Dinosaurs in decline tens of millions of years before their final extinction

Significance Whether dinosaurs were in decline before their final extinction 66 Mya has been debated for decades with no clear resolution. This dispute has not been resolved because of inappropriate data and methods. Here, for the first time to our knowledge, we apply a statistical approach that models changes in speciation and extinction through time. We find overwhelming support for a long-term decline across all dinosaurs and within all three major dinosaur groups. Our results highlight that dinosaurs showed a marked reduction in their ability to replace extinct species with new ones, making them vulnerable to extinction and unable to respond quickly to and recover from the final catastrophic event 66 Mya. Whether dinosaurs were in a long-term decline or whether they were reigning strong right up to their final disappearance at the Cretaceous–Paleogene (K-Pg) mass extinction event 66 Mya has been debated for decades with no clear resolution. The dispute has continued unresolved because of a lack of statistical rigor and appropriate evolutionary framework. Here, for the first time to our knowledge, we apply a Bayesian phylogenetic approach to model the evolutionary dynamics of speciation and extinction through time in Mesozoic dinosaurs, properly taking account of previously ignored statistical violations. We find overwhelming support for a long-term decline across all dinosaurs and within all three dinosaurian subclades (Ornithischia, Sauropodomorpha, and Theropoda), where speciation rate slowed down through time and was ultimately exceeded by extinction rate tens of millions of years before the K-Pg boundary. The only exceptions to this general pattern are the morphologically specialized herbivores, the Hadrosauriformes and Ceratopsidae, which show rapid species proliferations throughout the Late Cretaceous instead. Our results highlight that, despite some heterogeneity in speciation dynamics, dinosaurs showed a marked reduction in their ability to replace extinct species with new ones, making them vulnerable to extinction and unable to respond quickly to and recover from the final catastrophic event.

Whether dinosaurs were in a long-term decline or whether they were reigning strong right up to their final disappearance at the Cretaceous-Paleogene (K-Pg) mass extinction event 66 Mya has been debated for decades with no clear resolution. The dispute has continued unresolved because of a lack of statistical rigor and appropriate evolutionary framework. Here, for the first time to our knowledge, we apply a Bayesian phylogenetic approach to model the evolutionary dynamics of speciation and extinction through time in Mesozoic dinosaurs, properly taking account of previously ignored statistical violations. We find overwhelming support for a long-term decline across all dinosaurs and within all three dinosaurian subclades (Ornithischia, Sauropodomorpha, and Theropoda), where speciation rate slowed down through time and was ultimately exceeded by extinction rate tens of millions of years before the K-Pg boundary. The only exceptions to this general pattern are the morphologically specialized herbivores, the Hadrosauriformes and Ceratopsidae, which show rapid species proliferations throughout the Late Cretaceous instead. Our results highlight that, despite some heterogeneity in speciation dynamics, dinosaurs showed a marked reduction in their ability to replace extinct species with new ones, making them vulnerable to extinction and unable to respond quickly to and recover from the final catastrophic event.
dinosaurs | evolution | speciation | phylogeny | Bayesian methods N onavian dinosaurs met their demise suddenly, coincident with the Chicxulub impact in Mexico around 66 Mya; however, whether there was any long-term trend toward declining diversity leading to the Cretaceous-Paleogene (K-Pg) boundary has been controversial and debated for decades (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14). This longstanding dispute has been prolonged partly because of differences in fossil datasets from different parts of the world and difficulties in rock dating but most importantly, methodological weaknesses-previous attempts have been nonphylogenetic, and analyses were conducted on simple time-binned tabulated data, resulting in a lack of statistical rigor (phylogenetic and temporal nonindependence have not been considered), and did not truly investigate evolutionary dynamics, such as speciation and extinction rates. In fact, patterns of speciation and extinction in dinosaurs have gone largely unstudied (8). Here, we study speciation dynamics (relationship between speciation and extinction rates) using an exclusively phylogenetic approach in a Bayesian framework.
If speciation and extinction rate were constant (but speciation was higher), we would expect to see a linear increase through time in the logarithm of the number of speciation events along each path of a phylogenetic tree (linear) (Materials and Methods and Fig. 1, model A). If speciation rate decreased through time but remained above extinction rate, then we would expect a curvilinear relationship (Fig. 1, models B and C). Such a relationship would reach an asymptote (speciation equals extinction) (Fig. 1, model B) and eventually, turn down as extinction rate surpasses speciation during the evolutionary history of the clade (Fig. 1, model C). The latter would correspond to a longterm pre-K-Pg demise in the case of dinosaurs. The distinction between such evolutionary dynamics can only be made using phylogenies with taxa sampled through time.

Results and Discussion
Using a phylogenetic generalized linear mixed model (GLMM) in a Bayesian framework (15) and three recent large comprehensive dinosaur phylogenies comprising 420 (8) and 614 taxa [two trees (16)], respectively, we found that the data are significantly better explained by a model, in which extinction rate exceeds speciation rate from ∼24 My before the K-Pg boundary, than the simpler alternative model [difference in deviance information criterion (ΔDIC) between linear and quadratic models >11] ( Fig. 2A and Table S1). Our findings are qualitatively identical across all three trees, and we report on results from one of the 614-taxon trees (16).
Because nonhomogeneity in evolutionary rates is widespread and common in nature (17)(18)(19) and dinosaurs are diverse-from the bipedal, carnivorous theropods to the quadrupedal, megaherbivorous sauropods-we might expect to find different speciation dynamics in the different dinosaurian subclades. When model parameters were estimated separately for each of the three main subclades (Ornithischia, Sauropodomorpha, and Theropoda), the same general pattern as in the total Dinosauria model was recovered but with extinction rates exceeding speciation rates earlier at 48-53 My before the K-Pg boundary (ΔDIC > 12) ( Fig.  2B and Table S1). Ornithischia here refers to nonhadrosauriform, nonceratopsid ornithischians, because the two Cretaceous subclades, Hadrosauriformes and Ceratopsidae, show speciation patterns distinct from other ornithischians; Lloyd et al. (8) also identified significant diversification shifts at the base of comparable clades [i.e., Euhadrosauria (here Hadrosauriformes) (SI Text) and Ceratopsidae]. In line with this finding, these two subclades show no signs of speciation slowdowns or downturns Significance Whether dinosaurs were in decline before their final extinction 66 Mya has been debated for decades with no clear resolution. This dispute has not been resolved because of inappropriate data and methods. Here, for the first time to our knowledge, we apply a statistical approach that models changes in speciation and extinction through time. We find overwhelming support for a long-term decline across all dinosaurs and within all three major dinosaur groups. Our results highlight that dinosaurs showed a marked reduction in their ability to replace extinct species with new ones, making them vulnerable to extinction and unable to respond quickly to and recover from the final catastrophic event 66 Mya.
(ΔDIC between linear and quadratic models >5 in favor of the linear model) (Figs. 2B, Inset and 3 and Table S1). Thus, the difference in the timing of the switch from slowdown to downturn in the Dinosauria model and for the three major clades is caused by the nonhomogeneity in speciation processes across dinosaurian groups. However, these two subclades combined only represent 14% of dinosaur species; over time, dinosaurs overwhelmingly experienced a reduction in their capacity to replace extinct species with new onesnet speciation per 1 My at the time that dinosaurs went extinct (66 Mya) was significantly below zero (speciation rate is less than extinction rate) (Fig. 3B) in the three major clades (Table S12)-and Hadrosauriformes and Ceratopsidae are the exceptions.
The most prominent downturn is seen in the sauropodomorphs, where speciation increases rapidly through the Triassic and Early Jurassic (an average of 0.137 speciation events for every 1 My) until ∼195 Mya; then, speciation rate starts to slow down, and extinction rate surpasses speciation rate at ∼114 Mya (Figs. 2B and 3). Early sauropodomorph lineages are numerous but not long-lasting, and taxa that originated earlier in geological time are successively replaced by younger ones. The near extinction of the diplodocoids at the end of the Jurassic 145 Mya did not affect high speciation rates (Fig. 3), and sauropodomorphs only began their decline ∼30 My into the Early Cretaceous (Fig. 3). The subsequent originations of titanosaurian taxa were not nearly enough to compensate for the continuous loss of sauropods throughout the remainder of the Cretaceous.
Speciation in theropods follows a slower increase (∼0.07 speciation events for every 1 My) with an early onset of speciation slowdown from the Late Triassic ∼215 Mya to the Early Cretaceous ∼120 Mya, when extinction rate exceeds speciation rate (Figs. 2B and 3). Although Theropoda contains one of the most morphologically diverse dinosaurian clades, the coelurosaurs, which include the giant carnivorous tyrannosaurs, parrotlike oviraptorosaurs, large pot-bellied therizinosaurs, ostrich-like ornithomimosaurs, small sickle-clawed dromaeosaurs, and birds (most of which are Cretaceous in age), they originated in the Early to Middle Jurassic (Fig. 3), much earlier than expected from apparent fossil occurrences (8). Clades appearing even earlier (e.g., ceratosaurs, megalosauroids, and allosauroids) also persist into the Late Cretaceous, all of which might suggest that the theropod speciation pattern would be a classic "early burst" or adaptive radiation-type speciation (20) with long protracted branches (8), corresponding to a speciation slowdown model. Although our results do show an initial burst of speciation events and a gradual and prolonged slowdown, consistent with an early burst model, the fact that extinction rate surpasses speciation A B C Fig. 1. Theoretical models of speciation through time. If speciation and extinction rate were constant through time (but speciation was higher) in dinosaurian history, we would expect to see a linear increase through time in the logarithm of the number of speciation events along each path of a phylogenetic tree (model A). If speciation rate decreased through time but remained above extinction rate, then we would expect a curvilinear relationship (models B and C). Such a relationship would reach an asymptote (speciation equals extinction; model B) and eventually, turn down as extinction rate surpassed speciation during the evolutionary history of the clade (model C). The latter would correspond to a long-term pre-K-Pg demise in the case of dinosaurs. rate highlights a more complex process in theropods (Figs. 2B  and 3). Because birds underwent a radiation in the Early Cretaceous after their appearance in the Middle to Late Jurassic, one might expect that their pattern of speciation would be distinct from that of nonavian theropods. However, when we allow separate coefficients (intercept, slope, and quadratic terms) in our model to be estimated for birds and nonavian theropods, the resulting regression parameters were not significant: in other words, the speciation dynamics in Mesozoic birds are not distinct from those of nonavian theropods (Tables S1-S10). This result is in line with recent findings of a high, sustained rate of change from the Late Triassic to the Early Cretaceous in the entire theropod lineage leading to Archaeopteryx and among the earliest birds (21,22).
Ornithischians show a similar increase to theropods (∼0.06 speciation events for every 1 My) to ∼192 Mya followed by a slowdown to ∼114 Mya, at which point extinction rate exceeds speciation rate (Figs. 2B and 3). Key morphofunctional features in oral food processing distinguish hadrosauriforms and ceratopsids from other ornithischians, permitting them to exploit major new food sources (23,24). Whether these herbivores were exploiting the new, small, fast-growing herbaceous angiosperms that became common and widespread as early as the Aptian-Albian (125-109 Mya) of the Early Cretaceous (25) is much debated. The powerful jaws and massive dental batteries of these herbivores might have been adapted to other tougher nonangiosperm plant food, and they benefited from a new adaptive complex in food processing.
Our results showing high levels of speciation in hadrosauriforms and ceratopsids, although consistent with previous findings (8), seem to contradict more recent work that suggests that these groups underwent a decline in morphological diversity during the last two stages of the Cretaceous of North America (13,26). These dinosaur species are morphologically and ecologically (at least at the family level) conserved (27), with most of the derived characteristics concentrated in their crania (24). Speciation can be high in these groups, despite the potentially low morphological diversity, because Cretaceous dinosaurs exhibited increased provincialism (28) (speciation arising from geographic isolation rather than sympatric niche partitioning), increased α-diversity (many more species with subtly varying skulls but identical postcrania sharing the herbivorous ecospace in single localities), and changing taxonomic composition of stable ecological community structures [ecological niches remain constant, but taxa filling those niches changed through time (27,29)].
An ecological limit on speciation or the filling of available niches (30,31) is commonly invoked to explain speciation slowdowns. Members of the same clade are more likely to compete for similar if not the same ecological niche or portions of ecospace (32,33), and the more numerous the number of contemporary lineages, the fewer the number of available niches. We tested such an effect by including a measure of intraclade niche competitioncladewise lineage diversity or the number of contemporary branches (including internal branches) for each taxon-in the model (SI Text). However, we find that cladewise lineage diversity is not significantly associated with speciation, and it does not explain the observed downturn; physical restrictions, such as geography or range sizes, could be more important. We can indirectly assess the influence of geography, such as segregation by geographic barriers (30), using Mesozoic eustatic sea-level reconstructions (34) as an additional covariate in our models. Although various hypotheses have been proposed regarding the influence of sea level on biodiversity in dinosaurs (35), the most compelling suggests that increasing sea level results in fragmentation of large landmasses and can alter geographical distributions of habitats. In turn, continental fragmentation can lead to geographical segregation, reproductive isolation, and ultimately, speciation (30). Our results, for the first time to our knowledge, support this hypothesis-we find a significant positive effect of sea level on speciation [ΔDIC (five-group quadratic − five-group + sea-level models) >16; the proportion of the posterior distribution of the Markov-chain Monte Carlo (MCMC) estimates that crosses over zero multiplied by two (p-MCMC) <0.0010] (Tables S1-S10)-although the effect is small; for every 1-meter increase in sea level, speciation events increased by 0.2-0.25%. Horner et al. (29) observed that the emergence of transitional morphotypes coincides with marine transgressions in Late Cretaceous rocks of western North America, consistent with our finding that rising sea levels induce speciation. Importantly, the inclusion of sea level in any of our models does not diminish the temporal decline in species proliferation, despite the substantial rise of sea levels worldwide by some 150-200 m throughout the Cretaceous (Tables S2-S10).
Although we cannot positively identify a causal mechanism for the speciation downturn in dinosaurs, there are many possible global phenomena that occurred during the Cretaceous Period [e.g., the continued breakup of the supercontinents Laurasia and Gondwana (limiting free movement and eventual para-or peripatric speciation), intense prolonged volcanism (36), climate change (37)(38)(39), fluctuations in sea levels (34,40), and ecological interaction with rapidly expanding clades (41)]. To accurately identify causal mechanisms of Mesozoic dinosaurian demise, we recommend that future studies focus on a longer time period than just the last 10-20 My of the Cretaceous (4,13,42,43). In addition, our results highlight the importance of considering the expected increase in species number as clades expand and accounting for shared ancestry using phylogenetic approaches.
Our study represents the first, to our knowledge, explicitly phylogenetic statistical treatment of speciation dynamics in dinosaurs. Unlike previous nonphylogenetic attempts to study changes in dinosaur taxic diversity across geological time bins (8-10, 13, 35, 44, 45), our method is robust to sampling and other potentially confounding factors (SI Text and Tables S1-S10) and can statistically detect decreases in net speciation, which is difficult if not impossible to establish using conventional methods. Furthermore, by accounting for the effects of shared ancestry, we provide a more accurate picture of dinosaurian speciation dynamics than the simple summing of species records through time.
Our results show that dinosaurs were in decline for a much longer period than previously thought-extinction rate surpassed speciation rate at least 40 My before their final extinction. This prolonged demise leaves plenty of time for other animal groups to radiate and flourish as more and more ecological niches open up, most prominently the pre-K-Pg expansion of crown mammals (46). Although Mesozoic dinosaurs undoubtedly dominated the terrestrial megafauna until the end of the Cretaceous, they did see a reduction in their capacity to replace extinct species with new ones, making them more susceptible to sudden and catastrophic environmental changes, like those associated with the asteroid impact.

Materials and Methods
Phylogeny. We used three recent large comprehensive dinosaur phylogenies comprising 420 (8) and 614 taxa [two trees (16)]. Trees were scaled according to the midpoint time of each terminal stratigraphic range (16) using the "equal" scaling method (47) implemented in the paleotree R package (48). Additionally, we scaled the trees using two alternative sets of terminal dates, the first appearance dates and the last appearance dates, to assess the effects of tree scaling on model results.
GLMMs. We fitted GLMMs in a Bayesian framework through MCMC using the MCMCglmm R package (15). The total number of speciation events (node count) along the phylogenetic path for each taxon was modeled as the response variable, with the corresponding path length (time elapsed from root to tip) as the main effects predictor variable-this model formulation forms the null linear model (Fig. 1, model A). We also fitted a speciation slowdown model with the addition of a quadratic term (time 2 ) to the main effect. Incidentally, a quadratic model can also explain the opposite case, where speciation rate increases while extinction rate remains constant. We include phylogeny as a random effect to account for shared ancestry.
Separate intercepts, slopes, and quadratic terms were estimated for the three major dinosaurian clades (Sauropodomorpha, Theropoda, and Ornithischia; three-group model). Lloyd et al. (8) previously identified two significant diversification shifts in the Cretaceous ornithischians at the base of the clades Euhadrosauria (here Hadrosauriformes) and Ceratopsidae, and therefore, we estimated separate model coefficients (intercepts and slopes) for these groups from other ornithischians (five-group model).
Chains were run for 10 6 iterations, with sampling at every 1,000th iteration. We fitted a GLMM with a Poisson link to appropriately account for error structure in count data-although we discuss predicted curve shapes in log space, we did not log-transform node count for model fitting (49). MCMCglmm automatically accounts for overdispersion in the count data distribution. We used default priors (μ = 0 and V = I × 10 10 , where I is an identity matrix) for the fixed effects and parameter-expanded priors (V = 1, ν = 1, αμ = 0, and αV = 25 2 ) for the phylogenetic random effects (15).
Model fit was assessed using deviance information criterion (DIC) and inspection of model parameter significance (using the proportion of the posterior distribution of the MCMC estimate that crosses over zero multiplied by two). We determined the best fit model as the model with the lowest DIC score and a difference in DIC score compared with that of a base model (ΔDIC) that is greater than four. In the case where multiple models had nonsignificant differences in model fit (i.e., ΔDIC < 4), we inspected the significance of model parameters and selected the model with significant covariates (i.e., nonsignificant covariates were removed).
Extrinsic Factors. Because the fossil record has long been known to be incomplete (50,51), it is possible that the observed slowdown and downturn are byproducts of undersampling. This assumption would imply that there is a systematic downward bias in the phylogeny toward recent times, which would be counter to the usual expectation for poor sampling (50,51). Here, to test the effect of such biases, we fitted additional models with appropriate covariates, including stage-level formation counts (because formation count is widely reported to be associated with sampling bias) (9,10,12,35,44,52,53), taxon-specific formation counts (the number of formations in which a taxon is found), taxon-specific collection count (the number of fossil collections in which a taxon is represented), cladewise valid taxa counts (the known underrepresentation in the phylogeny) (54), fossil quality scores (state of preservation) (55), and body size (smaller taxa are less likely to be preserved) (56).
As an indirect measure of the influence of geography on speciation dynamics, such as segregation by geographic barriers (30), we used Mesozoic eustatic sea-level reconstructions (34) as an additional covariate in our models (mean sea-level value along each terminal branch). We also tested the ecological limit on clade diversification or the possible effects of niche saturation by adding a measure of intraclade diversity taken as the number of contemporary branches (including internal branches) for each taxon (the number of tips in time-sliced trees) (48). All data files are available in Datasets S1-S13.