2.3 | Analyses
To understand the evolution of behavioral niche in the Mygalomorphae and identify cases of niche convergence, we conducted ancestral state reconstruction (ASR) on our two behavioral characters. We compared the results of two methods: we conducted a maximum-likelihood (ML) approach (Pagel, 1999) on the genus-level phylogram and chronogram using thecorHMM R package (Beaulieu et al., 2021), and the Maximum Parsimony (MP) approach (Swofford & Maddison, 1987) on the supertree using Mesquite v3.51 (Maddison, 2008). For the ML reconstructions, we compared AICc scores across both alternate branch length sets (i.e., the chronogram and phylogram, see Wilson et al., 2022b) and across alternate state-transition models, and chose the branch-length set and model that minimized AICc (Appendix C).
Next, to visualize how mygalomorph somatic morphology relates to the behavioral niches that they inhabit, we conducted non-metric multi-dimensional scaling (NMDS) using the complete 55-character morphological dataset, revealing the position in two dimensional ‘morpho-space’ of all genera included in the study and the behavioral/ecological optima present in the infraorder. This analysis involved first calculating the Gower similarity coefficient (Gower, 1971) between all pairs of taxa based on the morphological characters, using the Claddis R-package (Lloyd, 2016) before using the resultant pairwise-similarity matrix to conduct the NMDS analysis, using the R-package vegan (Oksanen et al., 2013).
Finally, to identify the specific morphological features associated with different behavioral niches, and thereby better understand their function, we conducted a series of phylogenetic tests for correlated evolution between morphological features and behavior (Table 1). A morphological feature was tested for correlation with behavior if: (i) an association between the feature and behavior has been proposed previously in the literature; (ii) the function of the feature is known and is tied with a particular behavior; or (iii) a strong association between a feature and behavior was perceived while scoring characters for this study. We tested all selected morphological features for correlation with five key behaviors, all of which have evolved multiple times in mygalomorphs: (a) construction of a web (sheet, funnel, or curtain) at the entrance to the retreat; (b) opportunistic retreat construction (as opposed to construction of a burrow or nest); (c) construction of a burrow; (d) structural modification of the retreat entrance (with a purse, collar, turret, or trapdoor); and (e) construction of a hinged trapdoor at the retreat entrance.
We tested hypotheses in two steps. Firstly, we used the pairwise comparisons method (Maddison, 2000; Read & Nee, 1995) to test correlation between each morphological feature and all five behaviors. This method was applied as a stringent first pass because it is relatively robust to the ‘pseudoreplication problem’ that causes many other phylogenetic correlation tests to identify significant correlation in questionable scenarios (see Maddison & FitzJohn, 2015). Because this method does not consider branch lengths, it was conducted using the supertree to benefit from the additional taxa. The analysis was performed twice for each character, the first time using only pairs that contrasted in both characters (i.e., morphology and behavior), and the second time using pairs that varied in at least one of the two characters (i.e., morphology and/or behavior) (Maddison, 2000; Read & Nee, 1995). For each approach we identified 1000 alternative pairing schemes, and from these we took the highest possible P -value as our significance threshold, thereby reducing the chance of type-1 error.
After using this first step to identify significant cases of correlation, we then analysed these cases using Maximum Likelihood methods (sensu Pagel, 1994). For each case, we generated likelihood values using four different structured-Markov models: a model of independence (i.e., no correlation), and of morphological dependence on behavior, behavioral dependence on morphology, and morphological/behavioral interdependence (i.e., three alternate models of correlated evolution). We then estimated the delta-AICc for these four models to assess their relative strength. This allowed us not only to compare the aforementioned models of independence and dependence for each particular case (the best model will have a delta-AICc of 0), but also provided a way to compare hypotheses of correlation between a particular morphological feature and alternate behaviors, with the expectation that the strongest hypothesis will return the highest delta-AICc value for the independent model (indicating the relative weakness of this model compared to the strongest model of correlation for that feature/behavioral combination).