Multidimensional trait morphology predicts ecology across ant lineages

1. Understanding the link between ecology and morphology is a fundamental goal in biology. Ants are diverse terrestrial organisms, known to exhibit ecologically-driven morphological variation. While relationships between individual traits and ecologies have been identified, multidimensional interactions among traits and their cumulative predictive power remain unknown. Because selective pressures may generate convergent syndromes spanning multiple traits, we applied multivariate analyses across a wide sampling of taxa to assess ecomorphological variation in an integrative context.

2. How well does morphology predict ecology? Moreover, are there quantitatively-supported ant ecomorphs? We investigated the links between trait morphology and ecology by assembling a morphometric dataset spanning over 160 species within 110 genera. Because ants occupy a wide range of ecologies, we compiled natural history data on nesting microhabitat, foraging stratum, and functional role into 35 defined niche combinations. This tripartite ecological classification and our morphological dataset were optimized under dimension reduction techniques including Principal Component Analysis, Principal Coordinate Analysis, Linear Discriminant Analysis, and Random Forest supervised machine learning.

3. Our results describe ant ecomorphospace as comprising regions of shared, generalized morphology as well as unique phenotypic space associated with specialized ecologies. Dimension reduction and model-based approaches predict ecology with 77-85% accuracy and Random Forest analysis consistently outperforms LDA. While accounting for shared ancestry, we found eye, antennal scape, and leg morphology to be most informative in differentiating among ecologies. We also note some heterogeneity between trait significance in each ecological aspect (nesting niche, foraging niche, functional role). To increase the utility of ecomorphological classification we simplified our 35 observed niche combinations into 10 ecomorph syndromes, which were also predicted by morphology. The predictive power of these machine learning methods underscores the strong role that ecology has in convergently shaping overall body plan morphology across ant lineages. We include a pipeline for predictive ecomorphological modeling using morphometric data, which may be expanded with additional specimen-based and natural history data.

Methods
Our dataset includes ant trait measurements from 15 subfamilies, 113 genera, and 167 species, with associated ecological niche information. Morphometric sampling included linear measurements of 12 cephalic traits and 5 post-cephalic traits, with 5 additional traits that were not subsequently included in our analyses but are presented here, after preliminary analyses indicated overlap with other metrics, variation due to measurement artifacts, or lack of significance across all niche aspects. Fifteen selected traits have been previously correlated with ecology. All measurements were conducted on point-mounted specimens under stereo microscopy. Including raw measurements in dimension reduction techniques such as principal component analysis can result in body size driving the overwhelming majority of variation in a dataset, masking other potentially important contributors. To explore the impact of body size, we created two datasets for analyses: a dataset comprising raw measurements and a size-corrected dataset using only ratios. Tab 1 (all data (raw measurements)) is the raw measurements dataset; tab 2 (all data (ratio measurements)) is the size-corrected ratios dataset; tab 3 (tree-matched data (raw meas.)) holds a dataset of raw measurements pruned to taxa that are included in Blanchard and Moreau's 2017 comprehensive ant phylogeny to facilitate phylogenetic corrections; tab 4 (tree-matched data (ratio meas.)) includes the same taxa as in tab 3 but with the size-corrected ratio measurements.