Evolution of altitudinal migration in passerines is linked to diet

Abstract Bird migration is typically associated with a latitudinal movement from north to south and vice versa. However, many bird species migrate seasonally with an upslope or downslope movement in a process termed altitudinal migration. Globally, 830 of the 6,579 Passeriformes species are considered altitudinal migrants and this pattern has emerged multiple times across 77 families of this order. Recent work has indicated an association between altitudinal migration and diet, but none have looked at diet as a potential evolutionary driver. Here, we investigated potential evolutionary drivers of altitudinal migration in passerines around the world by using phylogenetic comparative methods. We tested for evolutionary associations between altitudinal migration and foraging guild and primary habitat preference in passerines species worldwide. Our results indicate that foraging guild is evolutionarily associated with altitudinal migration, but this relationship varies across zoogeographical regions. In the Nearctic, herbivorous and omnivorous species are associated with altitudinal migration, while only omnivorous species are associated with altitudinal migration in the Palearctic. Habitat was not strongly linked to the evolution of altitudinal migration. While our results point to diet as a potentially important driver of altitudinal migration, the evolution of this behavior is complex and certainly driven by multiple factors. Altitudinal migration varies in its use (for breeding or molting), within a species, population, and even at the individual level. As such, the evolution of altitudinal migration is likely driven by an ensemble of factors, but this study provides a beginning framework for understanding the evolution of this complex behavior.


| INTRODUC TI ON
Altitudinal migration is generally described as a seasonal movement from lower elevations to higher elevations for the breeding season and a downslope movement for the nonbreeding season (Barçante, Vale, & Alves, 2017;Hayes, 1995;Mackas et al., 2010).
Most studies on altitudinal migration have focused on the food abundance hypothesis rather than predation and climatic conditions, which are extremely challenging to study across a wide range of species and habitats. Some studies on altitudinal migration have provided evidence that frugivorous bird abundance is linked to fruit and flower abundance (Chaves-Campos, 2004; Kimura et al., 2001;Levey, 1988;Loiselle & Blake, 1991) while others have shown no evidence of this phenomenon (Boyle, 2010;Hart et al., 2011;Papeş, Peterson, & Powell, 2012;Rosselli, 1994). Boyle (2017), Chaves-Campos (2004), Kimura et al. (2001) and Pratt, Smith, and Beck (2017) suggested that food abundance drives uphill migration only, but this might depend of the species since Loiselle and Blake (1991) described downhill movement for some frugivorous species in Costa Rica when food was decreasing.
If altitudinal migration evolved as a strategy to track food resources, we would predict a link between diet (foraging guild) and altitudinal migration; however, the evidence for this relationship remains unclear. Frugivory has been suggested as a driver of altitudinal migration, in part because frugivorous altitudinal migrants have been observed more frequently at higher elevations in Costa Rica (Blake & Loiselle, 2000;Boyle, Conway, & Bronstein, 2011) and Nepal (Katuwal et al., 2016). However, Barçante et al. (2017) examined the foraging guild of all altitudinal migrant birds and showed that invertivorous altitudinal migrants are most abundant worldwide, except in the Neotropics where frugivores and nectivores are more abundant. Despite the fact that insect abundance in temperate regions is often posited as a major driver of the evolution of long-distance migration, little research has been dedicated to the role of insect abundance in the study of altitudinal migration even though insect intake might be crucial during the breeding season (Chaves-Campos, 2004;Levey, 1988) and invertivore bird species have been shown to vary in elevation seasonally in mountainous environments, such as Nepal (Katuwal et al., 2016).
Altitudinal migration has been observed in every zoogeographical region in the world (Barçante et al., 2017) although some hotspots seem to host a higher proportion of altitudinal migrants, such as the Himalayas and western North America (Boyle, 2017).
It is important to note, however, that some of this variation in the proportion of altitudinal migrants could result from a difference in sampling efforts across the world (Barçante et al., 2017).
Alternatively, environmental conditions in those regions, such as habitat availability and seasonality, may also favor the evolution of altitudinal migration.
Our goal was to examine potential drivers of the evolution of altitudinal migration in passerines. The order Passeriformes represents approximately half of the avifauna and 13% of them are described as altitudinal migrants, making them a good choice for this study. Of the 6,579 passerines species and subspecies recorded in this study, 830 species are considered altitudinal migrants and are distributed across 77 of the 137 families of Passeriformes ( Figure 1). Using a speciose and globally distributed group of birds, we conducted large-scale phylogenetic comparative analyses to examine evolutionary associations between altitudinal migration and diet (foraging guild) and habitat. In addition, we asked whether these associations differ depending on the zoogeographic region. We expected that frugivorous and nectivorous species were driven toward altitudinal migration in the Neotropics because they were tracking fruit and flower abundance which varies seasonally (Barçante et al., 2017;Chaves-Campo, 2004;Kimura et al., 2001;Levey, 1988;Loiselle & Blake, 1991). For every other region, invertivorous species would be driven toward altitudinal migration (Barçante et al., 2017). We also expected altitudinal migration to be evolutionary associated with forest habitats in the Neotropics because altitudinal migrants in Costa Rica (Stiles, 1988;Stiles & Clarke, 1989) and southeastern Brazil (Stotz, unpublished-see Stotz, Fitzpatrick, Parker, & Moskovits, 1996), for instance, include a high number of restricted-range and forest-dependent species.

| Ethics statement
No permits were required for this project.

| Data collection
We compiled data for species and subspecies of songbirds across the world, from supplementary material in Barçante et al. (2017) and Wilman et al. (2014), and data mining from two online databases: IUCN Red List and BirdLife Data Zone (retrieved in November 2018).
All entries were checked for nomenclature inconsistencies. Our universe consists of all 6,579 passerines in the IUCN Red List database, downloadable from their website https://www.iucnr edlist.org/ search after restricting (advanced) searches by taxonomy selecting, in the "search filters" option [Kingdom = Animalia; Phylum = Chordata; Class = Aves; Order = Passeriformes]. We associated four variables to each species: altitudinal migration status, primary habitat preference, foraging guild, and zoogeographic region.
A species was classified in our dataset as altitudinal migrant if its (common or scientific) name is listed in Barçante et al. (2017) either as altitudinal (238 species) or probable altitudinal migrant (592 species). BirdLife Data Zone provides, among many other information, the list of preferred breeding and nonbreeding habitats of a given species on the webpage http://dataz one.birdl ife.org/speci es/ facts heet/common_name-scien tific_name/details (where spaces are replaced by the character "-" on its common and scientific names).
Considering the great variety of habitats, we only used the major natural breeding habitat for each species and collapsed habitats into four major categories: dense habitat (forest + shrubland, 4,635 species), open habitat (grassland + savanna + open woodland + rocky areas, 563 species), water habitat (wetland + marine, 164 species), and generalist (species that occupied two or more major categories, 1,217 species). A total of 1,217 species occupied two or more major categories and were classified as generalists. To build the zoogeographic region (Newton & Dale, 2001), we downloaded from IUCN Red List website 13 lists of Passeriformes, each with all Passeriformes observed on a specific "Land Region" (selected in the "search filters" option) and translated those regions to a reduced set of zoogeographical regions as follows: "Caribbean islands" = Neotropical, "Antarctica" = Neotropical, "East Asia" = Indomalayan, "Europe" = Palearctic, "Mesoamerica" = Neotropical, "North Africa" = Checked individually; "North America" = Neartic. "North Asia" = Palearctic.

| Phylogeny
We downloaded the first 1,000 trees from Hackett backbone phylogenetic trees (Hackett et al., 2008
Brownian correlation and the maximum likelihood method were applied to each model. The models consisted of the response variable (altitudinal migration) coupled with each predictor individually (diet, habitat, and region), predictors paired together, or all predictors together. Two models also included an interaction; one between diet and region and one between habitat and region. The interaction was included to test whether the patterns of guild vary from one zoogeographical region to another as shown by Barçante et al. (2017); the same was applied to habitat. For the models with the interaction, we had to merge frugivore/nectarivore with seed/ plant material and vertebrate/fish/scavenger with invertivore, resulting in three diet categories: herbivore, omnivore, and invertivore. We ranked the models using Akaike's information criterion (AIC). We considered the top models competitive if they differed by <4 AIC units.

| RE SULTS
The best phylogenetic generalized least square model that predicted altitudinal migration included diet, region, an interaction between diet and region (Table 1 and Figure 2). The addition of habitat as a predictor did not improve the model's AIC. However, habitat was still associated with altitudinal migration (F 3 = 3.98, p = .0076; Figure 2b), with more altitudinal migrants in open habitat than dense habitat, water, and generalist. When we examined the terms in the top-ranked model, we found strong effects of foraging guild (F 2 = 6.48, p = .0016; Figure 2a), region (F 6 = 23.77, p < .0001; Figure 2c), and a foraging guild:region interaction (F 12 = 10.05, p < .0001). The interaction model revealed that herbivore/widespread (t = 4.75, p < .0001), omnivore/Palearctic (t = 3.26, p = .0011), and omnivore and herbivore/Nearctic (t = 7.43, p < .0001, t = 4.43, p < .0001) species were more likely to exhibit altitudinal migration (Table 2). The regions that revealed an evolutionary association between altitudinal migration and foraging guild were the Nearctic, Palearctic, and Widespread. However, the foraging guilds associated with altitudinal migration differed between these regions.

| D ISCUSS I ON
In the Nearctic, herbivore and omnivore species were more likely to be altitudinal migrants, a finding consistent with Boyle (2017).  (2004) and Levey (1988), who suggest that birds should follow fruit abundance during the nonbreeding season and insect abundance during the breeding season. Altitudinal migration would then be beneficial for omnivorous species. Omnivorous species are also linked to altitudinal migration in the Palearctic. Barçante et al. (2017) indicated that the proportion of frugivore/ nectarivore species that are altitudinal migrant in the Palearctic was lower than expected; our results demonstrating a disproportionate number of omnivorous species agrees with their findings.
For the species with a widespread distribution, herbivorous species were associated with altitudinal migration. This finding agrees with previous studies where herbivorous species have been indicated as altitudinal migrants all around the world (Blake & Loiselle, 2000;Guillaumet, Kuntz, Samuel, & Paxton, 2017;Katuwal et al., 2016;Kimura et al., 2001; but see Barçante et al., 2017). However, only 26 species are considered to have a widespread breeding distribution, so this interpretation should be taken with caution.
For the other regions (Neotropical, Indomalayan, Afrotropical, and Australasian), foraging guild was not directly associated with altitudinal migration, potentially due to the vast complexity of tropical ecosystems.
Habitat Altitudinal migration is challenging to study in part because of the variability in the expression of the behavior. For instance, some populations within the same species are altitudinal migrants while the others are resident (Boyle, 2017;Green, Whitehorne, Middleton, & Morrissey, 2015). There is also variation in the propensity to engage in altitudinal migration among individuals within a population (Boyle, 2008b(Boyle, , 2017Pratt et al., 2017;Rohwer et al., 2008) and within individuals across time (Hahn et al., 2004). In addition, most studies focus on the importance of altitudinal migration to birds moving to reach breeding grounds, but birds may also move up or downslope to reach molting grounds (Rohwer et al., 2008;Wiegardt et al., 2017). As such, this variation makes it extremely difficult to generalize and categorize birds as altitudinal migrants.
We suggest that more studies are needed about specifics of alti- Another limitation in our study is the lack of information for some regions (Barçante et al., 2017 The present study is the first to examine potential large-scale drivers of the evolution of altitudinal migration in passerines. Altitudinal migration has evolved independently in different regions of the world under the different environmental pressures coupled with varying life history characteristics. Our results have reinforced the idea that diet (foraging guild) played a major role in the evolution of altitudinal migration. However, the relationship between diet and altitudinal migration is complex and varies across different regions in the world. Given the prevalence of this behavior across foraging guilds, diet is clearly not the only factor that drove the evolution of altitudinal migration, but rather the evolution of this trait was likely driven by an ensemble of factors.

ACK N OWLED G M ENTS
We would like to thank S. Joly for providing drawing of birds used in Figure 1

CO N FLI C T O F I NTE R E S T S
None declared.

O PEN R E S E A RCH BA D G E S
This article has earned an Open Data Badge for making publicly available the digitally-shareable data necessary to reproduce the reported results. The data is available at https://doi.org/10.5061/ dryad.jwstq jq5n.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data are accessible on Dryad (https://doi.org/10.5061/dryad.jwstq jq5n).