A 450 million years long latitudinal gradient in age-dependent extinction

The fate of South America's endemic mammalian fauna in response to the most dramatic Cenozoic climate disruption

Around 34 Mya, the Eocene-Oligocene transition (EOT) marked the most dramatic global climatic cooling of the Cenozoic. On a planetary scale, paleontological evidence suggests that this transition was associated with major faunal turnovers, sometimes even regarded as a mass extinction crisis. In South America, there is no consensus on the response of the endemic mammals to this transition. Here, using a vetted fossil dataset and cutting-edge Bayesian methods, we analyzed the dynamics of South American mammal (SAM) diversification and their possible drivers across latitude (tropical vs. extratropical), taxonomic groups, and trophic guilds throughout the Eocene-Oligocene (ca. 56 to 23 Ma). Our results did not evidence any mass extinction among SAM at the EOT. Instead, they experienced a gradual and long-term diversity decline from the middle Eocene to the early Oligocene, followed by a sudden waxing-and-waning diversity associated with a large taxonomic-but not ecological-turnover. Tropical and extratropical lineages have had very distinct macroevolutionary histories. No effective change in the pace at which tropical lineages diversify was found, thus favoring the tropical stability hypothesis proposed by Wallace. Diversity-dependent effects, temperature, and Andean uplift were recovered as probable drivers of SAM diversification across the period. Contrasting evidence casts doubt on the common hypothesis primarily linking Oligocene faunal changes to grassland expansion. Our findings illustrate the uniqueness of the deep-time interplay between endemic SAM and their physical environment in a context of climatic shift, highlighting the need to consider regional idiosyncrasies for understanding the coevolution of life and climate.

EutherianCoP. An integrated biotic and climate database for conservation paleobiology based on eutherian mammals

We present a new database, EutherianCoP, of fossil mammals which lived globally from the Late Pleistocene to the Holocene. The database includes 13,972 fossil occurrences of 786 extant or recently extinct placental mammal species, plus 155,198 current occurrences for those of them which survived to the present. The occurrences are correlated with radiometric age information. For all species, we provide 32 different traits, inclusive of taxonomic, phenotypic, life history, biogeographic and phylogenetic information. Differently from any other compilation, the occurrences are complemented with estimates of past climatic conditions, including site-interpolated monthly and annual precipitation and temperature, leaf area index, megabiome type and net primary productivity, which are derived from transient paleo model simulations conducted with the Community Earth System Model 1.2 and the BIOME4 vegetation model. All data are further downloadable for further investigation.

Toward a Semi-Supervised Learning Approach to Phylogenetic Estimation

Models have always been central to inferring molecular evolution and to reconstructing phylogenetic trees. Their use typically involves the development of a mechanistic framework reflecting our understanding of the underlying biological processes, such as nucleotide substitutions, and the estimation of model parameters by maximum likelihood or Bayesian inference. However, deriving and optimizing the likelihood of the data is not always possible under complex evolutionary scenarios or even tractable for large datasets, often leading to unrealistic simplifying assumptions in the fitted models. To overcome this issue, we coupled stochastic simulations of genome evolution with a new supervised deep-learning model to infer key parameters of molecular evolution. Our model is designed to directly analyze multiple sequence alignments and estimate per-site evolutionary rates and divergence without requiring a known phylogenetic tree. The accuracy of our predictions matched that of likelihood-based phylogenetic inference when rate heterogeneity followed a simple gamma distribution, but it strongly exceeded it under more complex patterns of rate variation, such as codon models. Our approach is highly scalable and can be efficiently applied to genomic data, as we showed on a dataset of 26 million nucleotides from the clownfish clade. Our simulations also showed that the integration of per-site rates obtained by deep learning within a Bayesian framework led to significantly more accurate phylogenetic inference, particularly with respect to the estimated branch lengths. We thus propose that future advancements in phylogenetic analysis will benefit from a semi-supervised learning approach that combines deep-learning estimation of substitution rates, which allows for more flexible models of rate variation, and probabilistic inference of the phylogenetic tree, which guarantees interpretability and a rigorous assessment of statistical support.

Challenges in estimating species' age from phylogenetic trees

Aim

Species age, the elapsed time since origination, can give insight into how species longevity might influence eco-evolutionary dynamics, which has been hypothesized to influence extinction risk. Traditionally, species' ages have been estimated from fossil records. However, numerous studies have recently used the branch lengths of time-calibrated phylogenies as estimates of the ages of extant species. This approach poses problems because phylogenetic trees only contain direct information about species identity at the tips and not along the branches. Here, we show that incomplete taxon sampling, extinction, and different assumptions about speciation modes can significantly alter the relationship between true species age and phylogenetic branch lengths, leading to high error rates. We found that these biases can lead to erroneous interpretations of eco-evolutionary patterns derived from comparing phylogenetic age and other traits, such as extinction risk.

Innovation

For bifurcating speciation, the default assumption in most analyses of species age, we propose a probabilistic approach based on the properties of a birth-death process to improve the estimation of species ages. Our approach can reduce the error by one order of magnitude under cases of high extinction and a high percentage of unsampled extant species.

Main conclusion

Our results call for caution in interpreting the relationship between phylogenetic ages and eco-evolutionary traits, as this can lead to biased and erroneous conclusions. We show that, under the assumption of bifurcating speciation, we can obtain unbiased approximations of species age by combining information from branch lengths with the expectations of a birth-death process.

Trait-mediated speciation and human-driven extinctions in proboscideans revealed by unsupervised Bayesian neural networks

Species life-history traits, paleoenvironment, and biotic interactions likely influence speciation and extinction rates, affecting species richness over time. Birth-death models inferring the impact of these factors typically assume monotonic relationships between single predictors and rates, limiting our ability to assess more complex effects and their relative importance and interaction. We introduce a Bayesian birth-death model using unsupervised neural networks to explore multifactorial and nonlinear effects on speciation and extinction rates using fossil data. It infers lineage- and time-specific rates and disentangles predictor effects and importance through explainable artificial intelligence techniques. Analysis of the proboscidean fossil record revealed speciation rates shaped by dietary flexibility and biogeographic events. The emergence of modern humans escalated extinction rates, causing recent diversity decline, while regional climate had a lesser impact. Our model paves the way for an improved understanding of the intricate dynamics shaping clade diversification.

The reorganization of predator-prey networks over 20 million years explains extinction patterns of mammalian carnivores

Linking the species interactions occurring at the scale of local communities to their potential impact at evolutionary timescales is challenging. Here, we used the high-resolution fossil record of mammals from the Iberian Peninsula to reconstruct a timeseries of trophic networks spanning more than 20 million years and asked whether predator-prey interactions affected regional extinction patterns. We found that, despite small changes in species richness, trophic networks showed long-term trends, gradually losing interactions and becoming sparser towards the present. This restructuring of the ecological networks was driven by the loss of medium-sized herbivores, which reduced prey availability for predators. The decrease in prey availability was associated with predator longevity, such that predators with less available prey had greater extinction risk. These results not only reveal long-term trends in network structure but suggest that prey species richness in ecological communities may shape large scale patterns of extinction and persistence among predators.

DeepDive: estimating global biodiversity patterns through time using deep learning

Understanding how biodiversity has changed through time is a central goal of evolutionary biology. However, estimates of past biodiversity are challenged by the inherent incompleteness of the fossil record, even when state-of-the-art statistical methods are applied to adjust estimates while correcting for sampling biases. Here we develop an approach based on stochastic simulations of biodiversity and a deep learning model to infer richness at global or regional scales through time while incorporating spatial, temporal and taxonomic sampling variation. Our method outperforms alternative approaches across simulated datasets, especially at large spatial scales, providing robust palaeodiversity estimates under a wide range of preservation scenarios. We apply our method on two empirical datasets of different taxonomic and temporal scope: the Permian-Triassic record of marine animals and the Cenozoic evolution of proboscideans. Our estimates provide a revised quantitative assessment of two mass extinctions in the marine record and reveal rapid diversification of proboscideans following their expansion out of Africa and a >70% diversity drop in the Pleistocene.

Global impact and selectivity of the Cretaceous-Paleogene mass extinction among sharks, skates, and rays

The Cretaceous-Paleogene event was the last mass extinction event, yet its impact and long-term effects on species-level marine vertebrate diversity remain largely uncharacterized. We quantified elasmobranch (sharks, skates, and rays) speciation, extinction, and ecological change resulting from the end-Cretaceous event using >3200 fossil occurrences and 675 species spanning the Late Cretaceous-Paleocene interval at global scale. Elasmobranchs declined by >62% at the Cretaceous-Paleogene boundary and did not fully recover in the Paleocene. The end-Cretaceous event triggered a heterogeneous pattern of extinction, with rays and durophagous species reaching the highest levels of extinction (>72%) and sharks and nondurophagous species being less affected. Taxa with large geographic ranges and/or those restricted to high-latitude settings show higher survival. The Cretaceous-Paleogene event drastically altered the evolutionary history of marine ecosystems.

Integrating geographic ranges across temporal scales

Geographic ranges are a fundamental unit of biogeography and macroecology. Increasingly, paleontologists and ecologists alike are reconstructing geographic ranges of species from fossils, in order to understand the long-term processes governing biogeographic and macroevolutionary patterns. As these reconstructions have become increasingly common, uncertainty has arisen over the equivalency of paleo-ranges and modern ranges. Here, we argue geographic ranges are time-averaged at all temporal scales, and reflect the biotic and abiotic processes operating across the equivalent range of time and space scales. This conceptual framework integrates the study of geographic ranges reconstructed using modern and ancient data, and highlights the potential for ranges to illuminate processes responsible for diversity patterns over intervals spanning days to tens of millions of years of Earth history.

Stronger predation intensity and impact on prey communities in the tropics

The hypothesis that biotic interactions strengthen toward lower latitudes provides a framework for linking community-scale processes with the macroecological scales that define our biosphere. Despite the importance of this hypothesis for understanding community assembly and ecosystem functioning, the extent to which interaction strength varies across latitude and the effects of this variation on natural communities remain unresolved. Predation in particular is central to ecological and evolutionary dynamics across the globe, yet very few studies explore both community-scale causes and outcomes of predation across latitude. Here we expand beyond prior studies to examine two important components of predation strength: intensity of predation (including multiple dimensions of the predator guild) and impact on prey community biomass and structure, providing one of the most comprehensive examinations of predator-prey interactions across latitude. Using standardized experiments, we tested the hypothesis that predation intensity and impact on prey communities were stronger at lower latitudes. We further assessed prey recruitment to evaluate the potential for this process to mediate predation effects. We used sessile marine invertebrate communities and their fish predators in nearshore environments as a model system, with experiments conducted at 12 sites in four regions spanning the tropics to the subarctic. Our results show clear support for an increase in both predation intensity and impact at lower relative to higher latitudes. The predator guild was more diverse at low latitudes, with higher predation rates, longer interaction durations, and larger predator body sizes, suggesting stronger predation intensity in the tropics. Predation also reduced prey biomass and altered prey composition at low latitudes, with no effects at high latitudes. Although recruitment rates were up to three orders of magnitude higher in the tropics than the subarctic, prey replacement through this process was insufficient to dampen completely the strong impacts of predators in the tropics. Our study provides a novel perspective on the biotic interaction hypothesis, suggesting that multiple components of the predator community likely contribute to predation intensity at low latitudes, with important consequences for the structure of prey communities.

gen3sis: A general engine for eco-evolutionary simulations of the processes that shape Earth's biodiversity

Understanding the origins of biodiversity has been an aspiration since the days of early naturalists. The immense complexity of ecological, evolutionary, and spatial processes, however, has made this goal elusive to this day. Computer models serve progress in many scientific fields, but in the fields of macroecology and macroevolution, eco-evolutionary models are comparatively less developed. We present a general, spatially explicit, eco-evolutionary engine with a modular implementation that enables the modeling of multiple macroecological and macroevolutionary processes and feedbacks across representative spatiotemporally dynamic landscapes. Modeled processes can include species' abiotic tolerances, biotic interactions, dispersal, speciation, and evolution of ecological traits. Commonly observed biodiversity patterns, such as α, β, and γ diversity, species ranges, ecological traits, and phylogenies, emerge as simulations proceed. As an illustration, we examine alternative hypotheses expected to have shaped the latitudinal diversity gradient (LDG) during the Earth's Cenozoic era. Our exploratory simulations simultaneously produce multiple realistic biodiversity patterns, such as the LDG, current species richness, and range size frequencies, as well as phylogenetic metrics. The model engine is open source and available as an R package, enabling future exploration of various landscapes and biological processes, while outputs can be linked with a variety of empirical biodiversity patterns. This work represents a key toward a numeric, interdisciplinary, and mechanistic understanding of the physical and biological processes that shape Earth's biodiversity.

Re-evaluation of the "law of constant extinction" for ruminants at different taxonomical scales

The "law of constant extinction," proposed by Van Valen, states that long and short-lived taxa have equal chances of going extinct. This pattern of age-independent extinction was originally inferred using the fossil record of several different taxa and relied on survivorship curves built from the literal reading of the fossil record. Van Valen's seminal work was mostly done at higher taxonomic levels, hence its prevalence at the species level could not be directly inferred. The surprisingly few subsequent studies done at the species level have challenged the prevalence of age-independent extinction, but those have, for the most part, failed to explicitly incorporate inherent biases of the fossil record. Using a recent Bayesian framework that accounts for several of those biases, including the fact that very short-living lineages might never make to the record itself, we showed that Ruminantia species present age-dependent extinction, where extinction probability decreases with species age. An analysis at the genus level suggested age-independent extinction but further examination suggested that the pattern might be more complex than previously reported by Van Valen. Our results indicate that different taxonomic levels may present different extinction regimes, which could justify the development of new macroevolutionary theory and methods.

Investigating Biotic Interactions in Deep Time

Recent renewed interest in using fossil data to understand how biotic interactions have shaped the evolution of life is challenging the widely held assumption that long-term climate changes are the primary drivers of biodiversity change. New approaches go beyond traditional richness and co-occurrence studies to explicitly model biotic interactions using data on fossil and modern biodiversity. Important developments in three primary areas of research include analysis of (i) macroevolutionary rates, (ii) the impacts of and recovery from extinction events, and (iii) how humans (Homo sapiens) affected interactions among non-human species. We present multiple lines of evidence for an important and measurable role of biotic interactions in shaping the evolution of communities and lineages on long timescales.

The past and future human impact on mammalian diversity

To understand the current biodiversity crisis, it is crucial to determine how humans have affected biodiversity in the past. However, the extent of human involvement in species extinctions from the Late Pleistocene onward remains contentious. Here, we apply Bayesian models to the fossil record to estimate how mammalian extinction rates have changed over the past 126,000 years, inferring specific times of rate increases. We specifically test the hypothesis of human-caused extinctions by using posterior predictive methods. We find that human population size is able to predict past extinctions with 96% accuracy. Predictors based on past climate, in contrast, perform no better than expected by chance, suggesting that climate had a negligible impact on global mammal extinctions. Based on current trends, we predict for the near future a rate escalation of unprecedented magnitude. Our results provide a comprehensive assessment of the human impact on past and predicted future extinctions of mammals.