The cave bear (Ursus spelaeus s.l.) is an iconic extinct bear that inhabited the Pleistocene of Eurasia whose extinction causes are controversial. To identify the actual causes of the cave bear extinction, it is crucial to understand their feeding preferences. Here, we quantify shape descriptor metrics (DNE, RFI and OPCR) in three dimensional (3D) models of cave bear upper teeth (P⁴-M²) to make inferences on its controversial feeding behaviour. We use a comparative sample including representatives of all living bear species with known diets as a template. Our topographic analyses evidence that the complexity of upper tooth rows in living bears is more associated with the mechanical properties of the items consumed than with the type of food. Cave bears exhibit intermediate values on topographic metrics between those exhibited by the bamboo-feeder giant panda (Ailuropoda melanoleuca) and those taken by specialists in hard-mast consumption (Ursus arctos and Ursus thibetanus). The crown topography of cave bear upper teeth suggests a high efficiency to chew on tough vegetal resources of lower-quality, and no living bear is currently exploiting it. Our results align with a climatic-driven hypothesis to explain demise in cave bear populations during the late Pleistocene.

A dual-phase technique was used to produce polyvinylsiloxane-based molds (Virtual Putty and Light Body compounds) from original tooth rows (P⁴-M²). High-resolution replicas were obtained from molds using non-reflective polyurethane (Feroca Composites, Spain). Dental replicas were scanned at 0.2 mm resolution with a Roland LPX-600 located at the Central Research Services (University of Málaga, Spain). Meshes were merged and processed in Geomagic Studio 2012 (Geomagic, Inc. USA) to obtain entire enamel caps (EEC) of each tooth row cropped at cervical margin, smoothed and downsampled to 10,000 polygons faces. The EEC method was used here to prevent planometric footprint area lost and produce metric values for whole tooth shape. Meshes (*.ply file format) were then aligned orthogonal to the occlusal surfaces using MeshLab to mitigate impact of topographic metrics.