Molecular Phylogenetics and Evolution 59 (2011) 251–262

Contents lists available at ScienceDirect

Molecular Phylogenetics and Evolution

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

Deep mtDNA subdivision within Linnean species in an endemic radiationof tiger beetles from New Zealand (genus Neocicindela)

Joan Pons a,b, Tomochika Fujisawa a,c, Elin M. Claridge a,1, R. Anthony Savill d, Timothy G. Barraclough c,Alfried P. Vogler a,c,⇑a Department of Entomology, Natural History Museum, Cromwell Road, London SW7 5BD, UKb IMEDEA (CSIC-UIB), Miquel Marqués, 21 Esporles, 07190 Illes Balears, Spainc Division of Biology, Imperial College London, Silwood Park Campus, Ascot SL5 7PY, UKd Canterbury Museum, Rolleston Avenue, Christchurch, New Zealand

a r t i c l e i n f o a b s t r a c t

Article history:Received 19 January 2010Revised 15 November 2010Accepted 14 February 2011Available online 19 February 2011

Keywords:BiogeographyNew ZealandInvertebratesSpecies delimitationGMYC groupsDNA taxonomy

1055-7903/$ - see front matter � 2011 Elsevier Inc. Adoi:10.1016/j.ympev.2011.02.013

⇑ Corresponding author at: Department of EntomoloCromwell Road, London SW7 5BD, UK.

E-mail address: apv@nhm.ac.uk (A.P. Vogler).1 Present address: University of California Berkel

Moorea, French Polynesia, United States.

The invertebrate fauna of New Zealand is of great interest as a geologically tractable model for the studyof species diversification, but direct comparisons with closely related lineages elsewhere are lacking.Integrating population-level analyses with studies of taxonomy and clade diversification, we performedmtDNA analysis on Neocicindela (Cicindelidae, tiger beetles) for a broad sample of populations from 11 of12 known species and 161 specimens (three loci, 1883 nucleotides), revealing 123 distinct haplotypes.Phylogenetic reconstruction recovered two main lineages, each composed of 5–6 Linnean species whoseorigin was dated to 6.66 and 7.26 Mya, while the Neocicindela stem group was placed at 10.82 ± 0.48 Mya.Species delimitation implementing a character-based (diagnostic) species concept recognized 19 species-level groups that were in general agreement with Linnean species but split some of these into mostlyallopatric subgroups. Tree-based methods of species delimitation using a mixed Yule-coalescence modelwere inconclusive, and recognized 32–51 entities (including singletons), splitting existing species into upto 8 partially sympatric groups. These findings were different from patterns in the Australian sister genusRivacindela, where character-based and tree-based methods were previously shown to produce highlycongruent groupings. In Neocicindela, the pattern of mtDNA variation was characterized by high intra-population and intra-species haplotype divergence, the coexistence of divergent haplotypes in sympatry,and a poor correlation of genetic and geographic distance. These observations combined suggest a sce-nario of phylogeographic divergence and secondary contact driven by orogenetic and climatic changesof the Pleistocene/Pliocene. The complex evolutionary history of most species of Neocicindela due tothe relative instability of the New Zealand biota resulted in populations of mixed ancestry but not in ageneral loss of genetic variation.

� 2011 Elsevier Inc. All rights reserved.

1. Introduction

The invertebrate fauna of New Zealand has recently beensubjected to numerous studies of historical biogeography and phy-logeography (Wallis and Trewick, 2009). This surge of interest re-sulted in a better understanding of the origin of extant lineagesand the role of paleoclimatic and geological changes in the diversi-fication process. New Zealand is geologically derived from a frag-ment of continental crust that was part of the Gondwanansupercontinent, i.e. is expected to harbour an ancient fauna and

ll rights reserved.

gy, Natural History Museum,

ey, Gump Research Station,

flora, but instead most groups show patterns of species composi-tion and radiation usually attributed to young oceanic islands. Thismay be explained by the geographical isolation and extensive sub-mergence of New Zealand in the Oligocene (Landis et al., 2008)which caused the extirpation of most lineages. Phylogenetic anal-yses now have established that most endemic invertebrate lin-eages are derived from post-Oligocene colonizations (Goldberget al., 2008). However, many questions remain about the processesof diversification within New Zealand (Wallis and Trewick, 2009).Recent work addressed the closely linked issues of populationdivergence and geographic distributions which have been influ-enced by the complex orogeny (King, 2000; Landis et al., 2008),as well as Pleistocene glacial cycles (Buckley and Simon, 2007;Emerson and Wallis, 1995). Surveys of mtDNA variation now existfor a large number of species or genera, in particular for arthro-pods. These have attempted to infer the location of glacial refugia

252 J. Pons et al. / Molecular Phylogenetics and Evolution 59 (2011) 251–262

(Buckley et al., 2009; Marshall et al., 2009; Marske et al., 2009;McCulloch et al. 2010), the environmental constraints to dispersal(Leschen et al., 2008), the role of hybridization in species separa-tion (Buckley et al., 2006) or the taxonomic status and biogeo-graphic history of various insects (Buckley et al., 2006; Emersonand Wallis, 1995; Trewick, 2008) and arachnids (Boyer et al., 2007).

The emerging picture reveals great evolutionary complexity, asthese studies have commonly found incongruence with the namedLinnean species while genetic intra-specific variation was veryhigh. These studies also suggest very strong geographic structureof haplotype variation, while multiple distant haplogroups as-signed to the same Linnean species frequently co-occur at a singlelocality. The findings led some authors to query the utility ofmtDNA markers in DNA barcoding and DNA taxonomy generally(Boyer et al., 2007; Trewick, 2008). While complicated mtDNAphylogeographic patterns have been observed elsewhere, thesecases seem to be more frequent and more extreme among NewZealand groups.

Here, we add to this recent body of work with a phylogeograph-ic investigation of tiger beetles in the genus Neocicindela, whichform a well-defined endemic lineage with 12 named species.Neocicindela is fairly well known with regard to taxonomy, geo-graphic distribution and ecology (Rivalier, 1961; Savill, 1999),and so is an interesting group for studying diversification in NewZealand. Furthermore, Neocicindela has a well established sistergroup in the Australian genus Rivacindela (Pons et al., 2004), forwhich extensive DNA taxonomic and phylogeographic studies havebeen carried out (Pons et al., 2006). The latter study was instru-mental in establishing the General Mixed Yule Coalescent (GMYC)model for sequenced-based species delimitation separating speci-ation and coalescent processes (Pons et al., 2006). The methodcan be applied universally to mtDNA data for comparisons acrosstaxa and biogeographic regions.

Neocicindela is part of the subtribe Cicindelina, which is some-times referred to as genus Cicindela (sensu lato), a cosmopolitangroup that includes some 1000 species (Pearson, 1988; Pearsonand Cassola, 1992) and 50 genera and subgenera (Rivalier, 1961).As generalist predators of small arthropods, species differ in theirassociation with particular open-vegetation habitat types, such assand dunes, beaches, river banks and alpine meadows. Analysesof molecular clocks across the major lineages of Cicindelina placethe origin of Neocicindela to the late Miocene at 9.7–6.6 Mya atthe latest, based on the estimated age of the node that marks thesplit of Neocicindela from Rivacindela (estimated under limitedsampling of two species from each genus). Both lineages combinedare part of a larger clade of Australasian lineages whose age hasbeen dated to the mid-Miocene (15.9–12.7 Mya) (Pons et al.,2004). This places the split of Neocicindela from other cicindelidswell after the separation of New Zealand from Gondwana datedat c. 80 Mya (Landis et al., 2008; Trewick et al., 2007) and confirmsthat their distribution is not attributable to ancient vicariance, fol-lowing the general trends of New Zealand biota (Goldberg et al.,2008; Wallis and Trewick, 2009).

The Linnean taxonomy of cicindelines frequently recognizesgeographically confined subspecies separated by morphologicaldifferences in color patterns of the elytra and other body parts,likely in response selection pressure for background matching ofthe soil (Acorn, 1992). These differences have been shown to coin-cide with deep phylogenetic splits in mtDNA and other markers insome cases, for example in the separation of subspecies of theNorth American coastal Cicindela dorsalis on either side of theFlorida Peninsula (Vogler and DeSalle, 1993) and the sand duneassociated Cicindela limbata exhibiting distinct subgroups in Alaskaand Nova Scotia (Knisley et al., 2008). Likewise, studies on theEuropean Cicindela hybrida showed high levels of differentiationin mtDNA among local groups, although these mtDNA groups were

not always congruent with morphological differences and a nucle-ar marker (Cardoso et al., 2009). In Neocicindela, local differentia-tion is suggested by the recognition of geographically definedsubspecies in the coastal Neocicindela perhispida from North Islandthat has been subdivided based on elytral color pattern from thedark N. p. campbelli to the intermediate N. p. perhispida and the paleN. p. giveni. These forms reflect beach substrate grades from blackiron-sand in the south to pure white quartz in the north (Hadleyet al., 1988) and constitute a convincing case to support thehypothesized role of selection on body color variation for back-ground matching of the soil (Acorn, 1992). However, it is unclearto what degree these differences reflect historical subdivision alsodetectable in mtDNA markers.

The aim of this study is to establish a DNA-based framework forthe taxonomy and evolutionary analysis of Neocicindela, samplingmost species at multiple sites throughout their geographic ranges.Given the intra-specific variation and geographic structure fre-quently seen in the New Zealand fauna, we specifically focus onthe question of how to delimit species-level entities. We employboth ‘tree-based’ and ‘character-based’ (diagnostic) species con-cepts using quantitative methods (Vogler and Monaghan, 2007).These analyses on New Zealand lineages can be compared to theexisting work on the Australian Rivacindela (Pons et al., 2006),which may lead to the detection of region-specific evolutionaryprocesses.

2. Material and methods

2.1. Laboratory procedures and phylogenetic analysis

Eleven of the 12 named species of Neocicindela, one of whichseparated in three named subspecies, were included in this studyfor a total of 13 named taxa. These were collected at numeroussites across New Zealand (Fig. 1). Each taxon was represented bybetween 1 to 11 individuals. Specimens of Neocicindela spilleri werenot available due to its rarity. Range maps for all species of Neoci-cindela are well established (Savill, 1999) and demonstrate theexistence of several widespread species, including Neocicindelatuberculata, Neocicindela helmsi and Neocicindela parryi, while oth-ers are localized (Table 1; Fig. 1). Even the distribution of wide-spread species is very patchy on the local level due to the highhabitat specificity, but several well known collecting spots harbormultiple species either sympatrically or in close proximity. Theseplaces were sampled predominantly, with an attempt to covermultiple distant sites throughout the known species ranges. Multi-ple individuals were sampled from most of the sites, but supple-mented with a single specimen of each species collected on aseparate occasion (Table 1; Fig. 1).

DNA was extracted from hindlegs using the Qiagen DNeasyAnimal Tissue kit (Qiagen Inc., Valencia, CA, USA). Three regionsof the mitochondrial genome were analyzed, including 757 bp ofCytochrome Oxidase I (cox1), 357 bp of Cytochrome b (cob), and769 bp of 16S rRNA gene (rrnL) and adjacent tRNALeu (trnL2) andNADH dehydrogenase subunit 1 (nad1), referred to as the rrnL re-gion. Primer combinations and PCR conditions have been describedelsewhere: primers Pat and Jerry for cox1 (Simon et al., 1994),primers CB3 and CB4 for cob (Barraclough et al., 1999), and primers16Sa (Simon et al., 1994) and Alf1 (Vogler et al., 1993) for the rrnLregion. PCR fragments were sequenced in both directions on anABI377 automated sequencer. Australian members of the generaAbroscelis and Macfarlandia were used as outgroups (Pons et al.,2004). Newly generated sequences have been deposited in Gen-Bank under AF669501-FR669654 (cox1), FR670632-670773 (cob)and FR671200-FR671348 (rrnL region).

Fig. 1. Maps of collecting localities and known ranges of Neocicindela. The dots refer to locality records from Savill (1999). Collecting sites from the current study are labelledwith the site code given in Table 1 and the PAA group number from the population aggregation analysis.

J. Pons et al. / Molecular Phylogenetics and Evolution 59 (2011) 251–262 253

Parsimony analysis was conducted in PAUP⁄ 4.0b10 (Swofford,2002) under equal weighting of all characters and gaps coded asfifth character state with 1000 random addition replicates andkeeping no more than 10 trees in each random addition. Nonpara-metric bootstrapping was performed with 100 pseudoreplicates,each based on TBR searches with 50 random addition replicatesand keeping no more than 20 trees per replicate. Bayesian analyseswere conducted with MrBayes 3.04 (Huelsenbeck and Ronquist,2001) using the HYK + I + C model of sequence evolution selectedwith a hierarchical test of likelihood implemented in MODELTEST3.06 (Posada and Crandall, 1998). Bayesian analysis was conductedusing three heated and one cold Markov chains for 5,000,000 gen-erations, sampled at intervals of 1000 generations. Suitable burn-in(400,000 generations) and convergence of parameters were deter-mined using Tracer v1.3 (http://tree.bio.ed.ac.uk/software/tracer/).Trees (after burn-in) of two independent runs were combined in asingle majority consensus topology, and the frequencies of nodesin these trees were taken as a posteriori probabilities (Huelsenbeckand Ronquist, 2001). Convergence of posterior clade probabilities

were assessed with the software AWTY (Nylander et al., 2008) withthe cumulative and compare commands that displays the posteriorprobabilities of splits through generations in a single run or com-paring several runs. The statistical significance of alternative topol-ogies was assessed using the Shimodaira–Hasegawa test (SH;Goldman et al., 2000; Shimodaira and Hasegawa, 1999) in PAUP⁄4.0b10.

Age estimates of Neocicindela were made using the closestcicindelid outgroups in the genera Rivacindela, Abroscelis and Mac-farlandia, and fitting the age of this Australasian clade includingNeocicindela to 12.4 My, as estimated from a tree of global lineagesof Cicindela s. l. (=Cicindelina), for a rate of 3.34% and 0.76% diver-gence My�1 for cox1 and rrnL, respectively (Pons et al., 2004). Thiscalibration was initially obtained from dating of nodes in the phy-logeny of Cicindela from the New World, specifically those thatmark the disperisal of South American lineages into North Americaafter the closure of the Isthmus of Panama (Barraclough andVogler, 2002). The age estimated for the Neocicindela node wasthen used to calibrate the full clock-constrained tree generated

Table 1List of specimens and collecting data.

Species Distribution Habitat Code Locality District Latitude Longitude Date Indiv

N. austromontana SI, east Sandy river edge,alpine grassland

20 SI: Castle Hill Peak Canterbury �43.25 171.76 December 00 1

N. brevilunata NI, northeast Coastal beach, sanddunes

7 NI: Waipu Cove North Auckland �36.03 174.49 December 00 3

7B NI: Marsden Point North Auckland �35.84 174.50 December 00 393 NI: NE Lake Waipu Wellington �36.81 174.45 December 02 1

N. dunedensis SI, east Glacial loess 92 SI: Alexandra Otago �45.23 169.36 January 03 1N. feredayi SI, central; NI Sandy river edge 105 NI: Rangitikei River, NE

RataWellington �39.98 175.58 December 02 1

19 SI: Rangitata River Canterbury �43.65 170.95 December 00 1104 SI: 29 Km W Wairau

Valley, Wairau RiverMarlborough �41.63 173.21 January 03 1

N. hamiltoni SI, north-central Alpine grassland 21 SI: Porter HeightsSkifield

Canterbury �43.27 171.62 December 00 1

N. helmsi SI; NI central Sandy river edge 1A NI: Waiouru Wellington �39.46 175.68 December 00 32 NI: Waiouru Wellington �39.46 175.68 December 00 9

13 NI: Turoa Skifield Wellington �39.29 175.53 December 00 816 SI: Wairau River �

Wash StreamMarlborough �41.73 173.09 December 00 3

18 SI: Maruia River Nelson �42.25 172.22 December 00 1089 SI: Ikamatua Plain Nelson �42.28 171.66 January 03 1

N. latecincta SI, east Glacial loess 22 SI: Motukarara Canterbury �43.73 172.59 December 00 198 SI: 4 Km W Lake

ColeridgeCanterbury �43.38 171.56 January 03 1

103 SI: n/a Marlborough �41.88 174.11 January 03 1N. p. campbelli NI, central west coast Coastal beach, sand

dunes5 NI: Ngaruni Raglan

Harbour (Whaingaroa)South Auckland �37.79 174.89 December 00 6

90 NI: Port Waikato South Auckland �37.40 174.71 December 02 1N. p. giveni NI, northern east coast Coastal beach, sand

dunes8 NI: Rarawa Beach North Auckland �34.71 173.07 December 00 6

96 NI: Spirits Bay North Auckland �34.41 172.86 December 02 1N. p. perhispida NI, northern west coast Coastal beach, sand

dunes9 NI: Waipapakauri

BeachNorth Auckland �35.04 173.16 December 00 5

10 NI: Baylys Beach North Auckland �35.94 173.73 December 00 4101 NI: Mitimiti Beach North Auckland �35.43 173.28 December 02 1

N. parryi NI; SI Open forest floor,grassland, sandy riveredge

91 NI: SE Taupo Lake South Auckland �38.75 176.20 December 02 1

97 NI: 5.7 Km OakuraRiver, New Plymouth

Taranaki �39.15 173.91 December 02 1

4 NI: Karioi Wellington �39.44 175.50 December 00 912 NI: Turoa Skifield Wellington �39.29 175.53 December 00 1117 SI: Maruia River Nelson �42.25 172.22 December 00 11

N. sp. n/a 87 SI: 29 Km W WairauValley, Wairau River

Marlborough �41.63 173.21 January 03 1

N. tuberculata NI; SI north Grassland, coastalmudflats, beach

10B NI: Baylys Beach North Auckland �35.94 173.73 December 00 10

6 NI: Mangawhai Bay South Auckland �37.65 176.03 December 00 999 NI: 47 Km SE Taupo

LakeSouth Auckland �38.93 176.48 December 02 1

1 NI: Waiouru Wellington �39.46 175.68 December 00 688 NI: Beach NE Lake

WaipuWellington �36.81 174.66 December 02 1

15 SI: Wairau River -Wash Stream

Marlborough �41.73 173.09 December 00 6

N. waiouraensis NI central; SI north central Open grassland 3 NI: Waiouru Wellington �39.46 175.68 December 00 814 SI: Wairau River -

Wash StreamMarlborough �41.73 173.09 December 00 11

102 SI: 29 Km W WairauValley, Wairau River

Marlborough �41.65 173.21 January 03 1

254 J. Pons et al. / Molecular Phylogenetics and Evolution 59 (2011) 251–262

here. A strict clock was rejected (p < 0.01), hence we implementedrelaxed clock models that assume independent rates on differentbranches, i.e., there is no a priori correlation between a lineage’srate and that of its ancestor. Absolute node ages were estimatedby fitting Bayesian ML branch lengths using the semiparametricPenalized Likelihood (PL) in r8s 1.71 (Sanderson, 2002). The opti-mal smoothing value of 1 (Chi Square Error 1413.53) was favoredby cross-validation after testing four smoothing values (1, 10,100, and 1000). Sampling error due to stochastic rate changes thatmay affect branch length estimates and hence age estimates wasassessed estimating confidence intervals for node ages based onone hundred bootstrap replicates of the data and calculating

branch length on each of these in PAUP⁄, given the original treetopology and parameters estimated in the Bayesian analysis (Bald-win and Sanderson, 1998). Finally, branch lengths were made ultr-ametric using PL and the optimal smoothing value, and ageconfidence intervals estimated using the command profile in r8s.Node ages were estimated in BEAST 1.4.7 (Drummond and Ram-baut, 2007), enforcing a relaxed molecular clock with an uncorre-lated log–normal distribution and a Yule speciation model. Thedata were estimated using four independent partitions and mod-els: 1st codon sites (GTR + I + C), 2nd codon positions(HYK + I + C), 3rd codon sites (GTR + I + C) and finally structuralsites rrnL and trnL2 (HKY + I + C). The program optimized all

J. Pons et al. / Molecular Phylogenetics and Evolution 59 (2011) 251–262 255

parameters including topology except that we enforced the mono-phyly of Neocicindela. The BEAST analyses were run for 20 milliongenerations, sampling every 1000 generations, and the output wasanalyzed using Tracer 1.4 after discarding the first 2 million gener-ations. The most appropriate model for each partition was selectedwith a hierarchical test of likelihood implemented in MODELTEST3.06 (Posada and Crandall, 1998).

2.2. Analysis of species limits

Character-based species delimitation was implemented usingPopulation Aggregation Analysis (PAA) (Davis and Nixon, 1992).This method requires an a priori hypothesis of local populations,which here was based on the similarity of external appearance ofspecimens (i.e., assignment to a particular Linnean species) at a gi-ven locality. Populations defined in this way were then tested forthe presence of diagnostic (fixed) nucleotide changes unique to apopulation or set of populations separating them from all otherpopulations. DnaSP 4.10.9 (Rozas et al., 2003) was used to detectthese unique changes within a (sets of) populations. In addition,haplotype networks were created using statistical parsimony anal-ysis (Templeton, 2001), as implemented in TCS 1.3 (Clement et al.,2000). Separate haplotype networks are expected to correspondto single Linnean species or a few closely related species groups(Hart and Sunday, 2007; Pons et al., 2006). Therefore we also ap-plied PAA within these networks, testing if sequences obtainedfrom particular sites showed fixed mutations indicating separatePAA groups. This approach permits species delimitation via PAAbased on sequence information without using morphological orLinnean groups.

Tree-based methods were applied using the GMYC model thatuses the different rate of branching patterns within and among coa-lescence groups to define the species boundary (Fontaneto et al.,2007; Pons et al., 2006). In brief, the method optimizes a thresholdage that separates nodes specifying diversification events (clado-genesis) from nodes reflecting coalescence within species. Branchescrossing the threshold define genetic clusters that each obey an inde-pendent coalescent process. Confidence limits are provided whichcorrespond to threshold values ±2 logL units around the ML estimate(equation 6 of Pons et al. 2006), and significance of the GMYC modelis compared to the likelihood of the null model that the entire samplecan be fitted by a single neutral coalescent for all individuals (i.e.constitutes a single species). The GMYC method was implementedin the R package SPLITS (SPecies’ LImits by Threshold Statistics,http://r-forge.r-project.org/projects/splits/) to delimit species fromthe ultrametric tree and branching rates obtained after various tech-niques of applying a molecular clock (see above).

A similarity index was used to evaluate how well members ofGMYC groups correspond with named species: S = 2⁄Ncommon/(NGMYC + NSp), where NGMYC is the number of individuals in a GMYCgroup, NSp is the number of individuals in a named species andNcommon is the number of individuals that are both part of a GMYCgroup and a named species.

Matrices of uncorrected p-distance between species, populationsor clusters were obtained using Mega 4.0.1 (Tamura et al., 2007), andmatrices of geographical distance between sites were calculated inthe Geographic Distance Matrix Generator 1.2.2 available athttp://biodiversityinformatics.amnh.org/open_source/gdmg/.

3. Results

3.1. Sequence variation and phylogenetic analysis

Sequence data were obtained for 161 individuals of Neocicindela,with data missing for 7, 19 and 12 individuals, respectively, in

cox1, cob and rrnL, resulting in 123 distinct haplotypes. The alignedrrnL–trnL2 data included two deletions and one insertion whichcould be placed without ambiguity. Patterns of variation were typ-ical for insect mtDNA; 3rd codon positions were the most variable(65.16% of 511 informative sites), followed by rRNA/tRNA positions(16.83%), 1st codon positions (14.68%) and 2nd codon position(3.33%). All sites were A + T rich (first codon positions 59.6%, sec-ond 61.4%, third 72.7%, and rRNA/tRNA 76.0%). Nucleotide frequen-cies were homogeneous across species, except for 3rd codonpositions (p < 0.001; chi-square test of base homogeneity in PAUP).The ratio of transitions to transversions approached a plateauwhen uncorrected pairwise distances between sequences wereabout 10–11%. This was the level of divergence between Neocicin-dela species and outgroups (including Rivacindela) as well as thedeepest divergences within Neocicindela, but divergences werelower for almost all ingroup comparisons (Suppl. Fig. 1) suggestingthat branch length estimates in the ingroup were little affected bysaturation.

Bayesian (HYK + I + C model) and parsimony analyses of thecombined mtDNA data resolved well-supported groups of haplo-types that broadly agreed with the named (sub)species, butshowed clear phylogenetic subgroups within several widespreadspecies (Fig. 2) which also were geographically confined (seeFig. 1 for maps). Specifically, (i) N. perhispida was divided into threegeographically restricted groups corresponding to the named sub-species, and Neocicindela brevilunata formed an additional group(paraphyly of N. perhispida); (ii) Neocicindela wairouraensis wassplit in two clades, comprising individuals from the south of NorthIsland vs. the north of South Island; (iii) N. parryi also was sepa-rated into two lineages whose ranges divide in the southen NorthIsland where they occur sympatrically at one site; (iv) N. helmsiwas split broadly between groups confined to either North Islandand South Island, and a phylogenetically distant allopatric groupon South Island; (v) N. tuberculata, a widespread species commonon North Island, showed high variability but with less clearly de-fined substructure, while mtDNA haplotypes were paraphyleticwith respect to Neocicindela latecincta, its putative sister speciesthat occupies a non-overlapping, nearly contiguous range on mostof South Island.

Phylogenetic relationships at the species level recovered twowell-supported clades within Neocicindela (denoted as Neocicinde-la-1 and Neocicindela-2) in both parsimony and Bayesian treesearches (PP = 1, BS = 100), corresponding to a group of specieswith tridentate versus unidentate labrum, respectively (Savill,1999). In some tree searches Neocicindela was not recovered asmonophyletic with respect to its sister Rivacindela (not shown).However, in these cases topologies constrained for monophyly ofNeocicindela were not rejected as being significantly worse (n.s.;SH test). The basal branching point in Neocicindela-1 separated N.hamiltoni, a species of the subalpine zone. Neocicindela feredayi, aspecies with broader altitudinal range also reaching the subalpinezone, branched off at the next node in this clade. A further split inthis clade leads to Neocicindela austromontana, again from the sub-alpine zone, which was sister to the widespread N. helmsi (occur-ring on river edges) and N. perhispida (coastal beaches). Speciesof Neocicindela-2 generally included three clearly monophyleticlineages. The N. tuberculata/latecincta clade was a strongly sup-ported lineage grouping two grassland species. The two remaininglineages of Neocicindela-2 include N. waiouraensis, a grassland spe-cies, and N. parryi, a species of partly shaded open-forest habitat.This lineage is likely to include N. spilleri not available for this study(Savill, 1999).

Tree calibrations selecting only 21 distinct individuals (see be-low) and using previous age estimates for the Australasian group(Pons et al., 2004) put the descent from a common ancestor of Neo-cicindela to between 10.82 ± 0.48 Mya (range age of stem group

0.1 substitution/site

N. tuberculata.10B.1N. tuberculata.10B.8

N. tuberculata.10B.2N. tuberculata.15.4

N. tuberculata.6.7N. tuberculata.6.1N. tuberculata.10B.3N. tuberculata.10B.7

N. tuberculata.1.2N. tuberculata.10B.9

N. tuberculata.1.7N. tuberculata.10B

N. tuberculata.6.6N. tuberculata.6.2N. tuberculata.6.5N. tuberculata.10B.4N. tuberculata.10B.6

N. tuberculata.6.8N. tuberculata.6.4N. tuberculata.6

N. tuberculata.6.3N. tuberculata.10B.10

N. turbeculata 99N. tuberculata.1.1

N. tuberculata.1.4N. tuberculata.1

N. latecincta 103N. tuberculata.15.1N. tuberculata.15.2N. tuberculata.15.3

N. tuberculata.15N. tuberculata.1.3N. tuberculata.15.5

N. tuberculata.88N. latecincta.98

N. latecincta.22N. waiouraensis.14.2N. waiouraensis.14.3N. waiouraensis.14.4

N. waiouraensis.14.8N. waiouraensis.14N. waiouraensis.14.10N. waiouraensis.14.1N. waiouranensis.102

N. waiouraensis.14.5N. waiouraensis.14.6

N. waiouraensis.14.9N. waiouraensis.14.7

N. waiouraensis.3N. waiouraensis.3.2N. waiouraensis.3.1

N. waiouraensis.3.7N. waiouraensis.3.3N. waiouraensis.3.6

N. waiouraensis.3.5N. waiouraensis.3.8

N. dunedensis.92N. parryi.4N. parryi.4.2N. parryi.12.4

N. parryi.12.9N. parryi.12.10

N. parryi.4.6N. parryi.4.1N. parryi.4.3

N. parryi.4.4N. parryi1 91

N. parryi.4.5N. parryi.12.1N. parryi.12.2N. parryi.12.3N. parryi.12.8

N. parryi.12.5N. parryi.12.7N. parryi.12

N. parryi.12.6 N. parryi.17.8N. parryi.17.10

N. parryi.17.9N. parryi.17N. parryi.17.1

N. parryi.17.4N. parryi.17.2N. parryi.17.7N. parryi.17.3N. parryi.17.6

N. parryi.17.5N. parryi.4.7

N. parryi.4.10N. parryi.97

N. helmsi.13.1N. helmsi.13.8

N. helmsi.13.2N. helmsi.2.5N. helmsi.13.6

N. helmsi.2.3N. helmsi.2.4

N. helmsi.1a.8N. helmsi.2.2

N. helmsi.13.4N. helmsi.1a.5

N. helmsi.2.6N. helmsi.13.7N. helmsi.2.1

N. helmsi.13N. helmsi.2.7N. helmsi.2N. helmsi.13.5

N. helmsi.2.8N. helmsi.1a.6

N. helmsi.18.2N. helmsi.18.4

N. helmsi.89N. helmsi.18.8

N. helmsi.18.6N. helmsi.18.5

N. helmsi.18.7N. helmsi.18.1

N. helmsi.18.3N. helmsi.18.10N. helmsi.18.9

N. p.perhispida.10N. p.perhispida.10.3N. p.perhispida.10.4

N. p. perhispida101N. p.perhispida.10.2

N. p.perhispida.9N. p.perhispida.9.3

N. p.perhispida.9.1N. p.perhispida.9.4

N. p.perhispida.9.2N. p. campbelli.5.1N. p. campbelli.5.3N. p. campbelli.5.5

N. p. campbelli.5N. p. campbelli.5.4

N. p. campbelli.90N. p. campbelli.5.2

N. brevilunata.7BN. brevilunata.7B.6

N. brevilunata.7.1N. brevilunata.7B.7N. brevilunata.7N. brevilunata.93

N. brevilunata.7.2N. p.giveni.8.2N. p. giveni.96

N. p.giveni.8N. p.giveni.8.1N. p.giveni.8.6

N. p.giveni.8.3N. p.giveni.8.8N. helmsi.16

N. helmsi.16.1N. helmsi.16.2N. sp.87

N. austromontana.20N. feredayi.104

N. feredayi.105N. feredayi.19

N. hamiltoni.21

100/100

100/100

100/**

94/89

100/100

100/89

100/100

100/100

100/100

100/100

100/**

100/100

100/81

100/100

100/100

100/97

100/**

100/91

100/97

100/99

94/**

99/**

100/100

100/92

100/98

100/100100/100

100/100

100/**

99/74

100/84

100/99

100/†

98/†

100/96

**/100

**/96

100/†

helmsi

s.str.cam

pbelliginevi

brevilunata

feredayihamiltoni, net 13, PAA 19

dunedensis,

parryi

waiouraensis

tuberculata

helmsi

perhispida

latecincta

austromontana,

network 1

PAA 1

network 2

PAA 2PAA 3

PAA 5

network 4

PAA 6

net 5netw

ork 7PAA 8

PAA9

PAA10

network 8

PAA11PAA12

PAA13PAA14

Neocicindela 1

Neocicindela 2

PAA 16

PAA 4

net 10,

net 3,

Fig. 2. Tree from Bayesian analysis for the full set of 161 individuals. The extent of groups from statistical parsimony analysis (networks) and PAA are shown by vertical bars.Numbers on the branches are posterior probability values. The grey lines correspond to branches conforming to the branching rates in the ‘phylogeny’ portion of the tree inthe GMYC analysis. Nodes labelled with � are not present in the 50% majority rule consensus tree from parsimony analysis, and nodes labelled with �� are present but notsupported by high Bayesian posterior probability (PP < 90) or parsimony Bootstrap Support (BS < 70).

256 J. Pons et al. / Molecular Phylogenetics and Evolution 59 (2011) 251–262

J. Pons et al. / Molecular Phylogenetics and Evolution 59 (2011) 251–262 257

11.91–9.78 Mya) to 9.49 ± 1.33 Mya (range age of crown groups11.32–6.07; Table 2; Fig. 3). Estimated ages for the two main clades(basal nodes within each clade) were 6.66 and 7.26 Mya for Neoci-cindela-1 and Neocicindela-2, respectively, while ages for most lin-eages attributed to a species or subspecies name were in the rangeof 1.5–0.6 Mya. These values were estimated by transformingbranch lengths from a MrBayes tree and a single partition with aHKY + I + C model. To take into account uncertainty in topology,branch length and the clock model we performed a new analysisin BEAST using a relaxed clock with uncorrelated log normal distri-bution, and branch length were estimated by dividing the datasetin four partitions with independent models (see Material andMethods). Moreover, in the BEAST analysis we allowed uncertaintyon the calibrated node (12.4 ± 1.3, 95% confidence interval 14.54–10.26 Mya) based on confidence intervals estimated elsewhere(Pons et al., 2004). Ages from this analysis were very similar tothe previous ones but showed wider confidence intervals (Table 2).

3.2. Grouping procedures and species delimitation

Templeton’s statistical parsimony resulted in 13 independentnetworks separated by P19 mutational steps, the cut-off forP95% chance of homoplasy employed by the method to split upthe networks in this data set. The extent of the networks showedgood correspondence to the 13 named species, except that N.tuberculata and N. latecincta were lumped into a single network;three species (N. parryi, N. helmsi and N. waiouraensis) were subdi-vided into more than one network each; and network 8 includedindividuals assigned to N. helmsi, N. perhispida and N. brevilunata,i.e. it was inconsistent with named taxa (Fig. 2). PAA performedon these networks gave rise to further subdivision with 19 diag-nosable groups in total, in addition to two ungrouped divergentindividuals of N. latecincta (Fig. 2).

Sequence-based delimitation of coalescence groups using theGMYC method was performed on a clock-constrained tree madeultrametric using PL after rate smoothing in r8s. (A uniform clockwas strongly rejected for the Bayesian tree; likelihood ratio test 2

Table 2Dating of nodes in Neocicindela. The analysis was conducted in r8s under the GRT + I + C mEach node is numbered according to the topology in Fig. 3. The analyses were calibrated12.4 ± 2.5 my (Pons et al., 2004).

Node r8s

Mean St. dev. Ma

Rivacindela + Neocicindela 10.82 0.48 11Rivacindela 5.47 0.52 6Rivacindela East 3.51 0.38 4Rivacindela West 3.69 0.37 4Neocicindela 9.49 1.33 11Neocicindela-1 6.66 0.60 71 5.75 0.76 82 4.36 0.48 63 2.55 0.26 34 1.77 0.21 25 1.06 0.14 16 1.15 0.18 17 0.23 0.05 08 0.49 0.09 0

Neocicindela-2 7.26 0.53 89 5.38 0.55 610 1.71 0.22 211 3.87 0.51 512 0.60 0.12 113 0.69 0.15 114 0.39 0.09 015 1.24 0.16 116 0.92 0.19 117 0.19 0.06 0

logL = 575.12; p << 0.01, df = 159). The GMYC model was signifi-cantly better than the simple model under a single diversificationrate (p < 0.0001), but a multiple threshold model that permits dif-ferent node ages for the shifts of branching rate (Monaghan et al.,2009) did not result in further improvements (p = n.s.). Plots of MLagainst the number of lineages showed a broad hump-shaped peakaround the maximum, and a curious secondary peak correspond-ing to a much lower number of lineages (Suppl. Fig. 3). The maxi-mum likelihood solution resolved a surprisingly high total of 45GMYC entities, 23 of which were singletons (Fig. 4). The numberof GMYC units varied between 32–51 within the confidence inter-val of 2 logL units, or between 10–24 if singletons are excluded(Fig. 4). At the point of highest likelihood, none of the GMYC groupsproduced a perfect match with the named taxa, and only at themost conservative end of the confidence interval (resulting in 16GMYC groups) we correctly resolved three of the named taxa (N.p. campbelli, N. p. perhispida, N. brevilunata), while most othernamed species were split into numerous groups (Fig. 4). Whenthe GMYC method was carried out independently on each of thetwo main clades of Neocicindela, this reduced the number of enti-ties to 12 (range 6–23) for Neocicindela-1 and 20 (range 16–28)for Neocicindela-2 (Fig. 4, Suppl. Fig. 3) and slightly improved thefit with the named taxa. Overall the GMYC method provides a poorfit to the taxonomic species names, but unlike the PAA does notlump any of the named species.

Pairwise sequence variation between individuals was higherwithin the named taxa than within local populations, GMYCgroups and PAA groups (Fig. 5). This is unsurprising given the ge-netic substructure observed. Groups defined by PAA showedslightly higher intra-group variation and higher variance thanthose defined by the GMYC method. PAA groups generally aggre-gated several local populations into a single entity, while the GMYCgroups had a tendency to separate allopatrically distributed haplo-types. When comparing the inter-group divergences (Fig. 5), themeans were shifted to higher values in the Linnean species com-pared to the groups defined by GMYC and PAA. Likewise, diver-gences within groups were also higher in the Linnean species

odel and in BEAST with four partitions (PCG 1st : PCG 2nd : PCG 3rd: structural RNA).with the age of Rivacindela + Neocicindela + Macfarlandia + Abroscelis ancholaris set to

BEAST

x Min Mean Max Min

.91 9.78 10.81 13.31 8.38

.86 4.36 5.73 7.29 4.27

.32 2.50 3.76 4.86 2.76

.62 2.91 3.96 5.09 2.85

.32 6.07 10.13 12.58 7.87

.92 5.33 6.34 8.15 4.55

.20 4.41 5.45 7.04 3.97

.05 3.45 4.37 5.64 3.11

.30 1.96 2.78 3.64 1.99

.38 1.19 2.07 2.69 1.46

.53 0.75 1.23 1.65 0.83

.56 0.76 1.33 1.84 0.84

.39 0.13 0.26 0.41 0.13

.74 0.31 0.53 0.77 0.32

.57 5.97 7.44 9.42 5.59

.41 3.83 5.47 7.04 3.99

.27 1.26 1.79 2.41 1.21

.22 2.83 3.96 5.29 2.71

.05 0.38 0.61 0.91 0.35

.07 0.34 0.73 1.05 0.44

.60 0.21 0.45 0.68 0.25

.66 0.95 1.61 2.16 1.05

.66 0.63 0.97 1.39 0.61

.37 0.09 0.18 0.31 0.07

Rivacindela 22.8Rivacindela 31.3Rivacindela 115.1Rivacindela 291.1Rivacindela 112.6Rivacindela 289.1Rivacindela 51.5Rivacindela 14a.3Rivacindela 25a.3Rivacindela 76.6Rivacindela 94b.1Rivacindela 120.1Rivacindela 121.1Rivacindela 5.7Rivacindela 49.6Rivacindela 20.3Rivacindela 47.2Rivacindela 1a.3Rivacindela 12b.5Rivacindela 3b.1Rivacindela 12a.6Rivacindela 42.1Rivacindela 114.4Rivacindela 117.1Rivacindela 59a.6Rivacindela 142.2Rivacindela 67b.3Rivacindela 65.1Rivacindela 75.2Rivacindela 90.6Rivacindela 56.6Rivacindela 60.4Rivacindela 54b.2Rivacindela 137.1Rivacindela 62.1Rivacindela 70.6Rivacindela 71.2Rivacindela 63b.1Rivacindela 54a.1Rivacindela 80.1Rivacindela 59b.1Rivacindela 208.4Rivacindela 88.4Rivacindela 91.3Rivacindela 85.4Rivacindela 74.2N. tuberculata 10B.1N. turbeculata 99N. latecincta 98N. latecincta 22N. waiouraensis 3.8N. waiouraensis 14.2N. dunedensis 92N. parryi 4.2N. parryi 12.6N. parryi 4.7N. parryi 17.4N. p. campbelli 5.3N. p. perhipsidia 9.3N. brevilunata 7.2N. p. giveni 8.2N. helmsi 2.6N. helmsi 18.5N. helmsi 16.1N. austromontana 20N. feredayi 19N. hamiltoni 21M. arachnoidesA. ancholaris

10095

100

100100

100100

100

100

100

100

10099

100

100100

100

100

100

100

96

100 100

100

100

100

100

100

100

100

100

100

100

100

100

100

100100

100

100

100

100

100

99 100

100

Neocicindela 1

Neocicindela 2

Neocicindela +Rivacindela

Rivacindela

Rivacindela 1

Rivacindela 2

123456789101112 0

Neocicindela ***

100

Fig. 3. Chronogram of Neocicindela and Rivacindela at the species level. Representatives for each PAA group plus selected mtDNA lineages were used together with oneexemplar of each Wiens–Penkrot (Wiens and Penkrot, 2002) group of Rivacindela corresponding to an eastern and western clade (see Pons et al., 2006). Numbers above thebranches refer to Bayesian posterior probability values, numbers below branches refer to dated nodes with ages and confidence intervals given in Table 2. The monophyly ofNeocicindela (node labelled with ���) was enforced.

258 J. Pons et al. / Molecular Phylogenetics and Evolution 59 (2011) 251–262

02

46

810

Num

ber o

f GM

YC c

lust

ers

single

two clades

N. l

atec

inta

N. w

aiou

raen

sis

N. p

arry

i

N. d

uned

ensi

s

N. h

elm

si

N. p

. cam

belli

N. p

erhi

spid

a

N. t

uber

cula

ta

N. b

revi

luna

ta

N. p

. giv

eni

N. a

ustro

mon

tana

N. f

ered

ayi

N. h

amilt

oni

Fig. 4. Number of GMYC groups per named species. Bars show the number of GMYC groups at the low and high limit of the ±2 logL confidence interval and the optimal MLestimate in the middle. The pie charts show the proportion of specimens allocated to each of the GMYC groups when the group delimitation was carried out for all taxacombined (top row of pie charts) or separately on Neocicindela-1 and Neocicindela-2 (bottom row).

Neocicindelapopulations

Neocicindelataxonomic sp

NeocicindelaGMYC clusters

RivacindelaWP clusters

NeocicindelaPAA clusters

inter intra inter intra inter intra inter intra inter intra0.00

0.05

0.10

0.15

Fig. 5. Box plots showing pairwise sequence divergence within and betweengroups of Neocicindela defined in various ways. The graph also shows divergences inRivacindela, for comparison. WP groups are defined according to Wiens and Penkrot(2002) and closely correspond to PAA in this study. ‘Neocicindela populations’ refersto the samples obtained from a single site which may include a community ofdivergent species.

J. Pons et al. / Molecular Phylogenetics and Evolution 59 (2011) 251–262 259

than those defined under the quantitative grouping schemes whichdelimit the entities more finely. Notably, while in all cases the in-ter-group divergence is much greater than the intra-group diver-gence on average, the ranges of inter-group divergences are verywide, indicating that cut-off values of sequence divergence arenot useful to delimit these groups.

Finally, the data were compared to an equivalent data set for 46DNA-based groups in Rivacindela obtained for the same gene mark-ers and very similar analytical procedures (Pons et al., 2006). Thedivergences between groups were lower on average, indicatingthe generally greater rate of radiation in that lineage (Fig. 5). A testagainst Linnean groups in Rivacindela is not meaningful due to theincomplete alpha-taxonomy in this group. A further difference

between both data sets was with regard to the correlation of genet-ic distance with geographical distance. In Neocicindela, Mantel testsof matrix correlation of genetic and geographical distance wereweakly significant for the GMYC groups (r = 0.102; p = 0.016;one-tailed) and not significant for the 19 groups from the PAA anal-ysis (r = 0.052; p = 0.282). In contrast, in the 46 groups of Rivacin-dela the analysis established a high correlation of genetic andgeographical distance (r = 0.461; p = 0.0001).

4. Discussion

4.1. Species delimitation and geographic structure

We found incongruence of mtDNA-based groups with Linneanspecies in Neocicindela, which resulted mainly from the splittingof existing taxa. In most cases, these subgroups were allopatric,but several deeply subdivided lineages that were considered tobe a single Linnean species occurred in sympatry. The number ofputative species differed greatly under various grouping proce-dures. Most surprising was the extremely high number of entitiesin the GMYC analysis, even under the most conservative estimates(Fig. 4). This was in contrast to previous applications of this meth-od, which recovered groups that mostly coincided with morpho-logically (Ahrens et al., 2007; Fontaneto et al., 2007; Lahayeet al., 2008) or biogeographically (Papadopoulou et al., 2009a; Ponset al., 2006) delimited species. In addition, unlike the observationsfrom previous studies, the discriminatory power of the transitionin branching rates was surprisingly ill-defined, i.e. the numbersof GMYC groups varied substantially within a small interval ofprobabilities (±2 logL units). The analysis was sensitive to the tax-on choice (all Neocicindela vs. separate treatment of the two majorclades; Fig. 4), and a secondary peak in the likelihood surface sug-gested a much lower number of groups (Suppl. Fig. 2).

The results of the GMYC analysis also differed greatly fromquantitative procedures of species delineation under a diagnosticspecies concept with the PAA methodology (Cracraft, 1983; Davisand Nixon, 1992). This procedure split the 13 named (sub)speciesincluded in the analysis into 19 entities, 2 of which consist of

260 J. Pons et al. / Molecular Phylogenetics and Evolution 59 (2011) 251–262

ungrouped individuals (singletons), compared to 32–51 GMYCgroups. The difference is largely due to the amalgamation of widelydivergent haplotypes present at the same site into a single speciesby the PAA, as opposed to the recognition of separate GMYC groupsbased on branch length estimates. The GMYC method might recog-nize related haplotypes as sympatric species, rather than a single,highly variable population under the PAA, e.g. in populations as-signed to N. tuberculata (sites 1, 6, 10B), N. parryi (sites 4, 12), N.helmsi (site 18) and several others (Fig. 3).

The approach for assessing the genetic variation, in particularthe use of population-based or tree-based species concepts, there-fore greatly affected the establishment of species limits in Neocicin-dela. Multi-species assemblages are common in cicindelids,including syntopic occurrence of N. helmsi, N. tuberculata and N.waiourensis or N. helmsi and N. parryi at several locations (Table1). Their recognition as separate, sympatric species is unequivocaldue to great morphological differences. Additional species-levelgroups without such clearly recognizable traits may also co-exist.Under the population-based concept of the PAA, all co-occurringindividuals would be grouped into a single entity, which thereforehas a tendency to lump highly divergent groups. To ameliorate thiseffect we applied the PAA only within the individual networksfrom statistical parsimony analysis (TCS), rather than all individu-als co-occurring at a site, under the assumption that a single spe-cies would not span beyond a single mtDNA network (Hart andSunday, 2007). Only if this step is included, it is possible to detectsympatric species with the PAA. However, the use of TCS networksmay also generate artificial separations as, e.g., the sympatric PAAgroups at site 4 (within N. parryi) and site 18 (within N. helmsi) (seeFig. 1). Their status as closely related, sympatric species needs to betested further using unlinked genetic markers, morphology andecological traits.

Under the GMYC model, which does not take into account thelocality information, an even greater number of sympatric entitiesis recognized. This raises the question about the biological rele-vance of the GMYC groups in Neocicindela. First, we need to recog-nize the potential for artifacts in the recovery of GMYC groups frominaccurate reconstruction of the tree and branch lengths. In the cur-rent data set both measures may be affected by missing data (Mate-rial and Methods) and, in a few cases, the poor resolution ofsequencing chromatograms suggesting polymorphic sites. Thismay indicate the presence of nuclear mitochondrial pseudogenes(numts), as those reported for Rivacindela (Pons and Vogler,2005). In particular, individuals from sites 13 and 18 (N. helmsi) fre-quently showed double peaks in the sequencing chromatogramsindicating the co-amplification of numts or other secondary tem-plates resulting from heteroplasmy or even cross-contamination,although this could not be investigated here in more detail. Theseuncertainties in base calls may mislead the rates used for branchlength estimates, as is evident from the highly uneven branchlength of sister lineages, e.g. within N. helmsi (network 7) or N. waio-uraensis (network 2) (Fig. 2). This may affect the power of the GMYCmethod and the variability in ML estimates of the number and ex-tent of groups under different treatments of the data (Fig. 4).

Secondly, the recognition of different groups in Neocicindelawith tree-based vs. character-based methods may be affected bycomplex patterns of population separation and admixture, suchas the inevitable range shifts in the wake of geological and climaticchanges. In simulations, GMYC groups are formed under an islandmodel of deme structure (Slatkin, 1985) only when migration is ex-tremely low (Papadopoulou et al., 2008), i.e. the GMYC method rec-ognizes entities lacking gene exchange. Therefore it would beexpected that GMYC groups closely match the groups defined ongeographic coherence (as in the PAA), but this is clearly not thecase in Neocicindela. The outcome of cluster formation could bemore complex under episodic gene flow and secondary contact be-

tween subdivided populations that do not follow the neutral coa-lescent assumed by the model. The GMYC model will treat thesegroups as multiple cryptic species at a site, while the geography-based PAA lumps these groups. This is also evident from the highnucleotide diversity within some PAA groups compared to theGMYC groups (Fig. 5). Hence, the different results obtained withboth procedures are likely to reflect the complicated evolutionaryhistory of population separation and re-connection that has beenpostulated widely for lineages from New Zealand (Wallis and Tre-wick, 2009). Population genetics data and the use of unlinked lociwill have to confirm these conclusions.

The high intra-population diversity is likely the legacy of deepphylogeographic separation and admixture, rather than contempo-rary constraints to gene flow among individuals at a site. However,the latter would be implied under the GMYC criterion which recov-ers many groups of close relatives in sympatry, i.e. assumes theirgenetic separation despite co-occurrence at a site. As this seemsunlikely to be the cause of high genetic diversity at a site, giventhe above scenario of range movements, the geographically delim-ited local groups as defined by the PAA are considered a betterreflection of the species boundaries. We thus use of the 19 PAAgroups plus two divergent lineages of N. latecincta here repre-sented only as singletons (Fig. 3), for a total of 21 DNA-based enti-ties, as the best current representation of the species-leveldiversity of Neocicindela. The formal naming of subspecies in Neo-cicindela has been limited to three color variants of N. perhispida,but may need to be expanded, comparable to the proliferation ofrecognized subspecies in the taxonomically well studied Nearcticand Palearctic regions.

4.2. Implications for the evolution of New Zealand biotic diversity

The combination of phylogenetic and phylogeographic analysesfor the entire radiation of Neocicindela is a major step towards acomprehensive representation of mtDNA lineages across NewZealand. Our study adds to the growing body of data that establishgenetic patterns and biogeographical differentiation in NewZealand. Neocicindela represents a lineage that is generally associ-ated with open habitat types with limited vegetation cover, rang-ing from coastal beaches to alpine grassland, while N. parryi isthe only species associated with shaded forest floor (Savill,1999). Any interpretation of the patterns with regard to the impactof climate and geology on the biota, therefore, has to consider thehistorical distribution of these habitats.

Conformity of DNA data with the morphological classificationwas highest in the coastal R. perhispida where three named subspe-cies and R. brevilunata each are confined to stretches of coastline instrict allopatry while mtDNA clusters perfectly match these groups(PAA-11 to 14; Fig. 2). Some of the widespread species showedgeographic separation for example among North Island and SouthIsland populations, specifically (i) in N. parryi PAA-6 and PAA-7 vs.PAA-5; (ii) in N. waiouraensis PAA-3 vs. PAA-2; (iii) and in N. helmsiPAA-8 vs. PAA-9 and PAA-10. The latter also produced a highlydivergent group present at two sites in northern South Island la-belled as ‘new species’ by the collector (PAA-15). In addition, theless well sampled N. latecincta and N. feredayi each produced highlydivergent haplotypes, which might indicate deep genetic subdivi-sion in these species also, although sampling in N. latecincta wasinsufficient to confirm this. These broadly vicariant groups withinthe recognized Linnean species conform to a widely recognizeddisjunction in many taxa (the ‘‘beech gap’’) and are evidence ofthe postulated cladogenetic role of glaciations since the Pliocene(e.g. McCulloch et al., 2010).

While this corroborates the apparent ancient (Pliocene or Pleis-tocene) subdivision of geographically restricted populations inallopatry, several deeply subdivided lineages were found to occur

J. Pons et al. / Molecular Phylogenetics and Evolution 59 (2011) 251–262 261

in sympatry, e.g. in N. parryi (PAA-5 and PAA-7) and in N. tuberculata(divergent haplotypes within PAA-1). In addition, PAA-15 of N. helm-si was found in very close proximity to the distantly related otherlineages of this taxon (PAA-9 and PAA-10), although not in directsympatry. Similar observations of distantly related, sympatrichaplogroups have been made in studies of other arthropod groupsin New Zealand, mainly on South Island that was most affected byorogenetic and climatic changes of the Pliocene and Pleistocene.Sympatry of divergent lineages has been explained as the result offast in situ evolution of mtDNA in harvestman spiders (Boyer et al.,2007), but in insects it has been attributed to hybridization (Buckleyet al., 2006) or range movement of differentiated lineages (Marshallet al., 2008; Pratt et al., 2008; Trewick, 2008; Wallis and Trewick,2009). These range shifts seem to be directly linked to postglacialforest regrowth (Leschen et al., 2008) or retreat of glaciers in moun-tain habitat (Buckley and Simon, 2007; Trewick, 2008).

Range movements are closely dependent on the distribution ofhabitats, but in Neocicindela this dependency was affected by fre-quent evolutionary shift in habitat association. For example, theformation of alpine habitat is estimated to have occurred onlywithin the past 5 My or less (Goldberg et al., 2008), but in Neocicin-dela two lineages (N. austromontana and N. hamiltoni) apparentlyacquired an association with this habitat independently, demon-strating the plasticity in habitat traits in the largely tropical cicin-delids. Open loess and grassland habitats as those used by severalspecies of Neocicindela would have provided local glacial refugiawhich may continue to define the current ranges, e.g. in the highlylocalized N. helmsi groups PAA-9/10 and PAA-15, while also beingsubjected to range movements and contraction or expansion withthe shifting habitat availability. Range movements appear to becommon, as there was a poor correlation of geographic distanceand genetic distance for Neocicindela (see above).

The differences in genetic and geographical patterns are partic-ularly striking when compared to those in Rivacindela. This group isconfined to ephemeral lake systems in Australia’s desert interior(Kamoun and Hogenhout, 1996; Sumlin, 1997) and its diversifica-tion coincided with the break-up of large lake systems throughoutthe Pleistocene (Pons et al., 2006). Most GMYC groups in Rivacin-dela are confined to small ranges and all show similar habitat pref-erences (salt flats and sand dunes), while their intra-populationvariation was comparatively low (Fig. 5). In this respect, the pat-tern is very similar to the phylogeographic structure in the sanddune clade of N. perhispida/brevilunata whose ranges equally ap-pear to be the result of recent vicariance. Their ranges occupythe northern portion of North Island where the major effect of re-cent glacial cycles are changes in sea levels. This type of distur-bance, however, is not likely to cause population extinction, ashabitat shifts resulting from sea level changes may be slow andwould not require long-distance dispersal of populations. There-fore, these coastal habitats may have permitted the persistenceof populations in situ and produced geographically structured pat-terns (see also Papadopoulou et al., 2009a). This is unlike mostother species of Neocicindela which occupy other habitat typesand geographic regions where climatic changes likely caused rangeshifts. Taken together, these patterns support the notion that someparts of New Zealand, in particular the highlands and the southernlatitudes, and some habitat types, in particular forest habitats,were more strongly affected by glacial-induced range shifts, henceaffecting species in different ways depending on their habitat andclimate associations.

5. Conclusions

The combined phylogenetic and phylogeographic study ofNeocicindela revealed a great complexity of evolutionary patterns,

generally confirming the findings in other lineages of New Zealandinsects of deep phylogeographic structure, admixture of subdi-vided lineages, range movements and habitat shifts. Cicindelidsavoid cold climates and their distribution has to be seen in thiscontext, as habitat availability during the orogenetic and climaticchanges of the Pleistocene would have exerted strong selectionfor habitat shifts and range movements. The mix of dispersal andvicariance and the resulting high intra-population diversity inevi-tably create problems for taxonomy, as neat groupings are unlikelyto be recovered. In this situation, tree-based and population-basedmethods of species delimitation result in very differentconclusions.

Similarly complex patterns of variation in other New Zealandtaxa have raised questions about the validity of mtDNA for delim-itation or identification of species in general (Boyer et al., 2007;Trewick, 2008). However, despite the failure of the GMYC method-ology in some lineages of Neocicindela, in the beach-dwelling cladethe pattern of mtDNA subdivision was clearly in agreement withtaxonomic expectations, similar to what was shown previously inRivacindela, the Australian sister group (Pons et al., 2006). Theselatter cases support the power of mtDNA and the utility of theGMYC model to reveal biogeographically and morphologically sep-arable groups if lineage admixture is limited. The failure of theGMYC analysis in some lineages of Neocicindela therefore is indica-tion of the complex history of biota in New Zealand, and in partic-ular their evolutionary response to the Pleistocene glaciations.Unlike the situation in the Northern Hemisphere, where glaciationsresulted in large-scale extinctions and genetic homogeneity atnorthern latitudes, at southern continents instead preserved pat-terns of local separation and endemicity (Wallis and Trewick,2009). This is evident at latitudes even higher than New Zealand,e.g. in the Falkland Islands where populations of beetles show sim-ilarly high levels of intra-specific variation (Papadopoulou et al.,2009b) due to the local survival of ancient lineages during glacialmaxima and subsequent mixing. These processes resulted in great-er complexity of species level variation that is difficult to captureadequately by a rigid taxonomic system of species circumscription,of which Neocicindela is but another example.

Acknowledgments

We are grateful to David Brzoska for specimens of Neocicindelaand Saskia Bode and Lejla Buza for contributing to the DNAsequencing. This work was funded in part by Grant NER/A/S/2000/00489 of the Natural Environment Research Council (NERC).

Appendix A. Supplementary material

Supplementary data associated with this article can be found, inthe online version, at doi:10.1016/j.ympev.2011.02.013.

References

Acorn, J.H., 1992. The historical development of geographic color variation amongdune Cicindela in Western Canada (Coleoptera: Cicindelidae). In: Noonan, G.R.,Ball, G.E., Stork, N.E. (Eds.), The Biogeography of Ground Beetles of Mountainsand Islands. Intercept, Andover, UK, pp. 217–233.

Ahrens, D., Monaghan, M.T., Vogler, A.P., 2007. DNA-based taxonomy for associatingadults and larvae in multi-species assemblages of chafers (Coleoptera:Scarabaeidae). Mol. Phylogenet. Evol. 44, 436–449.

Baldwin, B.G., Sanderson, M.J., 1998. Age and rate of diversification of the Hawaiiansilversword alliance (Compositae). Proc. Natl. Acad. Sci. 95, 9402–9406.

Barraclough, T.G., Hogan, J.E., Vogler, A.P., 1999. Testing whether ecological factorspromote cladogenesis in a group of tiger beetles (Coleoptera: Cicindelidae).Proc. Roy. Soc. B 266, 1061–1067.

Barraclough, T.G., Vogler, A.P., 2002. Recent diversification rates in North Americantiger beetles (genus Cicindela). Mol. Biol. Evol. 19, 1706–1716.

262 J. Pons et al. / Molecular Phylogenetics and Evolution 59 (2011) 251–262

Boyer, S.L., Baker, J.M., Giribet, G., 2007. Deep genetic divergences in Aorakidenticulata (Arachnida, Opiliones, Cyphophthalmi): a widespread ‘miteharvestman’ defies DNA taxonomy. Mol. Ecol. 16, 4999–5016.

Buckley, T.R., Cordeiro, M., Marshall, D.C., Simon, C., 2006. Differentiating betweenhypotheses of lineage sorting and introgression in New Zealand alpine cicadas(Maoricicada Dugdale). Syst. Biol. 55, 411–425.

Buckley, T.R., Marske, K.A., Attanayake, D., 2009. Identifying glacial refugia in ageographic parthenogen using palaeoclimate modelling and phylogeography: theNew Zealand stick insect Argosarchus horridus (White). Mol. Ecol. 18, 4650–4663.

Buckley, T.R., Simon, C., 2007. Evolutionary radiation of the cicada genusMaoricicada Dugdale (Hemiptera: Cicadoidea) and the origins of the NewZealand alpine biota. Biol. J. Linn. Soc. 91, 419–435.

Cardoso, A., Serrano, A., Vogler, A.P., 2009. Morphological and molecular variation intiger beetles of the Cicindela hybrida complex: is an ‘integrative taxonomy’possible? Mol. Ecol. 18, 648–664.

Clement, M., Posada, D., Crandall, K.A., 2000. TCS: a computer program to estimategene genealogies. Mol. Ecol. 9, 1657–1659.

Cracraft, J., 1983. Species concept and speciation analysis. Curr. Ornithol. 1, 159–187.

Davis, J.I., Nixon, K.C., 1992. Populations, genetic variation, and the delimitation ofphylogenetic species. Syst. Biol. 41, 421–435.

Drummond, A.J., Rambaut, A., 2007. BEAST: Bayesian evolutionary analysis bysampling trees. BMC Evol. Biol. 7. Art. 214.

Emerson, B.C., Wallis, G.P., 1995. Phylogenetic relationships of the Prodontria(Coleotpera: Scarabaeidae: Melolonthinae), derived from sequence variation inthe mitochondrial Cytochrome Oxidase II gene. Mol. Phylogenet. Evol. 4, 433–447.

Fontaneto, D., Herniou, E.A., Boschetti, C., Caprioli, M., Melone, G., Ricci, C.,Barraclough, T.G., 2007. Independently evolving species in asexual bdelloidrotifers. PLoS Biol. 5, 914–921.

Goldberg, J., Trewick, S.A., Paterson, A.M., 2008. Evolution of New Zealand’sterrestrial fauna: a review of molecular evidence. Philos. Trans. Roy. Soc. B363, 3319–3334.

Goldman, N., Anderson, J.P., Rodrigo, A.G., 2000. Likelihood-based tests of topologiesin phylogenetics. Syst. Biol. 49, 652–670.

Hadley, N.F., Schultz, T.D., Savill, A., 1988. Spectral reflectances of 3 tiger beetlesubspecies (Neocicindela perhispida) – correlations with their habitat substrate.New Zeal. J. Zool. 15, 343–346.

Hart, M.W., Sunday, J., 2007. Things fall apart: biological species form unconnectedparsimony networks. Biol. Lett. 3, 509–512.

Huelsenbeck, J.P., Ronquist, F., 2001. MrBayes: bayesian inference of phylogenetictrees. Bioinformatics 17, 754–755.

Kamoun, S., Hogenhout, S.A., 1996. Flightlessness and rapid terrestrial locomotionin tiger beetles of Cicindela L. subgenus Rivacindela van Nidek from salinehabitats of Australia. Coleopts. Bull. 50, 221–230.

King, P.R., 2000. Tectonic reconstructions of New Zealand: 40 Ma to the present.New Zeal. J. Geol. Geophys. 43, 611–638.

Knisley, C.B., Woodcock, M.R., Vogler, A.P., 2008. A new subspecies of Cicindelalimbata (Coleoptera: Cicindelidae) from Alaska, and further review of theMaritima group by using mitochondrial DNA analysis. Ann. Entomol. Soc. Am.101, 277–288.

Lahaye, R., van der Bank, M., Bogarin, D., Warner, J., Pupulin, F., Gigot, G., Maurin, O.,Duthoit, S., Barraclough, T.G., Savolainen, V., 2008. DNA barcoding the floras ofbiodiversity hotspots. Proc. Natl. Acad. Sci. USA 105, 2923–2928.

Landis, C.A., Campbell, H.J., Begg, J.G., Mildenhall, D.C., Paterson, A.M., Trewick, S.A.,2008. The waipounamu erosion surface: questioning the antiquity of the NewZealand land surface and terrestrial fauna and flora. Geol. Mag. 145, 173–197.

Leschen, R.A.B., Buckley, T.R., Harman, H.M., Shulmeister, J., 2008. Determining theorigin and age of the Westland beech (Nothofagus) gap, New Zealand, usingfungus beetle genetics. Mol. Ecol. 17, 1256–1276.

Marshall, D.C., Hill, K.B.R., Fontaine, K.M., Buckley, T.R., Simon, C., 2009. Glacialrefugia in a maritime temperate climate: cicada (Kikihia subalpina) mtDNAphylogeography in New Zealand. Mol. Ecol. 18, 1995–2009.

Marshall, D.C., Slon, K., Cooley, J.R., Hill, K., Simon, C., 2008. Steady Plio-Pleistocenediversification and a 2-million-year sympatry threshold in a New Zealandcicada radiation. Mol. Phylogenet. Evol. 48, 1054–1066.

Marske, K.A., Leschen, R.A.B., Barker, G.M., Buckley, T.R., 2009. Phylogeography andecological niche modelling implicate coastal refugia and trans-alpine dispersalof a New Zealand fungus beetle. Mol. Ecol. 18, 5126–5142.

McCulloch, G.A., Wallis, G.P., Waters, J.M., 2010. Onset of glaciation drovesimultaneous vicariant isolation of alpine insects in New Zealand. Evolution64, 2033–2043.

Monaghan, M.T., Wild, R., Elliot, M., Fujisawa, T., Balke, M., Inward, D.J.G., Lees, D.C.,Ranaivosolo, R., Eggleton, P., Barraclough, T.G., Vogler, A.P., 2009. Acceleratedspecies inventory on Madagascar using coalescent-based models of speciesdelineation. Syst. Biol. 58, 298–311.

Nylander, J.A.A., Wilgenbusch, J.C., Warren, D.L., Swofford, D.L., 2008. AWTY (are wethere yet?): a system for graphical exploration of MCMC convergence inBayesian phylogenetics. Bioinformatics 24, 581–583.

Papadopoulou, A., Anastasiou, I., Keskin, B., Vogler, A.P., 2009a. Comparativephylogeography of tenebrionid beetles in the Aegean archipelago: the effectof dispersal ability and habitat preference. Mol. Ecol. 18, 2503–2517.

Papadopoulou, A., Bergsten, J., Fujisawa, T., Monaghan, M.T., Barraclough, T.G.,Vogler, A.P., 2008. Speciation and DNA barcodes: testing the effects of dispersalon the formation of discrete sequence clusters. Philos. Trans. Roy. Soc. B 363,2987–2996.

Papadopoulou, A., Jones, A.G., Hammond, P.M., Vogler, A.P., 2009b. DNA taxonomyand phylogeography of beetles of the Falkland Islands (Islas Malvinas). Mol.Phylogenet. Evol. 53, 935–947.

Pearson, D.L., 1988. Biology of tiger beetles. Annu. Rev. Entomol. 33, 123–147.Pearson, D.L., Cassola, F., 1992. World-wide species richness patterns of tiger beetles

(Coleoptera: Cicindelidae): indicator taxon for biodiversity and conservationstudies. Conserv. Biol. 6, 376–391.

Pons, J., Barraclough, T.G., Gomez-Zurita, J., Cardoso, A., Duran, D.P., Hazell, S.,Kamoun, S., Sumlin, W.D., Vogler, A.P., 2006. Sequence-based speciesdelimitation for the DNA taxonomy of undescribed insects. Syst. Biol. 55,595–609.

Pons, J., Barraclough, T.G., Theodorides, K., Cardoso, A., Vogler, A.P., 2004. Using exonand intron sequences of the gene Mp20 to resolve basal relationships inCicindela (Coleoptera: Cicindelidae). Syst. Biol. 53, 554–570.

Pons, J., Vogler, A.P., 2005. Complex pattern of coalescence and fast evolution of amitochondrial rRNA pseudogene in a recent radiation of tiger beetles. Mol. Biol.Evol. 22, 991–1000.

Posada, D., Crandall, K.A., 1998. MODELTEST: testing the model of DNA substitution.Bioinformatics 14, 817–818.

Pratt, R.C., Morgan-Richards, M., Trewick, S.A., 2008. Diversification of New Zealandweta (Orthoptera: Ensifera: Anostostomatidae) and their relationships inAustralasia. Philos. Trans. Roy. Soc. B 363, 3427–3437.

Rivalier, E., 1961. Démembrement du genre Cicindela Linné. IV. Faune indo-malaise.Rev. Franc. Entomol. 28, 121–149.

Rozas, J., Sanchez-DelBarrio, J.C., Messeguer, X., Rozas, R., 2003. DnaSP, DNApolymorphism analyses by the coalescent and other methods. Bioinformatics19, 2496–2497.

Sanderson, M.J., 2002. r8s, version 1.5. User’s Manual. University of California, Davis.Savill, R.A., 1999. A key to the New Zealand tiger beetles. Records of the Canterbury

Museum 13, 129–146.Shimodaira, H., Hasegawa, M., 1999. Multiple comparisons of log-likelihoods with

applications to phylogenetic inference. Mol. Biol. Evol. 16, 1114–1116.Simon, C., Frati, F., Beckenbach, A., Crespi, B., Liu, H., Flook, P., 1994. Evolution,

weighting, and phylogenetic utility of mitochondrial gene sequences and acompilation of conserved polymerase chain reaction primers. Ann. Entomol.Soc. Am. 87, 651–701.

Slatkin, M., 1985. Gene flow in natural populations. Annu. Rev. Ecol. Syst. 16, 393–430.

Sumlin, W.D., 1997. Studies on the Australian Cicindelidae XII: Additions toMegacephala, Nickerlea and Cicindela with notes (Coleoptera). Bull. WorldwideRes. 4 (4), 1–56.

Swofford, D.L., 2002. PAUP⁄: Phylogenetic Analysis using Parsimony. Version 4.0b.Sinauer Associates, Sunderland, MA.

Tamura, K., Dudley, J., Nei, M., Kumar, S., 2007. MEGA4: molecular evolutionarygenetics analysis (MEGA) software version 4.0. Mol. Biol. Evol. 24, 1596–1599.

Templeton, A.R., 2001. Using phylogeographic analyses of gene trees to test speciesstatus and processes. Mol. Ecol. 10, 779–791.

Trewick, S.A., 2008. DNA barcoding is not enough: mismatch of taxonomy andgenealogy in New Zealand grasshoppers (Orthoptera: Acrididae). Cladistics 24,240–254.

Trewick, S.A., Paterson, A.M., Campbell, H.J., 2007. Hello New Zealand. J. Biogeogr.34, 1–6.

Vogler, A.P., DeSalle, R., 1993. Phylogeographic patterns in coastal North AmericanTiger Beetles, Cicindela dorsalis, inferred from mitochondrial DNA sequences.Evolution 47, 1192–1202.

Vogler, A.P., DeSalle, R., Assmann, T., Knisley, C.B., Schultz, T.D., 1993. Molecularpopulation genetics of the endangered tiger beetle, Cicindela dorsalis(Coleoptera: Cicindelidae). Ann. Entomol. Soc. Am. 86, 142–152.

Vogler, A.P., Monaghan, M.T., 2007. Recent advances in DNA taxonomy. J. Zool. Syst.Evol. Res. 45, 1–10.

Wallis, G.P., Trewick, S.A., 2009. New Zealand phylogeography: evolution on a smallcontinent. Mol. Ecol. 18, 3548–3580.

Wiens, J.J., Penkrot, T.A., 2002. Delimiting species using DNA and morphologicalvariation and discordant species limits in spiny lizards (Sceloporus). Syst. Biol.51, 69–91.