Equal numbers of male and female mice were randomly assigned to either a paired (PA), unpaired (UN), shock-only (SO), paired reverse (PA-R), or unpaired reverse (UN-R) group for fear conditioning and fear extinction training (Figure 1). Data from the PA, UN, and SO groups were statistically tested for sex differences and the significant results from these analyses are listed in Table 1. Given that most comparisons did not yield significant differences, data from male and female mice were pooled for statistical comparisons between groups. Additionally, given the minimal sex differences observed within the PA, UN, and SO groups, we reduced the number of subjects in the PA-R and UN-R groups and, therefore, did not statistically test for sex differences.

Behavioral data from the PA, UN, PA-R, and UN-R groups during the second day of fear conditioning (CD2) were compared to observe how conditioned defensive behavior differs based on the associative value and stimulus order of the SCS. A two-way ANOVA was used to analyze the effect of trial and group on freezing during the tone and WN. There was no statistically significant interaction between the trial and group for tone-induced freezing (Figure 2A; F_{(12, 340)}=0.65, p=0.80); however, there was a significant main effect of the trial (F_{(4, 340)}=6.3, p<0.0001) and group (F_{(3, 340)}=19.6, p<0.0001). All groups showed little freezing to the WN (Figure 2B), and no significant interaction between trial and group (F_{(12, 340)}=0.76, p=0.69) or main effect of the trial (F_{(4, 340)}=2.07, p=0.08) were found. We did find a main effect of group (F_{(3, 340)}=3.00, p=0.03), which was attributed to greater freezing from the PA-R group during Trial 1.

An activity index was calculated for each mouse as a combined measurement of cue-induced locomotion with escape jumping (see Methods), and a two-way ANOVA was used to analyze tone- and WN-evoked activity indices (Figure 2C and D). The activity indices for all groups were very low during the tone period (Figure 2C), and there was no significant interaction between the trial and group (F_{(12, 340)}=0.98, p=0.47). The activity index during the tone decreased over trials, concomitant with the observed increase in freezing behavior (main effect of trial, F_{(4, 340)}=3.58, p=0.007). There was a significant effect of group (F_{(3, 340)}=2.7, p=0.045), which was attributed to the UN-R group displaying higher activity during Trial 2. While the WN-evoked activity indices in all groups showed no significant trial by group interaction (F_{(12, 340)}=0.15, p=0.99), or main effect of the trial (F_{(4, 340)}=0.54, p=0.70), a significant main effect of group (Figure 2D, F_{(3, 340)}=9.03, p<0.0001) was observed.

An ordinary one-way ANOVA was used to compare average freezing and activity indices between all groups, and Tukey’s multiple comparisons test was used for post-hoc comparisons. The PA and PA-R group showed significantly higher freezing during the tone than the UN group (Figure 2E, F_{(3, 68)}=9.56, p<0.0001; PA vs UN, p<0.0001; PA-R vs UN, p=0.046), and there was no significant difference between groups in their activity indices during tone (Figure 2F, F_{(3, 68)}=2.14, p=0.10). On the contrary, while there were no differences in WN-evoked freezing between groups (Figure 2G, F_{(3, 68)}=2.62, p=0.058), there was a significant interaction when analyzing activity indices during WN (Figure 2H, F_{(3, 68)}=3.04, p=0.035). However, post-hoc comparisons did not yield significant pairwise differences.

An ordinary one-way ANOVA was used to test for between-group differences in contextual freezing during the initial 3 min of the session preceding the first SCS presentation, and Tukey’s multiple comparisons test was used for post-hoc comparisons. There was a main effect of group for baseline contextual freezing (Figure 2I, F_{(3, 68)}=3.18, p=0.017), but significant differences were found only between the UN-R and SO groups (UN-R vs SO, p=0.013). Welch’s unpaired t-test was used to compare the differences in freezing during the pre-SCS and tone periods to determine the extent to which the tone increased freezing. The PA and PA-R groups showed significantly greater increases in freezing from pre-SCS to tone compared to the UN and UN-R groups (Figure 2J, F_{(3, 68)}=15.92, p<0.0001; PA vs UN, p<0.0001; PA vs UN-R, p=0.0009; PA-R vs UN, p<0.0001; PA-R vs UN-R, p=0.0055).

Given the interaction found in Figure 2H, we performed Welch’s unpaired t-test to conduct pairwise comparisons between average activity indices during WN. WN-evoked activity was higher in the PA group compared to the UN group (Figure 2K, PA vs UN, t_(43.08)=2.36, p=0.023) and the PA-R group (Figure 2L, PA vs PA-R, t_(31.62)=3.89, p=0.0005). Additionally, the PA-R group displayed lower WN-evoked activity than both of the UN groups (PA-R vs UN, t_(20.65)=2.65, p=0.015; PA-R mean ± SEM = 1.86±0.28; UN mean ± SEM = 5.39±1.31) and the UN-R group (PA-R vs UN-R, t_(9.215)=2.75, p=0.022; UN-R mean ± SEM = 8.84±2.52). There was no difference between the UN and UN-R groups with regards to WN-evoked activity (UN vs UN-R, t₍₁₄₎ = 1.214, p=0.24). While the PA and UN groups significantly differed in average activity indices during WN, the PA and UN-R groups did not (PA vs UN-R, t_(30.42)=0.9962, p=0.33; PA mean ± SEM = 12.54±2.73).

Overall, these data show that the respective changes in defensive behavior during tone and WN were significantly affected by the explicit pairing of SCS and shock during fear conditioning, and that the order of tone and WN presentation influenced the intensity of WN-evoked responses.

Although the UN or UN-R groups did not receive an associative pairing between SCS and shock like the PA and PA-R groups, all groups still displayed increased activity indices to WN (Figure 2D). To determine if this behavioral response was due to defensive flight or a more basic locomotor response, we investigated the occurrence of escape jumping and darting behaviors during the WN presentation on CD2. A substantial percentage of PA mice jumped during WN on every trial, and these jumps were distributed across the entire WN period (Figure 3A and B). In contrast, an exceedingly small percentage of the UN group jumped during WN (Figure 3C), and when jumps occurred, they occurred at the onset of the WN (Figure 3D). Both groups responded to shock with a similar number of jumps (Figure 3B and D). Trial-by-trial, PA mice displayed more jumping behavior across the WN period (Figure 3E) compared to UN mice (Figure 3F). Like the UN group, the reverse SCS groups also displayed lower jumping percentages to WN, with jumps in the PA-R group occurring rarely near the transition from WN to tone, (Figure 3G) and jumps in the UN-R group occurring near WN onset (Figure 3H). The PA group had the largest percentage of mice that jumped to WN during CD2 compared to all other groups (Figure 3I). Lastly, all groups exhibited jumps to shock (Figure 3J), with greater percentages of the unpaired and reverse cohorts responding with jumps.

Similar analyses were performed for darting behavior. A high percentage of PA mice showed darting during WN (Figure 4A), darts were specific for the WN, and they were spread across the stimulus period (Figure 4B). Mice in the UN group almost never darted during the tone or WN (Figure 4C and D). Furthermore, PA mice displayed darts across the WN period on every trial (Figure 4E), whereas UN mice did not (Figure 4F). The PA group also had the largest percentage of mice that darted during WN compared to all other groups (Figure 4G). Interestingly, we did not detect darting from the reverse groups during the SCS in CD2. The non-PA groups only rarely expressed escape jumping or darting, yet they did have elevated activity indices (Figure 2D). Therefore, we measured the average distance traveled over the WN period, and we used a one-way ANOVA to determine if these higher activity levels were due to a simpler locomotor response. The groups did not differ in distance traveled during preSCS (Figure 4H, F_{(3, 68)}=0.28, p=0.84), but we did observe differences during tone (Figure 4I, F_{(3, 68)}=3.34, p=0.024) and WN (Figure 4J, F_{(3, 68)}=9.83, p<0.0001). During the tone, the UN group had a significantly greater distance traveled than the PA group (PA vs UN, p=0.014), which is reflective of the elevated freezing during the tone in the PA group (Figure 2E). During the WN, the PA group had greater distance traveled than the UN (PA vs UN, p=0.0018) and the PA-R groups (PA vs PA-R, p<0.0001), but not the UN-R group (PA vs UN-R, p=0.083). Finally, all groups darted to shock in similar percentages (Figure 4K).

In summary, associative pairings of SCS and shock produced significant cue-induced freezing to the tone, as well as robust jumping and darting behaviors that occurred over the entirety of WN presentations. Altering the contingency between WN and shock in the unpaired or reversed SCS conditions profoundly reduced these defensive behaviors, but the increased activity in all groups during WN suggests that inherent properties of the WN interact with non-associative processes to induce locomotor responses. This is reflected in the increased distance traveled during WN in all groups compared to the tone period. Representative behavioral responses of the PA, UN, PA-R, and UN-R groups to the SCS during conditioning are provided in Video 1 and Video 2.

We next analyzed how the defensive ethogram of each group changed over the course of two extinction sessions. A two-way ANOVA was used to analyze the effect of trial and group on freezing during the tone, and Tukey’s multiple comparisons test was used for post-hoc comparisons. When analyzing tone-evoked freezing across extinction within the PA, UN, and SO groups (Figure 5A), a significant interaction between trial and group (F_{(62, 2208)}=2.3, p<0.0001) was found. For every trial except the last, the PA group exhibited a higher level of freezing during tone compared to the control groups (p<0.05 for Trials 1–15, for both sessions). Tone-evoked freezing presented similarly between the PA-R and UN-R groups (Figure 5B), yielding a significant interaction between trial and group (F_{(31, 576)}=1.63, p=0.018). The PA-R group maintained consistently high freezing during the tone across both extinction sessions, freezing more than the UN-R group for nearly all trials (p<0.05 for Trials 1, 3–16 on Ext1, all trials on Ext2).

To quantify the relative change in freezing over each extinction session, we calculated the difference in freezing between the first four trials and the last four trials for each session. An ordinary one-way ANOVA was used to analyze the effect of the group on changes in freezing during the tone, and Tukey’s multiple comparisons test was used for post-hoc comparisons. There was a significant difference between groups on the first day of extinction (Figure 5C, F_{(4, 87)}=11.9, p<0.0001), with the PA-R group being the only one to increase freezing during tone over the session (PA-R vs PA, p<0.0001; PA-R vs UN, p<0.0001; PA-R vs UN-R, p=0.007; PA-R vs SO, p=0.0032). Freezing for the PA and UN groups decreased similarly over the session (PA vs UN, p=0.99), and the SO group had significantly less change in freezing compared to the PA group (PA vs SO, p=0.021), which was attributed to the low level of freezing during the tone in the SO group (Figure 2A). A significant difference between groups was also detected for the second extinction session (Figure 5D, F_{(4, 87)}=19.2, p<0.0001). Only the PA group exhibited a decrease in freezing during tone compared to all other groups (PA vs UN, p<0.0001; PA vs PA-R, p=0.0007; PA vs UN-R, p<0.0001; PA vs SO, p<0.0001).

To determine if freezing during the tone was cue-induced, or was simply a continuation of contextual freezing, we calculated the difference between freezing in the pre-SCS period and freezing during the tone for each extinction session (Figure 5E and F). An ordinary one-way ANOVA was used to analyze the effect of the group, and Tukey’s multiple comparisons test was used for post-hoc comparisons. For both the first (Figure 5E, F_{(4, 87)}=105.5, p<0.0001) and second (Figure 5F, F_{(4, 87)}=137.5, p<0.0001) sessions of extinction, only the PA and PA-R groups increased freezing levels during the tone (p<0.0001 for all PA and PA-R comparisons to other groups for both sessions). The PA-R group showed a greater change in freezing from pre-SCS to tone than the PA group (Ext1, PA vs PA-R, p=0.0003; Ext2, PA vs PA-R, p<0.0001), whereas the UN, UN-R, and SO groups had equivalent freezing during the pre-SCS and tone periods (p>0.25 for all pairwise comparisons that excluded the PA and PA-R groups for both sessions). Taken together, these data suggest that the PA and PA-R groups associated the tone with threat, while freezing in the UN, UN-R, and SO groups was more indicative of contextual fear. Interestingly, while pairing either order of SCS with shock resulted in greater freezing during tone, we observed a phenotype more resistant to extinction within the PA-R group. These data indicate that the extinction of cue-induced freezing in the conditioned flight paradigm depends on its proximity to conditioned threat.

Activity indices were calculated to analyze the effect of extinction learning on behavioral responses during the WN. The PA, UN, and SO groups had elevated activity indices in the early trials of extinction training, yet only the PA group showed a decrease in WN-evoked activity (Figure 6A). A two-way ANOVA was used to analyze the effect of trial and group, and Tukey’s multiple comparisons test was used for post-hoc comparisons. There was a significant interaction between the trial and group (F_{(62, 2208)}=1.9, p<0.0001). Starting with the fifth trial of the first extinction session, the PA group expressed significantly less activity to WN compared to the UN and SO groups (p<0.05 compared to UN and SO for Trials 5, 6, and 8–13), and PA activity index scores remained below 1 for the entirety of the second extinction session (p<0.05 compared to UN and SO for Trials 1, 2, and 4–16). For the UN and SO groups, WN-evoked activity indices remained above 1 across extinction trials, indicating levels of movement that were higher during the WN than the pre-SCS period. There was no interaction between the trial and group (Figure 6B, F_{(31, 576)}=0.7, p=0.89) for the PA-R and UN-R groups. There was an effect of group (F_{(1, 576)}=64.7, p<0.0001), with activity being higher in the UN-R compared to the PA-R group. There was no significant effect of the trial (F_{(31, 576)}=0.66, p=0.92).

To illustrate the change in WN response over extinction, we calculated the difference in speed during WN between the first four and last four trials of each session. An ordinary one-way ANOVA was used to analyze the effect of the group, and Tukey’s multiple comparisons test was used for post-hoc comparisons. The PA group showed a significant decrease compared to the UN-R and SO groups (Figure 6C, F_{(4, 87)}=6.4, p<0.0001; PA vs UN-R, p=0.0052; PA vs SO, p=0.0003) and a near-significant decrease compared to the UN group (PA vs UN, p=0.054). There were no significant differences between groups during the second extinction session (Figure 6D, F_{(4, 87)}=1.47, p=0.22).

Interestingly, as WN-evoked activity decreased during extinction, the PA group developed and maintained a freezing response to WN, while the other groups displayed almost no freezing to WN (Figure 6E). A significant interaction between trial and group was detected (F_{(124, 2784)}=1.3, p=0.019), and the PA group displayed greater freezing than the UN, UN-R, and SO groups for a majority of the first extinction session (p<0.05 for Trials 4, 6, and 10–16) and for most of the second session (p<0.05 for Trials 1–13). Additionally, the PA group froze more during WN than the PA-R group for several trials of the second session (p<0.05 for Trials 2–6, 8–10). When comparing changes in WN-evoked freezing between the first and last four trials for the first extinction session (Figure 6F, F_{(4, 87)}=4.2, p=0.0036), the PA group displayed significant increases compared to the UN and SO groups (PA vs UN, p=0.0086; PA vs SO, p=0.017), but did not differ from the PA-R or UN-R groups (PA vs PA-R, p=0.18; PA vs UN-R, p=0.22). For the second session (Fig 6G, F_{(4, 87)}=8.1, p<0.0001), the PA group exhibited a significant decrease in WN-evoked freezing compared to the UN, UN-R, and SO groups (PA vs UN, p=0.0005; PA vs UN-R, p=0.0006; PA vs SO, p=0.0015), but not the PA-R group (PA vs PA-R, p=0.85). Collectively, these findings show changes in the magnitudes and modes of behavior to WN within the PA group across extinction, indicating that WN-evoked flight in the PA group can be extinguished and is associative.

To determine if the activity measured during extinction was related to defensive flight or a more basic locomotor response, we examined the expression of jumping and darting behaviors between groups. Within the first four trials of the first extinction session, PA mice displayed the most jumping behavior during WN, with the SO group displaying only two jumps occurring near the onset of WN, and the UN group displaying no jumps (Figure 7A). When examining darting behavior within the first four trials of extinction, minimal darting was observed during the tone period, the PA and SO groups displayed darting behavior spread across the WN period, while the UN group darted only a few times (Figure 7B). Jumps and darts were not present within the second extinction session for any group.

During the first extinction session, only the PA group displayed a concentration of jumping responses during WN within the first block of trials (Figure 7C), and jumps to WN rarely occurred in any other group (Figure 7E). All groups displayed variable amounts of darting to WN, with the PA, PA-R, and SO groups having the largest proportions of darters within the session (Figure 7D and F). Notably, the increase in darting in the PA group occurs around Trial 7, which is approximately the time point at which jumping is fully extinguished. Within the PA-R and SO groups, WN-evoked darting is distributed throughout the session.

These data suggest that escape jumping is largely an associative response that switches to darting as the psychological distance of threat increases; however, darting is controlled by associative and non-associative mechanisms. Representative behavioral responses of the PA, UN, and SO groups to the SCS between early and late periods of the first extinction session are provided in Video 3 and Video 4.

We previously observed heightened tail-rattling responses during the early trials of fear conditioning, which decreased with further conditioning (Borkar et al., 2020). Given that tail rattling has been shown to increase in the presence of uncertain threat (Salay et al., 2018), we measured tail rattling in all groups during extinction to determine the effects of associative and non-associative mechanisms on this defensive response. During the first extinction session, tail rattling behavior during SCS presentations was more prevalent in the UN, UN-R, and SO control groups and was most prominent during the tone period (Figure 8A and B). To monitor tail rattling within groups, cumulative behavioral frequencies were taken from the first and last four trials within each extinction session (reverse SCS groups were excluded due to lower N-values). During early extinction, SO and UN mice displayed more tail rattling than PA mice during the tone (Figure 8C). The frequency of tail rattling was lower in all three groups during the second extinction session, yet the UN and SO groups both displayed more than the PA group (Figure 8D). All three groups displayed similar levels of tail rattling by the end of the second session. Over the first extinction session, a larger proportion of the UN and SO groups displayed tail rattle during to tone (Figure 8E), and the UN group had the greatest proportion of tail rattle to WN (Figure 8F).

These data suggest that tail rattling during the SCS is mostly a non-associative defensive response that is suppressed when the SCS predictably signals threat.

We used C57BL/6 J mice (Jackson Laboratory, Bar Harbor, Maine, Stock #000664), aged 3–6 mo in this study. Equal numbers of males and females were used in all experiments. All mice were individually housed on a 12 hr light/dark cycle throughout the study with ad libitum access to food and water. Behavioral experiments were performed during the light cycle. All animal procedures were performed in accordance with institutional guidelines and were approved by the Institutional Animal Care & Use Committee of Tulane University.

Behavioral testing was performed in two contexts. Context A consisted of a 30 cm diameter transparent acrylic cylinder with a smooth acrylic floor, cleaned with 1% acetic acid between each subject. Context B consisted of a modular fear conditioning chamber (ENV-307W, Med Associates, Inc, Fairfax, Vermont) with metal grid flooring and walls of polycarbonate, stainless steel, and polyurethane, cleaned with 70% ethanol solution between sessions. Alternating current footshocks (ENV-414S, Med Associates, Inc) were delivered to the mice during conditioning in Context B. A programmable audio generator (ANL-926, Med Associates, Inc) generated auditory stimuli that were delivered at 75 dB in each context via an overhead speaker (ENV-224AM, Med Associates, Inc). A serial compound stimulus (SCS) was used as previously described (Fadok et al., 2017; Borkar and Fadok, 2021; Borkar et al., 2020). The SCS consisted of ten pips of tone (7.5 kHz, 0.5 ms at 1 Hz) followed by ten pips of white noise (0.5ms at 1 Hz), and the reversed SCS consisted of ten pips of white noise followed by ten pips of tone. Behavioral protocols were generated using Med-PC software (Med Associates, Inc) to control auditory stimuli and shock with high temporal precision.

Mice were randomly allocated to one of five groups: Paired (PA), Paired Reverse (PA-R), Unpaired (UN), Unpaired Reverse (UN-R), and Shock Only (SO). Behavioral testing took place over 5 d. For all days of the paradigm, PA-R and UN-R mice experienced the reversed SCS at identical presentation timings as their respective PA and UN counterparts. On Day 1 (Pre-Exposure), subjects were placed in Context A for a baseline period of 3 min, followed by four presentations of the SCS with a pseudorandom inter-stimulus interval (ISI) period of 90–100 s and a period of 40 s following the final SCS presentation, totaling 590 s per session. Day 2 and Day 3 (Conditioning) took place in Context B. On each conditioning day (CD1 and CD2), mice were subjected to one of three conditions after a 3- min baseline period. For all groups, each conditioning session lasted 820 s. PA mice (n=16 males, 16 females) and PA-R mice (n=5 males, 5 females) were presented with five pairings of the SCS co-terminating with a 1 s, 0.9 mA footshock, with pseudorandom ISI periods of 90–150 s and a period of 60 s following the final footshock of the session. UN mice (n=10 males, 10 females) and UN-R mice (n=5 males, 5 females) were presented with pseudorandom presentations of SCS and footshocks separate from one another with ISI periods of 40–60 s, with a period of 90 s following the final stimulus of the session. Stimuli were ordered such that the SCS could not reliably predict footshock. PA, PA-R, UN, and UN-R mice all received the same number of SCS and footshock presentations, only differing by SCS-footshock contingency. SO mice (n=10 males, 10 females) did not receive presentations of the SCS during conditioning and were given five footshocks with pseudorandom ISI periods of 120–160 s each session, with a period of 80 s following the final shock of the session. For all groups, stimulus timing and ISI differed between CD1 and CD2 to avoid predictable anticipation of stimuli before presentation. Days 4 and 5 (Extinction) took place in Context B, and each session consisted of 16 presentations of the SCS with pseudorandom ISI periods of 60–120 s, with a period of 50 s following the final SCS of the session. Each Extinction session (Ext1 and Ext2) lasted 1910s. Subjects were sacrificed after the conclusion of behavioral testing. Experiments were conducted in cohorts of 8–10 mice for a total of 10 replicates.

All sessions were recorded to video using a camera (Pike, Allied Vision, Stadtroda, Germany) mounted above the behavioral contexts with stimulus events encoded to the same files using TTL pulses (Omniplex, Plexon, Dallas, Texas). Contour tracking (Cineplex, Plexon) was used to automatically detect freezing based on frame-by-frame changes in pixels. Freezing behavior was defined as a complete cessation of movement for at least 1 s, and results were confirmed with an observer blinded to the condition. By determining a calibration coefficient using the known size of the behavioral context and the camera’s pixel dimensions, speed (cm/s) was extracted using the animal’s center of gravity. An activity index was calculated for each animal using a ratio of its speed during either the tone or WN stimulus (CS) period and its average speed from the combined 10 s periods prior to each SCS presentation (pre-SCS) during the session; the number of jumps performed during that stimulus period was then added to this ratio (Speed_CS/Speed_{avg pre-SCS}+Jumps). Previously we calculated flight scores per trial using speed from each trial’s pre-SCS period (Fadok et al., 2017; Borkar et al., 2020), but here we utilized an average from all pre-SCS periods in our calculations to avoid denominators that were close to or equal to 0, a complication noted by other groups (Hersman et al., 2020). Reflecting this change, we now refer to this measure of locomotor change as an ‘activity index’ instead of a ‘flight score’ as before. Escape jumps and tail-rattling behaviors were manually classified by an observer blinded to the condition. Jumps were defined as the period where the mouse had all four paws above the chamber floor. Tail rattling was defined as rapid back-and-forth vibrations of the tip and midsection of the tail. Darting behavior was detected and classified using machine learning software as described below and was defined as rapid bursts of movement across the floor of the chamber. Distance traveled over pre-SCS, tone, and WN periods was calculated per mouse by plotting its average speed per 0.5 s intervals and integrating the area under the curve.

When performing behavioral analyses that reported cumulative frequencies per group, 20 random subjects from the PA group were used to match the population sizes of the UN and SO control groups. Due to the lower number of subjects in the PA-R and UN-R groups, they were excluded from frequency-based comparisons.

Darts were scored using the program Simple Behavior Analysis (Nilsson et al., 2020) to generate a machine learning algorithm capable of automatically detecting the occurrence of the behavior of interest. To generate this model, top-down footage (640×480 pixel resolution, 30 frames per second) of 16 male and female C57BL/6 J mice that underwent SCS fear conditioning in Context B was collected and analyzed in DeepLabCut (DLC) (Mathis et al., 2018) to assign 8 tracking points (Nose, L Ear, R Ear, Center of Mass, L Flank, R Flank, Tail Base, Tail Tip) to subjects. The DLC markerless tracking model was generated using manually assigned points from ~2000 frames trained using the ResNet50 Neural Network for 125,000 iterations. 2370 frames containing darting behavior were identified and added to the training set for SimBA. The darting start point was defined as the first frame in which the mouse began accelerating from a resting position, and the endpoint was defined as the last frame before the mouse returned to a full stop. Once the model was generated, all videos from all subjects were analyzed using a discrimination threshold of.37 and a minimum duration of 266 ms (eight frames).

Sample sizes for each group were justified via power analysis (α=0.05, power = 80%). Data were analyzed for statistical significance using Prism 9 (GraphPad Software Inc, San Diego, California). For all tests, the definition of statistical significance was p<0.05. All data were checked for normal distribution using the Shapiro-Wilk normality test (α=0.05). For pairwise comparisons between groups, an unpaired t-test with Welch’s correction was used to assess behavioral differences since all relevant datasets had normal distributions. One-way analysis of variance (ANOVA) was used to assess behavioral differences between all conditioning groups. Two-way ANOVA was used to assess interactions of time point and conditioning variant between groups, as well as interactions of stimulus and conditioning variant within groups. When either ANOVA yielded significant interactions, Tukey’s post-hoc multiple comparisons test was used to detect significant behavioral differences between groups.