Author Response:
Reviewer #1:
In this study, Sias and colleagues examined the neural mechanism underlying stimulus-outcome associations using a Pavlovian-to-instrumental transfer (PIT) task in rats. Rats were first trained in a Pavlovian conditioning task in which two different auditory stimuli (white noise or tone) predicted different outcomes (sucrose solution or food pellet). The rats were then subjected to an instrumental conditioning and a PIT test to examine stimulus-outcome associations. The authors first used fiber photometry to examine the bulk calcium signals from the basolateral amygdala (BLA) during Pavlovian conditioning, and found that a population of BLA neurons are activated at the onset of a conditioned stimulus and at the time of reward retrieval. The response was observed from the first day and the magnitude was relatively constant over the entire period (8 days), indicating that the population activity contained responses to novel auditory stimuli. The authors then performed optogenetic inhibitions of BLA neurons at the time of reward delivery and consumption during Pavlovian conditioning. Although the BLA inhibition did not affect the acquisition of Pavlovian approach to the reward port, it impaired a facilitation of pressing the lever associated with a specific outcome predicted by an auditory cue, supporting a role of BLA in learning to predict specific outcomes, not just reward generally. The authors also examined the role of interactions between BLA and the lateral orbitofrontal cortex (lOFC), first by inactivating lOFC axons in BLA, and then by a serial circuit disconnection experiment combining optogenetic and pharmacogenetic inhibitions of specific projections.
Although the role of BLA and lOFC in learning has been studied extensively, this study extends these studies by performing temporally specific inhibitions using optogenetics, axonal inactivation, and serial disconnection experiments. The finding that the BLA-lOFC circuit is not necessary for the acquisition of simple Pavlovian approaches but critical for outcome-specific stimulus-outcome associations is surprising. The authors performed sophisticated and difficult experiments, and the experiments are generally well done. The manuscript is clearly written, and the results are discussed carefully.
We appreciate this thoughtful evaluation of our manuscript.
I have one relatively minor concern regarding the description of the serial disconnection experiment. Overall, the manuscript provides interesting results and warrants publication at eLife.
1) The use of a serial circuit disconnection experiment (Figure 5) is elegant and informative. However, the authors could have achieved almost the same goal by bilateral inactivation of axonal terminals of lOFC->BLA projections during the encoding phase or BLA->lOFC projections during the retrieval phase.
We did these bilateral axonal terminal inactivation experiments. They showed us that the lOFCBLA pathway is involved in the learning (Figure 4) and the BLAlOFC pathway is involved in the retrieval (Lichtenberg et al., 2017) of stimulus-outcome memories. But these experiments are not capable of providing information on whether these pathways form a circuit. That is, whether BLAlOFC projection activity mediates the use of the associative information that is learned via activation of lOFCBLA projections or whether these pathways tap in to independent information streams. Our goal with the serial disconnection experiment was to address this specific circuit question. We have clarified the logic of this experiment.
- Results on Pg. 10: “But it remains unknown whether BLAlOFC projection activity mediates the use of the associative information that is learned via activation of lOFCBLA projections. That is, whether lOFCBLAlOFC is a functional stimulus-outcome memory encoding and retrieval circuit or whether lOFCBLA and BLAlOFC projections tap in to independent, parallel information streams. Indeed, stimulus-outcome memories are highly complex including multifaceted information about outcome attributes (e.g., value, taste, texture, nutritional content, category, probability, timing, etc.) and related consummatory and appetitive responses (Delamater & Oakeshott, 2007). Therefore, we next asked whether the lOFCBLA and BLAlOFC pathways form a functional stimulus-outcome memory encoding and retrieval circuit, i.e., whether the sensory-specific associative information that requires lOFCBLA projections to be encoded also requires activation of BLAlOFC projections to be used to guide decision making, or whether these are independent, parallel pathways, tapping into essential but independent streams of information. To arbitrate between these possibilities, we multiplexed optogenetic and chemogenetic inhibition to perform a serial circuit disconnection. We disconnected lOFCBLA projection activity during stimulus-outcome learning from BLAlOFC projection activity during the retrieval of these memories at the PIT test (Figure 5a)… …If BLAlOFC projection activity mediates the retrieval of the sensory-specific associative memory that requires activation of lOFCBLA projections to be encoded, then we will have bilaterally disconnected the circuit, attenuating encoding in one hemisphere and retrieval in the other, thereby disrupting the ability to use the stimulus-outcome memories to guide choice behavior during the PIT test. If, however, these pathways mediate parallel information streams, i.e., independent components of the stimulus-outcome memory, the expression of PIT should be intact because one of each pathway is undisrupted to mediate its individual component during each phase.”
Furthermore, if there are contralateral projections, the experimental design might have a problem. Please clarify these points.
This is a great point that we did not discuss as clearly as we could have. We appreciate the opportunity to clarify our logic. There are both ipsilateral and contralateral lOFCBLA projections. For this reason, we optically inactivated both the ipsilateral and contralateral lOFC input to the BLA of one hemisphere, leaving both the ipsilateral and contralateral lOFCBLA projections to the BLA of the other hemisphere intact. To achieve this, we expressed the inhibitory opsin ArchT bilaterally into the lOFC and placed the optical fiber unilaterally in the BLA. BLAlOFC projections are largely ipsilateral and so we expressed the inhibitory designer receptor hM4Di unilaterally in the BLA and put a guide cannula for CNO infusion over the hemisphere opposite to that in which we had placed the optical fiber. We have clarified this logic in the revised results and methods:
- Results Pg. 10 ¶2: “For the disconnection group (N = 10), we again expressed ArchT bilaterally in lOFC neurons (Figure 5b-d) to allow expression in lOFC axons and terminals in the BLA. This time, we implanted the optical fiber only unilaterally in the BLA (Figure 5b-d), so that green light (532nm, ~10mW), delivered again during Pavlovian conditioning for 5 s during the delivery and retrieval of each reward during each cue, would inhibit both the ipsilateral and contralateral lOFC input to the BLA of only one hemisphere. In these subjects, we also expressed the inhibitory designer receptor human M4 muscarinic receptor (hM4Di) unilaterally in the BLA of the hemisphere opposite to the optical fiber and in that same hemisphere placed a guide cannula over the lOFC near hM4Di-expressing BLA axons and terminals (Figure 5b-d). This allowed us to infuse the hM4Di ligand clozapine-n-oxide (CNO; 1 mM in 0.25 µl) prior to the PIT test to unilaterally inhibit BLA terminals in the lOFC, which are largely ipsilateral (Lichtenberg et al., 2017), in the hemisphere opposite to that for which we had inhibited lOFCBLA projection activity during Pavlovian conditioning. Thus, we optically inhibited the lOFCBLA stimulus-outcome learning pathway in one hemisphere at each stimulus-outcome pairing during Pavlovian conditioning, and chemogenetically inhibited the putative BLAlOFC retrieval pathway in the opposite hemisphere during the PIT test in which stimulus-outcome memories must be used to guide choice.”
- Methods on Pg. 19 ¶1: “The disconnection group (N = 10) was infused with AAV encoding the inhibitory opsin ArchT (rAAV5-CAMKIIa-eArchT3.0-eYFP; 0.3 µl) bilaterally at a rate of 0.1 µl/min into the lOFC (AP: +3.3; ML: ±2.5; DV: -5.4 mm from bregma) using a 28-gauge injector tip. Injectors were left in place for an additional 10 minutes. An optical fiber (200 µm core, 0.39 NA) held in a ceramic ferrule was implanted unilaterally (hemisphere counterbalanced across subjects) in the BLA (AP: -2.7; ML: ±5.0; DV: -7.7 mm from dura) to allow subsequent light delivery to both the ipsilateral and contralateral ArchT-expressing axons and terminals in the BLA of only one hemisphere. During the same surgery, in the hemisphere contralateral to optical fiber placement, a second AAV was infused unilaterally at a rate of 0.1 µl/min into the BLA (AP: -3.0; ML: ±5.1; DV: -8.6 from bregma) to drive expression of the inhibitory designer receptor human M4 muscarinic receptor (hM4Di; pAAV8-hSyn-hM4D(Gi)-mCherry, Addgene; 0.5 µl). A 22-gauge stainless-steel guide cannula was implanted unilaterally above the lOFC (AP: +3.0; ML: ±3.2: DV: -4.0) of the BLA-hM4Di hemisphere to target the hM4D(Gi)-expressing axonal terminals, which are predominantly ipsilateral.”
Also, the control experiments are now shown in Figure 5-2. It would be useful to have it in a main figure.
We have incorporated the ipsilateral control group data into the main Figure 5 (Pg. 11). As you can see below, because there were no differences between the two control groups (contralateral fluorophore only eYFP/mCherry & ipsilateral ArchT/hM4Di), we combined them into a single control group for comparison to the disconnection group. The individual data points in Figure 5 are coded by control group (eYFP/mCherry solid lines and circles, ipsilateral ArchT/hM4Di dashed lines and triangles). We also provide the data with the control groups disaggregated showing a comparison between all three groups in Figure 5-2 (Pg. 43)
- Results on Pg. 10 ¶2: “The control group received identical procedures with the exception that viruses lacked ArchT and hM4Di (N = 8). To control for unilateral inhibition of each pathway without disconnecting the circuit, a second control group (N = 8) received the same procedures as the experimental contralateral ArchT/hM4Di disconnection group, except with BLA hM4Di and the lOFC guide cannula in the same hemisphere as the optical fiber used to inactivate lOFC axons and terminals in the BLA (Figure 5-1). Thus, during the PIT test, for this group the BLAlOFC pathway was chemogenetically inactivated in the same hemisphere in which the lOFCBLA pathway had been optically inactivated during Pavlovian conditioning, leaving the entire circuit undisrupted in the other hemisphere. These control groups did not differ on any measure and so were collapsed into a single control group [(Pavlovian training, Training: F(2.2,31.3) = 12.96, P < 0.0001; Control group type: F(1,14) = 0.02, P = 0.89; Group x Training: F(7.98) = 0.76, P = 0.62) (PIT Lever presses, Lever: F(1,14) = 14.68, P = 0.002; Control group type: F(1,14) = 0.38, P = 0.55; Group x Lever: F(1,14) = 0.43, P = 0.52) (PIT Food-port entries, t14 = 0.72, P = 0.48)]. See also Figure 5-2 for disaggregated control data.”
Reviewer #2:
This manuscript aimed to dissociate two potential roles of the basolateral amygdala (BLA) in choice behavior: (1) contributing to sensory-specific stimulus-outcome memories or (2) assigning general valence to a reward-predictive cue. The authors used a well-validated Pavlovian-to-instrumental transfer (PIT) test with a series of circuit manipulations to show that lateral OFC to BLA projections are necessary for learning specific cue-outcome associations, rather than general valence, and that return BLA to lateral OFC projections are important for using that learned information in the PIT test.
Overall, this paper addresses a question that is important to anyone studying amygdala or orbitofrontal function. The study is well-designed, the multiplexed opto-chemogenetics experiment is particularly creative, and there are convincing results with appropriate controls.
We appreciate this thoughtful evaluation of our manuscript.
I only have a few minor questions about the calcium signals reported in the first portion of the manuscript. First, there is a steep rise in calcium signal in panel 1f, suggesting that the signal is time-locked to the cue. However, there is a qualitatively different response to rewards in 1g. Is this just because it's more difficult to time-lock to the animal's movements than an experimentally-controlled cue? Or is it possible that there's another source in the experimental set-up that could be triggering the response. For example, does the reward delivery make an audible sound?
You are absolutely right that the signal is not as time-locked to the reward collection because the rats collected the reward at somewhat variable times after delivery, which is, indeed, signaled by a subtle, but audible cue (pellet dispenser click or pump onset). To clarify this, we have now included Figure 1-4 (Pg. 35) showing the BLA calcium response to reward delivery. As you can see, the BLA reward response is also detectable when the data are aligned to the reward delivery, but there is still not as sharp of a response as that to the cue onset, likely owing to slight variability in the precise moment that the reward is perceived.
- Reference in Results Pg. 5 ¶2: “The same BLA reward response could also be detected when the data were aligned to reward delivery (Figure 1-4).”
Second, in Fig 2, is there any change in the reward response across training sessions, or is this signal also stable?
This is an interesting question, but unfortunately one we are not able to answer because we only recorded during one unpredicted reward delivery session after the last CSØ session. Because we saw that the BLA GCaMP response to the CSØ decreased and was nearly completely absent on the last day of exposure we wanted to make sure that this was not due to signal degradation over time, so we recorded during an unexpected reward session to serve as this positive control. We have now clarified this logic in the results.
- Results Pg. 6 ¶1: “To check whether the decline of the CSØ response was due simply to signal degradation over time, following the last CSØ session we recorded BLA calcium responses to unpredicted reward delivery. Rewards were capable of robustly activating the BLA (Figure 2g-i; peak; t5 = 2.93, P = 0.03; AUC; t5 = 4.07, P = 0.01). This positive control indicates that the decline of the BLA CSØ response was due to stimulus habituation, not signal degradation.”
Reviewer #3:
Summary:
This work tests the hypothesis that the reciprocal connections between the BLA and lOFC are needed to encode sensory-specific reward memories, as well as retrieve this same information once it has been learned in order guide decision making. The authors first use fiber photometry to measure the activity of excitatory BLA neurons during Pavlovian conditioning of two specific cues with two specific reward outcomes and find that transient responses are evident in BLA at cue onset and each time there is a cue contingent attempt to retrieve a reward. Using this information about event encoding in BLA, the authors go on to use optogenetics to inhibit BLA activity driven by lOFC inputs to BLA following reward retrieval attempts without affecting overall conditioned approach behavior. This manipulation has the effect of disrupting encoding of sensory-specific reward memories as it impairs the animals' subsequent performance on an outcome-specific Pavlovian instrumental transfer test. Since the authors have previously demonstrated that BLA inputs to lOFC are important for retrieving sensory-specific reward memories to affect decision making in the same PIT procedure, they go on to use an innovative serial disconnection approach using chemogenetic and optogenetic tools to show that inhibiting either pathway in opposing hemispheres, simultaneously, has comparable effects on outcome-specific PIT performance as bilateral inhibition of either pathway in isolation. Overall this is a compelling demonstration that inputs from BLA to lOFC and lOFC to BLA act in a coordinated manner to facilitate appetitive decision making.
Strengths:
These experiments build directly on the authors' prior demonstrations that lOFC projections to BLA are important for encoding incentive value but not for the retrieval of appetitive reward associations.
An elegant use of an outcome-specific Pavlovian instrumental transfer (PIT) procedure to demonstrate the important contributions of projections between the BLA and lOFC in encoding and retrieving stimulus-outcome reward associations.
The use of GCaMP measurements of BLA activity to temporally constrain optogenetic inhibition of lOFC inputs to BLA following reward retrieval, allowing specific conclusion about how encoding of stimulus-outcome memories mediated by lOFC inputs to BLA.
The authors utilize a measure of Pavlovian conditioned approach behavior to convincingly demonstrate that the effects of their optogenetic manipulations during Pavlovian conditioning on behavior during PIT is sensory specific and due to potentially confounding changes in motivation or learning.
We appreciate this thoughtful evaluation of our manuscript.
Weaknesses:
The conditioned approach responses appear to asymptote after two out of the eight Pavlovian conditioning sessions. Although the authors have run a control experiment in which they show that novelty contributes to the GCaMP responses measured in BLA at cue onset in early sessions, they do not clearly demonstrate learning related changes in GCaMP responses across sessions to either cue or reward retrieval. Thus, it isn't necessarily clear how quickly the sensory-specific reward memories are formed in BLA and if repeated stimulus-outcome pairings, particularly once general approach behavior reaches asymptote, actually serve to increasingly strengthen the memory.
We agree with this limitation that our report and are actively working to address these interesting questions in our ongoing work. Indeed, a learning related-change in the BLA response can only be inferred from the present data and is not directly demonstrated. In the present experiment the nature of the memory is tested after learning, precluding understanding of the precise time course of the development of the sensory-specific stimulus-outcome memory. Future work should incorporate an online neural and/or behavioral assessment of sensory-specific reward memory encoding during learning to well address this important question.
No explanation is provided for how the transient BLA GCaMP responses at cue onset sustain stimulus-outcome memory encoding at the time of reward. A straightforward account would be a sustained response to the cue that overlaps with the GCaMP response to reward retrieval. In addition there is no attempt to transiently inactivate the entire BLA or specific pathways at cue onset to determine how simple cue encoding affects subsequent performance in the PIT paradigm.
This is an excellent point. We were somewhat surprised to see only a transient response to the CS onset. This suggests to us that perhaps there is a more sustained response elsewhere in the brain (or even in a different cell type in the BLA). Perhaps this sustained response follows the transient response detected here.
We also agree that it is an important question (and limitation of the current work) of whether the BLA response to the cue is important for S-O memories. This is also a question we are addressing on in our ongoing work. We have acknowledged both this limitation/interesting question in the revised manuscript.
- Discussion Pg. 13 ¶1: “Future work is needed to reveal the precise information content encoded by BLA neurons during reward experience that confers their function in the formation of stimulus-outcome memories, though BLA neurons will respond selectively to unique food rewards (Liu et al., 2018), which could support the generation of sensory-specific reward memories. Whether BLA cue responses are also important for encoding stimulus-outcome memories is another important question exposed by the current results.”
The multiplexed chemogenetic and optogenetic serial disconnection approach is too coarse a manipulation to support the claim that reciprocal connections between the BLA and lOFC support encoding and retrieval of the same information. To make this claim it is necessary to use detailed functional assays of the activity in each pathway to determine what information they code during the Pavlovian conditioning and PIT procedures.
We completely agree with this excellent point. We appreciate the reviewer pointing out how our language led to an interpretation that is not supported by the current data. Indeed, the data do not show whether the same information is transmitted between lOFCBLA and BLAlOFC and that need not be the case for these projections to function in a circuit. To remedy this, we have removed the ‘same information’ language throughout the manuscript, including in the abstract (Pg. 2), results (Pg. 9-11), discussion (Pg. 13-14), and methods (Pg. 20-21). We have brought our framing and interpretation of the disconnection results much closer to the present data. For example:
Results Pg. 10 ¶1: “Therefore, we next asked whether the lOFCBLA and BLAlOFC pathways form a functional stimulus-outcome memory encoding and retrieval circuit, i.e., whether the sensory-specific associative information that requires lOFCBLA projections to be encoded also requires activation of BLAlOFC projections to be used to guide decision making, or whether these are independent, parallel pathways, tapping into essential but independent streams of information.”
Results Pg. 11 ¶1: “…indicating that the lOFC and BLA form a bidirectional circuit for the encoding (lOFCBLA) and use (BLAlOFC) of appetitive stimulus-outcome memories.”
Discussion Pg. 14 ¶1: “Here, using a serial disconnection procedure, we found that during reward choice BLAlOFC projection activity mediates the use of the sensory-specific associative information that is learned via activation of lOFCBLA projections. Thus, lOFCBLAlOFC is a functional circuit for the encoding (lOFCBLA) and subsequent use (BLAlOFC) of sensory-specific reward memories to inform decision making.”
We have also included the important caveat that future work with detailed characterization of the activity of each pathway is needed to draw conclusions on the information content conveyed by each pathway:
- Discussion Pg. 14 ¶2: “The precise information content conveyed by each component of the lOFC-BLA circuit and how it is used in the receiving structure is a critical follow-up question that will require a cellular resolution investigation of the activity of each pathway.”