The unit of analysis
In our analyses, we used the 91 World Wildlife Foundation ecoregions of the Neotropics (Olson et al., 2001) included within the total extent of rodent range maps (-55.98º S to 12.63º N, -91.66º E to -34.79º W). These ecoregions have an average area of 16.91 ± 21.87 square degrees. By design, the WWF ecoregions consider regional species pools, represent homogeneous areas in terms of biota and climates, capture major environmental heterogeneity at a global scale, and are objectively classifiable into major habitat types (Olson et al. 2001; Smith et al. 2018). Furthermore, ecoregion ecotones represent meaningful boundaries between biological communities (Smith et al. 2018), a property highly desirable considering our hypothesis. The Neotropical ecoregions embrace a striking diversity of habitats and have changed in position due to climate change over geological time (Costa et al. 2017). Furthermore, such changes were more severe at ecoregion ecotones than at their cores (Mayle et al. 2004; Mayle and Power 2008).
We began our analyses by building an empty raster of 0.25º cell size. We used 0.25º cell size to ensure sufficient sample size in ecotones and cores. Next, we determined the coordinates of the cell centroid of ecoregions in order to obtain points at several distances from ecoregion ecotones (Fig. 1). For each ecoregion, we measured the geographical distance between each point and the ecoregion boundary using thedist2Line function (‘geosphere’ package, Hijmans 2019). As we were interested in comparing the tip-based metrics between points at the ecotone and core, we defined ecotone points as the 10 points closest to the ecoregion boundary, whereas we defined core points as the 10 points farthest from the ecoregion boundary (Fig. 1). Our total sample size was 1,820 points: 910 in cores and 910 in ecotones from 91 different ecoregions.
We obtained the identity of sigmodontine rodents whose ranges overlap the centroid points in the core and ecotone of ecoregions. We used a buffer of 0.125-degree width (half of cell size) to obtain the species composition around the points; a width of 0.125 degrees also avoided the overlap between buffers, which would result in high spatial dependence in rodent composition between neighboring points. After obtaining point-scale composition, we continued the analyses with species occurring exclusively in the ecotone or core of each ecoregion. We used the range maps of 350 of 384 species listed in Patton et al. (2015) for which we could calculate tip-based metrics. Nomenclature and classification mainly followed accounts in Patton et al. (2015), updated where necessary (see Maestri et al. 2017 and references therein). Range maps are available in Dryad Digital Repository (Maestri et al. 2019, http://dx.doi.org/10.5061/dryad.8vt6s95).