Humans use internal models to construct and update a sense of verticality.

Internal models serve sensory processing, sensorimotor integration and motor control. They could be a way to construct and update a sense of verticality, by combining vestibular and somatosensory graviception. We tested this hypothesis by investigating self-orientation relative to gravity in 39 normal subjects and in subjects with various somatosensory losses showing either a complete deafferentation of trunk and lower limbs (14 paraplegic patients after complete traumatic spinal cord injury) or a gradient in the degree of a hemibody sensory loss (23 hemiplegic patients after stroke). We asked subjects to estimate, in the dark, the direction of the Earth vertical in two postural conditions-upright and at lateral whole body tilt. For upright conditions, verticality estimates were not different from the direction of the Earth vertical in normal (0.24° ± 1; P = 0.42) and paraplegic subjects (0.87° ± 0.9; P = 0.14). The within-subject variability was much greater in hemiplegic than in normal subjects (2.05° ± 1.15 versus 1.06° ± 0.4; P < 0.01) and greater in paraplegic than in normal subjects (1.13° ± 0.4 versus 0.72° ± 0.4; P < 0.01). These findings indicate that, even if vestibular graviception is intact, somaesthetic graviception contributes to the sense of verticality, leading to a more robust judgement about the direction of verticality when vestibular and somaesthetic graviception yield congruent information. As expected, when normal subjects were tilted, their verticality estimates were biased in the direction of the body tilt (5.55° ± 3.9). This normal modulation of verticality perception (Aubert effect), was preserved in hemiplegics on the side of the normoaesthetic hemibody (ipsilesional) (6.09° ± 6.3), and abolished both in paraplegics (1.06° ± 2.5) and in hemiplegics (0.04° ± 6.7) on the side of hypoaesthetic hemibody (contralesional). This incongruence did not exist in deafferented paraplegics who exclusively used vestibular graviception with a similar efficacy no matter what the lateral body position. The Aubert effect was not an on-off phenomenon since the degree of hemiplegics' somatosensory loss correlated with the modulation of verticality perception when they were tilted to the side of hypoaesthetic hemibody (r = -0.55; P < 0.01). The analysis of anatomical correlates showed that the Aubert effect required the integrity of the posterolateral thalamus. This study reveals the existence of a synthesis of vestibular and somaesthetic graviception for which the posterolateral thalamus plays a major role. This corresponds to a primary property of internal models and yields the neural bases of the Aubert effect. We conclude that humans construct and update internal models of verticality in which somatosensory information plays an important role.