Experimental design, behavior, histology and calcium imaging.

a. Experimental design illustrating discriminative fear conditioning with either a 15 kHz CS⁺ or a 3 kHz CS⁺ (experimental groups), and no-shock controls tested at identical time points but never exposed to footshock. All groups were tested with 3 and 15 kHz tones and two intermediate frequencies (7 and 11 kHz) on days 1, 15 and 30 after conditioning. b. Behavioral responses across testing days (CS⁺15 kHz: n = 27 mice; CS⁺3 kHz: n = 22 mice; no-shock controls: n = 13 mice). Two-way repeated-measures ANOVA: CS⁺15 : effect of day F(2,52) = 13.28, p < 0.001, frequency: F(3,78) = 85.09, p < 0.001, day × frequency interaction: F(6,156) = 3.79, p = 0.002; CS⁺3: effects of day F(2,42) = 14.42, p < 0.001, frequency: F(3,63) = 58.81, p < 0.001, day × frequency interaction: F(6,126) = 1.41, p = 0.217; no-shock controls: effects of day: F(2,20) = 1.38, p = 0.276, frequency: F(3,30) = 1.43, p = 0.253; day × frequency interaction: F(6,60) = 0.335, p = 0.916. Significant Tukey multiple comparisons are denoted by asterisks, *** p < 0.001. c. Representative histological section showing GCaMP6f expression. Imaging depth did not exceed 300 μm, restricting recordings to PL. d. Example calcium imaging data showing raw fluorescence signals, deconvolved activity, thresholded events and corresponding activity traces.

Distribution of stable and dynamic cell types across sessions.

a. Activity plots from an animal trained with a 15 kHz CS+, ordered by activity level. Activity plots illustrate positive sound-responsive neurons (bottom), negative sound-responsive neurons (top), and mixed or non-responsive neurons (middle) before and after sound presentation. b. Proportions of neuronal response types across experimental groups and testing days. Number of cells: CS+15: conditioning: 3,834; day 1: 3,177; day 15: 4,186; day 30: 3,693, CS+3 conditioning: 1,628; day 1: 1,381; day 15: 1,881; day 30: 1,833, control: conditioning: 1,118; day 1: 986; day 15: 1,386; day 30: 563 (2 mice only yielded data up to day 15).c. Venn diagrams showing the proportions of consistently active neurons (neurons active across all retrieval sessions), partially active neurons (active in two sessions), and transiently active neurons (active in a single session). The proportions of cells in the different categories were similar across groups, with the exception of temporary cells on Day 30, which were more abundant in the control group. However, this difference reached significance only for the comparison between the control and CS+15 groups (p < 0.05; Table S1). Diagrams and analyses for control mice include only mice recorded for 30 days.

Population activity of positive sound responders shows emotional graded threat-value patterns of activity in response to tones.

a-c. Population responses in animals trained with a 15 kHz CS+ (a), a 3 kHz CS+ (b), and no-shock controls (c). In all groups, upper panels show positively responsive neurons and lower panels negatively responsive neurons. Boxplots show the median (center line), interquartile range (box), and whiskers extending to ±1.5× the interquartile range; points outside the whiskers represent individual observations beyond this range. CS+15: positively modulated: day 1: F(3,18) = 9.963, p < 0.001; day 15: F(3,18) = 9.973, p < 0.001; day 30: F(3,18) = 6.627, p = 0.003; negatively modulated: day 1: F(3,18) = 2.483, p = 0.094; day 15: F(3,18) = 1.877, p = 0.178; day 30: F(3,18) = 2.753, p = 0.073). CS+3: positively modulated: day 1: F(3,12) = 6.899, p = 0.006; day 15: F(3,12) = 9.247, p = 0.002; day 30: F(3,12) = 6.123, p = 0.009; negatively modulated: (day 1: F(3,12) = 0.87, p = 0.484; day 15: F(3,12) = 0.512, p = 0.641; Day 30: F(3,12) = 1.448, p = 0.278); no shock control: positively modulated: day 1: F(3,15) = 0.527, p = 0.670; day 15: F(3,15) = 1.852, p = 0.181; day 30: F(3,9) = 1.046, p = 0.418; negatively modulated: day 1: F(3,13) = 1.205, p = 0.347; day 15: F(3,13) = 1.375, p = 0.294; day 30: F(3,9) = 0.95, p = 0.457. Significant Tukey multiple comparisons are denoted by asterisks, *p < 0.05, **p < 0.01, ***p < 0.001.

Generalized linear model (GLM) analysis of the relative contributions of tone identity and freezing behavior to neuronal activity.

a. Regression coefficients (β) for positive tone-selective neurons in conditioned mice exhibit opposite monotonic gradients across tones after controlling for freezing. This graded pattern is absent in non-shock controls. b. The freezing-to-tone ratio remains below 1 for every tone in all groups, indicating that tone identity explains more variance than freezing in positive tone-selective neurons. c. The proportion of freezing-dominant neurons is low across all tones, indicating that freezing does not dominate neuronal responses. d. When pooled across tones, the proportion of freezing-dominant neurons remains stable across retrieval sessions. e. Negative tone-selective neurons exhibit suppressive tone coefficients across all frequencies with only weak frequency dependence, indicating the absence of pronounced graded responses. f. Freezing-to-tone ratios are higher than in positive responders but remain below 1 across tones in conditioned mice. In non-shock controls, the ratio approaches 1, indicating that freezing contributes similarly to tone identity. g. The proportion of freezing-dominant neurons is higher than in positive responders but remains a minority across all tones and groups. h. The proportion of freezing-dominant neurons remains relatively stable across retrieval sessions for each group. Controls displayed a lightly higher proportion of freezing-dominant neurons but this difference was not significant. Significant Tukey multiple comparisons are denoted by asterisks, *p < 0.05, **p < 0.01, ***p < 0.001.

Population vector similarity across tones and time.

a-c. Population similarity maps for all tone pairs across the time course of sound presentation in the CS+15 (a), CS+3 (b), and no-shock control (c) groups. d. Schematic illustrating the color scale and map orientation; the y-axis denotes the earlier tone in the comparison, and the x-axis denotes the later tone. e-f. Box plots showing population similarity during the first 5 s following tone onset, quantified relative to the CS+ for the CS+15 (e, F(3,36) = 12.025, p < 0.001) and CS+3 (f, F(3,24) = 7.435, p = 0.004) groups. Significant Tukey multiple comparisons are denoted by asterisks, *p < 0.05, **p < 0.01, ***p < 0.001.

Clustering of PL subpopulations based on mutual information.

a. Schematic of the mutual information (MI)–based clustering pipeline. An unsorted MI matrix computed from simultaneously recorded prelimbic (PL) neurons was first subjected to spectral clustering to identify primary clusters based on shared information structure, independent of response sign. The MI matrix was then reordered according to these primary cluster assignments. Within each primary cluster, MI values were combined element-wise with a corresponding sign matrix encoding the direction of correlation between cell pairs, yielding a signed MI matrix. Spectral clustering was subsequently applied independently within each primary cluster to identify secondary subclusters with distinct signed interaction patterns. Each subcluster was assigned a unique label, and all labels were combined to generate the final clustered signed MI matrix, enabling separation of positively and negatively modulated sound-responsive neurons. b. Final clustered signed MI matrices for each experimental and control groups. Matrices are sorted by cluster labels for the CS+15 group (top), CS+3 group (middle), and No Shock group (bottom). Color scale indicates signed MI strength (HI to LO). c. Average stimulus-aligned population responses for clusters showing strong positive modulation to individual tones. C.1–C.4 show primary responses to 3 kHz, 7 kHz, 11 kHz, and 15 kHz tones, respectively, across groups. d. Average stimulus-aligned population responses for clusters showing strong negative modulation to individual tones. D.1–D.4 show primary responses to 3 kHz, 7 kHz, 11 kHz, and 15 kHz tones, respectively, across groups.

Graded-tone responsive neurons are only present in experimental groups.

a-b. Average stimulus-aligned population responses for clusters showing graded tone responses in animals trained with a 15 kHz CS+, showing positive (a) and negative (b) response patterns. c-e. Average stimulus-aligned population responses for clusters showing graded tone responses in animals trained with a 3 kHz CS+, showing positive (c-d) and negative (e) response patterns. f-g. Analysis of stability of neuronal identity within graded tone clusters across retrieval sessions. Benjamini–Hochberg corrected significance denoted by asterisks. h-i. Baseline/stimulus firing rate ratio (BSR) showing changes in activity of graded clusters only present in the experimental CS+15 (h) and CS+3 (i) groups across days. Significant Tukey multiple comparisons are denoted by asterisks, *p < 0.05, **p < 0.01, ***p < 0.001.

GLM analysis of clustered neurons.

a–h, Positive (a–d) and negative (e–h) single tone-selective neurons. a. Tone β coefficients in conditioned mice form opposite gradients peaking at the CS+ that are absent in non-shock controls. b. The freezing-to-tone ratio remains below 1 across preferred tones and retrieval sessions. c. Within-cell comparisons confirm that tone coefficients exceed freezing coefficients in all groups. d. The proportion of freezing-dominant neurons is low and stable across retrieval sessions. e. Negative tone-selective neurons exhibit suppressive β coefficients with weak frequency dependence and no graded responses in controls. f. The freezing-to-tone ratio remains below 1 but is higher than in positive responders and more variable in non-shock controls. g. Tone responses exceed freezing responses but by a smaller margin than in positive responders. h. The proportion of freezing-dominant neurons is higher than in positive responders but remains stable across retrieval sessions. i–p. Positive (i–l) and negative (m–p) graded neurons. i. Freezing-controlled tone β coefficients exhibit strong CS+-centered gradients. j. The freezing-to-tone ratio is well below 1, indicating that tone identity explains substantially more variance than freezing. k. Within-cell comparisons confirm that tone coefficients greatly exceed freezing coefficients. l. Only a small, stable fraction of positive graded neurons is freezing-dominant. m. Negative graded neurons exhibit strong suppressive gradients. n. The freezing-to-tone ratio remains below 1. o. Tone responses exceed freezing responses by approximately threefold. p. The proportion of freezing-dominant neurons remains low across retrieval sessions. Significant Tukey multiple comparisons are denoted by asterisks, *p < 0.05, **p < 0.01, ***p < 0.001.

(a–c) Discrimination ratios (DRs) were used to quantify discrimination between each frequency and the CS+ across testing days in the CS+15 group (a), CS+3 group (b), and control animals (c). In both experimental groups, discrimination increased over time, particularly between the CS+ and the CS− and between the CS+ and the frequency adjacent to the CS−. Discrimination between the CS+ and its adjacent frequency also increased across testing days in the CS+3 group but not in the CS+15 group (Tukey’s multiple comparisons). In contrast, control animals showed no consistent changes in discrimination over time. Repeated-measures ANOVAs with frequency comparison and testing day as factors revealed significant effects of frequency comparison, testing day, and their interaction in the CS+15 group (frequency: F(2,52) = 44.362, p < 0.001; day: F(2,52) = 4.822, p = 0.012; interaction: F(4,104) = 3.421, p = 0.011). In the CS+3 group, significant effects of frequency comparison and testing day were observed, but not their interaction (frequency: F(2,42) = 24.415, p < 0.001; day: F(2,42) = 5.499, p = 0.008; interaction: F(4,84) = 0.23, p = 0.921). No significant effects were detected in control animals (frequency: F(2,20) = 1.21, p = 0.319; day: F(2,20) = 0.767, p = 0.477; interaction: F(4,40) = 0.154, p = 0.960). Asterisks denote Tukey’s multiple comparisons: p < 0.05; ** p < 0.01; *** p < 0.001.

Sex differences in behavior.(a-c) Sex differences in CS⁺15-trained mice.

No sex differences were observed on any testing day (effect of sex: day 1, F(1,25) = 1.52, p = 0.23; day 15, F(1,25) = 0.55, p = 0.467; day 30, F(1,25) = 0.17, p = 0.681). On all days, there was a significant main effect of frequency, indicating successful learning in both males and females (day 1, F(3,75) = 33.46, p < 0.001; day 15, F(3,75) = 26.32, p < 0.001; day 30, F(3,75) = 51.47, p < 0.001). No sex × frequency interactions were detected on any testing day (p > 0.05). Post hoc comparisons showed that on day 1, mice discriminated 3 kHz from 7, 11 and 15 kHz (p < 0.05), but generalized among the higher frequencies (p > 0.05). On days 15 and 30, mice discriminated 3 kHz from 7, 11 and 15 kHz (p < 0.05) and 7 kHz from 11 and 15 kHz (p < 0.05), while generalization persisted between 11 and 15 kHz (p > 0.05). (d-f) Sex differences in CS⁺3-trained mice. No main effect of sex was observed on any testing day (day 1, F(1,20) = 1.24, p = 0.279; day 15, F(1,20) = 0.78, p = 0.388; day 30, F(1,20) = 0.003, p = 0.956). A significant main effect of frequency was present across all days, indicating learning in both sexes (day 1, F(3,60) = 40.65, p < 0.001; day 15, F(3,60) = 30.17, p < 0.001; day 30, F(3,60) = 35.19, p < 0.001). Post hoc comparisons indicated discrimination among all frequencies across days (p < 0.05), except between 11 and 15 kHz (p > 0.05). Significant sex × frequency interactions were observed on day 1 (F(3,60) = 6.91, p < 0.001) and day 15 (F(3,60) = 3.55, p < 0.02), reflecting sex-specific differences at 11 kHz on day 1 and at 3 kHz on day 15. These interaction effects were not present on day 30 (F(3,60) = 0.76, p = 0.522) and likely reflect increased behavioral variability rather than stable sex differences.

Evaluation of cell registration quality.

(a) Distribution of equivalent neuronal footprint diameters (µm) for all registered neurons. (b) Representative spatial footprints recorded on Day 1 (red), Day 15 (green), and Day 30 (blue), along with the merged image showing neurons tracked throughout the experiment (white). (c) Total number of neurons recorded in each experimental group and the percentage of neurons successfully registered across all sessions. Note that two control animals were excluded because they lost the GRIN lens after Day 15. (d–f) Distributions of centroid shifts between matched neurons in the retrieval sessions relative to the conditioning session for the CS+3 (d), CS+15 (e), and control (f) groups. The red line indicates the median equivalent neuronal footprint diameter (21.4 µm). The upper left corner of each panel shows the median centroid shift and the percentage of registered neurons with centroid shifts smaller than the median footprint diameter.

Population activity of consistently active neurons.

(a-b) Population activity of consistently active neurons from animals trained with a 15 kHz CS+ (a) or a 3 kHz CS+ (b). Upper panels show positively tone-responsive neurons and lower panels show negatively tone-responsive neurons. Adjacent boxplots display areas under the population response curves (AUC). Boxplots show the median (center line), interquartile range (box), and whiskers extending to ±1.5× the interquartile range; points outside the whiskers represent individual observations beyond this range. Consistently active positive responders exhibit graded responses across test days (day 1 to day 30). *p < 0.05; **p < 0.01.

Population activity of emerging-retained neurons.

Emerging-retained neurons become part of the active ensemble after conditioning and remaining active after that. (a-b) Population activity of emerging-retained neurons from animals trained with a 15 kHz CS+ (a) or a 3 kHz CS+ (b). Upper panels show positively tone-responsive neurons and lower panels show negatively tone-responsive neurons. Adjacent boxplots display areas under the population response curves. Emerging-retained neurons show greater variability on day 1 compared with later test days. Only positive responders show significant graded valence. *p < 0.05; **p < 0.01.

Population activity of transiently active neurons.

Transiently active neurons were active on only a single test day. (a-b) Population activity of transiently active neurons from animals trained with a 15 kHz CS+ (a) or a 3 kHz CS+ (b). Upper panels show positively tone-responsive neurons and lower panels show negatively tone-responsive neurons. Adjacent boxplots display areas under the population response curves. Transiently active neurons did not show graded population responses on day 1; graded responses emerged by day 15 and were maintained through day 30 in positive sound responder neurons. *p < 0.05; **p < 0.01.

Clustering of PL subnetworks based on signed mutual information for the CS+ 15 kHz group per testing session.

(a-b) Average stimulus-aligned population responses for clusters showing positive (a) or negative (b) modulation to individual tones (3, 7, 11, or 15 kHz, upper panels) or graded emotional tuning (a-b, bottom panels).

Clustering of PL subnetworks based on signed mutual information for the CS+ 3 kHz group per testing session.

(a-b) Average stimulus-aligned population responses for clusters showing positive (a) or negative (b) modulation to individual tones (3, 7, 11, or 15 kHz, upper panels) or graded emotional tuning (a-b, bottom panels).

Clustering of PL subnetworks based on signed mutual information for the no shock control per testing session.

(a-b) Average stimulus-aligned population responses for clusters showing positive (a) or negative (b) modulation to individual tones (3, 7, 11, or 15 kHz).

a–d, Baseline-to-stimulus firing rate ratio (BSR) illustrating changes in positively responding neurons within tone-specific clusters shown in Fig. 5, for the CS+15 (red), CS+3 (blue), and control groups. (a) BSR of neurons in clusters primarily responsive to 3 kHz (Fig. 5c.1), (b) 7 kHz (Fig. 5c.2), (c) 11 kHz (Fig. 5c.3), and (d) 15 kHz (Fig. 5c.4). ANOVA results are reported in Table S3; Tukey’s multiple-comparison tests indicate significance (p < 0.05; p < 0.01; p < 0.001).

Statistical analyses corresponding to the Venn diagrams shown in Fig. 2.

One-way ANOVAs were used when the assumptions of normality and homogeneity of variance were met; otherwise, Kruskal–Wallis tests were performed. Significant Kruskal–Wallis tests were followed by Dunn’s multiple-comparisons tests. Effect sizes are reported as partial eta squared (ηp²) for ANOVAs and epsilon squared (ε²) for Kruskal– Wallis tests.

Statistics comparing areas under the curve for the CS+15, CS+3, and No shock control groups across all frequencies and testing sessions.

(a-b) Three-way mixed model ANOVAs using Group (CS+15, CS+3, No shock), Frequency (3, 7, 11, 15 kHz), and Time (days 1, 15, 30) as variables for positive (a) and negative (b) cell responders. Simple effects were calculated using Holm correction.

Statistics corresponding to consistently active, emerging-retained, and transiently active cells.

One-way ANOVAs with repeated measures were used when normality and equal variance tests were passed to test the effect of frequency. Friedman tests were used (χ² with Kendall’s W) when assumptions were violated, Friedman tests were used (χ² with Kendall’s W).

Statistical analysis of GLM-derived measures for cells used in the population curves shown in Fig. 4.

Two-way ANOVAs with repeated measures were conducted to evaluate effects of group (CS+15, CS+3, or No shock), frequency (3, 7, 11, 15 kHz) or day (days 1, 15, 30), and their interactions. Tukey post hoc comparisons are shown where simple effects were tested.

Statistical analyses identifying neuronal clusters that remained stable over time, defined as neurons that retained the same functional response profile on days 15 and 30 as on day 1.

These analyses correspond to the clusters shown in Figs. 6 and 7, as indicated in the table. For each cluster, the observed proportion of neurons that preserved their response profile was compared with a null distribution, with multiple comparisons controlled using the Benjamini– Hochberg procedure. Only the graded neuronal clusters in each experimental group (Fig. 7) exhibited significant stability across time. No graded clusters were identified in the no-shock control group.

Statistical analyses corresponding to the normalized baseline-to-stimulus firing rate (BSR) for the neuronal clusters shown in Figs. S7–S9.

BSRs for the clusters shown in Fig. S7-S9 are presented in Fig. S10, whereas BSRs for the graded clusters shown in Figs. S7-S8 are presented in Figs. 7h–i. For pure tone-selective clusters, two-way repeated-measures ANOVAs were used to evaluate the effects of group (CS+15, CS+3, or No Shock), frequency (3, 7, 11, and 15 kHz), and their interaction. For graded clusters, one-way repeated-measures ANOVAs were used to evaluate the effect of time. Tukey’s multiple-comparisons tests are reported in Figs. S10 and 7h–i.

Statistical analysis of positive and negative modulated single-frequency and graded neurons shown in Fig. 8.

Two-way ANOVAs with repeated measures were conducted to evaluate effects of group (CS+15, CS+3, or No shock), frequency (3, 7, 11, 15 kHz), day, condition (tone vs. freezing), and their interactions. Tukey post hoc comparisons are shown where simple effects were test