henhaobio
Sign in

bioRxivPaleontologyPreprint

Locomotor Trends Among Early Mammals Illuminated by Predictors of Arboreality from the Mammalian Appendicular Skeleton

Jonathan A. NationsiD, Lucas N. WeaveriD, David M. GrossnickleiD

PDFDOICodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

Arboreal locomotion is a common trait among extant mammals, and it likely evolved independently in numerous Mesozoic and Cenozoic mammalian lineages. We review evidence for the evolution of arboreality within lineages of early mammals and their stem relatives (Synapsida) while also emphasizing the uncertainty of locomotor inferences of fossil lineages. An improved understanding of the evolutionary origins and prevalence of arboreality in early mammals requires robust links between morphological traits and locomotor behaviors that can be applied to fossils. To this end, we use a large comparative dataset of linear measurements taken on extant mammalian postcranial specimens to evaluate the skeletal proxies that best differentiate arboreality from other locomotor modes. Consistent with previous research, we find that measures of digit length and elbow proportions are especially powerful predictors of climbing. Locomotor predictions based on single measurements or ratios are improved when phylogeny is incorporated into models. Nonetheless, combining multiple functionally relevant traits into our models yields robust predictions of locomotor mode even without incorporating phylogenetic information--this is especially encouraging for fossil groups with unstable phylogenies. We test these locomotor proxies in three early-mammalian case studies--Multituberculata, Plesiadapiformes, and early Theria. We find that multituberculates had a wide locomotor breadth whereas plesiadapiforms all had a high probability of climbing, consistent with previous studies. However, some Mesozoic therians in our sample are predicted to be ground-dwellers, conflicting with the hypothesis that the ancestral therian was scansorial or arboreal. The predictive tools presented here will improve locomotor inferences for small mammals, further illuminating the prevalence and importance of arboreality in early mammalian evolution.

In-depth summary

Claude Opus 5.5 · xhigh effort · 2 Nov 2026 · from the full text of version 1 (14 Jul 2025) · CC BY-NC-ND 4.0

Bayesian multilevel models of 69 appendicular skeletal metrics across 236 extant mammal species predict climbing at 95% accuracy and indicate that several Mesozoic therians, including Juramaia, were terrestrial ground-dwellers.

Inferring arboreal locomotion in fragmentary Mesozoic mammal fossils is complicated by small body sizes, unstable phylogenies, and reliance on single traits without accounting for phylogenetic covariance and intraspecific variation. Evaluating 69 appendicular metrics across 236 extant mammalian species with Bayesian multilevel regression identifies the strongest skeletal predictors of climbing and enables multi-trait fossil predictions without phylogenetic placement.

Key results

  1. ShownCombining the 14 most accurate appendicular measurements and functional indices in a Bayesian multiple regression model predicts mammalian climbing with ~95% accuracy even without phylogenetic information.

    Across 427 specimens from 236 (or 237) extant species, single-metric binary predictions reach 94-100% accuracy with the phylogenetic correlation matrix included but drop to 50-85% without phylogeny; combining the top 14 metrics restores binary prediction accuracy to ~95% without phylogenetic input (and >90% across 9 of 11 fossil subset models). · 3.5.1 Testing for Overfitting and Accuracy; 3.6.1 The Effect of Phylogenetic Correlations on…; 3.6.3 Multiple Regression Greatly Improves the…; 5.2 Predictions of Fossil Mammal Locomotion

  2. ShownManual and pedal digit proportions, along with proximal humerus width and ulnar olecranon process metrics, are the strongest individual predictors of climbing when phylogeny is excluded.

    Without phylogenetic information, the pedal digit length ratio PES2 achieves the highest single-metric binary accuracy at 85%, while MANUS, PES, and the humerus proximal width ratio (HPI) each exceed 80% binary accuracy and achieve >55% exact-rank (>80% within one rank) ordinal accuracy. · 3.4.2 Binary Model Results; 3.5.1 Testing for Overfitting and Accuracy; 4.1 Which Measurements Best Predict Climbing in…; 4.1.1 Pedal Metrics Predict Climbing as Well as…

  3. ShownMultiple regression predictions for 14 fossil mammal taxa classify three of five Mesozoic therians (including Jurassic Juramaia) as terrestrial, challenging the consensus that early therians were ancestrally scansorial or arboreal.

    In ordinal and binary multiple regression models across 5 Mesozoic therians, Eomaia scansoria and Sinodelphys szalayi are predicted as scansorial or arboreal, whereas Juramaia sinensis, Ambolestes zhoui, and Cokotherium jiufotangensis have high probabilities of being terrestrial non-climbers. · 5.1 Sampling of Fossil Mammals; 5.2.1 Results of Binary Fossil Predictions; 5.2.2 Results of Ordinal Fossil Predictions; 6.1 Were Therians Ancestrally Ground-Dwellers?

  4. ShownApplied to Multituberculata and Plesiadapiformes, multi-trait models predict arboreality across Paleocene plesiadapiforms and Ptilodus kummae, but predominantly terrestrial habits among Mesozoic multituberculates.

    Across 4 middle-late Paleocene plesiadapiforms (Plesiadapis cookei, Ignacius clarkforkensis, Torrejonia wilsoni, and Dryomomys szalayi), binary multiple regression predicts climbing for all four (with Dryomomys split between terrestrial and scansorial in ordinal models); among 5 multituberculates, Paleocene Ptilodus is strongly predicted as arboreal whereas Mesozoic taxa are mostly predicted as terrestrial. · 5.1 Sampling of Fossil Mammals; 5.2.1 Results of Binary Fossil Predictions; 5.2.2 Results of Ordinal Fossil Predictions; 6.2 Many Mesozoic Paleoenvironments Promoted…

Methods, data and reproducibility

Compiled 26 linear appendicular measurements across 427 specimens of 236 (or 237) extant mammal species to derive 69 metrics (26 log-linear traits, 17 functional indices, and 26 log-shape ratios). Fitted Bayesian multilevel binary and 4-rank ordinal regression models in brms/Stan with species and phylogenetic group-level effects and missing-data imputation, then predicted locomotion for 14 fossil taxa.

  • Linear appendicular skeletal measurements (49 compiled, 26 retained with <50% missing data) and body mass across extant non-volant terrestrial/amphibious mammals from 70 families in 21 orders · multiple species · n = 427 specimens of 236 (or 237) species (2 monotreme, 72 marsupial, 163 placental) · 10.5281/zenodo.10144438
  • Binary locomotor dataset ('climb' vs 'no climb') of extant mammal species · multiple species · n = 95 climb species (150 specimens), 142 no-climb species (277 specimens)
  • Ordinal locomotor dataset (Inter-substrate, Terrestrial, Scansorial, Arboreal) of extant mammal species · multiple species · n = 65 inter-substrate, 77 terrestrial, 25 scansorial, 70 arboreal species
  • Appendicular skeletal measurements of Mesozoic and Paleocene fossil mammals (5 Mesozoic Theria, 5 Multituberculata, 4 Plesiadapiformes) from literature and ImageJ photographs · multiple species · n = 14 fossil species
Methods & tools
R, brms (v2.19.6), Stan, ImageJ, panTheria, EltonTraits, Animal Diversity Web
Software & environment
R with brms package v2.19.6 and Stan (4 chains, 2,000 warm-up and 2,000 sampling iterations per model), ImageJ.
Limitations (4)
  • Categorizing continuous locomotor behaviors into discrete binary or four-level ordinal ranks is a simplification, and defining scansoriality (~10% of the dataset; 25 species, 42 specimens) is subjective due to incomplete natural history records. · stated by the authors
  • Multiple regression models containing up to 14 correlated morphological predictors sacrifice biological interpretability of individual skeletal traits in exchange for higher predictive accuracy. · stated by the authors
  • Incomplete preservation of fossil skeletons required excluding log-shape ratios and geometric-mean size predictors and fitting separate subset multiple regression models (ranging from 2 to 14 metrics, such as only Ppl and Uol for Torrejonia wilsoni). · stated by the authors
  • Because the sampled Mesozoic therians are often classified as early eutherians and complete early metatherian skeletons are lacking in the fossil record, the models cannot definitively distinguish whether crown Theria as a whole or only Eutheria were ancestrally terrestrial. · stated by the authors

Read this if you study mammalian evolutionary morphology, paleobiology, or phylogenetic comparative methods and want a Bayesian framework for predicting locomotor ecology from postcranial measurements.

Citations
3

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISAlfieri '26Toussaint '25Amson '18Rickman '23Brocklehurs '24Huizenga '26
72% semantic matchPaleontology
Fabio Alfieri, Julia Arias‐Martorell, Carla Argilés-Esturgó+1 more·Zoological Journal of the Linnean Society·Jan 13, 2026· 3 citations