1. We recorded eye movements in four normal human subjects during refixations between targets calling for various combinations of saccades and vergence. We confirmed and extended prior observations of 1) transient changes in horizontal ocular alignment during both pure horizontal saccades (relative divergence followed by relative convergence) and pure vertical saccades (usually divergence for upward and convergence for downward saccades); 2) occasional, high-frequency (20-25 Hz), conjugate oscillations along the axis orthogonal to the main saccade; and 3) the speeding up of horizontal vergence by both horizontal and vertical saccades. 2. To interpret these findings, we developed a hypothesis for the generation of vergence to step changes in target depth, both with and without associated saccades. The essential features of this hypothesis are 1) the transient changes in horizontal ocular alignment during pure horizontal saccades reflect asymmetries in the mechanical properties of the lateral and medial rectus muscles causing adduction to lag abduction; 2) pure vergence movements in response to step changes in target depth are generated by a neural network that uses a desired change in vergence position as its input command and instantaneous vergence motor error (the difference between the desired change and the actual change in vergence) to drive vergence premoter neurons; and 3) the facilitation of horizontal vergence by saccades arises from nonlinear interactions in central premotor circuits. 3. The hypothetical network for generating pure vergence to step changes in target depth is analogous in structure to the local feedback model for the generation of saccades and has the same conceptual appeal. With the assumption of a single nonlinearity describing the relationship between a vergence motor error signal and the output of the neurons that generate promoter vergence velocity commands, this model generates pure vergence movements with peak velocity-amplitude relationships and trajectories that closely match those of experimental data. 4. Several types of models are proposed for the central, nonlinear interaction that occurs when saccades and vergence are combined. Common to all models is the idea that omnidirectional pause neurons (OPN), which are thought to gate activity for saccade burst neurons, also gate activity for saccade-related vergence. In one model we hypothesize the existence of a separate class of saccade-related vergence burst neurons, which generate premotor horizontal vergence commands but only during saccades. In a second model we hypothesize separate right eye and left eye saccadic burst neurons that receive not only conjugate, but also equal but oppositely directed vergence error signals. In this way the difference between the outputs of the right eye and left eye saccade burst neurons produces a saccade-related horizontal vergence command. In a third model we propose that facilitation of vergence during saccades is a result of a multiplication (an increase in the gain of premotor vergence velocity neurons selectively during saccades). 5. The results of simulations of these models and comparison with our experimental data favor the first and third models, which either incorporate a separate class of saccade-related vergence burst neurons, or assume a change in property of premotor vergence velocity neurons caused by the lifting of OPN inhibition during saccades. Simulations of these models lead to a number of predictions about the properties of neurons within the brain stem that generate vergence commands, both with and without associated saccades. Electrophysiological experiments are needed to confirm or refute these hypotheses.
