The BLA is critical for learning that previously neutral cues can predict future punishments or rewards and for assigning valence to those cues (Baxter and Murray, 2002; Janak and Tye, 2015). The BLA receives dense cholinergic input (Woolf, 1991; Zaborszky et al., 2012) and we speculated that, since ACh signaling is involved in both attention and several types of learning (Picciotto et al., 2012), it could be essential for learning about cues that predict salient events, such as reward delivery. Based on data showing that ACh neurons fire in response to unexpected or salient events (Hangya et al., 2015), we also hypothesized that ACh release might vary as mice learn a cue-reward contingency. Therefore, we designed a cue-reward learning task in which food-restricted mice were trained to perform a nose poke when signaled by a cue (tone) to receive a palatable reward (Ensure) on a 30 s variable intertrial interval (ITI; Figure 1A–D). We injected adeno-associated virus (AAV) carrying an improved version of the fluorescent ACh sensor GRAB_ACh3.0 (ACh3.0; Jing et al., 2018; Jing et al., 2019) construct into the BLA of mice and implanted an optical fiber above the BLA to record ACh signaling during the cue-reward learning task (Figure 2A + Figure 2—figure supplement 1A).

Figure 1

Download asset Open asset

Figure 2 with 9 supplements see all

Download asset Open asset

During the Pre-Training phase of the task, mice received reward and receptacle light presentation for performing a nose poke in the active port during tone presentation (Figure 1C, purple active nose poke coincident with tone) but there was no consequence for an incorrect nose poke (Figure 1C, red active nose poke not coincident with tone). Animals quickly learned to make a high number of responses over the course of each Pre-Training session. In this paradigm, mice obtained most available rewards by day 5 of Pre-Training (Figure 2B, blue shaded region). However, this phase of training did not promote learning of the cue-reward contingency, (i.e. that they should only nose poke during tone presentation) seen by the high number of incorrect nose pokes (Figure 2—figure supplement 2A, blue shaded region). Mice performed roughly eightfold more incorrect nose pokes than correct nose pokes, suggesting that mice were not attending to the task contingency. The Training phase of the task was identical to Pre-Training except incorrect nose pokes resulted in a 5 s timeout, during which the house light was illuminated, that concluded with a restarting of the ITI timer (Figure 1D, red active nose poke not coincident with tone). On day 1 of the Training phase, all animals earned fewer rewards (Figure 2B, pink shading) and, while still high, incorrect nose pokes dropped (Figure 2—figure supplement 2A, pink shading). Animals that did not meet acquisition criterion by day 9 (defined as consistently earning 20 or more rewards per session, Figure 2B, white horizontal line) were moved to a 20 s variable ITI to promote responding (Figure 2B, pink shading day 10). Following the change in ITI, mice acquired the cue-reward behavior at different rates. After acquisition, animals were switched to Extinction training in which correct nose pokes did not result in reward delivery, and all mice decreased nose poke responding (Figure 2B + Figure 2—figure supplement 2A, orange shading).

During Pre-Training, when there were high numbers of both correct and incorrect nose pokes, there was a large increase in ACh release following correct nose pokes, which were followed by reward delivery and receptacle light, but not incorrect nose pokes (Figure 2C + Figure 2—figure supplement 1B–C). We used bootstrapped confidence intervals (bCIs) to determine when transients were statistically significant (bCI did not contain the null of 0 [Jean-Richard-Dit-Bressel et al., 2020]). Correct, but not incorrect, nose poke trials consistently showed a sustained, significant increase in fluorescence close to the time of nose poke onset (Figure 2C). We also observed a significant decrease in fluorescence for most mice around 2–4 s after correct nose poke, which corresponds to the time of reward retrieval.

ACh release occurred in response to different events as mice learned the task (data for individual mice are shown in Figure 2D + Figure 2—figure supplement 1D–G and averaged data across all mice at key time points in the task is shown in Figure 2E + Figure 2—figure supplement 1H). During Pre-Training rewarded trials, the highest levels of ACh release occurred close to the time of correct nose pokes (NP), with a smaller peak at the time of reward retrieval (entry into the reward receptacle, Rec). As Training began, the ACh release during reward trials shifted dramatically toward the time of reward retrieval, likely because the animals were learning that many nose pokes did not result in reward delivery. Incorrect nose pokes that triggered a timeout were also followed by a modest but non-significant increase in BLA ACh levels (Figure 2—figure supplement 2B–H). As mice began to learn the contingency (Figure 2E + Figure 2—figure supplement 1H, 10 rewards), the peak ACh release during rewarded trials shifted back to the time of the correct nose poke response. As animals approached the acquisition criterion (Figure 2E + Figure 2—figure supplement 1H, Acq.), ACh level significantly increased at the time of the tone, suggesting that as animals learned the cue-reward contingency, the tone became a more salient event. At this time point, there was still a peak at the time of reward, but its magnitude was diminished. After task acquisition, the increase in ACh following correct nose pokes remained but was diminished, and incorrect nose pokes did not elicit apparent ACh release (Figure 2—figure supplement 2C–H, Acq.). During Extinction, ACh release to tone onset diminished. We replicated this experiment in an independent cohort of mice and found similar results (Figure 2—figure supplements 3–4). Mice in this replicate cohort learned in a similar fashion (Figure 2—figure supplement 3B + 4A) but met the acquisition criteria faster than initial mice because aspects of the behavioral setup were optimized (3D printed wall extensions) to allow the imaging apparatus to be used inside sound attenuating chambers (see Materials and methods section). One difference observed in this group that learned the task more rapidly, was small magnitude, but significant, increases in BLA ACh release following tone onset late in Pre-Training (Figure 2—figure supplement 3C–I). As behavioral performance during the Training phase increased, ACh release to tone onset became more pronounced, as in the initial cohort.

In order to determine the source of the ACh released in the BLA during cue-reward learning, we recorded calcium dynamics as a measure of cell activity of ChAT⁺ NBM terminal fibers in the BLA (NBM-BLA), since the NBM is a major source of cholinergic input to the BLA (Jiang et al., 2016; Woolf, 1991; Zaborszky et al., 2012). We injected AAV carrying a Cre-recombinase-dependent, genetically-encoded calcium indicator (DIO-GCaMP7s) into the NBM of ChAT-IRES-Cre mice and implanted an optical fiber above the ipsilateral BLA (Figure 2F + Figure 2—figure supplement 5A–C). As with the ACh3.0 sensor, there was a significant increase in NBM-BLA cholinergic terminal activity following correct, but not incorrect, nose pokes (Figure 2H + Figure 2—figure supplement 5D–E). NBM-BLA cholinergic terminal activity evolved across phases of the reward learning task as was seen for ACh levels in the BLA (data for each mouse shown in Figure 2I + Figure 2—figure supplement 5F–G, averaged across all mice at key time points in the task shown in Figure 2J + Figure 2—figure supplement 7G). Strikingly, NBM-BLA cholinergic terminal activity most closely followed correct nose pokes in Pre-Training and shifted primarily to tone onset as mice learned the contingency during Training. As in the replication cohort for the ACh sensor, small magnitude, but significant, increases in terminal activity were observed following tone onset late in Pre-Training (Figure 2J + Figure 2—figure supplement 7G). Incorrect nose pokes that resulted in a timeout in Training sessions were followed by a modest increase in NBM-BLA cholinergic terminal activity before task acquisition (Figure 2—figure supplement 6B–E). During Extinction, activity of NBM-BLA terminals following tone onset diminished. These findings were replicated in an independent cohort of mice, which we combined for across-mouse statistical analyses (Figure 2—figure supplements 7–8).

In order to record NBM-BLA cholinergic terminal activity and BLA ACh levels simultaneously in the same mouse, we injected AAV carrying a construct for Cre-recombinase dependent red-shifted genetically-encoded calcium indicator (DIO-jRCaMP1b) into the NBM of ChAT-IRES-Cre mice, ACh3.0 sensor into the ipsilateral BLA, and implanted a fiber above the BLA (Figure 2—figure supplement 9A–E, mouse 1). DIO-jRCaMP1b was also injected into the NBM of a wild-type littermate so Cre-mediated recombination would not occur to control for any crosstalk between the ACh3.0 and jRCaMP1b channels. While this was only a single animal and proof of principle for future studies, we found that NBM-BLA cholinergic terminal activity coincided with ACh levels (Figure 2—figure supplement 9F–G). Importantly, this relationship between ACh release and NBM-BLA terminal fiber activity was not explained by signal crosstalk (Figure 2—figure supplement 9H–I), further indicating that the BLA ACh measured comes at least in part from the NBM.

Glutamatergic principal cells are the primary output neurons of the BLA (Janak and Tye, 2015), and their firing is modulated by NBM-BLA cholinergic signaling (Jiang et al., 2016; Unal et al., 2015). BLA principal neurons can increase their firing in response to cues as animals learn cue-reward contingencies (Sanghera et al., 1979; Schoenbaum et al., 1998; Tye and Janak, 2007). Calcium/calmodulin-dependent protein kinase (CaMKII) has been shown to be a marker for glutamatergic BLA principal cells (Butler et al., 2011; Felix-Ortiz and Tye, 2014; McDonald, 1992; Tye et al., 2011). To determine whether ACh modulates principal neuron activity during cue-reward learning, we injected AAV carrying a Cre-recombinase dependent genetically encoded calcium indicator (DIO-GCaMP6s) into the BLA of CaMKIIα-Cre mice to record BLA principal cell activity during the learning task (Figure 3A + Figure 3—figure supplement 1A). As was seen for BLA ACh levels, there was a significant increase in BLA CaMKIIα cell activity following correct and a modest decrease in activity following incorrect nose pokes on the last day of Pre-Training (Figure 3B). However, the activity peaked later after the correct nose poke response (~2.5 s) compared to the ACh3.0 signal (~0.5 s) and appeared to align more tightly with reward retrieval (Figure 3—figure supplement 1B). As mice learned the task (Figure 3C + Figure 3—figure supplement 2A), BLA CaMKIIα cell activity increased first in response to reward and, after acquisition of the task, to the reward-predictive cue (individual data for each mouse shown in Figure 3D + Figure 3—figure supplement 1E–F, and averaged data across all mice at key time points in task is shown in Figure 3E + Figure 3—figure supplement 1G–H).

Figure 3 with 4 supplements see all

Download asset Open asset

During Pre-Training, the highest levels of BLA CaMKIIα cell activity followed reward retrieval. In addition, during the first few days of Training, BLA CaMKIIα cell activity after reward retrieval was higher than it was during Pre-Training, and the magnitude of response decreased as mice learned the contingency and earned more rewards, ultimately reaching similar intensity to that observed during Pre-Training. Concurrently, as mice approached acquisition of the task (Figure 3C, white horizontal line), BLA CaMKIIα cell activity significantly increased in response to tone onset (Figure 3D-E + Figure 3—figure supplement 1E–H, Acq.), suggesting that the recruitment of BLA CaMKIIα cell activity likely reflects the association of the cue with a salient outcome (Lutas et al., 2019; Sengupta et al., 2018). Incorrect nose pokes that triggered a timeout did not elicit a different response in CaMKIIα cell activity compared to before timeouts were incorporated (Figure 3—figure supplement 2B–G). In an independent cohort of mice, those with more posterior fiber tip placements (mice 4 + 7) replicated the primary findings (Figure 3—figure supplements 3–4).

Since ACh released by NBM-BLA terminals during Training shifted to tone onset during acquisition of cue-reward learning (Figure 2E,J), we hypothesized that ACh may potentiate learning the cue-reward contingency. We, therefore, tested whether increasing ACh release in BLA during learning could alter cue-reward learning by injecting AAV carrying a Cre-recombinase-dependent channelrhodopsin-EYFP (AAV-DIO-ChR2-EYFP) construct bilaterally into the NBM of ChAT-IRES-Cre transgenic mice and placing fibers over the BLAs to optically stimulate cholinergic terminals originating from the NBM selectively (Figure 4A + Figure 4—figure supplement 1). Optical control over ChAT⁺ NBM cells was verified by ex vivo slice recordings, depolarizations followed light flashes and clear action potentials were observed ex vivo (Figure 4B + Figure 4—figure supplement 2). After operant familiarization, ChAT⁺ NBM-BLA terminals were stimulated via bilateral optical fibers (2 s, 20 Hz, 25 ms pulses) triggered by a correct nose poke throughout both Pre-Training (Figure 4C) and Training (Figure 4D). Stimulation usually occurred during at least a portion of all three components of a rewarded trial: tone, correct nose poke, and reward retrieval, since these events were often separated by short latencies.

Figure 4 with 5 supplements see all

Download asset Open asset

As seen in previous experiments, during the Pre-Training phase animals made a high number of nose poke responses over the course of each session, obtained most available rewards by the last day (Figure 4E + Figure 4—figure supplement 3A, blue shading), and committed a very high number of incorrect nose pokes (Figure 4F + Figure 4—figure supplement 3B, blue shading). There were no differences in rewards earned (main effect of group (EYFP versus ChR2) in a two-way repeated-measures ANOVA, F (1, 9)=1.733, p=0.2205) or incorrect nose pokes (main effect of group (EYFP versus ChR2) in a two-way repeated-measures ANOVA, F (1, 9)=0.002433, p=0.9617) between the EYFP control (n = 5) and ChR2 (n = 6) groups during the Pre-Training phase (Figure 4E-F + Figure 4—figure supplement 3A–B, blue shading), suggesting that increasing BLA ACh signaling was not sufficient to modify behavior during the Pre-Training phase of the task.

On Day 1 of the Training phase, all animals earned fewer rewards (Figure 4E + Figure 4—figure supplement 3A, pink shading) and incorrect nose pokes remained high (Figure 4F + Figure 4—figure supplement 3B, pink shading). As the animals learned that a nose poke occurring outside of the cued period resulted in a timeout, both control EYFP and ChR2 groups learned the contingency and improved their performance, resulting in acquisition of the cue-reward task (20 rewards earned). However, significant group differences emerged, such that ChR2 mice earned significantly more rewards than EYFP controls (Figure 4E + Figure 4—figure supplement 3A, pink shaded; main effect of group (EYFP versus ChR2) in a two-way repeated-measures ANOVA, F (1, 9)=9.434, p=0.0133), and there was a significant Day × Group (EYFP versus ChR2) interaction (two-way repeated-measures ANOVA, F (11, 99)=3.210, p=0.0009). ChR2 mice also made significantly fewer incorrect nose pokes than control mice (Figure 4F + Figure 4—figure supplement 3B, pink shaded; two-way repeated-measures ANOVA, F (1, 9)=12.67, p=0.0061), suggesting that the ChR2 group learned the tone-reward contingency more quickly than the EYFP group. EYFP mice were able to reach the same peak cue-reward performance as the ChR2 group only after 4–6 additional days of training. Once peak performance was achieved, there was no difference in extinction learning between the groups (main effect of group (EYFP versus ChR2) in a two-way repeated-measures ANOVA, F (1, 9)=2.293, p=0.1643). While sex differences in the behavior were not formally tested side by side, an independent cohort of male mice (EYFP n = 7, ChR2 n = 7, Figure 4—figure supplement 4) was tested to determine whether both male and female mice would respond to ACh stimulation, revealing similar trends during Training for rewards earned (Figure 4—figure supplement 3C,E, pink shaded; two-way repeated-measures ANOVA, Group main effect (EYFP versus ChR2): F (1, 12)=3.636, p=0.0808, Day × Group interaction: F (11, 132)=3.033, p=0.0012) and incorrect nose pokes (Figure 4—figure supplement 3D,F, red shaded; two-way repeated-measures ANOVA, Group main effect (EYFP versus ChR2): F (1, 12)=4.925, p=0.0465).

In order to determine if optical stimulation of NBM-BLA cholinergic terminals improved performance in the task by increasing the rewarding value of the outcome, rather than enhancing cue-reward learning by some other means, we allowed mice to nose poke for optical stimulation rather than for Ensure (Figure 4—figure supplement 5A). There were no differences between the EYFP control and ChR2 groups (two-way repeated-measures ANOVA, F (1, 9)=0.6653, p=0.4357). We also tested whether NBM-BLA cholinergic terminal activation was reinforcing on its own by stimulating these terminals in a real-time place preference test. Mice were allowed to explore two similar compartments to determine baseline preference, and NBM-BLA cholinergic terminals were then stimulated in one of the two chambers to determine whether it increased time spent in the simulation-paired chamber. There was no difference between groups (Figure 4—figure supplement 5B, main effect of group (EYFP versus ChR2) in a two-way repeated-measures ANOVA, F (1, 9)=0.1311, p=0.7257) in place preference, confirming that optical activation of NBM-BLA cholinergic terminals is not innately rewarding. Stimulation of NBM-BLA cholinergic terminals also did not lead to changes in nose poke behavior in an uncued progressive ratio task (Figure 4—figure supplement 5C, main effect of group (EYFP versus ChR2) in a two-way repeated-measures ANOVA, F (1, 12)=0.0009814, p=0.975). Locomotor behavior was also not significantly affected by NBM-BLA cholinergic terminal activation (Figure 4—figure supplement 5D, two-way repeated-measures ANOVA, F (1, 9)=0.05804, p=0.8150.) Finally, to determine whether there was any effect of NBM-BLA cholinergic terminal stimulation on preference for, or avoidance of, a stressful environment, mice were tested for changes in time spent in the dark or light side due to laser stimulation in the Light/Dark Box test, and there were no differences between the groups (Figure 4—figure supplement 5E–F, unpaired t-tests, number of crosses: p=0.3223; time in light side: p=0.1565).

ACh signals through multiple receptor subtypes, with rapid, ionotropic signaling mediated through stimulation of nAChRs, and metabotropic signaling mediated through stimulation of mAChRs (Picciotto et al., 2012). To determine which ACh receptors were involved in this cue-reward learning task, mice were injected intraperitoneally with saline (n = 8), mecamylamine (non-competitive nicotinic antagonist, Mec, n = 9), scopolamine (competitive muscarinic antagonist, Scop, n = 8), or a combination of both antagonists (Mec+Scop, n = 9) 30 min prior to Pre-Training and Training, during the same epochs of the task in which optical stimulation was administered (Figure 5A). Like optical stimulation, blockade of ACh receptors during the Pre-Training phase of the task had no effect on rewards earned (Figure 5B + Figure 5—figure supplement 1A, blue shading, main effect of Group (antagonist) in a two-way repeated-measures ANOVA, F (3, 30)=1.285, p=0.2973) or on the large number of incorrect nose pokes (Figure 5C + Figure 5—figure supplement 1B, blue shading, main effect of Group (antagonist) in a two-way repeated-measures ANOVA, F (3, 30)=1.496, p=0.2356). By contrast, blockade of muscarinic signaling abolished the ability of mice to learn the correct cue-reward contingency during the Training period (Figure 5B + Figure 5—figure supplement 1A, pink shading, two-way repeated-measures ANOVA, Antagonist main effect: F (3, 30)=23.13, p<0.0001, Day x Antagonist interaction: F (33, 330)=10.79, p<0.0001), with these mice maintaining high levels of incorrect nose pokes for the duration of Training compared to Saline and Mec treated mice (Figure 5C + Figure 5—figure supplement 1B, pink shading, main effect of Group (antagonist) in a two-way repeated-measures ANOVA, F (3, 30)=25.64, p<0.0001). Saline and Mec groups were not significantly different in any phase of the task, including across Extinction (Figure 5B-C + Figure 5—figure supplement 1A–B, orange shading, main effect of Group (antagonist) in a two-way repeated-measures ANOVA, F (1, 15)=1.201, p=0.2903). We have shown that this dose of mecamylamine delivered i.p. has significant effects in tests of anxiety-like behavior and responses to inescapable stress. In addition, chronic treatment with mecamylamine at this dose is sufficient to decrease BLA c-fos immunoreactivity (Mineur et al., 2007). Consistent with the inability to acquire the cue-reward contingency, mice treated with Scop or Mec+Scop also obtained very few rewards during Extinction (Figure 5B + Figure 5—figure supplement 1A, orange shading). The antagonists had no effect on locomotion as measured by beam breaks (Figure 5—figure supplement 1C, one-way ANOVA, F (3, 30)=0.5074, p=0.6802).

Figure 5 with 1 supplement see all

Download asset Open asset

Acetylcholine is often thought of as a neuromodulator (Picciotto et al., 2012), and the window for cholinergic effects on synaptic plasticity varies across ACh receptor subtypes (Gu and Yakel, 2011). It is therefore possible that ACh signaling may result in intracellular signaling changes that outlast the cue presentation window. In order to determine if the effect of NBM-BLA stimulation is dependent upon the timing of correct nose poke and laser stimulation contingency, we repeated the experiment in an independent cohort of mice with an additional non-contingent ChR2 group that received the same number of stimulation trains as the contingent ChR2 group, but in which light stimulation was explicitly unpaired with task events (Figure 6A + Figure 6—figure supplement 1). As in the previous experiment, there were no differences between the EYFP control (n = 6) and stimulation groups (contingent-ChR2 n = 5 and non-contingent ChR2 n = 5) during Pre-Training (Figure 6B-C + Figure 6—figure supplement 2A–B, blue shading; main effect of group (EYFP versus contingent-ChR2 versus non-contingent ChR2) two-way repeated-measures ANOVAs; rewards earned: F (2, 13)=0.7008, p=0.5140; incorrect nose pokes: F (2, 13)=0.3906, p=0.6843). However, the non-contingent ChR2 group was not significantly different from the contingent ChR2 group during the training period with respect to number of rewards earned (two-way repeated-measures ANOVA, F (1, 8)=0.09147, p=0.7700) or incorrect nose pokes (two-way repeated-measures ANOVA, F (1, 8)=0.3681, p=0.5609), but both ChR2 groups were significantly better than the EYFP control group (Figure 6B-C + Figure 6—figure supplement 2A–B, pink shading; two-way repeated-measures ANOVAs; rewards earned: Group (EYFP versus contingent-ChR2 versus non-contingent-ChR2) main effect: F (2, 13)=7.254, p=0.0077; Day × Group interaction: F (22, 143)=1.861, p=0.0164. Incorrect nose pokes: Group main effect: F (2, 13)=4.884, p=0.0262). These results demonstrate that ChR2-mediated ACh release does not have to be time-locked to the cue, nose poke, or reward retrieval to improve performance of the task, suggesting that ACh may alter the threshold for neuronal plasticity for cue-reward pairing over a much longer timescale than might be expected based on results from the ACh3.0 recording and NBM-BLA recordings, which could be consistent with the involvement of mAChR signaling in this effect. As in the previous experiment, once all groups reached criterion for acquisition of the cue-reward contingency, there were no differences between any of the groups during Extinction (Figure 6B-C + Figure 6—figure supplement 2A–B, orange shaded; two-way repeated-measures ANOVA, F (2, 13)=0.04229, p=0.9587).

Figure 6 with 2 supplements see all

Download asset Open asset

We have shown that ACh release in the BLA is coincident with the stimulus that was most salient to the animal at each phase of the task. Use of the fluorescent ACh sensor was essential in determining these dynamics (Jing et al., 2018; Jing et al., 2019). Previous microdialysis studies have shown that ACh is released in response to positive, negative, or surprising stimuli, but this technique is limited by relatively long timescales (min) and cannot be used to determine when cholinergic transients align to given events in an appetitive learning task and how they evolve over time (Sarter and Lustig, 2020). In this cue-reward learning paradigm, when there was no consequence for incorrect nose-poking (Pre-Training phase), animals learned to perform a very high number of nose pokes and received a large number of rewards, and BLA ACh signaling peaked following correct nose pokes. Both the behavioral response (nose poking that was not contingent with the tone) and the ACh response (linked to the correct nose poke) suggest that the animals were not attending to the tone during most of the Pre-Training phase of the task, but rather were attending to the cues associated with reward delivery, such as the reward light or the sound of the pump that delivered the reward. Consistent with this possibility, in the next phase of the task when mice received a timeout for responding if the tone was not presented, performance of all groups dropped dramatically. Interestingly, in animals that had difficulty learning the cue-reward contingency, during early Training sessions ACh release shifted to reward retrieval, likely because this was the most salient aspect of the task when the majority of nose pokes performed did not result in reward. Finally, as mice acquired the contingency between tone and reward availability, the tone also began to elicit ACh release in the BLA, suggesting that mice learned that the tone is a salient event predicting reward availability. Since there are multiple sources of ACh input to the BLA, it was important to determine whether NBM cholinergic neurons were active during the periods when ACh levels were high (Woolf, 1991). Recordings from cholinergic NBM-BLA terminal fibers showed similar dynamics to ACh measurements, suggesting that the NBM is a primary source of ACh across the phases of cue-reward learning.

Perhaps the most well-known example of dynamic responding related to learning cue-reward contingencies and encoding of reward prediction errors is the firing of dopaminergic neurons of the ventral tegmental area (VTA; Schultz, 1998). After sufficient pairings, dopaminergic neurons will fire in response to the cue that predicts the reward, and no longer to the rewarding outcome, which corresponds with behavioral changes that indicate an association has been formed between conditioned stimuli (CS) and unconditioned stimuli (US). It should be noted that dopamine signaling is not unique in this learning-related dynamic response profile. Serotonergic neuronal responses also evolve during reward learning in a manner distinct from dopaminergic neurons (Zhong et al., 2017). Plasticity related to learning has also been observed in cholinergic neurons in the basal forebrain complex during aversive trace conditioning, such that after several training days, neuronal activity spans the delay between CS and US (Guo et al., 2019). Additionally, a recent study suggested that ACh may signal a valence-free reinforcement prediction error (Sturgill et al., 2020). Future studies on the selective inputs to NBM to BLA cholinergic neurons would be of interest to identify the links between brain areas involved in prediction error coding.

We found that BLA CaMKIIα cells were most reliably activated following reward retrieval before contingency acquisition (both when they were receiving several rewards but no timeouts in Pre-Training and few rewards early in Training). Similar to the recording of ACh levels, after acquisition, the tone began to elicit an increase in BLA CaMKIIα cell population activity. However, activity of CaMKIIα neurons differed from ACh signaling in the BLA in important ways. ACh was released in response to the salient events in the task that were best able to predict reward delivery or availability. By contrast, the activity of BLA CaMKIIα neurons was not tightly time-locked to correct nose poking, and instead followed reward retrieval until acquisition, when activity increased in response to tone onset. The divergent dynamics of ACh release and CaMKIIα neuron activity underscores that ACh’s role in the BLA is to modulate, rather than drive, the activity of CaMKIIα neurons, and therefore may alter dynamics of the network through selective engagement of different populations of GABA interneurons (Unal et al., 2015).

Neuronal activity and plasticity in the BLA is required for both acquisition of appetitive learning (conditioned reinforcement) and fear conditioning, however the inputs that increase activity in the structure during salient events likely come from many brain areas (McKernan and Shinnick-Gallagher, 1997; Rogan et al., 1997; Tye et al., 2008). In particular, dopaminergic inputs to the BLA are important for acquisition of conditioned reinforcement and for linking the rewarding properties of addictive drugs to cues that predict their availability (Cador et al., 1989). Our results indicate that ACh is a critical neuromodulator upstream of the BLA that is responsive to salient events, such as reward availability, motor actions that elicit reward, and cues that predict reward. We show here that increasing endogenous ACh signaling in the BLA caused mice to perform significantly better than controls in an appetitive cued-learning task. Heightened ACh release during learning of a cue-action-reward contingency led to fewer incorrect responses and increased acquisition rate in both female and male mice. The optical stimulation was triggered by correct nose poke, thus the cholinergic NBM-BLA terminal fiber stimulation overlapped with all three salient events: tone, nose poke, and reward retrieval, since the tone terminated 2 s after correct nose poke. We chose this stimulation pattern, as opposed to linking optical stimulation to tone onset, to ensure stimulation was dependent on behavioral responses. Therefore, stimulation did not precisely recapitulate the ACh release profile observed in mice that had already acquired the task (when ACh increases following tone onset). This suggests that behaviorally-contingent increases in BLA ACh are sufficient to enhance task acquisition (but see below). It is also possible that optogenetic-mediated ACh release may last longer than endogenous, tone-evoked release. A simultaneous stimulation and recording approach would be required to compare ACh release under both conditions (Pisansky et al., 2019). It is important to note that basal forebrain neurons have been demonstrated to co-release ACh and GABA (Ma et al., 2018; Saunders et al., 2015), and cholinergic basal forebrain neurons that project to the BLA have been shown to co-express a glutamate transporter (Ma et al., 2018; Nickerson Poulin et al., 2006). Thus, it is possible that both fiber photometry and optogenetic results could be influenced, in part, by co-release of other neurotransmitters from ChAT-positive neurons. Future studies employing additional fluorescent neurotransmitter sensors (Marvin et al., 2013; Marvin et al., 2018; Marvin et al., 2019) could help understand the interaction between the different signals employed by basal forebrain neurons.

It is possible that ACh improved learning by increasing the intensity of the reward, potentiating the learned association, improving discrimination, or a combination of these phenomena. However, increasing ACh release in the BLA was not inherently rewarding, because it did not support self-stimulation or real-time place preference. This is at odds with a recent study that found stimulation of NBM-BLA cholinergic terminals could induce a type of place-preference and modest self-stimulation (Aitta-aho et al., 2018). Perhaps slight differences in targeting of ChR2 infusion or differences in the behavioral paradigm could be responsible for the lack of direct rewarding effects of optical ChAT terminal stimulation in the current study. Other recent work (Jiang et al., 2016) has demonstrated that stimulating this NBM-BLA cholinergic pathway is sufficient to strengthen cued aversive memory, suggesting that the effect of ACh in the BLA may not be inherently rewarding or punishing but instead potentiates plasticity in the BLA, allowing learning of cue-outcome contingencies. Similarly, it is possible that ACh alters motor activity. However, there were no effects of optical stimulation on locomotion or responding in the inactive nose poke port. In addition, during the Pre-Training phase when there was no consequence for incorrect nose pokes, all groups earned the same number of rewards, regardless of optical stimulation or pharmacological blockade of ACh receptors, suggesting that ACh is not involved in the motor aspects of the task or the value of the reward. Indeed, differences emerged only during the Training phase, when attention to the tone was critical to earn rewards. Further, incorrect nose poking remained high for mice administered scopolamine. This suggests that scopolamine-treated animals were seeking the reward, as in the operant familiarization and Pre-Training phases of training, but were unable to learn that they should only nose poke in response to the tone.

Cell-type-specific expression of AChRs and activity-dependent effects place cholinergic signaling at a prime position to shape BLA activity during learning. For instance, late-firing interneurons in the BLA exhibit nAChR-dependent EPSP’s when no effect is seen on fast-spiking interneurons, while principal neurons can be either excited or inhibited through mAChRs, depending on activity level of the neuron at the time of cholinergic stimulation (Unal et al., 2015). BLA mAChRs can support persistent firing in principal neurons and can be important for the expression of conditioned place preference behavior, as well as trace fear conditioning (Baysinger et al., 2012; Egorov et al., 2006; McIntyre et al., 1998). Similar to studies of trace fear conditioning, in which activity of the network over a delay period must be maintained, we found that metabotropic (mAChRs) but not ionotropic (nAChRs) ACh receptors were required for learning the contingency of this cue-reward task. The timing of cholinergic signaling can be a critical factor in the induction of synaptic plasticity in other brain regions, so we hypothesized that the enhancement of cue-reward learning observed might be dependent upon when NBM-BLA terminal fibers were stimulated with respect to tone presentation and/or behavioral responses (Gu and Yakel, 2011). However, we found that heightened ACh signaling in the BLA improved behavioral performance even when stimulations were explicitly unpaired with the cue or correct nose poking. This suggests that the effect of increased cholinergic signaling in the BLA is long lasting, and that stimulation during a learning session is sufficient to potentiate synaptic events linking the cue to a salient outcome—even if CS and/or reward delivery are presented tens of seconds later. Given the findings from fiber photometry recordings, which showed endogenous ACh release was time-locked to salient stimuli during the task and evolved with learning, it is surprising that time-locking of exogenous ACh release was not necessary for enhancement of cue-reward learning. Coupled with pharmacological evidence demonstrating that muscarinic signaling is necessary for reward learning in this task, these results suggest the involvement of metabotropic signaling downstream of muscarinic receptors that outlasts the initial cholinergic stimulation.

To conclude, the abundant ACh input to the BLA results in ACh release in response to stimuli that predict reward in a learned cue-reward task. Increasing cholinergic signaling results in accelerated learning of the cue-reward contingency. These findings are consistent with the hypothesis that ACh is a neuromodulator that is released in response to salient stimuli and suggests that ACh signaling may enhance neuronal plasticity in the BLA network, leading to accelerated cue-reward learning.

All procedures were approved by the Yale University Institutional Animal Care and Use Committee (protocol: 2019–07895) in compliance with the National Institute of Health’s Guide for the Care and Use of Laboratory Animals. Experiments were performed in mice of both sexes, in keeping with the NIH policy of including sex as a biological variable. Sex of mice in behavioral graphs is indicated by circles for females and squares for males.

Female and male heterozygous mice with Cre recombinase knocked into the choline acetyltransferase (ChAT) gene (ChAT-IRES-Cre, B6;129S6-Chat^{tm2(cre)Lowl/J}, Jackson Laboratory, Bar Harbor, ME) were bred in house by mating ChAT-IRES-Cre with C57BL6/J mice. CaMKIIα-Cre (Tg(Camk2a-cre)2Gsc) mice obtained from Ronald Duman (Casanova et al., 2001; Wohleb et al., 2016) were bred in house as above. C57BL6/J mice were obtained from The Jackson Laboratory at 6–10 weeks of age, and tested at 5–7 months of age, following at least 1 week of acclimation. All mice were maintained in a temperature-controlled animal facility on a 12 hr light/dark cycle (lights on at 7:00 AM). Mice were group housed 3–5 per cage and provided with ad libitum food and water until undergoing behavioral testing. Mice were single housed 1–3 weeks before surgery to facilitate food restriction and body weight maintenance.

Surgical procedures for behavior were performed in fully adult mice at 4–6 months of age, age-matched across conditions. For viral infusion and fiber implantation, mice were anesthetized using isoflurane (induced at 4%, maintained at 1.5–2%) and secured in a stereotactic apparatus (David Kopf Instruments, Tujunga, CA). The skull was exposed using a scalpel and Bregma was determined using the syringe needle tip (2 µL Hamilton Neuros syringe, 30 gauge needle, flat tip; Reno, NV).

For fiber photometry surgeries, 0.4 µL of AAV9 hSyn-ACh3.0 (Vigene Biosciences Inc) to measure BLA ACh levels (Figure 2A-E + Figure 2—figure supplements 1–2) was delivered unilaterally to the BLA (A/P; −1.34 mm, M/L + or - 2.65 mm, D/V −4.6 mm, relative to Bregma) of ChAT-IRES-Cre or wild-type C57BL6/J mice at a rate of 0.1 µL/min. The needle was allowed to remain at the infusion site for 5 min before and 5 min after injection. A mono fiber-optic cannula (1.25 mm outer diameter metal ferrule; 6 mm long, 400 µm core diameter/430 µm outer diameter, 0.48 numerical aperture (NA), hard polymer cladding outer layer cannula; Doric Lenses, Quebec City, Quebec, Canada) was implanted above the BLA (A/P; −1.34 mm, M/L + 2.65 mm, D/V −4.25 mm) and affixed to the skull using opaque dental cement (Parkell Inc, Edgewood, NY). For BLA CaMKIIα cell calcium dynamic recordings (Figure 3 + Figure 3—figure supplements 1–2), 0.5 µL of AAV1 Syn-FLEX-GCaMP6s-WPRE-SV40 (Addgene, Watertown, MA) was injected into the left BLA using the same procedure and coordinates but was injected into CaMKIIα-Cre mice. Cholinergic NBM-BLA terminal fiber calcium dynamic recording (Figure 2F-J + Figure 2—figure supplements 5–8) surgeries were performed as above except AAV1-Syn-FLEX-jGCaMP7s-WPRE (Addgene) was infused unilaterally into the NBM (A/P: - 0.7 mm, M/L + or - 1.75 mm, D/V – 4.5 mm) of ChAT-IRES-Cre mice, with the optical fiber being placed above the ipsilateral BLA. The jRCaMP1b + ACh3.0 surgeries to simultaneously measure cholinergic NBM-BLA terminal fiber calcium dynamics and BLA ACh levels (Figure 2—figure supplement 9) consisted of both the NBM and BLA infusions above, except AAV1 Syn-FLEX-NES-jRCaMP1b-WPRE-SV40 (Addgene) was infused the NBM of ChAT-IRES-Cre mice. The RCaMP sham mouse (Figure 2—figure supplement 9E,H) was a wild-type littermate and thus had no jRCaMP1b expression. pAAV.Syn.Flex.GCaMP6s.WPRE.SV40 (Addgene viral prep # 100845-AAV1; http://n2t.net/addgene:100845; RRID:Addgene_100845), pGP-AAV-syn-FLEX-jGCaMP7s-WPRE was a gift from Douglas Kim and GENIE Project (Addgene viral prep # 104491-AAV1; http://n2t.net/addgene:104491; RRID:Addgene_104491), and pAAV.Syn.Flex.NES-jRCaMP1b.WPRE.SV40 (Addgene viral prep # 100850-AAV1; http://n2t.net/addgene:100850; RRID:Addgene_100850) were gifts from Douglas Kim and GENIE Project (Chen et al., 2013; Dana et al., 2016; Dana et al., 2019).

Mice were allowed to recover in a cage without bedding with a microwavable heating pad underneath it until recovery before being returned to home cage. For 2 d following surgery, mice received 5 mg/Kg Rimadyl i.p (Zoetis Inc, Kalamazoo, MI) as postoperative care.

For optical stimulation experiments (Figure 4, Figure 6 + Figure 4—figure supplements 1–5 + Figure 6—figure supplements 1–2), surgeries were performed as above except as follows: 0.5 µL of control vector (AAV2 EF1a-DIO-EYFP) or channelrhodopsin (AAV2 EF1a-DIO-hChR2(H134R)-EYFP; University of North Carolina Gene Therapy Center Vector Core, Chapel Hill, NC) was delivered bilaterally into the NBM (A/P: - 0.7 mm, M/L ± 1.75 mm, D/V – 4.5 mm) of ChAT-IRES-Cre mice. Mono fiber-optic cannulas (1.25 mm outer diameter zirconia ferrule; 5 mm long, 200 µm core diameter/245 µm outer diameter, 0.37 NA, polyimide buffer outer layer cannula; Doric Lenses) were inserted bilaterally above the basolateral amygdala (BLA, A/P; −1.22 mm, M/L ± 2.75 mm, D/V −4.25 mm). Mice were randomly assigned to EYFP or ChR2 groups, controlling for average group age.

For ex vivo local field potential electrophysiology experiments (Figure 4B), the NBM was injected with DIO-ChR2-EYFP as described above, except mice were 8 weeks of age (see Supplemental Methods for current clamp recording methods). The coronal brain slices containing the NBM were prepared after 2–4 weeks of expression. Briefly, mice were anesthetized with 1× Fatal-Plus (Vortech Pharmaceuticals, Dearborn, MI) and were perfused through their circulatory systems to cool down the brain with an ice-cold (4°C) and oxygenated cutting solution containing (mM): sucrose 220, KCl 2.5, NaH₂PO₄ 1.23, NaHCO₃ 26, CaCl₂ 1, MgCl₂ 6 and glucose 10 (pH 7.3 with NaOH). Mice were then decapitated with a guillotine immediately; the brain was removed and immersed in the ice-cold (4°C) and oxygenated cutting solution to trim to a small tissue block containing the NBM. Coronal slices (300 μm thick) were prepared with a Leica vibratome (Leica Biosystems Inc, Buffalo Grove, IL) after the tissue block was glued on the vibratome stage with Loctite 404 instant adhesive (Henkel Adhesive Technologies, Düsseldorf, Germany). After preparation, slices were maintained at room temperature (23–25°C) in the storage chamber in the artificial cerebrospinal fluid (ACSF; bubbled with 5% CO₂% and 95% O₂) containing (in mM): NaCl 124, KCl 3, CaCl₂ 2, MgCl₂ 2, NaH₂PO₄ 1.23, NaHCO₃ 26, glucose 10 (pH 7.4 with NaOH) for recovery and storage. Slices were transferred to the recording chamber and constantly perfused with ACSF with a perfusion rate of 2 mL/min at a temperature of 33°C for electrophysiological experiments. Cell-attached extracellular recording of action potentials was performed by attaching a glass micropipette filled with ACSF on EYFP-expressing cholinergic neurons with an input resistance of 10–20 MΩ under current clamp. Blue light (488 nm) pulse (60 ms) was applied to the recorded cells through an Olympus BX51WI microscope (Olympus, Waltham, MA) under the control of the Sutter filter wheel shutter controller (Lambda 10–2, Sutter Instrument, Novato, CA). All data were sampled at 3–10 kHz, filtered at 3 kHz and analyzed with an Apple Macintosh computer using Axograph X (AxoGraph). Events of field action potentials were detected and analyzed with an algorithm in Axograph X as reported previously (Rao et al., 2008).

Pre-processing of raw data was performed using a modified version of a MATLAB (MathWorks, Natick, MA) script provided by Doric. The baseline fluorescence (F0) was calculated using a first-order least mean squares regression over the ~30 min recording session. Second-order least mean squares regressions were used when photobleaching of the sensor was more pronounced, as in the case of NBM-BLA terminal fiber recordings. The change in fluorescence for a given time point (ΔF) was calculated as the difference between it and F0, divided by F0, which was multiplied by 100 to yield %ΔF/F0. The %ΔF/F0 was calculated independently for both the signal (465 nm) and reference (405 nm) channels to assess the degree of movement artifact. Since little movement artifact was observed in the recordings (Figure 2—figure supplement 1B–C, Figure 2—figure supplement 5D–E, Figure 3—figure supplement 1C–D, tan lines), the signal %ΔF/F0 was analyzed alone (the code provided allows for running the entire analysis pipeline with the reference channel %ΔF/F0 if desired). The %ΔF/F0 was z-scored to give the final Z%ΔF/F0 reported here. For the BLA CaMKIIα cell recordings (Figure 3—figure supplement 1C–D), the reference channel displayed some mirroring (moving in the opposite direction) compared to the signal. This is likely because 405 nm is not the ‘true’ isosbestic point for GCaMP and we were instead measuring some changes in calcium-unbound GCaMP rather than calcium-insensitive GCaMP signal alone (Barnett et al., 2017; Kim et al., 2016a; Sych et al., 2019). Graphs and heatmaps for averaged traces aligned to actions were based on licking bout epoch filtering code from TDT (Alachua, FL; link in code comments).

Combined action heatmaps were generated in MATLAB (2020a) by analyzing data 5 s preceding tone onset (rewarded trials only) to 5 s after receptacle entry. Actions were aligned despite variable latencies by evenly splitting a maximum of 4 s post-tone onset/pre-correct nose poke and 1 s post-correct nose poke/pre-receptacle entry for each trial within a day. The resulting aligned trials were averaged to generate daily averages that made up the rows of the individual animal heatmaps. Blanks in the rows of heatmaps (black time bins) indicate time bins added for alignment, meaning that no trials for that day had a latency that stretched the entire window. Only rewarded trials where the mouse entered the receptacle within 5 s after nose poke were analyzed. Full or partial training days were excluded from analysis if there were acquisition issues such as the patch cord losing contact with the fiber or behavioral apparatus malfunction. Lack of trials (fewer than three) for analysis or recording issues led to missing rows of fiber photometry data in the heatmap despite having behavioral data, in which case these rows were skipped rather than adding entire blank rows. Due to individual differences in behavior, across-mouse average data was calculated by using a selection of days in which behavior was roughly similar or milestones such as the first and last day of Pre-Training, first day earning 10 rewards in Training, first day crossing acquisition threshold (and maintaining afterward), last day of Training, last day of Extinction (with three or more rewarded trials that met analysis criteria). Additional days were included in across-mouse average heatmaps when possible. Incorrect nose poke heatmaps were generated by averaging signals for 5 s before and 5 s after incorrect nose pokes that were not preceded by an incorrect nose poke in the last 5 s. The incorrect nose poke heatmaps averaged across mice were generated using the same selection of days as the combined action heatmaps for a given experiment.

Bootstrapped confidence intervals (bCIs) of the Z-scored % ΔF/F0 fiber photometry data within and across mice were generated using the methods described in Jean-Richard-Dit-Bressel et al., 2020 for the following events: tone onset, correct nose poke, receptacle entry, and incorrect nose poke. For the within-mouse analysis by day, trials were aligned to event onset, and bCIs were generated for events that had at least 3 trials from 5 s before to 10 s after each event. Each series of data were bootstrapped 10,000 times and a two-sided 99% confidence interval was constructed. Data were considered significantly different from baseline (Z% ΔF/F0 = 0) when their 99% bCIs did not contain zero for an interval of time designated by a consecutive threshold of 0.5 s.

To avoid comparing vastly different numbers of trials, in graphs where correct and incorrect nose pokes were plotted together, incorrect nose pokes were downsampled to match the number of correct nose poke trials. For Incorrect Nose Pokes graphs where last Pre-Training Day and Training Day 1 were plotted together, both days were downsampled to the number of correct nose pokes on the last Pre-Training Day.

For the combined action bCI plots (tone onset, correct nose poke, and receptacle entry), the selection of days for each mouse matched that of the cohort-averaged combined action heatmaps. The three-event plots were combined by cropping to match the maximum latencies used in the combined action heatmaps. For the across-mouse averaged bCI plots, analyses were carried out as above except the bootstrapping used mouse trial averages. The mean lines for across-mouse averaged bCI plots were calculated by taking the mean of all individual trials together. The NBM-BLA cholinergic terminal fiber experiment required combining the two independent cohorts to obtain n ≥ 3. For the incorrect nose poke bCI plots, the number of trials used for each day was downsampled to 20 if a mouse performed more than 20.

Male wild-type C57BL/6J mice were injected i.p. 30 min before each Pre-Training and Training session with a volume of 10 mL/kg with the following compounds: 1× DPBS (Thermo Fisher Scientific, Waltham, MA), 1 mg/kg mecamylamine hydrochloride (Millipore Sigma, St. Louis, MO), 0.5 mg/kg (-) scopolamine hydrochloride (Millipore Sigma), or 1 mg/kg mecamylamine + 0.5 mg/kg scopolamine (Figure 5 + Figure 5—figure supplement 1).

After completion of behavioral experiments, animals were anesthetized with 1× Fatal-Plus (Vortech Pharmaceuticals). Once there was no response to toe-pinch, mice were transcardially perfused with 20 mL ice-cold 1× DPBS followed by 20 mL 4% paraformaldehyde (PFA, Electron Microscopy Sciences, Hatfield, PA). Brains were extracted and post-fixed for at least 1 d in 4% PFA at 4°C and transferred to 30% sucrose (Millipore Sigma) for at least 1 dy at 4°C. Brains were sliced 40 µm thick on a self-cooling microtome and stored in a 0.02% sodium azide (Millipore Sigma) PBS solution. Brain slices were washed in PBS, blocked for 2–3 hr (0.3% Triton X-100, American Bioanalytical, Canton, MA; 3% normal donkey serum, Jackson ImmunoResearch, West Grove, PA), then incubated overnight with primary antibodies (1:1000 + 1% normal donkey serum). Slices were then washed in PBS and incubated with secondary antibodies (1:1000) for 2 hr, washed, stained with DAPI for 5 min, washed, mounted, and coverslipped with Fluoromount-G (Electron Microscopy Sciences). All incubations were at room temperature. Microscope slides were imaged using a FLUOVIEW FV10i confocal microscope (Olympus). Injection sites and fiber placements were designated on modified Allen Mouse Brain Atlas figures (Lein et al., 2007). Mice were excluded from analyses if fluorescence was not observed at injection sites or if fiber tips were not identified at the intended site.

Operant behavioral data saved by MEDPC IV was transferred to Excel using MPC2XL. Data were organized in MATLAB and analyzed in Prism (V8.3.0, GraphPad Software, San Diego, CA). Differences between groups and interactions across days for Training were evaluated using Two-Way Repeated Measures ANOVAs. We computed the required sample size for a 90% power level with an alpha of 0.05 by estimating the control (EYFP) group mean would be 10 rewards and the mean experimental (ChR2) group would be 20 rewards with a standard deviation of 5. We utilized a power calculator for continuous outcomes of two independent samples, assuming a normal distribution. The result was six samples per group. Each manipulation experiment started with at least six mice included in each group (Sealed Envelope, 2020). In each experiment, each animal within a group served as a biological replicate. These studies did not include technical replicates. Masking was not applied during data acquisition but data analyses were semi-automated in MATLAB and performed blind to condition.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

These studies were supported by grants DA14241, DA037566, MH077681. LW, DT, and PR were supported by NS022061, MH109104 from the National Institutes of Health, and by the intramural programs of NINDS and NIMH. X-BG was supported by DA046160. RBC was supported by T32-NS007224. This work was funded in part by the State of Connecticut, Department of Mental Health and Addiction Services but this publication does not express the views of the Department of Mental Health and Addiction Services or the State of Connecticut. The views and opinions expressed are those of the authors. We thank Samantha Sheppard for the use of her mouse illustration and animal care assistance and Nadia Jordan-Spasov for genotyping and laboratory help. Li Jiang performed the ex vivo current-clamp recordings. Angela Lee and Wenliang Zhou provided helpful input into experimental planning. Colin Bond, Marcelo Dietrich, Usman Farooq, Onur Iyilikci, Sharif Kronemer, Matthew Pettus, and Zach Saltzman provided insightful discussion and assistance with analysis and figure design. Ralph DiLeone, Stephanie Groman, Hyojung Seo, and Jane Taylor offered helpful discussion about experimental design and analysis. The support teams at Doric Lenses (Alex Côté and Olivier Dupont-Therrien) and Tucker-Davis Technologies provided discussion, analysis support, and MATLAB code assistance.

Animal experimentation: All procedures were approved by the Yale University Institutional Animal Care & Use Committee in compliance with the National Institute of Health's Guide for the Care and Use of Laboratory Animals. (protocol: 2019-07895).

1. Received: March 27, 2020
2. Accepted: September 17, 2020
3. Accepted Manuscript published: September 18, 2020 (version 1)
4. Accepted Manuscript updated: September 22, 2020 (version 2)
5. Version of Record published: October 1, 2020 (version 3)

This is an open-access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.