Species limits and phylogeography of Newportia (Scolopendromorpha) and implications for widespread morphospecies

Abstract The genus Newportia Gervais, 1847, includes some 60 nominal species distributed in the Caribbean islands and from Mexico to central South America. Modern keys to species and subspecies are available, greatly facilitating identification, but some species are based on few specimens and have incomplete documentation of taxonomically-informative characters. In order to explore genetic variability and evolutionary relationships within geographically-widespread morphospecies, specimens of Newportia (Newportia) stolli (Pocock, 1896) and Newportia (Newportia) divergens Chamberlin, 1922, two nominal species distinguished principally by differences in suture patterns on T1, were sequenced for mitochondrial 16S rRNA and cytochrome c oxidase subunit I (COI) genes from populations in southern Mexico, Guatemala, Honduras and Brazil. Newportia (Newportia) stolli is paraphyletic with respect to Newportia (Newportia) divergens within a clade from Guatemala, Honduras, and Chiapas (Mexico), most trees being consistent with a single loss of a connection between the anterior transverse suture on T1, whereas specimens of “Newportia (Newportia) stolli” from Brazil are not closely allied to those from the Mesomerican type area. The widespread morphospecies Newportia (Newportia) monticola Pocock, 1890, was sequenced for the same loci from populations in Costa Rica, Colombia and Brazil, finding that specimens from these areas do not unite as a monophyletic group. Samples of Newportia (Newportia) oreina Chamberlin, 1915, from different regions of Mexico form geographic clusters that resolve as each other’s closest relatives. These results suggest that some widespread species of Newportia may be taxa of convenience more so than natural groupings. In several cases geographic proximity fits the phylogeny better than taxonomy, suggesting that non-monophyletic species do not result from use of inappropriate molecular markers. Molecular identification is possible for specimens missing taxonomically informative morphological characters, notably damaged specimens that lack the ultimate leg pair, a protocol that may also apply to other taxonomically difficult genera that are prone to damage (such as Cryptops).


Introduction
Newportia Gervais, 1847 is a species-rich Neotropical genus that belongs to the family Scolopocryptopidae, encompassing blind Scolopendromorpha with 23 leg-bearing segments, pectinate second maxillary claws, and kinked and pineapple-shaped processes in the gizzard (Shelley and Mercurio 2005;Koch et al. 2009Koch et al. , 2010. Newportia has until recently been classified as one of two genera in the subfamily Newportiinae, distinguished from Tidops Chamberlin, 1915, by different forcipular structures (Chagas-Júnior 2011). Phylogenetic analyses based on multi-locus molecular sequence data have, however, indicated that Tidops nests within Newportia rather than being the sister group, as does another clade that had been assigned to a separate subfamily, the Mesoamerican Ectonocryptopinae (Vahtera et al. 2013).
The geographic distribution of Newportia (including Tidops, Ectonocryptops Crabill, 1977, andEctonocryptoides Shelley &Mercurio, 2005 as subgenera: Vahtera et al. 2013) extends from northern Mexico throughout Central America and the Caribbean islands to Paraguay. Most species of Newportia have tarsus 2 of the ultimate leg divided into five to nearly 40 tarsomeres, or with indistinct separation of tarsi 1 and 2. Currently some 60 nominal species or subspecies are recognised (Minelli et al. 2006 and onwards;Schileyko 2013). In many species, diagnostic features involve the spinose processes on the ultimate prefemora and femora and the number of tarsomeres, all inconvenient characters because individuals frequently lose these legs when collected.
We propose a solution to the taxonomic impediment of missing ultimate legs by using mitochondrial sequence data to supplement identifications. We also explore phylogeographic patterns within and between select species of Newportia from Mexico and Central America using parsimony and maximum likelihood methods. The resultant phylogenies allow the taxonomic value of purportedly diagnostic morphological characters to be evaluated and for the limits of morphospecies to be tested.

Methods
Thirty-four specimens of Newportia from Mexico, Guatemala, Honduras, and Costa Rica were sorted mostly from collections made by the LLAMA (Leaf Litter Survey of Mesoamerica) project, deposited in the Museum of Comparative Zoology (MCZ), Harvard University, Cambridge Massachusetts, USA and accessible through the dedicated data base MCZbase (http://mczbase.mcz.harvard.edu). All tissues were fixed in absolute ethanol and thus were amendable to DNA sequencing.
Identifications were made using the most recent key for N. (Newportia) (Schileyko, 2013), supplemented with taxonomic descriptions in modern literature (Schileyko and Minelli 1998;Chagas-Júnior and Shelley 2003), standard monographs (Attems 1930), original descriptions, and examination of type material designated by R. I. Pocock in The Natural History Museum (London) and or by R. V. Chamberlin in the MCZ.
LLAMA specimens keyed to either N. Chamberlin, 1915, or N. (N.) divergens Chamberlin, 1922. All LLAMA specimens were sequenced for two mitochondrial loci: 16S rRNA and cytochrome c oxidase subunit I (COI). These loci were selected because they vary both within and between species, and even between individuals from geographically close populations. The 34 LLAMA samples were supplemented with N. (Newportia) and N. (Ectonocryptoides) sequences from our previous work (Vahtera et al. 2013), nine new Newportia specimens from five localities in Amazonas and Roraima, Brazil, and novel sequences for an individual of N. (N.) pusilla Pocock, 1893, from Ecuador (see Table 1 for morphospecies determinations and locality data).
Total DNA was extracted from the legs utilizing the NucleoSpin®Tissue kit (Macherey-Nagel). Samples were incubated overnight. PCR amplifications were performed with illustra TM PuReTaq TM Ready-To-GoTM PCR Beads (GE Healthcare). The COI fragments were amplified using primer pair HCO1490 (Folmer et al. 1994 and HCOout (Carpenter and Wheeler 1999) and the 16S rRNA fragments using primer pair 16Sa/16Sb (Xiong and Kocher 1991;Edgecombe et al. 2002). The normal amplification cycle for COI consisted of an initial denaturation step (2 min at 95 °C), followed by 35 cycles of denaturation (1 min at 95 °C), annealing (1 min at 43 °C) and extension (1.5 min at 72 °C), followed by a final extension step (4 min at 72 °C). For the 16S rRNA fragment the cycle consisted of an initial denaturation step (2 min at 94 °C), followed by 35 cycles of denaturation (30 s at 94 °C), annealing (30 s min at 43 °C) and extension (1 min at 72 °C), followed by a final extension step (7 min at 72 °C). Visualization of the PCR products was done by 1 % agarose electrophoresis using Midori Green Advanced DNA Stain and FastGene® GelPic LED Box (Nippon Genetics, GmbH).
Samples were purified using ExoSAP-IT (Affymetrix) and sent to FIMM (Institute for Molecular Medicine Finland) for sequencing. Chromatograms were visualized and assembled using Sequencer 5.0.1 (Gene Codes Corp., Ann Arbor, Michigan, USA). Sequence alignment editor Se-Al (Rambaut 1996) was used to visualize the sequences simultaneously. GenBank registrations for new sequences are listed in Table 1.
Parsimony analysis was conducted with POY ver. 5.1.1 (Wheeler et al. 2014) run in 16 nodes in the high-performance supercluster Taito at CSC (IT-Center of Science), Finland. A timed search of three hours was first performed on the unaligned data set. The resulting tree was used as the starting tree for the next round in which an additional timed search of six hours was performed. Parameter set 111 (indel/transversion and transversion/transition costs all equal) was used throughout the searches and branch lengths were reported using the newly implemented command "report ("file_name. tre", trees:(total, branches:true))". Nodal support was calculated using parsimony jackknifing (Farris et al. 1996).
Additional analyses used a probabilistic approach with the maximum likelihood program RAxML ver. 8.0.22 (Stamatakis 2014). For these, multiple sequence alignments (MSA) were first estimated with MUSCLE ver. 3.6 (Edgar 2004) and then trimmed using Gblocks ver. 0.91b (Castresana 2000;Talavera and Castresana 2007) to remove areas of ambiguous alignment. Since COI sequences showed no length variation, they were not trimmed in Gblocks. The amount of 16S rRNA data that remained after trimming was 59% of the original 585 positions. The two data sets were concatenated using SequenceMatrix (Vaidya et al. 2011) and the concatenated data were analyzed with RAxML in the CIPRES Science Gateway (Miller et al. 2010). A unique general time reversible (GTR) model was specified for each partition independently. Nodal support was estimated using the rapid bootstrap algorithm (applying the Majority Rule Criterion) using the GTR-CAT model (Stamatakis et al. 2008).

Results
The combined analysis of both COI and 16S fragments using parsimony as the optimality criterion resulted in two most parsimonious (MP) trees of length 4625 steps.  Table 1 for coordinates of samples).
The strict consensus tree (Fig. 2) shows these two trees are almost identical, differing only in the placement of two Brazilian specimens of N. (N.) stolli in relation to each other. Comparing strongly supported clades, the maximum likelihood tree (lnL -14054.372302: Fig. 3) shows the same major geographic and taxonomic groupings as the parsimony tree. This congruence is noteworthy because the data sets analyzed under these two optimality criteria were different (unaligned in POY and analyzed using the concept of dynamic versus static homologies with some regions removed in RAxML), as are the resampling methods (jackknifing and bootstrapping, respectively). Parts of the trees that are incongruent between the two analyses involve nodes that received low resampling supports in both analyses (e.g., the positions of N. (N.) adisi and Brazilian specimen 89b relative to other species). Both analyses depict substantial branch lengths both within and between species, with only a few instances of no (or minimal) variation between specimens from the same or geographically close populations.

Mexico
Costa Rica phyletic with respect to Ectonocryptoides (and presumably Ectonocryptops). A previous scolopendromorph phylogeny (Vahtera et al. 2013) had also indicated affinity between N. (N.) oreina and N. (Ectonocryptoides) quadrimeropus; analyses based on combined molecular and morphological data resolved them as sister-groups, although only one individual of each was then available. We note that N. oreina possesses a shorter tarsus than most congeners. The phylogeny interprets the ancestral condition of the ultimate leg tarsi of Newportia as being elongate and divided into tarsomeres, with the relatively short tarsus 2 of N. (N.) oreina being a possible precursor to the stout tarsi of the submerged taxon, "Ectonocryptopinae". This transformation series increases the plausibility of the subclavate "ectonocryptopine" ultimate legs being derived from an ancestor with flagelliform tarsi, a result that was already strongly signaled by molecular phylogenies (Vahtera et al. 2013) and is reinforced by the current trees. A Mesoamerican clade uniting N. (N.) stolli and N. (N.) divergens from Mexico (Chiapas), Guatemala and Honduras is recovered in both parsimony and likelihood analyses (Figs 2, 3), though resampling methods did not strongly support it (JK <50, BS 57). N. (N.) divergens is resolved as monophyletic in the POY analyses but is nested within a paraphyletic N. (N.) stolli, implying a single loss of the median part of the anterior transverse suture on T1 (Fig. 2). However, there is no jackknife support for the divergens clade. In contrast, the likelihood analysis did not support monophyly of N. (N.) divergens; six individuals from Guatemala and Honduras resolve as a wellsupported clade (BS 98), but two others from Honduras (81, 82) are grouped with two Mexican N. (N.) stolli specimens, albeit with weak nodal support.
Specimens identified as N. (N.) stolli from the Brazilian Amazon do not unite with supposed congeners from Mesoamerica but are instead most closely related to other taxa from the same region, i.e., a specimen identified as N. (N.) monticola (91) and N. (Tidops) collaris. This result implies that N. (N.) stolli is polyphyletic and an indistinct segmentation of ultimate tarsus 2 has multiple (convergent) origins. This character had once served as the basis for recognising a subgenus N. (Scolopendrides), e.g., in the classification of Bücherl (1974), but this taxon is not used in current classifications (Schileyko and Minelli 1998). We re-examined the N. (N.) stolli specimens again in light of the signal for non-monophyly in the phylogenetic analysis, attempting to recognize any morphological character(s) that would separate the specimens from Brazil from those from Mesoamerica. However, we found no distinctive characters between the samples; the specimens appear to be morphologically indistinguishable and using the existing keys they would all be identified as N. (N.) stolli with confidence.
Costa Rican specimens of N. (N.) monticola unite as a monophyletic group (JK 100, BS 99) in both analyses. In the maximum likelihood tree (Fig. 3) a Colombian specimen of N. (N.) monticola (103974) is resolved as a sister taxon to the Costa Rican clade but this relationship is not found in the parsimony tree (Fig. 2). In neither analysis did a Brazilian specimen identified as N. (N.) monticola unite with the other supposed conspecifics.
The two included specimens of N. (N.) pusilla, one from Ecuador (specimen 86) and the other from Brazilian Amazonas (specimen 90), likewise do not form a clade but instead are situated in different parts of the tree. The Brazilian specimen conforms to "Amazonian type pusilla" of Schileyko and Minelli (1998), characterized by rudimentary paramedian sutures on T1 (in contrast to their complete absence in other populations). Both analyses group this Brazilian specimen together with N. (N.) longitarsis stechowi but since there is no strong resampling support in either analysis (JK <50,BS 69), the question about its identity and closest relative remains unclear.
We also included a few Newportia specimens that could not be identified morphologically since they lacked ultimate legs, were juveniles, or did not key out to any known species. A specimen (54) from Costa Rica has a unique character combination and is apparently a distinct species but lacks its ultimate legs. In the POY analysis it groups together, although with weak support, with the Costa Rican N. (N.) monticola clade. A very distinctive Brazilian specimen (89b) with all tarsi bipartite and tarsus 2 of the ultimate leg undivided groups at the base of the Mexican N. (N.) oreina/N. (E.) quadrimeropus clade in the parsimony analysis. However, there is poor resampling support for this grouping and it is instead allied to species with indistinctly segmented ultimate tarsus 2 and the Brazilian clade in the likelihood tree. The poor support values and topological instability under different analytical conditions render the affinities of this undescribed species uncertain.

Discussion
Some of the specimens used in this study were either of small size because of the collection methods employed (and thus may not have been appropriate for keying using traditional criteria formulated for mature specimens) or were missing their taxonomicallyinformative ultimate legs. Nonetheless, several such specimens could be identified with a high degree of accuracy because their sequence data placed them within clades whose nomenclature could be established based on standard external morphological characters. An example is provided by a juvenile from Brazil (92) that is in poor condition and cannot be identified to species. However, the analysis shows it to be a juvenile of a Brazilian clade assigned to N. (N.) stolli. This approach is likely to be valuable in other groups of taxonomically-difficult centipedes that rely heavily on characters of the ultimate leg pair but often lack those legs in fixed specimens, such as Cryptops, where the numbers of tibial and tarsal saw teeth are fundamental taxonomic characters. The identification of developmental stages or adults without key taxonomic characters is becoming standard for many groups of animals, including other arthropod groups, such as insects (Monaghan et al. 2009;Gattolliat and Monaghan 2010) and arachnids (Fernández et al. 2014).
Some morphologically delimited species were found to be monophyletic groups, like N. (N.) divergens in the parsimony analysis, but others were paraphyletic or polyphyletic. This could be interpreted as a failure of the taxonomic characters traditionally used to delimit species or a failure in reconstructing an accurate tree by the markers selected. The second option is unlikely for the reasons outlined below, especially the biogeographical patterns exhibited in many clades where "distinct" species from the same regions tend to cluster together and not with their supposed conspecifics from other geographical regions. In particular N. (N.) stolli formed a series of geographic groupings that in part were paraphyletic with respect to sympatric species (specifically, to N. (N.) divergens in Mesoamerica) or in other cases were found to be distantly related (Brazilian "N. (N.) stolli"). The first pattern is consistent with N. (N.) stolli being a grade united by a plesiomorphy (a continuous anterior transverse suture on T1), some parts of which are most closely related to a species defined by an apomorphic state (i.e., loss of the median extent of the anterior transverse suture). The tree topology, however, suggests that the Brazilian specimens identified as N. (N.) stolli are misidentified. Newportia (N.) monticola is likewise a questionable taxon, the monophyletic Costa Rican group never uniting with a specimen of the same putative species from Brazil and only variably so with one from Colombia. Brazilian N. (N.) monticola and N. (N.) stolli unite in a well-supported clade (JF and BS 100), indicating that, in this instance, geography is a better predictor of relationships than taxonomy. It is noteworthy that N. (N.) stolli and N. (N.) monticola are among the most geographically widespread "species" of Newportia, but our results suggest that the wide distribution is partly an artifact of morphologically-based identifications. The same evidently applies to N. (N.) pusilla, a morphospecies that is regarded as ranging from St. Vincent through Colombia to the Brazilian Amazon (Schileyko and Minelli 1998;Chagas-Júnior et al. 2014). Polyphyly of this species in the molecular trees suggests that its diagnostic characters (absent or rudimentary paramedian sutures on T1 and a lack of ventral spinose processes on the ultimate leg femora) evolved convergently in different regions.
Centipede systematics, still strongly influenced by mid 20 th Century conceptualisations of species (see Edgecombe 2007), primarily assumes polymorphic and geographically widespread entities. The existing concepts that N. (N.) monticola and N. (N.) stolli are widespread throughout much of Central and South America exemplify where morphospecies do not appear to correspond to clades but rather are classes defined by combinations of characters. In these instances, molecular tools may prove to be invaluable for species delimitations, and novel morphological characters will need to be identified to rediagnose polyphyletic species.