Analysis of host rock
To better constrain depositional history, a single sample (~ 5 × 5 × 2 cm) of moderately-to-well consolidated sedimentary material, found next to the skeletons, was acquired for investigation. Thin section analysis showed the rock to be a matrix-supported, medium-coarse brecciated tuff, following the classification scheme of White and Houghton9.
Framework components are either lithic fragments (i.e. being multi-phase and retaining initial texture) or mineral fragments (i.e. being > 90% a single phase). Sizes range from 0.02 to 5.96 mm with a mean of 0.46 mm (lithic fragments), and 0.02–0.43 mm with a mean of 0.11 mm (mineral fragments). Of the 228 mineral fragments characterized, 73% and 16% are either feldspar or quartz, respectively, with less abundant phases including Fe/Ti-oxides (6%), biotite-phlogopite (2%), amphiboles (1%), pyroxenes (1%), and apatite (1%; Supplementary Fig. S2A). Of the 169 lithic clasts characterized, 88%, 8%, and 4% were determined to be of igneous (volcanic), unknown, or sedimentary precursor types, respectively (Supplementary Fig. S2B). Those of volcanic origin are commonly (~ 60%) observed to contain phenocrysts. Selected examples are shown in detail in Supplementary Fig. S3. Matrix material consists predominantly of fine ash-sized smectite (nontronite), quartz, feldspar, and heulandite (Supplementary Fig. S4). Further, thin section mapping shows an average matrix: framework clast ratio of 70:30. In the limited amount of material available for study, neither sedimentary structures (such as grading or stratification) nor bioturbation features are observed.
Previous investigations of the Yixian Formation have suggested sedimentary material to have originated from multiple volcanic events, which caused associated lahars, pyroclastic flows, and ashfalls10,11,12,13. Here, the observed diversity in lithic clast characteristics, such as phenocryst phase composition (Supplementary Fig. S2C) and colour (Supplementary Fig. S2D), is consistent with these having originated from multiple (though predominantly volcanic) primary lithologies, consistent with the high degree of physical mixing associated with a pyroclastic origin. High-energy deposition is also evidenced by poor sorting, high angularity, and variable roundness (Supplementary Fig. S5), as well as the common presence of truncated phenocrysts at edges of lithic fragments (Supplementary Fig. S3). Calcite cement is also observed, commonly occurring at the margins of individual lithoclasts as well as in fractures throughout the rock (Supplementary Fig. S3). The abundance of smectite (nontronite) in the matrix material suggests diagenetic alteration of original mafic volcanic materials.
Assessment of preserved individuals
The entombed individuals represent the small ceratopsian dinosaur Psittacosaurus lujiatunensis entangled with the even smaller gobiconodontid mammal Repenomamus robustus (Fig. 1, Supplementary Fig. S6). Skeletal measurements for both are given in Table 1.
Figure 1 Psittacosaurus lujiatunensis-Repenomamus robustus pair (WZSSM VF000011) locked in mortal combat. Insets depict (left to right): hand of R. robustus wrapped around lower jaw of P. lujiatunensis, teeth of R. robustus embedded in forearm of P. lujiatunensis, hind foot of R. robustus wrapped around lower hindlimb of P. lujiatunensis. Scale bar equals 10 cm. Full size image
Table 1 Measurements for Psittacosaurus lujiatunensis-Repenomamus robustus pair (WZSSM VF000011) in mm. Full size table
The dinosaur skeleton is complete. The skull preserves all three autapomorphies that diagnose P. lujiatunensis (prefrontal width less than 50% that of the nasal, quadratojugal-squamosal contact along the anterior margin of the quadrate shaft, jugal-quadrate contact posteroventral to the laterotemporal fenestra; Supplementary Fig. S7)14. The dinosaur is lying prone, with its hindlimbs folded on either side of the body. The neck and tail curl to the dinosaur’s left. Based on femoral circumference scaling15 and applied developmental mass extrapolation16, we estimate its body mass to have been 10.6 ± 6.0 kg at time of death. We estimate its age to have been at least 6.5 years, and more likely closer to 10 years, based on established femur length-age relationships17,18.
The body of the mammal coils to the right and sits atop the left side of the dinosaur. The skeleton is nearly complete; only the distal end of the tail is missing. In life, the mammal would have been several centimeters longer than preserved (Table 1). Although diagnostic dental and mandibular characters are poorly exposed on the individual and so cannot be verified, the comparably small size of the animal, and its weak sagittal and lambdoid crests and zygomatic arches (Supplementary Fig. S8), suggest that it is R. robustus and not the larger R. giganticus, which is also known from the Lujiatun Member4. The mammal’s left hand grips the lower jaw of the dinosaur (which is dislocated and displaced rostrally), and its left hindleg is trapped within the folded left leg of the dinosaur, the hindfoot gripping the dinosaur’s left shin, immediately below the knee (Fig. 1; Supplementary Fig. S9). The mammal died while biting two of the dinosaur’s left anterior dorsal ribs; its mandible plunges downward into the indurated sediment to firmly clasp the bones (Fig. 1; Supplementary Fig. S9). These two ribs appear to be broken, based on their slight misalignment with the remaining ribcage, but the breaks are obscured, and it is not possible to determine with certainty whether the ribs were broken in life or due to taphonomic processes. Developmental mass extrapolation and long bone circumference-body mass scaling relationships provide a mass of 3.43 ± 1.42 kg for the mammal. Cheek tooth eruption and wear are commonly used as indicators of maturity in mammals19, but this is not possible for the implicated R. robustus individual because the teeth are obscured. The femur of the mammal is 15% shorter than that of the R. robustus holotype (Institute of Vertebrate Paleontology and Paleoanthropology [IVPP] specimen V 12549), and the obliteration of the internasal and interfrontal sutures (Supplementary Fig. S8) does not appear as extensive as in the holotype. Nevertheless, the long bone epiphyses of the mammal are fused, indicating that growth was nearing cessation. Probably, this individual was a subadult when it died.
Assessment of fossil association
The intimate and intertwined nature of the skeletons is remarkable, and suggests that this fossil association is authentic, not forged. Although fossil forgeries have been reported from the Jehol Group of China before, these typically involve the simple juxtaposition of two or more independent fossils20,21, and do not replicate the tangled nature of the skeletons documented here. It might be argued that the broken and slightly displaced anterior ribs of the dinosaur indicate tampering, given the otherwise mostly intact nature of the skeletons. However, there has been postdepositional displacement of some other bones, including the lower jaw of the dinosaur (Supplementary Fig. S6B), and the distal manual and pedal phalanges (Fig. 1, Supplementary Fig. S9A,C) and distal tail vertebrae of the mammal (Fig. 1); similar displacement of the ribs is therefore possible, particularly if they had fractured prior to burial. To convince ourselves of the authenticity of the fossil, we prepared and exposed the left dentary of the mammal, which had not yet been revealed at the time of acquisition, and found that it, too, plunges into the matrix to clasp the dinosaur’s ribs (Supplementary Fig. S10).
The association of the two animals also could not have resulted from passive taphonomic processes; the intact nature of the skeletons (except for some of the lighter distal limb and tail elements) indicates that they could not have been transported any appreciable distance prior to deposition. Rather, the animals almost certainly were buried where they died, both events having occurred closely in time, if not simultaneously.
The clutching hands and feet of the mammal, its biting jaws, and its position atop the dinosaur indicate that the mammal was clearly the aggressor in the preserved interaction, in agreement with the inferred carnivorous lifestyle of Repenomamus robustus4 (Psittacosaurus lujiatunensis was almost certainly herbivorous14). Nevertheless, the nature of the interaction is not immediately obvious. It is possible that the mammal was scavenging the carcass of the dinosaur when the two became buried. This proposed scenario would account for the large size of the dinosaur relative to the mammal (terrestrial predators usually favour prey that are not much larger than themselves, particularly when hunting alone22,23), and the fact that the mammal was biting the ribs of the dinosaur when it died, which would otherwise have been difficult to access (but not impossible—see below) on a living prey item. However, while plausible, we cite three lines of evidence that challenge this hypothesis. First, the bones of the dinosaur are otherwise devoid of tooth marks, which are commonly left by carnivorous mammals while scavenging24. Second, it seems unlikely that the two animals would have become so entangled, were the dinosaur dead prior to the arrival of the mammal. Third, the scavenging scenario does not predict the position of the mammal atop the dinosaur, since the mammal could presumably just as easily have eaten the dinosaur from ground level.
We propose instead that the two animals were buried in an act of predation on the part of the mammal, only for both to have been entombed by a sudden lahar-type volcanic debris flow (Fig. 2). This hypothesis would explain the entwined nature of the skeletons, wherein the left hindfoot of the mammal became trapped within the folded left leg of the dinosaur when it collapsed to the ground. It would also account for the lack of tooth marks and other indications of scavenging on the dinosaur’s skeleton, and for the mammal’s position atop the dinosaur, as though to subdue its weakened prey. To address the question of whether the dinosaur was too large to have been reasonably preyed upon by the mammal, we examined the relationship between predator body mass and prey maximum body mass among terrestrial carnivorans using phylogenetic generalized least squares (PGLS) regression. We modeled several evolutionary scenarios, of which a Pagel’s lambda model best fits the data (Fig. 3, Supplementary Table S2). Our fossil association falls well within the 95% prediction intervals for both solitary and pack hunters. We therefore cannot reject the hypothesis that this association preserves a doomed predation event on the part of the mammal, despite its smaller size. By analogy, although wolverines (Gulo gulo) are typically opportunistic feeders of large prey, lone individuals are also known to occasionally hunt animals many times their own size, including moose (Alces alces), caribou/reindeer (Rangifer tarandus), and domestic sheep (Ovis spp.)25,26. Least weasels (Mustela nivalis) have similarly been reported as occasionally attacking much larger capercaillie (Tetrao spp.), hazelhen (Tetrastes bonasia), and hare (Lepus spp.)27,28.
Figure 2 Life restoration showing Repenomamus robustus grappling with Psittacosaurus lujiatunensis. Artwork by Michael Skrepnick. Reproduced with permission. Full size image
Figure 3 Phylogenetic generalized least squares models. (A) Brownian motion, (B) Ornstein–Uhlenbeck, (C) Pagel’s λ, (D) ACDC. Linear models show the relationship between predator body mass and maximum prey body mass for solitary (blue) and pack (green) hunters. Shaded areas represent the 95% confidence intervals; dotted lines represent 95% prediction intervals. The association between the Repenomamus robustus and Psittacosaurus lujiatunensis documented here (red star), and the predicted association of somatically mature examples of these species (black star), are well within the 95% prediction intervals for each model and hunting style. Full size image
It may seem unlikely that the mammal was biting the exposed ribs of the dinosaur if it were not already long deceased; however, feeding on live prey happens commonly in carnivorous mammals, including African wild dogs (Lycaon pictus), spotted hyenas (Crocuta crocuta), and jackals (Canis mesomelas and C. aureus)29,30. In fact, after an initial struggle, the prey may ultimately give up on self-defence, opting instead to passively lay down in a state of exhaustion and deep shock29. This depiction is not very unlike the position assumed by the dinosaur described here. Kleptoparasitism by large predators on the open African savannah can significantly alter the hunting and feeding habits of smaller species31,32, and the eating of still-living pretty by African wild dogs is conceivably one such adaptation. The larger carnivorous theropods of the Early Cretaceous Lujiatun ecosystem might have posed an equal threat to Repenomamus spp., motivating a similarly rapacious feeding behaviour in the mammals.
Could an adult P. lujiatunensis have eventually outgrown the prey size threshold of R. robustus, and thereby avoided further predation from the latter species? We examined this question in the same way as above, plotting the adult body masses of these two species on our regression estimates (Fig. 3). Again, an adult P. lujiatunensis (body mass ≈ 23.5 kg) plots well within the expected maximum prey size threshold of an adult R. robustus (body mass ≈ 5.54 kg). It is therefore plausible that P. lujiatunensis remained vulnerable to predation from R. robustus throughout its lifetime. Threats from the still larger Repenomamus giganticus, the tyrannosauroid Dilong paradoxus, and an as-yet undescribed carnosaur ensured that P. lujiatunensis were ever-vigilant during ‘Lujiatun time’. Despite undersampling of the otherwise fossiliferous Lujiatun Member (Fig. 4A), it is clear that P. lujiatunensis was an abundant prey item on the Early Cretaceous landscape, considering raw and taphonomically-corrected count data and estimated standing crop biomass (Fig. 4B–D).