Some species rely on others to obtain part, or in extreme cases all, of their food. For example, ant-following birds exploit insects that are disrupted by swarms of army ants. Such behavioral specialization is often linked to the evolution of morphological traits that enhance resource exploitation. In this study, I investigated whether morphological traits covary with ant-following behavior. I predicted that traits than enhance competitive ability, like a heavier weight and signals of dominance, should be more prevalent in species that rely more on ants to obtain food. Other morphological traits were also expected to diverge reflecting the unique lifestyle of ant-following birds. I focused on Neotropical families that include obligate ant-following species, as well as others that follow ants frequently, occasionally, or not at all—thus providing a gradient of ant-following behavior. Using a phylogenetic framework that incorporated primary lifestyle and foraging behavior, I identified several morphological changes in ant-following species. Compared to non-following species, obligate species were generally heavier with a significantly reduced wing area due to shorter wings and secondaries. They also had shorter tails and narrower, shorter bills. Additionally, bare periorbital areas, which are thought to act as signals in competitive interactions, were more common among obligate species than in other groups. Facultative ant-following species exhibited intermediate phenotypes and species that rarely follow ant swarms showed no significant morphological differences from those that never do. Overall, the findings support the hypothesis that morphological traits covaried with ant-following behavior in Neotropical birds.

Data collection

I searched the literature for studies reporting ant-following behavior in Neotropical birds. I focused on suboscine passerine families where the behavior occurs frequently including Thamnophilidae, Dendrocolaptidae, Conopophagidae, and Formicariidae (Martínez et al. 2021). Within the suboscines, I excluded the Furnariidae, a large family where ant-following is not common and when it occurs considered rare (Roberts et al. 2000; Pizo and Melo 2010; Sazima and D'Angelo 2015; O'Donnell 2017). Ant-following is also considered rare in the Tyrannidae (Willis 1983b), another large family closely related to the Furnariides, one of the three infraorders of suboscine passerines (Harvey et al. 2020).

I gathered information on ant-following behavior from species accounts in Birds of the World (https://birdsoftheworld.org) and also performed searches on Google Scholar combining each species name and the key words ant or ant-following. Based on available evidence, ant-following behavior for each species was classified into one of the following categories: obligate (species that forage exclusively at swarms), facultative (species that attend swarms frequently but not exclusively), rare (species that make occasional or opportunistic use of swarms), absent (well-known species with no documentation of ant-following), or unknown (species with little information on diet or behavior).

I gathered additional information on morphological traits and ecology for each species. For foraging behavior, I distinguished between sallying, gleaning or bark foraging (Remsen and Robinson 1990) using accounts from Birds of the World. Sallying involves visual searches for prey from a perch followed by snatches of distant prey typically on the ground for ant-following species. Gleaning involves more continuous searches for food and consumption of food items directly from the substrate. Bark foraging involves searching for prey beneath or at the surface of bark. I used a published standardized classification for primary lifestyle, breeding range geographic location, and morphology (Tobias et al. 2022). For primary lifestyle, species were classified as either insessorial or terrestrial based on the most common location of foraging activities. Based on occurrence maps, this source also provided the centroid of latitude for the breeding range of each species. Standardized measurements of morphological traits from this source included average body mass, tarsus length, secondary length, wing length, tail length, beak length (to the culmen), beak width, and beak depth. Wing measurements were also combined to produce two indices: hand-wing index, a proxy of dispersal ability, and approximated wing area based on wing length and secondary length. For eye size, I used transverse diameter of the eye obtained from a large study of preserved specimens (Ritland 1982). This measurement is known to vary across species as a function of ecological traits such as lifestyle and foraging behavior (Ausprey 2021; Beauchamp 2023). Finally, I examined photographs of each species to document the presence or absence of bare periorbital areas in adult birds.

Data analysis

I excluded for analysis species where ant-following could not be determined. I used the ‘phylolm’ R-package to build phylogenetic linear Gaussian models (Ho et al. 2020). I built one model for each morphological trait. The set of independent variables included: body mass, absolute centroid of the breeding range latitude, foraging behavior, ant-following behavior, and lifestyle. To account for phylogenetic relatedness, I included the variance-covariance distance matrix using the most recent consensus tree for suboscine birds based on genomic markers (Harvey et al. 2020) (Fig. 1). As the presence of bare periorbital areas is a binary variable, I used a logistic regression approach provided in the same package. Bare periorbital areas did not occur in species that do not follow ants. Therefore, I combined the rare and absent categories for this particular analysis only.

All morphological measurements were log₁₀-transformed. All quantitative variables were scaled before analysis for ease of interpretation of model estimates. I ran each model under three different evolutionary scenarios (Brownian motion model, Ornstein-Uhlenbeck model, Pagel’s lambda model) and compared AICs to determine the model with the best fit. In all cases, the best fit was provided by

Pagel’s lambda model, which was subsequently used for statistical inference. I assessed multicollinearity using variance inflation factors (VIF). All VIFs were smaller than 2.3 showing no multicollinearity issues (O’Brien 2007). With the ‘rr2’ R-package (Ives and Li 2019), I calculated the R2 for the full models including the phylogenetic signal and ecology and for partial models including only phylogeny or ecology.