A polyphyodont, semiaquatic eutriconodontan mammal from the Early Cretaceous of China

Mammals originated through the progressive acquisition of key morphological and developmental innovations that underlie their evolutionary success across diverse ecological niches¹. Among these, diphyodonty and a fully detached middle ear are widely regarded as two key diagnostic features of extant mammals and represent fundamental transitions in mammalian evolution^2,3. Extant mammals typically exhibit ante-molar diphyodont dentition, with two generations of teeth at all or some loci in the incisors, canines, and premolars, in contrast to polyphyodonty in reptiles, in which multiple generations occur throughout the dental series². Reconstructing dental replacement patterns in the fossil record is challenging because temporally resolved growth series are rarely preserved, and interpretations of Mesozoic mammalian dentitions rely largely on indirect evidence from tooth wear, crown morphology, and phylogeny from limited specimens^4,5. As a result, both the timing of the origin of diphyodonty and the number of times it evolved within Mammalia remain uncertain. In particular, the relationship to mammalian life history traits—such as determinate growth, lactation, and endothermy—has yet to be fully resolved. Furthermore, although detachment of the middle ear from the mandible is a hallmark of extant mammals, the specific developmental and evolutionary pathways by which the auditory ossicles separated from the lower jaw remain debated^3,6. Here, we report a nearly complete skeleton of an Early Cretaceous eutriconodontan mammal that preserves a rare combination of polyphyodonty, a reduced ossified Meckel’s cartilage, and fully differentiated auditory ossicles. This exceptional specimen provides critical anatomical evidence regarding the acquisition of key mammalian traits and refines our understanding of the evolutionary trajectory leading to extant mammals.

Systematic paleontology

Class Mammalia Linnaeus, 1758⁷

Order Eutriconodonta Kermack, Mussett and Rigney, 1973⁸

Family incertae sedis

Dongoconodon platycauda, gen. et sp. nov.

Holotype

Skeleton lacking left humerus, ulna, and radius. IMMNH-PV01700 (Inner Mongolia Museum of Natural History, Hohhot, Inner Mongolia, China) (Fig. 1; Supplementary Figs. 1 and 2).

Locality and age

Lower Cretaceous Jiufotang Formation at Changzigou locality, Sanjiazi Town of Lingyuan City, Liaoning, China, dated to approximately 120 million years ago⁹.

Etymology

“Dong” (for Prof. Junshe Dong, in honor of his contribution to vertebrate paleontology), “conodon” (Greek, combining cone and tooth); “platy” (Greek, flat), “cauda” (Latin, tail).

Diagnosis

Dental formula I3/i2 C1/c1 P3/p3 M4/m4 [incisors, canines, premolariforms, molariforms]. Differs from other eutriconodontans in having anteroposteriorly elongated but transversely short, plate-like proximal caudal vertebrae. Resembles gobiconodontids (sensu Martin et al.¹⁰) in having some molariforms that have replacements, but differs in the absence of a procumbent, enlarged caniniform lower first incisor and the presence of distinctive ectoflexus on upper molariforms. Further differs from gobiconodontids in dental formula, with one premolariform more than Repenomamus and Spinolestes; one upper incisor more than Gobiconodon. Further differs from Repenomamus in having the upper first incisor smaller than the remaining incisors and in lacking an elongated, tapered end of the ossified Meckel’s cartilage. Further differs from Spinolestes in having a distinctive distal lingual cingulid and cuspule f on the lower molariforms. Differs from Amphilestes and Acinacodus in having a deeper horizontal ramus of the mandible and a Meckelian groove parallel to the ventral border. Further differs from Amphilestes in having one lower premolariform and one molariform less, and from Acinacodus in having one lower premolariform more and two fewer lower incisors and one lower molariform less. Differs from Jeholodens in having two fewer lower incisors, one fewer upper incisor, an additional premolariform, and the presence of cuspules e and f on the lower molariforms. Differs from Yanoconodon in having one upper incisor more, one additional premolariform and molariform, and the presence of a lingual cingulum on the upper molariforms. Differs from Liaoconodon and Jueconodon in having an additional premolariform, a two-rooted lower penultimate premolariform, and a cuspule f on the lower molariforms. Differs from the molariforms of triconodontids in the presence of cusps e and f and unequal heights of principal cusps and the absence of tongue-in-groove interlocking.

Description

Dongoconodon is represented by a nearly complete articulated skeleton (Fig. 1) with a well-preserved cranium and mandibles (Supplementary Fig. 1), lacking only the right humerus, ulna, and radius. The cranium is triangular and posteriorly broad, with paired septomaxillae forming horizontal plates that contribute to the lateral roof of the external nares. The squamosal zygomatic process is massive, dorsoventrally deep, and spans more than half of the zygomatic arch. The mandible is robust, with a slightly convex ventral margin and maximum depth beneath the last molariform; the mandibular condyle is bulbous and lies approximately level with the alveolar margin. The axial skeleton comprises 7 cervical, 21 thoracic, 5 lumbar, 3 sacral, and at least 19 caudal vertebrae, together with 22 ribs, the last of which is interpreted as a lumbar rib. The vertebral centra are dorsoventrally compressed, whereas the proximal caudal centra are elongate and plate-like. The humerus has a spherical, transversely elongate head that faces posterolaterally without projecting above the greater tubercle, a prominent deltopectoral crest extending along two-thirds of the shaft, a large entepicondyle, and a broad olecranon fossa. The radius is relatively straight, with a mediolaterally compressed shaft and a slightly expanded distal end. The ulna bears a moderately developed olecranon process, a shallow, elongate trochlear notch and a mediolaterally compressed shaft. The femur has a medially offset, rounded head and a dorsoventrally flattened shaft that is only slightly longer than the tibia. The fibula is more slender and slightly longer than the tibia, with an enlarged parafibula, a distally expanded shaft, and no distal styloid process. A detailed description of the dentition is provided in Supplementary Note 1.

Phylogenetic analysis

Most recent phylogenetic analyses have eutriconodontans as monophyletic^6,11,12, although in one study¹³ they are a paraphyletic assemblage outside the clade of Multituberculata and Theria (placentals and marsupials). The placement of the monophyletic Eutriconodonta varies from outside Mammalia¹¹ to differing positions within the lineage that includes Theria^6,12. We conducted a parsimony analysis to test the position of Dongoconodon. The strict consensus tree (Supplementary Fig. 3) has Eutriconodonta as sister to Trechnotheria (which includes “symmetrodonts” and therians) as in Wang et al.³ and Mao et al.¹². Within Eutriconodonta are two main clades: one uniting Dongoconodon and Gobiconodontidae¹⁰, including the largest Mesozoic mammals, the badger-size Repenomamus¹⁴ and the opossum-size Gobiconodon ostromi¹⁵, and the other for Liaoconodon, triconodontids, and jeholodontids.

Dental replacement in Dongoconodon

A complication with some fossil taxa concerns the definition of a molar. For extant mammals, a molar is an unreplaced posterior tooth¹⁶, determined through the study of ontogeny. Fossils seldom preserve sufficient ontogenetic evidence to ascertain the pattern of dental replacement. As a proxy for ontogeny, paleontologists use morphology. Molars in fossils are usually identified by their complex crown morphology and posterior jaw position; premolars are situated more anteriorly, are generally simpler, and when a replacement is known, it is usually simpler than its deciduous predecessor. Recently, differential wear has been used as the main criterion to distinguish the premolar-molar boundary in morganucodontans¹⁷. However, using wear alone may not be applicable for all fossils; in the docodontan Haldanodon, for example, the direction of increasing wear in the molar series changes during ontogeny from mesially stronger to distally stronger¹⁸.

Some gobiconodontid eutriconodontans have complex posterior teeth with similarly complex replacements^10,19. Because those teeth do not fit the single generation definition of a molar, the term molariform was introduced for the complex posterior postcanines and premolariforms for the simpler anterior ones¹⁹. Identifying postcanines in gobiconodontids by ontogeny versus complex crown morphology results in different dental formulae. For example, by morphology, Spinolestes¹⁰ has two premolariforms and four molariforms, but as the four molariforms are replaced, by the ontogenetic definition, it has six premolars and no molars. Given that the premolar-molar boundary of the ontogenetic definition is not known for most eutriconodontans, to make comparisons across the clade, the complex crown morphology definition is uniformly adopted here. By that criterion, Dongoconodon has three premolariforms and four molariforms, with the anterior two molariforms having replacements (Fig. 2).

Dongoconodon has 11 tooth positions in the upper jaw and ten in the mandible (see Supplementary Note 1). Teeth are in different stages of eruption, and the replacement teeth are in different states of formation. The most telling quadrant of the dentition for interpreting the replacement pattern in Donogoconodon is the right lower mandible (Fig. 2). Anterior to the erupted canine are three unerupted teeth within the bone, a first incisor with its replacement followed by a second incisor. In the postcanine dentition, the second and third molariforms are fully erupted. To interpret this arrangement as a diphyodont dentition requires that these molariforms fully erupted before what would be two deciduous incisors, given that the first incisor has what would be a permanent replacement. Such a pattern does not occur in any extant²⁰ or extinct mammals. An alternative interpretation followed here is that the delayed unerupted incisors are not deciduous teeth, but are from a second generation. Given that the first incisor has a replacement tooth, that replacement represents a third generation (i.e., polyphyodonty).

In general, ante-molariform teeth in Dongoconodon are not erupted or are less worn than the erupted first two molariforms, which supports their identification as from a second generation. Given that some of these ante-molariforms have replacements, a third generation is known for the canine and for some incisor and premolar loci. The lower primary right first molariform has been shed, and its complex replacement is within the bone. Below the replacement is a small enamel cap that appears to be a third generation. The only other mammal to exhibit any polyphodonty is Gobiconodon borissiaki, which, based on multiple specimens, is interpreted to have three generations of some molariforms^21,22.

The prevailing view is that ante-molar diphyodonty originated once at the base of Mammaliaformes (encompassing Late Triassic taxa such as Morganucodon and crown Mammalia)^2,23 (Supplementary Fig. 4). This consensus is questioned by the recent report of ante-molar diphyodonty 20 million years earlier in the Norian eucynodont Brasilodon²⁴, as well as by the presence of polyphyodonty in the Early Cretaceous eutriconodontans Gobiconodon borissiaki^21,22 and now also Dongoconodon. These eutriconodontans are in a clade with other taxa characterized by molariform replacement (Supplementary Fig. 4). Based on their large caniniform incisors and jaw shape, gobiconodontids are interpreted to be carnivorous^14,25, with their anterior dentition used for catching and ripping prey. Dongoconodon shows similar anterior dentition and jaw shape, suggesting a carnivorous diet. Dietary specializations have been shown to drive lineage diversification in the first phase of the mammalian radiation in the Early Jurassic²⁶. A second phase in the Early Cretaceous also filled a diversity of ecomorphological niches and included gobiconodontids and Dongoconodon. The reappearance of molariform replacement and polyphyodonty in this lineage likely results from their carnivorous specializations. Given that the actual dental eruption pattern remains unknown for most relevant Mesozoic fossil taxa, it is uncertain how many times diphyodonty arose or switched to polyphyodonty in some lineages. Plasticity in the dentition may be more widespread in mammals as molariform replacement is also documented in the stem therian lineage²⁷ (Supplementary Fig. 4).

Ossified Meckel’s cartilage and detachment of the mammalian middle ear

Meckel’s cartilage (MC) forms in all mammalian embryos²⁸ and can be subdivided into proximal, central, and distal regions, each with distinct ontogenetic fates^29,30 (Fig. 3a–c). The proximal region gives rise to the malleus and incus; the distal region is incorporated into the mandibular symphysis; the central region is further differentiated into an anterior portion that becomes incorporated into the mandible, and a posterior portion that becomes ligamentous. Most Mesozoic mammals have a sulcus of varying sizes on the lingual surface of the mandible that is interpreted as housing the central (and in some cases distal) MC, which in turn connects to the auditory ossicles via the proximal MC³¹ (Fig. 3d, f). Through this connection, the auditory ossicles are inferred to have functioned dually in chewing and hearing. In those Mesozoic mammals where the Meckelian sulcus is entirely absent, the auditory ossicles are considered to be detached from the mandible without a bony connection (Fig. 3h), as in all extant mammals, and function exclusively in hearing. Although the MC has not been found in any fossils, an ossified Meckel’s cartilage (OMC) was described for the first time in 2001 in the Early Cretaceous eutriconodontan Repenomamus³². Subsequently, an OMC has been reported in some morganucodontans³³, docodontans³⁴, eutriconodontans³⁵, “symmetrodonts³⁶”, and even one eutherian³⁷.

The loss of bone in the OMC (or cartilage in the central part of the MC) is widely regarded as a critical step in the bony detachment of the mammalian middle-ear apparatus from the mandible. A recent hypothesis regarding the therian lineage (marsupials and placentals) proposed that this process was initiated at the junction between the OMC and the auditory ossicles, based on the Early Cretaceous “symmetrodont” Origolestes, in which a small gap between these elements was reported⁶. This gap was interpreted as freeing the ossicles for an exclusive auditory function, while leaving an elongate, curved OMC of uncertain function projecting from the mandible. A recent reinterpretation of the same specimen concludes that this gap is an artifact³⁸ and instead proposes that the initial breakdown occurred between the mandible and MC and progressed posteriorly. This scenario is congruent with the ontogenetic fate of MC in extant mammals³⁰ (Fig. 3a–c).

Dongoconodon preserves the OMC as well as all the elements of the bony middle-ear apparatus with malleus, incus, stapes, and ectotympanic (Fig. 3f, g and Supplementary Fig. 5). The OMC is straight, shaped like a flattened spindle with a twist in the middle. Its mandibular end is mediolaterally compressed and has a small area of contact with a flat facet on the mandible, ventral to the mandibular foramen, while the tympanic end is dorsoventrally compressed, where it contacts the ectotympanic and malleus.

Among eutriconodontans, two primary types of OMC are now known, one with a mandibular attachment nearly 60% of the OMC length and a second with an attachment approximating 25%. In the first type, as seen in Repenomamus³⁵, Liaoconodon³⁹ (Fig. 3d, e), and Yanoconodon⁴⁰, the ends of the OMC radically differ: the mandibular end is a narrow rod fit into a long Meckelian sulcus, and the tympanic end is broad and dorsoventrally compressed; there is a curve in the middle that moves the OMC medially away from the mandible. In the second type, as seen in Dongoconodon (Fig. 3f, g), the mandibular end is much shorter and not a tapered rod. The only OMC reported in Gobiconodon³⁵ is incomplete, but its shallow facet on the mandible resembles that in Dongoconodon, suggesting it also had a short OMC with a small area of mandibular contact. Dongoconodon and Gobiconodon³⁵ have a narrow second groove dorsal to the OMC facet that extends forward under the penultimate molariform. Given that there are no likely neurovascular or muscular candidates^41,42, the occupant of the dorsal groove is likely a ligament or soft tissue structure related to the OMC-mandibular contact.

The two OMC types in eutriconodontans (long and short mandibular attachments) are joined by a third morphology: no mandibular attachment with the absence of the Meckelian sulcus (e.g., Corviconodon⁴³, Fig. 3h). We view these three morphologies as a transformation series mirroring extant mammal ontogeny with the direction of reduction from anterior to posterior and the detachment of the auditory ossicles occurring between the OMC and mandible³⁰. These three morphologies (contact long, short, and absent) can be found in representatives of other mammalian lineages, including the one leading to placentals, with the Meckelian sulcus long, for example, in Prokennalestes⁴⁴ and Eomaia⁴⁵, short in Ambolestes⁴⁶ and Cokotherium⁴⁷, and absent in Microtherulum⁴⁷ and Montanalestes⁴⁸. These widespread morphologies and a detachment of MC at the mandible account for the multiple origins in different mammalian lineages of detached ossicles dedicated to hearing.

Semiaquatic adaptations

Dongoconodon is estimated to weigh between 115 and 190 grams (Supplementary Note 2), roughly the size of a water vole. It has a well-preserved skeleton that shows features occurring in extant semiaquatic mammals^49,50. Its femur has a dorsoventrally flattened shaft and mediolaterally broadened distal epiphysis (Fig. 1). Its hands and feet are remarkably like those of the platypus, Ornithorhynchus (Fig. 4a–f), broad and elongated compared to the limbs. The metacarpals and metatarsals increase in length from medial to lateral, and those for digits two through four are mediolaterally compressed at their proximal ends. The manual and pedal phalanges are dorsoventrally flattened, and the midshafts of the proximal phalanges for digits two through five are broader than the midshafts of their respective metacarpals and metatarsals. The fifth metacarpal contacts the lateral aspect of the hamate in the distal carpal row rather than head-on. Digits one through four increase in length in both the hands and feet. Together, these features allow the manual and pedal digits to spread, increasing the maximal plantar surface area for propulsion during the power phase of the swimming stroke^51,52. Several of the manual phalanges have flanges on their dorsal margins, resembling those in the platypus where the interdigital webbing attaches (Fig. 4c, d).

Caudal vertebrae in mammal tails are divided into three regions: proximal, transitional, and distal⁵³. The proximal vertebrae have anterior zygapophyses that articulate with the posterior zygapophyses of the preceding vertebra, and the transitional region ends with the longest caudal vertebrae. Dongoconodon preserves 19 caudal vertebrae, four proximal, six transitional, and nine distal (Fig. 5 and Supplementary Fig. 6a); the presence of one or more additional distals cannot be ruled out. The proximal caudal vertebrae have relatively broader and longer transverse processes and flattened centra (Supplementary Fig. 6d–f), indicative of a flattened tail. Caudal vertebrae 1–10 are box-like with short but proximodistally elongated transverse processes, which progressively decrease in width. The distal caudal vertebrae are rounded and elongated.

Based on external morphology, tails of semiaquatic mammals are characterized as tapered, keeled, or paddle-like, with the propulsive power of the tail produced with body undulations increasing from the first to the last⁵⁴. Tapered tails are elongate and narrow to a point, with some dorsoventral flattening (e.g., nutria Myocastor). Keeled tails are mediolaterally compressed (e.g., muskrat Ondatra), and paddle tails (e.g., beaver Castor, platypus Ornithorhynchus, Jurassic Castorocauda [Supplementary Fig. 6c]) are highly flattened dorsoventrally with a broad end. The caudal vertebrae of these tail types differ considerably (Fig. 5). The paddle tails have longer proximal than transitional regions with wide transverse processes for muscle attachment extending into the transitional region. Tapered tails have longer transitional regions than proximal regions, and these are roughly subequal in keeled tails. Keeled tails have tall spinous processes and chevrons that extend into the transitional region for increased muscle attachment, and tapered tails have the least area for muscle attachment with simple transitional vertebrae. We interpret a tapered type tail for Dongoconodon, with the box-like, flattened transitional caudal vertebrae resembling those of the nutria Myocastor, a semiaquatic echimyid South American rodent with a lightly haired tail⁵⁵.

To assess the locomotor mode of Dongoconodon, we performed a canonical variate analysis on 13 functional indices from 124 extant species with known locomotor modes (Supplementary Data 1) and then projected Dongoconodon into the resulting morphospace. Our results show that Dongoconodon falls within the locomotor morphospace formed by the nine semiaquatic species (Fig. 4g), with a 100% posterior probability supporting a semiaquatic locomotor mode. Several other Mesozoic mammaliforms have been interpreted as semiaquatic, i.e., the docodontan Castorocauda⁵⁶ and the eutriconodontans Liaoconodon and Yanoconodon⁵⁷. None of these fossils has a tail resembling that of Dongoconodon (Supplementary Fig. 6b, c). Castorocauda has a broad, scaly tail that resembles that of the modern beaver, Castor, which is used in paddling propulsion⁵⁶, while Liaoconodon and Yanoconodon have caudal vertebrae without flattened transverse processes^39,40. Because of its unique morphologies, the swimming mode of Dongoconodon differed from those of these three fossils, with its flattened, tapered tail used mainly in steering and stabilization, as in the nutria, and its robust forefeet providing the main propulsive force, as in the platypus.

The central tenet of early mammalian evolution is that key mammalian characters evolved incrementally in Mesozoic lineages, with the mammaliaform Morganucodon representing a pivotal intermediate as it exhibits diphyodonty², precise occlusion⁵⁸, and increased rapid juvenile growth and determinate adult growth²³, all traits shared with crown mammals. Eutriconodontans, however, complicate this linear scenario, exhibiting a mosaic of states across these characters broadly present in extant mammals (Supplementary Fig. 4). Triconodon is reportedly diphyodont⁵⁹ with precise occlusion⁶⁰, whereas Dongoconodon has polyphyodonty, with Gobiconodon borissiaki^21,22 at a minimum having molariform polyphyodonty. The Meckelian sulcus similarly varies in eutriconodontans, ranging from the elongated form resembling that seen in Morganucodon at one extreme to the absence of the sulcus with detached ossicles as in extant mammals at the other. This diversity indicates that key mammalian traits did not necessarily evolve in a tightly linked fashion. Instead, eutriconodontans demonstrate that dental, occlusal, and auditory features could evolve independently, producing complex combinations of ancestral and derived states.

The specimen examined in this study (IMMNH-PV01700) was collected in 2016 from Changzigou, Liaoning Province, China, and is permanently deposited in the Inner Mongolia Museum of Natural History (IMMNH). The collection was conducted under Permit 300, issued to IMMNH by the Ministry of Natural Resources of China on 30 April 2016. The specimen has remained in China and was not exported or imported for the purposes of this study; therefore, no export or import permits were required. All research on IMMNH-PV01700 was undertaken with permission from the Inner Mongolia Museum of Natural History and in compliance with all relevant ethical regulations and legal requirements governing the collection and scientific use of paleontological specimens.

Computed tomography scanning

The specimen was scanned on the General Electric Phoenix v|tome|x M at Yinghua NDT, Shanghai, China. The scanning had the following parameters: voltage of 180 kV, current of 100 µA, and voxel size of 21.95 µm. The three-dimensional reconstructions were created with the software Mimics (version 16.1).

Multivariate analyses

Canonical Variate Analysis (CVA, multivariate linear discriminant analysis) is utilized to maximize morphological separations between locomotor groups of extant mammals based on 13 limb bone functional indices (Supplementary Data 1). Reconstructing the locomotor modes of Mesozoic mammals based on extant taxa is inherently challenging, as these early mammals retain numerous plesiomorphic features inherited from non-mammalian cynodonts and display an evolutionary trajectory that diverges significantly from that of extant mammals. Recent analyses suggest that functional indices derived from limb element measurements provide more robust proxies for inferring locomotor adaptations in Mesozoic mammals⁶¹. Additionally, Mesozoic mammal specimens are typically preserved in a compressed state, limiting available data to linear measurements of long bones. In this study, functional indices based on long limb bone measurements—used in previous studies—were calculated. These indices, which are typically length-by-length or width-by-length skeleton ratios, describe the proportions of skeletal elements.

The extant mammal dataset used in this study is based on Weaver and Grossnickle⁶², which consists of 31 linear postcranial measurements for 133 extant species, to which we added five additional semiaquatic species from the collections of the Carnegie Natural History Museum (Supplementary Data 2). These 138 extant species belong to 17 different orders, including Monotremata, Didelphimorphia, Microbiotheria, Diprotodontia, Dasyuromorphia, Peramelemorphia, Lipotyphla, Carnivora, Primates, Dermoptera, Scandentia, Rodentia, Lagomorpha, Cingulata, Hyracoidea, Afrosoricida, and Macroscelidea. Following Chen and Wilson⁵⁷, these species were categorized into eight groups based on their common locomotor modes: scansorial, arboreal, gliding, terrestrial, saltatorial, semifossorial, semiaquatic, and fossorial. Because Dongoconodon lacks key skeleton adaptations for gliding and saltatorial locomotion (e.g., elongate radius and tibia, gracile femur), these locomotor groups were excluded from the analysis to minimize potential bias. The final dataset comprised 13 functional indices across 124 extant species. Postcranial measurements of Dongoconodon were obtained using digital calipers (Supplementary Data 2).

The CVA analysis generated a multidimensional morphospace representing the morphological disparity of extant mammals categorized by locomotor mode (Supplementary Data 3). The reported species, Dongoconodon, was subsequently projected into this morphospace. Its positioned relationships to group centroids were then analyzed to predict its most probable locomotor classification. The six-mode CVA revealed that the first three canonical variates (CVs) collectively explained 95.5% of the total variance. CV1, explaining 54.7% of the total variance, showed strong positive correlation with crural index (CI, tibia length divided by femur length), humerus epicondyle breadth (HEB, humerus distal end width divided by humerus length), femur epicondyle breadth (FEB, femur distal end width divided by femur length), humerus proximal breadth (HPI, humerus proximal end width divided by humerus length) and femur robustness index (FRI, transvers femur diameter divided by femur length), but strong negative correlation with manus proportion index of the proximal phalanx (MANUS, proximal phalanx length divided by metacarpal length). CV2 accounted for 31.9% of the total variance, correlated positively with MANUS and the manus proportion index of the intermediate phalanx (MANUS2, intermediate phalanx length divided by metacarpal length), and negatively with metacarpal robustness index (MRI, metacarpal width divided by length) and HPI. CV3 accounted for 8.89% of the total variance, exhibited positive correlations with MANUS2 and MRI, but negative correlations with brachial index (BI, radius length divided by humerus length) and CI.

Distinct clustering patterns among the six locomotor groups emerged in the CV1-CV2-CV3 morphospace (Fig. 4g). The semiaquatic, arboreal, and fossorial groups occupied segregated regions in separate corners, whereas scansorial, terrestrial, and semifossorial groups aggregated along a different direction with partial overlap. The semiaquatic group was well separated from the other five groups, presenting a typically shortened tibia (low CI), robust humerus (high HEB and HPI), and strong femur (high FEB and FRI). The analysis achieved an overall 61.3% accuracy across the 124 extant species. Classification success rates were: semiaquatic (77.8%), arboreal (76.3%), scansorial (61.5%), terrestrial (55.6%), fossorial (52.6%), and semifossorial (33.3%).

The CVA shows that Dongoconodon is most likely a semiaquatic species, as it falls within the morphospace of the semiaquatic group (Fig. 4g), with 100% posterior probability. This conclusion can also be supported by several characteristics of Dongoconodon, such as a significantly high value of FEB and FRI, and relatively high HEB and HPI, which are similar to those of extant semiaquatic species.

Phylogenetic analysis

The data matrix consisting of 105 taxa and 594 characters was analysed using maximum parsimony. In our description of Dongoconodon above, we provided a rationale for using the terms molariform and premolariform, which we follow in our taxon-character matrix. For Dongoconodon, one of the few taxa where the pattern of dental replacement is known, premolariforms are scored based primarily on second-generation teeth and molariforms based on first-generation teeth. In scoring dental replacement characters, we only score those taxa where replacement teeth or the pattern of dental replacement is known. The analysis was conducted in TNT⁶³, utilizing a new technology search (sectorial search, ratchet, and tree fusing) set to 100 iterations. The search procedure was followed by a traditional search. All characters were unweighted and nonadditive. The search resulted in five most parsimonious trees of length 2810 (with a consistency index of 0.271 and a retention index of 0.706). The strict consensus tree (2823 steps; consistency index = 0.270; retention index = 0.704) of these five most parsimonious trees is presented in Supplementary Fig. 4.

Nomenclatural acts

This published work and the nomenclatural acts it contains have been registered in ZooBank, the online registration system for the International Code of Zoological Nomenclature. The ZooBank Life Science Identifiers for this publication are http://zoobank.org/urn:lsid:zoobank.org:pub:3706565C-8424-4407-8627-AC96E8405D96 and http://zoobank.org/urn:lsid:zoobank.org:act:2C9ED6E7-21EC-4096-9911-DBF437230ED1.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.