Multi-Stable Bimodal Perceptual Coding within the Ventral Premotor Cortex

Neurons of the primate ventral premotor cortex (VPC) respond to tactile or acoustic stimuli. But how VPC neurons process and integrate information from these two sensory modalities during perception remains unknown. To investigate this, we recorded the activity of VPC neurons in two trained monkeys performing a bimodal detection task (BDT). In the BDT, subjects reported the presence or absence of a tactile or an acoustic stimulus. Initial single-cell analyses revealed a diverse range of responses during the BDT: purely tactile, purely acoustic, bimodal and others that exhibited sustained activity during the decision maintenance delay—between the stimulus offset and motor report. To further explore the VPC’s role in the BDT, we applied dimensionality reduction techniques to uncover the low-dimensional latent dynamics of the neuronal population and conducted parallel analyses on a recurrent neural network (RNN) model trained on the same task. Neural trajectories associated with tactile responses diverged strongly from those related to acoustic responses. Conversely, during the stimulus-absent trials the neural dynamics remained at rest. During the delay, the trajectories demonstrated a pronounced rotational dynamic toward a subspace orthogonal to the sensory response space, supporting memory maintenance in stable equilibria. This suggests that the network dynamics can sustain distinct stable states corresponding to the three potential task outcomes. Using low-dimensional modeling, we propose a universal dynamical mechanism underlying the transition from sensory to mnemonic processing, consistent with our experimental and computational observations. These findings show that the VPC contains neurons capable of bimodal coding and that its population can integrate competing sensory information and maintain decisions throughout the delay period, regardless of the sensory modality.