Molecular Phylogenetics and Evolution 93 (2015) 234–248

Contents lists available at ScienceDirect

Molecular Phylogenetics and Evolution

journal homepage: www.elsevier.com/locate/ympev

Comparison of five methods for delimitating species in Ophion Fabricius, a diverse genus of parasitoid wasps (Hymenoptera, Ichneumonidae) q ⇑ Marla D. Schwarzfeld , Felix A.H. Sperling

Department of Biological Sciences, CW 405 Biological Sciences Building, University of Alberta, Edmonton, Alberta T6G 2E9, Canada article info abstract

Article history: DNA taxonomy has been proposed as a method to quickly assess diversity and species limits in highly Received 6 January 2015 diverse, understudied taxa. Here we use five methods for species delimitation and two genetic markers Revised 31 July 2015 (COI and ITS2) to assess species diversity within the parasitoid genus, Ophion. We searched for compen- Accepted 4 August 2015 satory base changes (CBC’s) in ITS2, and determined that they are too rare to be of practical use in delim- Available online 8 August 2015 iting species in this genus. The other four methods used both COI and ITS2, and included distance-based (threshold analysis and ABGD) and tree-based (GMYC and PTP) models. We compared the results of these Keywords: analyses to each other under various parameters and tested their performance with respect to 11 ABGD Nearctic species/morphospecies and 15 described Palearctic species. We also computed barcode accumu- Compensatory base change DNA barcoding lation curves of COI sequences to assess the completeness of sampling. The species count was highly vari- GMYC able depending on the method and parameters used, ranging from 47 to 168 species, with more ITS2 conservative estimates of 89–121 species. Despite this range, many of the Nearctic test species were fairly Poisson tree processes robust with respect to method. We concluded that while there was often good congruence between Species delimitation methods, GMYC and PTP were less reliant on arbitrary parameters than the other two methods and more easily applied to genetic markers other than COI. However, PTP was less successful at delimiting test spe- cies than was GMYC. All methods, as well as the barcode accumulation curves, indicate that several Palearctic species remain undescribed and that we have scarcely begun to appreciate the Nearctic diver- sity within this genus. Ó 2015 Elsevier Inc. All rights reserved.

1. Introduction Many studies using DNA taxonomy, particularly of the COI gene, have delimited species with distance-based clustering methods. Species are a basic unit in studies of ecology, biogeography, and For example, species have been defined as terminal monophyletic evolution, and are essential to applications in conservation or bio- clusters on neighbor-joining trees (Hajibabaei et al., 2006), by a logical control (Claridge et al., 1997; Sites and Marshall, 2004). But standard interspecific threshold of 1–3% (Hebert et al., 2003a,b, most of Earth’s species remain undescribed, while taxonomists are 2004; Smith et al., 2009, 2013; Stalhut et al., 2013; too few and traditional taxonomy is too slow to deal fully with this Strutzenberger et al., 2011; Tang et al., 2012), or by an interspecific unknown diversity (Brooks and Hoberg, 2000; Godfray, 2002; distance of 10X the intraspecific distance (Hebert et al., 2004). Wheeler, 2004). With the advent of rapid DNA sequencing, molec- While it is widely acknowledged that a single threshold divergence ular taxonomy has been proposed for quickly assessing species will not apply to all taxa (Cognato, 2006; Hendrich et al., 2010; diversity in diverse, little-known taxa (Blaxter, 2004; Pons et al., Monaghan et al., 2009), pre-defined thresholds have nonetheless 2006; Tautz et al., 2003). In particular, the Barcode of Life project been implicitly or explicitly used as criteria for assessing species has popularized use of half of the cytochrome oxidase I (COI) gene diversity in a number of studies (Barrett and Hebert, 2005; of mitochondrial DNA as a standardized ‘‘barcode” for species Hebert et al., 2003a; Smith et al., 2009, 2013; Strutzenberger identification and discovery (Hebert et al., 2003a,b; Miller, 2007). et al., 2011). As well, the universality of barcoding gaps is contro- versial, with several studies finding significant overlap between intra- and interspecific divergences (Meyer and Paulay, 2005;

q Wiemers and Fiedler, 2007). The challenge of objectively determin- This paper was edited by the Associate Editor S.L. Cameron. ing this gap led to the development of the automated barcode gap ⇑ Corresponding author at: Canadian National Collection of Insects, Arachnids and Nematodes, 960 Carling Ave., Ottawa, Ontario K1A 0C6, Canada. discovery (ABGD) algorithm, which infers confidence intervals for E-mail address: [email protected] (M.D. Schwarzfeld). intraspecific divergences, and recursively partitions the sequences http://dx.doi.org/10.1016/j.ympev.2015.08.003 1055-7903/Ó 2015 Elsevier Inc. All rights reserved. M.D. Schwarzfeld, F.A.H. Sperling / Molecular Phylogenetics and Evolution 93 (2015) 234–248 235 according to the empirically observed divergences (Puillandre of conspecificity, as closely related species may not have any CBC’s et al., 2012a). (Müller et al., 2007; Wiemers et al., 2009). CBC’s therefore only Distance-based methods have been criticized for being purely indicate a minimum number of species among the specimens phenetic, rather than incorporating evolutionary theory (DeSalle, examined. 2007; Rubinoff et al., 2006; Will and Rubinoff, 2004; Will et al., Ophion is a genus of large-sized (10–20 mm) nocturnal para- 2005). Some explicitly evolutionary approaches have been devel- sitoid wasps in the subfamily Ophioninae of the family oped to address this concern, such as coalescence or tree-based Ichneumonidae. Unlike the majority of ophionine genera, Ophion methods, and these may be used to delimit large numbers of spe- are most diverse in temperate regions, with few species in tropical cies (Carstens and Dewey, 2010; Fujita et al., 2012; Leaché and habitats (Gauld, 1980, 1988). Ophion are large, abundant, and come Fujita, 2010; O’Meara, 2010; Yang and Rannala, 2010). One of these readily to lights; however, Nearctic species have received almost methods, the generalized mixed Yule coalescent (GMYC) model, no taxonomic attention despite being highly apparent and easily determines the point at which coalescent branching patterns collected. Seventeen Nearctic species are currently described (within species) transition to Yule patterns (between species) (Schwarzfeld and Sperling, 2014; Yu et al., 2012), yet it has been (Fontaneto et al., 2007; Pons et al., 2006). This method has shown estimated that the fauna consists of approximately 50 species promise for identifying species boundaries in diverse, little known (Gauld, 1985). In the Palearctic region, Ophion are better known, taxa (Astrin et al., 2012; Ceccarelli et al., 2012; Monaghan et al., with 79 described species (Yu et al., 2012), but there have been 2009); however, the results depend on the choice of parameters no species revisions that span the Palearctic. The fauna has been used to construct the ultrametric tree (Ceccarelli et al., 2012). most comprehensively examined in the United Kingdom, with a The Poisson Tree Processes model (PTP) is another tree-based number of revisions (Gauld, 1973, 1976, 1978) culminating in a method that uses a non-ultrametric phylogenetic tree as input, comprehensive revision of 16 known species (Brock, 1982). thus avoiding the challenges of constructing an accurate ultramet- Ophion are notoriously difficult to identify as they are morpho- ric tree (Zhang et al., 2013). The method delimits species under the logically homogenous among species and yet variable within spe- assumption that the number of substitutions within a species is cies (Brock, 1982; Gauld, 1980). They are thus well-suited to the significantly lower than the number of substitutions between application of molecular taxonomic methods as a first step toward species. species delimitation. Here we estimate Ophion species diversity Regardless of the delimitation method used, reliance on a single using tree-based methods (GMYC, PTP), distance-based methods genetic marker may give misleading results (Dupuis et al., 2012). (ABGD, threshold analysis), and by assessing compensatory base While COI has many advantages for studies of species delimitation, changes of ITS2. The study is mainly based on Canadian specimens, such as its ease of amplification and relatively rapid rate of muta- with the addition of specimens from the United States and repre- tion (Li et al., 2010; Monti et al., 2005; Williams et al., 2012), it may sentatives of all species, except one, that have been described from not accurately delimit species due to factors such as introgression, the United Kingdom. retained ancestral polymorphisms, or coamplification of nuclear pseudogenes (numts) (Cognato, 2006; Dupuis et al., 2012; Funk 2. Methods and Omland, 2003; Meier et al., 2006; Schmidt and Sperling, 2008). It is therefore important to use additional data, such as from 2.1. Specimen collection other genes or morphology, in an integrative taxonomic analysis to more accurately delimit species (Dayrat, 2005; Roe and Sperling, A total of 674 specimens of Ophion were included in this study, 2007; Schlick-Steiner et al., 2010; Schwarzfeld and Sperling, 2014). including 572 from Canada, 81 from Europe, 19 from the United The internal transcribed spacer 2 (ITS2) of ribosomal RNA is States, and 2 from Costa Rica (Supplementary Table 1). Most another rapidly evolving genetic marker that has proved useful Nearctic specimens were newly collected and sequenced for this for phylogenetic analyses at the species level (Alvarez and Hoy, study, while 25 were sequenced from alcohol-preserved material 2002; Campbell et al., 1993; Gomez-Zurita et al., 2000; Hung on loan from the Canadian National Collection of Insects, et al., 2004; Li et al., 2010; Wagener et al., 2006). Ribosomal RNA Arachnids and Nematodes, and 92 COI sequences from British is found in multiple tandem repeats within the genome, which Columbia were included from the Barcode of Life database undergo concerted evolution so that (in most taxa) the copies are (BOLD), courtesy of J. DeWaard (project code: ICHBC). Specimens identical (Hillis and Davis, 1988; Ji et al., 2003; but see Keller selected for sequencing represent the range of morphological vari- et al., 2006). Because of these multiple copies, ITS2 is relatively ation observed in over 4000 specimens from a variety of habitats. easy to amplify. However, the high rate of mutation and the pres- Eighty specimens from 15 species of Palearctic Ophion were also ence of numerous insertion-deletion events (indels) can make newly sequenced, all of which are from the United Kingdom except accurate alignment of this gene challenging, particularly among for one specimen from Spain and one from France. Finally, one more distantly related species (Coleman and Vacquier, 2002). additional sequence of O. obscuratus Fabricius (also from the UK) ITS2 has a highly conserved secondary structure, with characteris- and two sequences of O. flavidus Brullé from Costa Rica were tic paired domains and unpaired regions (Coleman, 2007). included from Genbank and BOLD. With the exception of Incorporating this secondary structure can improve alignment of sequences obtained from GenBank or BOLD, all Nearctic taxa used ITS2, thus facilitating phylogenetic reconstruction, and also per- in the current study were identified by M. D. Schwarzfeld. mits the discovery of compensatory base changes (CBC’s) Palearctic taxa were identified by G. R. Broad (Natural History (Coleman and Vacquier, 2002; Wiemers et al., 2009; Wolf et al., Museum, London), with the exception of O. forticornis, which was 2013). These are substitutions where two paired nucleotides (often identified by M. R. Shaw (National Museums Scotland, Edinburgh). quite far apart in the linear sequence) both mutate such that the original pairing is maintained (Coleman, 2009; Coleman and Vacquier, 2002; Gutell et al., 1994). Müller et al. (2007) determined 2.2. Sequencing that even a single CBC between two individuals indicates that enough evolutionary time has passed for reproductive isolation DNA was extracted from a single hind leg and amplified follow- to arise, resulting in separate species status. Such CBC’s do not ing methods described in Schwarzfeld and Sperling (2014).We cause reproductive isolation directly, instead simply indicating a sequenced COI using the primers lco hym (5’–CAA ATC ATA AAG correlation (Wolf et al., 2013). As well, a lack of CBC’s is not proof ATA TTG G–3’) and hco out (5’–CCA GGT AAA ATT AAA ATA TAA 236 M.D. Schwarzfeld, F.A.H. Sperling / Molecular Phylogenetics and Evolution 93 (2015) 234–248

ACT TC–3’) (Schulmeister, 2003), which produce a 676 base pair the tree into the program Conserve (Agapow and Crozier, 2008). fragment equivalent to the ‘‘barcode” region (Hebert et al., We excluded all sequences, and then randomly added sequences 2003a,b). We analyzed ITS2 using the primers ITS2-F (5’-GGG in sets of 10 (Nearctic, Canada and Alberta datasets) or sets of 5 TCG ATG AAG AAC GCA GC-3’) and ITS2-R (5’-ATA TGC TTA AAT (UK dataset), and calculated the phylogenetic diversity (PD) of TCA GCG GG-3’) which anneal to the flanking 28S and 5.8S genes the tree at each successive addition. These results were plotted (Navajas et al., 1994). Sequences are deposited in NCBI’s to assess whether the phylogenetic diversity was approaching an GenBank, and accession numbers are listed in Supplementary asymptote. Table 1. We sequenced COI for 565 Ophion specimens. A few were not suc- 2.4. ITS2 secondary structure and CBC’s cessfully sequenced in both directions, or were otherwise incom- plete; these sequences were included with gaps coded as missing The complete protocol used to determine the secondary struc- data. The 95 Ophion sequences obtained from BOLD and Genbank ture of ITS2 is described in Supplementary Data 2. Briefly, we esti- were sequenced using different primers (see BOLD website for mated the structure for an arbitrarily chosen sequence using the sequencing protocols) and were slightly shorter in length. RNAfold webserver, using the minimum free energy method with Alignment was unambiguous, and was confirmed by translating default settings (Hofacker, 2003). We imported this sequence and nucleotides to amino acids in Mesquite (Maddison and Maddison, structure into the ITS2 database (Koetschan et al., 2010; Schultz 2011). The data matrix can be found deposited on the Dryad data et al., 2006; Selig et al., 2008) and used it as a template for folding repository (available at datadryad.org, doi: 10.5061/dryad.hf155). the remaining sequences. If any sequences had less than 90% sim- Since analysis of ITS2 secondary structure (see Section 2.4 and ilarity to the existing template, one of the anomalous sequences Supplementary Data 2) was sensitive to missing data, we only was directly folded in RNAfold and added as an additional template included sequences that were complete and lacked ambiguous to the database. We repeated this iterative process until all regions; the final dataset included 394 ITS2 sequences. Excluding sequences were folded; the final dataset consisted of eight the flanking 28S and 5.8S regions, the unaligned sequences ranged sequences that were folded directly and 386 sequences that were from 717 to 1053 base pairs in length. A subset of these (380 spec- folded using homology. Sequences and structures were aligned imens) were successfully sequenced for both ITS2 and COI using 4Sale, a program designed to perform structural alignments (Table 1). of RNA sequences (Seibel et al., 2006), and a matrix of compen- satory base changes (CBC’s) was exported. The final alignment (excluding the 28S and 5.8S flanking regions) was 1750 base pairs 2.3. Barcode accumulation curves in length. This alignment was used for all phylogenetic analyses described below, and can be found on the Dryad data repository All species-delimitation algorithms are dependent on sampling (datadryad.org, doi: 10.5061/dryad.hf155). We calculated a (Bergsten et al., 2012; Lohse, 2009; Meyer and Paulay, 2005; neighbor-joining tree based on the CBC distance matrix using the Papadopoulou et al., 2009). We assessed the completeness of sam- program CBCAnalyzer (Wolf et al., 2005). pling by calculating ‘‘barcode accumulation curves” of the COI sequences, following the procedure outlined by Smith et al. 2.5. Species delimitation (2009). Similar to species accumulation curves used in ecological studies, barcode accumulation curves should approach an asymp- We analyzed sequence data using four quantitative methods of tote as fewer unique haplotypes are added to the specimen pool, species delimitation: generalized mixed Yule coalescent analysis indicating that most of the diversity has been sampled. Since the (GMYC) (Fontaneto et al., 2007; Pons et al., 2006); Poisson tree pro- vast majority of specimens were from Canada, we calculated the cesses model (PTP) (Zhang et al., 2013); automatic barcode gap dis- curve for Canadian material alone (n = 561), as well as a separate covery (ABGD) (Puillandre et al., 2012a); and threshold analysis curve including all Nearctic specimens (n = 580). We also calcu- using the program jMOTU (Jones et al., 2011). Each analysis lated an accumulation curve for specimens from the province of uses different terminology to refer to the delimited taxa Alberta, since this is the most densely sampled region (n = 342). (GMYC = ‘‘entities”; PTP = ‘‘phylogenetic species”; ABGD = ‘‘groups” Finally, we calculated a barcode accumulation curve including only or ‘‘hypothetical species”; jMOTU = ‘‘molecular taxonomic units” Palearctic specimens (n = 78). These curves are not based on ran- or ‘‘MOTU”), thus acknowledging that they may not represent dom samples and do not represent comparable regions. We use biologically meaningful species. For clarity and consistency, we them to indicate the pattern of sequence accumulation within are using the term ‘‘putative species” in place of each of the above the current study, but do not expect them to represent the true terms, while the term ‘‘species” is applied only to taxa that are diversity of each region. delimited more robustly by multiple data sources. To calculate the curves, we first constructed a neighbor-joining tree for each dataset in MEGA version 5 (Tamura et al., 2011), using 2.5.1. GMYC the K2P model and pairwise deletion of missing data, and imported Models for the construction of the ultrametric trees required for the GMYC analysis were selected using the program

Table 1 PartitionFinder version 1.0.1 (Lanfear et al., 2012), with each gene Summary of specimens sequenced from each geographic region represented in this analyzed as a single partition. For the ITS2 dataset, the selected study. Palearctic specimens are all from the western Palearctic region, primarily the model was TVMef + G, and COI used HKY + I + G. We obtained the United Kingdom. ultrametric trees using BEAST version 1.7.5 (Drummond et al., COI only ITS2 COI & ITS2 COI ITS2 Total No. 2012) on the CIPRES Science Gateway computing cluster (Miller only total total specimens et al., 2010), with ITS2 and COI analyzed separately. Identical hap- Canada 198 11 363 561 374 572 lotypes were removed using the program Alter (Glez-Peña et al., USA 12 0 7 19 7 19 2010), and outgroups were excluded. The choice of tree prior can Costa 20 0 2 0 2 impact the results of the phylogenetic analysis (Ceccarelli et al., Rica 2012). For example, the coalescent tree prior is often more appro- Palearctic 68 3 10 78 13 81 priate for population-level data and recently diverged taxa, Total 280 14 380 660 394 674 whereas the Yule prior is recommended for species-level data M.D. Schwarzfeld, F.A.H. Sperling / Molecular Phylogenetics and Evolution 93 (2015) 234–248 237

(Drummond et al., 2006). We therefore calculated trees using a molecular and morphological both suggest that this species is basal random starting tree, the uncorrelated lognormal relaxed clock within Ophion (Schwarzfeld, unpublished data). model, and two different tree models (coalescent and Yule- PTP was conducted using the online web service available at process) for a total of four analyses for each marker (ITS2 and COI). http://species.h-its.org/ptp (Zhang et al., 2013). This server con- Each analysis was run with either four (COI coalescent, COI ducts both the original maximum likelihood PTP, as well as a newly Yule, ITS2 coalescent) or five (ITS2 Yule) independent threads, developed Bayesian version of PTP (bPTP) that adds Bayesian sup- and log files and tree files were combined using LogCombiner. port values to the delimited species. However, our analyses failed The two COI analyses were run for 40 million generations, with to show convergence even with the maximum number of genera- trees sampled every 4000 generations and a burnin of 25%. tions (500,000) available on the web server; we have therefore Combined logfiles were examined in Tracer version 1.5 (Rambaut only included the maximum likelihood results in this paper. We and Drummond, 2009) to assess convergence; all combined esti- conducted PTP on trees including all sequences (COIall, ITS2all), mated sample sizes (ESS) were >200 with the exception of a single and on trees including only those specimens with sequences avail- parameter (mean.Rate, ESS = 179) using the Yule prior. The ITS2 able for both genes (COIcommon, ITS2common). Non-overlapping analysis with the coalescent prior was run for 80 million genera- sequences were pruned from the ML trees in R, prior to running tions, sampling every 4000 generations, with a burnin of 25%; all PTP. ESS were >200 except for two parameters (mean.Rate, treemodel. Root) that were between 100 and 200. Obtaining convergence in 2.5.3. ABGD the ITS2 analysis with the Yule prior was more challenging. The Automatic barcode gap discovery (ABGD) infers clusters of final analysis included three threads of 160 million generations sequences by recursively partitioning the data using a range of and two threads of 80 million generations, sampled every 4000 prior intraspecific sequence divergences and by calculating a generations. The first 25% of the longer runs, and the first 50% of model-based confidence limit for intraspecific divergence at each the shorter runs were removed as burnin. Two parameters failed iteration (Puillandre et al., 2012a). We first calculated distance to show convergence (coeffofVariation and covariance, ESS <100); matrices for COI and ITS2 in MEGA 5.1 using both K2P corrected however, the remaining parameters were either P200 or very distances and uncorrected p-distances. Separate matrices were cal- nearly so. All analyses were resampled in LogCombiner (part of culated for all specimens (COIall and ITS2all), or only including spec- the BEAST package) to give a combined total of approximately imens that were successfully sequenced for both genes (COIcommon, 15,000 trees. Maximum clade credibility trees (the tree in each ITS2common). We calculated the number of clusters using the ABGD analysis with the highest product of posterior clade probabilities) online tool (Puillandre et al., 2012a; available at http://wwwabi. were produced from the combined tree files in TreeAnnotator (part snv.jussieu.fr/public/abgd/abgdweb.html). Analyses of COI used of the BEAST package), using the mean heights option, and used as the default priors (prior intraspecific divergence = 0.001–0.1, input for GMYC analyses. divided into 10 partitions), since the program is optimized for We conducted GMYC analyses with the SPLITS package (avail- use with the COI barcoding gene, based on numerous studies. For able from http://r-forge.r-project.org/projects/splits/) using the ITS2, intraspecific sequence divergence has been less commonly program R (RStudio, 2012; R Core Development Team, 2013). assessed and is highly taxon-dependent. Within Ichneumonidae, Each tree was analyzed using both single and multiple threshold Wagener et al. (2006) found the intraspecific sequence divergence settings, including all haplotypes (COIall and ITS2all). To facilitate of Diadegma (subfamily Campopleginae) to be 0.2–0.6%. Within the comparison between genetic markers, we also ran the GMYC anal- Ophion scutellaris species group (Schwarzfeld and Sperling, 2014), yses only including taxa that were represented in both datasets intraspecific variation was similarly very low, ranging from a single

(COIcommon and ITS2common). Since in several instances a single haplotype per species to a maximum of 0.1%. We therefore ran the COI haplotype was represented by multiple ITS2 haplotypes, or ABGD analysis using prior intraspecific divergences of 0.0001–0.1 vice versa, the two analyses did not necessarily have the same (divided into 20 partitions). For each analysis we used both the number of individuals. We obtained the reduced trees by pruning default barcode relative gap width of 1.5 and a gap width of half non-overlapping taxa from ultrametric trees in R, prior to running the default (0.75) to increase the sensitivity of the analysis GMYC analyses. (Puillandre et al., 2012a)

2.5.2. PTP 2.5.4. Threshold analysis Poisson tree processes analyses used maximum-likelihood trees COI and ITS2 sequences were clustered into molecular taxo- as input. Models for the trees were determined using nomic units (MOTU) using the program jMOTU (Jones et al., PartitionFinder version 1.0.1 (Lanfear et al., 2012). For COI, each 2011). Commonly used thresholds for species delimitation based codon position was included as a potential partition, while ITS2 on COI are 1–3% (Bribiesca-Contreras et al., 2013; Hebert et al., was not partitioned. Only models that could be implemented in 2003a, b, 2004; Smith et al., 2009, 2013; Stalhut et al., 2013; the GARLI web service were included. For COI, the first and third Strutzenberger et al., 2011; Tang et al., 2012). Since computational codon position used GTR + I + G, while the second position used expense increases very little with more thresholds in the jMOTU F81 + G. ITS2 was analyzed using SYM + G. analysis, we calculated the number of species using a wider range Maximum likelihood trees were calculated using the web ser- of thresholds, from a single base pair difference (approximately vice for GARLI 2.1 (Bazinet et al., 2014; available at www. molecu- 0.15%) to a 50 base pair threshold (7.5% of the average sequence larevolution.org), using a neighbor-joining starting tree (calculated length). For the analysis of ITS2, we used the same range of thresh- in Mega 5.0; Tamura et al., 2011). The GARLI web service uses an olds (1–50 base pairs) as in the COI dataset, which is equivalent to adaptive approach wherein additional search replicates are based 0.1–5.7% of the average sequence length. Since the ITS2 sequences on the statistical probability that the best tree has been obtained contain numerous large indels, we used a range of low identity with 95% confidence. For ITS2 a total of 352 replicates were con- BLAST filters, from 80 to 97, to attempt to obtain the best align- ducted, while the COI analysis reached the maximum number of ment (Jones et al., 2010); however, the results were identical in 1000 replicates. In both cases, there was essentially no difference all cases. We therefore ran analyses on both the complete datasets in topology between the many best-scoring trees. COI trees were (COIall and ITS2all) and on the common datasets (COIcommon and rooted with Enicospilus, another genus in the subfamily ITS2common), using the default settings of a minimum 60% overlap Ophioninae, whereas ITS2 trees were rooted with O. minutus, since and low identity BLAST filter set to 97. 238 M.D. Schwarzfeld, F.A.H. Sperling / Molecular Phylogenetics and Evolution 93 (2015) 234–248

2.5.5. Comparison with morphology Comparisons of species-delimitation methods provide useful information into the performance of algorithms. However, even if multiple methods converge on the same results, there is no guar- antee that the clusters represent species. Morphological analysis of Ophion is ongoing, with many species thus far lacking clear mor- phological characters. Nevertheless, some species are morphologi- cally distinguishable. We selected 11 morphospecies to use as test species to determine whether or at what threshold they were recovered by various analyses. Seven of these species are the Nearctic members of the O. scutellaris species group (O. idoneus, O. importunus, O. keala, O. clave, O. aureus, O. brevipunctatus, and O. dombroskii) that were resolved using a combination of morphol- ogy, wing morphometrics and classical morphometrics (Schwarzfeld and Sperling, 2014). The other species include O. nigrovarius Provancher, O. slossonae Davis, and two undescribed, morphologically distinct species, sp. A and sp. B. Descriptions and Fig. 1. Barcode accumulation curves of all COI sequences from this study. photographs of the diagnostic features of each species are provided in Supplementary Data 3,orinSchwarzfeld and Sperling (2014). scale of sampling (Fig. 1). No dataset reached an asymptote, We calculated the average and maximum sequence divergence although the Alberta sequences may be approaching one. within each species (excluding those represented by singletons). Sequences from the UK were still in the steep initial part of the Distances are uncorrected p-distances, calculated in Mega 5.1 curve, precluding comparison with the other three datasets. The (Tamura et al., 2011), using the pairwise deletion option for gaps Nearctic closely follows the Canadian curve, since most specimens and missing data. For the species delimitation analyses, we exam- were included in both datasets. However, each specimen ined results from the COI and ITS2 datasets. We considered an all all sequenced from the southern United States (a total of ten speci- analysis to have successfully recovered a species if a single cluster mens from Arizona, Florida, Georgia, Kentucky, and New Mexico) included all of the sequenced individuals of a morphologically almost certainly represents an additional species, based on both defined species and no additional sequences. For the GMYC and genetic divergence and morphology (Schwarzfeld, unpublished PTP analyses, we reported if a species was successfully recovered data). This curve therefore greatly underestimates the true diver- (+), split between more than one cluster (Sp), or lumped with other sity for the Nearctic region. non-conspecific sequences (L). For the ABGD and jMOTU analyses, we listed the range of prior intraspecific divergences or thresholds, respectively, that successfully recovered the species, or recorded if 3.2. CBC analysis the species was never recovered (À). There was a maximum of 5 compensatory base changes (CBC’s) between any two specimens. The NJ tree derived from the CBC 2.5.6. Species delimitation of Palearctic Ophion matrix supported the overall topology of species groups within Since the Nearctic fauna is almost entirely undescribed, few the genus, as determined through a molecular phylogenetic analy- well-characterized species are available for assessing the perfor- sis of Ophion (Schwarzfeld, unpublished data). However, there mance of species-delimitation methods. Ophion from the UK have were no CBC’s within any species group except for two specimens been much more thoroughly studied, with a comprehensive mor- that differed from the rest of their group by one (O. obscuratus from phological revision and workable key to species (Brock, 1982). the U.K., DNA6917) or two (undescribed species from Arizona, However, these species have not been assessed with molecular DNA5569) CBC’s. Within species groups, CBC’s were thus generally data, and several species present potential taxonomic difficulties uninformative for distinguishing species. (Brock, 1982, G. Broad, pers. comm.). We have therefore analyzed the species limits of 15 of the 16 described British species, based 3.3. GMYC on COI sequence data. We restricted the analysis to COI as we had limited success sequencing these specimens for ITS2. Based The results of the GMYC analyses are found in Table 2. Three of on the results of the morphological comparisons described above, the single threshold analyses of the Yule prior trees (ITS2 , we selected a subset of analyses that were most effective for distin- all ITS2 , and COI ) failed to significantly reject the null guishing Nearctic species. We assessed each Palearctic species common common hypothesis that the sequences cannot be differentiated from a sin- using GMYC (coalescent prior, single threshold method), PTP, gle cluster. Nonetheless, the estimated numbers of clusters ABGD (relative gap width: 0.75, prior intraspecific divergence (excluding singletons) and putative species (=clusters plus single- 0.0046), and jMOTU (threshold: 7, 14, and 20 base pairs, or approx- tons, or ‘‘entities”) from these analyses were within the range of imately 1%, 2%, and 3% sequence divergence). While the 3% thresh- all other analyses. Across all analyses, the proportion of singletons old had only limited success in delimiting the Nearctic test species, ranged from 29% to 57% of the total number of GMYC species, with we included it because this threshold is commonly used in the bar- an average of 40%. coding literature (e.g. Hebert et al., 2003a; Tang et al., 2012). In general, the number of clusters was fairly robust with respect to both tree prior (Yule (Y) or coalescent (C)) and GMYC method

3. Results (single (S) or multiple (M)). For the COIall data, the number of clus- ters varied from 72 to 81, while in the smaller ITS2all dataset three 3.1. Barcode accumulation curves of the four analyses resulted in 52 or 53 clusters. The YM analysis

of the ITS2all data, however, only recovered 35 clusters. When only The three Nearctic barcode accumulation curves for COI (all common taxa were included, the number of clusters was also quite Nearctic specimens, Canadian only, and Alberta only) showed a robust between the two genes, with all analyses recovering 45–50 similar shape and, as expected, diversity increased with geographic clusters. M.D. Schwarzfeld, F.A.H. Sperling / Molecular Phylogenetics and Evolution 93 (2015) 234–248 239

Table 2 Results of generalized mixed Yule coalescent (GMYC) analysis. Method = number of thresholds; Clu = number of clusters; Spp = number of species (clusters + singletons); CI = confidence interval for the number of species; L-Null = likelihood of the null model (=all sequences form a single cluster); L-GMYC = likelihood of GMYC model; LRT = p-value of likelihood ratio test; COIall and ITS2all = all haplotypes included in analysis; COIcommon and ITS2common = only haplotypes represented by both markers were included.

Dataset Tree prior Method Clu Spp CI L-Null L-GMYC LRT

COIall Coalescent Single 72 114 100–137 3119.398 3135.499 <0.000 Multiple 83 150 129–167 3119.398 3145.07 <0.000 Yule Single 81 121 117–134 3057.645 3064.79 0.003 Multiple 79 124 88–125 3057.645 3073.966 0.000

ITS2all Coalescent Single 53 85 85–104 2326.429 2334.213 0.001 Multiple 52 95 93–96 2326.429 2339.508 0.000 Yule Single 53 75 62–81 2162.679 2166.034 0.0818a Multiple 35 51 47–104 2162.679 2173.709 0.001

COIcommon Coalescent Single 45 80 67–93 1728.258 1739.615 <0.000 Multiple 49 107 103–107 1728.258 1744.155 <0.000 Yule Single 50 86 2–106 1695.617 1698.06 0.180a Multiple 44 103 72–116 1695.617 1704.424 0.024

ITS2common Coalescent Single 45 76 53–94 2135.929 2143.237 0.002 Multiple 49 78 78–92 2135.929 2147.648 0.001 Yule Single 48 69 59–77 1990.495 1994.305 0.0545a Multiple 48 73 39–99 1990.495 2001.224 0.0181

a Not significant.

The total number of putative species was more variable. For the priors = 0.0010 and 0.0017, respectively; Fig. 3a). Over the remain-

COIall dataset, the number was fairly similar between three of the ing partitions, the pattern was analysis- and dataset-dependent, analyses (114–124); however, the CM analysis recovered 150 puta- though by a prior of 0.036 all analyses were recovering a single tive species. For the ITS2all analyses, the number of putative species cluster. The COIall analysis using the default gap width had a clear varied from 51 (YM) to 95 (CM). For the ‘‘common” datasets, there plateau in the number of species recovered, ranging from 66 to 82. was a minimum of 69 putative species (ITS2 YS) and a maximum of However, the pattern was less clear in the other analyses. The more 107 (COI CM). All of the COI analyses recovered more putative spe- sensitive analyses (gap width = 0.75) had a range of 52–168 puta- cies than did the corresponding ITS2 analyses. Within COIcommon, tive species (COIall) or 23–120 species (COIcommon) over the inter- the YM and CM analyses recovered far more putative species mediate priors, with the second and third partition recovering

(103 and 107, respectively) than did the corresponding single the same number of putative species. The COIcommon analysis with threshold analyses (86 and 80). the default gap width (1.5) had only a single intermediate parti- tion, with 48 putative species. 3.4. PTP The ITS2all and ITS2common ABGD analyses gave identical results, except that 0–3 fewer putative species were delimited in each par-

tition for ITS2common. For clarity, we have therefore only shown the Analysis of COIall gave a total of 89 putative species, while ITS2all analysis resulted in 54 putative species. It was fairly congruent ITS2all data in Fig. 3b. Using any prior intraspecific divergences at with GMYC where lineages were relatively deep (Fig. 2A), but more or below 0.001 resulted in the same number of putative species conservative with respect to splitting taxa when divergences were (COIall: 120; COIcommon: 117). With priors above 0.001, the number shallow (Fig. 2B). When only ‘‘common” taxa were included, sev- of putative species dropped rapidly. With a prior of 0.0013, 27 and eral more putative species were delimited in the COI analysis 59 putative species were recovered by the two ITS2all analyses (gap width 1.5 and 0.75, respectively) and with a prior of 0.0078 or (COIcommon: 84 putative species) compared to ITS2 (61 putative species). Interestingly, the ‘‘common” datasets tended to split taxa greater, fewer than 20 putative species were recovered by all in comparison to the ‘‘all” datasets, to the point that more putative analyses. species were delimited in the ITS2common analysis compared to ITS2all. 3.6. Threshold analysis

3.5. ABGD The number of putative species delimited for both COI and ITS2 initially dropped steeply as the threshold increased, eventually For the COI datasets, ABGD delimited the same number of puta- flattening out (Fig. 4). The curve was more gradual for COI, not tive species regardless of which distance matrix was used as the reaching a clear plateau until over 5% sequence divergence, input data (K2P or uncorrected p-distance). In the ITS2 analyses, whereas for ITS2, the curve flattened out by approximately 1% a few intermediate partitions had a maximum of three additional divergence. A total of 128, 90, and 47 putative species were delim- putative species with the uncorrected distance matrix, compared ited by the analysis of COIall at thresholds of approximately 1%, 2%, with the K2P distance matrix. Since the results were so similar, and 3%, respectively (7, 13, and 20 base pairs). At these thresholds, we are only reporting the results of the uncorrected analysis. the COIcommon dataset delimited 74, 43, and 20 putative species. The relative gap width had a strong influence on the intermedi- Comparing this to the ITS2common dataset, the most similar num- ate partitions of both the COI and ITS2 analyses (Fig. 3). For exam- bers of putative species were found at thresholds of 0.3%, 0.8%, ple, at a prior intraspecific divergence of 0.0017 (COI) or 0.0013 and 2.6–3% (3, 7, and 23–26 base pairs). (ITS2), the more sensitive analysis (gap width = 0.75) recovered over twice as many putative species as the default gap width 3.7. Test species (=1.5). The number of putative species recovered by each COI analysis The mean percent sequence divergence within each test species dropped precipitously between the first and second partition (COI ranged from 0.04% to 0.6% divergence for COI. The highest 240 M.D. Schwarzfeld, F.A.H. Sperling / Molecular Phylogenetics and Evolution 93 (2015) 234–248

Fig. 2. Demonstration of putative species limits as obtained through different quantitative delimitation methods. The phylogenetic tree is the coalescent tree of COI obtained through BEAST analysis; it is being used for representation, but is not intended to be a definitive phylogenetic tree for the genus. Section A of the tree demonstrates relatively good congruence between methods, while section B demonstrates wide-spread incongruence. Red = GMYC: c–s = coalescent prior, single threshold; c–m = coalescent prior, multiple threshold; y–s = Yule prior, single threshold; y–m = Yule prior, multiple threshold; Green = PTP; Orange = jMOTU: 1 = 1%, 2 = 2%, 3 = 3%; Blue = ABGD: a = 1.5 gap width, 2nd partition; b = 0.75 gap width, 2nd partition; c = 1.5 gap width, 4th partition; d = 0.75 gap width, 4th partition. Each bar represents a putative species. Black bars indicate the associated specimen(s) were not included within the putative species (i.e. taxa were not monophyletic with respect to this tree). Numbers are used to indicate when two separate bars belong to the same species (non-monophyletic on this tree). Identical haplotypes were removed for this phylogenetic analysis; therefore, some apparent singletons may be represented by multiple specimens in the complete dataset. M.D. Schwarzfeld, F.A.H. Sperling / Molecular Phylogenetics and Evolution 93 (2015) 234–248 241

Fig. 3. Results of ABGD analyses over a range of prior intraspecific divergences with two different gap widths (1.5 and 0.75). (A) COI: all = all sequences; common = only including sequences in common with ITS2. (B) ITS2: all sequences (ITS2common not shown, since results are equivalent).

CS, CM, and YS analyses of the ITS2 data each delimited 5 of the 10 species but which species were successfully delimited varied between the analyses. The YM analysis of ITS2 was the least suc- cessful GMYC analysis, recovering only two species. O. nigrovarius was the only species to be recovered by all GMYC analyses and both markers. Similarly, species were more successfully delimited in the PTP analysis of COI, compared to ITS2. However PTP in general was less successful than GMYC, with the COI analysis correctly delimiting four species out of eleven, while in the ITS2 analysis only a single species was successfully delimited (Table 4). Sp. B was never recovered by ABGD, being either split or lumped, depending on the parameters used (Table 5). All other species were recovered by at least one ABGD analysis. Again, the analyses of COI were generally more successful than the ITS2 anal- yses, with most species being successfully delimited over a range Fig. 4. Number of species delimited by jMOTU at thresholds of 0–50 base pairs of partitions. The more sensitive analysis of COI (gap width = 0.75) (equivalent to 7.5% (COI) or 5.7% (ITS2) sequence divergence); ITS2common not was most successful, with all species except for Sp. B being cor- shown as results are equivalent to ITS2 . all rectly delimited in at least one, and usually several, partitions. Within this analysis, the prior intraspecific divergence of 0.0046 Table 3 was the only partition that correctly delimited all species (exclud- Mean and maximum percent sequence divergence of COI and ITS2 within each ing Sp. B). ABGD of the ITS2 dataset recovered 7 of the 10 species in Nearctic test species that was represented by more than one individual. Divergences all partitions with prior intraspecific divergences 60.001); three of are uncorrected p-distances (%), with gaps and missing data treated by pairwise deletion. these species were also recovered in additional partitions. The remaining three species were never successfully delimited. Species COI mean COI max ITS2 mean ITS2max All of the test species were successfully delimited by jMOTU aureus 0.4 0.4 0.0 0.0 with the exception of Sp. B in the analysis of COI and O. clave in clave 0.2 0.3 0.0 0.0 the analysis of ITS2 (Table 5). In both cases, these species were split idoneus 0.04 0.3 0.0 0.0 keala 0.4 0.7 0.0 0.0 at low thresholds, and lumped with other sequences at higher nigrovarius 0.2 0.3 0.0 0.0 thresholds, but never delimited successfully. In the COI analysis, slossonae 0.3 0.4 – – all remaining species were delimited at a threshold of 1% sequence Sp. A 0.3 0.8 0.0 0.0 divergence, and all except O. brevipunctatus and O. dombroskii at 2% Sp. B 0.6 0.9 0.1 0.3 divergence. Only three species (O. nigrovarius, O. slossonae, and O. keala) were delimited at 3% divergence or higher. For the ITS2 anal- ysis, most species were delimited at very low divergences, with maximum divergence within any species was found in Sp. B, with only O. nigrovarius and Sp. A being successfully delimited at thresh- 0.9% divergence. ITS2 intraspecific divergence was almost non- olds of more than 1% divergence (9 base pairs). existent, with only sp. B showing any divergence (mean 0.1%, max- imum 0.3%) (Table 3). GMYC analysis of COI was generally more successful at correctly 3.8. Species delimitation of Palearctic Ophion delimiting species than was that of ITS2 (Table 4). Within the COI analysis, the coalescent or Yule tree priors were equally successful Compared to the Nearctic test species, sequence divergence at recovering the test species. The single threshold method was within several of the Palearctic ‘‘species” was high (Table 6). Nine most successful, correctly delimiting all species except O. bre- species had maximum intraspecific divergences of over 2%. Two vipunctatus and O. dombroskii, which were recovered as a single species (O. pteridis and O. perkinsi) were represented by a single cluster. The multiple-threshold method also lumped these two haplotype, and sequence divergence within O. ventricosus and O. species, and in addition failed to recover O. keala and Sp. A. The ocellaris were also very low (Table 6). 242 M.D. Schwarzfeld, F.A.H. Sperling / Molecular Phylogenetics and Evolution 93 (2015) 234–248

Table 4 Summary of congruence between generalized mixed-Yule coalescence analyses (GMYC) and Poisson tree processes (PTP) with eleven species/morphospecies of Nearctic Ophion. C = coalescent tree prior; Y = Yule tree prior; S = single threshold model; M = multiple threshold model; + = species successfully delimited; L = species lumped with other sequences; Sp = species split into >1 groups.

Species COI ITS2 COI ITS2 (n)(n) CS CM YS YM PTP CS CM YS YM PTP clave 5 2 + + + + + + + + L Sp/L aureus 24++++L +LLLL brevipunctatus 11LLLLL LLLLL dombroskii 11LLLLL LLLLL keala 7 7 + Sp + Sp Sp L Sp L L L importunus 11++++L +++LL idoneus 13 13 + + + + L L + L L L nigrovarius 52+++++++++Sp slossonae 3 0 + + + + + n/a n/a n/a n/a n/a sp. A 11 7 + L + L + + + + + + sp. B 9 9 + + + + L Sp Sp + L L

Table 5 Table 7 Summary of congruence between automatic barcode gap discovery (ABGD) and Number of putative species delimited for each morphologically identified species of threshold analyses with eleven species/morphospecies of Nearctic Ophion. ABGD used Ophion, based on the COI barcode region. All specimens are from the United Kingdom, two gap widths: 1.5 and 0.75; numbers represent the range of prior intraspecific with the exception of a single specimen of O. obscuratus from Spain and one specimen percent divergence that successfully delimited each species. jMOTU = threshold of O. forticornis from France. Analyses that successfully delimited a species are analysis using the program jMOTU; numbers indicate the range of thresholds, as a identified with +. The GMYC analysis used the single threshold method on an percentage of the mean sequence length, that successfully delimited each species; for ultrametric tree produced in BEAST, with a relaxed lognormal clock and coalescent both analyses, – indicates the species was never delimited. tree prior. PTP was conducted on a maximum likelihood tree obtained with GARLI. The ABGD analysis used a relative gap width of 0.75 and a prior intraspecific ABGD (%) jMOTU (%) divergence of 0.0046. The jMOTU analysis used thresholds of 7, 13 and 20 base pairs, Species COI ITS2 COI ITS2 or approximately 1%, 2%, and 3% sequence divergence.

1.5 0.75 1.5 0.75 Number of putative species clave 0.2–0.8 0.2–1.3 60.01 60.01 0.3–2.9 – Species n GMYC PTP ABGD jMOTU aureus – 0.5 60.01 60.01 1.1–2.0 0.1–0.3 1% 2% 3% brevipunctatus 0.1 0.1–0.5 – – 61.1 60.2 dombroskii 0.1 0.1–0.5 – – 61.1 60.1 brevicornis 32 2 + 3b 3b 2b keala 0.2–0.8 0.5–1.3 60.01 60.01 0.6–4.1 0.1 costatus 13 3a,c 4a,c 4a,c 5c 5c 4b,c importunus 0.1–0.8 0.1–1.3 <0.01–0.06 <0.01–0.06 62.3 60.8 crassicornis 42 2 2 22+ idoneus 0.2–0.8 0.2–1.3 60.01 60.01 0.3–2.3 0.1 forticornis 1+ + + +++ nigrovarius 0.5–0.8 0.5–1.3 <0.01–1.7 <0.01–1.7 0.5–5.2 0.1–4.3 longigena 1+ + + +++ slossonae 0.2–0.8 0.2–1.3 n/a n/a 0.5–4.5 n/a luteus 14 2 + 2 2 1b 1b sp. A 0.2–0.8 0.2–1.3 <0.01–0.4 <0.01–0.4 0.6–2.3 0.1–2.6 minutus 82 + 3 42+ sp. B – – – – – 0.5–0.8 mocsaryi 93a,c 2a,c 4a,c 5c 4c 2b,c obscuratus 73 4b 35b 5b 3b ocellaris 3+ + + +++ parvulus 32 2 2 222b perkinsi 2+ 2 + ++1b b b b Table 6 pteridis 3+ 1 ++11 Mean and maximum percent sequence divergence of COI within each Palearctic scutellaris 52 2 2 22+ species that was represented by more than one individual. Divergences are ventricosus 4+ + + 2++ uncorrected p-distances (%), with gaps and missing data treated by pairwise deletion. a Two sequences identified as O. mocsaryi and two identified as O. costatus were Species Mean Max recovered as a single species by ABGD and PTP, and as a monophyletic group split into two species by GMYC. brevicornis 1.0 2.1 b At least one of the delimited species lumped with other, presumably non- costatus 2.4 6.0 conspecific sequences. crassicornis 1.5 2.2 c One sequence identified as O. costatus was consistently found within one of the luteus 0.6 2.7 O. mocsaryi species – likely a misidentification of this specimen. minutus 1.4 3.7 mocsaryi 2.2 4.2 obscuratus 4.9 7.9 ocellaris 0.1 0.2 parvulus 9.5 9.5 divergent taxa. For example, at thresholds of 2% and 3% divergence, perkinsi 0.0 0.0 it combined O. pteridis and O. luteus, along with several Nearctic pteridis 0.0 0.0 species, into one large group, even though they differ by approxi- scutellaris 1.4 3.5 mately 6% sequence divergence. ventricosus 0.1 0.3 Of the 15 Palearctic Ophion species that were analyzed, three (O. forticornis, O. longigena, O. ocellaris) were delimited by all analyses, although two of these species (O. forticornis and O. longigena) were represented by singletons. O. ventricosus was recovered as a single ABGD, GMYC, and PTP were broadly congruent in their delimi- species by all except the 1% jMOTU analysis, and O. perkinsi by all tation of Palearctic species, with only slight differences in the num- except PTP (oddly, since it was represented by a single haplotype). ber of species within a few clades (Table 7). In comparison, while Also represented by a single haplotype, O. pteridis were never split the jMOTU analyses often split species into a similar number of into multiple species; however, according to the 2% and 3% jMOTU groups, in several instances it lumped these putative species with analyses, it was lumped with other sequences. All other species other sequences. In some cases jMOTU analyses combined highly were split to various extents by the different analyses. M.D. Schwarzfeld, F.A.H. Sperling / Molecular Phylogenetics and Evolution 93 (2015) 234–248 243

4. Discussion example, O. luteus, O. pteridis, and a number of Nearctic species, were grouped into a single MOTU, as there were apparently 4.1. Sampling enough intermediate sequences to act as a bridge between them. However, a trial analysis with the dataset restricted to O. luteus All quantitative species-delimitation methods can be misled if and O. pteridis retained them as separate MOTUs at thresholds of species are not adequately sampled, a common issue in diverse, up to 6% divergence. understudied taxa (Lim et al., 2012; Lohse, 2009; Papadopoulou Hamilton et al. (2014) recommend incorporating prior knowl- et al., 2009). In particular, undersampling can lead to either under- edge of natural history, genetic variation, and morphology in order estimation of species numbers (since some species are not sam- to design an effective sampling strategy for accurate species delim- pled) or overestimation (since some of the diversity within itation. This may be difficult to achieve with the first attempt at species may not be sampled, resulting in apparent genetic structur- delimiting species in an almost entirely unexplored genus, such ing) (Papadopoulou et al., 2009). However, determining what level as Ophion. However, as knowledge of the taxon accumulates, it of sampling is considered ‘‘adequate” is not a trivial task. Species should be possible to gradually ‘‘fill in the holes”, and thus refine ranges, geographical structuring, effective population sizes, and improve confidence in the delimitation results. intraspecific genetic variation, habitat niches (possibly including host-races in the case of parasitoids) all play a role in the observed 4.2. Comparison of ITS2 and COI as markers for species delimitation of data (Bergsten et al., 2012; Hamilton et al., 2014). In theory, even Ophion sampling across the genetic and geographic ranges of each species would improve delimitation methods; however in practice this is ITS2 in Ophion has a very different pattern of divergence com- likely impossible to achieve. For example, Bergsten et al. (2012) pared to COI, with high divergence between species groups, but found up to 70 individuals per species of diving beetle were very low intra- and interspecific divergence among closely related required to sample 95% of intraspecific variation. species (Schwarzfeld and Sperling, 2014; Schwarzfeld, unpublished In our study, none of the barcode accumulation curves reached data). Nevertheless, the branching pattern was largely congruent an asymptote, a common finding among diverse taxa such as par- with COI, and at least according to GMYC, a similar number of spe- asitoids (Fernández-Triana et al., 2011; Smith et al., 2009; Zaldívar- cies were delimited. Riverón et al., 2010). This study is almost entirely restricted to Due to its length, the presence of microsatellite regions, and the specimens from Canada and the northern United States. While high number of indels, ITS2 was more logistically challenging to the southern Nearctic region is almost entirely unexamined, the sequence and align for Ophion. However, both ITS2 and COI have more extensively studied Canadian fauna is also insufficiently sam- advantages and disadvantages with regards to species delimita- pled. The most densely sampled region is the province of Alberta, tion. In this study, COI successfully identified more test species in which provided 52% of the sequences for this study; the curve rep- GMYC and PTP, and the choice of priors or thresholds for ABGD resenting these sequences appears to be approaching an asymp- and jMOTU is more straightforward, due to the large number of tote, but even here, more sampling is needed to fully assess the studies focusing on this gene. In some studies, however, ITS2 has genetic diversity of these species. Included samples were collected been found to be more effective than COI at distinguishing closely haphazardly, in many cases as bycatch from moth collectors. related species (e.g. Wagener et al., 2006). As well, certain indels of Future studies should therefore place a high priority on sampling ITS2 can be a species-specific character, and have been used to suc- throughout a wider range of habitats and geographic regions. cessfully distinguish species of parasitic wasps in several studies A high proportion of singleton species were recovered in all (Klopfstein, 2014; Quicke et al., 2006; Van Veen et al., 2003). analyses, another typical result for diverse, understudied taxa While beyond the scope of the current study, more detailed exam- (Ceccarelli et al., 2012; Lim et al., 2012; Monaghan et al., 2009). ination of indels may resolve some of the ambiguously delimited The robustness of species-delimitation algorithms with regard to species of Ophion, undersampled species is an important consideration as singleton Despite the apparent advantages of COI in this study, relying on species are abundant in taxonomic and biodiversity studies (Lim this single gene may provide misleading results due to incomplete et al., 2012). The GMYC method has been found to be accurate even lineage sorting, introgression, bacterial endosymbionts, or nuclear with a high proportion of singleton species (Ceccarelli et al., 2012; mitochondrial pseudogenes (numts) (Dupuis et al., 2012; Funk and Monaghan et al., 2009; Talavera et al., 2013). In contrast, for accu- Omland, 2003; Raychoudhury et al., 2010; Schmidt and Sperling, rate results with ABGD, it is recommended that 3–5 sequences 2008; Song et al., 2008). In at least one species in this study, an should be included per species (Puillandre et al., 2012a). apparent numt (Gellissen et al., 1983; Lopez et al., 1994) was However, without a priori knowledge of the group being studied, amplified, as there is a homologous 2 base pair deletion in each this may be difficult to achieve, and at least in our study, the sin- sequence which would result in a shift in the codon reading frame gletons did not appear to negatively affect the results. PTP has been of the COI gene. Since the GMYC analysis of ITS2 also recovered the found to overestimate diversity when sampling is uneven between same species, it appears this numt may be taxonomically informa- species (Zhang et al., 2013), yet despite the uneven sampling in this tive. However not all numts are as readily recognizable, and can be study, PTP was more prone to underestimating the number of spe- amplified by chance in taxa that are not closely related. It is there- cies, compared to GMYC. However, this pattern may be indicated fore critical to include an independent molecular marker to verify by the increased splitting of putative species when sequences were the results obtained with COI. ITS2 is an important tool for delim- removed in the ‘‘common” analyses. Further studies of the effect of iting species in this genus, though the investigation of other rela- taxon sampling with regards to PTP are needed. tively rapidly evolving nuclear genes (e.g. CAD) (Sharanowski Increased sampling does not always lead to more accurate spe- et al., 2011; Wild and Maddison, 2008) may provide additional cies delimitation, particularly in threshold-based analyses. An and potentially more tractable species-level genetic markers. apparently distinct barcoding gap often disappears with increased sampling across the range of genetic variation of a given taxon 4.3. CBC analysis (Bergsten et al., 2012; Meyer and Paulay, 2005; Moritz and Cicero, 2004; Wiemers and Fiedler, 2007). In this study, at the The potential for a single compensatory base change (CBC) to commonly used 2% threshold, the single-linkage clustering algo- unambiguously identify species as being reproductively isolated rithm used by jMOTU clustered several distantly-related taxa. For has been a promising ‘‘magic bullet” for species delimitation 244 M.D. Schwarzfeld, F.A.H. Sperling / Molecular Phylogenetics and Evolution 93 (2015) 234–248

(Coleman, 2009; Müller et al., 2007; Wolf et al., 2013). The initial several partitions for both COI and ITS2. As well, there was no clear studies focused on plants and fungi, but did not assess the plateau in the estimated number of species with the more sensitive utility of CBC’s in animal taxa. The first test of CBC’s for species gap width, making the choice of prior a subjective but essential delimitation of animals was provided by Ruhl et al. (2010), task if this analytical method is to be used to assess species diver- where they found it to effectively delimit two morphologically sity. While both gap widths recovered most test species, in the COI indistinguishable species of flea beetle (Chrysomelidae: Altica). analysis, more species were recovered over a wider range of priors However, studies of blue butterflies (Lycaenidae; Wiemers et al., using the more sensitive gap width. The sensitivity of the ABGD 2009), mosquitoes (Alquezar et al., 2010), and budworm analysis to different gap widths is most apparent in one species- parasitoids (Smith et al., 2011) found CBC’s were too rare to group of apparently weakly diverged species (Schwarzfeld, unpub- distinguish closely-related species. The almost complete lack of lished data). For example, in the analysis of COI using an CBC’s between closely related Ophion species provides further intraspecific prior of 0.0017, the default analysis estimated 15 spe- evidence that for many insects, CBC’s are apparently too cies within this group, whereas the more sensitive analysis recov- scarce to be of practical use in distinguishing closely-related ered 81 species (Fig. 2b shows part of this group). The one test species. They are, however, potentially useful in defining higher species within this species group (Sp. B) was not accurately delim- taxonomic groups. ited in either case, being split with the more sensitive analysis and lumped with the default analysis. The sensitivity of the analysis to 4.4. Parameter selection and species delimitation the gap width may only be an issue within species complexes or recent radiations, which can confound any species-delimitation Two main parameters that must be selected for GMYC analyses method (Monaghan et al., 2006; Reid and Carstens, 2012). are the priors used to construct the phylogenetic tree, and the Nonetheless, the impact of gap width should be considered when choice of single- or multiple-threshold methods. While the using this method, since the choice of this parameter is largely multiple-threshold method of GMYC was developed to take into arbitrary. account different branching patterns and rates of evolution across While gap width also affected the intermediate partitions of the a tree (Monaghan et al., 2009; Papadopoulou et al., 2009), evidence ITS2 analysis this had little impact on the overall results, as only suggests it is consistently less accurate than the single-threshold three test species were recovered with any intraspecific divergence method (Esselstyn et al., 2012; Fujisawa and Barraclough, 2013). above 0.001. This supports the results seen in the O. scutellaris This study supports this finding with fewer test species being cor- group (Schwarzfeld and Sperling, 2014) where all species had a rectly delimited with this method in both the analyses of COI (both maximum of 0.1% intraspecific divergence. tree priors) and ITS2 (Yule prior). While the multiple-threshold Based on thresholds of 1–3%, there was almost a threefold dif- method did not always recover higher numbers of species than ference in the number of species estimated by jMOTU, from 47 spe- did the corresponding single analysis, all of the highest estimates cies at 3% divergence to 128 species at 1%. While 2% is one of the were obtained using this method, further supporting the conclu- most commonly used thresholds in DNA barcoding studies (e.g. sion of Esselstyn et al. (2012) that the multiple-threshold method Bribiesca-Contreras et al., 2013; Fernandez-Flores et al., 2013; overestimates the number of species. Smith et al., 2013; Strutzenberger et al., 2011), other thresholds Yule priors are generally considered more appropriate for anal- are also used, such as 1.6% (Smith et al., 2009), 2.3% (Young yses of speciation, whereas coalescent priors are more applicable et al., 2012), 3% (Hendrich et al., 2010; Tang et al., 2012), or 3.2% to intraspecific population level data (Drummond et al., 2006). In (Weigand et al., 2013). Although the choice of threshold is some- a GMYC analysis of a diverse genus of Braconidae, Ceccarelli times based on empirical studies (e.g. Hebert et al., 2003a; et al. (2012) found the accuracy of each prior depended on the gene Weigand et al., 2011), there is no reason to believe the same being analyzed, with the coalescent prior being more consistent thresholds will apply in different taxa; even within a taxon, there with morphology for COI data. However, this result may be can be highly variable patterns of intra- and interspecific diver- taxon-specific; in the current study, both the estimated number gence (e.g. Bergmann et al., 2013). of species and the accuracy of the analyses were largely unaffected Despite this range, with a standard 2% threshold the estimated by which tree prior was used to construct the phylogenetic tree. number of species (90) was within the range of the GMYC and the The birth–death prior is another alternative that may be appropri- intermediate ABGD analyses, and most test species were recov- ate for species diversification (Gernhard, 2008; Nee et al., 1994), ered. However, due to the ‘‘greedy algorithm” used in the single- though it was not tested in the current study. linkage clustering employed by this method (Jones et al., 2011), While fewer parameters must be selected for PTP, the analysis the delimitation of species boundaries sometimes differed consid- does depend on the accuracy of the input phylogenetic tree. We erably from the other two methods (Fig. 2). only tested a single ML tree, and therefore cannot determine if the use of different models or tree-building algorithms would 4.5. Comparison of methods affect the results. Tang et al. (2014) found that PTP was quite robust with respect to the phylogenetic methods used; however, We concluded that GMYC and PTP offer some distinct advan- the tendency of taxa to be more split when fewer specimens were tages over the other two methods of species delimitation. The lar- included again indicates the importance of taxon sampling for phy- gest advantage is that these methods incorporate evolutionary logenetic reconstruction. theory and are therefore much less reliant on the application of The results of the ABGD analyses were highly dependent on the arbitrary thresholds. Because of this, they are also more tractable gap width selected. In most recent studies that have employed this with genetic markers such as ITS2 that have less well- method, the default gap width of 1.5 is either explicitly used characterized intra- and interspecific patterns of diversity. With (Hendrixson et al., 2013; Kekkonen and Hebert, 2014; Tang et al., appropriate tree-building methodologies, these methods can easily 2012), or implicitly used (Crawford et al., 2012; Hamilton et al., be applied to any genetic marker. In comparison, choosing an 2014; Puckridge et al., 2013). However, Puillandre et al. (2012b) appropriate threshold or range of thresholds for a marker other increased this setting to 10 in order to decrease the sensitivity of than COI is particularly arbitrary, since no ‘‘standard” divergence their analysis to small gaps, while Weigand et al. (2013) used a has been accepted for species delimitation. range of more sensitive gap widths (0.01–0.9). In this study, reduc- Nonetheless, caution must be employed when using either of ing the gap width doubled the number of delimited species in these methods. One disadvantage of single-locus tree-based M.D. Schwarzfeld, F.A.H. Sperling / Molecular Phylogenetics and Evolution 93 (2015) 234–248 245 methods is that all species, by definition, must be monophyletic, 4.6. How many Ophion species are there? despite evidence that a high proportion of species are not mono- phyletic with respect to their gene trees (Funk and Omland, Based on COI (since many species were not successfully 2003). GMYC relies on the priors and parameters used to con- sequenced for ITS2), there was a wide range of estimates for the struct the ultrametric tree (Ceccarelli et al., 2012), and the con- total number of species present in this study. Considering all plau- struction of the trees can be computationally demanding. GMYC sible results of the quantitative species-delimitation analyses (i.e. may also overestimate species diversity compared to other meth- including all GMYC analyses; PTP; ABGD with gap widths of 1.5 ods (Kekkonen and Hebert, 2014; Hamilton et al., 2014; Miralles and 0.75 and priors of 0.0017–0.0077 and 0.0017–0.0129 and Vences, 2013; Paz and Crawford, 2012; Talavera et al., 2013; intraspecific divergence, respectively; and jMOTU thresholds of Weigand et al., 2013). PTP is far less computationally demanding 1–3% divergence), estimates ranged from 47 to 168 species of than the GMYC and since an ultrametric tree is not required, Ophion represented in this study. Excluding the more extreme esti- relies on fewer tree-building assumptions. It has been found to mates (i.e. including only single threshold GMYC analyses; PTP; be as accurate, if not more so, than GMYC when tested against ABGD analyses with a prior of 0.0046; and jMOTU results with a well-defined species (Dumas et al., 2015; Modica et al., 2014; threshold of 2%) resulted in estimates of 89–121 species (64–97 Tang et al., 2014; Zhang et al., 2013). In our study, it delimited of which are from the Nearctic region). considerably fewer putative species than the GMYC analyses, pos- The wide range of estimates emphasizes that all methods of sibly an advantage due to the hypothesized oversplitting by molecular species delimitation provide only ‘‘primary species GMYC. However, since it lumped several of the test species that hypotheses” (Puillandre et al., 2012b), which need to be assessed were correctly delimited by GMYC, its success likely depends to using additional methods. As a first step, these methods can be a a large extent on the phylogenetic patterns within the taxon in helpful tool to delimit hypothetical species; however, for compar- question. ative studies relying on species numbers (e.g. for conservation Both jMOTU and ABGD provide the flexibility of examining a applications or for analysis of patterns of biodiversity), it is vital range of thresholds or priors, respectively, which can be informa- that researchers also consider the parameters and potential short- tive in understanding the effects of the algorithms and the diver- comings of the delimitation method(s) used to obtain the sity within the group. However, the majority of barcoding studies estimates. use a single threshold to estimate species diversity (e.g. Our finding that most described British species were split may Bribiesca-Contreras et al., 2013; Fernandez-Flores et al., 2013; be due to the failure of delimitation methods for successfully cir- Smith et al., 2013; Strutzenberger et al., 2011; but see Laetsch cumscribing true biological species, or alternatively may indicate et al., 2012). A range of priors is more often examined with the presence of cryptic or previously unrecognized species. In ABGD (Paz and Crawford, 2012; Puckridge et al., 2013); however, several instances, putative species were not recovered as a mono- the impact of the relative gap width is rarely considered, even phyletic clade by phylogenetic analyses, indicating that multiple, though it can have a significant impact on the results obtained. A previously unrecognized species are almost certainly present. As further disadvantage of threshold analysis as implemented in well, the steep slope of the barcode accumulation curve for jMOTU is the grouping of distantly related species among taxa that these specimens suggests that genetic sampling is far from lack a clear barcoding gap. Despite these issues, distance-based complete. species-delimitation methods do not rest on the assumption of The use of a variety of analytical methods and parameters can reciprocal monophyly between species. It is therefore recom- aid in planning subsequent research to assess the hypothetical spe- mended that at least one distance-based and one tree-based cies obtained with these methods (Carstens et al., 2013). For exam- method be used in any species-delimitation study (Fontaneto ple, groups of sequences that are consistently delimited with a et al., 2015). variety of quantitative methods are good candidate species to All of the above methods are single-locus methods. While using examine in more detail for confirmatory morphology, ecology or single-locus methods on multiple loci independently allows quali- other sources of data that either support or refute the tative comparisons and reduces the potential biases from relying molecularly-defined species (Kekkonen and Hebert, 2014). In com- on a single molecular marker, several multi-locus methods have parison, species that are inconsistently delimited, while also been developed that can better account for some of the issues requiring an integrative taxonomy approach, are good candidates described above (e.g. gene-tree discordance, non-monophyletic for more fine-scaled analyses, such as quantitative morphometrics species, recently-diverged species with gene flow, etc.) (Dupuis (Lumley and Sperling, 2010; Schwarzfeld and Sperling, 2014)or et al., 2012; Fontaneto et al., 2015). Examples of multi-locus meth- population genetics (Shaffer and Thomson, 2007; Lumley and ods include: Bayesian phylogenetics and phylogeography (BP&P; Sperling, 2011). Yang and Rannala, 2010; Rannala and Yang, 2013); SpedeSTEM (Ence and Carstens, 2011); O’Meara’s heuristic model (O’Meara, 2010); and Bayes Factor Delimitation (BFD; Grummer et al., 2014). BP&P is one of the most widely-used methods; however, 5. Conclusion until recently it required a user-supplied guide tree, thus being less effective for unguided species-delimitation in entirely unknown Based on morphology, Gauld (1985) estimated a fauna of taxa. Recently, methods have been developed that do not require approximately 50 Nearctic species of Ophion. This study indicates previous knowledge of the included species, such as an unguided that the true number of Nearctic species is considerably higher version of BP&P (Yang and Rannala, 2014) and DISSECT (Jones and with more complete sampling will easily double that estimate. et al., 2015). Most multi-locus methods are computationally- The 17 described species of Ophion therefore represent just a min- expensive, and, depending on the study system, are arguably ute fraction of the total fauna. Similarly, even for the relatively ‘‘overkill” (Collins and Cruickshank, 2014). Nevertheless, as well-studied British fauna, there appear to be a surprising number sequencing becomes increasingly affordable, multi-locus methods of previously unrecognized species. It is widely acknowledged that will no doubt become the norm. Future studies of Ophion, and of there are vast numbers of undescribed species among tropical, species-delimitation studies in general, should include multi- small-bodied taxa (e.g. Quicke, 2012). However, this study shows locus methods. that even in a large-bodied, primarily temperate genus that is 246 M.D. Schwarzfeld, F.A.H. Sperling / Molecular Phylogenetics and Evolution 93 (2015) 234–248 readily attracted to lights and easily collected in large numbers, transcribed spacer (ITS2) and 28S rDNA sequences. Insect Molec. Biol. 2, 225– there is a wealth of diversity waiting to be discovered. 237. Carstens, B.C., Dewey, T.A., 2010. Species delimitation using a combined coalescent and information-theoretic approach: an example from North American Myotis Acknowledgments bats. Syst. Biol. 59, 400–414. Carstens, B.C., Pelletier, T.A., Reid, N.M., Satler, J.D., 2013. How to fail at species delimitation. Mol. Ecol. 22, 4369–4383. We are very grateful to Gavin Broad for loaning and donating Ceccarelli, F.S., Sharkey, M.J., Zaldívar-Riverón, A., 2012. Species identification in the British Ophion, for sending Ophion legs for sequencing, and for taxonomically neglected, highly diverse, neotropical parasitoid wasp genus Notiospathius (Braconidae: Doryctinae) based on an integrative molecular and freely sharing his knowledge of British Ophion. Thanks also to morphological approach. Molec. Phylogenet. Evol. 62, 485–495. Mark Shaw for sending a leg from O. forticornis. We wish to thank Claridge, M.F., Dawah, H.A., Wilson, M.R. (Eds.), 1997. Species: The Units of Jeremy deWaard for providing many specimens as well as nearly Biodiversity. Chapman and Hall, London. Cognato, A.I., 2006. Standard percent DNA sequence difference for insects does not 100 COI sequences through BOLD. This project would not have predict species boundaries. J. Econ. Entomol. 99, 1037–1045. been possible without the many collectors who provided speci- Coleman, A.W., 2007. Pan-eukaryote ITS2 homologies revealed by RNA secondary mens, especially: John Acorn, John Adams, Gary Anweiler, structure. Nucl. Acids Res. 35, 3322–3329. Coleman, A.W., 2009. Is there a molecular key to the level of ‘‘biological species” in Charley Bird, Heather Bird, Brett Bodeux, Bryan Brunet, Jason eukaryotes? A DNA guide. Molec. Phylogenet. Evol. 50, 197–203. Dombroskie, Julian Dupuis, Jessica Edwards, Dave Holden, Wes Coleman, A.W., Vacquier, V.D., 2002. Exploring the phylogenetic utility of ITS Hunting, Dave Langor, Nina Lany, Dave Lawrie, Lisa Lumley, Doug sequences for animals: a test case for abalone (Haliotis). J. Molec. Evol. 54, 246– 257. Macaulay, Boyd Mori, Ted Pike, Greg Pohl, and Chris Schmidt. We Collins, R.A., Cruickshank, R.H., et al., 2014. Known knowns, known unknowns, would also like to thank Martin Jones for assistance with jMOTU unknown unknowns and unknown knowns in DNA barcoding: a comment on and Matthias Wolf for help regarding 4Sale and ProfDistS. Thank Dowton. Syst. Biol. 63, 1005–1009. you to Seraina Klopfstein and two anonymous reviewers for their Crawford, A.J., Cruz, C., Griffith, E., Ross, H., Ibanez, R., Lips, K.R., Driskell, A.C., Bermingham, E., Crump, P., 2012. DNA barcoding applied to ex situ tropical constructive comments on an earlier version of this manuscript. amphibian conservation programme reveals cryptic diversity in captive This work was supported by: the Sustainable Forest Management populations. Mol. Ecol. Res. 13, 1005–1018. http://dx.doi.org/10.1111/1755- Network; Alberta Conservation Association Grants in 0998.12054. Dayrat, B., 2005. Towards integrative taxonomy. Biol. J. Linn. Soc. 85, 407–415. Biodiversity; Natural Sciences and Engineering Research Council DeSalle, R., 2007. Phenetic and DNA taxonomy; a comment on Waugh. BioEssays 29, (NSERC) and Alberta Ingenuity (Alberta Innovates) student scholar- 1289–1290. ships to MDS; and an NSERC Discovery grant to FAHS. Drummond, A.J., Ho, S.Y.W., Phillips, M.J., Rambaut, A., 2006. Relaxed phylogenetics and dating with confidence. PLoS Biol. 4, 699–710. Drummond, A.J., Suchard, M.A., Xie, D., Rambaut, A., 2012. Bayesian phylogenetics with BEAUti and the BEAST 1.7. Molec. Biol. Evol. 29, 1969–1973. Appendix A. Supplementary material Dumas, P., Barbut, J., Le Ru, B., Clamens, A.-L., d’Alençon, E., Kergoat, G.J., 2015. Phylogenetic molecular species delimitations unravel potential new species in Supplementary data associated with this article can be found, in the pest genus Spodoptera Guenée, 1852 (Lepidoptera, Noctuidae). PLoS ONE 10, e0122407. http://dx.doi.org/10.1371/journal.pone.0122407. the online version, at http://dx.doi.org/10.1016/j.ympev.2015.08. Dupuis, J.R., Roe, A.D., Sperling, F.A.H., 2012. Multi-locus species delimitation in 003. closely related animals and fungi: one marker is not enough. Mol. Ecol. 21, 4422–4436. Ence, D.D., Carstens, B.C., 2011. SpedeSTEM: a rapid and accurate method for References species delimitation. Mol. Ecol. Res. 11, 473–480. Esselstyn, J.A., Evans, B.J., Sedlock, J.L., Khan, F.A.A., Heaney, L.R., 2012. Single-locus Agapow, P.-M., Crozier, R.H., 2008. Conserve IV, . empirical application to Philippine round-leaf bats. Proc. Roy. Soc. B 279, 3678– Alquezar, D.E., Hemmerter, S., Cooper, R.D., Beebe, N.W., 2010. Incomplete 3686. concerted evolution and reproductive isolation at the rDNA locus uncovers Fernández-Flores, S., Fernández-Triana, J.L., Martínez, J.J., Zaldívar-Riverón, A., 2013. nine cryptic species within Anopheles longirostris from Papua New Guinea. BMC DNA barcoding species inventory of Microgastrinae wasps (Hymenoptera, Evol. Biol. 10, 392. http://dx.doi.org/10.1186/1471-2148-10-392. Braconidae) from a Mexican tropical dry forest. Mol. Ecol. Res. 13, Alvarez, J.M., Hoy, M.A., 2002. Evaluation of the ribosomal ITS2 DNA sequences in 1146–1150. separating closely related populations of the parasitoid Ageniaspis Fernández-Triana, J., Smith, M.A., Boudreault, C., Goulet, H., Hebert, P.D.N., Smith, A. (Hymenoptera: Encyrtidae). Ann. Entomol. Soc. Am. 95, 250–256. C., Roughley, R., 2011. A poorly known high-latitude parasitoid wasp Astrin, J.J., Stüben, P.E., Misof, B., Wägele, J.W., Gimnich, F., Raupach, M.J., Ahrens, D., community: unexpected diversity and dramatic changes through time. PLoS 2012. Exploring diversity in cryptorhynchine weevils (Coleoptera) using ONE 6, e14424. http://dx.doi.org/10.1371/journal.pone.0023719. distance-, character- and tree-based species delineation. Molec. Phylogenet. Fontaneto, D., Flot, J.-F., Tang, C.Q., 2015. Guidelines for DNA taxonomy, with a focus Evol. 63, 1–14. on the meiofauna. Mar. Biodiv. Online early, http://dx.doi.org/10.1007/s12526- Barrett, R.D.H., Hebert, P.D., 2005. Identifying spiders through DNA barcodes. Can. J. 015-0319-7. Zool. 83, 481–491. Fontaneto, D., Herniou, E.A., Boschetti, C., Caprioli, M., Melone, G., Ricci, C., Bazinet, A.L., Zwickl, D.J., Cummings, M.P., 2014. A gateway for phylogenetic Barraclough, T.G., 2007. Independently evolving species in asexual bdelloid analysis powered by grid computing featuring GARLI 2.0. Syst. Biol. 63, 812– rotifers. PLoS Biol. 5, e87. 818. http://dx.doi.org/10.1093/sysbio. Fujisawa, T., Barraclough, T.G., 2013. Delimiting species using single-locus data and Bergmann, T., Rach, J., Damm, S., DeSalle, R., Schierwater, B., Hadrys, H., 2013. The the Generalized Mixed Yule Coalescent approach: a revised method and potential of distance-based thresholds and character-based DNA barcoding for evaluation on simulated data sets. Syst. Biol. 62, 707–724. http://dx.doi.org/ defining problematic entities by COI and ND1. Mol. Ecol. Res. 136, 1069–1081. 10.1093/sysbio/syt033. Bergsten, J., Bilton, D.T., Fujisawa, T., Elliott, M., Monaghan, M.T., Balke, M., Fujita, M.K., Leaché, A.D., Burbrink, F.T., McGuire, J.A., Moritz, C., 2012. Coalescent- Hendrich, L., Geijer, J., Herrmann, J., Foster, G.N., Ribera, I., Nilsson, A.N., based species delimitation in an integrative taxonomy. Trends Ecol. Evol. 27, Barraclough, T.G., Vogler, A.P., 2012. The effect of geographical scale of sampling 480–488. on DNA barcoding. Syst. Biol. 64, 851–869. Funk, D.J., Omland, K.E., 2003. Species-level paraphyly and polyphyly: frequency, Blaxter, M.L., 2004. The promise of a DNA taxonomy. Philos. Trans. Roy. Soc. B 359, causes, and consequences, with insights from animal mitochondrial DNA. Ann. 669–679. Rev. Ecol. Evol. Syst. 34, 397–423. Bribiesca-Contreras, G., Solis-Marin, F.A., Laguarda-Figueras, A., Zaldívar-Riverón, A., Gauld, I.D., 1973. Notes on the British Ophionini (Hym., Ichneumonidae) including a 2013. Identification of echinoderms (Echinodermata) from an anchialine cave in provisional key to species. Entomol. Gaz. 24, 55–65. Cozumel Island, Mexico, using DNA barcodes. Mol. Ecol. Res. 13, 1137–1145. Gauld, I.D., 1976. Notes on the British Ophionini (Hym., Ichneumonidae). Part 3. The http://dx.doi.org/10.111/1755-0998.12098. identity of the species described by Morley, 1915 and Thomson, 1888 and a Brock, J.P., 1982. A systematic study of the genus Ophion in Britain (Hymenoptera, checklist of British species. Entomol. Gaz. 27, 113–117. Ichneumonidae). Tijdschr. Entomol. 125, 57–96. Gauld, I.D., 1978. Notes on the British Ophioninae (Hym., Ichneumonidae). Part 4. A Brooks, D.R., Hoberg, E.P., 2000. Triage for the biosphere: the need and rationale for revised key to the species of the genus Ophion Fabricius. Entomol. Gaz. 29, 145– taxonomic inventories and phylogenetic studies of parasites. J. Helm. Soc. Wash. 149. 67, 1–25. Gauld, I.D., 1980. An analysis of the classification of the Ophion genus-group Campbell, B.C., Steffen-Campbell, J.D., Werren, J.H., 1993. Phylogeny of the Nasonia (Ichneumonidae). Syst. Entomol. 5, 59–82. species complex (Hymenoptera: Pteromalidae) inferred from an internal Gauld, I.D., 1985. The phylogeny, classification and evolution of parasitic wasps of M.D. Schwarzfeld, F.A.H. Sperling / Molecular Phylogenetics and Evolution 93 (2015) 234–248 247

the subfamily Ophioninae (Ichneumonidae). Bull. Br. Mus. (Nat. Hist.) Entomol. Lim, G.S., Balke, M., Meier, R., 2012. Determining species boundaries in a world full 51, 61–185. of rarity: singletons, species delimitation methods. Syst. Biol. 61, 165–169. Gauld, I.D., 1988. A survey of the Ophioninae (Hymenoptera: Ichneumonidae) of Lohse, K., 2009. Can mtDNA barcodes be used to delimit species? A response to Pon tropical Mesoamerica with special reference to the fauna of Costa Rica. Bull. Br. et al. 2006. Syst. Biol. 58, 439–442. Mus. (Nat. Hist.) Entomol. 57, 1–309. Lopez, J.V., Yuhki, N., Masuda, R., Modi, W., O’Brien, S.J., 1994. Numt, a recent Gellissen, G., Bradfield, J.Y., White, B.N., Wyatt, G.R., 1983. Mitochondrial DNA transfer and tandem amplification of mitochondrial DNA to the nuclear genome sequences in the nuclear genome of a locust. Nature 301, 631–634. of the domestic cat. J. Mol. Evol. 39, 174–190. Gernhard, T., 2008. The conditioned reconstructed process. J. Theor. Biol. 253, 769– Lumley, L.M., Sperling, F.A.H., 2010. Integrating morphology and mitochondrial 778. DNA for species delimitation within the spruce budworm (Choristoneura Glez-Peña, D., Gómez-Blanco, D., Reboiro-Jato, M., Fdez-Riverola, F., Posada, D., fumiferana) cryptic species complex (Lepidoptera: Tortricidae). Syst. Entomol. 2010. ALTER: program-oriented format conversion of DNA and protein 35, 416–428. alignments. Nucl. Acids Res. Web Server issue. ISSN: 0305–1048. http:// Lumley, L.M., Sperling, F.A.H., 2011. Utility of microsatellites and mitochondrial dx.doi.org/10.1093/nar/gkq321. DNA for species delimitation in the spruce budworm (Choristoneura fumiferana) Godfray, H.C.J., 2002. Challenges for taxonomy. Nature 417, 17–19. species complex (Lepidoptera: Tortricidae). Mol. Phylogenet. Evol. 58, 232–243. Gomez-Zurita, J., Juan, C., Petitpierre, E., 2000. Sequence, secondary structure and Maddison, W.P., Maddison, D.R., 2011. Mesquite: a modular system for evolutionary phylogenetic analyses of the ribosomal internal transcribed spacer 2 (ITS2) in analysis. Version 2.75, . the Timarcha leaf beetles (Coleoptera: Chrysomelidae). Insect Mol. Biol. 9, 591– Meier, R., Shiyang, K., Vaidya, G., Ng, P., 2006. DNA barcoding and taxonomy in 604. Diptera: a tale of high intraspecific variability and low identification success. Grummer, J.A., Bryson, R.W., Reeder, T.W., 2014. Species delimitation using Bayes Syst. Biol. 55, 715–728. factors: simulations and application to the Sceloporus scalaris species group Meyer, C.P., Paulay, G., 2005. DNA barcoding: error rates based on comprehensive (Squamata: Phrynosomatidae). Syst. Biol. 63, 119–133. sampling. PLoS Biol. 3, 2229–2238. Gutell, R.R., Larsen, H., Woese, C.R., 1994. Lessons from an evolving rRNA: 16S and Miller, M.A., Pfeiffer, W., Schwartz, T., 2010. Creating the CIPRES Science Gateway 23S rRNA structures from a comparative perspective. Microbiol. Rev. 58, 10–26. for inference of large phylogenetic trees. Proceedings of the Gateway Hajibabaei, M., Janzen, D.H., Burns, J.M., Hallwachs, W., Hebert, P.D.N., 2006. DNA Computing Environments Workshop (GCE), 14 Nov. 2010, New Orleans, LA, barcodes distinguish species of tropical Lepidoptera. Proc. Nat. Acad. Sci. 103, pp. 1–8. 968–971. Miller, S.E., 2007. DNA barcoding and the renaissance of taxonomy. Proc. Nat. Acad. Hamilton, C.A., Hendrixson, B.E., Brewer, M.S., Bond, J.E., 2014. An evaluation of Sci. 104, 4775–4776. sampling effects on multiple DNA barcoding methods leads to an integrative Miralles, A., Vences, M., 2013. New metrics for comparison of taxonomies reveal approach for delimiting species: a case study of the North American tarantula striking discrepancies among species delimitation methods in Madascincus genus Aphonopelma (Araneae, Mygalomorphae, Theraphosidae). Mol. Phylog. lizards. PLoS ONE 8, e68242. http://dx.doi.org/10.1371/journal.pone.0068242. Evol. 71, 79–93. Modica, M.V., Puillandre, N., Castelin, M., Zhang, Y., Holford, M., 2014. A good Hebert, P.D.N., Cywinska, A., Ball, S.L., deWaard, J.R., 2003a. Biological identifications compromise: rapid and robust species proxies for inventorying biodiversity through DNA barcodes. Proc. Roy. Soc. Lon. B: Biol. Sci. 270, S96–S99. hotspots using the Terebridae (Gastropoda: Conoidea). PLoS ONE 9 (7), Hebert, P.D.N., Ratnasingham, S., deWaard, J.R., 2003b. Barcoding animal life: e102160. http://dx.doi.org/10.1371/journal.pone.0102160. cytochrome c oxidase subunit 1 divergences among closely related species. Monaghan, M.T., Balke, M., Pons, J., Vogler, A.P., 2006. Beyond barcodes: complex Proc. Roy. Soc. Lon. B (Suppl.) 270, S96–S99. DNA taxonomy of a South Pacific Island radiation. Proc. Roy. Soc. B 273, 887– Hebert, P.D.N., Stoeckle, M.Y., Zemlak, T.S., Francis, C.M., 2004. Identification of birds 893. through DNA barcodes. PLoS Biol. 2, e312. Monaghan, M.T., Wild, R., Elliot, M., Balke, M., Inward, D.J., Lees, D.C., Ranaivosolo, Hendrich, L., Pons, J., Ribera, I., Balke, M., 2010. Mitochondrial Cox1 sequence data R., Eggleton, P., Barraclough, T.G., Vogler, A.P., 2009. Accelerated species reliably uncover patterns of insect diversity but suffer from high lineage- inventory on Madagascar using coalescent-based models of species idiosyncratic error rates. PLoS ONE 5, e14448. delineation. Syst. Biol. 58, 298–311. Hendrixson, B.E., DeRussy, B.M., Hamilton, C.A., Bond, J.E., 2013. An exploration of Monti, M.M., Nappo, A.G., Giorgini, M., 2005. Molecular characterization of closely species boundaries in turret-building tarantulas of the Mojave Desert (Araneae, related species in the parasitic genus Encarsia (Hymenoptera: Aphelinidae) Mygalomorphae, Theraphosidae, Aphonopelma). Mol. Phylogenet. Evol. 66, 327– based on the mitochondrial cytochrome oxidase I gene. Bull. Entomol. Res. 95, 340. 401–408. Hillis, D.M., Davis, S.K., 1988. Ribosomal DNA: intraspecific polymorphism, Moritz, C., Cicero, C., 2004. DNA barcoding: promise and pitfalls. PLoS Biol. 2, e354. concerted evolution, and phylogeny reconstruction. Syst. Zool. 37, 63–66. Müller, T., Philippi, N., Dandekar, T., Schultz, J., Wolf, M., 2007. Distinguishing Hofacker, I.L., 2003. Vienna RNA secondary structure server. Nucleic Acids Res. 31, species. RNA 13, 1469–1472. 3429–3431. Navajas, M., Gutierrez, J., Bonato, O., Bolland, H.R., Mapangou-Divassa, S., 1994. Hung, Y.-T., Chen, C.A., Wu, W.-J., Lin, C.-C., Shih, C.-J., 2004. Phylogenetic utility of Intraspecific diversity of the Cassava Green Mite Mononychellus progresivus the ribosomal internal transcribed spacer 2 in Strumigenys spp. (Hymenoptera: (Acari: Tetranychidae) using comparisons of mitochondrial and nuclear Formicidae). Mol. Phylogenet. Evol. 32, 407–415. ribosomal DNA sequences and cross-breeding. Exp. Appl. Acarol. 18, 351–360. Ji, Y.-J., Zhang, D.-X., He, L.-J., 2003. Evolutionary conservation and versatility of a http://dx.doi.org/10.1007/BF00116316. new set of primers for amplifying the ribosomal internal transcribed spacer Nee, S., Holmes, E.C., May, R.M., Harvey, P.H., 1994. Extinction rates can be regions in insects and other invertebrates. Mol. Ecol. Notes 3, estimated from molecular phylogenies. Philos. Trans. Roy. Soc. Lond. B. Biol. Sci. 581–585. 344, 77–82. Jones, G., Aydin, Z., Oxelman, B., 2015. DISSECT: an assignment-free Bayesian O’Meara, B.C., 2010. New heuristic methods for joining species delimitation and discovery method for species delimitation under the multispecies coalescent. species tree inference. Syst. Biol. 59, 59–73. Bioinformatics 31, 991–998. Papadopoulou, A., Monaghan, M.T., Barraclough, T.G., Vogler, A.P., 2009. Sampling Jones, M. Ghoorah, A., Blaxter, M., 2010. jMOTU User Guide. Institute of error does not invalidate the Yule-coalescent model for species delimitation. A Evolutionary Biology, University of Edinburgh, 10 May 2010, Version 1.0. response to Lohse (2009). Syst. Biol. 58, 442–444. Jones, M., Ghoorah, A., Blaxter, M., 2011. JMOTU and Taxonerator: turning DNA Paz, A., Crawford, A.J., 2012. Molecular-based rapid inventories of sympatric barcode sequences into annotated operational taxonomic units. PLoS ONE 6, diversity: a comparison of DNA barcode clustering methods applied to e19259. http://dx.doi.org/10.1371/journal.pone.0019259. geography-based vs clade-based sampling of amphibians. J. Biosci. 37, 887–896. Kekkonen, M., Hebert, P.D.N., 2014. DNA barcode-based delineation of putative Pons, J., Barraclough, T.G., Gómez-Zurita, J., Cardoso, A., Duran, D.P., Hazell, S., species: efficient start for taxonomic workflows. Mol. Ecol. Res. 14, Kamoun, S., Sumlin, W.D., Vogler, A.P., 2006. Sequence-based species 706–715. delimitation for the DNA taxonomy of undescribed insects. Syst. Biol. 55, Keller, I., Chintauan-Marquier, I.C., Veltsos, P., Nichols, R.A., 2006. Ribosomal DNA in 595–609. the grasshopper Podisma pedestris: escape from concerted evolution. Genetics Puckridge, M., Andreakis, N., Appleyard, S.A., Ward, R.D., 2013. Cryptic diversity in 174, 863–874. flathead fishes (Sorpaeniformes: Platycephalidae) across the Indo-West Pacific Klopfstein, S., 2014. Revision of the western Palearctic Diplazontinae uncovered by DNA barcoding. Mol. Ecol. Res. 13, 32–42. (Hymenoptera, Ichneumonidae). Zootaxa 3801, 1–143. Puillandre, N., Lambert, A., Brouillet, S., Achaz, G., 2012a. ABGD, Automatic Barcode Koetschan, C., Förster, F., Keller, A., Schleicher, T., Ruderisch, B., Schwarz, R., Müller, Gap Discovery for primary species delimitation. Mol. Ecol. 21, 1864–1877. T., Wolf, M., Schultz, J., 2010. The ITS2 Database III: sequences and structures for Puillandre, N., Modica, M.V., Zhang, Y., Sirovich, L., Boisselier, M.-C., Cruaud, C., phylogeny. Nucl. Acids Res. 38, D275–D279. Holford, M., Samadi, S., 2012b. Large-scale species delimitation method for Laetsch, D.R., Heitlinger, E.G., Taraschewski, H., Nadler, S.A., Blaxter, M.L., 2012. The hyperdiverse groups. Mol. Ecol. 21, 2671–2691. phylogenetics of Anguillicolidae (Nematoda: Anguillicoloidea), swimbladder Quicke, D.L.J., 2012. We know too little about parasitoid wasp distributions to draw parasites of eels. BMC Evol. Biol. 12, 60. any conclusions about latitudinal trends in species richness, body size and Lanfear, R., Calcott, B., Ho, S.Y.W., Guindon, S., 2012. PartitionFinder: combined biology. PLoS ONE 7, e32101. http://dx.doi.org/10.1371/journalpone.0032101. selection of partitioning schemes and substitution models for phylogenetic Quicke, D.L.J., Mori, M., Zaldívar-Riverón, A., Laurenne, N.M., Shaw, M.R., 2006. analyses. Mol. Biol. Evol. 29, 1695–1701. Suspended mummies in Aleiodes species (Hymenoptera: Braconidae: Leaché, A.D., Fujita, M.K., 2010. Bayesian species delimitation in West African forest Rogadinae) with descriptions of six new species from western Uganda based geckos (Hemidactylus fasciatus). Proc. Roy. Soc. B 277, 3071–3077. largely on DNA sequence data. J. Nat. Hist. 40, 2663–2680. http://dx.doi.org/ Li, Y., Zhou, X., Feng, G., Hu, H., Niu, L., Hebert, P.D.N., Huang, D., 2010. COI and ITS2 10.1080/00222930601121288. sequences delimit species, reveal cryptic taxa and host specificity of fig- R Core Development Team, 2013. R: a language and environment for statistical associated Sycophila (Hymenoptera, Eurytomidae). Mol. Ecol. Res. 10, 31–40. computing. R Foundation for Statistical Computing, Vienna, Austria. 248 M.D. Schwarzfeld, F.A.H. Sperling / Molecular Phylogenetics and Evolution 93 (2015) 234–248

Rambaut, A., Drummond, A.J. 2009. Tracer, v. 1.5. Available from . GMYC model: insights from a butterfly survey. Methods Ecol. Evol. 4, 1101– Rannala, B., Yang, Z., 2013. Improved reversible jump algorithms for Bayesian 1110. species delimitation. Genetics 194, 245–253. Tamura, K., Peterson, D., Peterson, N., Stecher, G., Nei, M., Kumar, S., 2011. MEGA5: Raychoudhury, R., Grillenberger, B.K., Gadau, J., Bijlsma, R., van de Zande, L., Werren, Molecular Evolutionary Genetics Analysis using maximum likelihood, H., Beukeboom, L.W., 2010. Phylogeography of Nasonia vitripennis evolutionary distance, and maximum parsimony method. Mol. Biol. Evol. 28, (Hymenoptera) indicates a mitochondrial Wolbachia sweep in North America. 2731–2739. Heredity 104, 318–326. Tang, C.Q., Humphreys, A.M., Fontaneto, D., Barraclough, T.G., 2014. Effects of Reid, N.M., Carstens, B.C., 2012. Phylogenetic estimation error can decrease phylogenetic reconstruction method on the robustness of species delimitation accuracy of species delimitation: a Bayesian implementation of the general using single-locus data. Methods Ecol. Evol. 5, 1086–1094. mixed Yule-coalescent model. BMC Evol. Biol. 12, 196. Tang, C.Q., Leasi, F., Obertegger, U., Kieneke, A., Barraclough, T.G., Fontaneto, D., Roe, A.D., Sperling, F.A.H., 2007. Population structure and species boundary 2012. The widely used small subunit 18S rDNA molecule greatly delimitation of cryptic Dioryctria moths: an integrative approach. Mol. Ecol. underestimates true diversity in biodiversity surveys of the meiofauna. Proc. 16, 3617–3633. Nat. Acad. Sci. 109, 16208–19212. RStudio, 2012. RStudio: integrated development environment for R, Version Tautz, D., Arctander, P., Minelli, A., Thomas, R.H., Vogler, A.P., 2003. A plea for DNA 0.97.332. Boston, MA, Available at . taxonomy. Trends Ecol. Evol. 18, 70–74. Rubinoff, D., Cameron, S., Will, K., 2006. A genomics perspective on the Van Veen, F.J., Belshaw, R., Godfray, H.C., 2003. The value of the ITS2 region for the shortcomings of mitochondrial DNA for ‘‘barcoding” identifications. J. Hered. identification of species boundaries between Alloxysta hyperparasitoids 97, 581–594. (Hymenoptera: Charipidae) of aphids. Eur. J. Entomol. 100, 449–453. Ruhl, M.W., Wolf, M., Jenkins, T., 2010. Compensatory base changes illuminate Wagener, B., Reineke, A., Lohr, B., Zebitz, C.P.W., 2006. Phylogenetic study of morphologically difficult taxonomy. Molec. Phylogenet. Evol. 54, 664–669. Diadegma species (Hymenoptera: Ichneumonidae) inferred from analysis of Schlick-Steiner, B.C., Steiner, F.M., Seifert, B., Stauffer, C., Christian, E., Crozier, R.H., mitochondrial and nuclear DNA sequences. Biol. Control 37, 131–140. 2010. Integrative taxonomy: a multisource approach to exploring biodiversity. Weigand, A.M., Jochum, A., Pfenninger, M., Steinke, D., Klussman-Kolb, A., 2011. A Ann. Rev. Entomol. 55, 421–438. new approach to an old conundrum – DNA barcoding sheds new light on Schmidt, B.C., Sperling, F.A.H., 2008. Widespread decoupling of mtDNA variation phenotypic plasticity and morphological stasis in microsnails (Gastropoda, and species integrity in Grammia tiger moths (Lepidoptera: Noctuidae). Syst. Pulmonata, Carychiidae). Mol. Ecol. Res. 11, 255–265. http://dx.doi.org/ Entomol. 33, 613–634. 10.1111/j.1755-0998.2010.02937.x. Schulmeister, S., 2003. Simultaneous analysis of basal Hymenoptera (Insecta): Weigand, A.M., Jochum, A., Slapnik, R., Schnitzler, J., Zarza, E., Klussmann-Kolb, A., introducing robust-choice sensitivity analysis. Biol. J. Linn. Soc. 79, 245–275. 2013. Evolution of microgastropods (Ellobioidea, Carychiidae): integrating http://dx.doi.org/10.1046/j.1095-8312.2003.00233.x. phylogenetic and evolutionary hypotheses. BMC Evol. Biol. 13, 18. Schultz, J., Müller, T., Achtziger, M., Seibel, P.N., Dandekar, T., Wolf, M., 2006. The Wheeler, Q.D., 2004. Taxonomic triage and the poverty of phylogeny. Philos. Trans. internal transcribed spacer 2 database: a web server for (not only) low level Roy. Soc. Lon. B 359, 571–583. phylogenetic analyses. Nucl. Acids Res. 34, W704–W707. Wiemers, M., Fiedler, K., 2007. Does the DNA barcoding gap exist? – a case study in Schwarzfeld, M.D., Sperling, F.A.H., 2014. Species delimitation using morphology, blue butterflies (Lepidoptera: Lycaenidae). Front. Zool. 4, 8. http://dx.doi.org/ morphometrics, and molecules: definition of the Ophion scutellaris Thomson 10.1186/1742-9994-4-8. species group, with descriptions of six new species (Hymenoptera, Wiemers, M., Keller, A., Wolf, M., 2009. ITS2 secondary structure improves Ichneumonidae). ZooKeys 462, 59–114. phylogeny estimation in a radiation of blue butterflies of the subgenus Seibel, P.N., Müller, T., Dandekar, T., Schultz, J., Wolf, M., 2006. 4Sale – a tool for Agrodiaetus (Lepidoptera: Lycaenidae: Polyommatus). BMC Evol. Biol. 9, 300. synchronous RNA sequence and secondary structure alignment and editing. http://dx.doi.org/10.1186/1471-2148-9-300. BMC Bioinform. 7, 498. Wild, A.L., Maddison, D.R., 2008. Evaluating nuclear protein-coding genes for Selig, C., Wolf, M., Müller, T., Dandekar, T., Schultz, J., 2008. The ITS2 Database II: phylogenetic utility in beetles. Mol. Phylogenet. Evol. 48, 877–891. homology modelling RNA structure for molecular systematics. Nucl. Acids Res. Will, K.W., Mishler, B.D., Wheeler, Q.D., 2005. The perils of DNA barcoding and the 36, D377–D380. need for integrative taxonony. Syst. Biol. 54, 844–851. Shaffer, H.B., Thomson, R.C., 2007. Delimiting species in recent radiations. Syst. Biol. Will, K.W., Rubinoff, D., 2004. Myth of the molecule: DNA barcodes for species 56, 896–906. cannot replace morphology for identification and classification. Cladistics 20, Sharanowski, B.J., Dowling, A.P.G., Sharkey, M.J., 2011. Molecular phylogenetics of 47–55. Braconidae (Hymenoptera: Ichneumonoidea), based on multiple nuclear genes, Williams, P.H., Brown, M.J.F., Carolan, J.C., An, J., Goulson, D., Aytekin, A.M., Best, L.R., and implications for classification. Syst. Entomol. 36, 549–572. Byvaltsev, A.M., Cederberg, B., Dawson, R., Huang, J., Ito, M., Monfared, A., Raina, Sites Jr., J.W., Marshall, J.C., 2004. Operational criteria for delimiting species. Ann. R.H., Schmid-Hempel, P., Sheffield, C.S., Šima, P., Xie, Z., 2012. Unveiling cryptic Rev. Ecol. Evol. Syst. 35, 199–227. species of the bumblebee subgenus Bombus s. str. worldwide with COI barcodes Smith, M.A., Eveleigh, E.S., McCann, K.S., Merilo, M.T., McCarthy, P.C., et al., 2011. (Hymenoptera: Apidae). Syst. Biodivers. 10, 21–56. Barcoding a quantified food web: crypsis, concepts, ecology and hypotheses. Wolf, M., Chen, S., Song, J., Ankenbrand, M., Muller, T., 2013. Compensatory base PLoS ONE 6 (7), e14424. http://dx.doi.org/10.1371/journal.pone.0014424. changes in ITS2 secondary structures correlate with the biological species Smith, M.A., Fernández-Triana, J.L., Eveleigh, E., Gómez, J., Guclu, C., Hallwachs, W., concept despite intragenomic variability in ITS2 sequences – a proof of concept. Hebert, P.D.N., Hrcek, J., Huber, J.T., Janzen, D., Mason, P.G., Miller, S., Quicke, D. PLoS ONE 8, e66726. http://dx.doi.org/10.1371/journal.pone/0066726. L.J., Rodriguez, J.J., Rougerie, R., Shaw, M.R., Várkonyi, G., Ward, D.F., Whitfield, J. Wolf, M., Friedrich, J., Dandekar, T., Müller, T., 2005. CBCAnalyzer: inferring B., Zaldívar-Riverón, A., 2013. DNA barcoding and the taxonomy of phylogenies based on compensatory base changes in RNA secondary Microgastrinae wasps (Hymenoptera, Braconidae): impacts after 8 years and structures. In Silico Biol. 5, 291–294. nearly 20000 sequences. Mol. Ecol. Res. 13, 168–176. Yang, Z., Rannala, B., 2010. Bayesian species delimitation suing multilocus sequence Smith, M.A., Fernandez-Triana, J., Roughley, R., Hebert, P.D.N., 2009. DNA barcode data. Proc. Natl. Acad. Sci. USA 107, 9264–9269. accumulation curves for understudied taxa and areas. Mol. Ecol. Res. 9, 208– Yang, Z., Rannala, B., 2014. Unguided species-delimitation using DNA sequence data 216. from multiple loci. Mol. Boil. Evol. 31, 3125–3135. Song, H., Buhay, J.E., Whiting, M.F., Crandall, K.A., 2008. Many species in one: Young, M.R., Behan-Pelletier, V.M., Hebert, P.D.N., 2012. Revealing the hyperdiverse DNA barcoding overestimates the number of species when nuclear mite fauna of subarctic Canada through DNA barcoding. PLoS ONE 7, e48755. mitochondrial pseudogenes are coamplified. Proc. Nat. Acad. Sci. USA 105, Yu, D.S.K., van Achterberg, C., Horstmann, K., 2012. Taxapad 2012, Ichneumonoidea 13486–13491. 2011. Database on USB. Ottawa, ON, Canada, . Stalhut, J.K., Fernandez-Triana, J., Adamowicz, S.J., Buck, M., Goulet, H., Hebert, P.D. Zaldívar-Riverón, A., Martínez, J.J., Ceccarelli, F.S., de Jesús-Bonilla, V.S., Rodríguez- N., Huber, J.T., Merilo, M.T., Sheffield, C.S., Woodcock, T., Smith, M.A., 2013. DNA Pérez, A.C., Reséndiz-Flores, A., Smith, M.A., 2010. DNA barcoding a highly barcoding reveals diversity of Hymenoptera and the dominance of parasitoids diverse group of parasitoid wasps (Braconidae: Doryctinae) from a Mexican in sub-arctic environment. BMC Ecol. 13, 2. nature reserve. Mitoch. DNA 21, 18–23. Strutzenberger, P., Brehm, G., Fiedler, K., 2011. DNA barcoding-based species Zhang, J., Kapli, P., Pavlidis, P., Stamatakis, A., 2013. A general species delimitation delimitation increases species count of Eois (Geometridae) moths in a well- method with applications to phylogenetic placements. Bioinformatics 29, studied tropical mountain forest by up to 50%. Insect Sci. 18, 349–362. 2869–2876.