Computational protocol: Sexual size dimorphism in ground squirrels (Rodentia: Sciuridae: Marmotini) does not correlate with body size and sociality

Similar protocols

Protocol publication

[…] We measured CBL in 1527 specimens from museum collections. Our material covers 63 out of 68 species of ground squirrels recently recognized in the tribe Marmotini sensu []; for the number of species see []. Only adult, undamaged and located specimens (modus 15 per sex-species category) were examined. Adults were identified according to the degree of teeth abrasion or stage (adult/juvenile) was obtained directly from specimen tags. Our dataset can be biased by geographic intraspecific variability in body size documented in ground squirrels []. Nevertheless, taking into account the large variability in body size across species of the studied group, we expect that intraspecific variability does not largely influence the interspecific pattern. Moreover, only individuals determined as a same subspecies and/or collected within a restricted geographic area were measured in most species. Four Spermophilus species (S. ralli, S. pallidicauda, S. taurensis and S. brevicauda) and Marmota kastschenkoi were not included into our study, as we were not able to obtain enough specimens. All CBL measurements were taken by a single person (JM). Data on HFL for adult males and females of each species were taken from museum tags (in total available in 1392 specimens, Table ).Body mass is often used as a measure of body size in the SSD literature, including []. Nevertheless, body mass is not a good expression of body size in ground squirrels [], because many species considerably fluctuate in body mass throughout the year. For instance, data on body mass before and after hibernation are highly different e.g. [] and are not comparable to body mass in less seasonal species. Our measurements, CBL and HFL, represent the structural component of size that is less dependent on body conditions. Moreover, it is known that HFL in rodents is among those external measurements that reach their final size very early during postnatal ontogeny e.g. [,]. There is not a single perfect measurement expressing general body size (see e.g. []) and most body measurements use to be highly intercorrelated in morphometric analyses. For example, skull measurements are highly correlated with each other in ground squirrels []. To avoid this problem, we selected CBL and HFL, because they reflect size of very different body parts, although due to large variability in body size among ground squirrels, we can a priori expect that these measures will be highly correlated across all species included (we tested the correlation by Pearson test for males and females separately).The data on social organization were obtained in two ways. Most of information was taken from literature concerning ground squirrel sociality [,,]. Second, the level of sociality was classified using publications on general biology (see Table  for particular references) or our personal knowledge of the species biology (S. citellus). In this way, sociality was particularly assessed for the genus Ammospermophilus, species Callospermophilus madrensis, C. saturatus, Cynomys mexicanus, C. parvidens, Otospermophilus variegatus, Spermophilus citellus, S. suslicus, S. pygmaeus, S. fulvus, S. xanthoprymnus and Xerospermophilus mohavensis. Moreover, in the case of mentioned Palaearctic species, the social grades assignment was consulted with A. V. Tchabovski (N. A. Severtsov Institute of Ecology and Evolution, Russian Academy of Sciences) and V. Vohralík (Faculty of Science, Charles University in Prague). In classifying the species social grades, we especially consider descriptions of inter-individual contacts, age of first reproduction, age of dispersal, character of dispersal and density of individuals (see [] documenting correlation of sociality and the mentioned variables).Species were classified into five categories representing different levels of sociality defined by []: 1 – asocial, 2 – single-family female kin clusters, 3 – female kin clusters with a territorial male, 4 – polygynous harems with male dominance, 5 – egalitarian polygynous harems. Facultative monogamy, known in some marmots, was included in the last category, as it is usually interpreted as an extreme case of polygyny where males are unable to monopolize more than one female under harsh environmental conditions ([] and references therein). In cases of transient classification (i.e. level 1 – 2 in Urocitellus townsendii, Ictidomys tridecemlineatus, and 2 – 3 in Otospermophilus beecheyii) of species sociality in [], we adopted the social grades of []. Despite extensive literature search we were not able to obtain or estimate data about sociality in 17 species from our morphological dataset. These species are thus not included into the analyses concerning effects of sociality. The sociality scale in ground squirrels forms a continuum from the least social (grade 1) to the most social (grade 5) organization; species placed into same category have similar but not always identical social system []. Throughout this work, we assume that these social grades reflect the strength of sexual selection on males. However, we are aware that social systems are not always necessary surrogates for mating systems, e.g. []. We summarized information about species mating systems as well, but primary data for the group are scarce and frequently in contradiction, for a review see []. Moreover, the definition of the polygynous mating system is disputable in rodents [].The significance of SSD in a given species was tested by one-way ANOVAs. We expressed SSD as ratios of male to female mean CBL and HFL and tested whether the SSD estimation from CBL and HFL are correlated across species by Pearson´s product–moment correlation. We also tested whether these SSD indexes follow normal distribution across species by Kolmogorov-Smirnov test. Ordinary least square regression of log-transformed mean or maximal size measurements in males on log-transformed mean or maximal size measurements in females was used for testing the allometry of SSD and hence Rensch´s rule. Slope 1.0 was expected under isometric increase of male size with female size. The ANCOVA models with size measurements in males as the dependent continuous variable, size measurements in females as the continuous independent variable and degree of sociality as the independent categorical variable were used to test the relationship between SSD and sociality among species. We tested the effect of sociality on species body size by one-way ANOVAs with male, respectively female body size measurements as dependent variable and sociality as factor.Because species data are not independent, we performed analyses that take the phylogenetic relationship among species into account as well, specifically, the analyses of phylogenetic independent contrasts and PGLS [-]. We used the topology of the tribe published recently by []. Three species (U. cannus, U. mollis and A. nelsoni) are missing in their tree. We took the position of U. cannus and U. mollis from []. A. nelsoni is considered to diverged from A. leucurus, following [], we took it as sister to this species. See the Additional file for the whole composite tree. The branch length estimations for our tree are not available. Nevertheless, simulated studies showed that the independent contrasts method is sufficiently robust to errors in branch lengths []. We report results based on arbitrary, equal branch length both in PGLS models and in analyses of phylogenetic independent contrasts. In some cases, the phylogenetic contrasts based on equal branch length did not meet assumptions suggested by []. Therefore, in the analyses of independent contrasts, we used Graafen´s branch lengths instead as well, which did not lead to significant correlations between contrasts and branch length []. The results of all analyses of phylogenetic independent contrasts computed using either equal or Graafen´s branch lengths were equivalent; therefore, we report only the results for equal branch length.We used phylogenetic independent contrasts [] and PGLS models [,] - for recent application, discussion and minute description of the PGLS technique see e.g. [,] - to test for allometry in SSD, association of SSD with sociality and correlation between body size and sociality. In PGLS, the λ parameter is found by maximum likelihood. This parameter potentially varies between 0, indicating no effect of phylogenetic signal, and 1, corresponding to the analysis of independent contrasts where trait variation among species is predicted by phylogeny. Fit of nested PGLS models (models with and without a particular predictor) or PGLS models with and without a parameter restricted to a constant (e.g. λ can be restricted to 0 or 1, estimation of the λ parameter requires 1 degree of freedom) can be compared using a LR test: LRdf = −2 × [Lh (better-fitting model) – Lh (worse-fitting model)], where the best fitting model has the highest log-likelihood (Lh) score. The significance of this difference can be evaluated with a χ2 distribution with degrees of freedom equal to the difference in the number of parameters between the two competing models []. We used this approach for comparison of PGLS models with λ estimated by maximum likelihood with the analyses of raw data (λ restricted to 0 in PGLS) or of independent contrasts (λ restricted to 1) to estimate the impact of the phylogenetic signal for a given analysis. Further, we applied this procedure for comparison of PGLS models with and without predictor sociality to test the association of sociality on SSD.Degree of sociality is an ordered variable. In many cases, an ordered variable represent coarse information about an underlying continuous variable, which is also true in our case []. Thus, we used it in PGLS regressions and analysis of independent contrasts as a continuous variable. This approach may lead to increased Type I error largely due to potentially unequal distances between subsequent categories []. To control for this potential bias, we also coded sociality as a variable with two levels, that can be used as a dummy variable in regressions. In this approach, we coded sociality grades 1 and 2 as 0, and the grades 3, 4 and 5 as 1. This division is based on the expectation that different social organization should be associated with different potential for female monopolization, which should be connected with the strength of major selective pressure for male body size enlargement. And just the social grades 1–2 and 3–5 differ in the presence of the stable association of males with females.Because sample size was small for some species, we repeated all analyses after exclusion of species with the sample size in CBL less than 5 for a minority sex.We used an α level of significance of 0.05.All analyses were performed in Statistica 10.0 (Stat Soft 2011), the PDAP:PDTREE module [] within Mesquite 2.75 [] and Compare vers. 4.6b [] for independent contrasts and BayesTraits [] for PGLS models.Ethical note: The study is based on measurements of museum specimens and does not involve living animals. […]

Pipeline specifications

Software tools Statistica, Mesquite, BayesTraits
Applications Miscellaneous, Phylogenetics