Calcium imaging fiber photometry reveals that both mPFC and VI GABAergic neurons activate during spontaneous exploratory movement.

A, Parasagittal sections showing the optical fiber tract reaching mPFC or VI with GCaMP7f fluorescence expressed in GABAergic neurons around the fiber ending. The sections were aligned with the Allen brain atlas CCF. The right panels show schematics of the locations of optical fiber endings in mPFC and VI. B, ΔF/F calcium imaging time extracted around detected spontaneous movements recorded in mPFC (left) or VI (right). Time zero represents the peak of the movement. The upper traces show ΔF/F mean±SEM of all movement peaks (black), those that had no detected peaks 3 s prior (red), and peaks taken at a fixed interval >5 s (cyan). The lower traces show the corresponding movement speed for the selected peaks. Linear mixed effects model contrasts: (mPFC vs VI [-3 to -0.5 s]: all peaks t(10)=0.39 p=0.70; no peaks 3 s prior: t(10)= 3.43 p=0.0064; peaks > 5 s: t(10)= 3.2 p=0.009). C, Cross-correlation between movement and ΔF/F for the overall (left), translational (middle) and rotational (right) components in mPFC (red) and VI (cyan). D, Per session (dots) and mean±SEM (rectangle) linear fit (correlation, r) between overall movement and ΔF/F in mPFC (left) and VI (right), including the rotational and translational components. The lighter dots show the linear fit after scrambling one of the variables (shuffled). (n=7 mice in mPFC, n=5 mice in VI) All traces and averaged symbols in the paper are mean±SEM. If SEM is not visible, it is smaller than the trace/symbol thickness.

mPFC and VI GABAergic neurons differentially encode exploratory movement turning direction.

A, ΔF/F calcium imaging, overall movement, rotational movement, and angle of turning direction for detected movements classified by the turning direction (ipsiversive and contraversive; red and cyan) versus the side of the recording (implanted optical fiber) in mPFC. At time zero, the animals spontaneously turn their head in the indicated direction. The columns show all turns (left), those that included no turn peaks 3 s prior (middle), and peaks selected at a fixed interval >5 s (right). Note that the speed of the movements was similar in both directions (the y-axis speed is truncated to show the rising phase of the movement). B, Same as A for VI. C, Population measures (area, peak amplitude, and time to peak for traces 3 s around the detected peaks) of ΔF/F and overall movement for the different classified peaks in mPFC. Asterisks denote significant differences (p<0.05) between ipsiversive and contraversive movements. In the mPFC, neuronal activity did not differ between ipsiversive and contraversive turns across any condition (all turns: t(583)= 0.43 p=0.66; no turns 3 sec prior: t(583)= 0.35 p=0.73; turns per >5 sec : t(1157)= 1.5 p=0.13). D, Same as C for VI. VI neuron activity showed modest direction selectivity. Contraversive movements elicited significantly greater ΔF/F responses for all turns (t(583)= 3.6 p=0.0004), isolated turns per >5 sec (t(1157)= 3.31 p=0.0009), but not for turns following quiescence (no turns 3 sec prior: t(583)= 1.8 p=0.07). (n=7 mice in mPFC, n=5 mice in VI)

Behavioral performance in the cue-driven action tasks.

A, Arrangement of the shuttle box and tasks. The top table details the sequence of four tasks and the contingencies for each of the three CSs per task. Note that the tones do not apply to the final unsignaled US task (US, aversive unconditioned stimulus consisting of foot shock plus white noise). In this task, the full US, the foot shock component, or the white noise component are presented randomly and elicit a fast escape crossing (see text for further details). In the other three tasks (noUS, AA19, and AA39), three auditory tone conditioned stimuli (CS1–CS3) were presented randomly at three intensities (low, medium, high). During noUS, all CSs were neutral and crossings terminated the CS without consequence (Ig, Ignore contingency). During AA19, CS1 added the Active Avoid (AA) contingency; CS2 and CS3 remained neutral. During AA39, CS2 added the Passive Avoid (PA) contingency, while CS1 retained the AA contingency and CS3 remained neutral. The bottom table describes the trial structure for the three task contingencies (AA, PA, Ig). Each trial consists of a random intertrial interval (ITI), followed by a CS action interval. In the AA contingency this is followed by the US, if the mouse did not cross (actively avoid), which causes a fast crossing (escape, AA error). The PA contingency requires the mouse not to cross and applies a short US (0.5 s) if the mouse crosses (PA error). B. Behavioral performance across three sequential procedures: noUS, AA19, and AA39. Panels show percentage of actions (top), action latency (middle), and intertrial crossings (ITCs; bottom). Data are shown for combined mPFC and VI groups after confirming no group effect (χ²(1) = 0.95, p = 0.32). Other model factors (Task, CS, Tone Intensity) showed significant main effects and interactions (p < 0.0001). noUS. The number of spurious actions did not differ across CS1–CS3. Within each CS, tone intensity significantly modulated responding, with higher action rates at high compared to low intensity (t(2316) > 2.4, p < 0.05). CS2 at high intensity elicited more actions than CS2 at medium intensity (t(2315) = 5.6, p < 0.0001). AA19. Action rates during CS1 were higher than during noUS for all intensities (low: t(1585) = 13.02, p < 0.0001; medium: t(1575) = 15.50, p < 0.0001; high: t(1585) = 16.89, p < 0.0001). Actions during CS2 and CS3 did not differ from noUS at low or medium intensity but increased at high intensity (CS2: t(1594) = 6.97, p < 0.0001; CS3: t(1585) = 3.93, p = 0.012). Within AA19, action rates during CS1 exceeded those during CS2 (t(2315) = 24.5, p < 0.0001) and CS3 (t(2315) = 25.4, p < 0.0001), while CS2 and CS3 did not differ. Action latencies did not differ between noUS and AA19 for any CS. AA39. Action rates during CS1 exceeded those during CS2 and CS3 (CS1 vs. CS2: t(2316) = 21.45, p < 0.0001; CS1 vs. CS3: t(2316) = 28.6, p < 0.0001). Compared to AA19, overall responding decreased across all CSs (CS1: t(574) = 8.6, p < 0.0001; CS2: t(577) = 3.5, p < 0.0001; CS3: t(576) = 7.26, p < 0.0001). Actions during CS1 declined at all intensities (low: t(1599) = 6.07, p < 0.0001; medium: t(1599) = 5.43, p < 0.0001; high: t(1599) = 6.93, p < 0.0001), as did actions during CS3 (low: t(1606.7) = 4.47, p = 0.0012; medium: t(1599) = 4.29, p = 0.0026; high: t(1607) = 6.79, p < 0.0001). Actions during CS2 were reduced at low intensity (t(1611) = 3.4, p = 0.0018), but not at medium (t(1603) = 2.09, p = 0.10) or high intensity (t(1611) = 1.9, p = 0.052). CS2 elicited more actions than CS3 (t(2315) = 7.13, p < 0.0001). Action latencies increased in AA39 relative to AA19 for CS1 (t(1001) = 8.27, p < 0.0001) and CS2 (t(1222) = 7.5, p < 0.0001), but not for CS3 (t(1198) = 1.16, p = 0.68). CS1 latencies increased at medium (t(2160) = 5.51, p < 0.0001) and high intensity (t(2166) = 6.59, p < 0.0001), but not at low intensity (t(2170) = 3.15, p = 0.38). CS2 latencies increased at low (t(2222) = 5.53, p < 0.0001) and medium intensity (t(2210) = 4.34, p = 0.0042), but not at high intensity (t(2171) = 3.68, p = 0.061). ITCs. ITC frequency decreased across task phases from noUS to AA19 (t(271) = 2.78, p = 0.0057) and from AA19 to AA39 (t(271) = 3.34, p = 0.0018). During noUS, ITC rates did not differ among CSs. During AA19, ITCs were lower following CS1 than CS2 (t(2315) = 8.4, p < 0.0001) or CS3 (t(2315) = 4.37, p < 0.0001), and CS2 ITCs exceeded CS3 (t(2315) = 4.1, p < 0.0001). A similar pattern was observed in AA39, except that ITCs following CS2 and CS3 did not differ. C, Same data as in B, shown averaged across tone intensities.

Movement traces (overall speed) across task phases (noUS, AA19, AA39), cues (CS1–CS3), and cue intensities (low, medium, high), separated by Action (A) and No-Action (B) outcomes for all mPFC and VI mice.

Dashed boxes indicate correct or error responses for the active avoidance (AA) and passive avoidance (PA) contingencies; traces outside the boxes correspond to spurious actions or correctly ignored cues for the Ig contingency. Note that escapes are AA errors. Each panel overlays the three cue intensities. Some traces are intentionally truncated to highlight specific features, but the full trace is shown in C. C, Movement for the Action trials in A, replotted aligned to the action occurrence (From-Action is defined as the time at which the animal crosses the compartments door).

Marginal means of movement from mixed-effects models.

Population-level marginal means of movement speed estimated from linear mixed-effects models across all mice. Left panels (A) compare trial types across task phases; right panels (B) compare Action versus No-Action outcomes within each task and CS. Asterisks in left panels denote significant differences relative to the preceding task phase. Asterisks in right panels denote significant differences between Action and No-Action outcomes. Each panel row corresponds to one model window: baseline (pre-CS), orienting (0–0.5 s post-CS), action (0.5–7 s post-CS), and from-action (−2 to 2 s relative to crossing). Baseline window. Baseline movement differed across task phases and action outcomes. There were significant main effects of Task (χ²(2) = 52.0, p < 0.0001) and Outcome (χ²(1) = 199.9, p < 0.0001), as well as CS × Outcome (χ²(2) = 13.9, p = 0.001) and Task × Outcome (χ²(2) = 37.5, p < 0.0001) interactions. During noUS, baseline speed was higher on Action than No-Action trials across all CSs (t(3052) > 7, p < 0.0001). In AA19, baseline speed was higher for Action trials during CS2 (t(3082) = 4.44, p < 0.0001) and CS3 (t(3078) = 3.74, p = 0.00018), but not during CS1 (t(3116) = 0.38, p = 0.7). In AA39, baseline speed differed between Action and No-Action trials for CS3 (t(3096) = 7.43, p < 0.0001), CS2 (t(3098) = 4.1, p < 0.0001), and CS1 (t(3067) = 2.4, p = 0.015). Baseline movement decreased from noUS to AA19 (t(239) = 5.76, p < 0.0001) but did not differ between AA19 and AA39 (t(246) = 1.26, p = 0.2). Orienting window. Orienting responses varied as a function of task and tone intensity. A significant Task × Tone Intensity interaction was observed (χ²(4) = 25.5, p < 0.0001), with a Task × Tone Intensity × CS interaction (χ²(8) = 16.2, p = 0.04). At high intensity, orienting responses increased from noUS to AA19 for CS1 (t(3071) = 3.21, p = 0.0039) and CS3 (t(2861) = 3.4, p = 0.0019), but not for CS2 or at lower intensities. Action window. Movement during the action window showed main effects of CS (χ²(2) = 25.7, p < 0.0001) and tone intensity (χ²(2) = 73.4, p < 0.0001). Significant CS × Task (χ²(4) = 26.0, p < 0.0001) and CS × Tone Intensity (χ²(4) = 35.8, p < 0.0001) interactions were observed. Movement differed strongly between Action and No-Action trials (χ²(1) = 1274, p < 0.0001), with significant interactions between Outcome, CS, Task, and Tone Intensity (all p < 0.0001). From-Action window. From-Action is defined as the time at which the animal crosses the door. Movement aligned to the action event varied by CS (χ²(2) = 135.84, p < 0.0001), Task (χ²(2) = 165.79, p < 0.0001), and Tone Intensity (χ²(2) = 39.89, p < 0.0001). A CS × Task interaction was present (χ²(4) = 259.42, p < 0.0001), along with a CS × Task × Tone Intensity interaction (χ²(8) = 18.05, p = 0.02). noUS Action trials. During noUS, movement did not differ among CSs when intensities were pooled. At high intensity, CS2 differed from other CSs during noUS but not during AA19. AA19 Action trials. During AA19, movement during CS1 Action trials was greater than during noUS spurious actions (t(1871) = 8.82, p < 0.0001). Smaller increases were observed for CS2 (t(2284) = 4.2, p < 0.0001) and CS3 (t(2048) = 3.31, p = 0.0028), restricted to medium and high intensities. Within AA19, CS1 Action trials differed from CS2 (t(3102) = 3.14, p = 0.0033) and CS3 (t(3090) = 3.92, p = 0.00027). AA39 Action trials. In AA39, movement during CS2 Action trials differed from CS2 Action trials in AA19 (t(2653) = 5.91, p < 0.0001), with effects strongest at medium and high intensities. Movement during CS1 Action trials did not differ between AA19 and AA39, whereas CS3 Action trials decreased relative to AA19 (t(2381) = 3.22, p = 0.0028). Within AA39, CS2 Action trials differed from CS1 Action trials (t(3136) = 5.6, p < 0.0001). Comparable effects were observed in from-action–aligned analyses. AA39 No-Action trials. During AA39, movement during No-Action CS2 trials differed from No-Action CS2 trials in AA19 (t(1710) = 2.96, p = 0.0062). A similar difference was observed for CS1 No-Action trials (t(2408) = 2.63, p = 0.017), whereas no change was observed for CS3 No-Action trials (t(1466) = 0.23, p = 0.81).

Simple (AA19) and challenging (AA39) avoidance contingencies mixed-effects models of ΔF/F activity controlling for movement and baseline activity.

Population estimates of marginal means from ΔF/F fiber-photometry mixed-effects models for AA19 (A, B) and AA39 (C, D) tasks, shown separately for mPFC neurons (A, C) and VI neurons (B, D). Each panel set contains four rows corresponding to the baseline, orienting, action (CS-aligned), and from-action model windows. Within each row, the left and middle panels compare contingencies (Ignore, Active Avoid, Passive Avoid; Ig, AA, PA) for Action and No-Action trials, and right panel compare Action vs. No-Action outcomes. Transparency in the Action window indicates that movement was not controlled between outcomes in that model. A, B, AA19 models included Contingency (Ig, AA), Outcome, Group, and covariates. Baseline window. A significant Contingency × Outcome interaction was observed (χ²(1) = 12.82, p = 0.00034). No Contingency × Group or three-way interaction was detected. Orienting window. A significant Outcome × Group interaction was detected (χ²(1) = 5.25, p = 0.022). Action window. Two-way interactions were present among factors, with a significant Contingency × Group interaction (χ²(1) = 9.25, p = 0.0023). No three-way interaction was detected. Spurious actions during Ig trials did not differ from active avoids during AA trials in either region. From-Action window. (From-Action is defined as the time at which the animal crosses the door) A significant Contingency × Outcome × Group interaction was observed (χ²(1) = 9.56, p = 0.0019). No differences were detected between spurious actions and active avoids in either region. C, D, AA39 models included Contingency (Ig, AA, PA), Outcome, Group, and covariates. Baseline window. A significant Contingency × Outcome interaction was detected (χ²(2) = 20.77, p < 0.0001). No interaction with Group was observed. Orienting window. Significant Contingency × Outcome (χ²(2) = 24.87, p < 0.0001) and Contingency × Group interactions were detected (χ²(2) = 31.58, p < 0.0001), with no three-way interaction. Action window. Significant two-way interactions were observed, including a Contingency × Group interaction (χ²(2) = 24.14, p < 0.0001), as well as a significant Contingency × Outcome × Group interaction (χ²(2) = 6.35, p = 0.041). Spurious actions during Ig trials did not differ from active avoids in either region. Correct passive avoids differed from correct ignores in mPFC (t(1301) = 4.14, p < 0.0001). From-Action window. A significant Contingency × Outcome × Group interaction was detected (χ²(2) = 11.78, p = 0.0027). Active avoids differed from spurious actions in mPFC (t(1362) = 2.52, p = 0.011) but not in VI. E, Mean ΔF/F (top) and movement (bottom) traces for mPFC and VI neurons during the unsignaled escape task, showing interleaved trials of footshock alone, white noise (WN) alone, and the full US, aligned to US onset. F, Population estimates of marginal means from ΔF/F mixed-effects models (including movement covariates) during the escape window (0-4 s from US onset). In the baseline window, a strong effect of baseline movement was observed (χ²(1) = 14.31, p = 0.00015), with no main effects or interactions involving US or Group. During the escape window, significant effects of escape movement (χ²(1) = 12.09, p = 0.00052), baseline movement (χ²(1) = 4.15, p = 0.041), and baseline ΔF/F (χ²(1) = 140.14, p < 0.0001) were detected. After accounting for covariates, a significant main effect of US (χ²(2) = 33.02, p < 0.0001) and a US × Group interaction were observed (χ²(2) = 11.38, p = 0.0034). In mPFC, footshock alone differed from the full US (t(105) = 3.89, p = 0.00034) and white noise alone (t(105) = 6.13, p < 0.0001); no corresponding contrasts were significant in VI.

ΔF/F fiber photometry traces and corresponding movement traces for mPFC mice across task phases (noUS, AA19, AA39) and cues (CS1–CS3), separated by Action (A) and No-Action (B) outcomes.

For each cue (CS1, CS2, CS3; left, middle, right panels), traces from the three task phases are overlaid. Movement traces for each condition are shown in the lower panels. Arrows indicate correct or error responses for the active avoidance (AA) and passive avoidance (PA) contingencies. Some traces are intentionally truncated to highlight specific features, but the full trace is shown in C. C, Traces for the Action trials in A, replotted aligned to the action occurrence (From-Action is defined as the time at which the animal crosses the door).

Full mixed-effects models of ΔF/F activity across task phases controlling for movement and baseline activity.

Population estimates of marginal means from full mixed-effects models of ΔF/F fiber-photometry signals are shown for mPFC (A, B) and VI (C, D) GABAergic neurons. Models include all task phases (noUS, AA19, AA39) and incorporate covariates for window-specific movement, baseline movement, and baseline ΔF/F (excluded from the baseline window). Random effects were sessions nested within mice. Left panels (A, C) show comparisons across task phases for each CS and outcome; right panels (B, D) show comparisons between Action and No-Action outcomes within each task and CS. Asterisks in left panels denote significant differences relative to the preceding task phase; asterisks in right panels denote significant differences between Action and No-Action outcomes. Rows correspond to the baseline, orienting, action (CS-aligned), and from-action analysis windows (From-Action is defined as the time at which the animal crosses the door). Shaded bars in the action window indicate that movement was controlled within, but not between, outcomes. Baseline window. Baseline ΔF/F was strongly influenced by baseline movement (χ²(1) = 318.62, p < 0.0001). After controlling for movement, a significant Task × CS × Outcome interaction was observed (χ²(4) = 26.16, p < 0.0001), while the four-way Task × CS × Outcome × Group interaction was marginal (χ²(4) = 8.18, p = 0.085). Between-group contrasts revealed higher baseline ΔF/F in mPFC compared to VI for CS3 during noUS spurious-action trials (t(266) = 2.27, p = 0.024) and No-Action trials (t(50) = 2.48, p = 0.016). Within mPFC, baseline ΔF/F differed between Action and No-Action trials for CS1 and more prominently for CS3 during AA39. Across tasks, only CS3 showed a significant increase in baseline ΔF/F for Action trials from AA19 to AA39. In VI, baseline ΔF/F was higher for Action than No-Action trials across multiple CSs and task phases, including CS1 in noUS, CS2 in AA19, and CS3 in noUS and AA39. This relationship reversed for CS1 in AA19, where higher baseline VI activity preceded escapes (t(3086) = 3.36, p = 0.00078). Baseline ΔF/F changed in opposite directions for active avoids and escapes across AA19. Orienting window. Orienting-related ΔF/F was significantly modulated by orienting movement (χ²(1) = 96.09, p < 0.0001), baseline movement (χ²(1) = 44.08, p < 0.0001), and baseline ΔF/F (χ²(1) = 401.6, p < 0.0001). After accounting for these covariates, the Task × CS × Outcome interaction was marginal (χ²(4) = 8.87, p = 0.061), and the corresponding four-way interaction including Group was not significant (χ²(4) = 5.05, p = 0.28). Between-group contrasts revealed higher orienting ΔF/F in mPFC than VI for CS2 during AA39 correct passive avoids (t(33) = 2.64, p = 0.012). In mPFC, orienting ΔF/F differed between escapes and active avoids for CS1 and between spurious-action and No-Action trials for CS3 in AA19. Across tasks, orienting ΔF/F increased for CS2 No-Action and CS3 spurious-action trials from noUS to AA19, and for CS1 escapes and CS2 Action and No-Action trials from AA19 to AA39. In VI, significant outcome-related differences were limited to CS3 during AA39, and task-related increases were observed only for CS1 and CS3 No-Action trials from noUS to AA19. Action window. ΔF/F during the action window was strongly influenced by action-window movement (χ²(1) = 306.78, p < 0.0001), baseline movement (χ²(1) = 196.08, p < 0.0001), and baseline ΔF/F (χ²(1) = 1874.7, p < 0.0001). After controlling for movement within outcome categories, a significant Task × CS × Outcome interaction was observed (χ²(4) = 10.16, p = 0.037), along with a significant Task × CS × Outcome × Group interaction (χ²(4) = 10.88, p = 0.027). Between-group contrasts showed higher action-window ΔF/F in mPFC than VI for CS2 correct passive avoids (t(23) = 3.65, p = 0.0013) and CS1 escapes (t(22) = 3.50, p = 0.002) during AA39. In mPFC, ΔF/F increased across task phases for CS1 as active avoids and escapes emerged between noUS and AA19, and for CS2 with the emergence of passive avoid outcomes in AA39. CS3 also showed increases for both Action and No-Action trials from noUS to AA19. In VI, ΔF/F increases were observed for CS1 with the emergence of escapes in AA19 and for CS2 with the emergence of passive avoid errors in AA39, while CS3 showed increases primarily in No-Action trials. Differences between Action and No-Action trials were larger in VI than mPFC in this window. From-Action window. From-action ΔF/F was significantly modulated by from-action movement (χ²(1) = 154.73, p < 0.0001), baseline movement (χ²(1) = 40.47, p < 0.0001), and baseline ΔF/F (χ²(1) = 96.45, p < 0.0001). After controlling for these covariates, significant Task × CS × Outcome (χ²(4) = 18.09, p = 0.0019) and Task × CS × Group interactions were observed (χ²(4) = 22.65, p = 0.00015), with a marginal four-way interaction (χ²(4) = 9.6, p = 0.047). Between-group contrasts revealed higher from-action ΔF/F in VI than mPFC for CS2 (t(47) = 2.35, p = 0.022) and CS3 (t(38) = 3.20, p = 0.0027) spurious actions during noUS. In mPFC, from-action ΔF/F was higher for escapes than active avoids in AA trials during both AA19 and AA39. ΔF/F also increased for CS2 spurious actions from noUS to AA19 and for CS2 passive avoid errors in AA39 relative to spurious actions in AA19. In VI, outcome-related differences in the from-action window were limited, with enhanced activation during escapes observed only in AA39.

Movement-related neuron classes during a simple avoidance context (AA19 task).

A, Mean ΔF/F traces (top) and corresponding movement traces (bottom) for five mPFC GABAergic neuron classes identified by k-means clustering of the cross-correlation functions between speed and ΔF/F (inset a). Class 1 (black) and Class 2 (red) exhibited moderate positive correlations with movement, Class 3 (green) showed negative correlations, and Classes 4 (blue) and 5 (cyan) exhibited strong positive correlations, with Class 5 centered near zero lag. Traces are aligned to CS onset, for CS1–CS3, separated by contingency (Ignore, Ig; Active Avoid, AA) and behavioral outcome (action vs No-Action), aligned to CS onset. In AA19, CS1 signals active avoidance and an Action outcome is the correct response. B, Same as A, aligned from-action (defined as the time at which the animal crosses the door) for Action outcomes (no traces exist from-action for No-Action trials). C,D, Estimated marginal means of ΔF/F from linear mixed-effects models for the baseline, orienting, action, and from-action windows. Models compared contingencies (Ig vs AA) and outcomes (action vs No-Action) across neuron classes. C, contrasts across contingencies within outcomes. D, contrasts between outcomes within each contingency and neuron class. Transparency in the action window indicates that movement was not controlled between outcomes in that model. Baseline window. The model revealed a significant Contingency × Outcome interaction (χ²(1) = 44.01, p < 0.0001) and a significant Contingency × Outcome × Class interaction (χ²(4) = 11.21, p = 0.024), indicating class-dependent differences in pre-CS activity across contingencies and outcomes. In Classes 1 and 2, baseline ΔF/F differed between correct active avoids and spurious actions, while in Classes 1, 2, and 4, baseline activity differed between escapes and correctly ignored CSs. Orienting window. There was a significant main effect of neuron Class (χ²(4) = 124.33, p < 0.0001), indicating differences in orienting-related ΔF/F across classes. The Contingency × Outcome interaction with Class was weak (p = 0.08). In Class 3, orienting-related activity differed between active avoids and spurious actions. In Ig trials, orienting-related activity differed between spurious actions and correctly ignored CSs in Classes 2, 3, and 5. Action window. The model revealed a significant Contingency × Outcome interaction (χ²(1) = 20.47, p < 0.0001), with no significant three-way interaction involving Class (p = 0.08). Across classes, ΔF/F differed between escapes and correctly ignored CSs, except in Class 1. Differences between active avoids and spurious actions were class dependent, with opposite patterns observed in Classes 2 and 3. Comparisons between Action and No-Action outcomes are shown for completeness but are not controlled for movement in this window. From-Action window. A significant Contingency × Outcome × Class interaction was observed (χ²(4) = 17.33, p = 0.0016), indicating class-specific differences in action-aligned activity. Only Class 5 showed a difference between active avoids and spurious actions. In most other classes, ΔF/F differed between escapes and active avoids.

Movement-related neuron classes during a difficult avoidance context (AA39 task).

A, ΔF/F traces (top) and movement traces (bottom) for five mPFC GABAergic neuron classes identified by k-means clustering of cross-correlation functions between speed and ΔF/F (inset a). Class 1 (black) and Class 2 (red) neurons show moderate positive correlations with movement. Class 3 (green) neurons show negative correlations with movement. Class 4 (blue) and Class 5 (cyan) neurons show strong positive correlations, with Class 5 activity preceding Class 4 around zero lag. Traces are aligned to CS onset and shown separately for CS1–CS3 and behavioral outcome (action vs. No-Action), which may reflect correct responses, errors, or spurious actions depending on contingency. In AA39, CS1 signals active avoidance and an action is the correct response, whereas CS2 signals passive avoidance and an action is an error. B, Same as A, aligned from-action (defined as the time at which the animal crosses the door) for Action outcomes. C, Marginal mean ΔF/F estimates from linear mixed-effects models for the baseline, orienting, action, and from-action windows, comparing Ignore (Ig), Active Avoid (AA), and Passive Avoid (PA) contingencies across outcomes (action vs. No-Action) for each neuron class. D, Same as C, but comparing outcomes within each contingency for each neuron class. Transparency in the action window indicates that movement was not controlled between outcomes in that model. Baseline window. The mixed-effects model revealed a significant Contingency × Outcome interaction (χ²(1) = 194.01, p < 0.0001) and a Contingency × Outcome × Class interaction (χ²(4) = 66.83, p < 0.0001). Orienting window. A significant Contingency × Outcome interaction (χ²(2) = 106.17, p < 0.0001) and a significant Contingency × Outcome × Class interaction (χ²(8) = 81.89, p < 0.0001) were observed. Action window. The model revealed a significant Contingency × Outcome interaction (χ²(2) = 179.6, p < 0.0001) and a significant Contingency × Outcome × Class interaction (χ²(8) = 240.7, p < 0.0001). From-action window. A significant Contingency × Outcome × Class interaction was observed (χ²(8) = 257.09, p < 0.0001).

Distinct mPFC GABAergic neuron activation types during active avoidance modes.

A, k-means clustering of ΔF/F time series from individual mPFC GABAergic neurons within each active avoidance mode (defined in Fig. 5–Supplement 1) revealed three activation types (a–c). Traces are aligned to CS onset. Type a (1–3a) exhibited minimal modulation during avoidance. Type b (1–3b) showed a rapid increase in activity at CS onset followed by a gradual decline. Type c (1–3c) exhibited selective activation around the time of the avoidance action. Activation types are shown separately for each avoidance mode. Type b activation was strongest for Mode 1 and demodulated in Mode 3, while Type c activation was more robust in Modes 2 and 3. B, Same as A, aligned from-action (defined as the time at which the animal crossed the door to avoid). C, Population estimates of marginal means (ΔF/F) from linear mixed-effects models for the baseline, orienting, action, and from-action windows, comparing activation Types (a–c) across the three active avoidance modes. Baseline window. No differences across activation types or avoidance modes. Orienting window. Type b activation differed across avoidance modes (Mode 1 vs Mode 3: t(5748) = 2.67, p = 0.02). Action window. Type a and Type c activation was higher during Mode 2 (Type a: t(4871) = 3.62, p = 0.0008) and Mode 3 (Type a: t(5169) = 3.57, p = 0.0008) compared to Mode 1, whereas Type b showed no mode-dependent differences. From-Action window. Type b activation was lower during Mode 3 compared to Mode 2 (t(4356) = 4.69, p < 0.0001) and Mode 1 (t(4867) = 2.83, p = 0.0091). Type c activation was higher during Mode 2 (t(4727) = 12.03, p < 0.0001) and Mode 3 (t(4958) = 10.74, p < 0.0001) compared to Mode 1.

Activation of mPFC GABAergic neurons across three distinct signaled active avoidance modes.

A, k-means clustering of movement speed time-series from CS onset (gray panel) revealed three distinct active avoidance modes. Mode 1 avoids were initiated earlier after CS onset, whereas Mode 2 and Mode 3 avoids were delayed and faster (peak speed vs Mode 1: p < 0.0001), reflecting increasingly cautious responding. The top panel shows the average ΔF/F activity of all recorded mPFC neurons for each avoidance mode. mPFC activation was weaker during Mode 1, compared to Mode 2 and Mode 3. B, Same as A, aligned from-action (avoid) occurrence. C, Marginal means (ΔF/F) from the linear mixed-effect models for the baseline, orienting, avoidance, and from-action windows for the three active avoidance Modes. Action window. Mode 2 (t(4912) = 6.37, p < 0.0001) and Mode 3 (t(5246) = 4.86, p < 0.0001) avoids exhibited higher activation than Mode 1 avoids, and Mode 2 tended to have higher activation than Mode 3 avoids (t(4698) = 2.04, p = 0.04). Fron-Action window. Mode 2 vs Mode 1: t (4955) = 7.85, p < 0.0001; Mode 3 vs Mode 1: t(5236) = 4.27, p < 0.0001; Mode 2 vs Mode 3: t(4441) = 4.29, p < 0.0001.

Optogenetic inhibition of mPFC neurons has little effect on signaled avoidance.

A, Effects of continuous green light at three different powers (10-30 mW) on learning AA1 (black circles) followed by AA2 (red circles) in naïve mice expressing eArch3.0 in mPFC neurons. In AA1, the CS signals the active avoidance contingency. In AA2, ITCs are punished. On every trial, light was delivered during the CS and US to inhibit mPFC activity. Mice learned and performed the task like control mice, which are not shown (but see D, E). During AA2, performance remained unchanged, although mice exhibited the typical shift toward longer avoidance latencies. Catch trials (NoCs) consisted of blank trials without light or consequence for crossing. B, Same as A, but during subsequent AA3 task, in which CS1, CS1+Light, CS2, and CS2+Light trials were presented randomly. Light trials included the three powers used in A. CS1 signals the active avoidance (AA) contingency, whereas CS2 signals the passive avoidance (PA) contingency. C, Optogenetic fiber placements in the mPFC and effect of blue light on neural activity in Vgat-Cre ChR2 mice. Top left, the schematic shows the range of depth placements tested in the mPFC of Vgat-Cre ChR2 mice in the sagittal plane (eArch3.0 fibers were around the dorsal portion of the range). Bottom, shows two examples of dorsal and ventral fibers from DAPI-stained sections. Example PSTH shows the effect of blue light in Vgat-Cre mice expressing ChR2 in GABAergic neurons recorded in freely behaving mice with an optrode implanted in mPFC. Continuous blue light for 1 s produces virtually complete sustained abolishment of multi-unit activity (MUA). The histogram is the average of 15 light trials. The same effects were observed in every mouse tested (n=3). D, Effects of continuous blue light in mPFC on learning AA1 followed by AA2 in naïve mice expressing ChR2 in GABAergic neurons. The columns represent groups of mice with optical cannulas in mPFC or VI, with a third No Opsin group with optical fibers in mPFC that did not express ChR2. The AA1 and AA2 tasks are like those in A, including catch trials. The right panel shows the contrasts between AA1 and AA2 for each group. E, Effect of blue light in the same three groups during AA3.