Universal prediction of vertebrate species age at maturity

Animal age at maturity can be used as a universal and simple predictor of species extinction risk. At present, methods to estimate age at maturity are typically species-specific, limiting comparisons among species, or are infeasible due to practical constraints. To overcome this, here we develop a universal predictor of species-level age at maturity for vertebrates. We show that modelling the frequency of 'CG' sequences (CpG sites) in gene promoter regions yields rapid predictions of vertebrate age at maturity. Our models predict age at maturity with remarkable accuracy and generalisability, with median error rates of 30% (less than 1 year) and are robust to genome assemblies of varying quality. We generate predictions for 1912 vertebrate species for which age at maturity estimates were previously absent from public databases. The predictions can be used to help to inform management decisions for the many species for which more detailed population information is currently unavailable. Earth's contemporary biodiversity crisis creates, in turn, an environmental management crisis. With more than one-fifth of all vertebrates under threat, efficient approaches to identify the most vulnerable species are urgently required. Approaches that enable rapid triage of the tens of thousands of species with unknown conservation status will allow for the best use of our limited wildlife management resources. Animal age at maturity is a universal predictor of extinction risk that could prove to be a tractable and rapid assessment tool for the many species lacking more precise information. Time to reach maturity determines both generation time and reproductive output and, thus, is a major driver of population growth rate in vertebrates. Despite its potential utility in conservation evaluations, age at maturity is difficult to measure in practice, particularly for species that are long-lived, rare, cryptic and/or have complex life histories. As a result, most methods to estimate age at maturity are indirect. Example techniques include predictive modelling based on physiological data in catsharks (Scyliorhinus canicula), a combination of mark-recapture, genotypic and geneology analyses in hawksbill turtles (Eretmochelys imbricata), or the measurement of steroid hormones in whale faeces. To enable the effective use of age at maturity as a comparative estimator of extinction risk among vertebrates, a universally applicable predictor of species age at maturity is required. The frequency of CpG sites in gene promoters has recently been shown to predict age at maturity in mammals as well as lifespan in mammals, fish, and other vertebrates. In vertebrates, gene promoters have distinct CpG distribution patterns that are highly predictive of epigenetic regulatory factors, such as DNA methylation and histone modification, and can be used as an indicator of gene expression and regulation. This ability to signify stable, dynamic, or repressed gene expression may explain CpG content's remarkable ability to predict species life history parameters. Here we use promoter CpG content to develop a universal method to predict species-level age at maturity in vertebrates. We include all species with suitable, publicly available genomic and age at maturity data to train and test the model (n = 1359 species). We evaluate improvements in prediction accuracy gained by developing group-specific models; one each for fish (n = 331), mammals (n = 550), and reptiles (including birds; n = 461). Promoter sequences are retrieved using homology to experimentally derived promoter sequences from humans (Homo sapiens) for the all-vertebrate and mammal-specific models, zebrafish (Danio rerio) for the fish-specific model and chicken (Gallus gallus) for the reptile model. Promoter CpG content is measured as CpG observed/expected ratio (CpG O/E; see ref. 17) and modelled with genomic GC percent and species order to account for any inherent differences in GC content. Prediction intervals are estimated using well-established uncertainty quantification methods. We then use genome sequence data to provide predictions and prediction intervals for 1912 species with previously unreported ages at maturity. The method is broadly applicable and will predict age at maturity for any vertebrate species for which whole genome sequence data can be obtained.

Alyssa M. Budd, Suk Yee Yong, Matthew J. Heydenrych, Benjamin Mayne, Oliver Berry, Simon Jarman