Enhanced Aversive Signals During Classical Conditioning in Dopamine Axons in Medial Prefrontal Cortex

Reviewer #1 (Public Review):

Summary:
In this manuscript, Abe and colleagues employ in vivo 2-photon calcium imaging of dopaminergic axons in the mPFC. The study reveals that these axons primarily respond to unconditioned aversive stimuli (US) and enhance their responses to initially-neutral stimuli after classical association learning. The manuscript is well-structured and presents results clearly. The utilization of a refined prism-based imaging technique, though not entirely novel, is well-implemented. The study's significance lies in its contribution to the existing literature by offering single-axon resolution functional insights, supplementing prior bulk measurements of calcium or dopamine release. Given the current focus on neuromodulator neuron heterogeneity, the work aligns well with current research trends and will greatly interest researchers in the field.

However, I would like to highlight that the authors could further enhance their manuscript by addressing study limitations more comprehensively and by providing essential details to ensure the reproducibility of their research. In light of this, I have a number of comments and suggestions that, if incorporated, would significantly contribute to the manuscript's value to the field.

Strengths:
-Descriptive.
-Utilization of a well-optimized prism-based imaging method.
-Provides valuable single-axon resolution functional observations, filling a gap in existing literature.
-Timely contribution to the study of neuromodulator neuron heterogeneity.

Weaknesses:
1. It's important to fully discuss the fact that the measurements were carried out only on superficial layers (30-100um), while major dopamine projections target deep layers of the mPFC as discussed in the cited literature (Vander Weele et al., 2018) and as illustrated in FigS1B,C. This limitation should be explicitly acknowledged and discussed in the manuscript, especially given the potential functional heterogeneity among dopamine neurons in different layers. This potential across-layer heterogeneity could also be the cause of discrepancy among past recording studies with different measurement modalities. Also, mentioning technical limitations would be informative. For example: how deep the authors can perform 2p-imaging through the prism? was the "30-100um" maximum depth the authors could get?

2. In the introduction, it seems that the authors intended to refer to Poulin et al. 2018 regarding molecular/anatomical heterogeneity of dopamine neurons, but they inadvertently cited Poulin et al. 2016 (a general review on scRNAseq). Additionally, the statement that "dopamine neurons that project to the PFC show unique genetic profiles (line 85)" requires clarification, as Poulin et al. 2018 did not specifically establish this point. Instead, they found at least the Vglut2/Cck+ population projects into mPFC, and they did not reject the possibility of other subclasses projecting to mPFC. Rather, they observed denser innervation with DAT-cre, suggesting that non-Vglut2/Cck populations would also project to mPFC. Discuss the potential molecular heterogeneity among mPFC dopamine axons in light of the sampling limitation mentioned earlier.

3. I find the data presented in Figure 2 to be odd. Firstly, the latency of shock responses in the representative axons (right panels of G, H) is consistently very long - nearly 500ms. It raises a query whether this is a biological phenomenon or if it stems from a potential technical artifact, possibly arising from an issue in synchronization between the 2-photon imaging and stimulus presentation. My reservations are compounded by the notable absence of comprehensive information concerning the synchronization of the experimental system in the method section. Secondly, there appear to be irregularities in Panel J. While the authors indicate that "Significant axons were classified as either reward-preferring (cyan) or aversive-preferring (magenta), based on whether the axons are above or below the unity line of the reward/aversive scatter plot (Line 566)," a cyan dot slightly but clearly deviates above the unity line (around coordinates (x, y) = (20, 21)). This needs clarification. Lastly, when categorizing axons for analysis of conditioning data in Fig3 (not Fig2), the authors stated "The color-coded classification (cyan/magenta) was based on k-means clustering, using the responses before classical conditioning (Figure 2J)". I do not understand why the authors used different classification methods for two almost identical datasets.

4. In connection with Point 3, conducting separate statistical analyses for aversive and rewarding stimuli would offer a fairer approach. This could potentially reveal a subset of axons that display responses to both aversive and appetitive stimuli, aligning more accurately with the true underlying dynamics. Moreover, the characterization of Figure 2J as a bimodal distribution while disregarding the presence of axons responsive to both aversive and appetitive cues seems somewhat arbitrary and circular logic. A more inclusive consideration of this dual-responsive population could contribute to a more comprehensive interpretation.

5. The contrast in initialization to novel cues between aversive and appetitive axons mirrors findings in other areas, such as the tail-of-striatum (TS) and ventral striatum (VS) projecting dopamine neurons (Menegas et al., 2017, not 2018). You might consider citing this very relevant study and discussing potential collateral projections between mPFC and TS or VS.

6. The use of correlation values (here >0.65) to group ROIs into axons is common but should be justified based on axon density in the FOV and imaging quality. It's important to present the distribution of correlation values and demonstrate the consistency of results with varying cut-off values. Also, provide insights into the reliability of aversive/appetitive classifications for individual ROIs with high correlations. Importantly, if you do the statistical testing and aversive/appetitive classifications for individual ROIs with above-threshold high correlation (to be grouped into the same axon), do they always fall into the same category? How many false positives/false negatives are observed?

"Our results remained similar for different correlation threshold values (Line 556)" (data not shown) is obsolete.

Reviewer #2 (Public Review):

Summary:
This study aims to address existing differences in the literature regarding the extent of reward versus aversive dopamine signaling in the prefrontal cortex. To do so, the authors chose to present mice with both a reward and an aversive stimulus during different trials each day. The authors used high spatial resolution two-photon calcium imaging of individual dopaminergic axons in the medial PFC to characterize the response of these axons to determine the selectivity of responses in unique axons. They also paired the reward (water) and an aversive stimulus (tail shock) with auditory tones and recorded across 12 days of associative learning.

The authors find that some axons respond to both reward and aversive unconditioned stimuli, but overall, there is a strong preference to respond to aversive stimuli consistent with expectations from prior studies that used other recording methods. The authors find that both of their two auditory stimuli initially drive responses in axons, but that with training axons develop more selective responses for the shock associated tone indicating that associative learning led to changes in these axon's responses. Finally, the authors use anticipatory behaviors during the conditioned stimuli and facial expressions to determine stimulus discrimination and relate dopamine axons signals with this behavioral evidence of discrimination. This study takes advantage of cutting-edge imaging approaches to resolve the extent to which dopamine axons in PFC respond appetitive or aversive stimuli. They conclude that there is a strong bias to respond to the aversive tail shock in most axons and weaker more sparse representation of water reward.

Strengths:
The strength of this study is the imaging approach that allows for investigation of the heterogeneity of response across individual dopamine axons, unlike other common approaches such as fiber photometry which provide a measure of the average population activity. The use of appetitive and aversive stimuli to probe responses across individual axons is another strength.

Weaknesses:
A weakness of this study is the design of the associative conditioning paradigm. The use of only a single reward and single aversive stimulus makes it difficult to know whether these results are specific to the valence of the stimuli versus the specific identity of the stimuli. Further, the reward presentations are more numerous than the aversive trials making it unclear how much novelty and habituation account for results. Moreover, the training seems somewhat limited by the low number of trials and did not result in strong associative conditioning. The lack of omission responses reported may reflect weak associative conditioning. Finally, the study provides a small advance in our understanding of dopamine signaling in the PFC and lacks evidence for if and what might be the consequence of these axonal responses on PFC dopamine concentrations and PFC neuron activity.

Reviewer #3 (Public Review):

Summary:

The authors image dopamine axons in medial prefrontal cortex (mPFC) using microprism-mediated two-photon calcium imaging. They image these axons as mice learn that two auditory cues predict two distinct outcomes, tailshock or water delivery. They find that some axons show a preference for encoding of the shock and some show a preference for encoding of water. The authors report a greater number of dopamine axons in mPFC that respond to shock. Across time, the shock-preferring axons begin to respond preferentially to the cue predicting shock, while there is a less pronounced increase in the water-responsive axons that acquire a response to the water-predictive cue (these axons also increase non-significantly to the shock-predictive cue). These data lead the authors to argue that dopamine axons in mPFC preferentially encode aversive stimuli.

Strengths:

The experiments are beautifully executed and the authors have mastered an impressively complex technique. Specifically, they are able to image and track individual dopamine axons in mPFC across days of learning. This technique is used the way it should be: the authors isolate distinct dopamine axons in mPFC and characterize their encoding preferences and how this evolves across learning of cue-shock and cue-water contingencies. Thus, these experiments are revealing novel information about how aversive and rewarding stimuli is encoded at the level of individual axons, in a way that has not been done before. This is timely and important.

Weaknesses:

The overarching conclusion of the paper is that dopamine axons preferentially encode aversive stimuli. This is prevalent in the title, abstract, and throughout the manuscript. This is fundamentally confounded. As the authors point out themselves, the axonal response to stimuli is sensitive to outcome magnitude (Supp Fig 3). That is, if you increase the magnitude of water or shock that is delivered, you increase the change in fluorescence that is seen in the axons. Unsurprisingly, the change in fluorescence that is seen to shock is considerably higher than water reward. Further, when the mice are first given unexpected water delivery and have not yet experienced the aversive stimuli, over 40% of the axons respond [yet just a few lines below the authors write: "Previous studies have demonstrated that the overall dopamine release at the mPFC or the summed activity of mPFC dopamine axons exhibits a strong response to aversive stimuli (e.g., tail shock), but little to rewards", which seems inconsistent with their own data]. Given these aspects of the data, it could be the case that the dopamine axons in mPFC encodes different types of information and delegates preferential processing to the most salient outcome across time. The use of two similar sounding tones (9Khz and 12KHz) for the reward and aversive predicting cues are likely to enhance this as it requires a fine-grained distinction between the two cues in order to learn effectively.

There is considerable literature on mPFC function across species that would support such a view. Specifically, theories of mPFC function (in particular prelimbic cortex, which is where the axon images are mostly taken) generally center around resolution of conflict in what to respond, learn about, and attend to. That is, mPFC is important for devoting the most resources (learning, behavior) to the most relevant outcomes in the environment. This data then, provides a mechanism for this to occur in mPFC. That is, dopamine axons signal to the mPFC the most salient aspects of the environment, which should be preferentially learned about and responded towards. This is also consistent with the absence of a negative prediction error during omission: the dopamine axons show increases in responses during receipt of unexpected outcomes, but do not encode negative errors. This supports a role for this projection in helping to allocate resources to the most salient outcomes and their predictors, and not learning per se. Below are a just few references from the rich literature on mPFC function (some consider rodent mPFC analogous to DLPFC, some mPFC), which advocate for a role in this region in allocating attention and cognitive resources to most relevant stimuli, and do not indicate preferential processing of aversive stimuli.

References:
1. Miller, E. K., & Cohen, J. D. (2001). An integrative theory of prefrontal cortex function. Annual review of neuroscience, 24(1), 167-202.
2. Bissonette, G. B., Powell, E. M., & Roesch, M. R. (2013). Neural structures underlying set-shifting: roles of medial prefrontal cortex and anterior cingulate cortex. Behavioural brain research, 250, 91-101.
3. Desimone, R., & Duncan, J. (1995). Neural mechanisms of selective visual attention. Annual review of neuroscience, 18(1), 193-222.
4. Sharpe, M. J., Stalnaker, T., Schuck, N. W., Killcross, S., Schoenbaum, G., & Niv, Y. (2019). An integrated model of action selection: distinct modes of cortical control of striatal decision making. Annual review of psychology, 70, 53-76.
5. Ridderinkhof, K. R., Ullsperger, M., Crone, E. A., & Nieuwenhuis, S. (2004). The role of the medial frontal cortex in cognitive control. science, 306(5695), 443-447.
6. Nee, D. E., Kastner, S., & Brown, J. W. (2011). Functional heterogeneity of conflict, error, task-switching, and unexpectedness effects within medial prefrontal cortex. Neuroimage, 54(1), 528-540.
7. Isoda, M., & Hikosaka, O. (2007). Switching from automatic to controlled action by monkey medial frontal cortex. Nature neuroscience, 10(2), 240-248.