Mediodorsal Thalamus Is Critical for Updating during Extradimensional Shifts But Not Reversals in the Attentional Set-Shifting Task

Synthesis

Thank you for your submission, the reviewers are largely in agreement that this is an interesting set of findings that add new knowledge to the literature. They have both included a series of points that require being addressed, as set out below. I would also add that as the study appears to only have used male mice, please could this be stated in the abstract and could there be a sentence added to the discussion stating that the results may not generalise to females.

Reviewer #1

This paper presents an interesting set of findings that MD lesions affect initial sensory discrimination (SD) and ED shifting of attention while sparing reversal learning and ID shifts of attention. These results are potentially important for understanding the role of rodent MD in cognitive flexibility. The experiments appear to be carefully conducted. The authors follow the well-established methods developed by Birrell and Brown (2000) to examine ID/ED attention shifting. MD lesions appear to be made skillfully and lesion extent analyzed quantitatively. C-Fos methods were used to assess neuronal activity in orbital and medial PFC 90 minutes following a second exposure to SD and ED training. These analyses lack necessary controls to determine whether MD lesions specifically affect the activity of these cortical areas during the behavioral tasks.

There are several issues that should be addressed in a revision of this paper.

As the authors note, Birrell and Brown found no effect of medial prefrontal lesions on initial SD. It is unclear why they predict that MD lesions should affect both SD and ED in the introduction (lines 73 and 74) while acknowledging that medial prefrontal lesions do not. This seems to be an important predicted distinction between MD and PFC function that should be addressed more directly.

The authors ascribe the decrease in trials to criterion for REV3 for the MD lesion group to effects of overtraining. This begs the question of why a similar effect not observed for REV 1 following SD overtraining here or for REV3 following ED overtraining in Birrell and Brown? Are there other potential explanations for this result?

The authors argue (paragraph beginning on line 529) that reduced correct 2nd choice responses represent a failure to adapt initial choice responses. Couldn’t this simply represent a failure to recognize a correct first choice (i.e failure to acquire the discrimination rule) leading to an incorrect response to the alternative pot?

The authors argue (paragraph beginning on line 558) that reduced c-Fos counts provide evidence that “MD-PFC interactions contribute to learning and cognitive flexibility”. It is not clear what to make of the c-Fos results. Comparable decreases were observed after a 90 m period that followed second exposure to SD and ED. However, there are no controls for nonspecific effects of MD lesions. Would comparable decreases have been observed for a group of rats given MD lesions with no behavioral training? Are decreases observed in OFC and PrL greater than in other areas of cerebral cortex? Given the lack of such basic controls, what do the c-Fos results add to this paper?

Reviewer #2

In this study the authors used excitatory lesions in the rat to study the impact of mediodorsal thalamus (MD) lesions on flexible behavior. Flexible behavior was tested in the Birrell & Brown odor-based rule shifting task that allows for assessing initial association, within perceptual dimension rule shifts (ID or reversal learning) and extradimensional rule shifts. Although the question addressed is not novel the data are important as they contradict previous findings and the discussion about these contradicting data will enhance our understanding of the role of the MD in flexible behavior. The experiments are well done. I have mostly comments about the selection and the interpretation of the data.

Figure 2: All animals completed the ID/ED task pre- and postoperatively. The authors should include both sets of data in Figure 2 so it can be seen how the animals performed pre-operatively. Also, I could not find in the method section that rats were tested pre- and postoperatively.

The authors interpret the deficit in the initial SD acquisition as that the rats have to learn a stable response strategy, which is impaired in MD lesioned rats. Once the rats have learned the strategy MD lesions have no effect on the learning of switched or reversed associations (as long as they are in the same perceptive domain). I have a problem with this interpretation because the rats have already acquired the response strategy pre-operatively. Thus, performing SD acquisition is not new for the rats. Is it possible that the lesioned rats forgot about or are not able to implement the previously learned SD-response strategy?

I don’t understand the argument of overtraining for the improved performance of MD lesioned mice in REV3. First, it takes MD lesioned mice longer to reach criterion in the ED shift condition thus doing more trials is not necessarily an overtraining. Second, if the mice were indeed overtrained, I would have expected a deficit and not an improvement in performance.

It is possible that the MD lesion animals never properly learned the rule. They may have been able to reach the criterion with a less firm understanding of the rule than the MD sham animals. In that case, they would also be quicker to learn the reversal as the rule that they would have to un-learn would not have as strong an association.

Figure 3: The rational and interpretation for the analysis of the 2nd correct choice is not clear to me. As per provided definition, the rat does a correct 2nd choice when the first bowl it visited was an incorrect bowl and the rat didn’t dig into the incorrect bowl. In this scenario if the rat doesn’t dig in the first bowl, this is for me an indication that the rat already understands the new rule based on previous trial outcomes. It is not clear to me what the authors mean by learning on the fly and why the deficit cannot be explained by an enhancement in perseverative responding or an inability to respond to negative feedback as explained in line 350. To address perseverative responding the authors should analyze the percentage of perseverative vs non-perseverative errors (e.g., as in Ferguson et al. 2018 Biol Psych).

There are alternative explanations for less correct 2nd choices. First, it possible that lesioned animals do less correct 2nd choices because they less often approached the incorrect bowl as the first bowl. Second, lesioned animals do less correct 2nd choices because they do more mistakes. By digging more often into the 1st bowl, when it is incorrect, they have less possibilities to do a 2nd correct choice.

Would it be easier for the animal to make the assessment of what is in each pot for trials where odor is the rule as opposed to texture? If odor is the dimension that is relevant, the animal might be able to smell the odor from a distance without approaching the pot and make a “2nd choice” without having to go visit both pots.

Figure 4: Reaction time seems not the appropriate term here as reaction times are usually in the range of milli-seconds. I would rather suggest to use the term “latency to dig”. Why is this measure only shown for error trials? It should be shown for both. The longer latencies to dig in the lesioned animals could be an indication of a change in motivation. This should be discussed.

Line 429: “The analysis of behavioral responses........is critical for monitoring and updating of choice response rules”. Again, I don’t see how the 2nd choice responses support this interpretation.

Figure 5-7: Even if the authors did not used unbiased stereological methods the authors still have to ensure that the sample areas are consistent between animals. The authors should indicate the anterior-posterior levels at which the three sections were taken and make sure that they were at the same level. They also should calculate the area that was sampled for each animal to make sure that there is no bias. How did the animals perform when repeating the two task conditions for the c-fos staining? Does repeating the task for a third time improve the behavior?

Ferguson et al. 2017 found that inhibiting the MD with the DREADD system in mice impaired initial acquisition and reversal learning but not ED set shifting using a different version of the Birrell & Brown odor based digging task. Parnaudeau et al. 2013 Neuron and 2015 Biological Psychiatry found a deficit in reversal learning but not in the initial acquisition of an operant based reversal learning task during DREADD-mediated MD inhibition. What explains the different outcomes between these two studies and the present study? Could it be related to inhibition restricted to the time window when the behavior is performed versus the chronic MD lesion? Alternatively, since animals had performed a reversal in the preoperative training, they are not naïve to it after the lesion, which may explain the absence of an effect of the lesion.

Author Response

ID/ED eNeuro - reviewer’s comments

Dear eNeuro Editor,

We are pleased to resubmit for publication in eNeuro a revised version of our manuscript REF eN-NWR-0162-21, titled “Mediodorsal thalamus is critical for updating during extra-dimensional shifts but not reversals in the attentional set-shifting task”.

We appreciate the thoughtful comments received from the reviewers, and we are grateful for the opportunity to address their concerns through this revision. Reviewers expressed very helpful comments and indicated a number of opportunities for improving the manuscript - opportunities that we took very seriously and have incorporated into our manuscript.

We have implemented the changes needed to address all the comments we received. The enclosed memo describes, briefly, how we have modified the original draft to address the reviewers’ concerns. To facilitate the reviewing process, and to make future communication more efficient, each individual comment was identified and our responses are indicated in red. Our implementation of the changes that reviewers suggested changed some of the results that we reported in the original draft making some effects stronger.

We are convinced that our attempt to address the reviewers’ concerns resulted in a stronger and more compelling manuscript, and we are grateful for the opportunity to contribute to eNeuro. We are looking forward to hearing from you.

Sincerely,

The Authors

Synthesis Statement for Author (Required):

Thank you for your submission, the reviewers are largely in agreement that this is an interesting set of findings that add new knowledge to the literature. They have both included a series of points that each require being addressed, as set out below. I would also add that as the study appears to only have used male mice, please could this be stated in the abstract and could there be a sentence added to the discussion stating that the results may not generalise to females.

This has been done, and to clarify the animals are male rats.

Reviewer #1

This paper presents an interesting set of findings that MD lesions affect initial sensory discrimination (SD) and ED shifting of attention while sparing reversal learning and ID shifts of attention. These results are potentially important for understanding the role of rodent MD in cognitive flexibility. The experiments appear to be carefully conducted. The authors follow the well-established methods developed by Birrell and Brown (2000) to examine ID/ED attention shifting. MD lesions appear to be made skillfully and lesion extent analyzed quantitatively. C-Fos methods were used to assess neuronal activity in orbital and medial PFC 90 minutes following a second exposure to SD and ED training. These analyses lack necessary controls to determine whether MD lesions specifically affect the activity of these cortical areas during the behavioral tasks.

There are several issues that should be addressed in a revision of this paper.

1) As the authors note, Birrell and Brown found no effect of medial prefrontal lesions on initial SD. It is unclear why they predict that MD lesions should affect both SD and ED in the introduction (lines 73 and 74) while acknowledging that medial prefrontal lesions do not. This seems to be an important predicted distinction between MD and PFC function that should be addressed more directly.

Thank you for this comment. To clarify, our position is that the MD and mPFC are working together in partnership, contributing different but complementary roles to goal-directed behaviours. Therefore, we are not wanting to suggest that the MD mimics PFC behaviours. Rather, the MD has its own unique pattern of interconnected inputs from subcortical and cortical structures that identify it as an important nodal structure within a wider neural network that includes frontal cortex. Similarly, the mPFC also has its own set of inputs so together the unique and overlapping inputs associated with the two structures complement each other. Dr Mitchell’s work in non-human primates has helped to clarify this position as research from her lab has shown that the MD is not simply mimicking PFC damage. For example, the MD contributes to new learning but is not involved in retention of preoperatively learnt strategies or visual discriminations. In contrast, the PFC does contribute to learning and retention (e.g., Mitchell et al 2007a, b; Mitchell and Gaffan, 2008).

Thus, we predicted MD contributes to learning because the Mitchell lab has identified that the primate MDmc (MDm in rodents that was lesioned in the current set of rats) contributes to learning complex visual discriminations associated with reward (Mitchell and Gaffan, 2008; Browning et al 2015; Chakraborty et al., 2016; Chakraborty et al., 2019). In addition, MD lesion rats are impaired when learning discriminations associated with reward (Mitchell and Dalrymple-Alford, 2005; Hunt and Aggleton, 1998; Courtiol and Wilson, 2019; with Mitchell and Chakraborty, 2013 providing an overview table of the effects of MD lesions in animals). Ferguson and colleagues (2017) also showed perturbations to the MD impair learning discriminations. Further, while we have shown that the animals with MD damage are impaired in learning, this damage does not render the animals completely amnesic. Instead, what we and others have observed is that when given further trials, the animals are typically able to learn the discriminations, and this was certainly true for the current set of rats.

We have adjusted the text in the introduction in several places to explain our reasons for running this experiment and our hypothesis further. The text is coloured in red in the manuscript.

2) The authors ascribe the decrease in trials to criterion for REV3 for the MD lesion group to effects of overtraining. This begs the question of why a similar effect not observed for REV 1 following SD overtraining here or for REV3 following ED overtraining in Birrell and Brown? Are there other potential explanations for this result?

Thank you for highlighting that this explanation is not yet fully clear. In subsequent work, the Brown lab has identified that rats who are over-trained take less trials to complete reversals. (Dhawan SS, Tait DS, Brown VJ. More rapid reversal learning following overtraining in the rat is evidence that behavioural and cognitive flexibility are dissociable. Behav Brain Res. 2019 May 2; 363: 45-52. doi: 10.1016/j.bbr.2019.01.055.). This behavioral performance has been identified and studied for many decades in psychology (e.g. Reid, 1953; Lovejoy, 1966). Lovejoy, E. (1966). Analysis of the overlearning reversal effect. Psychological Review, 73(1), 87-103, doi: 10.1037/h0022687

We believe that the additional trials required by the MD lesion rats, relative to the MD Sham controls, to learn the new discrimination during the ED subtask caused an overexposure to all of the sensory dimensions. Therefore, the MD lesion group may have developed, through overtraining, a greater understanding of the rewarded and unrewarded stimulus dimensions. This may then have increased the salience of the predictive cues and reduced/eliminated the number of factors that can led to an error. Others have also shown similar effects in rats during studies to assess model based and model free learning (e.g. Nilsson et al., 2015, doi: 10.1016/j.neubiorev.2015.06.015 provides a good review).

Given our explanation of overtraining, we still would not predict to see an enhanced effect on REV1 as the SD (simple discrimination) subtask does not use both sensory dimensions. The two types of sensory discriminations (odour and digging medium) are only introduced for the CD (compound discrimination) subtask, which is run immediately after the SD subtask. REV1 follows the CD subtask. The rats did not experience overtraining on the CD task as they completed it in a similar amount of trials as the MD Sham controls. Throughout these two further subtasks (CD and REV1), the rats had to apply the strategy, (i.e. respond to the rewarded sensory dimension (either odour or digging media) that was learnt during the SD).

Finally, as indicated, we had expected the MD lesion rats to be impaired in reversal learning but this wasn’t the case. However, all the animals had been trained to perform the task preoperatively prior to their lesion (as also identified by Reviewer 2), so all the rats were familiar with the concept of reversal learning and knew that it could occur within the task structure (Costa et al, 2015: doi: 10.1523/JNEUROSCI.1989-14.2015 ; Jang et al 2015: doi: 10.1523/JNEUROSCI.1594-15.2015).

We attribute the differences between our results and the study of Birrell and Brown (2000) to the differences of the investigated areas. Birrell and Brown investigated the role of the medial prefrontal cortex in ID/ED. The lesions were centered on prelimbic cortex and included damage to infralimbic cortex. However, in the rat, the MD projects to medial prelimbic cortex, orbitofrontal cortex and cingulate (Cg1) cortex. A large body of research indicates that separate rat prefrontal cortex subregions differentially contribute to learning, memory, and other cognitive functions (for a review, see Dalley, Cardinal, & Robbins, 2004; Uylings, Groenewegen, & Kolb, 2003). Differential involvement of prefrontal cortex areas has been observed in tests that examine sustained attention, working memory, memory consolidation, and task switching (Birrell & Brown, 2000; Chudasama et al., 2003; Delatour & Gisquet-Verrier, 2000; Dias & Aggleton, 2000; Ragozzino et al., 2003; Rudebeck et al., 2006). For example, several previous studies showed that anterior cingulate lesions did not impair reversal learning in two-choice discrimination tests (Becker et al., 1981; Bussey et al., 1997; Chaillan et al., 1997; Joel et al., 1997; Meunier et al., 1991; Warburton et al., 1998). Gemma and colleagues (2006) showed a deficit in incorporating information about previous action outcomes to guide subsequent behavior. Lateral OFC inactivation preferentially impaired performance during reversal phases. In contrast, prelimbic inactivation caused an apparent improvement in performance by increasing the number of reversals completed. This was associated with enhanced sensitivity to recently rewarded actions and reduced sensitivity to negative feedback. Infralimbic inactivation had no effect, whereas the anterior cingulate appeared to play a permissive role in this form of reversal learning. Hence, we argue that the impairments and changes we report in the current study are the combination of the effects of the dysfunction of MD and its cortical targets.

We have updated our discussion to capture these main points.

3) The authors argue (paragraph beginning on line 529) that reduced correct 2nd choice responses represent a failure to adapt initial choice responses. Couldn’t this simply represent a failure to recognize a correct first choice (i.e. failure to acquire the discrimination rule) leading to an incorrect response to the alternative pot?

Thank you for this comment. However, we wish to highlight that during the ID/ED task, a correct choice is when the rat digs the bowl containing the right dimension. During the task we have noticed that the rats exhibited one of two behaviours before making a choice, they either visit the compartment containing the correct/rewarded dimension and dug to get the reward (correct choice), or visited the opposite compartment, explored the bowl without digging, then switched to the compartment containing the correct dimension and then dug. The latter is considered as a correct 2nd choice response. A failure to acquire the discrimination rule resulted in exploring and digging the incorrect bowl.

For this analysis, we are trying to capture what aspects of learning are impaired after permanent MD lesion. We agree with the proposition that the rats may have simply not correctly identified a correct first choice. However, we disagree with this as a complete explanation because if so, this would suggest the rats are perhaps slower to learn due to a sensory deprivation or are only perseverative with their responding. During learning, there are many strategies that could be used to acquire the new rule for optimal responding. We think that the MD, via thalamocortical interactions, is helping the cortex to compute the optimal signals as a consequence of the here and now within a given trial. With the MD gone, the PFC is still receiving multiple signals but the relevant ones are no longer identifiable. This may be the result of changes to coherence or in density of firing or via disruptions linked to excitation and inhibition in the PFC. So, while the rat is still able to learn and make decisions, it is not able to do so quickly and optimally in the here and now of a single trial. Thus, with the MD gone and this additional route of signal transfer no longer being conducted to the cortex, we think that the rats are not able to quickly learn to adapt after the task has changed (ED subtask) and instead it takes them longer to determine which stimulus dimension is now the correct one to guide optimal behavior. What we observe in their behavior is that the MD lesion rats return to sampling the unrewarded stimulus or previous location of the bowl because other brain regions had already directed the motor responses to go with this option at the start of the trial. We consider that the reduced correct 2nd choice responses represent a failure to adapt the initial choice response during a particular trial and thus a failure of within trial behavioural flexibility.

4) The authors argue (paragraph beginning on line 558) that reduced c-Fos counts provide evidence that “MD-PFC interactions contribute to learning and cognitive flexibility”. It is not clear what to make of the c-Fos results. Comparable decreases were observed after a 90 m period that followed second exposure to SD and ED. However, there are no controls for nonspecific effects of MD lesions. Would comparable decreases have been observed for a group of rats given MD lesions with no behavioral training? Are decreases observed in OFC and PrL greater than in other areas of cerebral cortex? Given the lack of such basic controls, what do the c-Fos results add to this paper?

We agree that sufficient controls have not be utilized for this part of the study so we have decided to remove the c-Fos imaging from the paper. We needed to have included a lesion and MD Sham group that sat in the cage but it was not possible for this study.

Reviewer #2

In this study the authors used excitatory lesions in the rat to study the impact of mediodorsal thalamus (MD) lesions on flexible behavior. Flexible behavior was tested in the Birrell & Brown odor-based rule shifting task that allows for assessing initial association, within perceptual dimension rule shifts (ID or reversal learning) and extradimensional rule shifts. Although the question addressed is not novel the data are important as they contradict previous findings and the discussion about these contradicting data will enhance our understanding of the role of the MD in flexible behavior. The experiments are well done. I have mostly comments about the selection and the interpretation of the data.

1) Figure 2: All animals completed the ID/ED task pre- and postoperatively. The authors should include both sets of data in Figure 2 so it can be seen how the animals performed pre-operatively. Also, I could not find in the method section that rats were tested pre- and postoperatively.

Thank you very much for pointing this out. We have now clarified in the methods that preoperative training occurred in the task. In the original draft, we had only mentioned preoperative training in the results section. For Figure 2, we have now included both preoperative and post-operative data.

2) The authors interpret the deficit in the initial SD acquisition as that the rats have to learn a stable response strategy, which is impaired in MD lesioned rats. Once the rats have learned the strategy MD lesions have no effect on the learning of switched or reversed associations (as long as they are in the same perceptive domain). I have a problem with this interpretation because the rats have already acquired the response strategy pre-operatively. Thus, performing SD acquisition is not new for the rats. Is it possible that the lesioned rats forgot about or are not able to implement the previously learned SD-response strategy?

It is true that the SD (simple discrimination) subtask is not new for the rats because they were run preoperatively on all subtasks of the ID/ED task in the same order as the postoperative test. However, they had not acquired the response strategy preoperatively to complete the SD postoperatively as the final subtask ran preoperatively was REV3, which has two sensory dimensions (odour and digging media, i.e. four stimuli). In contrast, the SD subtask uses only one sensory dimension (odour or digging media with two stimuli to discriminate), so the SD subtask is different for the animals to encounter after their surgery.

In the preoperative session, the SD is the first subtask the rats encounter. However, in the postoperative session while it is still the first subtask they encounter, it is following on from REV3 of the preoperative session and surgery (MD Sham or Lesion) and at least two weeks to recover and return to 85% of baseline weight. Given all of this, it is possible that the rats had forgotten about the strategy to conduct a simple discrimination but we disagree that this is the case as monkeys with MD lesions are still able to implement a previously learnt response strategy (Mitchell et al, 2007 J Neurosci). Instead, what we think is that the rats with MD lesions are unable to quickly update/change to a different response strategy (i.e. revert back to dealing with only one sensory dimension with two stimuli) in addition to rapidly learning this new sensory dimension discrimination. However, as indicated in response comment 1) to Reviewer 1, while the MD lesion rats were impaired in learning they were not densely amnesic and instead just needed more trials to acquire the correct response strategy.

3) I don’t understand the argument of overtraining for the improved performance of MD lesioned mice in REV3. First, it takes MD lesioned mice longer to reach criterion in the ED shift condition thus doing more trials is not necessarily an overtraining. Second, if the mice were indeed overtrained, I would have expected a deficit and not an improvement in performance.

Please see our above response [comments 2)] in reply to Reviewer 1, as we cover the explanation of overtraining there. In addition, to clarify, we were using rats in this research, not mice.

4) It is possible that the MD lesion animals never properly learned the rule. They may have been able to reach the criterion with a less firm understanding of the rule than the MD sham animals. In that case, they would also be quicker to learn the reversal as the rule that they would have to un-learn would not have as strong an association.

This is also a possible explanation, and one we cover in the manuscript (line 444 onwards). The standard version of the Birrell and Brown (2000) task that we used has been adapted, if rats are not able to learn the rule. This deficit in rule learning would be displayed with impaired performance during the CD, REV1, and ID shifts. In the adapted version of the task, additional ID shifts have been introduced so that the rats with profound learning deficits may be able to learn the rule prior to the ED subtask being introduced. This has been the case for rats with other thalamic lesions, namely the anterior thalamus (ATN: Wright et al 2015) and the nucleus reuinens (NRe: Linley et al 2016). However, rats with MD lesions can and did learn the rule. As indicated by their performance in the task, they were not impaired in the CD, REV1 or ID shifts. Thus, we propose that the learning deficit is one that is related to not being able to rapidly adapt and update behavioral responses (choices) during the trial as a consequence of the rule change rather than an overall deficit in learning per se.

5) Figure 3: The rational and interpretation for the analysis of the 2nd correct choice is not clear to me. As per provided definition, the rat does a correct 2nd choice when the first bowl it visited was an incorrect bowl and the rat didn’t dig into the incorrect bowl. In this scenario if the rat doesn’t dig in the first bowl, this is for me an indication that the rat already understands the new rule based on previous trial outcomes. It is not clear to me what the authors mean by learning on the fly and why the deficit cannot be explained by an enhancement in perseverative responding or an inability to respond to negative feedback as explained in line 350. To address perseverative responding the authors should analyze the percentage of perseverative vs non-perseverative errors (e.g., as in Ferguson et al. 2018 Biol Psych).

Any tendency to perseverate in responding to a previously correct stimulus would impair reversal learning. However, our MD rats do not show impairments during the reversals, i.e. they are not perseverating (like OFC lesion rats). Interestingly, OFC lesion rats (McAlonan and Brown, 2003) also had a preoperative test session that exposed them to the reversal but this didn’t mitigate the postoperative test impairment. The OFC lesion rats, in contrast, were not impaired in the ED shift. Thus, this ID/ED task, allows for different types of behavioral flexibility and cognitive control to be assessed within the same session.

Reversals in this task are a special case of an intra-dimensional (ID) shift. Like the ID shift, reversal learning does not require attention to be reoriented, because the attended dimension remains the same as the previous subtasks (SD, CD). However, unlike the ID shift in which all the exemplars are novel, in the reversal, the exemplars remain the same as the just completed subtask (CD for REV1 and ID for REV2): the previously correct exemplar is now incorrect, and the rat must respond to the previously incorrect exemplar.

Therefore, given that the rats are not showing perseverative responding during reversal learning, it suggests that the ED shift deficit for our MD lesion rats is not caused by a problem with perseverative responding either (although rats may make perseverative responses) because the rats are as able as controls to abandon a response to a previously reinforced stimulus. Rather, the rats were unable to learn quickly, within the trial, if they had chosen the incorrect bowl to dig in and instead, continue to dig in the incorrect bowl (i.e. they made less 2nd choice correct responses than the MD Sham controls).

It was not possible to divide performance into epochs corresponding to a period of perseverative responding and a period of learning (Ferguson et al. 2018 Biol Psych; Hunt and Aggleton, 1998), because learning was rapid in the control group, meaning that there were not enough trials. In order to reach criterion, the rats must complete 6 consecutive correct trials. In the MD lesion group, during the ED subtasks, we observed that the rats would sometimes pick two or three correct bowls in a row and then switch to the other (incorrect) bowl for a couple of trials and then back and forth. In monkeys with MD lesions, we have also observed that they don’t perseverate in their responding (Mitchell et al., 2007; Chakraborty et al. 2016). Instead, MD lesion monkeys adopt a strategy that causes an increase in switching their choice responses between all the available options (Chakraborty et al. 2016). We were able to conduct this additional analysis with the monkeys as they were completing 300 trials per day. This type of behavioral responding suggests that the monkeys and rats are aware of the task structure but they are requiring many more trials to be able to develop a hypothesis about the optimal stimuli that will provide the reward most often.

6) There are alternative explanations for less correct 2nd choices. First, it possible that lesioned animals do less correct 2nd choices because they less often approached the incorrect bowl as the first bowl. Second, lesioned animals do less correct 2nd choices because they do more mistakes. By digging more often into the 1st bowl, when it is incorrect, they have less possibilities to do a 2nd correct choice.

The left-right position of presentation in the cage was a pseudorandom series determined in advance. The trials are forced choice so they have to respond to one of the bowls. They are motivated to complete the task and receive their daily food after testing. Their weight is reduced to 85% of free-feeding weight and all the rats readily ate the Honey Nut Cheerios cereal. The criterion to complete each subtask is 6 consecutive correct choices. All animals had the same opportunity to approach the correct and the incorrect bowls. When the rats made a mistake by digging in the incorrect bowl, this was scored as error not a correct 2nd choice response. Because lesioned animals make more errors, we evaluated their performance as percentage of correct 2nd choice responses out of the total number of correct choices (please see Fig.3).

7) Would it be easier for the animal to make the assessment of what is in each pot for trials where odor is the rule as opposed to texture? If odor is the dimension that is relevant, the animal might be able to smell the odor from a distance without approaching the pot and make a “2nd choice” without having to go visit both pots.

This is a good question to explore. We’ve conducted additional analyses to investigate this. We have included these details in the results section from line 361 onwards.

We analyzed the rats’ errors made reaching criterion during the ED subtask, as a consequence of whether they experienced a sensory dimension shift from odor to digging medium or digging medium to odor, which indicated that the increase in errors occurs irrespective of the direction of the attentional shift. The univariate ANOVA with Lesion_type and Sensory Shift as the between-subject factors and errors to criterion as the dependent measure revealed a main effect of Lesion_type, F(1, 21) = 106.03, p < 0.001, but no main effect of Sensory Shift, F(1, 21) = 3.84, p= 0.063 and no interaction, F(1, 21) = 0.19, p=0.666.

8) Figure 4: Reaction time seems not the appropriate term here as reaction times are usually in the range of milli-seconds. I would rather suggest to use the term “latency to dig”. Why is this measure only shown for error trials? It should be shown for both. The longer latencies to dig in the lesioned animals could be an indication of a change in motivation. This should be discussed.

Thank you for this suggestion, we have adjusted the figures to show “latency to dig” instead of “reaction time”. We have not included latencies on correct choices as this is a forced choice task with a criterion of 6 consecutive correct responses indicating acquisition in the task. All rats completed the correct trials well. Thus, we have only included latency to dig for the error trials to be able to highlight that all the rats were motivated to perform the task and this wasn’t the reason for their deficit. While the MD lesion rats were slower to respond on error trials during the first three subtasks (SD, CD and REV1), they were not unmotivated to perform the task, rather they were attending, learning the rewarded sensory dimension more slowly, but actively engaged in each of the subtasks and overall testing session.

9) Line 429: “The analysis of behavioral responses........is critical for monitoring and updating of choice response rules”. Again, I don’t see how the 2nd choice responses support this interpretation.

When the rats made more 2nd choice responses that means that they could adapt and change their response within a trial according to the environmental contingencies via monitoring and updating their choice response.

We have adapted this sentence now on line 415 onwards as indicated below.

The analyses of behavioral responses indicated that the MD lesion rats made significantly fewer correct 2nd choice responses within a trial suggesting that information transfer via the MD is critical for the rapid, within trial monitoring and updating of an optimal choice response.

10) Figure 5-7: Even if the authors did not used unbiased stereological methods the authors still have to ensure that the sample areas are consistent between animals. The authors should indicate the anterior-posterior levels at which the three sections were taken and make sure that they were at the same level. They also should calculate the area that was sampled for each animal to make sure that there is no bias. How did the animals perform when repeating the two task conditions for the c-fos staining? Does repeating the task for a third time improve the behavior?

We have removed the c-fos imaging and analyses from the manuscript as there are insufficient control data.

11) Ferguson et al. 2017 found that inhibiting the MD with the DREADD system in mice impaired initial acquisition and reversal learning but not ED set shifting using a different version of the Birrell & Brown odor based digging task. Parnaudeau et al. 2013 Neuron and 2015 Biological Psychiatry found a deficit in reversal learning but not in the initial acquisition of an operant based reversal learning task during DREADD-mediated MD inhibition. What explains the different outcomes between these two studies and the present study? Could it be related to inhibition restricted to the time window when the behavior is performed versus the chronic MD lesion? Alternatively, since animals had performed a reversal in the preoperative training, they are not naïve to it after the lesion, which may explain the absence of an effect of the lesion.

In response to the Parnaudeau et al. 2013 Neuron and 2015 Biological Psychiatry studies, during both the lever press operant box task and the T-maze task, the initial acquisition of the correct discrimination was delayed for the mice with the DREADDs activated (e.g. please see Figure 3B discrimination and Figure 4 from the Neuron paper). The comparison they made in Figure 3B was for the trials conducted on session 7 of learning only. In addition, the volume of DREADDs that were infused into the mice were extensive (0.5 μl per hemisphere - overall total of 1 μl in each mouse). By comparison in our rats’ brains in the current study and previous studies conducted during my PhD (Mitchell and Dalrymple-Alford 2005, 2006), we infused (0.20 μl in the anterior MD and then 0.18 μl in the posterior MD - overall total of 0.38 μl per rat). From my extensive experience of conducting infusions into the dorsal thalamus (ATN, MD) in rodents and monkeys, the use of this larger volume in mice would mean that the DREADDs had almost certainly spread to adjacent dorsal thalamic structures (likely the ANT, intralaminar nuclei (ILn), and potentially downwards to the nucleus reuniens (NRe)). All of these adjacent structures are also interconnected to the medial PFC in rodents. Thus, after i.p. CNO injections all areas (e.g. the MD and ATN, ILn, NRe) that received the DREADDs would be inactive, which would create widespread temporary dysfunction across several neural networks. The larger volume of DREADDs used in these studies was also suggested as a difficulty for interpretation by Wolff and Vann (2019, J Neurosci). Thus unfortunately, from the Parnaudeau and colleagues’ studies, it is difficult to be able to conclude that the performance deficits they observed were only caused by inhibition within the MD. In addition, typically permanent perturbations to the MD do not cause deficits in working memory for rats, monkeys or humans so the Parnaudeau results are not similar to what is observed in other studies either.

For the Ferguson et al (2017) study, the initial deficit in acquisition after MD perturbations is similar to what we observed in our ID/ED task - we have now included this reference in our introduction. However, we suggest the difference from our results for the reversal learning is that the reversal tasks used across ours and their studies were not the same. For example, the Ferguson et al task consisted of an initial association phase where the odour was always rewarded followed with a rule shift phase where the media was always rewarded, then the reversal was performed (the previously unrewarded media is therefore rewarded), which we think made it difficult for the mice to get the associations sorted during the reversals - potentially a deficit in updating reward associations (which rats and monkeys with MD lesions are impaired at doing). Thus, the mice may not have learnt the required strategy to get the task sorted as they were performing concurrent associative learning. In our task, the rats initially learnt which sensory dimension was correct (similarly impaired in acquisition as the Ferguson mice are too) but they went on to learn concurrent discriminations applying a response strategy that involved the same sensory dimension. Thus, the task structure and current strategy is key to performing in the task, as is a core difference between the two types of reversal learning.

We have adapted the discussion to incorporate these points.