Computational protocol: Phylogenetic plant community structure along elevation is lineage specific

Similar protocols

Protocol publication

[…] Alignment of rbcL and matK sequences within each plant family was performed with Clustal W algorithm in Mega (Tamura et al. ) and Seaview (Gouy et al. ). Profile alignment was used to align sequences between families before manual inspection of the final alignment. The final rbcL and matK matrix consisted of 3092 nucleotides in 231 species. Abies alba Mill., and Picea abies (L.) H. Karst were included as outgroup species. The best model of DNA substitution was tested using jModelTest (Posada ), which resulted in the selection of the GTR + Γ for both DNA regions. Bayesian inference (BI) was performed in MrBayes (Huelsenbeck and Ronquist ) using the selected model. The prior distributions relied on four Markov chain Monte Carlo (MCMC) chains of 30 million generations sampling species every 1000 generations. Convergence of the independent run was assessed by checking the log-likelihood and sampled model parameters in Tracer (Drummond and Rambaut ). The initial 10,000 trees were discarded, leaving 20,000 trees for estimation of the maximum clade credibility consensus tree.Estimation of divergence times was performed with Beast (Drummond and Rambaut ) with the GTR + Γ model of evolution. Specifically, nine fossils obtained from the study by Magallon and Castillo () (Table ) were used as minimal age constraints for plant stem (Brassicaceae and Polygonaceae) and crown groups (Apiales, Dipsacales, Ericales, Malpighiales, Rosaceae, eudicots, and angiosperms). The searches were run assuming an uncorrelated lognormal relaxed molecular clock and Yule process for speciation rates. The calibration points took a lognormal distribution (Table ) with the means and standard deviation chosen to reflect our confidence in the fossils used. The MCMC chain was run for 80 million generations, with trees sampled every 1000 generations. Convergence was also assessed in Tracer by checking the effective sample size (ESS) of the model parameters and assessing the stability of posterior probabilities on individual nodes from the 95% highest posterior density (HPD) estimates (e.g., Rabosky et al. ). The first 40,000 trees were discarded as burn-in, before reconstructing the molecular dated tree. The resulting phylogenetic trees were checked against the Angiosperm Phylogeny Group tree for accepted relationships among plant orders and families (APG III Group ). The outgroup species were removed from the calibrated tree to perform all the subsequent analyses. [...] To test our second hypothesis (H2), that communities are more clustered at older nodes, we computed the NRI for each of the 693 communities using the package Picante (Kembel et al. ) in conjunction with the Geiger library (Harmon et al. ). NRI is the same as the negative standardized effect size mean pairwise phylogenetic distance (MPD) among species in a community and is given in equation (Webb et al. ; Kembel et al. ): (2) where “obs” is the observed community, “rnd” is the random community and “SD” is the standard deviation (Webb et al. ).We chose this index because it is sensitive to phylogeny-wide patterns, and the computation of phylogenetic structure explicitly provides the statistical power to unravel the dominant phylogenetic pattern in a community (Webb et al. ; Kembel et al. ). The calculation was based on subtree phylogenetic distances in each community, present at each node, tested against 9999 null communities. The null model randomizations were based on random shuffling of taxa within the set of taxa present in a given community, while maintaining species richness and prevalence (Kembel and Hubbell ; Parra et al. ). This ensured that the NRI was only influenced by the species pool that subtend from the node of interest. A total of 230 phylogenetic tree nodes were estimated (except for terminal nodes with only two species). The number of communities at each node that was used to estimate average NRI values is provided in . We estimated average NRI, along with the deviation from the expected null distribution (i.e., the standard deviation of the mean NRI at each node), separately for all of the species descending from each phylogenetic tree node. These values were then plotted on the phylogenetic tree to distinguish the main trend at each node. Average NRI values of 71 (30.9%) nodes with only two species could not be estimated during the analyses, since NRI is an effect size measure that relies on more than one comparison. Positive values of NRI indicate phylogenetic clustering, while negative values indicate phylogenetic overdispersion (Webb ).We also related average NRI to node ages, to better distinguish how the time scale on a phylogeny may affect the detection of phylogenetic patterns across a particular species pool. As variations in the species richness of lineages may create a methodological bias in the level of relatedness detected between species contained in that lineage (Webb ), we used spearman's rank correlations to determine whether the species richness of lineages had a significant correlation with the patterns of phylogenetic assembly at each node. Finally, we measured the effects of lineage age and the community size (or species richness) at nodes on average NRI using a generalized linear model (GLM); this analysis was performed to differentiate the influence of age from the phylogenetic patterns observed at each node. […]

Pipeline specifications