Heterogeneous morphological evolution during the Great Dying

Using a deep learning method that extracts morphological features from images of marine fossils, we explore morphological disparity dynamics over a time series of 4 million years, spanning the Permian–Triassic mass extinction event.
Published in Ecology & Evolution
Heterogeneous morphological evolution during the Great Dying
Like

Share this post

Choose a social network to share with, or copy the shortened URL to share elsewhere

This is a representation of how your post may appear on social media. The actual post will vary between social networks

Overview

Morphology provides insights into biotic evolutionary processes, and also explicit analysis of biological forms allows documentation and interpretation of the evolution, success, and demise of fossil groups. Hence, morphological study has been a hot topic over the last several decades. Disparity or morphological diversity is one of the most intuitive ways to quantify morphological variations in a sample of taxa and has been fundamental to understanding evolutionary processes in deep time. Furthermore, morphological disparity and taxonomic diversity are distinct measures of biodiversity, typically expected to evolve synergistically. However, evidence from mass extinctions leads to a drastic loss of taxonomic diversity, and I have been puzzled by how morphological evolution progresses during critical periods. Current studies are limited, and sometimes show controversial results due to different approaches (Villier and Korn, 2004; Dai et al., 2021). Therefore, a more comprehensive method needs to be developed to consistently investigate more clades during mass extinction events.

Figure 1. Schematic illustration of the DeepMorph that automatically extracts morphological traits for each clade.

Using DeepMorph to extract fossil features from images

The work on this study started in 2022, when I had some experience with deep learning methods. I started trying to use deep learning to study the morphology of fossils. I investigated and trained many deep convolutional neural network models, including supervised and unsupervised approaches. Finally, I developed an analytical pipeline, termed DeepMorph (Fig. 1), that integrates deep learning techniques for fossil feature extraction from images with geometric morphometrics. The DeepMorph allows us to test the signatures of morphological extinction and its selectivity of different clades under a consistent quantification perspective. Therefore, we compiled a comprehensive database of fossil specimen images from six extensively documented marine clades across the Permian-Triassic mass extinction (PTME), namely, ammonoids, bivalves, articulated brachiopods, gastropods, ostracods, and conodonts. Our dataset includes 599 genera (covering 83% of genera recorded in the Paleobiology Database, Fig. 2) represented by 656 images, spanning about 3.44 Myr (i.e., 254.14 Ma – 250.7 Ma). Besides, we also used simulations to test whether they are consistent with several selective extinction patterns.

Figure 2. Generic diversity distribution of clades and its coverage compared to the paleobiology database (PBDB).

Morphological evolution and selectivity

Our DeepMorph approach automates the analysis of fossil specimen contours, effectively capturing main variations in morphologies, leading to a two-dimensional embedding that distinctly separates various morphotypes. This embedding offers an approximation of morphospace, aligning with observations derived from traditional morphological analyses. We used the sum of variances (SOV) to quantify the morphological disparity, which calculates the sum of the variances along the morphospace axes. Morphological analysis revealed substantial disparity loss (-17%–-76%) in five clades across the PTME, except for conodonts (+36%, Fig. 3). Besides, an exciting finding is that null hypothesis test revealed four distinct morphological evolution patterns (Fig. 4): lateral selective extinction from ammonoids, marginal selective extinction from brachiopods and ostracods, non-selective extinction from bivalves and gastropods, and negligible morphological extinction from conodonts.

Figure 3. Morphological evolution of six clades across PTME.

Clades that suffered a selective extinction demonstrated higher loss of complex and ornamented forms, while clades that did not experience morphologically selective extinctions preferred to maintain their morphspace (Fig. 3d and e). The presence and intensity of morphological selectivity likely reflect the variations in environmental tolerance thresholds among different clades (Song et al., 2024). Our results highlight that the PTME had heterogeneous morphological selective impacts across clades, offering new insights into how mass extinctions can reshape biodiversity and ecosystem structure.

Figure 4. Four distinct patterns of morphological evolution were identified during the PTME. a, Lateral selective extinction from ammonoids. b, Marginal selective extinction, including brachiopods and ostracods. c, Non-selective extinction, containing bivalve and gastropod. d, Negligible morphological extinction from conodonts.

Increasing evidence that some morphological traits such as size and complexity are linked to higher extinction rates in both ancient and modern events (Payne et al., 2016), suggests that morphological studies of fossils can help to assess modern extinction risks (Raja et al., 2021). Besides, DeepMorph and other deep learning methods demonstrated the powerful capability of automatically analyzing large-scale data with high efficiency, and there will be more interdisciplinary research between deep learning and evolutionary paleobiology.

 

References

Dai, X., Korn, D. & Song, H. Morphological selectivity of the Permian-Triassic ammonoid mass extinction. Geology 49, 1112–1116 (2021).

Payne, J. L., Bush, A. M., Heim, N. A., Knope, M. L. & McCauley, D. J. Ecological selectivity of the emerging mass extinction in the oceans. Science 353, 1284–1286 (2016).

Raja, N. B. et al. Morphological traits of reef corals predict extinction risk but not conservation status. Glob. Ecol. Biogeogr. 30, 1597–1608 (2021).

Song, H. et al. Respiratory protein-driven selectivity during the Permian-Triassic mass extinction. The Innovation 5, 100618 (2024).

Villier, L. & Korn, D. Morphological disparity of ammonoids and the mark of Permian mass extinctions. Science 306, 264–266 (2004).

Please sign in or register for FREE

If you are a registered user on Research Communities by Springer Nature, please sign in

Follow the Topic

Paleontology
Life Sciences > Biological Sciences > Evolutionary Biology > Paleontology
Evolutionary Biology
Life Sciences > Biological Sciences > Evolutionary Biology