Encoding without generation in the medial prefrontal cortex during cautious goal-directed actions

Adaptive behavior under threat requires deciding when to act and when to withhold action to avoid harm, often under conditions where movement, arousal, and task demand covary. Medial prefrontal cortex (mPFC) activity is widely associated with such control, yet it remains unclear whether this activity reflects causal action generation or broader evaluative processes shaped by behavioral state. Here, we combined fiber photometry, single-cell calcium imaging, mixed-effects modeling, and optogenetic inhibition to examine how GABAergic neurons in mouse mPFC represent cues, actions, and outcomes during a series of learned avoidance tasks of increasing complexity that promote cautious responding. By explicitly controlling for baseline activity and movement, we show that much apparent task-related activity in mPFC reflects movement and cue-evoked signals that are also present in a control cortical region, the visual cortex. mPFC GABAergic neurons showed little encoding of simple avoidance contingencies but broadly encoded punished outcomes. A small subset of neurons with strong movement sensitivity encoded more demanding avoidance contingencies requiring selection between action generation and deferment. For equivalent avoidance actions, distinct neuronal populations preferentially encoded either cue onset or the action. Despite this encoding, optogenetic inhibition of mPFC had minimal effects on the learning or performance of the different contingencies. These findings reveal a dissociation between neural encoding and causal necessity, indicating that mPFC GABAergic activity primarily reflects evaluative and contextual aspects of cautious avoidance behavior rather than direct control of action execution. Significance statementAvoiding danger often requires deciding when to act and when to hold back. The dorsomedial prefrontal cortex (mPFC) is widely assumed to support this control, yet the contributions of its neurons have remained unclear. Using neural population and single-neuron recordings together with targeted inhibition in mice performing learned threat-guided tasks, we show that GABAergic mPFC neurons are activated by cues, actions, and outcomes, but are not required for executing the behavior itself. These findings suggest that the mPFC primarily evaluates and contextualizes threat-motivated actions rather than generating them, highlighting a common mismatch between neural encoding and causal necessity. Synopsis for ReviewersNote the following organizational features: O_LIGiven the hierarchical structure of the experiments, all analyses use linear mixed-effects models, with sessions nested within mice as random effects. Models include relevant covariates, such as movement and baseline activity, to disentangle their contributions from task contingency-related effects. C_LIO_LITo facilitate reading flow, statistical details supporting the Results are reported in the figure legends. Additional statistical values can be provided or relocated to the main text if preferred. C_LIO_LIFigures and their legends are placed adjacent to the corresponding results to improve readability, but high-resolution versions are also at the end. In all population plots, the symbols, and traces are Mean{+/-}SEM. If error bars are not visible, they are smaller than the symbol and trace. We can adjust this as requested. C_LIO_LISeveral analyses are presented as supplemental figures to streamline the main narrative; figure order and presentation can be adjusted as requested. C_LI The Results are organized into six segments: O_LImPFC and visual cortex GABAergic neurons are sensitive to movement. Using fiber photometry, we show that GABAergic neurons in both mPFC and a control cortical region, visual cortex (VI), exhibit strong sensitivity to movement (Fig. 1; Fig. 1-S1). These findings motivated the inclusion of movement and baseline activity as covariates in neural analyses to control their effects. C_LIO_LIBehavioral performance across a series of avoidance tasks. Mice were trained in tasks of increasing difficulty (outlined in Fig. 2A). Animals were first exposed to three neutral cues (noUS), followed by cues predicting different contingencies. In AA19, CS1 signals active avoidance, and then in AA39, CS2 signals passive avoidance. Behavioral performance is shown in Fig. 2, associated movement traces in Fig. 2-S1, and mixed-effects models of movement dynamics in Fig. 2-S2. C_LIO_LIFiber photometry reveals limited encoding of simple avoidance but encoding of aversive outcomes. Fiber photometry recordings from mPFC and VI during the tasks show that mPFC GABAergic neurons do not robustly encode simple avoidance contingencies but exhibit sensitivity to more complex task demands and punished errors. Analyses also tested neural responses to unsignaled aversive stimulation. Results are shown in Fig. 3, with photometry traces in Fig. 3-S1 and a full model spanning all task phases in Fig. 3-S2. C_LIO_LISingle-cell recordings reveal movement-sensitive classes with distinct contingency encoding. Miniscope imaging was performed in mPFC during AA19 and AA39. Neurons were classified based on their correlation with movement and then tested for contingency encoding. This revealed class-specific differences, including a subset of neurons that encode avoidance-related variables. Results for AA19 are shown in Fig. 4, and for AA39 in Fig. 4-S1. C_LIO_LIAvoidance actions segregate into distinct movement modes with dissociable neural encoding. Avoidance actions were classified based on their movement profiles, revealing three response modes that differ in response latency and vigor, reflecting varying degrees of behavioral caution. Clustering neural activity within each mode identified neurons that selectively encode cue onset versus action execution (Fig. 5 and Fig. 5-S1). C_LIO_LIOptogenetic inhibition of mPFC has minimal effects on avoidance behavior. Using two complementary optogenetic approaches (eArch3.0 and Vgat-ChR2), we tested the causal contribution of mPFC neurons to avoidance behavior. Inhibition produced minimal effects on learning or performance across tasks. Behavioral effects and optogenetic validation are shown in Fig. 6. C_LI