PreprintPDF Available

Resolving spatial complexities of hybridization in the context of the gray zone of speciation in North American ratsnakes ( Pantherophis obsoletus complex)

May 2020

May 2020

DOI:10.1101/2020.05.05.079467

License
CC BY-NC-ND 4.0

Authors:

Frank T Burbrink

American Museum of Natural History

Marcelo Gehara

Rutgers, The State University of New Jersey

Estimating species divergence with gene flow has been crucial for characterizing the gray zone of speciation, which is the period of time where lineages have diverged but have not yet achieved strict reproductive isolation. However, estimates of divergence times and gene flow often ignores spatial information, for example the formation and shape of hybrid zones. Using population genomic data from the eastern ratsnake complex ( Pantherophis obsoletus ), we infer phylogeographic groups, gene flow, changes in demography, the timing of divergence, and hybrid zone widths. We examine the spatial context of diversification by linking migration and timing of divergence to the size, shape, and types of hybridization (e.g., F1, backcrosses) in hybrid zones. Rates of migration between lineages are associated with the width and shape of hybrid zones. Timing of divergence is not related to migration rate across species pairs and is therefore a poor proxy for inferring position in the gray zone. Artificial neural network approaches are applied to understand how landscape features and past climate have influenced population genetic structure among these lineages prior to hybridization. The Mississippi River produced the deepest divergence in this complex, whereas Pleistocene climate and elevation secondarily structured lineages.

Available via license: CC BY-NC-ND 4.0

Content may be subject to copyright.

Title: Resolving spatial complexities of hybridization in the context of the gray zone of

speciation in North American ratsnakes (Pantherophis obsoletus complex)

Frank T. Burbrink1*

Marcelo Gehara1,2

Edward A. Myers1,3

1Department of Herpetology, The American Museum of Natural History, Central Park West and

79th Street, New York, NY 10024 USA

2Department of Biological Sciences, Rutgers University Newark, 195 University Ave,

Newark, NJ 07102

3Department of Vertebrate Zoology, National Museum of Natural History, Smithsonian

Institution, Washington, DC, USA

*Corresponding author – fburbrink@amnh.org

.CC-BY-NC-ND 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted May 7, 2020. . https://doi.org/10.1101/2020.05.05.079467doi: bioRxiv preprint

Abstract:

Estimating species divergence with gene flow has been crucial for characterizing the gray

zone of speciation, which is the period of time where lineages have diverged but have not yet

achieved strict reproductive isolation. However, estimates of divergence times and gene flow

often ignores spatial information, for example the formation and shape of hybrid zones. Using

population genomic data from the eastern ratsnake complex (Pantherophis obsoletus), we infer

phylogeographic groups, gene flow, changes in demography, the timing of divergence, and

hybrid zone widths. We examine the spatial context of diversification by linking migration and

timing of divergence to the size, shape, and types of hybridization (e.g., F1, backcrosses) in

hybrid zones. Rates of migration between lineages are associated with the width and shape of

hybrid zones. Timing of divergence is not related to migration rate across species pairs and is

therefore a poor proxy for inferring position in the gray zone. Artificial neural network

approaches are applied to understand how landscape features and past climate have influenced

population genetic structure among these lineages prior to hybridization. The Mississippi River

produced the deepest divergence in this complex, whereas Pleistocene climate and elevation

secondarily structured lineages.

Key Words: Eastern Nearctic, migration, isolation, neural networks, reproductive isolation, cline,

hybrid zone

Wide-ranging species complexes that cross numerous biogeographic barriers provide

opportunities to better understand how changing landscapes affects diversification, gene flow,

demography, and the formation of hybrid zones. Lineages within species complexes ranging

across heterogeneous landscapes are likely at different stages of the speciation process reflecting

how specific environmental and biogeographic barriers have uniquely altered changes in gene

flow and other demographic processes in a particular complex (Myers et al. 2020). Even at the

same barrier within related groups of organisms, the timing of divergence and degree of gene

flow can be species specific and dependent on when they encountered the barrier (Riddle 2016;

Myers et al. 2019). Alternatively, generalist species not as constrained to specific habitats may

not show a correlation between genetic variation and landscape features (Joseph and Wilke 2007;

Makowsky et al. 2009; Lourenço et al. 2017).

Lineage divergence can be placed in the context of the gray zone of speciation, which

defines the range of time where speciation proceeds from early population differentiation with

unfettered gene flow to nearly reproductive isolation defined by low rates of gene flow (de

Queiroz, 2007; Hewitt, 2008; Jackson et al., 2016; Roux et al., 2016). The ability to delimit

species therefore may be correlated with their position in the gray zone, which has previously

been assessed by relating measures of genetic isolation and migration (the genealogical

divergence index; GDI) with species-delimitation probabilities (Jackson et al. 2016; Leaché et al.

2019). Moreover, given a model of speciation that proceeds by Dobzhansky-Muller

incompatibilities and genetic drift alone, the degree of gene flow might be predicted by the age

of lineage formation relative to population size and thus position within the gray zone (Orr 1995;

Gavrilets 2004; Singhal and Moritz 2013). Clearly some bias exists here; enhanced gene flow

between the youngest divergences may never proceed to the later stages in the gray zone as they

may rapidly collapse into a single lineage (Garrick et al. 2019), such as in the case where

physical barriers to gene flow disappear. On the other hand, strong selection may prevent young

lineages from exchanging alleles at particular loci, shortening the time in the gray zone (Mayr

1963; Barton 2010; Feder et al. 2012; Roux et al. 2016; Edwards et al. 2020). Therefore, finding

a threshold that can determine if two species are unique given only generations since divergence

may fail in cases where rates of gene flow between lineages drastically change over time and

space (Gourbière and Mallet 2010; Nosil et al. 2017).

Conceptualizing the gray zone without a spatial demographic component therefore may

be inadequate for understanding speciation and delimiting species. Isolation-migration models

(Hey and Nielsen 2004; Hey 2010) are coalescent estimators of historical processes and can

assess gene flow throughout time, reflecting position in the gray zone of speciation. These

models, however, lack any spatial component necessary to understand the underlying geographic

and environmental processes influencing migration. Studies of hybrid zones on the other hand

can clarify the location of migration and rates of contemporary gene flow, provided selection

does not eliminate F1 hybrids or backcrossing. If species pairs rapidly adapt to unique habitats

across their distribution, then age of divergence and degree of gene flow may be unrelated

relative to neutral processes when compared across taxa (Nosil and Crespi 2006; Agrawal et al.

2011; Nosil 2012; Karrenberg et al. 2019). Therefore, divergence over landscape can show how

selection at a barrier can be discordant from the timing of origin of the lineages (Barton and

Hewitt 1985; Harrison 1993; Jiggins and Mallet 2000; Gay et al. 2008; Seehausen et al. 2014;

Stankowski et al. 2017). Combining spatial information with genetic data will better help

understand how genetic variation is structured by geographic distance, environment, and the

presence of hybrid zones. This should provide a clearer understanding of why lineages maintain

particular rates of gene flow along a biogeographic barrier that has resulted in a hybrid zone.

Moreover, investigating gene flow in space determines if lineages were generated via isolation

and migration or are simply artificially partitioned groups with continuous gene flow over the

landscape defined by IBD (Wright 1943; Frantz et al. 2010; Bradburd et al. 2018).

To understand the factors shaping reproductive isolation in a species complex, we ask

whether divergence time predicts rates of gene flow or cline width across space in eastern

100

ratsnakes (Pantherophis obsoletus complex). This complex is a wide-ranging group of four

101

species found throughout the forested regions of the Eastern Nearctic and nearby Chihuahuan

102

Desert that have diverged at three unique biogeographic barriers. The primary divergence in this

103

complex occurred at the Mississippi River, which has consistently been identified as one of the

104

main biogeographic barriers in the ENA (Robison 1986; Burbrink et al. 2000; Soltis et al. 2006;

105

Brandley et al. 2010; Zellmer et al. 2012; Myers et al. 2020). Further east, the Appalachian

106

Mountains and Apalachicola/Chattahoochee River System (AARS) have noted barrier effects in

107

other organisms and likely contributed to the divergence between P. alleghaniensis and P.

108

spiloides (Walker and Avise 1998; Burbrink et al. 2000; Soltis et al. 2006). West of the

109

Mississippi River, divergences occurred at the transition between temperate forests to the rocky

110

areas on the western edge of the Edwards Plateau into the Chihuahuan Desert and isolated the

111

species P. obsoletus and P. bairdi (Fig.1; Lawson and Lieb 1990; Burbrink et al. 2000; Burbrink

112

2001). Since the Pliocene, the ENA has experienced numerous climate change events associated

113

with glacial cycles that forced species into refugia and compressed populations; upon climate

114

amelioration populations expanded into formerly glaciated areas (Hewitt 2000; Bintanja and van

115

de Wal 2008; Burbrink et al. 2016). Using dense population sampling and genomic-scale data,

116

we combine isolation-migration estimates with spatial information to understand species

117

divergence in the context of geographic distance, contemporary and historical environments, and

118

biogeographic boundaries. We also examine how migration rates and divergence dates relate to

119

the width and types of hybrids (F1, backcrosses) in these hybrid zones at biogeographic

120

boundaries. We also investigate demographic changes through time to understand if population

121

expansion occurs among all lineages as predicted given glacial cycling. Addressing these

122

questions provides an integrated view of phylogeographic history over a physically and

123

historically complex landscape and yields a clearer understanding of speciation processes and

124

delimitation in the gray zone.

125

126

127

Methods

128

129

Dataset

130

We sampled 288 individuals liberally covering the range of all species within the

131

Pantherophis obsoletus complex (Fig.1; Dryad XXX). DNA was extracted from all samples

132

using Qiagen DNeasy Blood & Tissue Kits and samples were screened for quality using broad-

133

range Qubit Assays. We used services from RAPiD Genomics (https://www.rapid-

134

genomics.com/services/) to generate 5472 baits and to sequence 5060 conserved elements

135

(UCEs) loci following the protocols from (Faircloth et al. 2012) and (Sun et al. 2014). These

136

markers have been used to address phylogeographic/population genetic and deeper phylogenetic

137

questions (Branstetter and Longino 2019; Younger et al. 2019). We mapped UCE reads to a

138

Chromium 10x Pantherophis spiloides genome (in prep), removed loci containing >50% missing

139

samples, and removed individual specimens missing >30% of all alignments (details on the

140

assembly of UCE loci are available in Supporting Information Material 1). We produced both

141

phased locus datasets, used in PHRAPL and PipeMaster, and single SNPs/locus for all other

142

analyses, furthermore we filtered the unlinked SNP dataset to remove alleles with a minor allele

143

frequency <0.1 and used these data for population structuring analyses.

144

145

Geographic groupings

146

We estimated population structure by comparing Discriminant Analysis of Principal

147

Components (DAPC; Jombart et al., 2010) in adegenet v2.1.2 (Jombart 2008). We compared this

148

to estimated effective migration surfaces (EEMS v; Petkova et al., 2016), which models effective

149

migration rates over geography to represent regions where migration is low in cases where

150

genetic dissimilarity increases rapidly, thus providing a uniquely distinct view of the location of

151

population clusters relative to biogeographic barriers as compared to DAPC (details on these

152

population grouping methods and other methods for assessing structure are available in

153

Supporting Information Material 1). We also examined fixation indices (FST) estimated from Nei

154

(1973) in adegenet v2.1.2 (Jombart 2008) across all loci comparing spatially adjacent lineage

155

pairs from DAPC .

156

157

Isolation, migration, and historical demography

158

We generated a coalescent-based species tree with SNAPP v1.3.0 (Bryant et al. 2012) to

159

understand relationships among the four taxa identified from clustering analyses and test species

160

delimitation. Because generating a tree in SNAPP using all individuals was computationally

161

intractable, we used four individuals per taxon that were sampled from different regions of the

162

taxon’s distribution. We estimated SNAPP trees using BEAST v2.5.2 (Bouckaert et al. 2014;

163

details on parameter setting for SNAPP are available in Supporting Information Material 1). We

164

assessed four alternative species delimitation models including four taxa, three taxa (collapsing

165

P. bairdi/P. obsoletus or P. alleghaniensis/P. spiloides), and two taxa (collapsing P. bairdi/P.

166

obsoletus and P.alleghaniensis/P. spiloides).

167

Using the four inferred lineages, we examined which models best described the origins of

168

these groups given divergence time, historical demographic change, and migration between

169

spatially adjacent taxa. We tested 2,300 candidate isolation-migration models including all

170

possible topologies with or without migration between all spatially adjacent pairs using the

171

program PHRAPL (Jackson et al. 2016) implemented in R. To further test that four genetic

172

lineages exist, we ran pairwise comparisons between the best selected model and models of three

173

populations where sister species were collapsed into a single entity; four Pantherophis species

174

against one model where P. obsoletus and P. bairdi were collapsed, and another where P.

175

spiloides and P. alleghaniensis were collapsed (details on PHRAPL runs are available in

176

Supporting Information Material 1).

177

To estimate gene flow, timing of divergence, and demographic change, we used

178

PipeMaster (Gehara et al., 2017; Gehara et al. in review) to simulate genetic data and perform

179

approximate Bayesian computation (ABC) and supervised machine-learning. Here we used the

180

best model selected in the PHRAPL analysis as a template (same topology and migration

181

parameters) to generate three competing models: (i) an isolation migration model with constant

182

population size for each lineage and constant migration, (ii) demographic change along each

183

lineage and constant migration through time and demographic change, and (iii) population size

184

change with migration occurring after the Last Glacial Maximum. We simulated 100,000 data

185

sets of 54 summary statistics and performed ABC rejection with 0.01 tolerance level to select the

186

best of these three models (Table S1). We then took the selected model and performed rejection

187

using a 0.1 tolerance level with a neural network regression using a single layer and 20 nodes to

188

estimate the model parameters.

189

190

Spatial population genetics

191

We used artificial neural networks (ANN; Lek and Guégan 1999; Legendre and Fortin

192

2010; Legendre et al. 2011) to understand how contemporary and historical aspects of the

193

landscape and climate generated genetic diversity. We first compared niche models of current

194

climates between adjacent species pairs to determine if contemporary niches were significantly

195

different prior to using these variables for understanding their effects on genetic structure (see

196

Supporting Information Material 1 for details). We then used Nei’s distance (Nei, 1972) to

197

estimate genetic distance among all individuals in the R package adegenet. For predictor

198

variables, we used 1) geographic distances measured as pairwise great-circle distances among all

199

points to account for isolation-by-distance using the R package fossil (Vavrek 2020), 2) binary

200

categorization of isolation east and west of the Mississippi River, 3) elevation, and 4) bioclimatic

201

variables at 2.5 mins representing current climate, Last Glacial Maximum (21kya), Last

202

Interglacial (130 kya), Pleistocene Marine Isotope Stage 19 (787 kya) and Mid-Pliocene

203

Warming (3.2 Ma) all obtained from the PaleoClim database (Brown et al. 2018). For the

204

bioclimatic data and elevation, we extracted parameter values given the latitude and longitude for

205

each individual sample. To reduce dimensionality of the bioclimatic variables, we transformed

206

the former using principal components analyses using the R package raster (Hijmans et al. 2014).

207

Spatial distances were transformed using principal coordinates analyses in the R package ape

208

(Paradis and Schliep 2019). For principal coordinate/component analyses we retained axes that

209

account for >95% of variance.

210

We used regression-based ANN to determine which variables predict genetic distances

211

and rank those with the highest model importance. Machine learning methods can infer non-

212

linear interactions among many predictor variables, regardless of the distribution or variable type

213

and can generate a range of models of varying complexity (Lek et al. 1996; Zhang 2010;

214

Libbrecht and Noble 2015; Sheehan et al. 2016). To conduct the ANN regressions we used the R

215

package caret (Kuhn 2008) and partitioned the data into the standard 70% training and 30% test

216

sets (Lek et al. 1996; Zhang 2010; Burbrink et al. 2020). This analysis was run using 1,000

217

maximum iterations to ensure convergence. We resampled the data using the default 25 bootstrap

218

replicates to reach convergence over the following parameters: weight decay, root mean squared

219

error, r2, and mean absolute error. We assessed the power of these models by recomposing the

220

test and training sets 100 times and compared three test statistics (root mean squared error, r2,

221

and mean absolute error) to those from randomized response variables for each of these 100 re-

222

estimated models. We ranked the most important variables over these replicates after accounting

223

for multicollinearity among variable affects using the R package Rnalytica (see Supporting

224

Information Material 1). We also compared results from our ANN analysis to those using

225

redundancy analyses (RDA, see Supporting Information Material 1 for details).

226

227

Hybrid-zone dynamics

228

To understand if the composition and width of hybrid-zones differed among adjacent

229

lineages we first estimated the composition of individuals and their locations (parentals, F1s,

230

backcrosses). We first used snapclust (Beugin et al. 2018) in adegenet to assign pure parentals

231

(P1 and P2), backcrosses and F1s between the following spatially adjacent pairs of taxa, herein

232

referred to by geographic area: 1) eastern (P. allegahniensis/P. spiloides), 2) central (P.

233

spiloides/P. obsoletus), and 3) western (P. obsoletus/P. bairdi). This method maximizes the

234

likelihood of two fixed panmictic populations using a geometric approach with an expectation-

235

maximization algorithm. We estimated group membership first using the “k-means” option in the

236

function. To check the probability that snapclust identified hybrids correctly with these data, we

237

isolated pure parental individuals (membership >0.95), used the hybridize function to generate

238

the same number of hybrids and parentals and then determined the probability that snapclust

239

finds the correct number of hybrids and parentals.

240

We then fit clines in the program HZAR v.2-9 (Derryberry et al. 2014) for the same

241

species pairs using individual genetic assignments from DAPC to determine if cline widths differ

242

among groups. We compared these to ancestry alternatively estimated in snaplcust (see

243

Supporting Information Material 1 for details). Using the Gaussian cline model, we estimated the

244

center and width of the cline and determined if these sigmoidal distributions have significant tails

245

by fitting the following models: 1) no tails, 2) right tail only, 3) left tail only, 4) mirrored tails,

246

and 5) both tails estimated independently (see Derryberry et al. 2014). We fit these models to our

247

data using AICc and ran the MCMC chains for 5x106 generations, thinned by 5x103 generations,

248

and estimated stationarity using ESS >200 in the R package CODA (Plummer et al. 2006). We

249

note that making transects through multiple locations along each hybrid zone would better define

250

the width throughout the length of the cline, though this was not possible here for all clines.

251

Using equations from Barton and Gale (1993) and applied in Bailey et al. (2015) we

252

addressed selection against hybridization for each cline. Using estimates of maximum lifetime

253

dispersal (s) for ratsnakes, here 0.937-4.3 km (Weatherhead and Blouin-Demers pers comm.),

254

we predict the width of the cline under neutrality over a range of generation times (T) since the

255

origin of each sister pair and group using the equation:

! " #$%&'

(

We also predict selection

256

at the center of the cline, also using s and estimates of cline width (w) from HZAR using the

257

equation:

258

* "

. We estimate these given the range of dispersal estimates and estimates for the widths

259

of clines.

260

261

All scripts, genetic, and spatial data are available on Dryad XX.

262

263

264

Results

265

Data and population genetic structure

266

After filtering loci for presence in 70% or more individuals, we generated 2,491 UCE

267

loci. For all downstream analyses we randomly sampled one SNP per locus yielding a total of

268

846 SNPs for 238 individuals for an average of 13.96% missing data.

269

Both EEMS and DAPC find population structure that generally matched the geographic

270

ranges of P. alleghaniensis, P. spiloides, P. obsoletus, and P. bairdi (Fig. 1 and Fig. S1;

271

Burbrink, 2001). All three EEMS generated similar acceptance proportions for all proposal types

272

(12-51%). Total DAPC assignment probabilities without apriori species groupings were 0.975

273

(P. bairdi = 1.0, P. obsoletus = 0.96; P. spiloides = 1.0, and P. alleghaniensis = 0.97). Similarly,

274

all three EEMS runs suggest low migration at the MR and the area separating the P. bairdi and

275

P. obsoletus in west Texas. East of the MR, estimated low migration occurred near the

276

Appalachian Mountains, though this area is a complex mixture of isolation and gene flow (Fig.

277

1).

278

These estimates of population structure all showed P. alleghaniensis as occurring mostly

279

in the Florida Peninsula and north along the east coast to Virginia/Delaware (not as far north as

280

in Burbrink 2001), P. spiloides is found from western Florida to the Mississippi River up through

281

the Midwest and east to the northeastern US, P. obsoletus is found west of the Mississippi River

282

in the forested regions of the Midwest, and P. bairdi is distributed in the Chihuahuan Desert of

283

west Texas (Figs.1-2).

284

285

Isolation-migration processes

286

Our SNAPP analyses produced a phylogeny with a root separating the taxa east and west

287

of the Mississippi River and then a sister relationship between P. bairdi and P. obsoletus and

288

then P. spiloides and P. alleghaniensis (Fig. 2), consistent with Burbrink et al., (2000). Bayes

289

factors (BF) showed the four-taxon model as decisively superior to the three-taxon (BF =35.40)

290

and two-taxon model (BF=2022.46).

291

Using those four lineages we filtered models of divergence using PHRAPL. Models with

292

all four taxa, incorporating migration between spatially adjacent lineages was preferred with the

293

same topology as found using SNAPP. DAIC between the best ranked model and next model was

294

> 20, suggesting high confidence in model selection (Table S2).

295

The demographic model estimated from PipeMaster that incorporated historical

296

population size change and constant migration (IMD) best fit the data (posterior probability 1.0;

297

see PCA plots in Fig. S2; Figs. 2 & 3). Median migration rates were highest between eastern

298

groups (P. alleghaniensis to P. spiloides = 2.79 and P. spiloides to P. alleghaniensis = 3.79

299

individuals/generation), lower in the central groups across the Mississippi River (P. spiloides to

300

P. obsoletus = 1.96 and P. obsoletus to P. spiloides = 1.28 individuals/generation), and lowest in

301

western groups (P. bairdi to P. obsoletus= 0.55 and P. obsoletus to P. bairdi = 0.65

302

individuals/generations; Table 1). Divergence occurred earliest in the central lineages at the

303

Mississippi River at 1.08 My (95% CI =0.553 to 1.674 My), then the eastern lineages at 0.650

304

My (95%CI = 0.225 to 0.983 My) and western lineages at 0.367 My (95% CI 0.367 to 0.875 My;

305

Table 1). Divergence dates and average rates of migration were poorly correlated (ρ=0.264,

306

P=0.83, n=3). For all taxa we inferred population expansion from ancestral Ne sizes ranging from

307

5,200 to 5,600 to modern sizes at 138,000 to 281,000. The posterior distributions for all

308

parameters were different from the uniform priors, though wide distributions of several

309

parameters suggested some uncertainty in estimates (Fig. 3).

310

311

Spatial population genetics and hybrid zones

312

We used ANN to determine what features of the landscape and environmental layers

313

through time best predict genetic structure (Fig.4). Contemporary niches between taxa were

314

significantly different for all taxa (see Supporting Information Material 1 for details, Fig. S3).

315

Estimates of accuracy using ANN were high (>90%) for all comparisons. Most genetic structure

316

can be predicted by the Mississippi River showing 100% variable importance (Table S3).

317

Distance also played a role in structuring these data (although they were partialed out in RDA

318

analyses yielding similar conclusions regarding other landscape variables, see Supporting

319

Information Material 1). For those lineages east of the Mississippi River, population structure

320

was predicted by climate at the Last Glacial Maxima and elevation (variable importance=100).

321

For those lineages west of the Mississippi River, climate during Pleistocene Marine Isotope

322

Stage 19 and elevation structured these lineages (variable importance = 100).

323

Using snapclust, we showed that the number of hybrids of all types decreases with

324

species pair comparisons from east to west (Fig. 5). The parentals, hybrids and backcrosses were

325

predicted with a mean accuracy of 0.99 in simulations. All hybrid types were spatially oriented

326

between parental types where the number of hybrids and backcrosses between the eastern groups

327

approached the number of parental individuals (49%), for the central groups the hybrids and

328

backcrosses to parental ratio was 5.9%, and for the western groups there were no hybrids

329

detected (Fig. 5). It is important to note, however, that sampling was more concentrated between

330

P. alleghaniensis/P. spiloides compared to the other pairs.

331

For the eastern lineages, HZAR showed strong support for a right-tailed model

332

(ΔAICDAPC =-31.96 - -107.42), whereas central lineages at the Mississippi River (ΔAICDAPC = -

333

4.26 - -8.869) and western lineages at the Chihuahuan Desert and Edward’s Plateau (ΔAICDAPC =

334

-5.51 - -221.35) both supported a no-tails model (Fig. 6; Table S4 & S5). However, estimating

335

tailed models for the latter two comparisons may have required more samples than were

336

available. The width of clines decreased from the eastern lineages (width=126 km, 1st and 3rd

337

quartiles 114-138 km, tail length =53 km, 1st and 3rd quartiles 44-61 km), to central (width = 98

338

km, 1st and 3rd quartile 89-107 km), to the western lineages (width = 8 km, 1st and 3rd quartile =4-

339

15 km; Table 1). Sharply declining hybrid zones characterized the central and western lineages,

340

whereas the cline for eastern lineages showed a long right tail extending north.

341

Hybrid zone widths under neutrality should be larger than the actual widths given that

342

migration between sister pairs likely occurred prior to the LGM (Fig. S4). Contact time for each

343

of these pairs to approach the same actual width sizes all occurred within the Holocene: 516-

344

10,785 ybp (eastern lineages), 261-5424 ybp (central lineages) and 3-27 ybp (western lineages).

345

We also showed that selection against hybrids at the center of hybrid zones ranged from a

346

median of 0.3 % (0.02-0.6%) for eastern, 0.5% (0.03 to 1%) for central, and 79% (1.5-100%) for

347

western lineages.

348

When using a single SNP per locus we showed that 30% of loci from central lineages,

349

10.9% of loci for western lineages, and only 1.9% of loci for eastern lineages attained an Fst

350

>0.35. Given demographic histories between comparisons, this threshold for Fst is generally

351

considered extreme population disconnection beyond which advantageous loci will be exchanged

352

between populations (Wright 1978; Frankham et al. 2010).

353

354

Discussion

355

Grey Zone Dynamics

356

To understand the factors shaping reproductive isolation in a young species complex, we

357

ask whether divergence time predicts rates of gene flow or cline width across space in eastern

358

ratsnakes. This species complex is composed of four divergent lineages, with P. alleghaniensis

359

and P. spiloides meeting in the southeastern US at the intersection of subtropical and temperate

360

areas, P. spiloides and P. obsoletus separated at the Mississippi River, and P. obsoletus and P.

361

bairdi meeting at the connection on the forested and rocky habitats on the western edge of the

362

Edwards Plateau. All three spatially adjacent pairs of taxa show variable migration each having

363

unique cline shapes. The western lineages (P. obsoletus, P. bairdi) show very low migration

364

rates, a cline equivalent to only 0.35% of the parental range, fixation (Fst > 0.35) in 10.9% of

365

loci, and likely selection against hybrids. This is in comparison with central lineages (P.

366

obsoletus, P. spiloides) showing more migration, a cline occupying 5% of the parental ranges,

367

but with 35% of loci showing fixation, and possibly lower selection. Finally, eastern lineages (P.

368

alleghanienis, P. spiloides) have incomplete reproductive isolation (showing numerous F1 and

369

backcrosses) with only 1.9% of loci showing fixation, and high migration rates with a clinal tail

370

(Figs. 3 & 6). Where reproductive isolation occurs given low numbers of migrants/generation,

371

there is increased fixation among loci and steeper clines occur. The inverse is also true; when

372

reproductive isolation is incomplete, higher rates of gene flow less fixation of loci, and variable

373

cline widths occur. Importantly, however, reproductive isolation, gene flow estimates, and hybrid

374

clines show no association with timing of divergence.

375

Hybrid zone dynamics help to understand the interactions between pairs of taxa as

376

characteristics of the hybrid zone reveal whether there is selection against hybrids (tension

377

zones), selection for hybrids (bounded superiority), or ecological gradients with reduced fitness.

378

They also reveal whether selection is endogenous (genomic) or exogenous (environmental;

379

Barton and Hewitt 1985; Barton and Gale 1993). Our models indicate that migration at the

380

Mississippi River and in the eastern lineages occurred through the Pleistocene suggesting some

381

form of a hybrid zone existed through major climate change cycles. Given this age and assuming

382

selective neutrality, we demonstrate that these hybrid zones should be much larger, in most cases

383

exceeding the ranges of parental taxa (Fig. S4). The widths of the current hybrid zones are

384

maintained over time, preserving the historical and geographic identity of the parental taxa.

385

While there is some selection against hybrids as estimated from hybrid zone width and dispersal,

386

depending on species-pair comparison, it is possible that other factors may be constraining these

387

hybrid zones like fitness changes along an ecological gradient or isolation by environment (Case

388

and Taper 2000; Case et al. 2005; McEntee et al. 2018). There is support for this hypothesis

389

given significant differences among niches for all adjacent pairs of lineages (Fig S3).

390

391

Influence of barriers and environment on population structure

392

These ratsnake lineages are structured by a complex of physical barriers, historical

393

climate profiles, and geographic distance. Machine learning approaches show that the

394

Mississippi River has served as a strong barrier for isolating lineages of ratsnakes, forming

395

groups east and west of this river during the mid-Pleistocene at 1.08 My (95% CI =0.553 to

396

1.674 My; Figs 1&4, Table 1). These results are consistent with previous mtDNA and

397

morphological conclusions (Burbrink, 2001; Burbrink et al., 2000). The Mississippi River has

398

served as a strong barrier to gene flow given evidence from historical estimates of increased

399

drainage (Cox et al. 2014) and numerous unrelated organisms show genetic disconnection across

400

the river (Robison 1986; Burbrink 2002; Soltis et al. 2006; Burbrink et al. 2008; Pyron and

401

Burbrink 2009; Brandley et al. 2010; Satler and Carstens 2017; Myers et al. 2020).

402

Hybridization, while low (1-2 individuals/generation, Fig.3), does occur in a zone (~100km) east

403

of the Mississippi River between the central lineages (P. obsoletus and P. spiloides; Fig. 6).

404

Furthermore, divergence in this group can be predicted by elevation, geographic distance, and

405

mid-Pleistocene climate change (787kya). The timing of the mid-Pleistocene climate change falls

406

within the timing of divergence estimated from PipeMaster (Fig.4).

407

The Mississippi River is responsible for the earliest divergence in this complex but other

408

geographic features are also important. East of the Mississippi River, divergence between P.

409

alleghaniensis and P. spiloides was influenced by elevation, climate at the LGM, and geographic

410

distance after accounting for variable inflation (Fig 4). For these taxa, divergence occurred at the

411

Appalachian Mountains and Apalachicola/Chattahoochee River System, which suggests that

412

changes in elevation at the Appalachian Mountains, in part, played a role in isolating lineages or

413

maintaining divergence. Finer-scale testing is necessary to tease apart the effects of these river

414

systems, ancient embayments, and connections to uplifted areas. Although the timing of

415

divergence predates the LGM, it is possible that repeated 100,000 year glacial cycles isolated

416

these taxa into refugia, as has been suggested previously for ratsnakes and other organisms

417

(Waltari et al. 2007; Noss et al. 2015). Alternatively, Pantherophis alleghaniensis may have

418

diverged in an isolated Florida, due to sea level changes during interglacials, and subsequently

419

expanded north and west. The concentration of the yellow, striped morphology in the range of P.

420

alleghaniensis in Florida and parts of the southeastern coast of the US indicates that this isolation

421

likely resulted in these morphologically unique taxa (Schultz 1996; Burbrink 2001). This area of

422

isolation, defined as a suture zone between the Florida peninsula and the continental US has been

423

found for many taxa (Remington 1968; Swenson and Howard 2005; Ruane et al. 2015). While

424

these two forms can be delimited with genetic data and occupy distinctly different geographic

425

regions, reproductive isolation is not complete. However, niches are significantly different and P.

426

alleghaniensis is typically found in the Florida peninsula and coastal plain environments (Bailey

427

1995; Burbrink 2001), whereas P. spiloides is found throughout the remainder of the forested

428

habitats east of the Mississippi River including the ecoregions defined as interior river valleys

429

and hills, interior plateau, Appalachian habitats, southeastern and southcentral plains, and

430

Mississippi River valley. Uncertainty in the identity of taxa using morphology where these taxa

431

meet is likely due to extensive gene flow in hybrid zones at barriers following postglacial range

432

expansion (Fig. 1, 3; Burbrink 2001; Gibbs et al. 2006).

433

Divergence between P. bairdi and P. obsoletus occurred in the western edge of the

434

Edwardas Plateau, with the former extending westward into xeric forested or Chihuahuan desert

435

scrub habitats always associated with rocky habitats, whereas the latter extends east into Texas,

436

preferring forested habitats, (Lawson and Lieb 1990; Werler and Dixon 2000). Population

437

genetic structure between these taxa is associated with mid-Pleistocene climate, elevation, and

438

geographic distance (Fig. 4). Dates of origin of these taxa are within the mid-Pleistocene,

439

suggesting that past climate played a strong role in promoting population divergence. During this

440

period, the Chihuahuan Desert was much drier relative to the late-Pleistocene (Graham and Mead

441

1987; Metcalfe et al. 2002; Metcalfe 2006) and this increased aridity could have driven

442

ecological divergence between this species pair. While divergence time is not the oldest here,

443

migration rates are lowest, and a hybrid zone is narrow. Previous research shows limited

444

hybridization with backcrosses between these taxa occurring in a narrow region of the

445

southwestern Edwards Plateau (Lawson and Lieb 1990; Vandewege et al. 2012).

446

Despite each lineage occupying uniquely different habitats, population responses to

447

Pleistocene climate change were similar (Fig. 3). Given the divergence time for each group, Ne

448

has expanded by 50 times since the Pleistocene. These estimates are consistent with other

449

predictions in the ENA, where 75% of the tested vertebrates show population size expansion and

450

75% of tested snakes show synchronous expansion of Ne, likely coinciding with the retreat of

451

glaciers at various times in the Pleistocene or during the Holocene (Bintanja and van de Wal

452

2008; Burbrink et al. 2016). However, Pantherophis bairdi has a range only in a small area of

453

the west Texas and isolated populations in northeastern Mexico, where habitats may have been

454

indirectly affected by glacial cycles.

455

456

Taxonomy and delimitation in the genomic age

457

Delimiting species using genetic data, while useful, has proved controversial (Bauer et al.

458

2011; Sukumaran and Knowles 2017; Leaché et al. 2019). In particular, most genetic

459

delimitation methods fail to account for gene flow (but see Flouris et al. 2019), do not consider

460

spatial information, and place individuals into predefined groups prior to testing (O’Meara

461

2010). These methods all assume species diverge at an evolutionary time scale where mutations

462

accumulate through drift or selection (Rannala 2015). And while simulations show that a

463

threshold above one migrating individual/10 generations will cause model-based delimitation

464

methods to fail at finding two species (Zhang et al. 2011), it is not clear how pulses or

465

inconsistent migration affect species transitioning through the gray zone. Importantly, Wright

466

(1931) indicated that up to 1,000 individuals migrating/generation may not prevent populations

467

from drifting given effective population size and spatial segregation.

468

Modern thresholds for species delimitation (Jackson et al. 2016; Leaché et al. 2019) use

469

GDI (>0.7), estimates of species divergence time over population size (2τ/θ>1⁠), absolute

470

divergence time (104) and maximum migration rates (M=Nm<1), metrics that are all likely

471

influenced by sampling effort, location of samples sequenced, and detection of hybrids. For

472

example, if large hybrid zones exist but remain poorly sampled then estimates of migration will

473

be reduced, GDI will be high, and the probability of MSC estimators will suggest multiple

474

species. Conversely, if only hybrid zones are sampled then all metrics will predict only a single

475

species. For all pairs of ratsnakes 2τ/θ>1 and divergence time is always greater than 104.

476

However, migration rates remain high between adjacent pairs in the eastern lineages. This forces

477

us to address how we are defining a species in the context of hybrid zones, how we infer

478

reproductive isolation beyond MSC methods, and how selection is maintaining hybrid zones. For

479

all lineages of ratsnakes, hybrid zones are smaller than predicted by neutral selection if these

480

hybrid zones formed in the Pleistocene. It is possible that these hybrid zones formed after the

481

retreat of the last glacial cycle during the Holocene. However, geographic lineage identity has

482

been maintained for >133,000 generations in all pairs, regardless of migration, and niches are

483

unique among all lineages. Moreover, where hybrid zones extend over large areas, gene flow

484

likely varies throughout space and may have changed over time (Barton and Hewitt 1985; Mallet

485

2007), particularly in response to glacial cycles. Thus, identifying and delimiting species in the

486

context of admixture is complex. Still, maintaining lineage and geographic identity, even with a

487

hybrid zone suggests that selection may have maintained unique evolutionary trajectories.

488

Here we have demonstrated that these four species, delimited previously using mtDNA

489

and morphology (Burbrink, 2001; Burbrink et al., 2000) can be recognized as distinct taxa given

490

methods of spatial grouping and species delimitation, though ranges in the eastern-most lineage

491

are restricted much further south than previously thought. However, migration rates between P.

492

spiloides and P. alleghaniensis (Fig. 3) are high enough to draw some concern that these taxa

493

should not be recognized given that they may not represent independent evolutionary trajectories

494

(de Queiroz, 2007; Rannala, 2015; Zhang et al., 2011). Unique niches in the subtropical areas of

495

the southeastern US and morphology provides some evidence for isolation between these taxa:

496

the striped forms (with variable ground colors of orange, yellow, green, and gray) are indicative

497

of P. alleghaniensis whereas adults with dark saddle patterns on grey, olive, brown, or black

498

ground colors (or completely black with no pattern) are found west of Florida and east of the

499

Mississippi River, throughout the Midwest and up to the northeastern US and Canada represent

500

P. spiloides. However, hybridization north and west of Florida is likely represented by a dulling

501

of this yellow ground color and mixed patterned individuals (Schultz 1996; see Supporting

502

Information Material 1 for more information on taxonomy).

503

504

Future Directions

505

To better understand the complexity of speciation processes, an even more integrated

506

whole-genomic approach with dense population sampling should be implemented. Even when

507

researchers have demonstrated that integrating spatial information along with estimates of timing

508

and gene-flow among groups presents a more comprehensive picture of speciation over a

509

complex landscape, there still remain several areas of investigation. For instance, estimating

510

gene flow and cline shape using detailed transects across various intersections along hybrid

511

zones would provide a better understanding of how migration changes in different habitats at

512

different latitudes and what type of hybrid zone is present. Additionally, better integrating

513

spatial, habitat, and migration estimates would allow researchers to test dynamic speciation

514

models that could be used to understand how migration changes over the landscape at various

515

times throughout the history of these taxa. Finally, using whole genome data along clines will

516

help determine if selection is occurring in the hybrid zones and determine the function of loci

517

showing evidence of selection and resisting gene flow.

518

519

Author contributions: FTB and EAM contributed to data collection, all authors contributed to

520

analyses, FTB wrote the initial draft, and all authors edited subsequent versions.

521

522

Acknowledgments: We thank R. Glor, R. Brown, and L. Welton at KU and C. Austin and D.

523

Dittman at LSUMN and K. Krysko and FLMNH, D. Shepard at Louisiana Tech University, and

524

A. McKelvy for providing tissues. We also thank L. and G. Derryberry for help with HZAR. We

525

also thank E. Chen for help extracting DNA. EAM was funded by the Gerstner Scholar/Theodore

526

Roosevelt and Peter Buck/Walter Rathbone Bacon fellowships. We thank D. Frost, A. Leaché,

527

A. Bauer, and K. deQuieroz for helpful discussions on species delimitation, type specimens, and

528

admixed type specimens.

529

530

Data Accessibility Statement: All data and unique scripts are available on Dryad XXX.

531

532

533

534

Literature Cited

535

Agrawal, A. F., J. L. Feder, and P. Nosil. 2011. Ecological Divergence and the Origins of

536

Intrinsic Postmating Isolation with Gene Flow. Int. J. Ecol. 2011:1–15.

537

Bailey, R. G. 1995. Description of the ecoregions of the United States. 2nd ed. USDA Forest

538

Service.

539

Bailey, R. I., M. R. Tesaker, C. N. Trier, and G.-P. Saetre. 2015. Strong selection on male

540

plumage in a hybrid zone between a hybrid bird species and one of its parents. J. Evol. Biol.

541

28:1257–1269.

542

Barton, H. H., and K. Gale. 1993. Genetic analysis of hybrid zones. Pp. 13–45 in Hybrid Zones

543

and the Evolutionary Process. Oxford University Press, New York.

544

Barton, N. H. 2010. What role does natural selection play in speciation? Philos. Trans. R. Soc.

545

Lond. B. Biol. Sci. 365:1825–40.

546

Barton, N. H., and G. M. Hewitt. 1985. Analysis of Hybrid Zones. Annu. Rev. Ecol. Syst.

547

16:113–148.

548

Bauer, A. M., J. F. Parham, R. M. Brown, B. L. Stuart, L. Grismer, T. J. Papenfuss, W. Böhme,

549

J. M. Savage, S. Carranza, J. L. Grismer, P. Wagner, A. Schmitz, N. B. Ananjeva, and R. F.

550

Inger. 2011. Availability of new Bayesian-delimited gecko names and the importance of

551

character-based species descriptions. Proc. R. Soc. B Biol. Sci. 278:490.

552

Beugin, M.-P., T. Gayet, D. Pontier, S. Devillard, T. Jombart, and T. Hansen. 2018. A fast

553

likelihood solution to the genetic clustering problem. Methods Ecol. Evol. 9:1006–1016.

554

Bintanja, R., and R. S. W. van de Wal. 2008. North American ice-sheet dynamics and the onset

555

of 100,000-year glacial cycles. Nature 454:869–72.

556

Bouckaert, R., J. Heled, D. Kühnert, T. Vaughan, C.-H. Wu, D. Xie, M. A. Suchard, A.

557

Rambaut, and A. J. Drummond. 2014. BEAST 2: A software platform for Bayesian

558

evolutionary analysis. PLoS Comput. Biol. 10:e1003537.

559

Bradburd, G. S., G. M. Coop, and P. L. Ralph. 2018. Inferring continuous and discrete

560

population genetic structure across space. Genetics 210:33–52.

561

Brandley, M. C., T. J. Guiher, R. A. Pyron, C. T. T. Winne, and F. T. Burbrink. 2010. Does

562

dispersal across an aquatic geographic barrier obscure phylogeographic structure in the

563

diamond-backed watersnake (Nerodia rhombifer)? Mol. Phylogenet. Evol. 57:552–560.

564

Branstetter, M. G., and J. T. Longino. 2019. Ultra-conserved element phylogenomics of New

565

World Ponera (Hymenoptera: Formicidae) illuminates the origin and phylogeographic

566

history of the endemic exotic ant Ponera exotica. Insect Syst. Divers. 3.

567

Brown, J. L., D. J. Hill, A. M. Dolan, A. C. Carnaval, and A. M. Haywood. 2018. PaleoClim,

568

high spatial resolution paleoclimate surfaces for global land areas. Sci. Data 5:180254.

569

Bryant, D., R. Bouckaert, J. Felsenstein, N. A. Rosenberg, and A. RoyChoudhury. 2012.

570

Inferring species trees directly from biallelic genetic markers: bypassing gene trees in a full

571

coalescent analysis. Mol. Biol. Evol. 29:1917–1932.

572

Burbrink, F. T. 2002. Phylogeographic analysis of the cornsnake (Elaphe guttata) complex as

573

inferred from maximum likelihood and Bayesian analyses. Mol. Phylogenet. Evol. 25:465–

574

476.

575

Burbrink, F. T. 2001. Systematics of the eastern ratsnake complex (Elaphe obsoleta). Herpetol.

576

Monogr. 1–53.

577

Burbrink, F. T., Y. L. Chan, E. A. Myers, S. Ruane, B. T. Smith, and M. J. Hickerson. 2016.

578

Asynchronous demographic responses to Pleistocene climate change in Eastern Nearctic

579

vertebrates. Ecol. Lett. 19:1457–1467.

580

Burbrink, F. T., F. Fontanella, R. A. Pyron, T. J. Guiher, C. Jimenez, and R. Alexander Pyron.

581

2008. Phylogeography across a continent: The evolutionary and demographic history of the

582

North American racer (Serpentes : Colubridae : Coluber constrictor). Mol. Phylogenet.

583

Evol. 47:274–288.

584

Burbrink, F. T., F. G. Grazziotin, R. A. Pyron, D. Cundall, S. Donnellan, F. Irish, J. S. Keogh, F.

585

Kraus, R. W. Murphy, B. Noonan, C. J. Raxworthy, S. Ruane, A. R. Lemmon, E. M.

586

Lemmon, and H. Zaher. 2020. Interrogating genomic-scale data for Squamata (lizards,

587

snakes, and amphisbaenians) shows no support for key traditional morphological

588

relationships. Syst. Biol. 69:502–520.

589

Burbrink, F. T., R. Lawson, and J. B. Slowinski. 2000. Mitochondrial DNA phylogeography of

590

the polytypic North American rat snake (Elaphe obsoleta): A critique of the subspecies

591

concept. Evolution (N. Y). 54:2107–2118.

592

Case, T. J., R. D. Holt, M. A. McPeek, and T. H. Keitt. 2005. The community context of species’

593

borders: ecological and evolutionary perspectives. Oikos 108:28–46.

594

Case, T. J., and M. L. Taper. 2000. Interspecific competition, environmental gradients, gene

595

flow, and the coevolution of species’ borders. Am. Nat. 155:583–605.

596

Cox, R. T., D. N. Lumsden, and R. B. Van Arsdale. 2014. Possible relict meanders of the

597

Pliocene Mississippi River and their implications. J. Geol. 122:609–622.

598

De Queiroz, K. 2007. Species concepts and species delimitation. Syst. Biol. 56:879–886.

599

Derryberry, E. P., G. E. Derryberry, J. M. Maley, and R. T. Brumfield. 2014. hzar : hybrid zone

600

analysis using an R software package. Mol. Ecol. Resour. 14:652–663.

601

Edwards, S., R. Hopkins, and J. Mallet. 2020. Speciation. Pp. 296–318 in S. Scheiner and D.

602

Mindell, eds. The Theory of Evolution. University of Chicago Press, Chicago.

603

Faircloth, B. C., J. E. McCormack, N. G. Crawford, M. G. Harvey, R. T. Brumfield, and T. C.

604

Glenn. 2012. Ultraconserved Elements Anchor Thousands of Genetic Markers Spanning

605

Multiple Evolutionary Timescales. Syst. Biol. 61:717–726.

606

Feder, J. L., S. P. Egan, and P. Nosil. 2012. The genomics of speciation-with-gene-flow. Trends

607

Genet. 28:342–350.

608

Flouris, T., X. Jiao, B. Rannala, and Z. Yang. 2019. A Bayesian implementation of the

609

multispecies coalescent model with introgression for phylogenomic analysis. Mol. Biol.

610

Evol., doi: 10.1093/molbev/msz296.

611

Frankham, R., J. D. (Jonathan D. . Ballou, and D. A. (David A. Briscoe. 2010. Introduction to

612

conservation genetics. Cambridge University Press.

613

Frantz, A. C., L. C. Pope, T. R. Etherington, G. J. Wilson, and T. Burke. 2010. Using isolation-

614

by-distance-based approaches to assess the barrier effect of linear landscape elements on

615

badger (Meles meles) dispersal. Mol. Ecol. 19:1663–1674.

616

Garrick, R. C., J. D. Banusiewicz, S. Burgess, C. Hyseni, and R. E. Symula. 2019. Extending

617

phylogeography to account for lineage fusion. J. Biogeogr. 46:268–278.

618

Gavrilets, S. 2004. Fitness landscapes and the origin of species. Princeton University Press.

619

Gay, L., P.-A. Crochet, D. A. Bell, and T. Lenormand. 2008. Comparing clines on molecular and

620

phenotypic traits in hybrid zones: a window on tension zone models. Evolution (N. Y).

621

62:2789–2806.

622

Gehara, M., A. A. Garda, F. P. Werneck, E. F. Oliveira, E. M. da Fonseca, F. Camurugi, F. d. M.

623

Magalhães, F. M. F. M. Lanna, J. W. Sites, R. Marques, R. Silveira-Filho, V. A. São Pedro,

624

G. R. Colli, G. C. Costa, and F. T. Burbrink. 2017. Estimating synchronous demographic

625

changes across populations using hABC and its application for a herpetological community

626

from northeastern Brazil. Mol. Ecol. 26:4756–4771.

627

Gibbs, H. L., S. J. Corey, G. Blouin-Demers, K. A. Prior, and P. J. Weatherhead. 2006.

628

Hybridization between mtDNA-defined phylogeographic lineages of black ratsnakes

629

(Pantherophis sp.). Mol. Ecol. 15:3755–3767.

630

Gourbière, S., and J. Mallet. 2010. Are species real? The shape of the species boundry with

631

exponential failure, reinforcement, and the “missing snowball.” Evolution (N. Y). 64:1–24.

632

Graham, R. W., and J. I. Mead. 1987. Environmental fluctuations and evolution of mammalian

633

faunas during the last deglaciation in North America. Pp. 371–402 in North America and

634

Adjacent Oceans During the Last Deglaciation. Geological Society of America, Boulder,

635

Colorado 80301.

636

Harrison, R. G. 1993. Hybrid zones and the evolutionary process. Oxford University Press.

637

Hewitt, G. 2000. The genetic legacy of the Quaternary ice ages. Nature 405:907–913.

638

Hewitt, G. M. 2008. Speciation, hybrid zones and phylogeography - or seeing genes in space and

639

time. Mol. Ecol. 10:537–549.

640

Hey, J. 2010. Isolation with migration models for more than two populations. Mol. Biol. Evol.

641

27:905–920.

642

Hey, J., and R. Nielsen. 2004. Multilocus methods for estimating population sizes, migration

643

rates and divergence time, with applications to the divergence of Drosophila pseudoobscura

644

and D-persimilis. Genetics 167:747–760.

645

Hijmans, R. J., J. van Etten, and J. Checg. 2014. Geographic Data Analysis and Modeling [R

646

package raster version 2.5-8]. Comprehensive R Archive Network (CRAN).

647

Jackson, N. D., B. C. Carstens, A. E. Morales, and B. C. O’Meara. 2016. Species delimitation

648

with gene flow. Syst. Biol. 66:799–812.

649

Jiggins, C. D., and J. Mallet. 2000. Bimodal hybrid zones and speciation. Trends Ecol. Evol.

650

15:250–255.

651

Jombart, T. 2008. adegenet: a R package for the multivariate analysis of genetic markers.

652

Bioinformatics 24:1403–1405.

653

Jombart, T., S. Devillard, and F. Balloux. 2010. Discriminant analysis of principal components: a

654

new method for the analysis of genetically structured populations. BMC Genet. 11:94.

655

Joseph, L., and T. Wilke. 2007. Lack of phylogeographic structure in three widespread

656

Australian birds reinforces emerging challenges in Australian historical biogeography. J.

657

Biogeogr. 34:612–624.

658

Karrenberg, S., X. Liu, E. Hallander, A. Favre, J. Herforth0Rahmé, and A. Widmer. 2019.

659

Ecological divergence plays an important role in strong but complex reproductive isolation

660

in campions (Silene ). Evolution (N. Y). 73:245–261.

661

Kuhn, M. 2008. Caret: Classification and Regression Training Package. R package version 6.0-

662

77.

663

Lawson, R., and C. S. Lieb. 1990. Variation and hybridization in Elaphe bairdi (Serpentes:

664

Colubridae). J. Herpetol. 24:280.

665

Leaché, A. D., T. Zhu, B. Rannala, and Z. Yang. 2019. The spectre of too many species. Syst.

666

Biol. 68:168–181.

667

Legendre, P., and M. J. Fortin. 2010. Comparison of the Mantel test and alternative approaches

668

for detecting complex multivariate relationships in the spatial analysis of genetic data. Mol.

669

Ecol. Resour. 10:831–844.

670

Legendre, P., J. Oksanen, and C. J. F. ter Braak. 2011. Testing the significance of canonical axes

671

in redundancy analysis. Methods Ecol. Evol. 2:269–277.

672

Lek, S., M. Delacoste, P. Baran, I. Dimopoulos, J. Lauga, and S. Aulagnier. 1996. Application of

673

neural networks to modelling nonlinear relationships in ecology. Ecol. Modell. 90:39–52.

674

Lek, S., and J. F. Gue. 1999. Artificial neural networks as a tool in ecological modelling , an

675

introduction. 120:65–73.

676

Libbrecht, M. W., and W. S. Noble. 2015. Machine learning applications in genetics and

677

genomics. Nat. Rev. Genet. 16:321–332.

678

Lourenço, C. R., K. R. Nicastro, C. D. McQuaid, R. M. Chefaoui, J. Assis, M. Z. Taleb, and G. I.

679

Zardi. 2017. Evidence for rangewide panmixia despite multiple barriers to dispersal in a

680

marine mussel. Sci. Rep. 7:10279. Nature Publishing Group.

681

Makowsky, R., J. Chesser, and L. J. Rissler. 2009. A striking lack of genetic diversity across the

682

wide-ranging amphibian Gastrophryne carolinensis (Anura: Microhylidae). Genetica

683

135:169–183. Springer Netherlands.

684

Mallet, J. 2007. Hybrid speciation. Nature 446:279–283. Nature Publishing Group.

685

Mayr, E. 1963. Animal species and evolution. Belknap Press of Harvard University Press,

686

Cambridge,.

687

McEntee, J. P., J. G. Burleigh, and S. Singhal. 2018. Dispersal predicts hybrid zone widths

688

across animal diversity: Implications for species borders under incomplete reproductive

689

isolation. bioRxiv 472506. Cold Spring Harbor Laboratory.

690

Metcalfe, S. E. 2006. Late Quaternary environments of the northern deserts and central

691

transvolcanic belt of Mexico. Ann. Missouri Bot. Gard. 93:258–273. Missouri Botanical

692

Garden Press.

693

Metcalfe, S., A. Say, S. Black, R. McCulloch, and S. O’Hara. 2002. Wet Conditions during the

694

Last Glaciation in the Chihuahuan Desert, Alta Babicora Basin, Mexico. Quat. Res. 57:91–

695

101.

696

Myers, E. A., A. D. McKelvy, and F. T. Burbrink. 2020. Biogeographic barriers, Pleistocene

697

refugia, and climatic gradients in the southeastern Nearctic drive diversification in

698

cornsnakes ( Pantherophis guttatus complex). Mol. Ecol. 29:797–811. John Wiley & Sons,

699

Ltd.

700

Myers, E. A., A. T. Xue, M. Gehara, C. L. Cox, A. R. Davis Rabosky, J. Lemos0Espinal, J. E.

701

Martínez0Gómez, and F. T. Burbrink. 2019. Environmental heterogeneity and not vicariant

702

biogeographic barriers generate community0wide population structure in desert0adapted

703

snakes. Mol. Ecol. 28:4535–4548.

704

Nei, M. 1973. Analysis of gene diversity in subdivided populations. Proc Natl Acad Sci USA

705

70:3321–3323.

706

Nosil, P. 2012. Ecological Speciation. Oxford University Press, Oxford.

707

Nosil, P., and B. J. Crespi. 2006. Ecological divergence promotes the evolution of cryptic

708

reproductive isolation. Proc. R. Soc. B-Biological Sci. 273:991–997.

709

Nosil, P., J. L. Feder, S. M. Flaxman, and Z. Gompert. 2017. Tipping points in the dynamics of

710

speciation. Nat. Ecol. Evol. 1:1. Nat Ecol Evol.

711

Noss, R. F., W. J. Platt, B. A. Sorrie, A. S. Weakley, D. B. Means, J. Costanza, and R. K. Peet.

712

2015. How global biodiversity hotspots may go unrecognized: lessons from the North

713

American Coastal Plain. Divers. Distrib. 21:236–244.

714

O’Meara, B. C. 2010. New heuristic methods for joint species delimitation and species tree

715

inference. Syst. Biol. 59:59–73.

716

Orr, H. A. 1995. The population genetics of speciation: the evolution of hybrid incompatibilities.

717

Genetics 139:1805–13. Genetics Society of America.

718

Paradis, E., and K. Schliep. 2019. ape 5.0: an environment for modern phylogenetics and

719

evolutionary analyses in R. Bioinformatics 35:526–528. .

720

Petkova, D., J. Novembre, and M. Stephens. 2016. Visualizing spatial population structure with

721

estimated effective migration surfaces. Nat. Genet. 48:94–100. Nature Publishing Group.

722

Plummer, M., N. Best, K. Cowles, and K. Vines. 2006. CODA: Convergence diagnosis and

723

output analysis for MCMC. R News 6:7–11.

724

Pyron, R. A., and F. T. Burbrink. 2009. Lineage diversification in a widespread species: roles for

725

niche divergence and conservatism in the common kingsnake, Lampropeltis getula. Mol.

726

Ecol. 18:3443–3457.

727

Rannala, B. 2015. The art and science of species delimitation. Curr. Zool. 61:846–853.

728

Remington, C. L. 1968. Suture-zones of hybrid interaction between recently joined biotas. Pp.

729

321–428 in Evolutionary Biology. Springer US, Boston, MA.

730

Riddle, B. R. 2016. Comparative phylogeography clarifies the complexity and problems of

731

continental distribution that drove A. R. Wallace to favor islands. Proc. Natl. Acad. Sci. U.

732

S. A. 113:7970.

733

Robison, H. W. 1986. Zoogeographic implications of the Mississippi River Basin. Pp. 267–285

734

in The Zoogeography of North American Freshwater Fishes. Wiley, New York.

735

Roux, C., C. Fraïsse, J. Romiguier, Y. Anciaux, N. Galtier, and N. Bierne. 2016. Shedding light

736

on the grey zone of speciation along a continuum of genomic divergence. PLoS Biol.

737

14:e2000234.

738

Ruane, S., O. Torres-Carvajal, and F. T. Burbrink. 2015. Independent demographic responses to

739

climate change among temperate and tropical milksnakes (Colubridae: Genus

740

Lampropeltis). PLoS One 10:e0128543.

741

Satler, J. D., and B. C. Carstens. 2017. Do ecological communities disperse across biogeographic

742

barriers as a unit? Mol. Ecol. 26:3533–3545.

743

Schultz, K. D. 1996. A monograph of the colubrid snakes of the genus Elaphe Fitzinger. Koelz

744

Scientific Books, Havlicuv.

745

Seehausen, O., R. K. Butlin, I. Keller, C. E. Wagner, J. W. Boughman, P. A. Hohenlohe, C. L.

746

Peichel, G.-P. Saetre, C. Bank, A. Brannstrom, A. Brelsford, C. S. Clarkson, F.

747

Eroukhmanoff, J. L. Feder, M. C. Fischer, A. D. Foote, P. Franchini, C. D. Jiggins, F. C.

748

Jones, A. K. Lindholm, K. Lucek, M. E. Maan, D. A. Marques, S. H. Martin, B. Matthews,

749

J. I. Meier, M. Most, M. W. Nachman, E. Nonaka, D. J. Rennison, J. Schwarzer, E. T.

750

Watson, A. M. Westram, and A. Widmer. 2014. Genomics and the origin of species. Nat

751

Rev Genet 15:176–192.

752

Sheehan, S., Y. S. Song, E. Buzbas, D. Petrov, A. Boyko, and A. Auton. 2016. Deep learning for

753

population genetic inference. PLOS Comput. Biol. 12:e1004845.

754

Singhal, S., and C. Moritz. 2013. Reproductive isolation between phylogeographic lineages

755

scales with divergence. Proceedings. Biol. Sci. 280:20132246.

756

Soltis, D. E., A. B. Morris, J. S. McLachlan, P. S. Manos, and P. S. Soltis. 2006. Comparative

757

phylogeography of unglaciated eastern North America. Mol. Ecol. 15:4261–4293.

758

Stankowski, S., J. M. Sobel, and M. A. Streisfeld. 2017. Geographic cline analysis as a tool for

759

studying genome-wide variation: a case study of pollinator-mediated divergence in a

760

monkeyflower. Mol. Ecol. 26:107–122.

761

Sukumaran, J., and L. L. Knowles. 2017. Multispecies coalescent delimits structure, not species.

762

Proc. Natl. Acad. Sci. U. S. A. 114:1607–1612.

763

Sun, K., K. A. Meiklejohn, B. C. Faircloth, T. C. Glenn, E. L. Braun, and R. T. Kimball. 2014.

764

The evolution of peafowl and other taxa with ocelli (eyespots): a phylogenomic approach.

765

Proc. R. Soc. B Biol. Sci. 281:20140823.

766

Swenson, N. G., and D. J. Howard. 2005. Clustering of contact zones, hybrid zones, and

767

phylogeographic breaks in North America. Am. Nat. 166:581–591.

768

Vandewege, M. W., D. Rodriguez, J. P. Weaver, T. D. Hibbitts, M. R. J. Forstner, and L. D.

769

Densmore. 2012. Evidence of hybridization between Elaphe bairdi and Elaphe obsoleta

770

lindheimeri including comparative population genetics inferred from microsatellites and

771

mitochondrial DNA. J. Herpetol. 46:56–63. Society for the Study of Amphibians and

772

Reptiles.

773

Vavrek, M. J. 2020. Fossil. https://cran.r-project.org/web/packages/fossil/index.html.

774

Walker, D., and J. C. Avise. 1998. Principles of phylogeography as illustrated by freshwater and

775

terrestrial turtles in the southeastern United States. Annu. Rev. Ecol. Syst. 29:23–58.

776

Waltari, E., R. J. Hijmans, A. T. Peterson, Á. S. Nyári, S. L. Perkins, and R. P. Guralnick. 2007.

777

Locating Pleistocene refugia: comparing phylogeographic and ecological niche model

778

predictions. PLoS One 2:e563.

779

Werler, J. E., and J. R. Dixon. 2000. Texas snakes : identification, distribution, and natural

780

history. 1st ed. University of Texas Press, Austin.

781

Wright, S. 1978. Evolution and the Genetics of Populations. Vol. 4. Variability within and

782

among Natural Populations. University Chicago Press.

783

Wright, S. 1931. Evolution in Mendelian populations. Genetics 16:97–159.

784

Wright, S. 1943. Isolation by Distance. Genetics 28:114–38. Genetics Society of America.

785

Younger, J. L., P. Dempster, Á. S. Nyári, T. O. Helms, M. J. Raherilalao, S. M. Goodman, and S.

786

Reddy. 2019. Phylogeography of the Rufous vanga and the role of bioclimatic transition

787

zones in promoting speciation within Madagascar. Mol. Phylogenet. Evol. 139:106535.

788

Zellmer, A. J., M. M. Hanes, S. M. Hird, and B. C. Carstens. 2012. Deep phylogeographic

789

structure and environmental differentiation in the carnivorous plant Sarracenia alata. Syst.

790

Biol. 61:763–777.

791

Zhang, C., D.-X. Zhang, T. Zhu, and Z. Yang. 2011. Evaluation of a bayesian coalescent method

792

of species delimitation. Syst. Biol. 60:747–761.

793

Zhang, W. 2010. Computational Ecology: Artificial Neural Networks and Their Applications.

794

World Scientific.

795

796

797

798

799

800

801

802

803

804

805

806

Table 1. Measures of divergence dates in millions of years (My), migration rates (first number is

807

first taxon into second taxa, e.g., Pantherophis alleganiensis into P.spiloides, and second vice

808

versa), and migration rates (individuals/generation) for each contacting group in the

809

Pantherophis obsoletus complex.

810

811

Group

Age (My)

Migration Rate (ind/gen)

Cline Width (m)

Eastern (P. alleghaniensis x P.spiloides)

0.65 (0.367-0.875)

2.79/3.79

125 (114-138)

Central (P.spiloides x P.obsoletus)

1.08 (0.55-1.67)

1.96/1.28

97 (88-107)

Western (P.obsoletus x P. bairdi)

0.367 (0.37-0.88)

0.65/0.55

8 (4-15)

812

Figures and Legends

813

814

815

816

Fig.1. Distribution and population delimitation for the ratsnakes Pantherophis obsoletus

817

complex. (A) Map showing range and geographic features: Chihuahuan Desert Province, the

818

Mississippi River, and the Appalachian Mountains and Apalachicola/Chattahoochee River

819

System (AARS). (B) Estimated effective migration surfaces (EEMS) showing areas of both high

820

(blue) and low migration (brown). (C) Population structure using Discriminant Analysis of

821

Principal Components (DAPC) shows the location of four geographically distinct lineages. (D)

822

Bi-plots of PC space shows relative distances among the four taxa using DAPC.

823

824

825

826

827

Fig.2. The results of SNAPP species-tree estimation for the four taxa over all loci (A) and

828

models showing Isolation Migration (IM) and Isolation Migration with Demographic Change

829

(IMD; best-fit model) and Isolation with Recent Migration and Demographic Change (IMRD)

830

estimated in PipeMaster. Colored arrows represent migration among spatially adjacent pairs. (B).

831

Photographs representing some of the typical color patterns seen in ratsnakes: 1) Pantherophis

832

alleghaniensis (photo F. Burbrink), 2) P. alleghaniensis (photo N. Claunch), 3) P. spiloides

833

(photo A. Meier), 4) P. obsoletus (photo D. Shepard), and 5) P. bairdi (E. Myers).

834

835

836

837

838

Fig.3. Results from PipeMaster showing estimates and directionality of gene flow (A),

839

divergence times between groups and taxon pairs (B), and changes in population sizes over time

840

(C).

841

842

843

844

845

846

847

848

849

850

851

852

853

854

855

856

857

858

859

860

861

Fig.4. Artificial neural network results showing accuracy predicting genetic structure among all

862

lineages and species pairs (left) and ranked variable importance (filtered for multicollinearity in

863

inset box, right).

864

865

Fig.5. Results from estimating hybrid types between taxon pairs using snapclust showing the

866

spatial distribution of parentals and hybrids along with their frequencies.

867

Longtiude

−110 −100 −90 −80 −70

25 30 35 40 45

Latitude

−110 −100 −90 −80 −70

25 30 35 40 45

Longtiude

Latitude

−110 −100 −90 −80 −70

25 30 35 40 45

Latitude

P. spiloides

P. alleghaniensis

Backcross (P. alleghaniensis)

Backcross (P. spiloides)

P. spiloides

P. obsoletus

Backcross (P. spiloides)

Backcross (P. obsoletus)

P. obsoletus

P. bairdi

Backcross

(P.alleghaniensis)

F1 P.allegheninsis

P.spiloides

0 10 20 30 40 50

Backcross

(P.spiloides)

Backcross

(P.obsoletus) P.obsoletus

P.spiloides

0 20 40 60 80 100

Backcross

(P.spiloides)

P.bairdi

P.obsoletus

0510 15 20

Longtiude

Eastern Lineages

Central Lineages

Western Lineages

868

Fig.6. Cline model results for all three species pairs using Gaussian Cline (HZAR) model given

869

locations and admixture estimates using DAPC.

870

Eastern Lineages (Pantherophis alleghaniensis x Pantherophis spiloides)

-80.57 Longtitude

25.40 Latitude

~Homestead, FL

Center = 797 km

Width =98 km

-97.42 Longitude

35.18 Latitude

~Oklahoma City, OK

-104.02 Longitude

30.68 Latitude

~McDonald Observatory, TX

Center =617 km

Width = 8 km

Frequency

Distance (km)

0 500 1000 1500 2000

0.0 0.2 0.4 0.6 0.8 1.0

+++

++++

+++

+ + +

+++

++ +

+++

Center = 753 km

Width = 126 km

Right Tail = 53 km

0 500 1000 1500 2000

0.0 0.2 0.4 0.6 0.8 1.0

++ ++ +

++ ++ ++++ + ++ ++ + +++

+ +++ ++++++ + +++++ + ++++++ +++ ++ ++ +++++ ++++++ ++++ + +++ +++ +++ ++ ++ + +++++++++++ + + +++ ++++ + + ++++++++++ +++ + +++ ++++++++ ++

Frequency

Distance (km)

Central Lineages (Pantherophis spiloides x Pantherophis obsoletus)

0 200 400 600 800 1000 1200 1400

0.0 0.2 0.4 0.6 0.8 1.0

+++ ++ + ++

++ +++ +++ + ++ +++ + + +++ +++ +

Western Lineages (Pantherophis bairdi x Pantherophis obsoletus)

Supplemental Material

871

DRYAD XXX

872

Genomic Dataset

873

Geographic Localities

874

R scripts

875

876

Supporting Information Material 1

877

Additional Methods

878

879

Table S1 – Priors and summary stats for PipeMaster; estimating migration, timing of divergence,

880

and historical demography

881

Table S2 – PHRAPL Results (open with R read.table() or Excel)

882

Table S3 – RDA and ANN Results

883

Table S4 – HZAR Results for snapclust assignments

884

Table S5 – HZAR results for DAPC assignments

885

886

Fig.S1. Delimiting populations using sparse nonnegative matrix factorization (SNMF) and

887

spatial Principal Component analysis (sPCA) showing the location of four geographically

888

distinct lineages.

889

890

Fig. S2 – PipeMaster model fit of observed data to simulations under IM, IMD, and IMD-LGM

891

models for PC1-PC10.

892

893

Fig. S3 – Symmetric background niche similarity tests for spatially adjacent pairs (eastern,

894

central, and western lineage comparisons) for the D and I statistics.

895

896

Fig. S4 – Prediction of cline width under neutrality given origin times for each species pair: A)

897

Pantherophis alleghaniensis/P. spiloides, B) P. spiloides/P.obsoletus, and C) P.

898

obsoletus/P.bairdi.

899

900

901

Supporting Information Material 1

902

Supplemental Methods and Results

903

904

Dataset

905

906

We sampled 288 individuals liberally covering the range of all species within the

907

Pantherophis obsoletus complex (Fig.1; Dryad XXX). DNA was extracted from all samples

908

using Qiagen DNeasy Blood & Tissue Kits and samples were screened for quality using broad-

909

range Qubit Assays (https://www.thermofisher.com/us/en/home/industrial/spectroscopy-

910

elemental-isotope-analysis/molecular-spectroscopy/fluorometers/qubit/qubit-assays.html). We

911

used services from RAPiD Genomics (https://www.rapid-genomics.com/services/) to generate

912

5472 baits and to sequence 5060 conserved elements (UCEs) loci following the protocols from

913

(Faircloth et al., 2012) and (Sun et al., 2014).

914

Raw sequence reads were trimmed of adapter contamination using illumiprocessor

915

(Faircloth, 2013), a wrapper around the trimmomatic package (Bolger et al., 2014). These

916

sequence capture data were then assembled by mapping reads to all UCEs found within a

917

Chromium 10x assembled Pantherophis spiloides genome (Burbrink and Myers, in prep) after

918

2,500 base pairs of flanking DNA around UCE baits were pulled from this genome assembly

919

using the Phyluce commands phyluce_probe_run_multiple_lastzs_sqlite and

920

phyluce_probe_slice_sequence_from_genomes (Faircloth, 2016). Sequence reads for each

921

sample were then mapped to the ‘reference’ UCE dataset using bwa v0.7.17 (Li and Durbin,

922

2009). Samtools was used to convert sam format to bam files and alignments were soft clipped

923

using the CleanSam tool in Picard (http://broadinstitute.github.io/picard/). To phase these data,

924

we mapped reads back to each individual sample’s alignments. To do this, bam files were

925

converted to fastq using samtools mpileup (Li et al., 2009), fastq files were converted to fasta

926

format via seqtk (https://github.com/lh3/seqtk) and all missing sites were removed using seqkit

927

(Shen et al 2016). These newly created fasta files were then used as a reference sequence for

928

phasing the data following the seqcap_pop pipeline (Harvey et al., 2016). Briefly we followed

929

this pipeline and used GATK (McKenna et al., 2010) to mark duplicates, call, realign, and

930

annotate/mask indels, call and annotate SNPs via GATK, restrict SNP calling to high quality

931

SNPs, and then conducted read backed phasing. Finally, we created a sample specific fasta file of

932

all phased loci for every sample using ‘add_phased_snps_to_seqs_filter.py’ available via the

933

seqcap_pop pipeline (Harvey et al., 2016). All samples were combined into locus specific fasta

934

files using a perl script developed in (Myers et al., 2019). Locus specific fasta files were aligned

935

using muscle as implemented in the R package ape (Paradis and Schliep, 2019) after removing

936

individuals from alignments that fell below the 25% quantile of the distribution of sequence

937

lengths for each locus. All fasta files were conservatively trimmed of missing data resulting in

938

alignments with no missing sites (i.e., trimming to the shortest sequence in the alignment). We

939

filtered these alignments by removing all loci that were missing >50% sequenced samples and

940

removed all individuals that were missing from >30% of all alignments. Finally, we created a vcf

941

file from these alignments and using vcftools (Danecek et al., 2011) and filtered this for minor

942

allele frequency >10% retaining only 1 SNP/locus for population assignment analyses (Linck

943

and Battey, 2019).

944

945

946

Geographic groupings

947

We estimated population structure by comparing Discriminant Analysis of Principal Components

948

(DAPC; Jombart et al., 2010) and spatial Principal Component analysis (sPCA; Jombart et al.,

949

2008) in adegenet v2.1.2(Jombart 2008), sparse nonnegative matrix factorization (SNMF;

950

Frichot et al., 2014) in LEA v1.4.0 (Frichot and François 2015), and estimated effective

951

migration surfaces (EEMS v; Petkova et al., 2016), which accounts for IBD . Each of these

952

methods use different assumptions for grouping individuals and we compare congruence among

953

them.The model-free method DAPC sequentially estimates K-means for clustering and selects

954

the best fit to infer genetic clusters in the absence of prior group identification. Using the

955

package adegenet (Jombart 2008) in R (R Core Team 2015) we first transformed data using PCA

956

by choosing a large number of PC axes (n=200), then picking the number of discriminant

957

functions yielding large F-statistics (>2000) and optimal number of groups using BIC (Bayesian

958

inference criterion). Because using a large number of the PCs may yield arbitrary solutions, we

959

estimated the optimal a-score, which measures this bias by calculating the difference between

960

the actual cluster assignment probabilities and randomly assigned probabilities (predicting 15-23

961

PCs). We also used cross-validation with a 90% training and 10% test dataset to estimate the

962

average predicted success for each group.

963

Because DAPC does not account for spatial structure, such as IBD, we also used sPCA.

964

Similar to DAPC, this method does not require data to be in Hardy-Weinberg or linkage

965

equilibrium. sPCA also adds an element of space by isolating global structure, representing

966

disconnected groups or clines, from local structure, representing repulsion (selection against

967

similar genetic types co-occurring), and random noise. Using adegenet, we estimated sPCA

968

given SNPs and locations of each sample using up to 15 positive eigenvalues (global structures)

969

and 15 negative eigenvalues (local structures). To connect individual locations to build a

970

connection network, we used Delaunay triangulation. We also implemented Monte-Carlo tests

971

with 100 permutations using the function spca_randtests to determine if significant local or

972

global spatial structures exist.

973

We also assessed ancestral coefficients using SNMF in the package LEA (Frichot and

974

François 2015). This methodology is comparable to ADMIXTURE (Alexander et al. 2009) and

975

Structure (Pritchard et al. 2000) methods, but runs 10-30x faster using genomic-scale data. This

976

method produces a least-squares estimate of ancestry populations given K ancestral populations

977

We estimated ancestry coefficients over 1-6 K using 100 repetitions per K, 100 iterations per

978

algorithm, and a tolerance of 0.05. To determine the appropriate number of groups, we used a

979

cross-validation technique with the entropy criterion, which masks genotypes to fit a particular

980

model to each K.

981

Finally, to compare these methods of grouping individuals into populations, we estimated

982

migration rates over the landscape using EEMS (Petkova et al. 2016). This method models

983

effective migration rates over geography to represent regions where migration is low in cases

984

when genetic dissimilarity increases rapidly, thus providing a uniquely distinct view of the

985

location of population clusters relative to biogeographic barriers as compared to DAPC, sPCA,

986

and SNMF. We ran EEMS 3 times for 3x106 generations with burnin set at 1x106 generations,

987

thinned by 9,999 generations, with 1,000 demes covering the known range of the P. obsoletus

988

complex (Burbrink 2001).

989

Both EEMS and DAPC find population structure, generally matched the geographic

990

ranges of P. alleghaniensis, P. spiloides, P. obsoletus, and P. bairdi (Fig. 1 and Fig. S1;

991

Burbrink, 2001). Total DAPC assignment probabilities were 0.975 (P. bairdi = 1.0, P. obsoletus

992

= 0.96; P. spiloides = 1.0, and P. alleghaniensis = 0.97). Comparing these assignments to those

993

from SMNF, we found that only P. alleghaniensis was misclassified as P. spiloides 8.5% of the

994

time. Similarly, SNMF cross validation predicted four groups (cv = 0.599) as compared to two

995

(cv=0.627), three (cv=0.607), and five (cv=0.601). These four groups generally match the same

996

ranges as those found in DAPC. sPCA showed the presence of four groups, with PC1 showing

997

the separation of groups at the MR and the two lineages east of this river, whereas PC2-5 shows

998

divergence in all taxa. Importantly, global spatial structures were significant (observed spatial

999

structure test statistic = 75.47, random expectation = 28.14; P < 0.009). Similarly, all three

1000

EEMS runs suggest low migration at the MR and the area separating the P. bairdi and P.

1001

obsoletus in west Texas. East of the MR, estimated low migration occurred near the Appalachian

1002

Mountains, though this area is a complex mixture of isolation and gene flow (Fig. 1). All three

1003

EEMS generated similar acceptance proportions for all proposal types (12-51%).

1004

1005

Running SNAPP

1006

We set forward and reverse mutation rates to 1. We applied a gamma distribution (α = 2, β =200)

1007

for the speciation rate (λ) prior and the snapprior was set to α = 1, β =250, and κ = 1. To test the

1008

four alternative species delimitation models including four taxa, three taxa (collapsing P.

1009

bairdi/P. obsoletus or P. alleghaniensis/P. spiloides), and two taxa (collapsing P. bairdi/P.

1010

obsoletus and P.alleghaniensis/P. spiloides we used a stepping-stone analysis with the

1011

PathSampleAnalyser with 72 steps and a chain length of 200000, burnin percentage at 50%, and

1012

a preBurnin of 10,000. We checked for stationarity using Tracer v1.7.1 (Drummond and

1013

Rambaut 2007) determining that estimated sample size (ESS) were > 200 for all parameters.

1014

Running PHRAPL

1015

PHRAPL uses gene-tree distances between simulated and empirical data to rank models using

1016

approximate likelihoods and AICs. To generate the empirical observations we estimated

1017

unrooted gene trees for each loci using IQ-Tree 1.6.12 (Nguyen et al. 2015) and testing all

1018

substitution models per locus prior to estimation. We rooted the trees using midpoint rooting and

1019

performed 100 subsamples of two individuals per population. To reduce model space, we

1020

allowed only one population size parameter (all populations have equal size) and one migration

1021

parameter (all migration rates are equal). We then ran 1000 simulations per model, searching

1022

over a grid of four CollapseStarts values and five MigrationStarts values.

1023

1024

Niche models

1025

We determined if the four taxa show uniquely distinct niches. We quantified niche

1026

similarity using methods introduced in Warren et al. (2008) using Schoener’s D and Hellinger

1027

distance I (Schoener 1968). We used the georeferenced genetic samples here as inputs,

1028

predictions of species identity from DAPC, and 19 bioclimatic variables cropped to the eastern

1029

half of the US (Fick and Hijmans 2017). Estimates of D and I between adjacent species pairs

1030

were calculated and significance of niche identity was derived from 500 random permutations of

1031

lineage identity and re-estimation of the ecological niche model (ENM) using maxent (Phillips

1032

and Dudik 2008) in the R program ENMTools (Warren et al. 2010). Because underlying

1033

differences between lineage-pair regions may not be different from the exact locations of

1034

individuals estimating niche identity we used the background test to produce a null distribution

1035

of ENM symmetrically between a lineage and randomized occurrence points within the adjacent

1036

range of the other adjacent lineage (using a radius of 20km and 1000 points). Therefore, if D and

1037

I from the identity tests were significantly lower than the background test null distribution then

1038

similarity between species, provided that accessible habitat exists in the region, is rejected

1039

(Warren et al. 2008).

1040

Identity tests for niche overlap for both the D and I statistics were significant (P <

1041

1.6x10-12 - 4.02 x10-214) for all adjacent pairs. Dissimilarity for the background tests was

1042

significantly lower than the null distributions for all pairs of taxa (P=0.003 to 0.019; Fig. S3).

1043

These results suggests niches overlap between pairs of taxa less than expected by chance.

1044

1045

1046

Spatial genetics using RDA

1047

We predicted genetic distances from environmental, elevation, river, and distance variables using

1048

RDA in the R package Vegan (Dixon 2003), which avoided problems with spatial data using

1049

Mantel tests (Legendre et al. 2015). This method was used to compare results from the ANN

1050

analyses. RDA is an asymmetric canonical analysis here used to determine the significance of

1051

predictor axes (Legendre et al. 2011) at generating predict genetic structure over a landscape

1052

(McGaughran et al. 2014; Noguerales et al. 2016). We ran this using the capscale function where

1053

dissimilarity data were ordinated with metric scaling and then passed to RDA. All variable

1054

distances were used to predict genetic distance and significance of each variable was tested using

1055

anova-like permutation tests to examine the joint effect of predictor variables given partial

1056

effects from spatial distance to account for IBD. Also, given that some of these variables might

1057

be correlated, we assessed multicollinearity and reran RDA using the reduced number of

1058

predictor variables. To examine how multicollinearity among variables affects model inference

1059

given linear dependence among these independent variables (De Veaux and Ungar 1994; Hastie

1060

et al. 2001; Shan et al. 2006; Dormann et al. 2013), we used stepwise variable inflation factors in

1061

the package Rnalytica (https://rdrr.io/github/software-analytics/Rnalytica/). The function

1062

stepwise.vif selects non-correlated variables by excluding those with the highest variable

1063

inflation factor above five.

1064

Estimates of accuracy using ANN were high (>90%) for all comparisons and yielded

1065

similar conclusions to RDA. Here, RDA demonstrated that most of the structure for all taxa can

1066

be predicted by the Mississippi River, MIS19 (climate at 787Kya), and elevation (P=0.001-

1067

0.002) and these predictors similarly showed 100% variable importance using ANN (Table S3).

1068

Distance also played a role in structuring these data, though they were partialed out in RDA. For

1069

those lineages east of the MR, population structure was predicted by LGM and elevation (RDA:

1070

P = 0.001, ANN: variable importance=100). For those lineages west of the MR, we

1071

demonstrated that MIS19 and elevation structure these populations (RDA: P= 0.001, ANN:

1072

variable importance = 100).

1073

1074

Hyrbid zone modeling in HZAR using snaplust and DAPC assignments:

1075

We also fit clines for the same species pairs using individual genetic assignments: pure parental

1076

(0,1), backcross (0.25, 0.75), and F1 (0.5,0.5) and, alternatively, using adegenet prediction

1077

probabilities in the program HZAR v.2-9 (Derryberry et al. 2014). Using the Gaussian cline

1078

model, we estimated the center and width of the cline and determined if these sigmoidal

1079

distributions have significant tails by fitting the following models: 1) no tails, 2) right tail only,

1080

3) left tail only, 4) mirrored tails, and 5) both tails estimated independently (see Derryberry et al.

1081

2014). We fit these models to our data using AICc and ran the MCMC chains for 5x106

1082

generations, thinned by 5x103 generations, and estimated stationarity using ESS >200 in the R

1083

package CODA (Plummer et al. 2006).

1084

We found agreement for best-fit model between spaclust and DAPC assignments and

1085

parameter estimates. For the eastern lineages, HZAR showed strong support for a right-tailed

1086

model (ΔAICsnapclust =-1.91 - -6.08, ΔAICDAPC =-31.96 - -107.42), whereas central lineages at

1087

the MR (ΔAICsnapclust =-4.28 - -8.70, ΔAICDAPC = -4.26 - -8.869) and western lineages at the

1088

Desert and Edward’s Plateau (ΔAICsnapclust =-5.51 - -221.35, ΔAICDAPC = -5.51 - -221.35) both

1089

supported a no-tails model (Fig. 6; Table S4 & S5).

1090

1091

1092

Genomics and the ICZN

1093

Another problem likely to be encountered by other researchers using genomic data is that

1094

previously named type specimens at particular type localities may be from regions that show

1095

some admixture. For instance, the type locality of the eastern-most species, P. alleghaniensis is

1096

“the summit of the Blue Ridge in Virginia and the Highlands of the Hudson” (Holbrook 1836;

1097

Schultz 1996). Both areas are likely composed of admixed individuals and the International Code

1098

of Zoological Nomenclature (ICZN) forbids naming species based on hybrids (Article

1099

1.3.3;ICZN, 1999) though the intention of this ICZN article did not consider proportions of

1100

admixture. One possible solution would be to use the next available name from Florida, where

1101

no admixture is present (eliminating P. quadrivittata), which would be P.deckerti (type locality:

1102

Lower Matecumber Key, FL; Brady 1932). Another solution would be to designate a Neotype

1103

from a locality showing no admixture and retain P. alleghaniensis. If P. alleghaniensis/P.

1104

spiloides were not considered unique, then P. alleghaniensis has priority by 18 years (Holbrook

1105

1836; Duméril et al. 1854).

1106

The type locality for the remaining taxa, P. spiloides (New Orleans, Louisiana; Duméril

1107

et al. 1854), P. obsoletus (On the Missouri River from the Vicinity Isle au Vache to Council

1108

Bluff; Say, 1823), and P. bairdi (Fort Davis, Apache Mountains, Jeff Davis Co.: Texas; Yarrow

1109

1880) are all from regions that likely represent individuals not admixed (Fig.1 & 5, Burbrink,

1110

2001). Species delimitation methods, estimates of fixation for loci, and observation that these

1111

four organisms have remained geographically distinct in the face of gene flow throughout the

1112

Pleistocene suggests that these taxa should be considered unique species though hybrid zones

1113

exist. We acknowledge difficulties recognizing the eastern lineages as distinct and could argue

1114

for recognizing them as a single taxon, P. alleghaniensis.

1115

1116

Table S1. Priors and summary stats for PipeMaster for the IM, IMD, IMD-LGM models;

1117

estimating migration, timing of divergence, and historical demography.

1118

1119

1120

PARAMETER

PRIOR.1

PRIOR.2

DISTRIBUTION

Ne0.pop1

20000

500000

runif

Ne0.pop2

20000

5e+05

runif

Ne0.pop3

20000

5e+05

runif

Ne0.pop4

20000

5e+05

runif

mig0.1_2

runif

mig0.1_3

runif

mig0.1_4

runif

mig0.2_1

runif

mig0.2_3

runif

mig0.2_4

runif

mig0.3_1

runif

mig0.3_2

runif

mig0.3_4

runif

mig0.4_1

runif

mig0.4_2

runif

mig0.4_3

runif

join1_2

30000

1000000

runif

join3_4

30000

1000000

runif

join2_4

500000

1700000

runif

1121

IMD

1122

PARAMETER

PRIOR.1

PRIOR.2

DISTRIBUTION

Ne0.pop1

10000

500000

runif

Ne0.pop2

10000

500000

runif

Ne0.pop3

10000

500000

runif

Ne0.pop4

10000

500000

runif

mig0.1_2

rtnorm

mig0.1_3

rtnorm

mig0.1_4

rtnorm

mig0.2_1

rtnorm

mig0.2_3

rtnorm

mig0.2_4

rtnorm

mig0.3_1

rtnorm

mig0.3_2

rtnorm

mig0.3_4

rtnorm

mig0.4_1

rtnorm

mig0.4_2

rtnorm

mig0.4_3

rtnorm

Ne1.pop1

1000

10000

runif

Ne1.pop2

1000

10000

runif

Ne1.pop3

1000

10000

runif

Ne1.pop4

1000

10000

runif

t.Ne1.pop1

3000

170000

runif

t.Ne1.pop2

3000

170000

runif

t.Ne1.pop3

3000

170000

runif

t.Ne1.pop4

3000

170000

runif

join1_2

30000

1000000

runif

join3_4

30000

1000000

runif

join2_4

500000

1700000

runif

1123

IMD-LGM

1124

PARAMETER

PRIOR.1

PRIOR.2

DISTRIBUTION

Ne0.pop1

10000

500000

runif

Ne0.pop2

10000

500000

runif

Ne0.pop3

10000

500000

runif

Ne0.pop4

10000

500000

runif

mig0.1_2

rtnorm

mig0.1_3

rtnorm

mig0.1_4

rtnorm

mig0.2_1

rtnorm

mig0.2_3

rtnorm

mig0.2_4

rtnorm

mig0.3_1

rtnorm

mig0.3_2

rtnorm

mig0.3_4

rtnorm

mig0.4_1

rtnorm

mig0.4_2

rtnorm

mig0.4_3

rtnorm

Ne1.pop1

1000

10000

runif

Ne1.pop2

1000

10000

runif

Ne1.pop3

1000

10000

runif

Ne1.pop4

1000

10000

runif

t.Ne1.pop1

3000

170000

runif

t.Ne1.pop2

3000

170000

runif

t.Ne1.pop3

3000

170000

runif

t.Ne1.pop4

3000

170000

runif

mig1.1_2

runif

mig1.1_3

runif

mig1.1_4

runif

mig1.2_1

runif

mig1.2_3

runif

mig1.2_4

runif

mig1.3_1

runif

mig1.3_2

runif

mig1.3_4

runif

mig1.4_1

runif

mig1.4_2

runif

mig1.4_3

runif

t.mig1.1_2

8700

runif

t.mig1.1_3

8700

runif

t.mig1.1_4

8700

runif

t.mig1.2_1

8700

runif

t.mig1.2_3

8700

runif

t.mig1.2_4

8700

runif

t.mig1.3_1

8700

runif

t.mig1.3_2

8700

runif

t.mig1.3_4

8700

runif

t.mig1.4_1

8700

runif

t.mig1.4_2

8700

runif

t.mig1.4_3

8700

runif

join1_2

30000

1000000

runif

join3_4

30000

1000000

runif

join2_4

500000

1700000

runif

1125

Table S2 – PHRAPL Results. Available upon request (open with R read.table() or Excel).

1126

1127

Table S3. Results from RDA (F and P values) and ANN (variable importance [VAR_IMP]) for

1128

climate, distance, Mississippi River, and elevation variables . Significance of variables after

1129

variable inflation factors (VIF) are considered.

1130

1131

VARIABLES

RDA F VALUE

RDA P VALUE

P_VALUE_VIF

NN_VAR_IMP

NN_VAR_IMP_VIF

ALL TAXA

MISSISSIPPI

14.1683

0.0001

0.001

100

MPW

1.3642

0.06

LGM

1.6455

0.015

LIG

1.4935

0.024

MIS

1.7603

0.005

0.001

100

CURRENT

1.8522

0.009

ELEVATION

0.9627

0.458

0.002

100

DISTANCE

Factored

100

EAST_OF THE_MS

LGM

1.7148

0.001

100

LIG

2.0434

0.001

100

MIS

3.1272

0.001

CURRENT

1.8641

0.001

ELEVATION

1.0464

0.275

0.001

100

DISTANCE

Factored

100

WEST_OF_THE_MS

LGM

1.4705

0.013

LIG

1.1802

0.192

100

MIS

0.9094

0.632

0.001

100

CURRENT

0.9583

0.533

ELEVATION

0.626

0.984

0.31

100

DISTANCE

Factored

100

1132

1133

Table S4. Results from cline modeling using HZAR for each species pair grouping using

1134

snapclust ancestry estimates

1135

1136

PANTHEROPHIS_ALLEGHANIENSIS_X_P.SPILOIDES

MODEL

Center

Width

Tail_Right

Tail_Left

Tail_Mirror

AICc

ESS

NONE

729 (723-735)

141 (127-157)

-65.158

717-5200

RIGHT

729 (723-736)

141 (126-157)

859 (375-1599)

-67.067

530-37423

LEFT

729 (723-735)

141 (126-157)

1143 (593-1679)

-60.988

829-53146

MIRROR

729 (723-735)

141 (126-157)

880 (251-1636)

-61.575

321-53591

BOTH

730 (724-736)

141 (126-157)

769 (147-1562)

1226 (719-1723)

-62.885

358-23637

PANTHEROPHIS_SPILOIDES_X_P.OBSOLETUS

MODEL

Center

Width

Tail_Right

Tail_Left

Tail_Mirror

AICc

ESS

NONE

801 (795-807)

100 (90-110)

-535.0162

14399-95534

RIGHT

801 (795-808)

100 (90-110)

1290 (721-1816)

-530.734

676-79146

LEFT

801 (795-808)

100 (90-110)

1276 (709-1835)

-530.735

20873-100521

MIRROR

801 (795-807)

100 (90-110)

1307 (762-1848)

-530.732

10695-120834

BOTH

802 (795-808)

99 (90-110)

1299 (770-1830)

885 (323-1638)

-526.3181

612-32572

PANTHEROPHIS_BAIRDI_P.OBSOLE TUS

MODEL

Center

Width

Tail_Right

Tail_Left

Tail_Mirror

AICc

ESS

NONE

617 (575-670)

7 (3-13)

-150.277

5-1e5

RIGHT

617 (575-670)

10 (4-17)

1204 (613-1803)

-144.766

14-109080

LEFT

188 (635-2094)

70 (6-123)

647 (440-970)

71.072

3-523

MIRROR

642 (590-876)

12 (5-182)

757(37-1578)

58.219

1-915

BOTH

616 (574-660)

9 (4-16)

1218 (617-1810)

1198 (609-1810)

-138.23

9-98456

1137

1138

1139

1140

1141

1142

1143

1144

1145

1146

1147

1148

1149

1150

1151

1152

Table S5. Results from cline modeling using HZAR for each species pair grouping using

1153

adegenet ancestry estimates.

1154

1155

PANTHEROPHIS_ALLEGHANIENSIS_X_P.0SPILOIDES0

MODEL0

Center&

Width&

Tail_Ri ght&

Tail_Left&

Tail_Mi rror&

AICc&

ESS&

NONE0

938&(918-954)&

552&(507-599)&

62.806&

1393-2686&

RIGHT0

753(747-760)&

141&(126-157)&

53&(45-61)&

-44.61&

970-3782&

LEFT0

912&(883-939)&

610&(544-681)&

188&(59-1433)&

-60.988&

129-514&

MIRROR0

760&(746-911)&

155&(116-502)&

91&(66-177)&

64.53&

6-350&

BOTH0

750&(740-757)&

120&(64-136)&

53&(47-60)&

819&(278-1559)&

-12.65&

33-1717&

PANTHEROPHIS_SPILOIDES_X_P.OBSOLETUS0

MODEL0

Center&

Width&

Tail_Ri ght&

Tail_Left&

Tail_Mi rror&

AICc&

ESS&

NONE0

797&(790-803)&

97&(89-107)&

-528.48&

12066-16809&

RIGHT0

797&(790-803)&

98&(88-107)&

11335&(809-1857)&

-524.22&

363-29907&

LEFT0

797&(790-803)&

98&(89-107)&

1281&(718-1841)&

-524.22&

4278-36224&

MIRROR0

797&(790-803)&

98&(89-107)&

1302&(763-1851 )&

-524.22&

3565-41252&

BOTH0

797&(791-804)&

97&(88-106)&

1327&(791-1864)&

890&(235-1646)&

-519.79&

184-7761&

PANTHEROPHIS_BAIRDI_P.OBSO LETUS0

MODEL0

Center&

Width&

Tail_Ri ght&

Tail_Left&

Tail_Mi rror&

AICc&

ESS&

NONE0

617&(575-660)&

7&(4-13)&

-150.277&

10-34416&

RIGHT0

617&(575-661)&

10&(5-20)&

1204&(595-1793)&

-144.766&

6-24620&

LEFT0

652&(595-1874)&

7&(3-81)&

647&(465-1515)&

71.072&

7-400&

MIRROR0

617&(574-661)&

11&(5-19)&

1219(622-1807)&

-144.76&

9-38425&

BOTH0

617&(574-660)&

8&(3-13)&

1211&(619-1806 )&

1235&(635-1824 )&

-138.29&

2-367770&

1156

1157

1158

1159

1160

1161

1162

1163

1164

1165

1166

1167

1168

1169

1170

1171

1172

1173

1174

1175

1176

1177

1178

1179

1180

1181

1182

1183

1184

1185

Fig.S1. Delimiting populations using sparse nonnegative matrix factorization (SNMF) and

1186

spatial Principal Component analysis (sPCA) showing the location of four geographically

1187

distinct lineages.

1188

1189

1190

1191

1192

1193

1194

1195

1196

1197

1198

1199

1200

Fig. S2 – PipeMaster model fit of observed data to simulations under IM, IMD, and IMD-LGM

1201

models for PC1-PC10.

1202

−4

0 10 20

PC1

PC2

IMD

IMD-LGM observed

−4

0 10 20

PC1

PC3

−5.0

−2.5

0.0

2.5

0 10 20

PC1

PC4

−3

0 10 20

PC1

PC5

−4

−2

0 10 20

PC1

PC6

−4

−2

0 10 20

PC1

PC7

−2.5

0.0

2.5

0 10 20

PC1

PC8

−2

0 10 20

PC1

PC9

−2

0 10 20

PC1

PC10

Observed

Fig. S3 – Symmetric background niche similarity tests for spatially adjacent pairs (eastern,

1203

central, and western lineage comparisons) for the D and I statistics.

1204

1205

Fig. S4 – Prediction of cline width under neutrality given origin times for each species pair: A)

1206

Pantherophis alleghaniensis/P. spiloides, B) P. spiloides/P.obsoletus, and C) P.

1207

obsoletus/P.bairdi.

1208

0 1 2 3 4 5 6

0 1000 2000 3000 4000 5000

Cline Width (km)

0 2 4 6 8 10

0 1000 2000 3000 4000 5000 6000

0 1 2 3

0 1000 2000 3000

Time (Years x 100,000)

median dispersal

max dispersal

median dispersal

max dispersal

median dispersal

max dispersal

Pleistocene

Actual cline width

Pleistocene

Actual cline width

Pleistocene

Actual cline width

P.alleghaniensis

P.spiloides

P.obsoletus

P.bairdi

1209

Phylogenetics of Mud Snakes (Squamata: Serpentes: Homalopsidae): A Paradox of Both Undescribed Diversity and Taxonomic Inflation

Article

Full-text available

Feb 2021

Mud snakes (Serpentes: Homalopsidae) are a family of 54 described, mainly aquatic, species primarily distributed throughout mainland Southeast Asia and the Indo-Australian Archipelago. Although they have been the focus of prior research, the basic relationships amongst genera and species remain poorly known. We used a combined mitochondrial and nuclear gene dataset to infer their phylogenetic relationships, using the highest levels of taxon and geographic sampling for any homalopsid phylogeny to date (64% generic and 63% species coverage; 140 individuals). Our results recover two reciprocally monophyletic groups: the fangless Brachyorrhos and its sister clade comprised of all rear-fanged homalopsids. Most genera and interspecific relationships were monophyletic and strongly supported, but intergeneric relationships and intraspecific population structure lack support. We find evidence of both undescribed diversity as well as cases of taxonomic inflation within several species. Tree-based species delimitation approaches (mPTP) support potential new candidate species as distinct from their conspecifics and also suggest that many named taxa may not be distinct species. Divergence date estimation and lineage-through-time analyses indicate lower levels of speciation in the Eocene, with a subsequent burst in diversification in the Miocene. Homalopsids may have diversified most rapidly during the Pliocene and Pleistocene, possibly in relation to tectonic shifts and sea-level fluctuations that took place in Sundaland and the Sahul Shelf. Our analyses provide new insights on homalopsid taxonomy, a baseline phylogeny for the family, and further biogeographic implications demonstrating how dynamic tectonics and Quaternary sea level changes may have shaped a widespread, diverse family of snakes.

Evidence for rangewide panmixia despite multiple barriers to dispersal in a marine mussel

Article

Full-text available

Dec 2017

Oceanographic features shape the distributional and genetic patterns of marine species by interrupting or promoting connections among populations. Although general patterns commonly arise, distributional ranges and genetic structure are species-specific and do not always comply with the expected trends. By applying a multimarker genetic approach combined with Lagrangian particle simulations (LPS) we tested the hypothesis that oceanographic features along northeastern Atlantic and Mediterranean shores influence dispersal potential and genetic structure of the intertidal mussel Perna perna. Additionally, by performing environmental niche modelling we assessed the potential and realized niche of P. perna along its entire native distributional range and the environmental factors that best explain its realized distribution. Perna perna showed evidence of panmixia across >4,000 km despite several oceanographic breaking points detected by LPS. This is probably the result of a combination of life history traits, continuous habitat availability and stepping-stone dynamics. Moreover, the niche modelling framework depicted minimum sea surface temperatures (SST) as the major factor shaping P. perna distributional range limits along its native areas. Forthcoming warming SST is expected to further change these limits and allow the species to expand its range polewards though this may be accompanied by retreat from warmer areas.

Zoogeographic implications of the Mississippi River basin

Article

Full-text available

Jan 1986

Henry W. Robison

Comparative phylogeography clarifies the complexity and problems of continental distribution that drove A. R. Wallace to favor islands

Article

Full-text available

Jul 2016

Brett R Riddle

Deciphering the geographic context of diversification and distributional dynamics in continental biotas has long been an interest of biogeographers, ecologists, and evolutionary biologists. Thirty years ago, the approach now known as comparative phylogeography was introduced in a landmark study of a continental biota. Here, I use a set of 455 studies to explore the current scope of continental comparative phylogeography, including geographic, conceptual, temporal, ecological, and genomic attributes. Geographically, studies are more frequent in the northern hemisphere, but the south is catching up. Most studies focus on a Quaternary timeframe, but the Neogene is well represented. As such, explanations for geographic structure and history include geological and climatic events in Earth history, and responses include vicariance, dispersal, and range contraction-expansion into and out of refugia. Focal taxa are biased toward terrestrial or semiterrestrial vertebrates, although plants and invertebrates are well represented in some regions. The use of various kinds of nuclear DNA markers is increasing, as are multiple locus studies, but use of organelle DNA is not decreasing. Species distribution models are not yet widely incorporated into studies. In the future, continental comparative phylogeographers will continue to contribute to erosion of the simple vicariance vs. dispersal paradigm, including exposure of the widespread nature of temporal pseudocongruence and its implications for models of diversification; provide new templates for addressing a variety of ecological and evolutionary traits; and develop closer working relationships with earth scientists and biologists in a variety of disciplines.

Deep Learning for Population Genetic Inference

Article

Full-text available

Mar 2016
PLOS COMPUT BIOL

Given genomic variation data from multiple individuals, computing the likelihood of complex population genetic models is often infeasible. To circumvent this problem, we introduce a novel likelihood-free inference framework by applying deep learning, a powerful modern technique in machine learning. Deep learning makes use of multilayer neural networks to learn a feature-based function from the input (e.g., hundreds of correlated summary statistics of data) to the output (e.g., population genetic parameters of interest). We demonstrate that deep learning can be effectively employed for population genetic inference and learning informative features of data. As a concrete application, we focus on the challenging problem of jointly inferring natural selection and demography (in the form of a population size change history). Our method is able to separate the global nature of demography from the local nature of selection, without sequential steps for these two factors. Studying demography and selection jointly is motivated by Drosophila, where pervasive selection confounds demographic analysis. We apply our method to 197 African Drosophila melanogaster genomes from Zambia to infer both their overall demography, and regions of their genome under selection. We find many regions of the genome that have experienced hard sweeps, and fewer under selection on standing variation (soft sweep) or balancing selection. Interestingly, we find that soft sweeps and balancing selection occur more frequently closer to the centromere of each chromosome. In addition, our demographic inference suggests that previously estimated bottlenecks for African Drosophila melanogaster are too extreme.

Dispersal Predicts Hybrid Zone Widths across Animal Diversity: Implications for Species Borders under Incomplete Reproductive Isolation

Article

Mar 2020

Hybrid zones occur as range boundaries for many animal taxa. One model for how hybrid zones form and stabilize is the tension zone model, a version of which predicts that hybrid zone widths are determined by a balance between random dispersal into hybrid zones and selection against hybrids. Here, we examine whether random dispersal and proxies for selection against hybrids (genetic distances between hybridizing pairs) can explain variation in hybrid zone widths across 131 hybridizing pairs of animals. We show that these factors alone can explain ∼40% of the variation in zone width among animal hybrid zones, with dispersal explaining far more of the variation than genetic distances. Patterns within clades were idiosyncratic. Genetic distances predicted hybrid zone widths particularly well for reptiles, while this relationship was opposite tension zone predictions in birds. Last, the data suggest that dispersal and molecular divergence set lower bounds on hybrid zone widths in animals, indicating that there are geographic restrictions on hybrid zone formation. Overall, our analyses reinforce the fundamental importance of dispersal in hybrid zone formation and more generally in the ecology of range boundaries.

Phylogeography of the Rufous Vanga and the role of bioclimatic transition zones in promoting speciation within Madagascar

Article

Jun 2019

Madagascar is known as a biodiversity hotspot, providing an ideal natural laboratory for investigating the processes of avian diversification. Yet, the phylogeography of Madagascar's avifauna is still largely unexamined. In this study, we evaluated phylogeographic patterns and species limits within the Rufous Vanga, Schetba rufa, a monotypic genus of forest-dwelling birds endemic to the island. Using an integrative taxonomic approach, we synthesized data from over 4000 ultra-conserved element (UCE) loci, mitochondrial DNA, multivariate morphometrics, and ecological niche modeling to uncover two reciprocally monophyletic, geographically circumscribed, and morphologically distinct clades of Schetba. The two lineages are restricted to eastern and western Madagascar, respectively, with distributions broadly consistent with previously described subspecies. Based on their genetic and morphological distinctiveness, the two subspecies merit recognition as separate species. The bioclimatic transition between the humid east and dry west of Madagascar likely promoted population subdivision and drove speciation in Schetba during the Pleistocene. Our study is the first evidence that an East-West bioclimatic transition zone played a role in the speciation of birds within Madagascar.

ape 5.0: an environment for modern phylogenetics and evolutionary analyses in R

Article

Jul 2018
BIOINFORMATICS

After more than fifteen years of existence, the R package ape has continuously grown its contents, and has been used by a growing community of users. The release of version 5.0 has marked a leap towards a modern software for evolutionary analyses. Efforts have been put to improve efficiency, flexibility, support for 'big data' (R's long vectors), ease of use, and quality check before a new release. These changes will hopefully make ape a useful software for the study of biodiversity and evolution in a context of increasing data quantity. Availability: ape is distributed through the Comprehensive R Archive Network: http://cran.r-project.org/package=apeFurther information may be found athttp://ape-package.ird.fr/.

Do ecological communities disperse across biogeographic barriers as a unit?

Article

Apr 2017
MOL ECOL

Biogeographic barriers have long been implicated as drivers of biological diversification, but how these barriers influence co-occurring taxa can vary depending on factors intrinsic to the organism and in their relationships with other species. Due to the interdependence among taxa, ecological communities present a compelling opportunity to explore how interactions among species may lead to a shared response to historical events. Here we collect single nucleotide polymorphism (SNP) data from five commensal arthropods associated with the Sarracenia alata carnivorous pitcher plant, and test for co-diversification across the Mississippi River, a major biogeographic barrier in the southeastern United States. Population genetic structure in three of the ecologically dependent arthropods mirrors that of the host pitcher plant, with divergence time estimates suggesting two of the species (the pitcher plant moth Exyra semicrocea and a flesh fly Sarcophaga sarraceniae) dispersed synchronously across this barrier along with the pitcher plant. Patterns in population size and genetic diversity suggest the plant and ecologically dependent arthropods dispersed from east to west across the Mississippi River. In contrast, species less dependent on the plant ecologically show discordant phylogeographic patterns. This study demonstrates that ecological relationships may be an important predictor of co-diversification, and supports recent suggestions that organismal trait data should be prominently featured in comparative phylogeographic investigations. This article is protected by copyright. All rights reserved.

Isolation by distance

Article

S. Wright

Multispecies coalescent delimits structure, not species

Article

Jan 2017

Significance Despite its widespread application to the species delimitation problem, our study demonstrates that what the multispecies coalescent actually delimits is structure. The current implementations of species delimitation under the multispecies coalescent do not provide any way for distinguishing between structure due to population-level processes and that due to species boundaries. The overinflation of species due to the misidentification of general genetic structure for species boundaries has profound implications for our understanding of the generation and dynamics of biodiversity, because any ecological or evolutionary studies that rely on species as their fundamental units will be impacted, as well as the very existence of this biodiversity, because conservation planning is undermined due to isolated populations incorrectly being treated as distinct species.

Resolving spatial complexities of hybridization in the context of the gray zone of speciation in North American ratsnakes ( Pantherophis obsoletus complex)

Abstract

Recommended publications

Resolving spatial complexities of hybridization in the context of the gray zone of speciation in Nor...

Ecological Divergence and the History of Gene Flow in the Nearctic Milksnakes (Lampropeltis triangul...

Biogeographic Barriers, Pleistocene Refugia, and Climatic Gradients in the Southeastern Nearctic Dri...

Environmental Heterogeneity and Not Vicariant Biogeographic Barriers Generate Community Wide Populat...