Motion sickness induced by off-vertical axis rotation (OVAR)

Abstract

We tested the hypothesis that motion sickness is produced by an integration of the disparity between eye velocity and the yaw-axis orientation vector of velocity storage. Disparity was defined as the magnitude of the cross product between these two vectors. OVAR, which is known to produce motion sickness, generates horizontal eye velocity with a bias level related to velocity storage, as well as cyclic modulations due to re-orientation of the head re gravity. On average, the orientation vector is close to the spatial vertical. Thus, disparity can be related to the bias and tilt angle. Motion sickness sensitivity was defined as a ratio of maximum motion sickness score to the number of revolutions, allowing disparity and motion sickness sensitivity to be correlated. Nine subjects were rotated around axes tilted 10°–30° from the spatial vertical at 30°/s–120°/s. Motion sickness sensitivity increased monotonically with increases in the disparity due to changes in rotational velocity and tilt angle. Maximal motion sickness sensitivity and bias (6.8°/s) occurred when rotating at 60°/s about an axis tilted 30°. Modulations in eye velocity during OVAR were unrelated to motion sickness sensitivity. The data were predicted by a model incorporating an estimate of head velocity from otolith activation, which activated velocity storage, followed by an orientation disparity comparator that activated a motion sickness integrator. These results suggest that the sensory-motor conflict that produces motion sickness involves coding of the spatial vertical by the otolith organs and body tilt receptors and processing of eye velocity through velocity storage.

Appendix

Estimation of velocity from sequential activation of otolith polarization vectors

In this appendix, we describe the conceptual organization of the velocity estimation as presented in Raphan and Schnabolk (1988) (Fig. 10a), its neural network implementation (Fanelli et al. 1990) (Fig. 10b) and predictions of the estimated Bias Eye Velocity as a function of tilt angle and velocity of rotation (Fanelli et al. 1990) (Fig. 10c). Conceptually, the model implements the estimation of the angular velocity of a rotating wave of otolith afferent activation as the gravity vector sweeps around the head during rotation (Fig. 10a). This is accomplished as a discrete implementation of a ratio of temporal to spatial derivatives. The temporal derivatives are estimated from processing differences between neural activity for a given cell (n) and a central cell that has a delayed version of the activity (Delay (T)). The spatial derivatives are estimated from differences between nearest neighbor polarization vectors.

Fig. 10

a Model for estimation of head velocity ($\hat{\omega }$) during OVAR, with the pattern velocity estimation formula. b Neural network implementation of the model given in a. c Model predictions of the velocity estimation versus tilt angle for five simulated head velocities. The velocities are: 0°/s (open squares), 20°/s (closed circles), 40°/s (closed squares), 60°/s (open circles), and 80°/s (open triangles). The training tilts are 0, 30, 50, and 90°. The time delay is 0.75 s

The neural network implementation of Fig. 10b has a three-layer architecture with 16 neurons to estimate velocity. The estimate is a generalized inner product between temporal and spatial differences. This is a one-sided model, using information from one utricular macula. The difference between the left macula and the right macula estimation is due to mirror symmetry of the numbering of cells. A small leftward rotation of the left macula gives a positive estimate of head velocity, while a small leftward rotation of the right macula gives a negative estimate of head velocity. The K_n’s in the model implement a generalized inner product or bilinear form and account for the fact that the bias component of velocity due to OVAR is approximately independent of tilt angle after it has reached a normalization value.

The weights were adjusted to fit the data on Bias Velocity of head velocities of monkeys, ranging from 0 to 80°/s and tilts of the rotation axis of 0–90° in the model of Fig. 10b. There were 16 units in the input and internal layers. A periodic structure based on pairs of units denoted by A and B was chosen. Connection weights from neighboring units in the input layer converged onto two units in the internal layer. The connection weights were periodically repeated for the eight pairs of units in the internal layer, which because of this symmetry had identical characteristics. Note W₁₁ on the solid connection to A_N as well as the shaded one on the right, indicating an example of this symmetry. One such repeating structure is emphasized with the others suggested in gray. There is one output unit. The lowest level represents the equivalent acceleration of gravity on the otolith maculae. The dotted lines indicating pre-processing represent the function transforming the gravitational acceleration into otolith unit activations. The input layer represents the central projections of the otolith units as well as their delayed counterparts.

The key conclusion from this model is that the estimator predicts that the bias eye velocity normalizes at about 30° of tilt (Fig. 10c). For angles of tilt >30°, the estimation increases with rotational velocity and tends to saturate at about 60°/s, where the increments in estimate are no longer proportional to angular velocity. For tilts of the rotation axis less than 30°, the gain of the estimation is small relative to stimulus velocity and a substantial rotational velocity would be necessary to reach significant bias velocities.