Global body size distributions are shaped by selection pressures arising from biotic and abiotic factors such as temperature, predation and parasitism. Here, we investigated the ecological and evolutionary drivers of global latitudinal size gradients in an old insect order (Odonata; dragonflies and damselflies). Phylogenetic comparative analyses revealed that global size variation of extant taxa is negatively influenced by both regional avian diversity and temperature. Interestingly, fossil data show that the relationship between wing size and latitude has shifted: latitudinal size trends had initially negative slopes but became shallower or positive following the emergence of birds 150 MYA. These changing size-latitude trends over geological time were likely driven by bird predation and high dispersal ability of large dragonflies. Our results therefore suggest that latitudinal size gradients were shaped by temperature but also by predators driving the dispersal of large-sized clades out of the tropics and in to the temperate zone.

The phylogenetic tree that we use in this study was constructed from odonate DNA-sequences downloaded from GenBank, using a traditional morphological taxonomy as our backbone (Waller and Svensson 2017). Our phylogeny comprised 1322 taxa, about a fifth of all known extant odonate species (about 6400 in total) (Schorr and Paulson 2015; Waller et al. 2019). Further and more detailed information about phenotypic data collection and phylogeny reconstruction is provided elsewhere (Waller and Svensson 2017; Waller et al. 2019) and  from our open online resource “The Odonate Phenotypic Database”: http://www.odonatephenotypicdatabase.org/shiny/odonates/.

Size data (wing lengths) from extinct odonate fossil taxa were obtained from Clapham and Karr (2012). Although we use our previously published Odonata phylogeny in the present study (Waller and Svensson 2017) and we have published data on interspecific body size variation from the Odonate Phenotypic Database (Waller et al. 2019), all of the analyses in the present paper are new and have not been published elsewhere. Thus, there is no overlap between the results in the present study and our previous work.

All our statistical analysis have been carried out in the R statistical environment. We enclose a README-file in text-format, CSV-files with data and two R-files with code and functions to replicate all the analyses in the current paper.