Brain architecture of punishment learning

Abstract

Learning from punishment allows animals to suppress actions that produce adverse consequences while maintaining other rewarded behaviors. However, the brain mechanisms of this learning are poorly understood. Here, we combined instrumental behavioural analysis, whole-brain Fos mapping, spatial transcriptomics, computational network analysis, and chemogenetic inhibition in male and female mice. A within-subjects yoking procedure showed that suppression depended on the instrumental response–punisher contingency rather than matched shock exposure or embedded Pavlovian stimulus–shock relations. Whole-brain Fos network analysis showed marked reorganization of brain-wide Fos correlation structure after punishment learning. Punishment preserved modular, small-world organization characteristic of brain networks while reallocating regional community membership and increasing the centrality of the basolateral amygdala, zona incerta, and midbrain tegmentum. Spatial transcriptomics within these regions identified punishment-associated transcriptional programs in basolateral amygdala glutamatergic neurons, zona incerta GABAergic neurons, and multiple ventral midbrain GABAergic and dopaminergic populations. In silico deletion predicted that the basolateral amygdala, zona incerta, and rostral linear nucleus jointly support punishment learning. Consistent with this, multisite chemogenetic inhibition of these regions impaired punishment learning. Single-region inhibition revealed dissociable contributions of basolateral amygdala and zona incerta to within-session and between-session retention of punishment learning. Together, these findings show that punishment learning is supported by a brain network that enables animals to selectively suppress actions that produce adverse consequences. Significance statement Punishment learning is essential for adaptive behaviour because it allows animals to stop actions that produce harm while maintaining other rewarded actions. We show that this form of learning is not explained by shock exposure, Pavlovian fear, or activation of a single brain region. Instead, punishment learning reorganises brain-wide activity networks, recruits spatially structured transcriptional programs in specific neuronal populations, and depends on the function of the basolateral amygdala, zona incerta, and rostral linear nucleus of the raphe. These findings provide a multiscale account of how the brain learns from adverse consequences.