Previously, we developed a novel model for anxiety during motivated behavior by training rats to perform a task where actions executed to obtain a reward were probabilistically punished and observed that after learning, neuronal activity in the ventral tegmental area (VTA) and dorsomedial prefrontal cortex (dmPFC) represent the relationship between action and punishment risk (Park & Moghaddam, 2017). Here we used male and female rats to expand on the previous work by focusing on neural changes in the dmPFC and VTA that were associated with the learning of probabilistic punishment, and anxiolytic treatment with diazepam after learning. We find that adaptive neural responses of dmPFC and VTA during the learning of anxiogenic contingencies are independent from the punisher experience and occur primarily during the peri-action and reward period. Our results also identify peri-action ramping of VTA neural calcium activity, and VTA-dmPFC correlated activity, as potential markers for the anxiolytic properties of diazepam.

Subjects

Male and female Long-Evans (bred in house n=8) and Sprague-Dawley (Charles River n=5) rats were used. Animals were pair-housed on a reverse 12 h:12 h light/dark cycle. All experimental procedures and behavioral testing were performed during the dark (active) cycle. All studies included both strains of male (n=7) and female (n=6) rats. All experimental procedures were approved by the OHSU Institutional Animal Use and Care Committee and were conducted in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals.

Initial Training & Punishment Risk Task (PRT)

The PRT follows previously published methods (Park & Moghaddam, 2017; Chowdhury et al., 2019). Rats were trained to make an instrumental response to receive a 45-mg sugar pellet (BioServe) under fixed ratio one schedule of reinforcement (FR1). The availability of the nosepoke for reinforcement was signaled by a 5-s tone. After at least three FR1 training sessions, PRT sessions began. PRT sessions consisted of three blocks of 30 trials each. The action-reward contingency remained constant, with one nose-poke resulting in one sugar pellet. However, there was a probability of receiving a footshock (300 ms electrical footshock of 0.3 mA) after the FR1 action, which increased over the blocks (0%, 6%, or 10% in blocks 1, 2 and 3, respectively). To minimize generalization of the action-punishment contingency, blocks were organized in an ascending footshock probability with 2-min timeouts between blocks. Punishment trials were pseudo-randomly assigned, with the first footshock occurring within the first five trials. All sessions were terminated if not completed in 180 mins.

Fiber Photometry Analysis

Peri-event analysis: Signals from the 465 (GCaMP6s) and 560 (tdTomato) streams were processed in Python (Version 3.7.4) using custom-written scripts similar to previously published methods (Jacobs & Moghaddam, 2020). Briefly, 465 and 560 streams were low pass filtered at 3 Hz using a butterworth filter and subsequently broken up based on the start and end of a given trial. The 560 signal was fitted to the 465 using a least-squares first order polynomial and subtracted from 465 signal to yield the change in fluorescent activity (ΔF/F= 465 signal - fitted 560 signal/ fitted 560 signal). Peri-event z-scores were computed by comparing the ΔF/F after the behavioral action to the 4-2 sec baseline ΔF/F prior to a given epoch. To investigate potential different neural calcium responses to receiving the footshock vs. anticipation, punished (i.e. shock) trials and unpunished trials were separated. Trials with a z-score value > 40 were excluded. From approximately 3,000 trials analyzed, this occurred on < 1% of trials.

Area under the curve (AUC) analyses: To represent individual data we calculated the AUCs for each subject. To quantify peri-cue and peri-action changes we calculated a change or summation score between 1 sec before (pre-event) and 1 sec after (post-event) cue onset or action execution. For the reward period, we calculated a change score by comparing 2 sec after reward delivery to the 1 sec prior to reward delivery. For punished trials, response to footshock was calculated as the change from 1 sec following footshock delivery compared to the 1 sec before footshock. Outliers were removed using GraphPad Prism’s ROUT method (Q=1%; Motulsky & Brown, 2006) which removed only three data points from the analysis.

Time Lagged Cross-Correlation Analysis: Cross-correlation analysis has been used to identify networks from simultaneously measured fiber photometry signals (Sych et al., 2019). For rats with properly placed fibers in the dmPFC and VTA, correlations between photometry signals arising in the VTA and dmPFC were calculated for the peri-action, peri-footshock and peri-reward periods using the z-score normalized data. The following equation was used to normalize covariance scores for each time lag to achieve a correlation coefficient between -1 and 1:

    Coef = Cov/(s¹*s²*n)

Where Cov is the covariance from the dot product of the signal for each timepoint, s¹ and s² are the standard deviations of the dmPFC and VTA streams, respectively, and n is the number of samples. An entire cross-correlations function was derived for each trial and epoch.

Comparison to Electrophysiology Results: Fiber photometry data for the third PRT session were compared to the average of the 50 msec binned single unit data (see Figure 4 of Park & Moghaddam, 2017). This third PRT session corresponds to the session electrophysiology data were collected from. To overlay data from the two techniques, data were lowpass filtered at 3 Hz and photometry data were downsampled to 20 Hz (to match the 50 msec binning). Data from both streams were then min-max normalized between 0 and 1 at the corresponding cue and action+reward epochs.

To assess the similarity of the two signals, we performed a Pearson correlation analysis between the normalized single unit and fiber photometry data for cue or action+reward epochs at each risk block, as well as between randomly shuffled photometry signals with single-unit response as a control. For significant Pearson correlations, we performed cross-correlation analysis (see above) to investigate if the photometry signal lagged behind electrophysiology given the slower kinetics of GCAMP6 compared to single-unit approaches (Chen et al., 2013).

Statistical Analysis

For FR1 training, trial completion was measured as the number of food pellets earned. Data were assessed for the first 3-4 training sessions. Action and reward latencies were defined as time from cue onset to action execution or from food delivery until retrieval, respectively. Values were assessed using a mixed-effects model with the training as a factor and post-hoc tests were performed using the Bonferroni correction where appropriate.

For the PRT, trial completion was measured as the percentage of completed trials (of the 30 possible) for each block. Action latencies were defined as time from cue onset to action execution. Data were analyzed using a two-way RM ANOVA or mixed effects model. Because there were missing data for non-random reasons (e.g. failure to complete trials in response to punishment risk) we took the average of risk blocks (blocks 2 and 3) and the no-risk block (block 1) to permit repeated measures analysis. We used mixed effects model if data were missing for random reasons. Risk and session were used as factors and post-hoc tests were performed using the Bonferroni correction where appropriate. When only two groups were compared a paired t-test or Wilcoxon test was performed after checking normality assumption through the Shapiro-Wilk test.

To assess changes in neural calcium activity, we utilized a permutation-based approach as outlined in (Jean-Richard-dit-Bressel et al., 2020) using Python (Version 3). An average response for each subject for a given time point in the cue, action, or reward delivery period was compared to either the first PRT or saline session. For each time point, a null distribution was generated by shuffling the data, randomly selecting the data into two groups, and calculating the mean difference between groups. This was done 1,000 times for each time-point and a p-value was obtained by determining the percentage of times a value in the null distribution of mean differences was greater than or equal to the observed difference in the unshuffled data (one-tailed for comparisons to 0% risk and FR1 data, two-tailed for all other comparisons). To control for multiple comparisons we utilized a consecutive threshold approach based on the 3 Hz lowpass filter window (Jean-Richard-dit-Bressel et al., 2020; Pascoli et al., 2018), where a p-value < 0.05 was required for 14 consecutive samples to be considered significant.

To assess AUC changes in photometry data, we compared all risk blocks and all sessions using ANOVA with factors risk block and session. Because not all subjects completed learning and diazepam data, we used an ordinary two-way ANOVA. Significant main effects and interactions were assessed with post-hoc Bonferroni multiple comparison tests.

To assess correlated activity changes as a function of risk or session, we took the peak and 95% confidence interval for the overall cross-correlation function. These values were compared by a two-way ANOVA with factors risk and session and utilized a post-hoc Bonferroni correction.

Other than permutation tests, all statistical tests were done using GraphPad Prism (Version 8) and an α of 0.05. Results for all statistical tests and corresponding figures can be found in Table 1 or supplemental figures.

Excluded Data

Outliers from latency analysis were removed when a data point was > 5 SDs above the mean across all blocks. This removed one data point from the analysis. In FR1 studies, data from one rat’s third and fourth session were excluded because the camera became misaligned with the patch cord and thus the last (fifth) FR1 session was used for analysis. In PRT studies, data from the dmPFC of one session for a rat was excluded due to lack of timestamp collection and one block of a session was excluded for two other rats because the control 560-nm LED failed for the dmPFC.

Four rats with VTA placement were excluded because fibers were placed outside the VTA or GCaMP6s expression was not observed. Several rats did not complete all phases of the experiment due to lost fiber implants, leaving the final sample sizes as n=9 and n=7 for dmPFC in learning and diazepam treatment stages, respectively, and n=4 for VTA in learning and diazepam treatment stages. 

Analysis Scripts

Data were analyzed using custom-written scripts in Python or R. Please see https://github.com/MoghaddamLab/Jacobs2022-eLife for analysis codes.

Source data may be opened using Microsoft Excel. One could also read them into Python or R which are open source.