Molecular Ecology (2016) 25, 4611–4631 doi: 10.1111/mec.13789

Population genomic evidence for multiple Pliocene refugia in a montane-restricted harvestman (Arachnida, Opiliones, Sclerobunus robustus) from the southwestern United States

SHAHAN DERKARABETIAN,*† MERCEDES BURNS,* JAMES STARRETT* and MARSHAL HEDIN* *Department of Biology, San Diego State University, 5500 Campanile Dr., San Diego, CA 92182-4614, USA, †Department of Biology, University of California Riverside, Riverside, CA 92521, USA

Abstract The integration of ecological niche modelling into phylogeographic analyses has allowed for the identification and testing of potential refugia under a hypothesis-based framework, where the expected patterns of higher genetic diversity in refugial popula- tions and evidence of range expansion of nonrefugial populations are corroborated with empirical data. In this study, we focus on a montane-restricted cryophilic harvest- man, Sclerobunus robustus, distributed throughout the heterogeneous Southern Rocky Mountains and Intermontane Plateau of southwestern North America. We identified hypothetical refugia using ecological niche models (ENMs) across three time periods, corroborated these refugia with population genetic methods using double-digest RAD- seq data and conducted population-level phylogenetic and divergence dating analyses. ENMs identify two large temporally persistent regions in the mid-latitude highlands. Genetic patterns support these two hypothesized refugia with higher genetic diversity within refugial populations and evidence for range expansion in populations found outside hypothesized refugia. Phylogenetic analyses identify five to six genetically divergent, geographically cohesive clades of S. robustus. Divergence dating analyses suggest that these separate refugia date to the Pliocene and that divergence between clades pre-dates the late Pleistocene glacial cycles, while diversification within clades was likely driven by these cycles. Population genetic analyses reveal effects of both isolation by distance (IBD) and isolation by environment (IBE), with IBD more impor- tant in the continuous mountainous portion of the distribution, while IBE was stronger in the populations inhabiting the isolated sky islands of the south. Using model-based coalescent approaches, we find support for postdivergence migration between clades from separate refugia.

Keywords: ddRAD-seq, isolation by environment, landscape genomics, refugial persistence, sky islands, southwestern United States Received 28 January 2016; revision received 11 July 2016; accepted 19 July 2016

and analytical methods (McCormack et al. 2013; Introduction Edwards et al. 2015; Garrick et al. 2015), and the quanti- The field of phylogeography has evolved considerably tative integration of geological, ecological and climatic since the original pioneering work of Avise et al. (1987). data, for example, have all served to elevate phylogeog- Improvements in sequence data collection techniques raphy as a multidisciplinary science (Avise 2009; Hick- erson et al. 2010). One such improvement has been the Correspondence: Shahan Derkarabetian, Fax: +1 619 594 5676; incorporation of environmental and bioclimatic data, in E-mail: [email protected] the form of ecological niche models (ENMs), into

© 2016 John Wiley & Sons Ltd 4612 S. DERKARABETIAN ET AL. phylogeographic and population genetic analyses (Car- regions of the SRMIP are relatively few. The vast major- stens & Richards 2007; Kozak et al. 2008; Chan et al. ity of phylogeographic and population genetic studies 2011). Not only do ENMs allow researchers to recon- of southwestern taxa analyse genetic differentiation and struct the full potential geographic distribution of a spe- population structure focusing only on small portions of cies, but also, with the use of reconstructed past the SRMIP, mainly the Madrean sky islands of southern bioclimatic conditions, the ostensible distribution of that AZ and NM (e.g. Barber 1999; Masta 2000; Downie species through time (Graham et al. 2010). The ability to 2004; Smith & Farrell 2005; McCormack et al. 2008; Ten- identify persistent and stable habitats has allowed nessen & Zamudio 2008; Ober et al. 2011; Bryson et al. researchers to adopt a hypothesis-based approach, 2013; Mitchell & Ober 2013; Hendrixson et al. 2015; where persistent regions are interpreted as hypothe- Manthey & Moyle 2015; Ober & Connolly 2015), or the sized refugia that produce predictable patterns of Southern Rocky Mountains (e.g. DeChaine & Martin genetic diversity (Carnaval et al. 2009). Hypotheses can 2005a; Knowles et al. 2007; Massatti & Knowles 2014). be generated regarding the expected genetic patterns of Only a single phylogeographic study of Sciurus aberti a species, which are then tested using empirical molecu- (Lamb et al. 1997) spans the entire mountainous SRMIP lar data (e.g. Devitt et al. 2013). region. Studies in this region attribute the depth of dif- Phylogeographic studies have confirmed several pre- ferentiation between sky island populations to the late dictable genetic patterns regarding populations that Pleistocene glacial cycles occurring <1.6 MA. A subset have persisted in refugia and those that have expanded of these studies have explicitly conducted divergence out of refugia into previously uninhabitable regions. dating analyses and concluded that isolated sky island Populations that have persisted in refugia will tend to populations are older than expected given the favour- show more diversity relative to populations outside able conditions for gene flow present during the last hypothesized refugia (e.g. Carnaval et al. 2009; Boyer glacial maximum (LGM) ~20 KA. Studies in other taxa et al. 2016). Conversely, populations inhabiting regions have also generally challenged the importance of late not predicted to be persistent will show evidence of Pleistocene glaciations (e.g. Klicka & Zink 1997). In range expansion, where there is a pattern of decreasing addition, the arthropods of the southwestern Madrean genetic diversity as geographic distance from the refu- sky islands are recently receiving much attention with gia increases (e.g. Gomez et al. 2005; Vandewoestijne & sustained research focusing on various aspects of their Van Dyck 2010; Jezkova et al. 2015). Here, populations biology and systematics (Moore et al. 2013; Meyer et al. at the expanding front show lower levels of diversity 2015). due to successive founder events by populations mov- Opiliones (more commonly called harvestmen) are ing into previously uninhabited areas (Hewitt 2000, arachnid taxa typically characterized by low vagility 2004). In temperate taxa of the Northern Hemisphere, and high ecological constraints. These biological charac- this is typically manifested as a latitudinal gradient due teristics make them ideal candidates for phylogeo- to postglacial northward expansion, where genetic graphic and biogeographic studies (e.g., Boyer et al. diversity decreases with increasing latitude as taxa 2007, 2016; Thomas & Hedin 2008; Giribet et al. 2010; moved northward following the retreat of glaciers. The Hedin et al. 2012; Schonhofer€ et al. 2013, 2015; DiDome- timing of diversification should be contemporaneous nico & Hedin 2016; Emata & Hedin 2016; Starrett et al. with the age of hypothetical refugia, and in the case of 2016). One such candidate is the genus Sclerobunus multiple hypothesized refugia, phylogeographic struc- Banks 1893, with 12 species distributed throughout ture should additionally reflect the separate histories of western North America. The widespread species Scler- distinct refugia (e.g. Hugall et al. 2002; Martınez-Solano obunus robustus (Packard 1877), first described from Col- et al. 2006; Devitt et al. 2013). Taken together, the detec- orado, was later redescribed and recorded with a very tion of these genetic patterns can corroborate the pres- broad distribution throughout the American southwest ence of refugia and elucidate the phylogeographic and Rocky Mountains (Briggs 1971). Following taxo- history of taxa and of focal regions and are particularly nomic changes proposed in a generic revision (Derkara- evident in dispersal-limited taxa. betian & Hedin 2014), this species is distributed The southwestern United States provides an excellent throughout the mountainous habitat of the SRMIP region for phylogeographic and landscape genetic stud- region. In the southern portion of this region, suitable ies due to its highly heterogeneous topography with habitat is found on the extremely discontinuous stretches of habitat of varying continuity. Here, we refer Madrean sky islands, isolated mountains surrounded to the mountains of Arizona (AZ), Colorado (CO), New by uninhabitable, low-lying desert and scrub grassland, Mexico (NM) and Utah (UT) as the Southern Rocky while further north the Southern Rocky Mountains are Mountains and Intermontane Plateau (SRMIP). Studies much more continuous with higher connectivity focusing on taxa distributed across all mountainous between suitable habitat. Sclerobunus robustus is a

© 2016 John Wiley & Sons Ltd SOUTHWESTERN MONTANE HARVESTMEN PHYLOGEOGRAPHY 4613 cryophilic microhabitat specialist, typically found Mexico may have acted as refugia during unfavourable underneath rotting logs and rocks (and sometimes in warmer conditions. However, the previous study of caves) in moist, high elevation conifer and aspen forests S. robustus was limited in that it relied almost entirely (Briggs 1971; Derkarabetian et al. 2010). on COI data and included no quantitative assessments A previous phylogeographic analysis of S. robustus of refugia. To establish a more accurate estimate (Derkarabetian et al. 2011) demonstrated that, despite of genealogical relationships and divergence dates of homogeneity in both somatic and genitalic morphology, S. robustus, and to gain a better understanding of the the species is split into eastern and western lineages geologic and climatic processes contributing to popula- and includes six deeply genetically divergent, geo- tion structure, a multilocus data set is needed. In partic- graphically cohesive mitochondrial phylogroups ular, analysing data sets consisting of 100s–1000s of (Fig. 1). Divergence dating analyses based on mitochon- SNPs can lead to better estimates of population genetic drial cytochrome oxidase I (COI) sequence data using statistics even when low numbers of samples are biogeographic and geologic calibrations of outgroup included per population (Willing et al. 2012). taxa show that diversification between these phy- This study employs the recently developed ddRAD- logroups occurred during the Pliocene (~3.8–5.5 MA), seq protocol (Peterson et al. 2012) to obtain hundreds of while diversification within the clades occurred during loci, and associated single nucleotide polymorphisms the Pleistocene. Additionally, based on population (SNPs), which are then used in phylogeographic and genetic statistics and gene tree topologies, Derkara- landscape genomic analyses to explore the phylogeo- betian et al. (2011) suggested that the mid-latitude high- graphic history of the S. robustus complex. Our objec- lands of southern Utah/Colorado and northern New tives are to identify potential refugia, reconstruct a population-level phylogeny, estimate diversification dates and explore how geography and environment have shaped patterns of genetic diversity in S. robustus. Northwestern Specifically, we use ENM reconstructions to identify Southwestern 36 Northeastern regions of habitat that potentially served as refugia and Central Eastern 35 Southeastern 33 examine whether genetic patterns support the presence 34 Haviland 38 and location of these hypothesized refugia. Based on 39 37 previous phylogeographic studies integrating niche 32 31 2 modelling and genetic analyses (e.g. Carnaval et al. 26 27 23 28 2009; Devitt et al. 2013), we expect to see higher levels 1 25 30 4 24 of genetic diversity for populations found within 3 22 29 UT CO 21 hypothesized refugia, evidence of range expansion for AZ NM 14 populations found outside hypothesized refugia, and 15 high measures of phylogeographic structure between refugia. The broad distribution of S. robustus across the 13 SRMIP allows a comparison of genetic patterns in dif- 10 fering geographic backdrops, in this case, the isolated Madrean sky islands of southern Arizona and New Mexico vs. the more contiguous regions of the Southern Rocky Mountains. 5 8 7 18 Materials and methods 19 6 16 12 20 11 Taxon sampling

17 We collected surface-dwelling S. robustus from 39 locali- 200 km 9 ties throughout its distribution (Fig. 1, Table S1, Sup- porting information). In this study, we use the term Fig. 1 Geographic distribution and sampling localities of ‘population’ as an equivalent to all specimens/samples S. robustus, coloured by clades recovered here (see Fig. 3) and from a single collecting locality (i.e. 39 populations). in Derkarabetian et al. (2011). Filled boxes indicate collecting localities where two samples were processed for ddRAD-seq. Specimens were preserved in 100% EtOH. For genetic – Half boxes indicate localities where a single sample was pro- analyses, 1 2 specimens of S. robustus were selected per cessed. Black circles indicate localities where other Sclerobunus population, depending on available sample size, with species have been collected. the exception of San Francisco Mountains, AZ, where

© 2016 John Wiley & Sons Ltd 4614 S. DERKARABETIAN ET AL. three samples were chosen (Fig. 1), for a total of 65 used 77 presence records (25% of which were used as specimens. Additionally, a single sample each of test points), including 43 records from recent fieldwork, S. nondimorphicus and S. idahoensis was included as out- plus 34 unique museum records that either had GPS group taxa. All cave populations of S. robustus were information associated with the specimens or could be excluded from this study as these cave populations are unequivocally georeferenced based on provided locality found at lower elevation and outside habitat deemed information: 19 records from the American Museum of suitable compared with surface-dwelling populations. Natural History, nine from the Museum of Comparative The bioclimatic variables representative of these cave Zoology and six from the Museum of Southwestern localities will not reflect the actual habitat requirements Biology. These records were accessed through Symbiota of surface Sclerobunus. Additionally, those cave popula- Collections of Arthropods Network database (SCAN; tions that have been included in phylogenetic analyses http://symbiota4.acis.ufl.edu/scan/portal). All Maxent are extremely recent derivations and not genetically dis- settings were left at default values. To extract biocli- tinct from geographically proximate surface popula- matic information for each collection locality, all 19 bio- tions. Cave populations, and their genetic affinities with climatic variables were added as layers to ArcMap surface populations, are more thoroughly discussed in implemented through ArcGIS Desktop version 10.3 Derkarabetian et al. (2010, 2011). (ESRI), and values were extracted using the ‘Extract Values to Points’ function. We used this data set to reduce the number of bioclimatic variables included in Identification of putative refugia the final model by implementing the method of Jezkova To identify regions of habitat that most likely served as et al. (2011) to identify highly correlated variables. This refugia, we operated under the premise that potential method generates a simplified climatic model with lar- refugia will be regions where climate reconstructions gely independent variables, reducing overparameteriza- support suitable habitat that are persistent through time tion. This resulted in variables 10 and 11 (mean and variable climatic conditions (Kozak et al. 2008). This temperature of warmest and coldest quarters) and 14, general framework has been utilized numerous times to 16 and 18 (precipitation of driest, wettest and warmest identify potential refugia (e.g. Knowles et al. 2007; Car- quarters) being removed. The final model included 14 naval et al. 2009; Graham et al. 2010; Chan et al. 2011). bioclimatic variables, in addition to altitude. Addition- For example, Devitt et al. (2013) used bioclimatic data to ally, the Community Climate System Model (CCSM4) reconstruct niche models for three time periods (cur- data sets were used at a downscaled resolution of 2.5 rent, Holocene, LGM data sets), which were then com- arc-minutes to model distributions during the LGM bined to identify regions of suitable habitat shared (~22 000 years ago) and mid-Holocene (~6000 years across time. In addition to reconstructing ENMs across ago) using only the 14 retained bioclimatic variables. these time periods, we conducted separate ENM analy- We combined the results of six separate ENM analy- ses on localities split into eastern and western lineages, ses where ecological niches were estimated for eastern which are both well-supported clades (Derkarabetian and western clades across the three temporal data sets. et al. 2011). Given their similar geographic distributions, All ASCII MAXENT results were opened as raster files in we treat the eastern and western lineages as ‘latitudinal ArcMap where they were converted to binary pres- replicates’ and assume that habitats predicted as suit- ence/absence maps using a threshold as determined by able to both lineages across time would more likely rep- the 10% minimum training logistic threshold. To iden- resent refugia. tify regions of overlap, binary maps were added ENMs were reconstructed in MAXENT version 3.3.3k together using the ‘Raster Calculator’ in the Spatial (Phillips et al. 2006) using 30 arc-seconds resolution data Analyst toolbox. sets of altitude and 19 bioclimatic variables (WorldClim version 1.4; Hijmans et al. 2005). While S. robustus likely Molecular data collection and bioinformatics selects suitable microhabitat (i.e. niche) on a fine scale, preferring the underside of fallen bark, rotting logs and Genomic DNA was extracted individually from whole rocks, these preferred microhabitats are restricted to specimens, with intestines removed, using a Qiagen moist high elevation forests, which are more directly DNeasy Blood and Tissue Kit (Qiagen, Valencia, CA, influenced by the coarse bioclimatic variables used in USA) as per manufacturer’s protocol. Libraries for ENM reconstructions. As such, we believe that the stan- ddRAD-seq were prepared using a modified version of dard bioclimatic variables have utility in this system, Peterson et al.’s (2012) protocol. Details of the library and other taxa with similar microhabitat specificities preparation protocol, including primers and adaptors (e.g. Rissler & Apodaca 2007; Bond & Stockman 2008; used, and sequencing methods can be found in Material Bond 2012; DiDomenico & Hedin 2016). In total, we S1 and Table S2 (Supporting information). Raw read

© 2016 John Wiley & Sons Ltd SOUTHWESTERN MONTANE HARVESTMEN PHYLOGEOGRAPHY 4615 data were analysed using the PYRAD version 3.0.5 soft- of Oregon and Washington (S. nondimorphicus) and the ware (Eaton 2014). This program assembles short read Northern Rocky Mountains of Idaho (S. idahoensis)isa RAD-seq data using graph-based clustering algorithms well-known vicariant event found in many other taxa that allow for indels and consists of seven steps: demul- (Brunsfeld et al. 2001; Shafer et al. 2010) and is dated at tiplexing, quality filtering, within sample clustering, 2–5 MA, coinciding with the formation of the Columbia error rate and heterozygosity estimation, base calling Plateau and subsequent rain shadow caused by the ris- and paralog detection, between sample clustering, and ing of the Cascade Mountains (Graham 1999). These alignment across samples. Following extensive prelimi- two species are 7.6% divergent for COI (uncorrected nary testing, the parameters used in all final analyses distance, Derkarabetian et al. 2010) indicating a rela- were as follows: maximum number of sites with quality tively older divergence that, coupled with estimated score <20 (NQual): 5, clustering threshold (Wclust): COI divergence rates of 3.54% per million years for 0.95, minimum coverage for a cluster (Mindepth): 6, arthropods (Papadopoulou et al. 2010) and 2.25% per maximum individuals with a shared heterozygous site million years for spiders (Bidegaray-Batista & Arnedo (MaxSH): 3. Multiple data sets were produced which 2011), correspondingly dates to this geologic event. varied in the minimum taxon coverage needed for a Although geologic evidence is not ideal for calibrating locus to be included (MinCov). MinCov values were 65, divergence analyses (Heads 2010), fossils are not known 48, 32, 16 and 4, although we only discuss the results of for this genus (or family). However, many studies of the 32 (~50% coverage) and 16 (~25% coverage) MinCov harvestmen taxa with similar biological characteristics data sets (hereafter called the m32 and m16 data sets, (i.e. low vagility microhabitat specialists with high eco- respectively). All other parameters were kept at default logical constraints) indicate that they are ideally suited values. PYRAD was run on the University of California, and highly informative for biogeographic studies across Riverside Institute for Integrative Genome Biology many timescales (e.g. Boyer et al. 2007; Boyer & Giribet (UCR-IIGB) Linux cluster. Unless indicated otherwise, 2009; Hedin et al. 2012; Schonhofer€ et al. 2013). all downstream analyses used matrices consisting of A PYRAD analysis using S. nondimorphicus, S. idahoensis S. robustus samples only. and three divergent S. robustus samples resulted in only 33 loci shared across the three species and could not be used to estimate a rate across all three species directly. Phylogenetics and divergence time estimation However, the general characteristics (e.g. average muta- Phylogenetic reconstruction and divergence dating tion rate, divergence) of ddRAD-seq loci across Scler- were conducted with the BEAST version 2.2.1 package obunus species should be similar within Sclerobunus (Bouckaert et al. 2014) on concatenated matrices where species. The level of genetic divergence for COI across assembled sequence reads were used as loci. We relied S. robustus (~8.3%) is greater than the COI divergence on a concatenation approach, as opposed to species between S. nondimorphicus and S. idahoensis (7.6%; tree methods, as this was required for estimating Derkarabetian et al. 2011), indicating that S. robustus divergence times using BEAST. We suggest an alterna- consists of deep species level divergences comparable tive approach in the Discussion. PARTITIONFINDER version to other Sclerobunus species. The average rate of evolu- 1.1.1 (Lanfear et al. 2012) was used to select the appro- tion across the S. nondimorphicus and S. idahoensis sam- priate model of evolution for the entire concatenated ples was calculated using a concatenated matrix of 945 matrix. Preliminary BEAST analyses were run, aimed at RAD loci (89 008 bp) by calibrating the split as a nor- testing strict vs. relaxed clock models, and Yule vs. mal distribution with mean of 3.5 and sigma value of coalescent tree models on the S. robustus concatenated 0.75. data set. These were run for 2 million generations Preliminary analyses using relaxed clock models sampling every 1000 with a 10% burn-in period. Multi- failed to converge, and analyses using the coalescent ple identical analyses each with four chains were con- model produced much older and largely unrealistic ducted to confirm similarity of tree topology and results (e.g. the root of S. robustus at ~27 MA). The Yule posterior distributions of parameters between indepen- model may be more appropriate as previous research dent runs. These preliminary analyses produced identi- (Derkarabetian et al. 2011), and ongoing studies with cal topologies. this data set indicate that S. robustus is likely a cryptic To provide a prior for the rate of evolution for the species complex. Final divergence dating analyses used estimation of divergence dates with the ddRAD-seq the strict clock, calibrated Yule and a nucleotide substi- data set, the single samples of S. nondimorphicus and tution model chosen with PARTITIONFINDER. Six indepen- S. idahoensis were used in a separate analysis to calcu- dent analyses were run for 10 million generations each, late a ‘background Sclerobunus ddRAD-seq rate’. The logging every 1000 steps. All six runs were combined geographic disjunction between the Cascade Mountains with LOGCOMBINER with 10% burn-in, and the resulting

© 2016 John Wiley & Sons Ltd 4616 S. DERKARABETIAN ET AL. clockRate parameter mean and variance were used as geographic plot (e.g. latitude and longitude) and an a priori calibration for divergence analyses using the rotated to maximize similarity. m32 and m16 S. robustus data sets. For the S. robustus divergence time analyses, the strict clock and Yule mod- Genetic, geographic and environmental correlations els were used along with the appropriate nucleotide model chosen by PARTITIONFINDER. The clockRate prior Geographic distance is commonly incorporated into was assigned a normal distribution (see results) match- explaining patterns of distribution and genetic differen- ing that identified in the biogeographic calibration anal- tiation in the form of isolation by distance (IBD; Wright ysis and ran for 10 million generations, logging every 1943; Rousset 1997), where the degree of genetic differ- 1000, with 10% burn-in. TRACER version 1.6 (Rambaut entiation between two populations is correlated with et al. 2014) was used to assess stationarity and check for their geographic distance. Recently, the importance of ESS values above 200, and TREEANNOTATOR version 2 environmental differences has been recognized and (Rambaut & Drummond 2014) was used to produce described as isolation by environment (IBE; Wang & maximum clade credibility trees. The resulting topology Summers 2010; Wang & Bradburd 2014). Many studies of major clades is identical to exploratory analyses have shown the importance of incorporating environ- using loci shared between ingroup samples and out- mental distances in some form when exploring patterns groups (S. nondimorphicus and S. idahoensis). Outgroup of genetic differentiation (e.g. Lee & Mitchell-Olds 2011; taxa were not included in these analyses because of Wang et al. 2013; Manthey & Moyle 2015). We evalu- very few shared loci with the ingroup. ated evidence for IBE with an environmental distance Additionally, we conducted Bayesian Skyline Plot matrix created by first extracting data for each biocli- analyses for all major lineages (Drummond et al. 2005) matic variable from each collecting locality (as for ENM and constructed a lineage through time plot (LTT; Nee analyses). This data set was then subject to PCA using et al. 1992; Harvey et al. 1994) using the R package APE the dudi.pca function in the ADE4 R package (Dray & version 3.0-11 (Paradis et al. 2004). LTT plots were Dufour 2007). The first two principal components were based on phylogenies including only one sample per then used as points to calculate a pairwise distance population. matrix for all populations, and IBE was assessed using IBDWS (Jensen et al. 2005). Additionally, we conducted a multiple regression analysis using the MMRR script Population genomic statistics and population structure (Multiple Matrix Regression and Randomization; Wang To maximize variability in the final data set while 2013). This method allows for tests of correlation simultaneously minimizing missing data for the calcu- between multiple distance matrices, assessing how a lation of population genetics statistics, all analyses dependent variable, in this case genetic distance, used the m32 data set. For all loci, we calculated sev- responds to multiple independent variables (e.g. geo- eral molecular diversity indices (gene diversity, num- graphic and environmental distance), and evaluates sig- ber of differences, heterozygosity, polymorphic loci), nificance through random permutations. All other and average FST for each collecting locality and clade correlations and regressions presented in the results with ARLEQUIN version 3.5.2.2 (Excoffier & Lischer were calculated using Pearson’s product–moment corre- 2010). We conducted an AMOVA to analyse genetic lation coefficient (r) in R 3.0.2 (R Core Team 2013). Any structure between the six major clades. Using a single correlations using within-population statistics were randomly sampled SNP from each variable locus based only on populations in which more than one (1265/1270 loci were variable), we explored popula- sample was successfully sequenced (i.e. singleton popu- tion structure with a principal components analysis lations were excluded). To assess the relationship (PCA) using the dudi.pca and find.clusters functions between geographic distance and genetic distance, we in the R package ADEGENET (Jombart 2008; Jombart & conducted isolation-by-distance (IBD) analyses using Ahmed 2011). Additionally, to examine the correlation IBDWS for all major clades and groupings (i.e. all S. ro- between genetic variation and geography, the first bustus, east vs. east clades, mtDNA clades). Pairwise two principal components and geographic coordinate geographic distances were calculated using the Geo- information for each collecting locality (Table S1, Sup- graphic Distance Matrix Generator version 1.2.3 (Ersts, porting information) were used in a Procrustes analy- American Museum of Natural History, Center for Biodi- sis conducted in the R package VEGAN (Oksanen et al. versity and Conservation, http://biodiversityinformat- 2015) assessing significance with 1000 permutations. ics.amnh.org/open_source/gdmg) where the genetic

Here, genetic variation in the form of the first two distance matrix consisted of pairwise FST values derived principal components is superimposed onto a from Arlequin.

© 2016 John Wiley & Sons Ltd SOUTHWESTERN MONTANE HARVESTMEN PHYLOGEOGRAPHY 4617

Migration estimation 3.0 and 100 000. Stationarity was assessed with ESS val- ues and manual inspection of the posterior distribu- Demographic model testing has been incorporated into tions. The number of migrants per generation (Nm) was phylogeographic analyses such that explicit demo- calculated by multiplying the mutation-scaled effective graphic parameters can be estimated and multiple alter- immigration rate by h, then dividing by an inheritance native phylogeographic scenarios can be modelled and scalar of four for nuclear loci. formally compared (e.g. Richards et al. 2007; Carstens et al. 2013; Barrow et al. 2015). To test for the possibility of migration between distinct refugia, we used the coa- Results lescent-based program FASTSIMCOAL2 (Excoffier et al. 2013). This program uses the site frequency spectrum Identification of potential refugia derived from SNP data to estimate parameters and test The layered binary ENMs identified two large regions observed data against expected data that is estimated that likely supported suitable and climatically stable through user-defined models. We tested two simple habitat in the mid-latitude regions: the San Juan/Tusas divergence models, one without postdivergence migra- Mountains along the western CO/NM border, and the tion and a second including postdivergence migration. southeastern edge of the Mogollon Rim along the AZ/ These models were tested on two pairs of sister clades NM border (Fig. 2). Other studies have suggested that that are hypothesized to have persisted in separate refu- for a given species, more central populations house the gia: the NW–SW clades and the CE–SE clades. We used highest degree of genetic diversity, demonstrating SNP data derived from rerunning step 7 of PYRAD with expansion out of refugia in these ‘mid-latitude’ regions only samples from the focal clades for each set of analy- (e.g. Hoskin et al. 2011; An et al. 2015; Peterson & ses included. MinCov values were set to be equal to Graves 2015). The combined ENMs also identify several 50% of total samples. The observed SFS files were cre- regions of persistent habitat: the Chuska and Zuni ated using a custom script (github.com/jordansatler/ Mountains of NM/AZ and the mountains of southwest- SNPtoAFS). For all analyses, the divergence time (Tdiv) ern UT. There are no known museum records for prior was also fixed to the mean divergence recovered from m32 analyses. We ran two sets of analyses, one in which the mutation rate was estimated, and another set where the mutation rate was fixed at 8.4E-9 mutations per generation based on estimates from Drosophila mela- nogaster (Haag-Liautard et al. 2007). Migration rates were given a relatively diffuse prior with log-uniform distribution (minimum of 1E-5, maximum 1E-2). All runs used the following options: -N 100 000 (maximum number of simulations to perform), -L 40 (maximum number of loops to perform) and -M 0.001 (stopping criterion, as minimum relative difference of parameters between iterations). Each analysis was repeated 20 times, and favoured models were chosen with AIC cal- culations. We also estimated migration rates between these two clade pairs using the program MIGRATE-N (Beerli & Felsenstein 2001; Beerli 2006; Beerli & Palczewski 2010). Sequence data were used as input, created by rerunning PYRAD as in FASTSIMCOAL2 analyses. Base frequencies were calculated from the data, mutation rates across loci were allowed to vary, and starting values for theta (h) and migration (M) were calculated using FST. Based on initial testing, we chose uniform priors for h (min. = 0, max. = 0.01, delta = 0.001) and M (min. = 0, Fig. 2 Binary ENMs for S. robustus. Grey regions denote ENM max. = 1000, delta = 100). Each model was run for based on current bioclimatic conditions across all populations 20 000 steps at in increment of 100, with 20 000 dis- of S. robustus. Black regions indicate persistent habitat across carded as burn-in. A static heating scheme was used eastern and western clades for current, mid-Holocene and LGM niche models. White dots indicate S. robustus collecting with four chains using temperature values of 1.0, 1.5, localities used in constructing ENMs.

© 2016 John Wiley & Sons Ltd 4618 S. DERKARABETIAN ET AL.

Sclerobunus from these areas, and specimens were not major mtDNA clades obtained in Derkarabetian et al. found during recent collecting attempts in the Zuni (2011): northwest (NW), southwest (SW), central eastern Mountains, AZ, and the mountains of southwestern (CE), southeastern (SE) and northeastern (NE). The Utah. Individual ENMs for eastern and western clades HAV mitochondrial clade (Haviland) is monophyletic and across all populations for current and LGM models with high support in the m16 phylogeny, while in the are presented in Figure S1 (Supporting information). m32 phylogeny it is paraphyletic with respect to the NE clade. Relationships among major clades and within- clade assignments recovered with the nuclear data pre- Molecular data sented here are identical to the clades recovered in In total, 67 samples were extracted, consisting of 65 Derkarabetian et al. (2011) using COI, with the excep- S. robustus from 39 collecting localities, one S. nondimor- tion of the placement and monophyly of HAV in the phicus and one S. idahoensis (Table S1, Supporting infor- m32 analysis. mation). On average, for the 65 S. robustus samples, Calibrated BEAST analyses using concatenated loci for sequencing resulted in 906 109.6 reads per sample S. nondimorphicus and S. idahoensis resulted in a mean (range: 172 984–1738 861), with 828 371.1 (91.4%) reads clockRate value of 1.83323E-3 substitutions per site per per sample (range: 155 453–1 596 212) passing quality million years (95% HPD interval 1.0418E-3, 2.8222E-3). control. Only 15 loci (15 unlinked SNPs) were shared Preliminary testing resulted in small (but significantly across all 65 samples of S. robustus. For S. nondimorphi- different from 0) ucld.stdev values (m32: mean 0.314, cus and S. idahoensis, 259 524 (230 004 passing QC) and HPD 0.246-0.388), indicating that the data are evolving 156 163 (141 440 passing QC) reads were sequenced, in almost clock-like pattern, and rate variation across respectively, resulting in 945 shared loci. Matrix statis- lineages is minor. More importantly, mean divergence tics are shown in Table 1, and sample sequencing statis- time and HPD estimates with strict and relaxed clock tics are shown in Table S1 (Supporting information). models were virtually identical. Divergence time estimates within S. robustus using this value as a rate calibration (normal distribution with Phylogenetics and divergence time estimation mean 1.83323E-3 and sigma 0.0004) are shown in Fig. 3 Final matrix statistics are presented in Table 1. All anal- and Table 2. In general, node ages increase with yses used the HKY + I + G substitution model chosen increasing number of loci. Mean ages for the root of for the entire concatenated matrix of 1270 loci by PARTI- S. robustus increased from 3.89 MA (95% HPD: 2.47– TIONFINDER. BEAST produced highly supported, nearly 5.73 MA) in the m32 data set to 4.27 MA (2.71–6.24 identical phylogenies for each data set of varying cover- MA) in the m16 data set. In the m32, the most recent age. In the m32 and m16 phylogenies, support mean divergence between any two major clades occurred increased as the number of loci increased. Across these at 2.72 MA (1.73–4.02 MA) for CE + SE clades, while in data sets, phylogenies were largely identical (Fig. 3 and the m16, this divergence time increased to 2.98 MA Figure S2, Supporting information), recovering the (1.91–4.39 MA). If, in the m32 analysis, NE and HAV are considered separate clades, the earliest between- clade divergence is 2.52 MA (1.57–3.69 MA). The mean Table 1 Matrix statistics for S. robustus only datasets. Matrix age of divergence within the most recently differentiated completeness was calculated by summing the number of loci – across all samples and dividing by the total number of loci clade (SW) is 1.44 MA (0.9 2.13 MA) in the m32, which multiplied by 65 (Table S1). PIS = parsimony-informative sites increased to 1.58 MA (0.98–2.29 MA) in the m16. Associated with an expansion out of refugia, it is m48 m32 m16 expected that there would be increases in both the size of populations and number of lineages postexpansion. Number of loci 498 1270 3303 Across all S. robustus, Bayesian skyline plots show two Mean N loci 428 884 1579 major increases in population size, one at the Pleis- Min 179 275 487 – ~ ~ Max 489 1176 2417 tocene Pliocene boundary 2.5 MA, and another 1.0 Matrix completeness (%) 85.94 69.58 47.79 MA (Fig. 3 inset). Similarly, skyline plots for all major Total variable sites 4686 11 431 24 232 clades, except NE + HAV, show an increase in popula- Total PIS 2277 5541 11 453 tion size with a major increase ~1.0 MA (Figure S3, Sup- Sampled unlinked SNPs 496 1265 3251 porting information). Here, the expectation of Sampled unlinked bi- 285 617 1355 population size increase is recovered where increases allelic SNPs are seen following isolation in separate Pliocene refugia Concatenated matrix 47 061 120 197 312 584 length and during Pleistocene glacial cycles. While the LTT plot (Fig. 3 inset) does show slight increases in lineage

© 2016 John Wiley & Sons Ltd SOUTHWESTERN MONTANE HARVESTMEN PHYLOGEOGRAPHY 4619

Time (MA) 5 4 3 2 1 0 Pliocene OP1130 (1) NW OP1202 (2) OP1148 (4) OP1149 (4) OP1173 (3) OP1174 (3) OP905 (5) SW OP951 (6) OP952 (6) OP916 (10) OP914 (10) OP915 (10) OP921 (7) OP2977 (9) OP2978 (9) OP935 (8) OP938 (8) OP964 (11) OP965 (11) OP945 (12) OP946 (12) CE OP2114 (14) OP877 (15) OP886 (13) OP887 (13) OP929 (16) OP1040 (17) SE OP1041 (17) OP893 (18) OP894 (18) OP889 (19) OP956 (20) HAV OP959 (20) OP1178 (21) OP1179 (21) OP1223 (22) OP1197 (23) OP1209 (24) OP1210 (24) OP2108 (25) NE OP2109 (25) OP1188 (26) OP1164 (27) OP1168 (27) OP1145 (28) OP2682 (29) 1.E0 OP2683 (29) Plio. Pleist. OP2104 (30) 40 OP2105 (30) OP1214 (31) OP1215 (31) 0 OP1122 (32) 30 OP1123 (32) OP2679 (33) OP1193 (34) 1.E-2 OP1194 (34) 20 OP2686 (35) OP1183 (36) Lineages OP1184 (36) 1.E-3 10 OP1135 (37) OP1136 (37) OP1218 (39) Effective population size Effective OP1219 (39) 0 1.E-4 OP1140 (38) 3 2 1 0 OP1141 (38)

Fig. 3 BEAST maximum clade credibility tree for the m32 data set. Clade colours correspond to those in Fig. 1. Node bars represent 95% HPD for divergence times. Blue HPD divergence bars indicate nodes with >0.95 posterior probability, and red bars indicate <0.95. Inset: Bayesian Skyline for all S. robustus (thick line with confidence intervals), and lineage through time plot (thin line) for one sample per population. numbers corresponding to these increases in population Genetic, geographic and environmental correlations size, particularly at ~1.0 MA, these increases do not sig- Molecular diversity indices and pairwise F estimates nificantly deviate from the null expectation of a con- ST for each clade are presented in Tables 3 and 4. The first stant rate of lineage diversification (Figure S3, two principal components from PCAs of 1265 SNPs Supporting information).

© 2016 John Wiley & Sons Ltd 4620 S. DERKARABETIAN ET AL.

Table 2 Mean estimated divergence dates for major clades and groupings, with 95% HPD in parentheses

m48 m32 m16

S. robustus root 3.212 [2.003, 4.713] 3.886 [2.473, 5.727] 4.266 [2.709, 6.242] Western clades 2.461 [1.543, 3.633] 2.776 [1.751, 4.098] 3.061 [1.928, 4.475] NW 1.761 [1.106, 2.628] 1.895 [1.196, 2.809] 2.117 [1.332, 3.103] SW 1.33 [0.822, 1.972] 1.441 [0.898, 2.129] 1.576 [0.979, 2.294] Eastern clades 2.646 [1.683, 3.917] 3.504 [2.251, 5.172] 3.956 [2.506, 5.784] CE + SE 2.081 [1.301, 3.078] 2.72 [1.731, 4.017] 2.982 [1.91, 4.386] CE 1.273 [0.775, 1.88] 1.536 [0.957, 2.26] 1.677 [1.049, 2.444] SE 1.355 [0.84, 1.997] 1.761 [1.115, 2.601] 2.003 [1.266, 2.938] HAV + NE 2.128 [1.338, 3.146] 2.8 [1.739, 4.089] 3.157 [1.985, 4.602] HAV NA NA 2.841 [1.791, 4.14] NE 1.311 [0.807, 1.929] 1.68 [1.059, 2.466] 1.967 [1.247, 2.877]

Table 3 Genetic diversity statistics for clades based on the m32 loci. Standard deviations of statistics included in parentheses

N # haplotypes Avg. # diffs Avg. gene diversity # polymorphic loci Expected heterozygosity

NW 6 12 12.7878 (6.164517) 0.021067 (0.011423) 39 0.4501 (0.09533) SW 15 30 9.35172 (4.40795) 0.015356 (0.008051) 69 0.26745 (0.13481) CE 4 8 10 (5.071548) 0.020704 (0.011943) 27 0.53714 (0.08674) SE 8 16 12.9 (6.106719) 0.023412 (0.01246) 58 0.37482 (0.1378) HAV 4 8 24.14285 (11.78276) 0.034539 (0.019157) 61 0.51531 (0.06839) NE 28 56 7.425974 (3.518801) 0.008429 (0.00443) 72 0.15894 (0.10663)

(Fig. 4, Figure S4, Supporting information) show three CE + SE, NE + HAV and NE clades. IBE correlations major groups corresponding to three main clades recov- were significant for most clades, the exceptions being ered in phylogenetic analyses: western (NW + SW), CE + SE clades and the individual NE and HAV clades. CE + SE and NE + HAV. BIC analyses using the MMRR analyses showed that environmental distance is find.clusters function retaining the first 65 components a significant independent predictor of genetic distance favoured six clusters, which corresponded to the across all S. robustus and within the SW and eastern mtDNA clades of previous work. In the m32 analysis clades, while both environment and geography were (Fig. 4), the NW and SW clades are distinct, while in significant and contributed nearly equally in the west- the m16 analysis (Figure S4, Supporting information) ern clade. Within the eastern clade, geographic dis- the HAV samples are separate from the NE samples. tances are the significant predictor of genetic distances Procrustes analyses on both the m32 (Fig. 5) and m16 for the CE + SE, NE + HAV and NE clades. The corre- (Figure S4, Supporting information) recovered signifi- lation between geographic distance and environmental cant similarity scores (>0.77, P-values = 0.001) between distance was small and not significant (r = 0.1097, a population’s location in SNP principal component P = 0.061, 1000 permutations). PCA of bioclimatic vari- space and their geographic location. The mean pairwise ables used in constructing the environmental distance geographic distance of populations in each clade (i.e. matrix largely differentiated the northern (NW, NE, overall size of distribution of a clade) and the mean HAV, CE) and the southern populations (SW, SE) along pairwise FST of those clades show a strong and signifi- PCA 1 (Figure S5, Supporting information), mirroring cant negative correlation (Fig. 6A), indicating that the the latitudinal break seen in the distribution of these clades with the smallest geographic distributions are clades. The northernmost southern population (popula- the most genetically differentiated. The AMOVA testing tion 10 SW clade, Fig. 1) and the southernmost northern genetic structure of the six clades indicated that differ- population (population 13 CE clade, Fig. 1) are in close ences among these clades accounted for the majority of proximity in the PCA (Figure S5, Supporting informa- variation (55.9%; Table 5). tion), reflecting their similar latitudes. Here, it is impor- Results of all IBD and IBE, including MMRR, analy- tant to note that the mountainous regions of New ses are shown in Table S3 (Supporting information). Mexico in between the northern and southern S. robus- IBD analyses recovered small but significant correla- tus clades (i.e. southern Sangre de Cristo Mountains, tions across all S. robustus, and within the western, Mt. Taylor, Sandia Mountains, Manzano Mountains) are

© 2016 John Wiley & Sons Ltd SOUTHWESTERN MONTANE HARVESTMEN PHYLOGEOGRAPHY 4621

Table 4 Pairwise FST values for each clade. Diagonal shows average FST within clades. 95% confidence interval in parentheses

NW SW CE SE HAV NE

NW 0.8914 (0.1029) SW 0.8931 (0.0213) 0.7286 (0.0557) CE 0.9689 (0.0138) 0.9339 (0.0173) 0.9526 (0.0511) SE 0.8867 (0.0312) 0.8220 (0.0399) 0.9040 (0.0375) 0.6583 (0.0956) HAV 0.9605 (0.0173) 0.9230 (0.0152) 0.9757 (0.0171) 0.8945 (0.0373) 0.9627 (0.0379) NE 0.9617 (0.0073) 0.9292 (0.0076) 0.9761 (0.0068) 0.8944 (0.016) 0.9612 (0.0104) 0.8428 (0.0355)

negative latitudinal correlation with genetic diversity NE (Fig. 6B). However, the only clades to show a signifi- HAV cant correlation between latitude and genetic diversity are the eastern (r = À0.9191, P < 0.0001), the SE (r = À0.9849, P = 0.0151) and the SW (r = À0.9449, P = 0.0045) clades. Within the eastern clade, the three CE + SE populations largely drive this correlation, and PCA2 although the NE clade shows a negative correlation, it is not significant (r = À0.2878, P = 0.3643). Within the NW NE + HAV clade, as latitude increases, the divergence CE −20 −10 0 10 20 SW time of a population from its nearest relative decreases SE additionally suggesting a pattern of northward expan- sion (Fig. 7). Conversely, for the SW clade, no signifi- −20 −10 0 10 20 cant correlation exists between divergence time of a PCA1 population from nearest relative and latitude Fig. 4 Results of principal components analysis for the m32 (r = 0.3845, P = 0.347). Correlations between genetic dif- SNP data set. ferentiation and latitude have been found in numerous studies (Sexton et al. 2014), many of which either exclusively inhabited by several other Sclerobunus spe- directly imply or explore this pattern in regard to post- cies (Fig. 1; Derkarabetian & Hedin 2014). glacial range expansion (e.g. Gomez et al. 2005; Vande- Across all S. robustus populations, there is a latitude- woestijne & Van Dyck 2010; Jezkova et al. 2015), where genetic component showing a strong and significant the expanding frontline populations show reduced

Fig. 5 Results of Procrustes analysis for the m32 SNP data set. Left pane: results 40 of Procrustes plot, with arrows indicating direction and magnitude of movement of each point in SNP principal component 38 space. Right pane: simplified geographic distribution of collecting localities. Pro- UT CO crustes plot was rotated 90° to corre- AZ NM spond with placement in the geographic 36 plot. Latitude

34

32 200 km

–112 –110 –108 –106 Longitude

© 2016 John Wiley & Sons Ltd 4622 S. DERKARABETIAN ET AL.

(A) postdivergence migration between refugial clades (Fig. 8A), with extremely low migration estimates. Con- HAV NE versely, FASTSIMCOAL2 analyses that do not use a fixed CE mutation rate favoured the model without postdiver- SW – NW gence migration for the NW SW pair only (Fig. 8B). MIGRATE N Mean Fst - analyses show some degree of migration SE between clade pairs (Table S4, Supporting information). r = −0.9498 A potentially confounding factor for these coalescent 0.6 0.7P = 0.8 0.0037 0.9 1.0 model-based analyses in this taxon may be the high 4.9 5.0 5.1 5.2 5.3 levels of population structure within clades, which may Log mean geographic distance (meters) lead to erroneous inferences of population parameters (B) r = –0.7995 (e.g. Chikhi et al. 2010; Heller et al. 2013). In this case, P = <0.00001 FASTSIMCOAL2 may be inferring the presence of migration to account for the relatively high levels of variation seen within each clade. We interpret the actual estimates of migration from these analyses with caution, given the ultimate uncertainty in generation times and mutation

Gene diversity rates. For these analyses, we assumed a generation time of 1 year. Generation and developmental times for 0 0.005 0.010 0.015 Sclerobunus have not explicitly been examined, but r = –0.7348 P = <0.0001 available evidence from other related harvestmen sug- gests the adult stage can be reached in 1 year (Gnaspini 2007), although S. robustus is known to overwinter as adults (Cokendolpher et al. 1993).

# polymorphic loci Discussion 0246810

32 34 36 38 40 Evidence for multiple refugia Latitude The integration of ENMs and genetic data can be used Fig. 6 Genetic and geographic correlations. (A) Correlation of to corroborate the presence of refugia and can also mean pairwise geographic distance vs. mean pairwise FST for reveal cases where multiple distinct refugia are identi- each clade, with 95% confidence intervals. FST calculated in ARLE- fied, for example, as have been found recently in other QUIN with each population considered a separate sample. (B) Lat- arachnid taxa (Bidegaray-Batista et al. 2016; Wachter itude vs. gene diversity and number of polymorphic loci of each et al. 2016). Previously, Derkarabetian et al. (2011) sug- population represented by more than one sequenced individual. gested the presence of refugia in the mountains of the Genetic statistics calculated using loci from the m32 data set. southern CO and northern NM border region. Our anal- genetic diversity relative to the stable ancestral popula- yses suggest the presence of two large refugial regions tions (Hewitt 2000, 2004). for S. robustus: one in the San Juan Mountains near the western portion of the CO/NM border and a second at the southeastern edge of the Mogollon Rim in southern Migration estimation AZ/NM. Additionally, it is clear that the remaining For both clade pairs, FASTSIMCOAL2 analyses using a fixed mountain ranges in CO were unlikely to serve as refu- mutation rate favoured the model that incorporated gia, suggesting northward expansion of the NE clade.

Table 5 AMOVA results across major clades

Variance Fixation Source of variation d.f. Sum of squares components Percentage of variation indices P-value

< Among clades 5 683.356 6.40967 Va 55.92 0.55917 FCT 0.0001 < Among populations within clades 33 483.357 4.14841 Vb 36.12 0.82095 FSC 0.0001 < Within populations 91 82.333 0.90476 Vc 7.89 0.82095 FST 0.0001 Total 129 1249.046 11.46284

© 2016 John Wiley & Sons Ltd SOUTHWESTERN MONTANE HARVESTMEN PHYLOGEOGRAPHY 4623

r = −0.64689

2.5 P = 0.0028 2.0 1.5 36 1.0

Divergence time Divergence 27

28 0.5 29 30 35 33 37.5 38.0 38.5 39.0 39.5 40.0 VI 34 Latitude 38 38 39 37 37 39 36 32 35 VI 34 33 31 V 32 V 31 30 29 II 26 27 IV 28 23 III 28 27 III 26 I 25 30 IV II 25 24 24 22 23 29 I 22 21 21 CO

3 2 1 0

Fig. 7 Phylogenetic and population genetic patterns for the NE clade. Top left: latitude vs. divergence time of each population from its nearest phylogenetic relative. Bottom left: m32 phylogeny of the NE clade, indicating sublineages in red. Right: geographic distri- bution and hypothesized expansion direction of the NE clade, with sublineages indicated.

The nuclear data presented here corroborate the expan- lack of genealogical exclusivity only in some of the sion of S. robustus out of multiple Pliocene refugia northernmost populations. occurring in these mid-latitude highlands. The San Juan Southwestern Sclerobunus species are diverse within Mountains harbour a high amount of genetic diversity the mid-latitude highlands, specifically the southern with representatives from 4/6 clades within ~150 km of Sangre de Cristo Mountains, which run parallel and are each other. A recent study focusing on dispersal-limited similar in latitude to the San Juan Mountains. In the mite harvestmen of Australia’s wet tropics found simi- Sangre de Cristo Mountains, three other Sclerobunus lar patterns, where higher present day taxonomic diver- species can be found (Derkarabetian & Hedin 2014). sity was statistically correlated with hypothesized Similar spatial patterns in species diversity are seen in climatically stable refugia (Boyer et al. 2016). Addition- montane mammals where the northern sky islands of ally, the three clades with the highest amount of inter- New Mexico hold a higher number of species relative nal genetic differentiation (HAV, CE and NW; Fig. 6A) to southern sky islands (Lomolino et al. 1989; Frey et al. are found almost entirely within the San Juan Moun- 2007). In a study by Lomolino et al. (1989) examining 27 tains. Additional support for this hypothesis can be sky islands across the SRMIP, the Mogollon Mountains seen in the genetic patterns of the NE clade. The NE were recorded as having the highest number of species, clade shows a strong phylogenetic pattern of northward while the other southern sky islands had a much lower expansion out of the San Juan refugium where diver- species richness, adding support to the evidence found gence times of sister populations decrease with increas- here that the Mogollon Mountains might have served ing latitude (Fig. 7), with a corresponding negative (but as a southern refugium for montane-restricted taxa. not significant) correlation of latitude and genetic diver- Population genetic data for the southern clades (SW sity. The previous COI analyses also showed a similar and SE) similarly suggest northward expansion, thus northward phylogenetic pattern for the NE clade, with supporting the second Pliocene refugia identified at the

© 2016 John Wiley & Sons Ltd 4624 S. DERKARABETIAN ET AL.

(A) 488 SNPs 956 SNPs that all areas currently inhabited by S. robustus should be identified as persistent considering that, when diver- Anc. Anc. 6341 44120 gence times are incorporated, all populations were likely present in their current distribution prior to the mid-Holocene and LGM. However, we argue that the use of downscaled LGM and Holocene data decreases sensitivity as downscaled data will not capture the NW SW CE SE smaller restricted patches of habitat that S. robustus 1.74E –10 3.2E –8 likely persisted in during unfavourable conditions (e.g. 839163 64458 270251 266 8.97E –7 4.98E –6 within canyons and ravines), and will only capture lar- ger persistent regions. These larger regions are more Tdiv = 2776000 (fixed) Tdiv = 2720000 (fixed) likely to have served as refugia during the warmer Plio- Mutation rate = 8.4E –9 (fixed) cene, for which bioclimatic data do not exist. Taken together, analyses from our study indicate the 488 SNPs 956 SNPs (B) presence of two large refugia in the SRMIP: the San Anc. Anc. Juan Mountains near the western CO/NM border and 31549 43270 the eastern end of the Mogollon Rim. Model-based analyses indicate the presence of some potential migra- tion between sister clades after initial separation in these distinct refugia. ENM predictions for the LGM indicate climatic conditions that likely supported con- NW SW CE SE 7.62E –10 nective habitat between these refugia, particularly for the eastern clades, while these refugia remain largely 6524665 2696954 51976 101 2.68E –5 separate in current and mid-Holocene reconstructions (Figure S1, Supporting information). To our knowledge, Tdiv = 2776000 (fixed) Tdiv = 2720000 (fixed) Mutation rate = 8.1E –10 Mutation rate = 4.47E –8 no other studies specifically indicate presence of mon- tane Pliocene refugia in either of these two regions. Fig. 8 Models chosen with FASTSIMCOAL2. (A) Favoured models Phylogeographic studies identifying refugia in the with a fixed mutation rate. (B) Favoured models with mutation southwest typically focused on relatively young taxa rate estimated. Point estimates of all parameters are included, with any refugia dating to the Pleistocene (e.g. Mitton which were taken from the iteration with the highest likeli- et al. 2000; DeChaine & Martin 2005b; Knowles et al. hood. The number of SNPs used in each analysis based on rerunning PYRAD is indicated above each model. 2007). It is important to note that while mean diver- gence dates between clades occur in the Pliocene, there is still some degree of uncertainty as the HPD values eastern edge of the Mogollon Rim. Both SW and SE partially extend into the early Pleistocene. However, the clades show a significant negative correlation between divergence times recovered using ddRAD-seq data in genetic diversity and latitude, although no clear phylo- this study are younger than, but still largely consistent genetic pattern of directionality is seen in either. How- with, previous work in Sclerobunus using traditional ever for the SW clade, genetic data show expansion out molecular markers (Derkarabetian et al. 2010, 2011). of a refugium that is further south than indicated by Additional phylogeographic studies of other codis- the ENMs. Early Pleistocene expansion resulted in tributed montane taxa of similar age will be needed to widespread but contiguous clades showing a typical corroborate the presence of these Pliocene refugia and pattern of decreasing genetic diversity with increasing whether they have shaped diversification patterns latitude (much like that seen currently in the NE clade). across taxa (e.g. Satler et al. 2016). Following this northward expansion, there was effec- tively simultaneous isolation of sky island populations Spatiotemporal patterns in Sclerobunus robustus that still retains a genetic footprint of northward expan- sion. Both LGM and mid-Holocene models predict The initial separation of S. robustus into eastern and some degree of connectivity between these populations western lineages occurred ~3.89 MA (2.47–5.73 MA). As (Figure S1, Supporting information), although the noted in Derkarabetian et al. (2011), the distributions downscaled data sets may not have the resolution to and break between eastern and western clades largely discern more fine-scale corridors. coincides with that seen in the tassel-eared squirrel Sci- A potential weakness of using an ENM approach for urus aberti found in mixed conifer forests of the SRMIP identifying persistent regions as refugia in S. robustus is (Lamb et al. 1997). Additionally, although currently

© 2016 John Wiley & Sons Ltd SOUTHWESTERN MONTANE HARVESTMEN PHYLOGEOGRAPHY 4625 restricted to the northern portion of the SRMIP, the the western clades with its formation beginning 6 MA montane grasshopper Melanoplus marshalli also shows a (Lucchitta 1989; McMillan et al. 2006). Additionally, cur- split into eastern and western clades (Knowles et al. rent bioclimatic conditions indicate many unsampled 2007). Based on divergence dates and Bayesian skyline sky islands with suitable habitat, several of which are plots, two episodes of population expansion occurred known to have locally derived cave populations of within S. robustus: first, an expansion coinciding with S. robustus (Welbourn 1999), suggesting the past, and the initial divergence of eastern and western lineages potentially present, suitability of habitat on these moun- into subclades that occurred near the Pliocene–Pleis- tains. It is uncertain how far south the southern clades tocene boundary 2.5–2.8 MA. The warmer climate dur- expanded; S. robustus are not known from Mexico, ing the Pliocene likely played a role in separating the although ENMs do indicate suitable habitat in the widespread and contiguous S. robustus into multiple northern Sierra Madre Occidental. units restricted to individual refugia separated by dis- persal barriers. Second, following isolation during war- Patterns of genetic differentiation mer climate, each individual clade then expanded ~1.0 MA during the cooler climate of the late Pleistocene, Across S. robustus clades, there is a pattern of decreasing where glacial cycles contributed to population-level genetic differentiation as overall range size increases divergence (e.g. Knowles 2001; DeChaine & Martin (Fig. 6A), where the most geographically restricted HAV 2005b). In addition, niche models for the LGM suggest and CE clades show the highest levels of internal genetic that the majority of the sampled populations and divergence. The distribution of the HAV clade is highly clades, particularly those in the southern sky islands, restricted showing the smallest distribution (mean pair- remained distinct despite availability of suitable connec- wise distance of ~79 km) yet shows the highest levels of tive habitat. genetic differentiation with a mean pairwise FST of 0.96, Despite uncertainty and regardless of the data set the highest average number of differences and the high- used, the divergence dates between major clades within est levels of gene diversity (Table 3). Conversely, the SW S. robustus are conservatively estimated to occur in the clade is the most widespread clade (~210 mean km pair- late Pliocene to early Pleistocene, indicating this species wise distance) and shows a comparatively lower FST of is relatively old. Conversely, divergences within major 0.73. A similar case is seen in the harvestmen genus clades fall within the late Pleistocene suggesting a role of Cryptomaster, where the relatively widespread the glacial cycles in differentiation among populations C. leviathan shows little intraspecific genetic divergence within major lineages. Although divergence dates esti- across multiple mountain ranges, while the much more mated here are younger than previous studies that used geographically restricted sister species C. behemoth multiple calibration points and a multilocus data set shows higher population genetic structure (Starrett et al. (Derkarabetian et al. 2010) or a COI-only data set 2016). The limited distributions seen in S. robustus clades (Derkarabetian et al. 2011), the general conclusions of are typical of short-range endemic taxa (Harvey 2002; pre-Pleistocene divergences between clades and Pleis- Harvey et al. 2011) with limited dispersal capabilities tocene divergences within clades are consistent across and distributions in discontinuous habitats, which can studies. Within the SW clade distributed throughout the lead to extreme levels of population structure (e.g. Keith Madrean sky islands, divergence dates between popula- & Hedin 2012; Hedin et al. 2015). Many species of Scler- tions in the m32 analysis ranged from 750 KA to 1.4 MA. obunus, and many harvestmen in general, can be consid- These dates largely coincide with other arthropod stud- ered short-range endemics (e.g. Ubick & Briggs 1992, ies in this region. For example, the spider Habronattus 2008; Boyer & Giribet 2009; Richart & Hedin 2013; pugilis (Masta 2000) and beetle Moneilema appresum Derkarabetian & Hedin 2014; DiDomenico & Hedin (Smith & Farrell 2005), both restricted to higher elevation 2016; Emata & Hedin 2016; Starrett et al. 2016). The nega- habitats in the Madrean sky islands, are relatively young tive correlation of range size and genetic differentiation species with population divergences occurring between is clear in S. robustus; however, it will be interesting to 300 KA and 1.7 MA. Similarly, other studies have sug- explore this pattern in other short-range endemics. gested much more recent, but still largely pre-LGM, In a recent meta-analysis, Sexton et al. (2014) found divergence estimates of <100 KA for the beetle Scaphino- that the majority of studies implicate IBE as a signifi- tus petersi (Ober et al. 2011; Mitchell & Ober 2013). cant factor in genetic differentiation either alone or in ENMs show suitable habitat throughout the moun- combination with IBD. Here, we show that IBD and IBE tainous regions of southwestern UT. Sclerobunus are not both play a role in shaping genetic patterns, although known from these mountains, and recent collecting in different clades of S. robustus. Across all populations, attempts were unsuccessful. In this case, the Colorado geographic and environmental distances both contribute River likely acted as a barrier to westward dispersal for to genetic differentiation of S. robustus, although

© 2016 John Wiley & Sons Ltd 4626 S. DERKARABETIAN ET AL. multiple regression analyses indicate that environmen- mutation rate of S. robustus. Very few loci were shared tal distance is a better predictor of genetic differentia- between the interspecific Sclerobunus taxa and S. robus- tion (Table S3, Supporting information). However, tus; therefore, the loci from the two interspecific samples within the major clades, the relative importance of these from which the mutation rate was based on may be variables seems to depend on the landscape connectiv- intrinsically different (divergence levels, mutation rate) ity. The eastern subclades (CE + SE and NE + HAV) from the loci of the 65 intraspecific S. robustus samples. suggest geographic distance as a stronger determinant The divergence levels between the interspecific of genetic patterns, particularly evident in the Sclerobunus are comparable to divergences levels within NE + HAV clade where populations are less genetically S. robustus, assuaging this discrepancy. A potential differentiated and found in highly contiguous habitat. shortcoming in our method is our choice of concatena- Conversely, in the western clade, environmental dis- tion of ddRAD-seq loci. Specifically, those loci that are tance plays a more important role in shaping genetic added when producing an m16 data set are likely be differentiation where the vast majority of populations more divergent than those loci that are only found in the are found on isolated sky islands separated by uninhab- m48 data set, as fewer individuals will have the ‘m16 itable low-lying desert and scrub grassland. Several loci’ due to allelic dropout. We recover an interesting studies have shown a similar situation where patterns pattern where estimated divergence dates increased of genetic differentiation are dependent on the continu- when the minimum taxon coverage value decreased ity of habitat (e.g. Keyghobadi et al. 2005; Alberto et al. (Table 2). When increasing the threshold for missing 2010; D’Aloia et al. 2014). These findings highlight the data (e.g. from m48 to m16), more divergent loci are importance of examining the full distribution of a spe- included in the data set. As demonstrated for ddRAD- cies when exploring the effects of geography and envi- seq in Arnold et al. (2013), more divergent loci that are ronment on the genetic differentiation of a species. sampled less will tend to show deeper divergence times, The contrasting genetic patterns seen between the SW and therefore increase node depths across the phy- and NE clades show two different temporal snapshots of logeny. Partitioning loci in such a way to account for the continuous changes in distribution and elevation variation in divergence level may reduce uncertainty in undergone by montane-restricted taxa as a result of the divergence dates across data sets. For example, loci interplay of climatic changes (Hewitt 2000, 2004) and could be partitioned according to which MinCov value niche conservatism (Wiens 2004; Wiens & Graham 2005). adds them to the data set (a partition of m32-only loci, a While both clades are nearly the same age, the genetic partition of the subset of only m48 loci, etc.). Despite patterns and connectivity of the contemporary NE clade these cautions, we reiterate the similarity of divergence may be reminiscent of a former more contiguously dis- dating estimates from ddRAD-seq loci with previous tributed SW clade when climate favoured more wide- estimates using more traditional markers (Derkarabetian spread and lower elevation forests. This formerly et al. 2010, 2011), all of which recover largely Pliocene widespread and continuous SW clade would likely have divergences among major clades. shown significant IBD, much like the current NE clade. However, the importance of IBD becomes insignificant after populations become isolated on sky islands and Acknowledgements local environmental conditions become more variable We thank Andrew Gottscho for initial training in ddRAD-seq across the clade, thus resulting in IBE. As our study sam- library preparation methods and for discussions regarding pled few specimens per population relative to the num- analyses. For assistance with fieldwork, we thank Joe Deas, ber of loci, the inclusion of more individuals per Angela DiDomenico, Damian Elias, Kristen Emata, Ryan Faw- population would allow more fine-scaled analyses of cett, Erika Garcia, Maureen McCormack, Casey Richart, Axel € gene flow and migration in these clades using model- Schonhofer, Dan Sitzmann, Eric Stiner and Steven Thomas. We thank Jordan Satler for providing custom scripts for creating based methods (Knowles & Carstens 2007). input files for fastsimcoal2 as well as helpful discussions. Con- versations with Paul Maier provided some useful insight. This Divergence dating with ddRAD-seq data research was funded by an NSF grant awarded to MH (DEB 1354558). Comments from Bryan Carstens and four anonymous Despite the increasing popularity of ddRAD-seq and reviewers helped to significantly improve the manuscript. similar methods, there are very few studies using these data to estimate divergence times (Stervander et al. 2015; see also Gottscho 2015). To calculate divergence times, References we made the assumption that the estimated mutation Alberto F, Raimondi PT, Reed DC et al. (2010) Habitat continu- rate between S. idahoensis and S. nondimorphicus based ity and geographic distance predict population genetic dif- on a biogeographic calibration is equivalent to the ferentiation in giant kelp. Ecology, 91,49–56.

© 2016 John Wiley & Sons Ltd SOUTHWESTERN MONTANE HARVESTMEN PHYLOGEOGRAPHY 4627

An B, Zhang L, Liu N, Wang Y (2015) Refugia persistence of Boyer SL, Markle TM, Baker CM, Luxbacher AM, Kozak KH qinghai-tibetan plateau by the cold-tolerant bird Tetraogallus (2016) Historical refugia have shaped biogeographical pat- tibetanus (Galliformes: Phasianidae). PLoS ONE, 10, terns of species richness and phylogenetic diversity in mite e0121118. harvestmen (Arachnida, Opiliones, Cyphophthalmi) endemic Arnold B, Corbett-Detig RB, Hartl D, Bomblies K (2013) RAD- to the Australian Wet Tropics. Journal of Biogeography, 7, seq underestimates diversity and introduces genealogical 1400–1411. biases due to nonrandom haplotype sampling. Molecular Briggs T (1971) The harvestmen of the family Triaenonychidae Ecology, 22, 3179–3190. in North America. Occasional Papers of the California Academy Avise JC (2009) Phylogeography: retrospect and prospect. Jour- of Sciences, 90,1–43. nal of Biogeography, 36,3–15. Brunsfeld SJ, Sullivan J, Soltis DE, Soltis PS (2001) Comparative Avise JC, Arnold J, Ball RM et al. (1987) Intraspecific phylo- phylogeography of northwestern North America: a synthesis. geography: the mitochondrial DNA bridge between popula- Special Publication-British Ecological Society, 14, 319–340. tion genetics and systematics. Annual Review of Ecology and Bryson RW, Riddle BR, Graham MR, Smith BT, Prendini L Systematics, 18, 489–522. (2013) As old as the hills: montane scorpions in southwest- Banks N (1893) The phalangida mecostethi of the United ern North America reveal ancient associations between bio- States. Transactions of the American Entomological Society, 20, tic diversification and landscape history. PLoS ONE, 8, 149–152. e52822. Barber PH (1999) Phylogeography of the canyon treefrog, Hyla Carnaval AC, Hickerson MJ, Haddad CF, Rodrigues MT, Mor- arenicolor (Cope) based on mitochondrial DNA sequence itz C (2009) Stability predicts genetic diversity in the Brazil- data. Molecular Ecology, 8, 547–562. ian Atlantic forest hotspot. Science, 323, 785–789. Barrow LN, Bigelow AT, Phillips CA, Lemmon EM (2015) Phy- Carstens BC, Richards CL (2007) Integrating coalescent and logeographic inference using Bayesian model comparison ecological niche modeling in comparative phylogeography. across a fragmented chorus frog species complex. Molecular Evolution, 61, 1439–1454. Ecology, 24, 4739–4758. Carstens BC, Brennan RS, Chua V et al. (2013) Model selection Beerli P (2006) Comparison of Bayesian and maximum-likeli- as a tool for phylogeographic inference: an example from the hood inference of population genetic parameters. Bioinformat- willow Salix melanopsis. Molecular Ecology, 22, 4014–4028. ics, 22, 341–345. Chan LM, Brown JL, Yoder AD (2011) Integrating statistical Beerli P, Felsenstein J (2001) Maximum likelihood estimation of genetic and geospatial methods brings new power to phylo- a migration matrix and effective population sizes in n sub- geography. Molecular Phylogenetics and Evolution, 59, 523–537. populations by using a coalescent approach. Proceedings of Chikhi L, Sousa VC, Luisi P, Goossens B, Beaumont MA (2010) the National Academy of Sciences, 98, 4563–4568. The confounding effects of population structure, genetic Beerli P, Palczewski M (2010) Unified framework to evaluate diversity and the sampling scheme on the detection and panmixia and migration direction among multiple sampling quantification of population size changes. Genetics, 186, 983– locations. Genetics, 185, 313–326. 995. Bidegaray-Batista L, Arnedo MA (2011) Gone with the plate: Cokendolpher JC, MacKay WP, Muma MH (1993) Seasonal the opening of the Western Mediterranean basin drove the population phenology and habitat preferences of montane diversification of ground-dweller spiders. BMC Evolutionary harvestmen (Arachnida: Opiliones) from Southwestern New Biology, 11, 317. Mexico. The Southwestern Naturalist, 38, 236–240. Bidegaray-Batista L, Sanchez-Garcia A, Santulli G et al. (2016) D’Aloia CC, Bogdanowicz SM, Harrison RG, Buston PM (2014) Imprints of multiple glacial refugia in the Pyrenees revealed Seascape continuity plays an important role in determining by phylogeography and paleodistribution modelling of an patterns of spatial genetic structure in a coral reef fish. endemic spider. Molecular Ecology, 25, 2046–2064. Molecular Ecology, 23, 2902–2913. Bond JE (2012) Phylogenetic treatment and taxonomic revision DeChaine EG, Martin AP (2005a) Marked genetic divergence of the trapdoor spider genus Aptostichus Simon (Araneae, among sky island populations of Sedum lanceolatum (Crassu- Mygalomorphae, Euctenizidae). ZooKeys, 252,1–209. laceae) in the Rocky Mountains. American Journal of Botany, Bond JE, Stockman AK (2008) An integrative method for 92, 477–486. delimiting cohesion species: finding the population-species DeChaine EG, Martin AP (2005b) Historical biogeography of interface in a group of Californian trapdoor spiders with two alpine butterflies in the Rocky Mountains: broad-scale extreme genetic divergence and geographic structuring. Sys- concordance and local-scale discordance. Journal of Biogeogra- tematic Biology, 57, 628–646. phy, 32, 1943–1956. Bouckaert R, Heled J, Kuhnert€ D et al. (2014) BEAST 2: a soft- Derkarabetian S, Hedin M (2014) Integrative taxonomy and ware platform for Bayesian evolutionary analysis. PLoS Com- species delimitation in harvestmen: a revision of the western putational Biology, 10, e1003537. North American genus Sclerobunus (Opiliones: Laniatores: Boyer SL, Giribet G (2009) Welcome back New Zealand: regio- Travunioidea). PLoS ONE, 9, e112001. nal biogeography and Gondwanan origin of three endemic Derkarabetian S, Steinmann DB, Hedin M (2010) Repeated and genera of mite harvestmen (Arachnida, Opiliones, Cyphoph- time-correlated morphological convergence in cave-dwelling thalmi). Journal of Biogeography, 36, 1084–1099. harvestmen (Opiliones, Laniatores) from montane western Boyer SL, Clouse RM, Benavides LR et al. (2007) Biogeography North America. PLoS ONE, 5, e10388. of the world: a case study from cyphophthalmid Opiliones, a Derkarabetian S, Ledford J, Hedin M (2011) Genetic diversifica- globally distributed group of arachnids. Journal of Biogeogra- tion without obvious genitalic morphological divergence in phy, 34, 2070–2085. harvestmen (Opiliones, Laniatores, Sclerobunus robustus) from

© 2016 John Wiley & Sons Ltd 4628 S. DERKARABETIAN ET AL.

montane sky islands of western North America. Molecular Graham A (1999) Late Cretaceous and Cenozoic History of North Phylogenetics and Evolution, 61, 844–853. American Vegetation. Oxford University Press, New York, Devitt TJ, Devitt SEC, Hollingsworth BD, McGuire JA, Moritz North of Mexico. C (2013) Montane refugia predict population genetic struc- Graham CH, VanDerWal J, Phillips SJ, Moritz C, Williams SE ture in the Large-blotched Ensatina salamander. Molecular (2010) Dynamic refugia and species persistence: tracking spa- Ecology, 22, 1650–1665. tial shifts in habitat through time. Ecography, 33, 1062–1069. DiDomenico A, Hedin M (2016) New species in the Sital- Haag-Liautard C, Dorris M, Maside X et al. (2007) Direct esti- cina sura species group (Opiliones, Laniatores, Phalangodi- mation of per nucleotide and genomic deleterious mutation dae), with evidence for a biogeographic link between rates in Drosophila. Nature, 445,82–85. California desert and Arizona sky islands. ZooKeys, 586, Harvey MS (2002) Short-range endemism amongst the Aus- 1–36. tralian fauna: some examples from non-marine environ- Downie DA (2004) Phylogeography in a galling insect, grape ments. Invertebrate Systematics, 16, 555–570. phylloxera, Daktulosphaira vitifoliae (Phylloxeridae) in the Harvey PH, May RM, Nee S (1994) Phylogenies without fossils. fragmented habitat of the Southwest USA. Journal of Biogeog- Evolution, 48, 523–529. raphy, 31, 1759–1768. Harvey MS, Rix MG, Framenau VW et al. (2011) Protecting the Dray S, Dufour AB (2007) The ade4 package: implementing the innocent: studying short-range endemic taxa enhances con- duality diagram for ecologists. Journal of Statistical Software, servation outcomes. Invertebrate Systematics, 25,1–10. 22,1–20. Heads M (2010) Old taxa on young islands: a critique of the Drummond AJ, Rambaut A, Shapiro B, Pybus OG (2005) Baye- use of island age to date island-endemic clades and calibrate sian coalescent inference of past population dynamics from phylogenies. Systematic Biology, 60,24–218. molecular sequences. Molecular Biology and Evolution, 22, Hedin M, Tsurusaki N, Macıas-Ordonez~ R, Shultz JW (2012) 1185–1192. Molecular systematics of sclerosomatid harvestmen (Opil- Eaton DA (2014) PyRAD: assembly of de novo RADseq loci for iones, Phalangioidea, Sclerosomatidae): geography is better phylogenetic analyses. Bioinformatics, 30, 1844–1849. than taxonomy in predicting phylogeny. Molecular Phyloge- Edwards SV, Shultz AJ, Campbell-Staton SC (2015) Next-gen- netics and Evolution, 62, 224–236. eration sequencing and the expanding domain of phylogeog- Hedin M, Carlson D, Coyle F (2015) Sky island diversification raphy. Folia Zoologica, 64, 187–206. meets the multispecies coalescent-divergence in the spruce- Emata KN, Hedin M (2016) From the mountains to the coast fir moss spider (Microhexura montivaga, Araneae, Mygalomor- and back again: ancient biogeography in a radiation of short- phae) on the highest peaks of southern Appalachia. Molecular range endemic harvestmen from California. Molecular Phylo- Ecology, 24, 3467–3484. genetics and Evolution, 98, 233–243. Heller R, Chikhi L, Siegismund HR (2013) The confounding Excoffier L, Lischer HE (2010) Arlequin suite ver 3.5: a new effect of population structure on Bayesian skyline plot infer- series of programs to perform population genetics analyses ences of demographic history. PLoS ONE, 8, e62992. under Linux and Windows. Molecular Ecology Resources, 10, Hendrixson BE, Guice AV, Bond JE (2015) Integrative species 564–567. delimitation and conservation of tarantulas (Araneae, Myga- Excoffier L, Dupanloup I, Huerta-Sanchez E, Sousa VC, Foll M lomorphae, Theraphosidae) from a North American biodi- (2013) Robust demographic inference from genomic and SNP versity hotspot. Insect Conservation and Diversity, 8, 120–131. data. PLoS Genetics, 9, e1003905. Hewitt G (2000) The genetic legacy of the quaternary ice ages. Frey JK, Bogan MA, Yates TL (2007) Mountaintop island age Nature, 405, 907–913. determines species richness of boreal mammals in the Amer- Hewitt GM (2004) Genetic consequences of climatic oscillations ican Southwest. Ecography, 30, 231–240. in the Quaternary. Philosophical Transactions of the Royal Soci- Garrick RC, Bonatelli IA, Hyseni C et al. (2015) The evolution ety of London B: Biological Sciences, 359, 183–195. of phylogeographic data sets. Molecular Ecology, 24, 1164– Hickerson MJ, Carstens BC, Cavender-Bares J et al. (2010) Phy- 1171. logeography’s past, present, and future: 10 years after. Giribet G, Vogt L, Gonzalez AP, Sharma P, Kury AB (2010) A Molecular Phylogenetics and Evolution, 54, 291–301. multilocus approach to harvestman (Arachnida: Opiliones) Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A (2005) phylogeny with emphasis on biogeography and the system- Very high resolution interpolated climate surfaces for global atics of Laniatores. Cladistics, 26, 408–437. land areas. International Journal of Climatology, 25, 1965–1978. Gnaspini P (2007) Development. In: Harvestmen: The Biology of Hoskin CJ, Tonione M, Higgie M et al. (2011) Persistence in Opiliones (eds Pinto-da-Rocha R, Machado G, Giribet G), pp. peripheral refugia promotes phenotypic divergence and spe- 455–472. Harvard University Press, Cambridge, Mas- ciation in a rainforest frog. The American Naturalist, 178, 561– sachusetts and London. 578. Gomez A, Vendramin GG, Gonzalez-Mart ınez SC, Alıa R Hugall A, Moritz C, Moussalli A, Stanisic J (2002) Reconciling (2005) Genetic diversity and differentiation of two Mediter- paleodistribution models and comparative phylogeography ranean pines (Pinus halepensis Mill. and Pinus pinaster Ait.) in the Wet Tropics rainforest land snail Gnarosophia bellen- along a latitudinal cline using chloroplast microsatellite denkerensis (Brazier 1875). Proceedings of the National Academy markers. Diversity and Distributions, 11, 257–263. of Sciences, 99, 6112–6117. Gottscho AD (2015) Lineage diversification of lizards (Phryno- Jensen JL, Bohonak AJ, Kelley ST (2005) Isolation by distance, somatidae) in southwestern North America: integrating web service. BMC Genetics, 6, 13. genomics and geology. Doctoral dissertation, University of Jezkova T, Olah-Hemmings V, Riddle BR (2011) Niche shifting California, Riverside. in response to warming climate after the last glacial

© 2016 John Wiley & Sons Ltd SOUTHWESTERN MONTANE HARVESTMEN PHYLOGEOGRAPHY 4629

maximum: inference from genetic data and niche assess- genomic evidence in montane sedges (Carex L.) from the ments in the chisel-toothed kangaroo rat (Dipodomys microps). Rocky Mountains. Evolution, 68, 2833–2846. Global Change Biology, 17, 3486–3502. Masta SE (2000) Phylogeography of the jumping spider Jezkova T, Riddle BR, Card DC, Schield DR, Eckstut ME, Cas- Habronattus pugillis (Araneae: Salticidae): recent vicariance of toe TA (2015) Genetic consequences of postglacial range sky island populations? Evolution, 54, 1699–1711. expansion in two codistributed rodents (genus Dipodomys) McCormack JE, Bowen BS, Smith TB (2008) Integrating paleoe- depend on ecology and genetic locus. Molecular Ecology, 24, cology and genetics of bird populations in two sky island 83–97. archipelagos. BMC Biology, 6, 28. Jombart T (2008) adegenet: a R package for the multivariate McCormack JE, Hird SM, Zellmer AJ, Carstens BC, Brumfield analysis of genetic markers. Bioinformatics, 24, 1403–1405. RT (2013) Applications of next-generation sequencing to phy- Jombart T, Ahmed I (2011) adegenet 1.3-1: new tools for the logeography and phylogenetics. Molecular Phylogenetics and analysis of genome-wide SNP data. Bioinformatics, 27, 3070– Evolution, 66, 526–538. 3071. McMillan ME, Heller PL, Wing SL (2006) History and Keith R, Hedin M (2012) Extreme mitochondrial population causes of post-Laramide relief in the Rocky Mountain oro- subdivision in southern Appalachian paleoendemic spiders genic plateau. Geological Society of America Bulletin, 118, (Araneae: Hypochilidae: Hypochilus), with implications for 393–405. species delimitation. Journal of Arachnology, 40, 167–181. Meyer WM III, Eble JA, Franklin K et al. (2015) Ground-dwell- Keyghobadi N, Roland J, Strobeck C (2005) Genetic differentia- ing arthropod communities of a Sky Island Mountain Range tion and gene flow among populations of the alpine butter- in Southeastern Arizona, USA: obtaining a baseline for fly, Parnassius smintheus, vary with landscape connectivity. assessing the effects of climate change. PLoS ONE, 10, Molecular Ecology, 14, 1897–1909. e0135210. Klicka J, Zink RM (1997) The importance of recent ice ages in Mitchell SG, Ober KA (2013) Evolution of Scaphinotus petersi speciation: a failed paradigm. Science, 277, 1666–1669. (Coleoptera: Carabidae) and the role of climate and geogra- Knowles LL (2001) Did the Pleistocene glaciations promote phy in the Madrean sky islands of southeastern Arizona, divergence? Tests of explicit refugial models in montane USA. Quaternary Research, 79, 274–283. grasshopprers. Molecular Ecology, 10, 691–701. Mitton JB, Kreiser BR, Latta RG (2000) Glacial refugia of lim- Knowles LL, Carstens BC (2007) Delimiting species without ber pine (Pinus flexilis James) inferred from the population monophyletic gene trees. Systematic Biology, 56, 887–895. structure of mitochondrial DNA. Molecular Ecology, 9,91– Knowles LL, Carstens BC, Keat ML (2007) Coupling genetic 97. and ecological-niche models to examine how past population Moore W, Meyer WM III, Eble JA (2013) Introduction to the distributions contribute to divergence. Current Biology, 17, Arizona Sky Island Arthropod Project (ASAP): systemat- 940–946. ics, biogeography, ecology, and population genetics of Kozak KH, Graham CH, Wiens JJ (2008) Integrating GIS-based arthropods of the Madrean Sky Islands. In: Merging Science environmental data into evolutionary biology. Trends in Ecol- and Management in a Rapidly Changing World: Biodiversity ogy and Evolution, 23, 141–148. and Management of the Madrean Archipelago III (eds Got- Lamb T, Jones TR, Wettstein PJ (1997) Evolutionary genetics tfried GJ et al. ), pp. 140–164. USDA (RMRS-P-67), Washing- and phylogeography of tassel-eared squirrels (Sciurus aberti). ton DC. Journal of Mammalogy, 78, 117–133. Nee S, Mooers AO, Harvey PH (1992) Tempo and mode of Lanfear R, Calcott B, Ho SY, Guindon S (2012) PartitionFinder: evolution revealed from molecular phylogenies. Proceedings combined selection of partitioning schemes and substitution of the National Academy of Sciences, 89, 8322–8326. models for phylogenetic analyses. Molecular Biology and Evo- Ober KA, Connolly CT (2015) Geometric morphometric and lution, 29, 1695–1701. phylogenetic analyses of Arizona Sky Island populations of Lee CR, Mitchell-Olds T (2011) Quantifying effects of environ- Scaphinotus petersi Roeschke (Coleoptera: Carabidae). Zoologi- mental and geographical factors on patterns of genetic differ- cal Journal of the Linnean Society, 175, 107–118. entiation. Molecular Ecology, 20, 4631–4642. Ober K, Matthews B, Ferrieri A, Kuhn S (2011) The evolution Lomolino MV, Brown JH, Davis R (1989) Island biogeography and age of populations of Scaphinotus petersi Roeschke on of montane forest mammals in the American Southwest. Arizona Sky Islands (Coleoptera, Carabidae, Cychrini). Zoo- Ecology, 70, 180–194. Keys, 147, 183–197. Lucchitta I (1989) History of the Grand Canyon and of the Col- Oksanen J, Blanchet FG, Kindt R et al. (2015) Package ‘vegan’, orado River in Arizona. Geologic evolution of Arizona: Arizona Community ecology package. Version 2-2. Available at: Geological Society Digest, 17, 701–715. https://cran.r-project.org/package=vegan. Manthey JD, Moyle RG (2015) Isolation by environment in Packard AS (1877) On a new cave fauna in Utah. Bulletin of white-breasted Nuthatches (Sitta carolinensis) of the Madrean the United States Geological Survey of the Territories, 3, Archipelago sky islands: a landscape genomics approach. 157–169. Molecular Ecology, 24, 3628–3638. Papadopoulou A, Anastasiou I, Vogler AP (2010) Revisiting Martınez-Solano I, Teixeira J, Buckley D, Garcıa-Parıs M (2006) the insect mitochondrial molecular clock: the mid-Aegean Mitochondrial DNA phylogeography of Lissotriton boscai trench calibration. Molecular Biology and Evolution, 27, (Caudata, Salamandridae): evidence for old, multiple refugia 1659–1672. in an Iberian endemic. Molecular Ecology, 15, 3375–3388. Paradis E, Claude J, Strimmer K (2004) APE: analyses of Massatti R, Knowles LL (2014) Microhabitat differences impact phylogenetics and evolution in R language. Bioinformatics, 20, phylogeographic concordance of codistributed species: 289–290.

© 2016 John Wiley & Sons Ltd 4630 S. DERKARABETIAN ET AL.

Peterson BJ, Graves WR (2015) Chloroplast phylogeography of Starrett J, Derkarabetian S, Richart CH, Cabrero A, Hedin M Dirca palustris L. indicates populations near the glacial (2016) A new monster from southwest Oregon forests: Cryp- boundary at the Last Glacial Maximum in eastern North tomaster behemoth sp. nov. (Opiliones, Laniatores, Travu- America. Journal of Biogeography, 43, 314–327. nioidea). ZooKeys, 555,11–35. Peterson BK, Weber JN, Kay EH, Fisher HS, Hoekstra HE Stervander M, Illera JC, Kvist L et al. (2015) Disentangling the (2012) Double digest RADseq: an inexpensive method for de complex evolutionary history of the Western Palearctic blue novo SNP discovery and genotyping in model and non- tits (Cyanistes spp.)–phylogenomic analyses suggest radiation model species. PLoS ONE, 7, e37135. by multiple colonization events and subsequent isolation. Phillips SJ, Anderson RP, Schapire RE (2006) Maximum Molecular Ecology, 24, 2477–2494. entropy modeling of species geographic distributions. Ecolog- Tennessen JA, Zamudio KR (2008) Genetic differentiation ical Modelling, 190, 231–259. among mountain island populations of the striped plateau R Core Team (2013) R: A Language and Environment for Statisti- lizard, Sceloporus virgatus (Squamata: Phrynosomatidae). cal Computing. R Foundation for Statistical Computing, Copeia, 3, 558–564. Vienna, Austria. URL http://www.R-project.org/. Thomas SM, Hedin M (2008) Multigenic phylogeographic Rambaut A, Drummond AJ (2014) TreeAnnotator v2. 1.2. Edin- divergence in the paleoendemic southern Appalachian opil- burgh: University of Edinburgh, Institute of Evolutionary ionid Fumontana deprehendor Shear (Opiliones, Laniatores, Biology. Triaenonychidae). Molecular Phylogenetics and Evolution, 46, Rambaut A, Suchard MA, Xie D, Drummond AJ (2014) Tracer 645–658. v1. 6. Computer program and documentation distributed by Ubick D, Briggs TS (1992) The harvestman family Phalangodi- the author. Available at: http://beast.bio.ed.ac.uk/Tracer dae. 3. Revision of Texella Goodnight and Goodnight (Opil- Richards CL, Carstens BC, Knowles LL (2007) Distribution iones: Laniatores). Texas Memorial Museum, Speleological modelling and statistical phylogeography: an integrative Monographs, 3, 155–240. framework for generating and testing alternative biogeo- Ubick D, Briggs TS (2008) The harvestman family Phalangodi- graphical hypotheses. Journal of Biogeography, 34, 1833–1845. dae. 6: revision of the Sitalcina Complex (Opiliiones: Lania- Richart C, Hedin M (2013) Three new species in the harvest- tores). Proceedings of the California Academy of Sciences, 59,1– men genus Acuclavella (Opiliones, Dyspnoi, Ischyropsali- 108. doidea), including description of male Acuclavella quattuor Vandewoestijne S, Van Dyck H (2010) Population genetic dif- Shear, 1986. ZooKeys, 311,19–68. ferences along a latitudinal cline between original and Rissler LJ, Apodaca JJ (2007) Adding more ecology into species recently colonized habitat in a butterfly. PLoS ONE, 5, delimitation: ecological niche models and phylogeography e13810–e13810. help define cryptic species in the black salamander (Aneides Wachter GA, Papadopoulou A, Muster C et al. (2016) Glacial flavipunctatus). Systematic Biology, 56, 924–942. refugia, recolonisation patterns, and diversification forces in Rousset F (1997) Genetic differentiation and estimation of gene Alpine-endemic Megabunus harvestmen. Molecular Ecology, flow from F-statistics under isolation by distance. Genetics, 25, 2904–2919. 145, 1219–1228. Wang IJ (2013) Examining the full effects of landscape hetero- Satler JD, Zellmer AJ, Carstens BC (2016) Biogeographic barri- geneity on spatial genetic variation: a multiple matrix regres- ers drive co-diversification within associated eukaryotes of sion approach for quantifying geographic and ecological the Sarracenia alata pitcher plant system. PeerJ, 4, e1576. isolation. Evolution, 67, 3403–3411. Schonhofer€ AL, McCormack M, Tsurusaki N, Martens J, Wang IJ, Bradburd GS (2014) Isolation by environment. Molecu- Hedin M (2013) Molecular phylogeny of the harvestmen lar Ecology, 23, 5649–5662. genus Sabacon (Arachnida: Opiliones: Dyspnoi) reveals Wang IJ, Summers K (2010) Genetic structure is correlated with multiple Eocene-Oligocene intercontinental dispersal events phenotypic divergence rather than geographic isolation in in the Holarctic. Molecular Phylogenetics and Evolution, 66, the highly polymorphic strawberry poison-dart frog. Molecu- 303–315. lar Ecology, 19, 447–458. Schonhofer€ AL, Vernesi C, Martens J, Hedin M (2015) Wang IJ, Glor RE, Losos JB (2013) Quantifying the roles of ecol- Molecular phylogeny, biogeographic history, and evolution ogy and geography in spatial genetic divergence. Ecology Let- of cave-dwelling taxa in the European harvestman genus ters, 16, 175–182. Ischyropsalis (Opiliones: Dyspnoi). The Journal of Arachnology, Welbourn WC (1999) Invertebrate cave fauna of Kartchner cav- 43,40–53. erns, Kartchner caverns, Arizona. Journal of Cave and Karst Sexton JP, Hangartner SB, Hoffmann AA (2014) Genetic isola- Studies, 61,93–101. tion by environment or distance: which pattern of gene flow Wiens JJ (2004) Speciation and ecology revisited: phylogenetic is most common? Evolution, 68,1–15. niche conservatism and the origin of species. Evolution, 58, Shafer A, Cullingham CI, Cot^ e SD, Coltman DW (2010) Of gla- 193–197. ciers and refugia: a decade of study sheds new light on the Wiens JJ, Graham CH (2005) Niche conservatism: integrating evo- phylogeography of northwestern North America. Molecular lution, ecology, and conservation biology. Annual Review of Ecology, 19, 4589–4621. Ecology, Evolution, and Systematics, 36, 519–539. Smith CI, Farrell BD (2005) Phylogeography of the longhorn Willing EM, Dreyer C, Van Oosterhout C (2012) Estimates of cactus beetle Moneilema appressum LeConte (Coleoptera: Cer- genetic differentiation measured by F ST do not necessarily ambycidae): was the differentiation of the Madrean sky require large sample sizes when using many SNP markers. islands driven by Pleistocene climate changes? Molecular PLoS ONE, 7, e42649. Ecology, 14, 3049–3065. Wright S (1943) Isolation by distance. Genetics, 28, 114.

© 2016 John Wiley & Sons Ltd SOUTHWESTERN MONTANE HARVESTMEN PHYLOGEOGRAPHY 4631

Fig. S1 ENMs for all S. robustus, and the western and eastern clades for current, mid-Holocene, and LGM bioclimatic conditions. S.D. and M.H. conceived and designed the experiments; S.D., M.B. and J.S. performed the experiments; S.D. Fig. S2 BEAST phylogeny and divergence date estimates for analysed the data; S.D. and M.H. contributed reagents/ the m16 dataset. Note all nodes are supported with posterior > materials/analysis tools; S.D., M.B., J.S. and M.H. probabilities 99. contributed to the writing of the manuscript. Fig. S3 Bayesian skyline plots for major clade groupings.

Fig. S4 SNP plots for the m16 dataset.

Fig. S5 PCA plot of environmental variable extracted from Data accessibility occurrence records.

Raw ddRAD-seq sequence read files have been submit- Table S1 Specimen locality information and pyRAD analysis ted to the NCBI Sequence Read Archive, Accession no. statistics. SRP077679 (BioProject PRJNA327427). Final concate- Table S2 Primers and barcodes used in ddRAD-seq library nated and SNP matrices, structures files used in PCA, preparation. and the environmental distance matrix have been deposited to the Dryad Digital Repository, http://dx. Table S3 Results of IBD, IBE, and MMRR analyses. Significant doi.org/10.5061/dryad.bq68f. values are indicated in bold. Table S4 Results of migrate-n analyses.

Material S1 ddRAD-seq library preparation and sequencing Supporting information protocol. Additional supporting information may be found in the online version of this article.

© 2016 John Wiley & Sons Ltd