Dissociating vestibular and somatosensory contributions to spatial orientation

Abstract

Inferring object orientation in the surroundings heavily depends on our internal sense of direction of gravity. Previous research showed that this sense is based on the integration of multiple information sources, including visual, vestibular (otolithic), and somatosensory signals. The individual noise characteristics and contributions of these sensors can be studied using spatial orientation tasks, such as the subjective visual vertical (SVV) task. A recent study reported that patients with complete bilateral vestibular loss perform similar as healthy controls on these tasks, from which it was conjectured that the noise levels of both otoliths and body somatosensors are roll-tilt dependent. Here, we tested this hypothesis in 10 healthy human subjects by roll tilting the head relative to the body to dissociate tilt-angle dependencies of otolith and somatosensory noise. Using a psychometric approach, we measured the perceived orientation, and its variability, of a briefly flashed line relative to the gravitational vertical (SVV). Measurements were taken at multiple body-in-space orientations (-90 to 90°, steps of 30°) and head-on-body roll tilts (30° left ear down, aligned, 30° right ear down). Results showed that verticality perception is processed in a head-in-space reference frame, with a systematic SVV error that increased with larger head-in-space orientations. Variability patterns indicated a larger contribution of the otolith organs around upright and a more substantial contribution of the body somatosensors at larger body-in-space roll tilts. Simulations show that these findings are consistent with a statistical model that involves tilt-dependent noise levels of both otolith and somatosensory signals, confirming dynamic shifts in the weights of sensory inputs with tilt angle.

Keywords: internal models; multisensory integration; vertical perception.

Fig. 1.

A: schematic representation of the Bayesian optimal integration model. The model contains three different stages. In the sensory input stage, physical signals are translated to sensory signals (^), which are assumed to be accurate, but contaminated with Gaussian noise. The Gaussian noise is assumed to be tilt-angle dependent for both the head-in-space (Ĥ_S) signal from the otoliths and the body-in-space (B̂_S) signal from the body somatosensors. In the coordinate transformation stage, the neck proprioceptive signal is used for a reference frame transformation of the body-in-space signal into an indirect head-in-space signal (Ĥ_S1 = B̂_S + Ĥ_B). To derive an optimal estimate of the head-in-space orientation (H̃_S), the cue combination stage of the model weights the indirect signal (green pathway) with the direct signal (blue pathway) and prior knowledge (red pathway) that our head is usually upright (centered around 0°). The relative contributions of these pathways (w_H1, w_HD, w_HP) depend on the Gaussian noise of the underlying signals. Finally, an optimal estimate of line-in-space is obtained for the subjective visual vertical (SVV) task by combining H̃_S with estimates of the eye-in-head orientation (Ẽ_H) and line-on-eye orientation (L̃_E). The latter is assumed to be veridical. Note that the cue combination stage is essentially a multiplication of the underlying probability distributions, resulting in a posterior head tilt angle distribution. The individual probability distributions of the sensory signals and the head tilt posterior are based on the parameters in Table 1 for the condition in which the body is upright and the head is tilted 30° right ear down (RED). B: simulations of the Bayesian optimal integration model for the systematic SVV error and variability (including SE), plotted against head-in-space (left) and body-in-space (right) orientations in the three different head-on-body tilt conditions: 30° left ear down (LED, orange), aligned (cyan), and 30° RED (magenta). Parameter values are based on previous research by Clemens et al. (2011) and Alberts et al. (2015). The broken lines in the plots indicate the 30° RED condition shown by the probability distributions in A.

Fig. 2.

SVV adjustments in a representative subject for the different head-on-body tilt conditions. A: proportion (P) of clockwise (CW) responses plotted against the line orientation relative to the gravitational vertical for different head-in-space (H) and body-in-space (B) combinations. Solid lines show the best-fit psychometric curves from which the SVV error [line-in-space orientation at which P(CW) = 0.5, broken line] and the variability (inversely related to the width of the curve) are extracted and plotted against head-in-space orientation in B and C. Systematic SVV error and variability measures from the psychometric curves in A are plotted as filled symbols and completed with systematic errors and variability of the remaining combinations of head and body-in-space orientation (open symbols).

Fig. 3.

A: mean systematic SVV error and variability (including 1 SE) across the 10 subjects are plotted against head-in-space orientation (left) and body-in-space orientation (right) for the three different head-on-body tilt conditions. Mean data are superimposed on the model simulations (including 1 SE) of Fig. 1B. B: sensory weights of the different sensors plotted against head-in-space orientation for the different head-on-body tilt conditions. Shaded areas are the variance in the sensory weights, based on the noise in the individual parameter settings of Table 1.

Fig. 4.

Simulations of the integration of trunk-graviceptive (solid) and cutaneous pressure (broken line-dot) sensors and the resulting body somatosensory-dependent variability (broken lines). Individual parameters are 2.5° + 0.05°/° tilt angle for the trunk graviceptive and 3.5° + 0.03°/° tilt angle for the cutaneous pressure sensors. The latter is shifted 90°, such that the lowest variability occurs when subjects lie on the side.