Morphological differentiation between species of Myrmelachista Roger (Formicidae:

Myrmelachista is an exclusively Neotropical ant genus. The ants from this genus nest and forage in live or dead trees, but data on its life cycle are relatively scarce. The taxonomy of this genus is considerably complex. The morphological characters of taxonomic importance are not well defined, and combining characteristics from the largest possible number of castes is thus essential. The goal of the present study was to present the results of multidimensional morphological analyses conducted based on morphometric measurements of workers and alates to help with the identification of Myrmelachista species. For that purpose, we gathered data from 277 specimens (workers, males and gynes) from eight species cohabiting Atlantic Forest areas. Eighteen morphometric variables were measured on workers and 20 on reproductive ants. Measurements related to eye size and mandible width were the variables more strongly associated with the two morphological clusters obtained based on the morphology of workers. However, the morphospace described by workers cannot be used to delimit the species of Myrmelachista


Introduction
Myrmelachista is an arboreal genus with 69 described species (Longino, 2006;Bolton, 2013).Its distribution is restricted to the Neotropical region, and 41% of the species occur in Brazilian territory (Kempf, 1972;Fernández & Sendoya, 2004).It is characterized by a visible antennal club with one or two segments, a character that is absent in the remaining Neotropical Formicinae (Fernández, 2003).Myrmelachista species have antennae with nine or ten segments; most species have nine-segmented antennae and are found mainly in Central America (only two have been described in South America), whereas ten-segmented species are mainly found in South America (only three species have been described in Central America and Mexico) (Longino, 2006).
The taxonomy of this genus is considered extremely complex (Wheeler, 1934;Snelling & Hunt, 1975;Longino, 2006).Morphological characters of taxonomic importance are not well defined, and it is necessary to combine information from all castes (workers and reproductive ants).Nakano et al. (2014) showed that some species may be separated using molecular characters, despite their high morphological similarity (Nakano, 2010).However, no detailed morphological study has been performed to evaluate species delimitation in Myrmelachista.
It is well known that many species can be difficult to diagnose and delimit, especially when using single operational criteria such as morphological characters or DNA markers (Cardoso et al., 2009;Ezard et al., 2010).Morphological data are useful for delimiting species, and arguments exist to refine approaches to species delimitation (Barr et al., 1985;Ezard et al., 2010;Seifert et al. 2014a).Some studies have suggested that multivariate analysis based on morphology provides further insights into species alpha-taxonomy and distribution of organisms (Seifert, 2002;Seifert et al. 2014ab).In this study, we present the results of multidimensional morphological analyses performed based on morphometric measurements of workers and alates (males and females) to help identify Myrmelachista species.

Collection sites and species
Collections were performed at three sites: (1) Francisco Affonso de Mello Municipal Natural Park, 23  (Fiaschi & Pirani, 2009;Colombo & Joly, 2010).Eight species of Myrmelachista can be found in this region (Suguituru et al., 2015), corresponding to 11% of the species recorded in the Neotropical Region and to 29% of the Brazilian species.Therefore, our study area exhibited high local species richness of the genus Myrmelachista, which is adequate to evaluate species discrimination tools, considering that these species are seldom recorded (Deyrup, 2003; observations on different published lists of ant species recorded in Brazil).
Species (or morphospecies) were identified by comparing the examined specimens with specimens deposited in the reference collection of the Museum of Zoology of the University of São Paulo.Vouchers were deposited in the myrmecofauna collection of the Alto Tietê Myrmecology Laboratory of the University of Mogi das Cruzes and in the Museum of Zoology of the University of São Paulo.

Number and morphology of specimens
The number of specimens per species is shown in Table 1.The morphological parameters were measured using a micrometer ruler coupled to a stereoscopic microscope.Eighteen morphological parameters were selected for workers and 20 for reproductive ants according to Longino (2006) and Silva and Brandão (2010).

Data Analysis
We used a method developed to delimit species morphometrically in a multivariate space without using an a priori definition.This approach for species delimitation identifies clusters in the morphospace when two or more well-separated groups are a better way of describing a given sample than one group (Ezard et al., 2010).The approach to species delimitation consists of four steps (based on Ezard et al., 2010): (1) extraction of orthogonal axes with robust covariance estimators (Croux et al., 2007); (2) reduction of the dimensionality of the orthogonal axes to only those with significant explanatory power (Jackson, 1993); (3) identification of the optimal number, shape and orientation of groups within the rotated dimension-reduced data (Fraley & Raftery, 2002).In step 1, each morphometric variable was centered on the median and scaled by percentile variability prior to reducing the dimensionality (i.e., the rotation was based upon the robust covariance estimates).In step 2, the number of components retained was determined using the broken stick criterion (Peres-Neto et al. 2005).In step 3, the dimension-reduced space was split into groups using cluster analysis, with optimal delimitation by K-means approaches (Klingenberg & Froese, 1991) combined with methods to estimate the shape and orientation of clusters in the morphospace, and Gaussian mixture models and a Bayesian approach were used to estimate the support for particular arrangements of clusters using iterative Expectation-Maximization methods for maximum-likelihood (Fraley & Raftery, 2002).The volume and shape can be equal or variable among axes (elliptical, round, diagonal or univariate), and the choice between competing models is made through the Bayesian Information Criterion (BIC).All analyses were performed in R version 3.1.1(R Core Team, 2014) and used the mclust (Fraley & Raftery, 2002), pcaPP (Filzmoser et al., 2014) and mvoutlier (Filzmoser & Gschwandtner, 2014) packages.

Workers
The robust PCA of 18 morphometric variables from 205 worker specimens from the Atlantic Forest populations revealed a continuum of variation but no discrete clusters within the Myrmelachista morphospace (Fig 1).The clustering on the dimension-reduced morphospace found weak evidence to reject the null hypothesis of homogeneous data in the worker measurements.The best model found two clusters of morphologies (ellipsoidal, varying volume, shape, and orientation) and did not delimit the worker individuals in the data set.The broken stick criterion suggested that two principal components should be retained, which represented 94% of the variation in the original data (Table 2).As expected, the first component is a size component, with variables loading

Males
The robust PCA of 20 morphometric variables from 21 male specimens suggested that five groups can be characterized mostly by their distinctive size and shapes in the plane of the first and second PCA axes (Fig 2).The clustering on the dimensionreduced morphospace found evidence in male measurements to reject the null hypothesis of homogeneous data in favor of the alternative hypothesis of more than one species of males.The best model found five clusters of morphology (univariate, equal variance) and delimited mainly males of M. catharinae (71%; N=7) and M. sp.7 (100%, N=4).M. arthuri was split in two groups, although very close in the morphospace (Fig 2).By contrast, individuals of M. sp.7 and M. ruszkii were equal in the morphospace; two individuals of M. catharinae were classified as a particular group.The broken stick criterion suggested that one principal component should be retained; PCI and PCII represented 89% of the variation in the original data (Table 2).The first component is a size component, with variables loading relatively weakly and positively into this component.Mandible width and petiole height were the variables with higher loads in axis I. Principal component II showed some allometry with positive loading of head width, maximum eye length, scape length, size, petiole size (length and width), femur length, distance between ocelli and ocelli width (the latter being strongly associated with PC II) (Table 2).relatively weakly and positively into this component.In the second component, mandible width and eye size loaded positively and relatively strongly with axis II (Table 2).

Gynes
The robust PCA of 20 morphometric variables from 51 gyne specimens showed that the species can be distinguished from each other with different degrees of consistency (Fig 3).The clustering on the dimension-reduced morphospace found evidence for more than one species in the gyne dataset.The best model found seven clusters of morphology (ellipsoidal, equal volume, shape and orientation) and delimited the gynes of M. catharinae (88%; N=9), M. sp.7 (100%, N=11), M. nodigera (100%, N=10) and M. gallicola (100%, N=2); individuals of M. ruszkii (N=19) were classified into two clusters suggesting high morphometric variability.The broken stick criterion suggested that two principal components should be retained, which represented 95% of the variation in the original data (Table 2).The first and second components explain 53% and 42%, respectively.The first component showed some allometry because petiole length loaded positively and relatively strongly (compared to the other variables loading negatively).Petiole length was the variable with higher loads in the first principal component; and head length and petiole length were the variables with higher loads in the second principal component (Table 2).Table 2.The loadings onto the robust principal components for workers, males and gynes specimens of Myrmelachista; larger absolute values indicate more influential traits on the dimension-reduced space.The first or second higher loadings in each axis were highlighted in bold.Workers: 18 variables and 205 specimens; Males: 20 variables and 21 specimens; Gynes: 20 variables and 51 specimens.

Workers Males Gynes
Component

Discussion
The Myrmelachista species analyzed in the present study exhibit ten-segmented antennae, a characteristic associated with a group considered to be very heterogeneous (Snelling & Hunt, 1975).Our results indicate that workers can not be identified based only on multivariate morphometric analyses.Separation of workers using morphometric data is a complex task because some studies show that the morphology of individuals of the same species originating from different colonies may vary to the point of being considered different species (Wheeler, 1934;Snelling & Hunt, 1975).
Surprisingly, the set of variables used to describe the morphometry of gynes was efficient at separating groups of species.The resulting groups of species were largely in agreement with the identification by comparison with worker specimens deposited in biological collections, the identity of which was confirmed through direct comparisons with type specimens.
Genetic analysis of M. catharinae workers revealed two groups (Nakano et al., 2014), which was not confirmed by morphometry.However, males of M. catharinae were classified into two groups, which suggests some phenotypic variability retrieved in our morphometric analyses.
The multivariate analysis indicated that characters such as head length and the sizes of the eye, ocelli and petiole may help with the delimitation of some species.In terms of function, these morphometric variables are mainly related to predation and locomotion (see review by Silva & Brandão, 2010).M. arthuri workers have been observed in the field preying on soldiers and minor workers of Atta sexdens Linnaeus, 1758, as well as on small beetles from different species (see image in Suguituru et al., 2015;GHP Castro personal communication), although this type of behavior is not usually reported for Myrmelachista species.
Our results clearly show that, unlike workers, the analysis of morphometric characters of males and especially of gynes enables the distinction of morphological groups corresponding to five of the named species in the Atlantic Forest sites studied (Tab.1).
Our observation that gynes of M. nodigera, M. catharinae and M. sp.7 are morphologically different corroborates the molecular data from Nakano et al. (2014).In contrast, the observation of different morphological groups for workers and gynes of M. ruszkii differs from the results of the molecular analyses (Nakano et al., 2014).The morphometric analysis indicated high phenotypic variability of M. ruszkii, and of males of M. arthuri.Longino (2006) stated that Myrmelachista species are more easily identified using reproductive ants, which is in accordance with our results.However, males and gynes of Myrmelachista are rarely recorded in studies of ant fauna, and the data available on Myrmelachista reproductive biology are still scarce (Nakano et al., 2012(Nakano et al., , 2013)), hindering the taxonomic resolution of this genus.Our results also indicate that the eyes, petiole and ocelli are important characters for species distinction when reproductive ants are collected.
The overlapping of morphometric characters of M. catharinae and M. arthuri workers could lead to the classification of these species under the same species name, which would then represent a morphologically and biologically highly variable species.This may also be the case for M. nodigera and M. gallicola.In both cases, the morphometric and molecular analyses produce consistent results.

Fig 2 .
Fig 2. Point clouds for males of Myrmelachista species in the Brazilian Atlantic Forest.The five clusters identified in the robust PCA are shown in different colors.Symbols: art= M. arthuri, cat = M. catharinae, rus= M. ruszkii, sp.7= M. sp.7.

Table 1 .
Total number of specimens per Myrmelachista species subjected to morphological analysis.