Attentional modulation of persistent visual representations after stimulus offset is coordinated across primary visual and anterior cingulate cortices

Flexible behavior in complex environments requires animals to infer which sensory inputs are relevant for guiding action and which need to be ignored. Yet, how cognitive control circuits apply contextual knowledge during decision-making is poorly understood. To address this question, we trained head-fixed mice on a cross-modal, context-dependent two-alternative forced-choice (2-AFC) decision-making task. In this task, animals simultaneously received visual and auditory inputs, and the behaviorally relevant modality switched covertly between blocks of trials, requiring animals to infer rule changes from reward feedback alone. Mice successfully adapted their behavior after rule switches and recovered to expert performance. Using simultaneous Neuropixels recordings in the primary visual cortex (V1) and the anterior cingulate cortex (ACC) during task performance, we found that neuronal responses in both regions showed mixed selectivity, multiplexing sensory, choice, and outcome signals. However, V1 neural responses more strongly encoded visual stimulus orientation, while ACC neural responses preferentially encoded outcome. Population decoding further revealed that V1 and ACC neurons maintained a persistent representation of task context throughout the trial, occupying a subspace approximately orthogonal to visual stimulus coding. This contextual representation strengthened progressively as mice adapted their behavior following each switch. Critically, context representations emerged in parallel in V1 and ACC, and decoding accuracies in these two regions were strongly correlated with each other and with behavioral performance across trial blocks. Orthogonal to this shared emergent and sustained context signal, visual stimulus decoding was enhanced when vision was task-relevant. This attentional enhancement in V1 occurred not only during the stimulus presentation period but also after stimulus offset. Similarly, in ACC, a persistent increase of visual stimulus identity encoding in visually relevant session blocks was present after stimulus offset. Taken together, these findings reveal a distributed, inter-areal representation of task context spanning sensory and frontal cortices, suggesting that V1 and ACC contribute to maintaining task-relevant sensory information beyond the immediate sensory response period to provide a circuit-level mechanism for cross-modal attentional control.