Morphological selectivity of the Permian-Triassic ammonoid mass extinction

Can we predict which kinds of organisms are more likely to survive extinctions? The answer would be quite different with regard to background and mass extinctions. For example, taxa with wide geographic ranges can buffer against background extinction, but not for mass extinctions (Dunhill and Wills, 2015). Mass extinctions in geological history usually act as a random nonselective pattern, affecting different clades simultaneously and globally (Payne and Finnegan, 2007). However, some clades with specific ecological traits show higher resilience to mass extinctions than others; for example, physiologically buffered groups were less affected by the Permian-Triassic mass extinction (PTME) than unbuffered groups (Bambach et al., 2002). Greater understanding of the selectivity of mass extinctions can shed light on extinction mechanisms and predict the extinction risk of living species.

The PTME was the most severe crisis in the Phanerozoic; it killed more than 80% of marine species (Stanley, 2016; Fan et al., 2020). It coincided with large environmental and climatic upheavals; e.g., global warming and oceanic anoxia (e.g., Song et al., 2012; Sun et al., 2012). Therefore, the PTME can provide an excellent deep-time experiment for testing selective extinction patterns. The PTME affected both taxa with wide and narrow geographic ranges without selectivity (Payne and Finnegan, 2007; Dai and Song, 2020); it did not show obvious selectivity on bivalves with different lifestyles, feeding types, or habitats (Huang et al., 2014). However, the PTME exhibited strong selectivity on other specific groups and ecological traits; e.g., marine invertebrates had much higher extinction rates than chondrichthyan fishes (Vázquez and Clapham, 2017). The degree of selectivity was thus highly variable among different clades, ecological traits, and geographic ranges.

Ammonoids suffered a diversity bottleneck during the PTME, with the survival of only three lineages, episageceratids, xenodiscids, and otoceratids (Wiedmann, 1973; Teichert and Rochester, 1986; Brayard et al., 2009). The PTME was thought to be nonselective in terms of ammonoid conch morphology, because geometric ratios of ammonoid conchs showed little change across the PTME (Villier and Korn, 2004; Korn et al., 2013; Monnet et al., 2015). However, whether selectivity occurred on other morphological characters, e.g., shell ornamentation and aperture shape, was not clear. Shell ornamentation and aperture shape are important morphological parameters because they may significantly affect hydrodynamic properties (Chamberlain and Westermann, 1976; Hebdon et al., 2020). In our study, we performed morphometric analyses based on a comprehensive morphological data set encompassing 127 ammonoid species to quantify the morphological evolution of ammo-noids across the Permian-Triassic boundary and test the morphological selectivity of the PTME.

We compiled a discrete morphological data set from published literature that contained 87 Changhsingian and 40 Griesbachian ammonoid species (Table S1 in the Supplemental Material¹), covering all valid genera known in this interval. Taxa in open nomenclature or with problematic classification were also included when they had a distinct morphology. We portioned the Griesbachian ammonoids into two groups: survivors and newcomers. Survivors were defined as initial postextinction species belonging to the surviving families and superfamilies; i.e., Otocerataceae, Episageceratidae, and Xenodiscidae (Wiedmann, 1973; Teichert and Rochester, 1986). Newcomers were all derived species and representatives of the newly originated families after the PTME; i.e., Ophiceratidae, Proptychidae, Mullericeratidae, and a family incertae sedis species Anotoceras nala.

Morphological characteristics employed in our study were based on the analysis by McGowan and Smith (2007) with some modifications (see Table S2). These characteristics encompass conch geometry, ornamentation, shape of the aperture and ontogeny (Fig. 1). Size was not included in this analysis because ammonoid shells were usually damaged during postmortem transportation and taphonomic processes. We usually coded the studied species based on its holotype, when it was well preserved. Non-holotype specimens were used in case of incomplete preservation of the holo-type. Intraspecific variation was not considered here owing to three reasons: (1) most of the studied species were described on the basis of a few specimens; (2) adding more specimens of well-sampled species will cause strong sampling biases, which will misrepresent the centroid of related morphospace; and (3) intraspecific variation is usually less than interspecific variation (Tanabe et al., 2003), and thus has little effect on the holistic morphospace.

We adopted the extinction space method (Korn et al., 2013) to quantify the morphological evolution of ammonoids during the PTME. This method is based on a two-dimensional morpho-space. It can be quantified by three parameters: sum of range (SOR), sum of variance (SOV), and position of centroid (POC). Three theoretical modes of extinction can lead to reductions in morphological disparity: random, marginal, and lateral extinctions. At the same extinction magnitude, SOR and SOV will show the smallest decrease in random extinction mode, an intermediate decrease in lateral extinction mode, and the largest reduction in marginal extinction mode; POC will not show prominent changes in random and marginal modes, but it will exhibit remarkable change in lateral mode.

We used the nonmetric multidimensional scaling (NMDS) method to construct an ammo-noid morphospace, because this is widely adopted for analyzing discrete morphological data. Two distance metrics (i.e., Euclidean and Gower) were used to calculate pairwise distances in PAST 3.0 software (Hammer et al., 2001). The performance of NMDS based on the Euclidean distance matrix (stress value = 0.157) was better than the Gower distance matrix (stress value = 0.216). Our analysis was therefore mainly based on the Euclidean distance matrix. Morphological diversity was calculated as the SOV and the SOR on the two NMDS axes. The bootstrap method was adopted to test the effect of sampling biases for SOV and SOR. We set a null hypothesis simulation wherein, given a random extinction of species with 10,000 replicates, the number of survivors is the same as the number of empirical survivors. Then, we compared the change of the centroid positions between the null hypothesis and empirical survivors. The distance between the centroids was calculated by Euclidean distance using their two NMDS coordinates. The simulations were performed using R version 3.5.3 (https://cran.r-project.org/bin/windows/base/old/3.5.3/).

The morphospace of the Changhsingian and Griesbachian ammonoids showed signifi-cant differences (Fig. 2A). The one-way permutational multivariate analysis of variance (PERMANOVA) test of their morphological characters also showed significant differences (p << 0.001). The Changhsingian ammonoids are mainly concentrated on the left side, represented by coarsely ornamented species, while the Griesbachian ammonoids are mainly clustered on the right side, dominated by smooth forms. Both the SOR and SOV of the Griesbachian ammonoids are smaller than the SOR and SOV of the Changhsingian ammonoids (Figs. 3A and 3B).

Compared with the overall Changhsingian ammonoid morphospace, the morphospace of the survivors is significantly reduced (Fig. 2C), reflecting differences in their morphological characters (one-way PERMANOVA test, p = 0.01). Survivors predominately belong to smooth or weakly ornamented morphotypes. The SOV of survivors does not decrease remarkably, because the survivors belong to morphologically distinct clades. Their SOR shows a significant reduction (Figs. 3A and 3B). Newcomers are mainly smooth and compressed forms, when compared with Changhsingian ammonoids (one-way PERMANOVA test, p << 0.001; Fig. 2C). SOR and SOV of newcomers are very low (Figs. 3A and 3B).

The centroid of the Griesbachian ammonoids is far away from the Changhsingian ammonoid centroid (Fig. 2A), and their distance in the morphospace is ∼0.072, which is significantly larger than the null hypothesis simulation (p << 0.001; Fig. 2B). The survivors also have a significantly distant centroid (∼0.047) compared to the Changhsingian species (p = 0.031; Fig. 2D). Therefore, a nonselective extinction mode is rejected. Our new results support a lateral selective extinction.

Our results reveal a lateral morphological selective mode for the PTME. Species with weak ornamentation or smooth shells were more likely to survive. Villier and Korn (2004) did not detect remarkable differences between the morphospaces of Changhsingian and Griesbachian ammonoid conchs; their results suggested that the postextinction assemblage shows slightly increased SOR and SOV. Subsequent works showed that Griesbachian ammonoids had low diversity but high disparity (McGowan, 2004; Brosse et al., 2013). In contrast to previous studies, our new results reveal that the survivors and all Griesbachian ammonoid species exhibit significantly lower disparity than those in the Changhsingian. Suture lines were not included in this study, but previous studies documented a significant drop in mean suture line complexity, associated with eliminations of the most complex and simplest suture lines during the PTME (Saunders et al., 1999). This extinction modality in ammonoid suture lines fits with the marginal selective extinction mode (Korn et al., 2013).

If we compare all Griesbachian species or species in surviving families/superfamilies with Changhsingian species to discuss selectivity, some postextinction species are involved in the analysis. This does influence the discussion of selectivity, since selective extinction only relates to pre-extinction taxa. Another potential problem is the huge magnitude of the PTME, which killed nearly all ammonoids, with only three lineages surviving (Wiedmann, 1973; Teichert and Rochester, 1986; Brayard et al., 2009). For a maximal reduction of the recovery effect and fitting with the huge magnitude of the PTME, we used three species, the oldest postextinction representative of each surviving lineage, i.e., Otoceras concavum, Episageceras dalailamae, and Hypophiceras triviale, for a test. The distance between their centroid and the centroid of the Changhsingian species was larger than the mean shift of random survival of three species (p = 0.084; Figs. 3C and 3D). Therefore, a selective extinction model is supported at species, family/superfamily, and whole-stage levels.

To test the robustness of our analysis, we used an alternative distance metric (Gower distance) to rerun the same analysis. Except for more overlap between the morphospaces of survivors and newcomers, the results based on the Gower distance were very similar to the Euclidean distance (Fig. 4A). The significantly selective signal also exists in this analysis (p = 0.002; Fig. 4B).

In our data set, the two species O. concavum and H. triviale have uncertain age constraints. Their occurrences are probably younger than the diverse Changhsingian assemblages from South China and Transcaucasia but older than the typical earliest Griesbachian assemblage. Unfortunately, their stratigraphic correlation with the conodont biostratigraphy is not highly resolved (Orchard and Tozer, 1997). Regardless of whether they were treated as Changhsingian or Griesbachian species, the results were very similar (Figs. S1 and S2).

The morphological shift from the Chang-hsingian coarsely ornamented ammonoids to the Griesbachian smooth ammonoids may reflect an ecological turnover. The relationship between ammonoid conch morphology and their life mode has received a lot of attention (Chamberlain and Westermann, 1976; Jacobs, 1992; Klug et al., 2016; Hebdon et al., 2020). However, due to the lack of preservation of their soft body, our understanding of their ecology is still incomplete (Naglik et al., 2015). The loss of coarsely ornamented ammonoids during the PTME suggests that they were more vulnerable to environmental perturbations, e.g., warming and anoxia (e.g., Sun et al., 2012; Song et al., 2012). Another possible explanation is phenotypic response to environmental stress, but clear evidence is lacking. Changhsingian ammonoids from Iran exhibited ornament simplification and size reduction prior to the PTME (Kiessling et al., 2018), which may suggest that being smooth can be an effective survival strategy under environmental stress. It is noted that more data are needed to test this pre-PTME morphological change. During the late Smithian crisis, which was a secondary extinction event in the Triassic (Stanley, 2009), spherical ammo-noids disappeared (Brosse et al., 2013). The loss of spherical forms also happened during the Triassic-Jurassic mass extinction (Smith et al., 2014). Ammonoids show complex morphological responses to different extinction events. Our work can stimulate further studies on selective extinction and morphological changes under environmental stress.

In contrast to a previously proposed morphological nonselective extinction mode, our new analyses based on a comprehensive ammonoid morphological data set reveal that the PTME was selective in ammonoid morphology. Weakly ornamented and smooth forms were more likely to survive. Changhsingian species usually had coarse ornamentations, while smooth species dominated the Griesbachian. This morphological shift probably indicates an ecological turnover within ammonoids near the Permian-Triassic boundary, caused by a series of environmental perturbations; e.g., global warming and anoxia.

We acknowledge Michael J. Benton and Li Tian for teaching morphometric methods during the short course in Wuhan (China) in 2019. We thank Kathleen Ritterbush, Matthew E. Clapham, Christian Klug, and an anonymous reviewer for their constructive suggestions. This research was funded by the National Natural Science Foundation of China (grant 41821001), the State Key R&D Project of China (grant 2016YFA0601100), the Strategic Priority Research Program of Chinese Academy of Sciences (grant XDB26000000), and the Fundamental Research Funds for National University, China University of Geosciences (Wuhan), and the Deutsche Forschungsgemeinschaft (Ko1829/18–1 and FOR 2332). This is Center for Computational and Modeling Geosciences (BGEG) publication 4.