Representational changes of latent strategies in rat medial prefrontal cortex precede changes in behaviour

Abstract

The ability to change behavioural strategies in the face of a changing world has been linked to the integrity of medial prefrontal cortex (mPFC) function in several species. While recording studies have found that mPFC representations reflect the strategy being used, lesion studies suggest that mPFC is necessary for changing strategy. Here we examine the relationship between representational changes in mPFC and behavioural strategy changes in the rat. We found that on tasks with a forced change in reward criterion, strategy-related representational transitions in mPFC occurred after animals learned that the reward contingency had changed, but before their behaviour changed. On tasks in which animals made their own strategic decisions, representational transitions in mPFC preceded changes in behaviour. These results suggest that mPFC does not merely reflect the action-selection policy of the animal, but rather that mPFC processes information related to a need for a change in strategy.

Figure 1. Two tasks with strategy changes.

(a) The MT-LRA task consists of a central track containing three low-cost T-choices, a high-cost T-choice at the top of the central track and two return-rails on which reward was delivered under either leftward, rightward or alternating contingencies. (b) On switch-days, the task contingency changed ∼20 min into the session. On this session, right laps were rewarded in the first part and left laps in the second. Rats were not removed from the track when the reward contingencies switched. See ref. . (c) The DD task consists of a repeated choice between larger-later and smaller-sooner options. The delay to the larger-later side was adjusted based on the rat's behaviour. (d) Rats typically showed three phases on the DD task, alternating to determine which side was delayed and what the starting delay was, titrating the delay by preferentially selecting one side or the other and then exploiting that delay by preferentially alternating between the two options, holding the delay constant. For analysis, neurophysiological relationships to behaviour was measured against six spatial sections on each task (Supplementary Fig. 1).

Figure 2. Behavioural change points.

(a) An example switch session, showing the time at which the reward contingency changed (solid line), and the time the animal's behaviour changed (dashed line). Lap meaning as in Fig. 1b. This session was a left to alternation day. (b) Histogram of time between the information switch and the animal's behaviour change. In all but one session, the primary behaviour change occurred after the information switch. (c) An example DD session, showing the time of detected behavioural change between the titration and exploitation phases. (d) Histogram of lap on which the behavioural change between the titration and exploitation phases occurred.

Figure 3. mPFC population firing changes follow task criterion changes but precede behavioural changes on the MT-LRA task.

(a) Lap–lap correlation of the population firing vectors for one session, with the switch lap marked (black line). (b) Average lap–lap correlation of the population firing vectors over all sessions aligned to the switch lap for each session. (c) Averaged Z-scored transition scores between matched-choice laps over all MT sessions aligned to the switch lap for each session (error bars show s.e.m., n=12). Normalization is bootstrap against interspike interval shuffles. (d) Averaged Z-scored transition scores aligned to the detected behavioural change (error bars show s.e.m., n=12). Asterisks indicate values significantly larger than zero (α=0.05) as indicated by a Z-test. Alternate normalizations are shown in Supplementary Fig. 4.

Figure 4. mPFC population firing on the DD task reveals a latent strategy transition.

(a) Lap–lap correlation of the population firing vectors for one session, with the detected behavioural change lap marked (dotted line). (b) The animal's behaviour on that session plotted as delay value by lap, with the behavioural change marked with a red line. (c) Lap–lap correlation of the mPFC population firing vectors aligned to the lap of detected maximal behavioural change. Although no clear transition can be seen, our transition-detection algorithm was able to find a clear transition ∼5–8 laps before the detected behavioural change (d). Asterisks indicate values significantly larger than zero (α=0.05) as indicated by a Z-test. Error bars show s.e.m., n=45. Alternate normalizations are shown in Supplementary Fig. 5.

Figure 5. VTE behaviours.

When rats come to difficult choices, they sometimes pause and re-orient back and forth before making their decision. This behaviour, termed VTE, is indicative of indecision and the flexible decision-processes necessary to accommodate it. VTE can be quantitatively measured with the Z-scored integrated absolute angular velocity through the choice point (zIdPhi, see Methods). (a) VTE is significantly increased after the forced strategy-switch on the MT-LRA task. (b) VTE is significantly increased around the moment of behavioural change on the MT-LRA task. (c) VTE is significantly increased before the moment of behavioural change on the DD task.

Figure 6. A transition tended to occur after the first few laps on MT-LRA sessions.

(a) Transition scores of each lap for each session aligned to the start of session, and ordered by the session with the earliest switch lap. We cannot assess transition scores in the first five laps of a sessions, so these first laps are represented as zero. Red line indicates the switch Lap on each session. (b) Average transition scores of all sessions shown in a. (c) Lap-by-lap correlation matrix of population firing averaged over all sessions, aligned to start of session. (d) Lap-by-lap correlation matrix of population firing, averaged over all sessions, aligned to lap of highest transition score in the first 15 laps of each session.

Figure 7. Correlation plots for the spatial population firing vector patterns for cells recorded across several days for two rats on the MT-LRA task.

Four cells were recorded across 6 days for rat 2 and five cells were recorded across 5 days for rat 3. White lines separate the individual sessions, and black lines indicate the switch in each session. Note the areas of heightened self-correlation at the beginning of each session that are correlated across sessions.