Neural correlates of two different types of extinction learning in the amygdala central nucleus

Extinction is a fundamental form of memory updating in which one learns to stop expecting an event that no longer occurs. This learning ensues when one experiences a change in environmental contingencies, that is, when an expected outcome fails to occur (simple extinction), or when a novel inflated expectation of a double outcome (overexpectation) is in conflict with the real outcome, and is a process that has been linked to amygdala function. Here, we show that in rats, the same neuronal population in the amygdala central nucleus updates reward expectancies and behaviour in both types of extinction, and neural changes in one paradigm are reflected in the other. This work may have implications for the management of addiction and anxiety disorders that require treatments based on the outcome omission, and disorders such as obesity that could use overexpectation, but not omission strategies.

. Neural firing at time of reward for the common population. a. Normalized neural firing of the common population during the post-reward period. Firing was greater when reward was delivered (following Overexpectation, blue, and Control, black) compared to when no reward was delivered (Omission, red). Firing following the Overexpectation compound is greater compared to following the Control compound during the early trials. b. Correlation of neural summation (Overexpectation -Control during the early trials) between the compounds and post-reward periods in the common population.   Distributions are shifted to the right for the Overexpectation unique population for the early but not late trials. Distributions are shifted to the left for the Omission unique population for the late but not early trials. Sessions categorized as exhibiting poor performance were based on behavioral data obtained from the Test, which showed no Overexpectation or Omission effects (see below). The behavioral and corresponding neural data from these sessions were analyzed separately from the data in the main text.
Analyses of the neural data from these failed sessions allow for further determining whether the changes in neural firing reported in the main text are only present in sessions that show evidence of Overexpectation Individual neuron analyses also confirmed no change across trials for any of the three compounds. Change indices representing firing during the early trials minus firing during the late trials were not different from zero for any of the compounds (Supplementary Figure 1c, max t(17)=-1.5, p>0.05). Finally, as neural firing did not reflect any change in associative relationships across the Compound Probe, no corresponding differences between the cues were seen on Test (see Supplementary Figure 1d). An ANOVA revealed an effect of cue F(2,38)=8.8, p<0.05), no effect of trials F(3,57)=1.9, p>0.05) and no interaction F(6,144)=2.0, p>0.05). To further examine the effect of cue, three paired-samples t-tests using Bonferonni correction revealed no difference between the Overexpectation and Control cues (t(38)<1, p>0.05) nor between the Omission and Control cues (t(38)=-1.7, p>0.05), but a difference between the Overexpectation and Omission cues (t(38)=2.9, p<0.05).

SUPPLEMENTARY NOTE 3. Neural Firing to reward: Common population
Neural responses to reward during the Compound Probe in the common population exhibited a mixed profile that tracked reward delivery as well as aspects of reward expectation. Neural firing (averaged across all 16 trials) during the reward period was greater compared to the compound period (t(25)=8.3, p<0.05), and firing was greater to reward compared to baseline for the two conditions that received pellets i.e.
Control (t(25)=7.6, p<0.05; Supplementary Figure 3a The above changes in neural firing following the presentation of the Overexpectation and Omission compounds are suggestive of the presence of an attentional or perhaps even an error-correcting signal.
Such an interpretation would require similar neural firing during the reward period between Overexpectation and Omission, with each differing in the same direction from Control. This was not the case. Further analyses revealed that firing to reward at the start of the phase differed between the conditions (F(2,50)=58.61, p<0.05, Supplementary Figure 3a) such that firing was greater at time of reward following the Overexpectation compound (vs. Control, q=3.53, p<0.05) and lower following the Omission compound (vs. Control,q=9.04,p<0.05). Interestingly, while the latter result is consistent with the difference in reward delivery between Omission (zero pellets delivered) and Control (two pellets delivered), the former difference is striking given that the reward delivered following the presentation of the Overexpectation and Control compound was identical (two pellets in both conditions). The difference between Overexpectation and Control suggests that firing during reward delivery may reflect not only presence of reward but also the expectation of reward for two reasons. Firstly, delivery of reward occurs during the last second of compound presentation. Secondly, the delivery of the first pellet represents the best temporal predictor of the arrival of a second pellet. Critically, the delivery of the second pellet would serve to predict the arrival of the additional overexpected pellets in the Overexpectation but not in the Control condition. This is supported by the finding that the neural summation seen during the compound predicts the additional expectation of two pellets at time of reward: The difference in firing between the Overexpectation and Control compounds during the early trials is correlated with the same difference during the reward period (r 2 =0.46, p<0.05, Supplementary Figure 3b).
Finally, the difference in firing at time of the reward period between Overexpectation and Control was adjusted: The difference in reward firing following these two compounds disappeared on subsequent days of Compound conditioning (each q<1).

SUPPLEMENTARY NOTE 4. Neural firing: Unique populations during Compound Probe.
Two  Figure 4). The distribution of these indices was positively shifted during the early trials (µ=0.53, t(10)=3.8, p<0.05), but centered on zero during the late trials (µ=-0.02, t(10)=-0.2 p>0.05), with a significant difference between them (t(10)=3.9, p<0.05). This shows that firing to the Overexpectation compound was greater compared to the Control compound at the start but equivalent by the end of the Compound Probe phase.
Similarly, 16 cells in the Omission-unique population also tracked changes in reward expectancy.
An index was computed representing the difference in firing between Omission and Control (Supplementary Figure 4). The distribution of the indices was centered on zero during the early trials (µ=-0.18, t(15)=0.9, p>0.05) but negatively shifted during the late trials (µ=-0.37, t(15)=-2.7, p<0.05), with a significant difference between them (t(15)=3.3, p<0.05). Thus, there were no differences in neural firing between the Omission and Control compounds during the early trials, and a clear difference emerged such that neural firing was lower to the Omission compared to the Control compound by the late trials. These analyses show that the population of cells that show a decline in firing uniquely to Overexpectation or uniquely to Omission tracked the change in reward expectancy in the Overexpectation and Omission conditions, respectively.

SUPPLEMENTARY NOTE 5. CN neural firing does not signal changes in attention.
Poor but not good predictors of outcomes command attention 22 . In the present task, both the Overexpectation and Omission compounds were poor predictors of their corresponding outcomes during the Compound Probe phase, whereas the Control compound consistently signaled its predicted outcome.
Therefore, it was expected that the Overexpectation and Omission compounds (but not the Control) would each command attention. In order to determine whether CN neurons signaled attention, we screened all 87 cells (irrespective of reward responses) for increase or decreases in neural firing during presentation of the Overexpectation and Omission compounds compared to the Control compound. Greater firing to the Overexpectation compared to the Control compound was obtained in seven neurons (that were not part of the reward-responsive population described in the main text and shown not signal attention) whereas three neurons were shown to exhibit the opposite pattern of results (i.e. lower firing rate, t-test, p<0.05). In the case of Omission, seven and one neurons were shown to exhibit greater and lower firing rate compared to the Control compound, respectively (t-test, p<0.05). All these cells were below chance (t's(999)=163.3, p's<0.05).