Extinction cascades, community collapse, and recovery across a Mesozoic hyperthermal event

The study examines how species interactions shape extinction selectivity and post-extinction recovery during a Mesozoic hyperthermal event. Traditional perspectives emphasize abiotic drivers (Court Jester) and document selectivity by latitude and physiological sensitivity, but often neglect trophic interactions that can buffer or amplify extinction risk. Embedding interaction networks is essential both to explain high extinction among pelagic predators via cascading effects and to evaluate how Red Queen processes influence recovery of biodiversity, structure and function. Focusing on the Early Toarcian Extinction Event (ETEE; ~183 Ma), which globally eliminated ~26% of marine genera and caused severe local losses in the Cleveland Basin (about 60% of marine species, including ~87% of benthic species), the authors ask: Who died and why (which traits and guilds were primary targets), how did secondary extinctions propagate through food webs, and did biodiversity, structure and function recover synchronously? They also test whether recovery signatures include a shift towards modern marine ecosystem structure associated with the Mesozoic Marine Revolution.

Prior studies link many mass extinctions to Large Igneous Province volcanism with rapid warming, ocean anoxia and acidification, and report selectivity by latitude and vulnerability to hypoxia, hypercapnia and acidification. However, most research neglects biotic interactions, despite evidence that many extinctions occur via cascading secondary effects. Recovery assessments based on taxonomic and functional diversity suggest recovery times of <1 to 50 Myr, but such approaches may not capture community structure or function. Previous work on the Toarcian documents extensive benthic losses and dysoxia/anoxia in the Cleveland Basin, opportunistic post-extinction bivalve dominance, and protracted benthic recovery. Network theory indicates that connectance, generality, vulnerability, and motifs (competition, omnivory, chains) relate to ecosystem function and stability, while trophic redundancy can reduce vulnerability to cascades. Macroecological hypotheses (e.g., Mesozoic Marine Revolution) predict increasing predation, productivity, and vertical structure in marine food webs through the Jurassic-Cretaceous.

Data comprised 38,670 fossil occurrences of 162 marine species (bivalves, brachiopods, gastropods, echinoderms, crustaceans, cephalopods), plus fish and trace fossils, from the Pliensbachian–Toarcian of the Cleveland Basin (Yorkshire, UK). Occurrences were partitioned into four intervals treated as communities: pre-extinction, post-extinction, early recovery, late recovery. Species were assigned ecological traits following the Bambach ecospace framework: motility (fast, slow, facultative, non-motile), tiering (pelagic, erect/epifaunal, surficial, semi-infaunal, shallow infaunal, deep infaunal), feeding mode (predator, suspension feeder, deposit feeder, miner, grazer), and body size (tiny ≤10 mm to gigantic >500 mm). A single primary producer node was added to each web. The Paleo Foodweb Inference Model (PFIM), a trait-based inferential model implemented in R, reconstructed feasible food webs by applying deterministic interaction rules requiring compatibility across motility, tiering, feeding and size. Trophic guilds were defined by unique combinations of foraging traits; additional attributes (motility, tiering collapsed to pelagic/epifaunal/infaunal, size class, calcification level) were recorded per guild for simulations. Extinction cascades were simulated on the pre-extinction web under 13 primary extinction scenarios: random; body size (large→small, small→large); tiering (infaunal→pelagic, pelagic→infaunal); motility (fast→non-motile, non-motile→fast); calcification (heavy→light, light→heavy); generalism (low→high, high→low); vulnerability (low→high, high→low). Primary deletions proceeded within trait levels in randomized sequences; secondary extinctions occurred when consumers lost all prey (R cheddar RemoveNodes cascade). Simulations stopped when richness matched the empirical post-extinction community (21 species). For each scenario, 50 replicates were generated. Simulated post-extinction webs were compared to the empirical post-extinction web using: (1) nine structural metrics, (2) frequencies of four motifs (S1 linear chains, S2 omnivory, S4 apparent competition, S5 direct competition), and (3) True Skill Statistic (TSS) comparing guild-level identity and loss. Community structure and function across all four empirical webs were quantified using richness (guild richness), connectance, maximum trophic level (chain length), generality (mean in-degree), vulnerability (mean out-degree), and motif counts (S1, S2, S4, S5). Robustness was assessed via generalized Rx curves (x=1–99%) under 500 random primary deletion sequences per network, reporting mean proportion of the community remaining after primary plus secondary extinctions.

• Primary extinction targeting tiering (infaunal > pelagic) best reproduced the empirical post-extinction community across both TSS (highest match) and multi-metric comparisons. This aligns with dysoxia/anoxia as the principal kill mechanism preferentially affecting seabed-dwelling taxa.
• Secondary extinctions concentrated among benthic secondary consumers (e.g., crustaceans, echinoderms) after the loss of their primary consumer prey under infaunal- and epifaunal-targeted scenarios.
• Scenarios targeting consumer generalism (both high→low and low→high) provided the next best matches, consistent with losses of active benthic intermediates under low oxygen and sensitivity of specialists (e.g., suspension feeders) leading to cascades. Traits such as body size or calcification did not outperform random scenarios.
• Post-extinction structure: relative to pre-extinction, connectance, generality, vulnerability, and maximum trophic level increased; linear chains (S1) and omnivory (S2) increased; apparent (S4) and direct (S5) competition decreased. The community shifted to a species-poor, densely connected web of generalists (consistent with Skeleton Crew hypothesis).
• Recovery dynamics were decoupled: connectance and generality rebounded by the early recovery, while richness, vulnerability, and competitive motifs recovered by the late recovery, requiring ~7 Myr to reach pre-extinction levels.
• Novel late recovery state: despite overall recovery, maximum trophic level and vertical motifs (chains, omnivory) exceeded pre-extinction values, indicating increased vertical structure consistent with the Mesozoic Marine Revolution.
• Robustness: all networks exhibited substantial secondary extinctions; pre- to post-ETEE robustness patterns were similar; robustness declined further in early recovery, then increased in late recovery to exceed pre-extinction robustness.
• Event context: globally ~26% of marine genera lost; in the Cleveland Basin ~60% of marine species (including ~87% of benthic species) went extinct, consistent with intense regional dysoxia/anoxia.

Embedding species interactions reveals that primary dysoxia-focused selectivity against infaunal and epifaunal guilds best explains the identity of losses and the community structure observed after the ETEE, including secondary cascades to higher trophic levels. This resolves otherwise puzzling high losses among predators via bottom-up prey loss. The post-extinction community exhibited reduced functional redundancy and elevated generalism, yielding dense consumer-resource linkages and diminished competitive motifs—consistent with theory that generalists preferentially survive mass extinctions and with the Skeleton Crew hypothesis. Recovery unfolded asynchronously: certain structural aspects (connectance, generality) recovered rapidly, while biodiversity and competitive motifs lagged, indicating functional and structural recovery can precede taxonomic recovery. The late recovery community’s greater trophic height and vertical interaction motifs imply a new stable state driven by macroevolutionary changes (increasing productivity and predation during the Mesozoic Marine Revolution). Robustness analyses show that structural reorganization during recovery, rather than the extinction event itself, altered vulnerability to cascades, culminating in a more robust late Toarcian network. These results link extinction selectivity, cascading dynamics, and macroevolutionary reorganization to explain both collapse and recovery trajectories, and suggest that modern-like vertical complexity can increase robustness to secondary extinctions.

The study demonstrates that Early Toarcian extinctions in the Cleveland Basin were driven by primary dysoxia/anoxia selectively targeting infaunal and epifaunal benthic guilds, with cascading secondary extinctions shaping the post-extinction web. The event transformed a diverse, functionally redundant community into a species-poor, densely connected, generalist-dominated network. Recovery of structure and function preceded full biodiversity recovery and required ~7 Myr, after which increased trophic height and vertical motifs indicate emergence of a new, more vertically complex and robust state associated with the Mesozoic Marine Revolution. This supports the Skeleton Crew hypothesis and suggests that macroevolutionary increases in productivity and predation can enhance ecosystem robustness. Future work could incorporate realised interaction networks with interaction strengths, expand to other regions and events for generality, and explore how trait-mediated rewiring influences robustness and recovery trajectories.

• Regional focus: analyses are based on the Cleveland Basin and may not represent global patterns.
• Network reconstruction uses feasible webs (all trait-permitted interactions) rather than realised, mechanistic networks, potentially overestimating link density and rewiring capacity.
• Interaction strengths are not quantified; stability inferences from structure alone may miss dynamical effects of heterogeneous strengths.
• Primary extinction scenarios are trait-sequenced and simplified; other environmental filters or combined stressors could influence selectivity.
• Robustness analyses rely on random primary deletion sequences and guild-level abstraction; species-level variation and non-random extinction sequences could yield different sensitivity.
• Temporal binning aggregates intervals and assumes stationarity of traits and rules within bins.