Climate and ecology predict latitudinal trends in sexual selection inferred from avian mating systems

Abstract

Sexual selection, one of the central pillars of evolutionary theory, has powerful effects on organismal morphology, behaviour, and population dynamics. However, current knowledge about geographical variation in this evolutionary mechanism and its underlying drivers remains highly incomplete, in part because standardised data on the strength of sexual selection is sparse even for well-studied organisms. Here, we use information on mating systems-including the incidence of polygamy and extra-pair paternity-to estimate the intensity of sexual selection in 10,671 (>99.9%) bird species distributed worldwide. We show that avian sexual selection varies latitudinally, peaking at higher latitudes, although the gradient is reversed in the world's most sexually selected birds-specialist frugivores-which are strongly associated with tropical forests. Phylogenetic models further reveal that the strength of sexual selection is explained by temperature seasonality coupled with a suite of climate-associated factors, including migration, diet, and territoriality. Overall, these analyses suggest that climatic conditions leading to short, intense breeding seasons, or highly abundant and patchy food resources, increase the potential for polygamy in birds, driving latitudinal gradients in sexual selection. Our findings help to resolve longstanding debates about spatial variation in evolutionary mechanisms linked to reproductive biology and also provide a comprehensive species-level data set for further studies of selection and phenotypic evolution in the context of global climatic change.

Fig 1. Potential latitudinal gradients in sexual selection and their putative mechanisms.

One hypothesis proposes that avian sexual selection increases with latitude because highly seasonal climates promote short, synchronous breeding seasons, leading to intense competition for mating opportunities in dense populations. This concept may apply primarily to higher trophic levels, with a gradient from monomorphic insectivores with year-round territories in the tropics (for example, Uniform Treehunter) to highly dimorphic migratory insectivores at higher latitudes (for example, Blackburnian Warbler). An alternative hypothesis predicts that sexual selection increases towards the tropics because climatic stability promotes year-round availability of food resources, including abundant fruit and flowers, leading to extreme polygamy in dietary specialists. This concept may apply primarily to lower trophic levels, with a gradient from monomorphic frugivores with biparental care at higher latitudes (for example, Japanese Waxwing) to highly dimorphic frugivores with female-only care in the tropics (for example, Wilson’s Bird-of-Paradise). Images are from Birds of the World, reproduced with permission of Cornell Lab of Ornithology.

Fig 2. Comparison of scores with alternative metrics of sexual selection.

Panels show results of Bayesian phylogenetic models assessing the relationship between sexual selection scores compiled for this study and 3 independent measures of sexual selection: (a) residual testes mass, (b) Bateman gradients (β_ss), (c) the opportunity for sexual selection (I_s) (see Methods). Dashed lines show the relationship for subsets of species with available data, using predictions from models accounting for phylogeny using a sample of 50 phylogenetic trees extracted from www.birdtree.org [50], grafted to the Prum and colleagues [61] genomic backbone. Effect sizes, p-values, and explained variance (R²) were calculated from fixed effects only, and therefore represent the specific variation explained by each predictor separate from phylogenetic effects. Data-point colour reflects increasing levels of sexual selection from low (blue) to high (red). In (b) and (c), point size is scaled by the number of individuals used to calculate sex-specific metrics. To compare sex-specific metrics with bidirectional scores, we selected the largest β_ss or I_s from each species, treating males and female metrics equally. We estimated I_s of 51 genetically monogamous species (with 0% extra-pair paternity) as zero. Excluding these species from I_s models produced similar results (Table B in S2 Text). Illustrations show exemplars of extremely low (Black Guillemot) and extremely high (Lance-tailed Manakin) I_s, respectively; images are from Birds of the World, reproduced with permission of Cornell Lab of Ornithology. The data underlying this figure can be found at https://doi.org/10.6084/m9.figshare.27255609.

Fig 3. Geographical distribution of sexual selection in birds.

(a) Worldwide variation in sexual selection scores for 9,836 bird species included in a global phylogeny (www.birdtree.org [50]), averaged for all species occurring in 5-km grid cells based on breeding range maps. Sexual selection was scored based on mating behaviour from monogamy (scored 0) to extreme polygamy (scored 4; Table 1). (b) Relationship between avian sexual selection and breeding latitude. (c) Geographical variation in data certainty, ranging from no evidence (scored 1) to strong evidence (scored 4; see Table A in S2 Text). (d) Relationship between data certainty and latitude. In (b) and (d), points denote mean sexual selection per 200-km grid cell (Behrmann projection); dashed lines were generated from spatial simultaneous autoregression (SAR) models predicting mean sexual selection strength. Additional SAR models on a subset of species with higher-quality data (scored 3–4 for data certainty) showed similar patterns (S1 Text and Table C in S2 Text). To reduce noise, cells with <10 species were excluded from all plots and models. To aid visualisation, plots were coloured using discrete intervals with an equal number of cells. Results are plotted using geographical range polygons provided by BirdLife International (www.datazone.birdlife.org) cropped to Earth’s land-surface using the BIO1 climate layer (www.chelsa-climate.org). The data underlying this figure can be found at https://doi.org/10.6084/m9.figshare.27255609.

Fig 4. Global distribution of sexual selection partitioned by trophic level and diet.

Upper panels show strength of sexual selection in primary (1^ry) consumers mapped globally (a) and plotted against latitude (b), as well as for frugivores only (c, d). Lower panels show strength of sexual selection in secondary (2^ry) consumers mapped globally (e) and plotted against latitude (f), as well as for invertivores only (g, h). We treated omnivores, carnivores and scavengers as secondary consumers. Sexual selection was scored from monogamy (0) to extreme polygamy (4; see Methods). For mapping, we averaged sexual selection scores in each 5-km grid cell based on all species with breeding ranges overlapping the cell. To aid visualisation, maps were coloured using discrete intervals with an equal number of cells. In scatterplots, points represent mean sexual selection per 200-km grid cell; dashed lines were generated from spatial simultaneous autoregression (SAR) models predicting mean sexual selection strength. Sensitivity analyses on a subset of species with higher-quality data (scored 3–4 for data certainty) showed similar patterns (Table D in S2 Text). To reduce noise, cells with <10 were excluded from all plots and models. Results are plotted using geographical range polygons provided by BirdLife International (www.datazone.birdlife.org) cropped to Earth’s land-surface using the BIO1 climate layer (www.chelsa-climate.org). The data underlying this figure can be found at https://doi.org/10.6084/m9.figshare.27255609.

Fig 5. The role of climate in regulating avian sexual selection.

(a) A family-level consensus phylogenetic tree showing the distribution of sexual selection scores across 194 families (n = 9,988 species). Terminal bars show average strength of sexual selection estimated for each family; longer bars indicate higher scores (see Methods). Coloured segments demarcate major clades; icons depict representative families. Panels (b–d) show average sexual selection scores for each family, partitioned by trophic level (b), migratory behaviour (c), and territoriality (d); black points show predicted mean values for each group from species-level Bayesian phylogenetic models; whiskers show 95% credible intervals. In (b), 1^ry = primary; 2^ry = secondary. Species traits shown in (b–d) are all hypothetically linked to temperature seasonality, which is also related to sexual selection scores (e); dashed line shows model fit from a species-level Bayesian phylogenetic model; shaded area denotes 95% credible intervals from model predictions. In (b–e), data points for each family are coloured by major clades (see phylogeny), with size and transparency of each data point scaled to within-family species richness. Statistics show effect size and p-value comparing main effects with reference groups: secondary consumer, no migration, and no territoriality, respectively. For ease of visualisation, data shown in panels are family-level averages, which do not fully reflect the stronger species-level correlations reported in statistics. Running the same models on high-certainty data produced similar results (Table F in S2 Text). The data underlying this figure can be found at https://doi.org/10.6084/m9.figshare.27255609. Icons were generated using R (rphylopic package) and are under the Creative Commons Attribution 4.0 International (CC BY 4.0) License.

Fig 6. Relative roles of ecology and climate as drivers of sexual selection in birds.

Results shown are from Bayesian phylogenetic models testing drivers of sexual selection in 9,836 species. Predictors include 3 life history variables (green), 1 climatic variable (pink), and 2 key interactions between diet and the dominant effects (territoriality and seasonality; blue). 1ry consumer = primary consumer. The reference groups for the 3 categorical predictors are as follows: secondary consumer; no migration; and no territoriality, respectively (see Methods for definitions). Models were run on a sample of 50 phylogenetic trees extracted from www.birdtree.org [50], grafted to the Prum and colleagues [61] genomic backbone. Dots show mean effect size estimates from 12,500 posterior draws. For each effect, broad bases of whiskers show 66% credible intervals; narrow tips of whiskers show 95% credible intervals. Coloured distributions indicate the spread of effect size estimates, generated from a sample of 1,000 posterior draws. Full statistical results are presented in Table F in S2 Text. The data underlying this figure can be found at https://doi.org/10.6084/m9.figshare.27255609.