Revealing parallel modulation on the sensory-motor decoder for smooth pursuit eye movements.

Primates use smooth pursuit eye movements to track moving objects. Pursuit is driven by visual commands to accelerate and supported by eye velocity feedback. We show that degraded motion reliability caused by reduced coherence in a pursuit target created from a moving patch of dots reduces the eye speed during the initiation of pursuit as well as during steady-state tracking. To understand the representation of image speed during pursuit and to ask why degraded motion reduces eye speeds, we recorded from isolated single neurons in the middle temporal area of extrastriate visual cortex (MT). Smooth pursuit is driven by sensory estimates of stimulus speed represented in MT. We sought to determine whether eye speed lags behind target speed for low dot coherence because (i) the speed representation in MT is compromised or (ii) the representation of image speed remains accurate in MT and eye speed is eroded in downstream circuits. We presented moving patches of dots of varying speeds and coherences while recording with microelectrodes from neurons in area MT. During pursuit initiation, the amplitude, but not the tuning, of MT responses depends on dot coherence. The population response gets noisier as coherence reduces the amplitude of neural (and eye movement) responses. To understand how MT drives the initiation and steady-state of pursuit, we asked whether we could decode appropriate motor commands from the MT population response and what were the properties of the successful decoders. During pursuit initiation, decoding eye speed required parallel pathways in a “gain-modulated vector averaging” decoder. One pathway estimated image speed by vector averaging and the other pathway computed the gain of sensory-motor transmission from the amplitude of the MT population response. To reproduce eye speed during steady-state tracking, yet another pathway was needed in the decoder. MT population activity is overall noisier during steady-state tracking but even for low dot coherence gain-modulated vector averaging predicts eye acceleration at a time when the eye is either stable at a speed well below target speed, or even decelerating. We could not account for the failure of eye speed to accelerate to target speed based on unresponsiveness of MT to image motion: pulsing the speed or coherence of the moving dots during steady-state tracking confirmed the responsiveness of area MT throughout pursuit. Instead, we propose a parallel effect of sensory-motor gain signals on the cerebellum’s eye velocity positive motor feedback that normally sustains steady-state eye speed.