Environmental variation and the evolution of large brains in birds

Environmental variability has long been postulated as a major selective force in the evolution of large brains. However, assembling evidence for this hypothesis has proved difficult. Here, by combining brain size information for over 1,200 bird species with remote-sensing analyses to estimate temporal variation in ecosystem productivity, we show that larger brains (relative to body size) are more likely to occur in species exposed to larger environmental variation throughout their geographic range. Our reconstructions of evolutionary trajectories are consistent with the hypothesis that larger brains (relative to body size) evolved when the species invaded more seasonal regions. However, the alternative—that the species already possessed larger brains when they invaded more seasonal regions—cannot be completely ruled out. Regardless of the exact mechanism, our findings provide strong empirical support for the association between large brains and environmental variability.


Supplementary
. Relative brain size (Mean ± SEM) in resident species as a function of latitude of the region where they occur. Residents from Higher latitude have larger brains than residents from lower latitudes (PGLS: p = 0.015, N=855, See Supplementary Table 1). Residents from higher (N=53), mid (N=335) and low (N=467) latitudes are represented by black, dark-grey and light-grey bars, respectively. Figure 3. Distribution of slopes (blue histograms) and p-values (red histograms) of PGLS models using 100 different phylogenetic trees linking brain size with PPC1 (a-b) and PPC2 (c-d), while controlling for log(body size).

Supplementary Figure 4.
Variation in productivity (Mean ± SEM), measured by the coefficient of variation of enhanced vegetation index (EVI) across 15 years during breeding season in the breeding distribution range (light grey) and during non-breeding season in the non-breeding distribution range (dark grey) of bird species exposed to different degree of environmental variation.

Supplementary Figure 5.
Relationship between relative brain size and migratory distance within avian orders. The lines are fitted on raw data, with a SEM interval; pvalues and R 2 are derived from PGLS models. Only orders with at least 10 species are presented. Figure 6. Distribution of the species used in the study exposed to different degrees of seasonality, divided in those residing the entire year at higher (yellow circles), medium (orange circles) and low latitudes (red circles). An alternative strategy to avoid seasonal changes is to migrate every year to avoid harsh winters, exhibiting long-distance migrations (dark-blue triangles) or short-distance travels (lightblue triangles). Each dot is plotted in the breeding centroid of the distribution area using the worldHires map from 'mapdata' R-package. Supplementary Figure 9. Phylogenetic regression between body and brain size (a), used to take into account allometric effects in brain size and obtain a relative measure based on the residuals (b).

Supplementary Tables
Supplementary Table 1. PGLS modelling variation in brain size (log-transformed) as a function of body size and latitude (with Low latitudes taken as reference for comparisons) for resident species. For partial migrants, if both breeding and wintering areas were available, the distance was calculated between these two areas. If not, the distance was calculated as the distance between resident centroid and non-breeding centroid or resident centroid and breeding centroid. We then plotted the migratory distance frequencies and we identified two clearly defined groups: birds migrating less than 5000 km and birds migrating more than 5000 km (Supplementary Fig. 7). This threshold was used to classify short and longdistance migrants. Finally, we also divided resident species into low latitudes (between 0º and 20º of latitude centroid of breeding regions), medium latitudes (between 20º and 40º) and high latitudes (above 40º of latitude). Therefore, we ended up with five categories (short-distance migrants, long-distance migrants, high latitude residents, medium latitude residents and low latitude residents) representing different selective regimes for 2 environmental variation (either characterized by degree of seasonality due to latitude differences or by mobility among regions). We then classified each of the 1,217 species for which we had information for brain size in one of these categories (See Supplementary Fig. 6, drawn using 'maps' 4 and 'mapdata' 5 R-packages.). Altitudinal movements and nomadic movements were not considered, so species following these movement patterns were pooled together within the category of resident species. Finally, birds that spent an important part of their life in open sea (e.g. pelagic birds) were neither considered because their migratory routes are largely unknown and the seasonality in their resources cannot be estimated using EVI information, as is the case for land species.
Environmental data. To characterise environmental variability, we used data from MODIS sensor, which was processed to generate the enhanced vegetation index (EVI) 6 .
EVI is a measure derived from the normalized difference vegetation index (NDVI). Both indices use chlorophyll radiation to estimate active leave density, which is a good proxy of primary production 7 . However, EVI has improved sensitivity in high biomass regions and improved vegetation monitoring through a decoupling of the canopy background signal and a reduction in atmosphere influences. EVI index is therefore a good proxy for primary productivity over time 8,9 . We used EVI time series from the available years (2000 to 2014) at 16 days of temporal resolution and 0.05º of spatial resolution 6 . We estimated EVI of each breeding and non-breeding area using 'sp' 10,11 , 'raster' 12 and 'rgdal' 13 , 'rgeos' 14 R-packages 15  consumed were given a score of 0.1. We considered information at the species level, and therefore gave to each food category the maximum value reported in any of the populations of the species (e.g. if one population only eats fruits but another population of the same species eats fruits and insects, "fruits" was given a score of 1 and insect a score of 0.5). Note however that for the great majority of species, details on diet composition at the population level were not available. We then estimated diet breadth using Rao's quadratic entropy as implemented in the r-package 'indicspecies' 17, 18