronogram (Fig. 3) in which we estimate that the Caribbean 302 Cyrtognatha split from its continental population as late as 13.2 MYA. This suggests that the genus 303 Cyrtognatha is much younger than the most reasonable possible vicariant timeframe. Likewise, the 304 reconstructed biogeographic patterns fail to support vicariant scenarios. First, the Jamaican lineage 305 split from Hispaniola soon after colonization (12.2 MYA) even though Jamaica was never a part of the 306 proto-Antilles. Secondly, Puerto Rico was a part of the proto-Antilles but was colonized only recently 307 (4.8 MYA). The results of Jamaican and Puerto Rican colonization from Hispaniola thus are most 308 consistent with a scenario of colonization by overwater dispersal. 309 The mid-Miocene (ca. 15 MYA) is considered as the start of the modern Earth 136 in that the climate 310 began to stabilize and the ocean currents started to take their current form. This combination of 311 events enabled the colonization of the Caribbean islands from eastern-northern parts of South 312 America for example via vegetation rafts passively drifting with water currents 137 . That also meant 313 that the wind directions and the hurricane paths most likely resembled those of today 138 , from East 314 to West direction 139 . In fact, hurricanes may create numerous dispersal/colonization opportunities, 315 especially for the organisms with poor active dispersal abilities 135,140 . Wind directions and tropical 316 storms are relevant for tetragnathid spiders like Cyrtognatha that disperse by ballooning and could 317 facilitate their colonization of the Caribbean islands in a stepping stone 141 or leap-frog 142 manner. 318 With the examination of the relationships in the time calibrated phylogeny (Fig. 3), we predict that a 319 single colonization of the Caribbean from the continental America occurred sometime between 10.5 320 and 20.7 MYA. The most likely scenario indicates the original colonization of the Greater Antilles 321 (Hispaniola; Fig. 4). Such patterns of colonization of Greater Antilles in Miocene are also evident in 322 many other lineages including vertebrates, invertebrates and plants 29,41,143-150 . A more rigorous test 323 of number and directionality of colonization pathways would require more thorough sampling across 324 potential source populations on the mainland. 325 Our inference of ancestral ranges proposes an early within island diversification of Cyrtognatha 326 ancestors occupying Hispaniola and predict that Hispaniola is the ancestral area for all Caribbean 327 clades (Fig. 4, Supplementary Fig. S5). The path of colonization does not resemble a straightforward 328 pattern such as the stepping-stone pattern. The colonization sequence seems more random or 329 resembles a "leap-frog" pattern. In our case the clear example of island being "leap frogged" is 330 Puerto   Notice that all putative species form exclusively single island endemic pattern. 723 using the most suitable model for our data (DIVALIKE + J, max_range_size = 2), revealed that 732 Hispaniola was most likely colonized first. Following colonization, Cyrtognatha diversified within 733 Hispaniola and subsequently dispersed from there to all other islands of the Caribbean. 734 735