Distinct representation of cue-outcome association by D1 and D2 neurons in the ventral striatum’s olfactory tubercle

  1. Nuné Martiros
  2. Vikrant Kapoor
  3. Spencer E Kim
  4. Venkatesh N Murthy  Is a corresponding author
  1. Department of Molecular & Cellular Biology and Center for Brain Science, Harvard University, United States

Abstract

Positive and negative associations acquired through olfactory experience are thought to be especially strong and long-lasting. The conserved direct olfactory sensory input to the ventral striatal olfactory tubercle (OT) and its convergence with dense dopaminergic input to the OT could underlie this privileged form of associative memory, but how this process occurs is not well understood. We imaged the activity of the two canonical types of striatal neurons, expressing D1- or D2-type dopamine receptors, in the OT at cellular resolution while mice learned odor-outcome associations ranging from aversive to rewarding. D1 and D2 neurons both responded to rewarding and aversive odors. D1 neurons in the OT robustly and bidirectionally represented odor valence, responding similarly to odors predicting similar outcomes regardless of odor identity. This valence representation persisted even in the absence of a licking response to the odors and in the absence of the outcomes, indicating a true transformation of odor sensory information by D1 OT neurons. In contrast, D2 neuronal representation of the odor-outcome associations was weaker, contingent on a licking response by the mouse, and D2 neurons were more selective for odor identity than valence. Stimulus valence coding in the OT was modality-sensitive, with separate sets of D1 neurons responding to odors and sounds predicting the same outcomes, suggesting that integration of multimodal valence information happens downstream of the OT. Our results point to distinct representation of identity and valence of odor stimuli by D1 and D2 neurons in the OT.

Editor's evaluation

Martiros et al. monitored the activity of dopamine D1 and D2-expressing neurons within the ventral striatum olfactory tubercle using 2-photon microscopy, as mice learned to associate odors and tones with either positive or aversive outcomes. Authors find differential roles for these neurons in the learning of odor valence and tone outcome associations. Overall, this study was deemed as an interesting and important contribution to our understanding of the neural basis of cue encoding.

https://doi.org/10.7554/eLife.75463.sa0

Introduction

Assigning value to stimuli in the external environment and subsequently adjusting behavior on the basis of those learned values is a primary function of the nervous system. Understanding this process is especially important because certain associations result in maladaptive behaviors such as compulsions, binge eating, and drug addiction (Everitt and Robbins, 2005; Keiflin and Janak, 2015; Wise and Koob, 2014). Olfaction or chemosensation is thought to have preceded other sensory modalities in evolution as the first direct link of the nervous system to the external environment (Ache and Young, 2005; Kaas, 2008; Schneider, 2013) to allow for such stimulus value learning. Olfactory sensory pathways to the forebrain remained relatively conserved across phyla while visual, auditory, and somatosensory information was routed in polysynaptic pathways through thalamus and cortex (Purves and Fitzpatrick, 2001; Schneider, 2013). Due to this, the olfactory system has uniquely direct access to limbic centers of the brain. In particular, the olfactory tubercle (OT), a ventral basal ganglia structure known to be involved in reward processing (Gadziola et al., 2015; Hagamen et al., 1977; Ikemoto, 2003; Ikemoto, 2007; Wesson and Wilson, 2011; Zhang et al., 2017), is a direct recipient of a stream of olfactory input from a distinct class of olfactory bulb neurons, the tufted cells (Igarashi et al., 2012). In addition to olfactory sensory input, the OT receives dense dopaminergic input from the ventral tegmental area (VTA), similar to other ventral striatal regions – suggesting that the OT is ideally suited to integrate these inputs to assign valence to olfactory stimuli based on experience. It has been described in humans that olfactory sensory cues may be more powerfully linked to emotional memories than other types of sensory cues (Miles and Berntsen, 2011; Pointer and Bond, 1998; Reid et al., 2015), and that odor associations can be important in psychological health and disease (Daniels and Vermetten, 2016; Herz, 2016), suggesting a possible specialized form of olfactory memory in limbic regions such as the OT.

The OT has been implicated in motivation as well as olfactory processing (Hagamen et al., 1977; Ikemoto, 2007; Koob et al., 1978; Koob and Volkow, 2010; Wesson and Wilson, 2011; Wright and Wesson, 2021). As an example, self-administration of cocaine into the OT was found to be even more reinforcing than administration into nucleus accumbens, a key region known to be involved in reward, motivation, and addiction (Ikemoto, 2003). Stimulation of OT neurons or dopaminergic terminals in the OT has been shown to be effective in inducing odor preference or approach (Fitzgerald et al., 2014; Gadziola et al., 2020; Zhang et al., 2017). Electrophysiological recordings in the OT have revealed that OT neurons differentiate between rewarded and unrewarded odors in a go/no-go task, and quickly track changing odor-outcome contingencies (Gadziola et al., 2015; Millman and Murthy, 2020). These effects may extend to humans with elevated OT activity and development of place preference in response to attractive odorants (Midroit et al., 2021). Moreover, the OT appears to be the olfactory processing site most strongly involved in tracking odor-outcome associations, as direct comparison to recordings in posterior piriform cortex revealed weaker representation of odor-reward contingency (Gadziola et al., 2020; Millman and Murthy, 2020). The much higher density of dopaminergic input to the OT as compared to piriform cortex is likely relevant in differentiating the functions of these two parallel olfactory processing regions.

This evidence establishes OT as a key region likely involved in learning odor-outcome associations and assigning emotional tags to odors. However, many questions remain unanswered regarding the nature of the information encoded by OT neurons and the role of the different neuronal types within the OT in this function. OT is a part of the ventral striatum and consists of spiny projection neurons (SPNs). Striatal SPNs are roughly divided into D1-type and D2-type dopamine receptor expressing SPNs, and these two groups of SPNs also have different output projection patterns (Bolam et al., 2000; Gerfen and Surmeier, 2011). This differentiation of OT projection neurons into D1 and D2 type is especially relevant when considering the role of dopaminergic input in shaping odor valence representation. D1- and D2-type dopamine receptors are thought to differ in terms of their response to dopamine, in particular the plasticity rules regulated by dopamine (Gerfen and Surmeier, 2011; Lovinger, 2010; Nicola et al., 2000). To address the role of these neuronal subtypes in the OT, we conducted the first (to our knowledge) two-photon imaging of specific neuronal types in the OT in behaving mice. We then used experimental manipulations to address questions about the role of the OT and these neuronal types in the stimulus identity-valence-response transformation function.

Results

To investigate the role of D1 and D2 receptor expressing neurons in the OT in odor valence learning, we imaged the activity of these neuronal populations in the OT using multiphoton microscopy, while mice learned to associate odors with aversive or rewarding outcomes. Adenoassociated virus that allows conditional expression of the calcium indicator GCaMP7s in the presence of Cre recombinase was injected into the right OT of Drd1-Cre and Adora2A-Cre mice and a 1 mm cannula was implanted targeting the OT (Figure 1A, Figure 1—figure supplement 1 ). After recovery, a Gradient-Index (GRIN) lens was placed in the cannula, mice were water restricted, habituated to head fixation, and trained on an odor-outcome conditioning task. Five monomolecular odors were coupled with graded outcomes ranging from aversive to rewarding (strong airpuff to nose, weak airpuff to nose, no outcome, small water drop, large water drop) (Figure 1B). Odor-outcome assignments were counterbalanced across mice and the five trial types were randomly interspersed across training sessions with 30 trials per day of each of the five odors. In each trial, the odor was presented for 1.5 s and the outcome (water or airpuff) was presented 1.3 s after odor onset. Prior to the first day of odor-outcome training, mice underwent a pre-training session (day 0) in which the airpuff and water outcomes were presented without any odors. In day 1 of odor-outcome training, mice quickly learned the odor-outcome contingencies and began licking in anticipation of water delivery in the period after odor onset prior to water delivery (Figure 1C). By training day 2, anticipatory licking was observed in implanted Drd1-Cre and Adora2A-Cre mice in response to rewarded odors 4 and 5, and little anticipatory licking was observed for non-rewarded odors 1–3 Figure 1D, Figure 1—figure supplement 2. The number of anticipatory licks was greater for the odor that predicted the larger reward, compared to that predicting smaller reward (2.7±0.35 vs. 2.0±0.24 on day 4 of training, p<0.0001 rank sum test, 30 trials each odor in six mice; Figure 1—figure supplement 3). This suggests that the mice can perceive graded reward outcomes in addition to categorizing odors as rewarding or aversive.

Figure 1 with 5 supplements see all
Two-photon calcium imaging of D1- and D2-type neurons in the olfactory tubercle (OT) during odor-outcome association learning.

(A) Cannula and Gradient-Index (GRIN) lens implantation targeting the OT in Drd1-Cre and Adora2A-Cre mice with AAV9-Syn-FLEX-GCaMP7s virus injection in the OT. (B) Odor-outcome task structure, odors 1–2 are paired with aversive airpuffs and odors 4–5 are paired with rewarding water drops in headfixed water restricted mice. Odor-outcome assignments are counterbalanced across mice. (C) Number of anticipatory licks in an example mouse in a 1 s period after odor onset and prior to water or airpuff delivery. Each training day has 30 trials of each of the five odors. (D) Mean number of anticipatory licks across 4 days of training in implanted Drd1-Cre mice (n=6, solid) and Adora2A-Cre mice (n=6, dashed). (E) Field of view of GCaMP7s expressing neurons in a Drd1-Cre mouse. (F) Example imaged neuron with activity in individual odor 5 trials in a single session. Dashed lines indicate odor onset and water onset. (G) Neuronal activity in field of view of a Drd1-Cre mouse in big airpuff and big water drop trials across days of training. Small gray dots indicate non-significantly responsive neurons. Day 0 indicates pre-training day in which no odors are presented, only water drops and airpuffs. Activity is shown in the 1 s period prior to outcome delivery, after odor onset in days 1 and 4.

OT neuronal activity robustly and bidirectionally reflects odor-outcome contingency

D1 neurons in six Drd1-Cre mice (88±13 neurons per mouse, 529 neurons in total) and D2 neurons in six Adora2A-Cre mice (56±18 neurons per mouse, 338 neurons in total) developed both excitatory and inhibitory responses to the task odors across days of training (Figure 1E–G), with the most rewarded odor typically eliciting the strongest neuronal responses after training. Within the OT, there are neurons in the islands of Calleja, which express Drd1 (but not A2A). We think the inclusion of neurons from the islands of Calleja to be rare because of their characteristically small size (~8 μm diameter) and dense clustering – neither of which appear to be the case in our imaged fields of view (Figure 1—figure supplement 4). The anticipatory licking of the mice without a GRIN lens implant is similar to that of the implanted mice (Figure 1—figure supplement 5), suggesting that the implanted mice are not impaired in their ability to learn the stimulus-outcome associations.

Task-related activity of the population of D1 neurons we recorded was strikingly similar during the two aversive trial types (Figure 2A, columns 1–2) and similar during the two rewarded trial types (Figure 2A, columns 4–5) although the identity of the odorants presented during these trials varied across mice (Figure 2—figure supplement 1). Overall, neurons followed similar activity patterns during the two rewarded trial types and during the two aversive trial types (Figure 2B–C). Correlation between the activities of groups of stimulus responsive neurons in each mouse, measured by cosine similarity, was higher for same outcome trials than for opposite outcome trials after the first training day (Figure 2D). D1 neurons were more likely to be activated by rewarding odors than aversive odors (Figure 2E, 38% activated by odor 5 vs. 18% activated by odor 1 on day 4, p<0.00001, Fisher’s exact test). The proportion inhibited by the same stimuli on day 4 was 18% and 13%, respectively, and not significantly different (p=0.7, Fisher’s exact test). Contrary to our expectation, we found that D2 neurons were also more likely to be activated by rewarding stimuli than aversive stimuli (Figure 2E, 36% activated by odor 5 vs. 21% activated by odor 1 on day 4, p=0.003, Fisher’s exact test; Figure 2—figure supplement 2). There was no significant difference between the proportion of D1 and D2 neurons activated in response to the most rewarding odor 5, on day 4 (38% and 36%, respectively, p=0.74, Fisher’s exact test).

Figure 2 with 2 supplements see all
D1-type neurons in the olfactory tubercle (OT) respond most strongly to rewarded odors and respond similarly to odors of similar outcomes.

(A) Activity of all activated (top) and inhibited (bottom) neurons from six Drd1-Cre mice in five trial types on day 4 of training. Vertical black lines indicate odor onset and outcome delivery time. Vertical dashed lines in the two right-most panels indicate average time of first lick by mice. Neurons are grouped by preferred stimulus, arrows on right indicate boundaries between groups. Neuronal activity in the two aversive trial types (columns 1–2) is similar and neuronal activity in the two rewarding trial types (columns 4–5) is similar. (B) Mean population activity of all activated and inhibited D1 neurons in five trial types. Odor onset at 1 s and outcome onset at 2.3 s. (C) 3D D1 neural population trajectories in five trial types. (D) Cosine similarity between groups of neurons responding to odor pairs of same valence or opposite valence (five neuronal subgroups as in A with pairwise comparisons, *p<0.05 rank sum test). (E) Proportions of significantly activated and inhibited D1 and D2 neurons on days 1–4 of training (day 4: n=529 D1 neurons, n=338 D2 neurons).

Individual D1 neurons are more likely to encode odor valence and D2 neurons more likely to encode odor identity

By using graded odor-outcome associations, we aimed to disambiguate neuronal coding for odor valence, odor motivational salience, and odor identity. We hypothesized that, in our task, idealized valence coding neurons would exhibit similar responses to odors of similar valence (Figure 3A, rows 1–2), salience coding neurons would exhibit similar responses to odors of high relevance regardless of whether the outcome positive or negative (Figure 3A, row 3), and odor identity responsive neurons would be most likely to respond to a single individual odor (Figure 3A, last row). In our dataset, the prevalence of neuronal responses consistent with odor salience coding was very low (3 of 529 D1 neurons were activated for high motivational salience odors 1 and 5, but not odors 2–4, as compared to 51 neurons activated for positive valence odors 4 and 5, but not odors 1–3, p<0.000001 Fisher’s test for D1 and D2 neurons). We formulated a metric to quantify salience (see Materials and methods, salience score) and found that responses consistent with salience coding were extremely rare (Figure 3—figure supplement 1). These data indicate that OT neuronal populations are very sensitive to the sign of the odor valence (aversive or rewarding). Thus, we focused the remainder of our analysis on odor valence or odor identity coding by D1 and D2 neurons.

Figure 3 with 1 supplement see all
Odor valence and identity coding in individual D1 and D2 olfactory tubercle (OT) neurons develops with learning.

(A) Hypothetical responses to the five task odors in idealized odor valence coding neurons, salience coding neurons, or odor identity coding neurons. (B) Calculation of valence scores for individual neurons by comparison of responses to odors of opposite valence to odors of similar valence. (C) Examples of four individual neurons with high odor identity coding scores and low valence coding scores. (D) Examples of four individual neurons with high valence coding scores and low identity coding scores. (E) Distributions of odor identity coding scores in D1 (blue, left) and D2 (pink, middle) neurons across 4 days of training. D1 neuron identity scores decrease and D2 neuron identity coding scores increase with training days. Right, day 4 distributions of identity scores of D1 and D2 neurons (**p<0.001, rank sum test). (F) Distributions of odor valence coding scores in D1 (blue, left) and D2 (pink, middle) neurons across 4 days of training. D1 neuron valence scores increase with training days. Right, day 4 distributions of valence scores of D1 and D2 neurons (***p<0.000001, rank sum test). Dotted lines show shuffled (control) distributions where the the odor-valence pairing has been randomized.

We assigned each significantly responsive neuron a valence score by computing a normalized difference between the neuron’s activity in response to odors of opposite valence and odors of same valence (Figure 3B, see Materials and methods). Identity scores were also calculated for each neuron by comparing the activity of the neuron to its most preferred stimulus to its next most preferred stimulus (see Materials and methods). Individual neurons varied widely in valence and identity scores, with some neurons having high identity scores and low valence scores (Figure 3C) and other neurons having high valence scores and low identity scores (Figure 3D). Interestingly, identity scores of D1 neurons decreased with training (p<0.05 for day 1 vs. 4, rank sum test), while identity scores of D2 neurons trended to increase with training (Figure 3E, left and middle). After training, identity scores of D2 neurons were significantly higher than those of D1 neurons (Figure 3E, right, p<0.001 rank sum test), with more D2 neurons exhibiting clear isolated activity in response to an individual odor. In contrast, odor valence coding scores of D1 neurons significantly increased with training (p<0.0001 for day 1 vs. 4, rank sum test) while those of D2 neurons did not (Figure 3F, left and middle). After training, odor valence coding scores of D1 neurons were significantly higher than those of D2 neurons (Figure 3F, right, p<0.000001 rank sum test), and higher than that of null distributions for valence scores calculated by shuffling odor-valence assignments (dashed lines in Figure 3F). A shuffling procedure was not appropriate for identity scores, since they were calculated by comparing the best odor response to the next best one without regard to the associated outcome.

D1 OT neurons encode odor valence in the absence of licking response or outcomes

In the odor-outcome association task, both odors of positive valence were accompanied by licking responses and water drop delivery, and both odors of negative valence were accompanied by airpuff delivery. As such, these trial types shared licking responses and unconditioned stimuli that likely contributed to the shared outcome-related neuronal responses we observed. In order to test whether activity of OT neurons reflected true valence coding of the odor stimulus, we constructed a probe session with two distinct training blocks. Block 1 of the session proceeded as previously described. After this first block, a 2 min break was introduced in which mice received 2 mL of free water droplets and became sated on water. Block 2 of the imaging was then conducted where mice did not lick in response to odor presentations and all outcomes (water and airpuff) were omitted (Figure 4A). We found that mice very rarely exhibited anticipatory licking in response to the rewarded odors in Block 2 (Figure 4B). Mice licked an average of 4.8±0.1 times in odor 5 trials in Block 1 and an average of 0.2±0.06 times in odor 5 trials in Block 2 (p<0.000001, rank sum test). In the following imaging data analysis, any Block 2 trials in which mice licked (5% of trials, or ~1 odor 5 trial per session) were excluded.

Figure 4 with 3 supplements see all
D1 neuronal valence coding is robustly preserved in the absence of licking response and absence of outcomes.

(A) Probe session structure. In Block 2 mice are sated with water, rarely exhibiting licking, and all outcomes are omitted. (B) Number of anticipatory licks in response to five-odor types in Blocks 1 and 2 (n=12 mice, 16 trials each odor type in each block). Rare trials with non-zero anticipatory licks in Block 2 were omitted from following neuronal analysis. (C) Activity of activated (top) and inhibited (bottom) D1 neurons in response to five-odor types in Block 2. Neurons are grouped by preferred stimulus, arrows on right indicate boundaries between groups. Black vertical lines indicate odor onset. No outcome was delivered. (D) 3D D1 neuronal population trajectories in Blocks 1 (solid lines) and 2 (dashed lines) in five trial types. (E) Cosine similarity between D1 and D2 neuronal activity averages in Blocks 1 and 2 for same valence and opposite valence odors (five neuronal subgroups as in C with pairwise comparisons, *p<0.05 rank sum test). (F) Distributions of D1 (left) and D2 (right) neuron valence scores in blocks 1 and 2 (n=294 D1 and 177 D2 neurons, **p<0.0001, paired t-test). Dotted lines show shuffled (control) distributions where the odor-valence pairing has been randomized. (G) Distributions of D1 and D2 neuron identity scores. Same neurons as in panel F.

In Block 2, in the absence of licking and outcomes, D1 neuronal responses to odors of previously learned negative valence (Figure 4C, columns 1–2) were still highly similar and responses to odors of previously learned positive valence were similar (Figure 4C, columns 4–5), as in the full task condition (Figure 2A). The Euclidean distance between D1 population activities in response to odor pairs of opposite valence was 68% greater than between odor pairs of same valence. Principal components of neuronal activity and cosine similarity measures further confirmed maintenance of strong valence representation in D1 neurons in Block 2 (Figure 4D–E). However, valence representation was no longer present in the D2 neuronal population in Block 2 (Figure 4E, Figure 4—figure supplement 1), with the population activity for odor pairs of same valence only 12% greater in Euclidean distance (compared to 68% for D1 neurons) than for odor pairs of opposite valence.

Individual neuron valence coding scores of D1 neurons decreased significantly in Block 2 as compared to the full task condition (Figure 4F, rank sum test p<0.0001) indicating that the presence of the licking response, the water and airpuff outcomes, as well as the motivated thirsty state in the mouse contributed significantly to the odor-outcome-related activity of the D1 neurons. However, the mean of the distribution of the D1 valence scores in Block 2 remained above zero (0.18±0.02; p=2.8151e-13, Wilcoxon signed rank test), with many D1 neurons maintaining strong valence representation. This valence score distribution for D1 neurons in Block 2 was significantly different from a control, shuffled distribution (Figure 4F; p=2.1025e-12, rank sum test). D1 neurons that maintained valence coding in the absence of licking and outcome were not distinguished by their relation to licking vigor in Block 2 (Figure 4—figure supplement 2A). Odor valence scores of D2 neurons decreased in the sated condition (Figure 4F, right, p<0.0001 rank sum test), and the mean of the distribution of valence scores of D2 neurons in Block 2 was no longer significantly different from zero. Valence scores of D1 neurons were significantly higher than valence scores of D2 neurons in Block 2 (p<0.01 rank sum test). Concurrently, with the omission of licking and outcomes, odor identity coding scores of both D1 and D2 neurons increased in Block 2 (Figure 4G, p<0.01 rank sum test). Salience scores calculated for D1 and D2 neurons in both blocks were indistinguishable or lower than shuffle controls, further supporting lack of salience coding (Figure 4—figure supplement 2B). While most of the analysis focused on rewarding or aversive odors, we noticed that responses to neutral odors trend with aversive odor responses in Block 1, but fall intermediate between rewarded and aversive odor responses in Block 2 Figure 3D, Figure 4—figure supplement 3.

Stimulus valence coding by D1 OT neurons is modality-sensitive

To further interrogate the stimulus valence coding property of OT neurons, the implanted Drd1-Cre mice (n=6) and Adora2A-Cre mice (n=4), which had been previously trained on the five-odor-outcome conditioning task, were then trained on a new task in which sound associations were introduced. The main motivation for these sound-odor association experiments was to determine whether sensory cues from a different modality known to activate OT neurons (Varga and Wesson, 2013; Wesson and Wilson, 2010) could become categorized with odor cues for combined valence coding. We have strong evidence from our earlier study (Millman and Murthy, 2020) that novel rewarded odors recruit the same OT neurons that respond to familiar rewarded odors, and therefore we wanted to see if adding a rewarding cue from a different modality will engage the same cells.

A new task was introduced in which two odor-outcome associations and two sound-outcome associations were presented. Odors 1 and 5 were paired with strong airpuff and large water reward as previously, and two sound tones (5 and 12 kHz) were used, also paired with strong airpuff and large water reward (Figure 5A). These sound stimuli were selected after a pilot behavioral study in a separate cohort of mice with no implants, designed to produce similar learning and anticipatory licking rates as the odor stimuli. After 1 training day with the new sound stimuli, the anticipatory licking rates of 10 mice in response to the rewarded sound stimulus were similar to that of the previously learned rewarded odor stimulus (Figure 5B). There was a significant difference between the number of licks for the rewarded and aversive sounds (p=8.0914e-28, two-sided Wilcoxon rank sum test), and a significant difference between the number of licks for the rewarded and aversive odors as before (p=8.2317e-54, two-sided Wilcoxon rank sum test). After 3 days of training on the new odor-sound associations, a probe session as described previously was conducted, in which mice were sated on water midway in the session and the odor and sound stimuli were then presented in Block 2 in the absence of licking or outcomes (Figure 5A).

Figure 5 with 3 supplements see all
Odors and sound tones associated with identical aversive and rewarding outcomes activate different D1 neuronal subpopulations in the olfactory tubercle (OT).

(A) Odor-sound association task structure. Three days of odor-sound training beginning at the end of the prior five-odor task training were conducted. Odors 1 and 5 associated with the strongest aversive and rewarding outcomes from previously learned five-odor task were preserved and two sound tones (5 and 12 kHz) were introduced with matching outcomes. Sound tone–outcome assignments were counterbalanced across mice. On day 4, a probe session was conducted as previously in which mice were sated prior to the second block. In the second block, mice did not exhibit licking and all outcomes were omitted. (B) Anticipatory licking of mice across three training days. By day 2, mice exhibit similar levels of anticipatory licking in response to the rewarded sound tone as to the rewarded odor (n=10 mice). (C) D1 neuronal population activity in Block 2 of day 4, in response to the learned odors and sound tones, and in the absence of licking, airpuffs, or water delivery. Neurons are grouped by preferred stimulus, arrows on right indicate boundaries between groups. Unlike in Figure 4C, distinct sets of D1 neurons are activated in response to odors and sound tones associated with identical outcomes. (D) Overlap in neurons responding to stimuli predicting similar aversive (orange) and rewarding (green) outcomes in the sated condition in the odor-odor task (top) and odor-sound task (bottom). Top orange, numbers of neurons in Block 2 of the odor-odor task that were activated in response to either aversive odor 1, aversive odor 2, or both of them (overlap), with % overlap indicated below. Top green, same for rewarding odors 5 and 4. Bottom, same for Block 2 of the odor-sound task (sounds in dashed lines). (E) Orange, mean activity of odor 1 activated neurons to the corresponding matched valence odor stimulus in Block 2 of the odor-odor (OO) task and matched valence sound stimulus in Block 2 of the odor-sound (OS) task (n=50 and 101 neurons). Green, same for odor 5 responsive neurons (n=100 and 134). (F) Distributions of valence scores of D1 neurons in the odor-odor task and odor-sound task in the full task condition with licking and outcomes. Shuffle control distributions are also shown as lines. (G) Same for stimulus-only condition in Block 2. (H) Principal components of D1 neuronal population activity in Blocks 1 and 2 in odor-odor task. (I) Same for odor-sound task. (J) Cosine similarity in the neuronal trajectories of stimulus responsive neuronal groups in the odor-odor task as compared to the odor-sound task in same valence vs. opposite valence stimulus pairs (four neuronal subgroups as in C with pairwise comparisons, *p<0.05 rank sum test).

In the stimulus-only condition in Block 2, we found that the same D1 neuronal populations which had previously strongly maintained similar activity patterns in response to odors of similar valence (Figure 4C) did not do so for odors and sounds of identical valence (Figure 5C). In the full-task condition when licking and outcomes (water or airpuff) were present, response similarity remained during rewarding and aversive trials (Figure 5I, Figure 5—figure supplement 1), but this similarity was lost in the stimulus-only condition. Only 5% greater Euclidean distance remained between D1 population activity in response to odor and sound of opposite valence as compared to odor and sound of same valence, unlike the 68% difference found in Block 2 of the five-odor task. Interestingly, although 9.2% of all D1 neurons (and 9% of D2 neurons) were significantly activated during at least one of the sound trial types in Block 2, these neurons were largely non-overlapping with those which responded to the odor stimuli (Figure 5D). We then directly compared the activity of neurons which were responsive to odors 1 and 5 in response to the matched valence odor stimuli (odors 2 and 4, in odor-odor task) and the matched valence sound stimuli (sounds 1 and 2, in odor-sound task). We found that aversive odor 1 responsive neurons became highly activated when aversive odor 2 was presented in Block 2 of the odor-odor task, but this was not the case when sound tone 1 was presented in the odor-sound task (Figure 5E, orange, p<0.000001 rank sum test for difference between odor-odor and odor-sound). Similarly, rewarding odor 5 responsive neurons became highly activated when rewarding odor 4 was presented in Block 2 of the odor-odor task, but this was not the case when rewarding sound tone 2 was presented (Figure 5E, green, p<0.000001 rank sum test). We also calculated valence scores for these neurons in the odor-odor task and the odor-sound task. In the standard task condition with licking and outcomes (Figure 5F), as well as in the sated condition with no licking or outcomes (Block 2, Figure 5G), the valence scores of D1 neurons in the odor-sound task were significantly lower than the valence scores of the neurons in the odor-odor task (p<0.0001, rank sum test). This occurred despite the presence of similar levels of anticipatory licking in response to the rewarded sound tone, and even though the odor- and sound-related outcomes in the odor-sound task were identically matched while those in the odor-odor task were not.

Finally, we compared the principal components of D1 neuronal population activity in response to the five odors in Blocks 1 and 2 in the odor-odor task and activity in response to the odor-sound stimuli in Blocks 1 and 2 in the odor-sound task. Principal components of D1 neuronal populations were very similar in odors predicting similar outcomes in the typical task condition (Figure 5H, top) and the sated condition with no licking or outcomes (Figure 5H, bottom). However, there was no such clear similarity of D1 neuronal population principal components in response to odors and sound tones predicting identical outcomes (Figure 5I), suggesting distinct neuronal activity trajectories in response to each of the four stimuli used in the odor-sound task. Measurement of the cosine similarity in the two conditions confirmed this interpretation (Figure 5J). D2 neurons in Block 2 of the odor-sound task did not display significant valence coding (Figure 5—figure supplements 2 and 3), which is not surprising given the lack of valence coding in Block 2 of the five-odor task (Figure 4G).

Discussion

The results of these experiments offer key insights into the function of OT neuronal circuitry during odor association learning. We find some surprising similarities and some key differences in the neuronal responsivity of D1- and D2-type OT neurons, in the first single neuron imaging experiments of specific neuronal subtypes in the OT. We found that D1 neuronal populations clearly responded to both positive and negative odor valence. This is the first demonstration of bidirectional odor valence coding by OT neurons, as no previous real-time recordings of OT neurons in response to aversive valence odors have been reported. We therefore can conclude that the OT is likely to be involved in learning about both positive and negative odor associations, rather than the alternative possibilities that the OT is only involved in learning rewarded odor associations, or that it encodes odor salience rather than signed odor valence. We note that a higher proportion of neurons was active in response to the rewarded odors than the aversive odors; however, this may be an effect of possible high value of the water reward outcome in water restricted mice as compared to the aversiveness of a relatively harmless airpuff to the nose. Surprisingly, similar proportions of D1 and D2 neurons responded to the five task odors, challenging our initial hypothesis that D2 neurons may respond more strongly to negatively reinforced odors, based on immediate early gene expression results (Murata et al., 2015; Murata et al., 2019). This suggests that models of dopamine acting to potentiate responses of D1R expressing neurons and de-potentiate responses of D2R expressing neurons or dopamine dips potentiating D2R expressing neurons (Bamford et al., 2018; Iino et al., 2020; Surmeier et al., 2007; Yagishita et al., 2014) are not sufficient to account for the neuronal activity we observed, in accordance with a more complex interplay between dopamine and the two striatal neuronal types (de Jong et al., 2019; Kutlu et al., 2021). Other inputs to D1- and D2-type neurons in the OT, including strong inputs from piriform cortex (White et al., 2019), could also create differentiated activity in these neurons.

We then addressed a long-standing question about the factors driving the responses of OT activity in odor-outcome behavioral paradigms. In previous OT recording experiments, go/no-go tasks were used where licking in response to the rewarded odors was required. In our classical conditioning task, mice similarly exhibited anticipatory licking at the onset of the rewarded odors, although the water delivery was not contingent on this response. In all of these conditions, the odor presentation itself was in every case coupled with the motor licking action of the mouse. Thus, it was unclear whether the neuronal activity recorded occurred as a result of the odor stimulus or the licking response. Previously, it was reported that the onset of the recorded neuronal responses preceded the onset of the licking action by ~200 ms (Gadziola et al., 2015; Millman and Murthy, 2020); however, this time lag is well within the time range typically seen in motor preparatory neuronal activity (Svoboda and Li, 2018; Tanaka et al., 2021) and could still be linked to the licking action or its preparation. To address this confound, we sated the mice on water and omitted the outcomes associated with the odors. We observed that in 95% of the trials in this condition, mice exhibited zero licks and we analyzed neuronal activity from only these trials. We observed that D1 OT neurons continued to respond to the learned odors and continued to strongly differentiate odor valence in the absence of licking or outcomes. This finding indicates that the critical sensory transformation step of integrating odor identity information and outcome information takes place in the OT D1 neurons, and that the odor-outcome-related activity seen in our recordings and previous recordings are not the result of unintended correlations with movement. This is especially notable given the low motivational state of the sated mouse in this experimental condition, suggesting that D1 OT neurons, at least temporarily, maintain odor valence memory even in conditions when the outcomes associated with the learned odors become less consequential. It should be noted, however, that in the condition where licking and water and airpuff delivery were present, the odor-outcome representation in D1 and D2 neurons was enhanced as compared to the stimulus-only sated condition. This indicates enhanced OT engagement during conditions in which the mice were motivated and behaving, consistent with other reports of increased neuronal activity modulation in sensory regions in attentive behaving mice as compared to passive stimulus presentation conditions (Busse et al., 2017; Carlson et al., 2018; Pakan et al., 2018).

We also note that D1 neurons are heterogeneous in terms of their valence coding, and a sizable fraction loses its valence coding in the absence of licking and outcome (Block 2). We did not find distinct populations of neurons responding to licking and valence separately (Figure 4—figure supplement 2A), indicating that licking and valence responsivity are intertwined factors. The heterogeneity of D1 responses could arise from natural stochasticity in cellular plasticity mechanisms involved in the generation of the firing patterns such that not all neurons (and connections) are modified in the desired direction. Our observed heterogeneity could, in part, be due to unintended labeling of additional neuronal types. It is unlikely that we imaged ventral pallidal neurons since we positioned our GRIN lens as ventrally as possible, confirmed the location histologically excluding mice in which we suspected more dorsal localization, and since expression of Drd1 and A2A is very low in the ventral pallidum (Figure 1—figure supplement 4A, Allen Brain Atlas). As discussed in the Results (Figure 1—figure supplement 4), it is unlikely that a significant number of neurons from the islands of Calleja were included in our analysis.

Our findings on the activity of OT D2 neurons provide an interesting contrast to the robust valence coding property of D1 neurons. While D2 neuronal populations differentiated between rewarded and aversive outcomes in the task condition involving licking and outcomes, this representation was significantly weaker than that of D1 neurons and disappeared in the stimulus-only condition when licking and outcomes were omitted. This result contrasts with photometry data in which reward contingency information was not observed in average D2 population activity in a go/no-go task (Gadziola et al., 2020), suggesting it is possible to differentiate rewarded trials in non-averaged D2 neuronal activity in a condition when the licking response is present. However, our data does demonstrate that reward contingency information is more readily represented by D1 neurons than D2 neurons. Individual D2 neurons were much more likely to respond to an individual odor out of the five task odors and exhibit little responsivity to any of the other four odors, suggesting odor-identity rather than odor-valence responsivity. Increased training resulted in increases in the valence-coding scores of D1 neurons and decreases in their odor-identity coding scores, while the opposite pattern was observed in D2 neurons, with increased odor-identity coding scores with training.

How D1 and D2 neuronal representations of learned odors diverge based on inputs to these neurons and the effect of dopamine onto them, and how they then subsequently shape behavior as they influence downstream targets are both important areas to investigate. Cre expressing neurons in the dorsal striatum from the Drd1-Cre and Adora2A-Cre mouse lines used in our imaging experiments have been previously shown to project via the distinct direct and indirect striatal output pathways (Gerfen et al., 2013), and as such have opposing overall effects on thalamus and cortex, potentiating behavioral output and inhibiting it, respectively. There is much less known about the differential projections of D1 and D2 neurons in the OT (Heimer et al., 1987; Zhou et al., 2003) and their function (Gadziola et al., 2020; Murata et al., 2015; Murata et al., 2019). Given the relatively weak odor-outcome representation by OT D2 neurons, the question also remains whether these neurons play a significant role in odor-outcome association learning, or whether D1 OT neurons predominantly contribute to this function.

Informed by previous studies showing that sound cues can also activate OT neurons (Varga and Wesson, 2013; Wesson and Wilson, 2010), we asked whether individual neurons respond similarly to odor and sound cues that predict the same reward or airpuff, which would point to multimodal integration within the OT to form a unified valence representation. Such a result might also be expected if neuronal responses were closely related to licking response and outcome, since the odor and sound cue lead to the same learned motor behavior. Consistent with previous results, we found that 24% of D1 neurons and 14% of D2 neurons were activated in response to at least one of the sound tones in the full task condition with licking and outcomes. However, unlike the large overlap and similarity between neuronal activity in response to different odors predicting similar outcomes, we found little overlap between neurons responding to odors and sounds predicting identical outcomes. This finding points to two conclusions.

First, corroborating our previous result, we can conclude that the odor valence-related OT activity was not primarily a result of the licking response of the mouse. We demonstrate that behaviorally, the anticipatory licking rates of the mice in response to the rewarded odor stimulus and the rewarded sound stimulus are similarly high after training, yet stimulus-outcome representation in the OT during the odor-sound task is minimal in the period prior to outcome delivery even in the presence of matched anticipatory licking rates. Second, we can conclude that stimulus valence representation by D1 OT neurons does not automatically generalize to multimodal stimuli. Our previous study (Millman and Murthy, 2020) demonstrated that introducing novel rewarded odors recruits the same valence cells that respond to familiar rewarded odors. Our current data indicate that adding a new stimulus of a different modality does NOT automatically recruit the same reward category cells. Sound-related responses in OT do occur and supra-additive effects of odor and sound stimuli have been reported (Wesson and Wilson, 2010). While there were clearly strong neuronal responses to sound tones in our experiments, our design did not include multiple sound tones for outcomes of the same valence as we did for odors in the original five-odor task. Therefore, we cannot be sure that a cell responding to a rewarding tone is simply because of sensory tuning, or because of valence coding. Future studies can address this important issue, since it will inform us about the possible specialized role of limbic brain regions such as the OT to store emotionally charged odor memories, a property that may be unique to the sensory modality of olfaction.

An unavoidable consequence of GRIN lens imaging of deep brain regions is tissue damage, but we are confident that our key conclusions are not affected by this issue. We made concerted efforts to minimize the damage caused by the implanted cannula. First, the cannula was constructed from highly biocompatible and thin-walled polyamide tubing and quartz floor which have previously been shown to minimize glial scar tissue (Bocarsly et al., 2015). Second, 2 mm of the cortex were removed by suction prior to the virus injection and cannula implantation to minimize pressure in the brain. Finally, the mice were allowed to recover for at least 1 month prior to the onset of behavioral training. After the extensive recovery period, mice learned the association tasks rapidly. The anticipatory licking of the mice with no GRIN lens implant is similar to that of the implanted mice (Figure 1—figure supplement 5), suggesting that the implanted mice are not impaired in their ability to learn the stimulus-outcome associations.

The main comparisons we make in our study are likely not a function of the damage caused by the lens. First, the differences in the valence coding of D1 and D2 neurons are unlikely to be a result of lens damage, since there is no reason to suspect that damage caused by the lens will differentially affect D1 and D2 neuronal activity in the OT. D1- and D2-type neurons in the OT and the striatum typically receive inputs from similar upstream structures and their inputs were not generally altered by the lens damage. Second, there is no reason to suspect that the robust valence coding in the absence of the licking and outcomes by D1 neurons could be a result of the lens damage. Third, there is no particular reason to suggest that the distinction between the odor and sound responsive neurons in the last set of experiments would be a result of the damage caused by the lens as the possible auditory cortical projections to the OT arrived from the posterior direction. Finally, the nature of the valence-related responses we observe in the OT are similar to those observed by others using tetrode recordings (Gadziola et al., 2015; Millman and Murthy, 2020) in which there was presumably less damage to the striatum.

In summary, we find that D1 OT neurons are more likely to encode learned odor valence than D2 neurons, and conversely less likely to encode odor identity. We also demonstrate, for the first time, that even when the licking response to rewarded odors is eliminated, OT D1 neurons continue to robustly encode odor valence suggesting that this stimulus to valence transformation by the OT precedes the motor action itself. These results also suggest that OT odor valence representation could inform downstream brain regions of the value of odor stimuli. Finally, we find that stimulus valence representation by OT neurons is limited to olfactory stimuli suggesting a specialized role of the OT in assigning emotional tags to odors based on previous experience. Further investigation into the relative contributions of D1 and D2 OT neurons to odor association learning and the neural mechanisms that result in the differential responses of these neuronal types is required. These neuronal imaging results suggest a specialized role for the OT in odor valence memory, and further studies can be conducted to assess the causal contribution of the OT to the hypothesized unique emotional qualities of olfactory memory.

Materials and methods

Animals

Adult male and female heterozygous B6.FVB(Cg)-Tg(Drd1-Cre)EY262Gsat/Mmucd and B6.FVB(Cg)-Tg(Adora2A-Cre)KG139Gsat/Mmucd (MMRRC) mice were 2–6 months of age at the start of the experiments. Due to the highly consistent colocalization of A2A receptors and D2 dopamine receptors in the striatum, no colocalization of A2A receptors with D1 dopamine receptors (Gerfen and Surmeier, 2011; Svenningsson et al., 1998), and previously established use of Adora2A-Cre mice for indirect pathway specific manipulation (Cui et al., 2013), we proceeded with the use of Adora2A-Cre mice to image D2-type neurons. All experiments were conducted with approved protocols and in accordance with Harvard University Animal Care Guidelines.

Cannula assembly

Custom-designed cannula were assembled in house. 6.2 mm length 1.1 mm diameter ultra-thin wall biocompatible polyimide tubing (MicroLumen) which was demonstrated to cause minimal inflammatory response in the brain (Bocarsly et al., 2015) was used for the walls of the cannula. 150 µm thickness quartz coverslips (Electron Microscopy Sciences) were cut to 1 mm diameter disks in a laser cutter and used as the floor of the cannula. Cut quartz disks were held with the assistance of vacuum under the view of a surgical microscope, attached to the polyimide tube with Norland Optical Adhesive NOA 68 (Edmund Optics), and adhesive was cured with UV light source (ThorLabs). Directly prior to surgical implantation, cannula were inspected and disinfected with the use of the UV light source.

Surgery

Naïve mice underwent surgical virus injection, cannula implantation, and head fixation plate implantation, prior to any behavioral training. Mice were anesthetized with an intraperitoneal injection of a mixture of xylazine (10 mg/kg) and ketamine (80 mg/kg) and placed in a stereotaxic apparatus. A 1.4 mm craniotomy was performed at 1.5 mm AP, 1.3 mm ML in the right hemisphere. A 22 G needle was used to suction 2 mm below the brain surface prior to virus injection. A pulled glass micropipette attached to a nanoinjector (MO-10, Narishige) was used to inject 400 nL of pGP-AAV9-syn-FLEX-jGCaMP7s-WPRE (Addgene) virus at a depth of 4.8 mm DV at a rate of 100 nL/min. Five minutes after the injection was completed, the glass pipette was raised out of the brain over the course of another 5 min. The sanitized cannula was then held lightly with a dental paper point inserted into its center and attached to the stereotaxic arm. The cannula was lowered over the course of 10 min to a depth of 4.9 mm DV. The cannula was secured onto the skull with cyanoacrylate glue and a head fixation plate was also glued to the skull behind the cannula. Dental cement (MetaBond) was then used to cover the skull and headplate. The opening of the cannula was covered with a silicone sealant (KwikSil). Mice were single housed after surgery. After a period of 4 weeks to allow for virus expression and the reduction of the inflammatory response to the insertion of the cannula, a 1 mm diameter, 3.4 mm long 0.5NA GRIN lens (ThorLabs) was inserted into the cannula and behavioral training and imaging begun.

Behavioral training

Mice were water restricted to reach 85–90% of their initial body weight and provided approximately 1–1.5 mL water per day in order to maintain desired weight. Mice were habituated to head fixation and drinking from water spout prior to initial training session. In the pre-training (day 0), mice were provided large water drops (20 µL), small water drops (10 µL), strong airpuff (10 PSI), and weak airpuff (5 PSI) in identical trial structure as full five-odor conditioning task, but odors were not used. Each mouse was then assigned five odor-outcome contingencies with the monomolecular odorants hexanal, limonene, anisole, eucalyptol, and heptanal. Odors were delivered via a custom-built olfactometer as described previously (Soucy et al., 2009) with a 1.5 L/min flow rate at a concentration of 20% for 1.5 s each. Odor-outcome contingencies were assigned so as to equally or close to equally counterbalance the odor-outcome contingencies across each cohort of mice. The delay between the solenoid valve opening (which is denoted as time 0 throughout the manuscript) and the arrival of the odor at the mouse’s nostrils was measured with a miniPID instrument (Aurora Scientific, Inc).

In days 1–4 of training, each of the five odors and associated outcomes were provided 30 times with 20 s inter-trial intervals. In 10% of trials (three trials of each trial type), the outcomes were omitted; however, this number of trials was not later found to be sufficient for the analysis of neuronal activity. Trial types were interspersed randomly across the session, with the constraint that equal numbers of each trial type occurred in the first and second half of each session to ensure equal trial type representation for the duration of each imaging session. Licking of the water delivery spout was measured throughout training and imaging with the use of a capacitance sensing Arduino circuit. Behavioral events control and recording was conducted with Python with adapted use of the ScopeFoundry platform (http://www.scopefoundry.org/) and National Instruments DAQ hardware.

Prior to Block 2 in the probe session in day 5 of training, 2 mL of free water was provided to the headfixed mouse over the course of 2 min. Fifteen trials of each trial type were then presented in identical task structure as in Block 1, however all water and airpuff outcomes were omitted.

Training in the odor-sound task occurred 1–2 days after the completion of the five-odor task training. Odors 1 and 5 with strong airpuff and large water drop outcomes were preserved, and 5 and 12 kHz sound tones were introduced paired with the same strong airpuff and large water drop outcomes. As in previous task, 30 trials of each condition were randomly interspersed across session time. Onset and duration (1.5 s) of the sound stimuli and associated outcomes was identical to that of the odor stimuli. After 3 days of odor-sound task training, a probe session with two blocks was conducted as in the original five-odor task.

Two-photon imaging of calcium activity

A custom-built microscope was used for in vivo imaging as described previously (Kapoor et al., 2016; Petzold et al., 2009). Imaging was conducted at 5 Hz with an air objective (10×, Leica) at 930 nm using a Ti:sapphire laser (Chameleon Ultra, Coherent) with a 140 fs pulse width and 80 MHz repetition rate. Image acquisition, scanning, and stimulus delivery were controlled by custom-written software (Kapoor, 2022; deposited in GitHub) in LabVIEW (National Instruments). Prior to two-photon imaging, the position of the GRIN lens and approximate neuronal imaging plane was determined with a camera. The head fixation plate was mounted on an adjustable pitch and roll platform (ThorLabs) which allowed for manual adjustment of the lens angle to parallel alignment with the objective. The depth of the imaging plane was adjusted each day to closely match that of the previous imaging days, capturing the same or highly overlapping neuronal populations across days of training.

Calcium imaging data analysis

Imaging data was motion corrected with Non-Rigid Motion Correction (NoRMCorre) (Pnevmatikakis and Giovannucci, 2017). The activity of single neurons was then isolated and background subtracted with the use of CaImAn (Giovannucci et al., 2019) followed by manual refinement. The number of putative calcium transient events in each neuron was then quantified based on a criterion of activity 3 standard deviations above a temporally proximal baseline lasting longer than 5 frames, and neurons with less than two recorded transients were not used in the analysis. Due to baseline fluorescence fluctuations in single neurons, the activity of each neuron in individual trials was normalized to a 1 s pre-trial baseline to isolate trial event-related fluorescence changes. Due to the 5 Hz imaging rate and following use of a minimal locally weighed smoothing filter to reduce noise in recordings, in some cases event-triggered changes in fluorescence may appear to begin in 1–2 frames prior to the event time.

Valence Score=mean (differences between opposite valence odors)mean (differences between same valence odors)|maximum responseminimum response|
Identity Score=Strongest odor responsesecond strongest odor responsestrongest odor response
Salience Score=mean (differences between different salience odors)mean (differences between similar salience odors)|maximum responseminimum response|

Odor valence coding scores and odor identity coding scores were computed for neurons which had significant mean activity deviations of >30% from pre-trial baseline as shown below. Due to the variability in the neutral odor responses which often tracked closely with either the aversive odors or the rewarded odors in neurons which clearly differentiated between the rewarded and aversive odors, the neutral odor responses were not included in the numerator of the valence score formula (but they could still feature in the denominator if neutral odors caused maximum or minimum response for a given neuron). A salience score was calculated analogous to the valence score using the formula noted above. Distributions of odor valence scores and odor identity scores across neuronal types and task conditions were compared with the non-parametric Wilcoxon rank sum test. We generated null distributions by shuffling the odor-valence pairing. Since the identity scores were calculated by comparing the best odor response to the next best one, the shuffling procedure does not affect the identity scores and was not done for those distributions. Matrix difference measures of neuronal population activity were obtained by taking the Euclidean norm of the difference between population activity in response to pairs of stimuli used in the task. The mean norm for stimuli pairs of opposite valence was then compared to the mean norm for stimuli pairs of the same valence. Neuronal population activity dimensionality reduction and trajectory analysis was conducted with the use of the DataHigh toolbox (Cowley et al., 2013) with all D1 and D2 neurons collected in the dataset and trial types used as input conditions.

Histological confirmation of imaging site

After completion of imaging experiments, mice were transcardially perfused, and the brains were removed from the skull. Coronal floating sections were cut using a vibratome (Leica VT1000S). Brain sections were imaged using the Zeiss Axio Scan slide scanner at the Harvard Center for Biological Imaging to visualize the location of GCaMP expression and the location of cannula tip. Brain section images were matched and overlaid with the Paxinos and Franklin Mouse Brain Atlas cross-sections to identify imaging location. Six of eleven implanted Drd1-Cre mice had confirmed OT imaging locations, while in others the tip of the cannula was located in ventral pallidum or nucleus accumbens. Eight of eleven implanted Adora2A-Cre mice had confirmed OT imaging locations, but only six of eight produced satisfactory imaging results. Only mice with confirmed OT imaging locations and successful imaging results were used in later analysis.

Statistical analysis

We primarily used the non-parametric Wilcoxon rank sum test (referred to simply as rank sum test in the main text) and Fisher’s exact test, since we could not assume normally distributed data in most cases. Comparisons involving t-tests were paired and two-tailed.

Data availability

Raw data, source data, metadata, and analysis code have been uploaded into Dryad linked to this manuscript.

The following data sets were generated
    1. Martiros N
    2. Murthy V
    3. Kapoor V
    4. Kim S
    (2021) Dryad Digital Repository
    Two-photon imaging of D1 and D2 type neurons in the olfactory tubercle of behaving mice.
    https://doi.org/10.5061/dryad.6hdr7sr28

References

  1. Book
    1. Purves D
    2. Fitzpatrick D
    (2001)
    Neuroscience: Central Projections of the Olfactory Bulb
    Sinauer Associates.
  2. Book
    1. Schneider J
    (2013)
    Brain Structure and Its Origins
    MIT Press.

Decision letter

  1. Alicia Izquierdo
    Reviewing Editor; University of California, Los Angeles, United States
  2. Michael J Frank
    Senior Editor; Brown University, United States
  3. Daniel W Wesson
    Reviewer; University of Florida, United States

Our editorial process produces two outputs: i) public reviews designed to be posted alongside the preprint for the benefit of readers; ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "Distinct representation of cue-outcome association by D1 and D2 neurons in the olfactory striatum" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Michael Frank as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Daniel W Wesson (Reviewer #1).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Martiros et al. monitored the activity of dopamine D1 and D2-expressing neurons within the ventral striatum olfactory tubercule (OT) using 2-photon microscopy, as mice learned to associate odors and tones with either positive or aversive outcomes. D1 neurons robustly encoded learned odor valence, and D1 neurons maintained an odor valence representation even when appetitive or aversive unconditioned stimuli were removed. In contrast, D2 neurons were more selective to odor identity than D1 neurons, and odor valence coding in D2 neurons was dependent on odor outcome associations. Finally, authors show that in D1 neurons, odor and tone outcome associations recruit largely non-overlapping neuronal ensembles. Overall, this study was deemed by reviewers as an interesting and important contribution, however, they did identify a need for added details about data analyses and statistical comparisons for a clearer interpretation.

Essential revisions:

1) Authors should provide more details about the specific cell types being monitored in the D1 and A2A-cre mice since cre is present in neurons, not just medium spiny / striatal projection neurons. The authors did not exclude these other cells in their analysis or experimentally so any revision should be clear about inclusion of other types of neurons (e.g., D3).

2) The function of the different subpopulations of 'valence coding' D1 neurons should be clarified in both the Results and Discussion. Relatedly, it is unclear how neuronal responses to neutral odors are taken into consideration when calculating valence scores. It is suggested to include the distributions of valence scores with shuffled cell-odor pairs as reference points. Statistical tests for comparing odor valence and identity distributions should be described in more detail.

3) It is unclear how odor identity (and sound frequency) is represented in neuronal response patterns. The authors should show neuronal responses grouped by mice / odorant-unconditioned stimulus combinations, in addition to showing neuronal responses grouped by the trial outcomes only.

4) There was concern about a large portion of the brain damaged by the placement of the GRIN lens. The implication of this damage needs to be discussed.

5) It is suggested to better link behavior to neural responses- specifically the timing of licking with the GCaMP response.

6) Authors should provide a clearer rationale for the tone learning study. The motivation for the odor-tone association experiments is unclear. The main result is that different subpopulations of OT neurons encode odor and tone associations. However, this does not appear to be a question the authors set out to address a priori.

Reviewer #1 (Recommendations for the authors):

This study is a much needed addition to the field, and the manuscript is excellent and well done. I have a few number of major and a longer list of minor issues I'd appreciate being addressed.

Figure 4F2 – the authors package these results that most D1 neurons maintain valence coding in absence of instrumental responding or conditioned outcomes, and while this is true, a good amount do not. It would be beneficial for the authors to elaborate upon what those two different subpopulations of D1 neurons are doing both in Results and Discussion.

The authors do show that some cells display auditory valence coding, yet conclude across the population there is no such activity. It would be helpful for the authors to describe activity of the cells which do in fact show learned responses to the tones. Related to this, in some places the authors describe their results as "not modality invariant". Perhaps this could be rephrased to read, "modality variant" to avoid double negatives. Finally, isn't the finding that there are responses to multisensory stimuli and that these responses are dependent upon learning reason to reconsider titling this paper as originating from the "olfactory striatum"? These results show this is not simply an "olfactory" structure, and as described in Wesson 2020 (Tubular Striatum), the concept of calling the olfactory tubercle the "olfactory striatum" lost ground decades ago. Of course, I'm all for scholarly creativity of the authors and am not asking them to change the title, but more to think about what message this sends to readers.

Overall, it would be helpful to see the timing of licking, even if on average, to help relate the gcamp response to behavior better. This is important since the authors used a fixed ITI of 20s and normalized each neuron's activity to a 1s pre-trial baseline. But can we be sure there was no anticipatory-related activity during that time? More specifically, it would be nice to show a new figure/inset of Figure 1C, day 1 and/or 4 in raster-format (similar to 4B perhaps) so readers can gain an appreciation for the timing of the licks specifically in this panel, if possible (it's reported in the first results paragraph that mice began licking after odor onset before water delivery, but it is not shown as such).

I'd appreciate the authors elaborating upon the argument from their Discussion that tubercle neurons "encodes odor salience rather than signed odor valence". I'm not sure that I appreciate how the authors results directly disambiguate this, nor am I even clear how one could do in the context of like experiments.

Reviewer #2 (Recommendations for the authors):

1) It is unclear how neuronal responses to neutral odors are taken into consideration when calculating valence scores. For example, a neuron with a strong (positive) valence score would be expected to respond to the rewarded odor but not to the neutral odor or odors associated with the air puff. The current analysis appears to ignore neutral odors and should be expanded to include neutral odors.

2) In figure 3, the authors claim that for D1 but not D2 neurons, the valence score increases with training. Valence scores in naive mice are expected to be zero – this should be reflected in the data. To help interpretability, the distributions of valence scores with shuffled cell-odor pairs should be included as reference points.

Furthermore, figure 1 suggests that there is a rage of anticipatory lick numbers on day one, likely reflecting differences in learning during the first session. When analyzing imaging data over time, the authors should compare data from mice with similar behavioral performance.

3) In figure 4, the authors state that the mean valence score of D1 neurons remains above zero for block 2 trials – the mean valence score should be indicated in the figure. Statistical tests for comparing odor valence and identity distributions should be described in more detail.

4) The motivation for the odor-tone association experiments described in figure 5 is unclear. Without characterizing tone-outcome learning in OT neurons in the described experimental setting, the multimodal task is difficult to interpret. Furthermore, the response properties of neurons in the 'full task condition' are hardly analyzed (supplementary figure 4), instead, the analysis focuses on block 2 trials and is difficult to follow.

The main result is that different subpopulations of OT neurons encode odor and tone associations. However, this does not appear to be a question the authors set out to address and may be a trivial result given the well-established differences in neuronal connectivity.

5) It is unclear how odor identity (and sound frequency) is represented in neuronal response patterns in figures 2 and 4. The authors should show neuronal responses grouped by mice / odorant-unconditioned stimulus combinations, in addition to showing neuronal responses grouped by the trial outcomes only.

Reviewer #3 (Recommendations for the authors):

p. 2

The term "addictive" is not appropriate to describe intracranial self-administration of cocaine. Use "reinforcing" or "rewarding", instead.

Suppl. Figure 1A

OT D1R neurons do not project to the substantia nigra or ventral tegmental area, or OT D2R neurons do not project to the globus pallidus. It is confusing to mention such projections that are most likely caused by viral diffusions to the nucleus accumbens and dorsal striatum. Clarify the points that the authors are making here.

Figure 3A

Dotted circles used to describe their identifies are not shown on the actual diagram.

p.3

Licking response may not be labeled as "instrumental". Essentially, the study ued a Pavlovian conditioning procedure rather than operant conditioning procedure. Therefore, the conditioned licking response is a Pavlovian conditioned response.

p. 4

For clarity, rewrite the sentence that contains the phrase "non-zero licks were excluded."

Rewrite the sentence "Odors 1 and 5, which... were combined with two sound tones". This sentence does not convey what the authors are trying to describe. The two modalities were combined in the new experiment; odors did not get combined with tones, which has a completely different meaning.

p. 4 and Figure 5A,

Avoid the phrases "odor and sound pairs" and "odor-sound pairs". In the psychology/behavior conditioning literature, they mean that odors are paired with sounds.

Figure 5B

The plot suggests that by day 3, the mice may not have fully discriminated between sounds 1 and 2. The mice displayed anticipatory licks with sound 1 while they did not with odor 1.

Figure 5F-5G and Suppl. Figure 5B-5C

It is more informative to show data separately between O and S for the plots of the OS experiment.

Suppl. Figure 4

Airpuff-paired sound triggered little or no conditioned response from the D1 neurons. What does this mean?

Suppl. Figure 5B-5C

The Y-axis is not labeled.

p. 6

What do water- or airpuff-paired neural signals encode when they are present in the absence of behavior?

The last full paragraph on p. 6

The discussion is confusing. The authors mention that D1R neurons and D2R neurons have differential projections and that they have confirmed the fact in Fg. S1A. However, the authors do not discuss how projection patterns of the D1R and D2R neurons in the dorsal striatum and the nucleus accumbens are relevant to those of the OT. OT D1R and D2R neurons primarily project to the ventral pallidum (Heimer et al. 1987, Zhou et al. 2003); it is not clear how these neurons differentially affect downstream circuits.

Heimer L, Zaborszky L, Zahm DS, Alheid GF. 1987. The ventral striatopallidothalamic projection: I. The striatopallidal link originating in the striatal parts of the olfactory tubercle. J Comp Neurol 255: 571-91

Zhou L, Furuta T, Kaneko T. 2003. Chemical organization of projection neurons in the rat accumbens nucleus and olfactory tubercle. Neuroscience 120: 783-98

The paragraph starting from the bottom of p. 6 and ending at the top of p. and

It is not coherently written. It is hard to follow the arguments.

p. 7

The authors state "D1 OT neurons selectively and bidirectionally encode learned odor valence, unlike D2 neurons". This statement sounds too strong. The sub-heading in the result section on p. 3 characterizes the results more sensibly as follows: "Individual D1 neurons are more likely to encode odor valence and D2 neurons more likely to encode odor identity".

https://doi.org/10.7554/eLife.75463.sa1

Author response

Essential revisions:

1) Authors should provide more details about the specific cell types being monitored in the D1 and A2A-cre mice since cre is present in neurons, not just medium spiny / striatal projection neurons. The authors did not exclude these other cells in their analysis or experimentally so any revision should be clear about inclusion of other types of neurons (e.g., D3).

We acknowledge the potential heterogeneity of the cells we monitored in these mice. Two concerns raised by Reviewer 3 are: (1) cells in the islands of Calleja (IC), which has been shown to express D1 in addition to D3 receptors may be included in the imaging and (2) we may have inadvertently imaged cells in the ventral pallidum, which is located just dorsal to the OT. We address these concerns in detail in the response to Reviewer 3 below (Weakness, point 2). We add some text in the Discussion, acknowledging these points (Discussion third paragraph).

2) The function of the different subpopulations of 'valence coding' D1 neurons should be clarified in both the Results and Discussion. Relatedly, it is unclear how neuronal responses to neutral odors are taken into consideration when calculating valence scores. It is suggested to include the distributions of valence scores with shuffled cell-odor pairs as reference points. Statistical tests for comparing odor valence and identity distributions should be described in more detail.

We now discuss the different subpopulations (Discussion third paragraph), and we explain in detail in the Response to reviewer 1 (first point). We have also considered different metrics for describing valence that would include neutral odors, and discuss why we converged on the particular metric we use. We explain in detail our response below (Reviewer 2, point 1). We have now used shuffled distributions for comparison (Figure 3F, Figure 4FG, Figure 5FG and the associated text in the Results section) – thanks for this important suggestion. Finally, we describe statistical tests in greater detail throughout the manuscript.

3) It is unclear how odor identity (and sound frequency) is represented in neuronal response patterns. The authors should show neuronal responses grouped by mice / odorant-unconditioned stimulus combinations, in addition to showing neuronal responses grouped by the trial outcomes only.

Thank you for this suggestion. To better illustrate the neuronal responses to specific odor-outcome pairings, we have added a supplementary figure with neurons grouped by mouse / odorant-outcome combination (Figure 2 figure supplement 1 and Figure 5 figure supplements 2 and3). The basic patterns of results hold for individual mice.

4) There was concern about a large portion of the brain damaged by the placement of the GRIN lens. The implication of this damage needs to be discussed.

We have added some text discussing this damage (Discussion, eigth paragraph), which is an unfortunate outcome of the experimental procedure. We address the issue in two ways. First, we argue that the principal findings in our work on the OT neuronal activity are unlikely to be related to this damage. Second, we show that the behavior of the mice is not detectably affected by the damage (Figure 1 figure supplement 5). We discuss these issues in detail below in the response to reviewer 3.

5) It is suggested to better link behavior to neural responses- specifically the timing of licking with the GCaMP response.

We appreciate this suggestion and have compared the timing of licking and the neuronal (calcium) responses. Details are below in the response to Reviewer 1 (forth point). Data are now shown in Figure 1 figure supplement 3.

6) Authors should provide a clearer rationale for the tone learning study. The motivation for the odor-tone association experiments is unclear. The main result is that different subpopulations of OT neurons encode odor and tone associations. However, this does not appear to be a question the authors set out to address a priori.

The tone learning experiments were actually an integral part of the study design. We used tones to determine whether adding another stimulus of a different modality will recruit the same cells for valence coding. We have strong evidence from our earlier study (Millman and Murthy, J Neurosci 2020) that adding new odors does indeed recruit same valence cells, so we wanted to see if adding a rewarding cue from a different modality will engage the same cells. Although the use 4 tones (2 for aversion and 2 for reward) might be even cleaner, nevertheless, our current data indicate that adding a new stimulus does NOT automatically recruit the same reward category cells. We present a detailed response below under Reviewer 2, point 4. We explain these points in the manuscript now (Results section, first paragraph when sound experiments are presented, and Discussion seventh paragraph), and also acknowledge the limitations.

Reviewer #1 (Recommendations for the authors):

This study is a much needed addition to the field, and the manuscript is excellent and well done. I have a few number of major and a longer list of minor issues I'd appreciate being addressed.

Figure 4F2 – the authors package these results that most D1 neurons maintain valence coding in absence of instrumental responding or conditioned outcomes, and while this is true, a good amount do not. It would be beneficial for the authors to elaborate upon what those two different subpopulations of D1 neurons are doing both in Results and Discussion.

This is a question of great interest to us. In response to the Reviewer’s comment, we conjectured that it may be possible that a subpopulation of licking-responsive D1 neurons with high valence scores in Block 1 may have stopped differentiating between rewarded and unrewarded odors in Block 2. In order to test this, we correlated the activity of each neuron on a trial-by-trial basis in the pre-outcome period to the number of anticipatory licks in the pre-outcome period in Block 1. Activity of individual neurons was considered to be significantly correlated to anticipatory licking if the absolute value of Pearson’s correlation coefficient was > 0.5 and the p value was < 0.05 in either odor 4 or odor 5 trial types (16 trials each odor type). We then compared the distribution of the valence scores of Block 1 licking correlated and non-correlated neurons in Blocks 1 and 2 (see Figure 4—figure supplement 2). In contrast to the hypothesis that licking correlated neurons may be less likely to encode valence in Block 2, in the absence of licking, we found that these neurons maintained high valence scores in Block 2. In both groups of neurons, the distribution of valence scores in Block 2 was significantly different from that of the shuffled distribution (Wilcoxon Rank sum test p < 0.01). Rather, the valence scores of the Block 1 licking correlated neurons were significantly higher in Block 2, in the absence of licking, than those of the non-licking correlated neurons in Block 2 (Wilcoxon Rank sum test p < 0.0001). This suggests that there is not a separate population of neurons responding to licking and valence separately, but rather that licking and valence responsivity are intertwined factors, such that even in the absence of licking the same neurons continue to encode odor valence.

We continue to be interested in the factors that lead to the decrease in the valence coding in Block 2 and suggest that the animal’s low motivational state in Block 2 (sated on water, lack of outcomes, and engagement) likely results in lower overall engagement of the OT neuronal circuitry. In addition, in almost all coding frameworks, not every cell has directly interpretable responses and many factors will clearly influence firing. It is also plausible that any cellular plasticity mechanisms involved in the generation of the firing patterns will have stochastic elements such that not all neurons (and connections) will be modified in the desired direction. Another way to interpret the data is that the particular task we have engages only a small part of the overall capacity of the OT neural circuit – so one can expect only a fraction of the neurons to be altered by this task, leaving many as reservoirs for other functions. We continue to be interested in this question, but more directed experimentation may be required to explain the changes in the neuronal activity levels and valence coding between Blocks 1 and 2 and contrast it with typical representational drift. We have added this data in Figure 4 figure supplement 2B, and discuss this issue in the Discussion section, paragraph 3.

The authors do show that some cells display auditory valence coding, yet conclude across the population there is no such activity. It would be helpful for the authors to describe activity of the cells which do in fact show learned responses to the tones.

In the present version of the study, we only used a single aversive tone and a single rewarding tone. While there were clearly strong neuronal responses to these sound tones, we don’t have multiple sound tones for outcomes of the same valence as we did for odors in the original five odor task. Therefore, we cannot be sure that a cell responding to a rewarding tone is simply because of sensory tuning, or because of valence coding. It is certainly possible that if we had used multiple sound tones of the same valence, we would observe valence coding for sounds as we observed for odors in the five odor task. The primary question we asked with the existing experiments was whether neurons responding to odor of the same valence also responded to auditory stimuli of the same valence, and we found that this was not common – hence the perhaps confusing phrasing of “not modality invariant” (which we have replaced with “modality-sensitive”). We note that our earlier study (Millman and Murthy J Neurosci 2020) offered strong evidence that adding new rewarded odors will indeed recruit same valence cells that respond to familiar rewarded odors. Our current data indicate that adding a new stimulus of a different modality does NOT automatically recruit the same reward category cells.

Related to this, in some places the authors describe their results as "not modality invariant". Perhaps this could be rephrased to read, "modality variant" to avoid double negatives. Finally, isn't the finding that there are responses to multisensory stimuli and that these responses are dependent upon learning reason to reconsider titling this paper as originating from the "olfactory striatum"? These results show this is not simply an "olfactory" structure, and as described in Wesson 2020 (Tubular Striatum), the concept of calling the olfactory tubercle the "olfactory striatum" lost ground decades ago. Of course, I'm all for scholarly creativity of the authors and am not asking them to change the title, but more to think about what message this sends to readers.

We have rephrased the relevant text as “modality-sensitive” (in the Abstract and in the main text). We appreciate the point made by this reviewer about the multisensory nature of this structure and the use of the term olfactory striatum. After considering all alternatives, we prefer to use the classical term olfactory tubercle and have titled the paper as “Distinct representation of cue-outcome association by D1 and D2 neurons in the ventral striatum’s olfactory tubercle

Overall, it would be helpful to see the timing of licking, even if on average, to help relate the gcamp response to behavior better. This is important since the authors used a fixed ITI of 20s and normalized each neuron's activity to a 1s pre-trial baseline. But can we be sure there was no anticipatory-related activity during that time? More specifically, it would be nice to show a new figure/inset of Figure 1C, day 1 and/or 4 in raster-format (similar to 4B perhaps) so readers can gain an appreciation for the timing of the licks specifically in this panel, if possible (it's reported in the first results paragraph that mice began licking after odor onset before water delivery, but it is not shown as such).

Thank you for this excellent suggestion. We have plotted the time of the first lick post-odor onset for each of the four training days in D1 and D2 mice, and the distribution of the times of the first lick (we show this in Figure 1 figure supplement 2). We find that the time of the first lick is on average 0.72s ± 0.3s (std dev) after odor onset. Rarely, the mice would lick immediately after the opening of the odor valve, but this was not in response to only the rewarding odors but occurred also in response to other odor types. The timing of the first lick we observe here is later than had been observed in previous studies (Gadziola et al., 2015, Millman and Murthy, 2020) because in those studies a licking response was required to receive the water reward, whereas in our study the delivery of the water was not contingent on the licking response. Hence, the mice tended to start licking closer to the time of the water delivery, which occurred at 1.3s after the odor onset. The late onset of the licking response that we quantified further strengthens the argument that the valence-related activity we observed in the first 0.5s after odor onset is not a result of the licking response but rather in response to the positively reinforced odors instead. As displayed in Figure 2A and B, the activity of OT neurons often peaks by the 0.5s after odor onset well preceding the licking itself. Per the reviewer’s suggestion, we will place dashed lines in the neuronal population activity plots to indicate the average time of licking onset in the rewarded trials.

I'd appreciate the authors elaborating upon the argument from their Discussion that tubercle neurons "encodes odor salience rather than signed odor valence". I'm not sure that I appreciate how the authors results directly disambiguate this, nor am I even clear how one could do in the context of like experiments.

We propose that salience coding neurons would respond similarly to odors of high relevance (e.g. odors predicting strong airpuff or big water reward) whereas valence coding neurons would respond similarly to odors predicting outcomes of the same sign (e.g. two odors predicting water reward), as in the hypothetical responses illustrated in Figure 3A. Given the distinctly different population and individual neuronal activity to odors predicting positive and negative outcomes, we quickly concluded that it was rare for neurons to exhibit response patterns consistent with salience coding without regard to outcome direction. We quantify this in the following portion of the text “the prevalence of neuronal responses consistent with odor salience coding was very low (3 of 529 D1 neurons were activated for high motivational salience odors 1 and 5, but not odors 2-4, as compared to 51 neurons activated for positive valence odors 4 and 5, but not odors 1-3, p < 0.000001 Fisher’s Test for D1 and D2 neurons).”

In response to the Reviewer’s question, and to further elaborate the difference between valence and salience coding, we added a salience score calculated analogously to the valence score as follows:

Salience Score=mean(differences between different salience odors)  mean(differences between similar salience odors)|maximum response minimum response|

In this calculation, odor pairs 3-5 and odor pairs 3-1 were considered to have different salience (neutral odor v odor with highly salient outcome). Odor pairs 1-5 and 2-4 were considered of more similar salience because they produced large and medium outcomes, respectively. According to this calculation, a neuron with salience-like responses (as shown for example in Figure 3A, third row) should receive a high positive salience score. In contrast, a neuron with very different responses to outcomes of opposite sign (rewarding v aversive) but similar magnitude, should receive a negative salience score inconsistent with salience coding. As expected, we find that there are very few D1 or D2 neurons in either Block 1 or 2 of the task with have positive salience scores. This indicates that the incidence of neurons with responses consistent with salience coding is highly rare in the OT. If anything, a comparison with the shuffle controls indicates that there is greater than chance occurrence of negative scores (as expected if valence coding is strongly prevalent, skewing the salience scores). We present this data in Figure 3 figure supplement 1 and Figure 4 figure supplement 2A now, and mention in the Results.

Reviewer #2 (Recommendations for the authors):

1) It is unclear how neuronal responses to neutral odors are taken into consideration when calculating valence scores. For example, a neuron with a strong (positive) valence score would be expected to respond to the rewarded odor but not to the neutral odor or odors associated with the air puff. The current analysis appears to ignore neutral odors and should be expanded to include neutral odors.

Thank you for this clarifying point. We did not sufficiently address the responses of the neurons in neutral odors in the original manuscript and have added results to better demonstrate the neutral odor responses, which are interesting. Neuronal responses to neutral odor trials were not always intermediates between aversive and rewarding odor responses. Instead, neutral odor responses often tracked closely with the aversive odors, and more rarely tracked closely with the rewarded odors or fell between the two (see examples in Figure 3D of our manuscript). In Block 1 of the last day of training, valence responsive neurons (those which clearly differentiated between rewarded and aversive odors) tended to respond to neutral odors similarly as to the aversive odors (panel A in Figure 4—figure supplement 3, left peak in distribution); however, there appeared to be a second smaller group of neurons which responded to neutral odors similarly as they did to the rewarded odors. This was also true of D2 neurons in Block 1 which differentiated between the rewarded and aversive odors (see panel C, Figure 4—figure supplement 3). In Block 2, when the licking and outcomes were omitted, this distribution shifted to the right indicating that neutral odor responses tended to be an intermediate between aversive and rewarding odor responses or track more closely with the rewarded odor responses (see panel B, Figure 4—figure supplement 3). This suggests that the similarity in the responses to the aversive and neutral odors in Block 1 was likely in part due to the fact that the mice were not licking in response to the aversive and neutral odors, but were licking in response to the rewarding odors. The same trend is observed in the principal components of the D1 neuronal activity in Blocks 1 and 2 (Figure 5H in manuscript), where neutral odor responses trend with aversive odor responses in Block 1, but fall intermediate between rewarded and aversive odor responses in Block 2.

In determining the formula for the valence score calculation, we aimed to identify neurons which clearly differentiated between the two aversive and the two rewarding odors.

valence score= mean(differences\ between\ opposite\ valence\ odors)mean(differences between same valence odors)|maximum responseminimum response

Due to the variability in the neutral odor responses which often tracked closely with either the aversive odors or the rewarded odors in neurons which clearly differentiated between the rewarded and aversive odors, the neutral odor responses were not included in the numerator of the valence score formula. However, while the numerator of the equation does not incorporate neutral odor responses, the denominator would take into account neutral odor responses if they were either the minimum or maximum of those observed. We have added these points in the Methods section when describing valence scores, and present the panels in Figure 4 figure supplement 3.

2) In figure 3, the authors claim that for D1 but not D2 neurons, the valence score increases with training. Valence scores in naive mice are expected to be zero – this should be reflected in the data. To help interpretability, the distributions of valence scores with shuffled cell-odor pairs should be included as reference points.

To provide a point of comparison, we have now incorporated the Reviewer’s great suggestion of the valence scores calculated with shuffled odors in all of the valence score distribution plots (Figure 3, 4, 5). As the identity scores were calculated on the basis of the best odor response as compared to the next best one, the shuffling procedure does not affect the identity scores and was not done for those distributions. These are now shown in Figures 3F, 4F and 5FG, and mentioned in the text.

With regards to the naïve mice – we found that the mice learned the odor association task quickly, such that by the second half of the training session of the first day of training with the odors the mice had learned the associations (Figure 1C and D). Due to the very small number of trials during which the mice were exposed to the odors but had not yet learned the odors associations (< 10 trials per odor), we are unfortunately unable to calculate neuronal valence coding in the naïve state (see individual mice licking rates in Figure 1 figure supplement 3).

Furthermore, figure 1 suggests that there is a rage of anticipatory lick numbers on day one, likely reflecting differences in learning during the first session. When analyzing imaging data over time, the authors should compare data from mice with similar behavioral performance.

The added variability in the day 1 licking rates is in part due to the learning process, but is likely also because in two of the six Drd1-Cre mice we imaged, in the first training day the lick detection did not pick up any licks due to a malfunction (thus those two mice were not included in the day 1 mean). This only occurred on the first day of two D1 mice and not on any other training days or mice. The individual licking rates of the four remaining D1 mice (Figure 1 figure supplement 3) all indicate that the mice began to learn the odor associations within the first day to varying degrees. While we would have liked to have imaging data to analyze at a stage when the mice have not yet learned the odor associations, they learned the task too quickly within < 10 trials of the first session in most cases.

Based on the Reviewer’s suggestion, we plotted the anticipatory licking of each mouse separately across training days (Figure 1 figure supplement 3). We found that by day 2 all mice exhibited clear knowledge of the rewarded odors, and this was maintained through day 4. Note that the mice were not required to lick to receive the water reward, and there were no consequences for licking in response to the aversive and neutral odors. The absolute anticipatory licking rates varied slightly and were modulated by many factors beyond odor-outcome learning including the motivational level (thirst) of the mouse and overall tendency of the mouse to lick. We also confirmed that the mice differentiate between the small water reward predicting odor and the large water reward predicting odor. In day 4 of training, they did an average of 2.and anticipatory licks in response to odor 5 and an average of 2.0 anticipatory licks in response to odor 4 (p < 0.0001 Wilcoxon Rank Sum test, 30 trials each odor in 6 mice). This suggests that the mice learned the gradient of odor outcomes rather than simply categorizing odors into rewarding or aversive.

Due to the quick learning and ceiling effect of the behavior, we only compare neuronal activity in days 1 and 4 in our analysis. The three main neuronal imaging results we report (D1 and D2 neuron type comparison – Figure 2 and 3, Block 1 and sated Block 2 comparison – Figure 4, and sound-odor v odor-odor task comparison – Figure 5) are all from imaging experiments conducted at the end of the training. By day 4, all mice clearly differentiated between the rewarded and unrewarded odors (see day 4 in Figure 1—figure supplement 3) and we concluded that it is reasonable to combine the neuronal activity recorded from the mice. We show the individual mouse licking data in Figure 1 figure supplement 3.

3) In figure 4, the authors state that the mean valence score of D1 neurons remains above zero for block 2 trials – the mean valence score should be indicated in the figure. Statistical tests for comparing odor valence and identity distributions should be described in more detail.

Thank you for this suggestion. The mean valence score for D1 neurons in Block 1 is 0.39, the mean valence score for D1 neurons in Block 2 is 0.18 and there is a bimodal distribution of valence scores in Block 2 with one of the modes centered around zero and one of the modes centered around 0.4 similar to the mean in Block 1. The non-parametric two-tailed Wilcoxon rank sum test was used in all statistical comparisons of valence and identity score distributions. The Wilcoxon signed rank test indicates that the distribution of the overall valence scores of D1 neurons in Block 2 is significantly different from zero p = 2.8151e-13. Per the Reviewer’s suggestion, we also added the shuffled valence score distribution for the same neurons and find that the Block 2 D1 neuron valence score distribution is significantly different from the shuffled distribution (p = 2.1025e-12, Wilcoxon Rank Sum test). We have added this information in the main text in the Results section (rather than in the figure legends, to avoid clutter).

4) The motivation for the odor-tone association experiments described in figure 5 is unclear. Without characterizing tone-outcome learning in OT neurons in the described experimental setting, the multimodal task is difficult to interpret. Furthermore, the response properties of neurons in the 'full task condition' are hardly analyzed (supplementary figure 4), instead, the analysis focuses on block 2 trials and is difficult to follow.

The main result is that different subpopulations of OT neurons encode odor and tone associations. However, this does not appear to be a question the authors set out to address and may be a trivial result given the well-established differences in neuronal connectivity.

Thank you for raising this point which was not previously explained well. We had three motivations for conducting the sound-odor association experiment. First, it has previously been demonstrated that neurons in the tubercle can become activated in response to auditory tones (Varga and Wesson, 2013, Wesson and Wilson, 2010), so we wondered whether the valence related activation we observed in response to odors extended to auditory stimuli of the same valence as well. Second, as we learned that this was not the case, we found that the sound trials could serve as an additional control for checking whether the valence-related neuronal activity we observed was due to the instrumental licking response. The mice licked at similar rates in response to the rewarding odor and sound stimuli, and yet we did not observe the same similarity in neuronal responses to these odor and sound stimuli as we did to odors of the same valence in the five-odor task. This strengthens the argument that the odor valence-related activity we record in the original association task is not purely a result of the shared licking response to the two rewarded odors. Third, in the long term, we are interested in investigating the question of whether odor associations have unique qualities such as being more long-lived or more likely to elicit strong emotional responses due to the direct convergence of olfactory bulb input and dopaminergic input in limbic brain regions such as the olfactory tubercle. For this reason, we found it useful to use sound associations and related neuronal responses to gather evidence to begin to address this question. We clarify this in the manuscript in the Results section, when we first describe the odor-sound experiments and Figure 4.

The reviewer has correctly pointed out that in the main text we selected to focus on the neuronal imaging results in the final blocked version of the sound-odor association task. This was done primarily for the sake of brevity, to focus on the most direct comparison of the neuronal responses to the stimulus-only condition to the sounds and odors. In doing so, we could directly compare valence coding in the stimulus only condition in the odor-odor task and in the odor-sound task and eliminate the effects of the licking responses and the water/airpuff outcomes which were shared in the two tasks. We do display the distribution of the valence scores in the full-task condition in the odor-odor and odor-sound tasks in Figure 5F, the Block 1 and Block 2 neuronal activity principal components in Figure 5I, and the full-task neuronal activity in Figure 5 figure supplement 1. We also show the cosine similarity for the O-O and O-S task in Blocks 1 and 2 in Figure 5J.

5) It is unclear how odor identity (and sound frequency) is represented in neuronal response patterns in figures 2 and 4. The authors should show neuronal responses grouped by mice / odorant-unconditioned stimulus combinations, in addition to showing neuronal responses grouped by the trial outcomes only.

Thank you for this suggestion. To better illustrate the neuronal responses to specific odor-outcome pairings, we have added a supplementary figure with neurons grouped by mouse / odorant-outcome combination. The basic patterns of results hold for individual mice. These are shown in Figure 2 figure supplement 1 and Figure 5 figure supplement 1.

Reviewer #3 (Recommendations for the authors):

p. 2

The term "addictive" is not appropriate to describe intracranial self-administration of cocaine. Use "reinforcing" or "rewarding", instead.

We have changed in to “reinforcing”.

Suppl. Figure 1A

OT D1R neurons do not project to the substantia nigra or ventral tegmental area, or OT D2R neurons do not project to the globus pallidus. It is confusing to mention such projections that are most likely caused by viral diffusions to the nucleus accumbens and dorsal striatum. Clarify the points that the authors are making here.

We apologize for the confusion. The reviewer is correct in that the OT D1 and D2 neurons do not project to SNc/VTA and GP. This figure was included to validate the mouse models used – that the D1 and D2 mouse lines have differential projections, and the virus spillover in regions near the OT (for example, other ventral striatal regions) will allow us to trace projections to the SNc and VTA. However, this is needlessly confusing, and these mouse lines have been extensively validated previously, therefore we have removed this figure panel.

Figure 3A

Dotted circles used to describe their identifies are not shown on the actual diagram.

Apologies – we have added these now.

p.3

Licking response may not be labeled as "instrumental". Essentially, the study ued a Pavlovian conditioning procedure rather than operant conditioning procedure. Therefore, the conditioned licking response is a Pavlovian conditioned response.

We agree with this assessment – we have changed instances of the term “instrumental” to “licking” in the text.

p. 4

For clarity, rewrite the sentence that contains the phrase "non-zero licks were excluded."

OK. We have changed this to “any trials in which the mice licked were excluded”.

Rewrite the sentence "Odors 1 and 5, which... were combined with two sound tones". This sentence does not convey what the authors are trying to describe. The two modalities were combined in the new experiment; odors did not get combined with tones, which has a completely different meaning.

OK, we have changed this to: A new task was introduced in which two odor-outcome associations and two sound-outcome associations were used. The odors 1 and 5 were paired with strong airpuff and large water reward as previously, and two sound tones were used (5kHz and 12kHz) also paired with strong airpuff and large water reward.

p. 4 and Figure 5A,

Avoid the phrases "odor and sound pairs" and "odor-sound pairs". In the psychology/behavior conditioning literature, they mean that odors are paired with sounds.

This is a fair point. We have changed the text to remove pairs, and just refer to them as “odor and sound”.

Figure 5B

The plot suggests that by day 3, the mice may not have fully discriminated between sounds 1 and 2. The mice displayed anticipatory licks with sound 1 while they did not with odor 1.

We agree with the Reviewer that there is more anticipatory licking in response to the aversive sound 1 than to the aversive odor 1. However, the mice tend to lick marginally more both in response to the rewarded and aversive sounds – possibly due to the startling nature of the sound onset. In day 3, the mice clearly and strongly discriminated between sounds 1 and 2. They perform an average of 3.and anticipatory licks in response to sound 2 and 1.1 licks in response to sound 1 (p = 8.0914e-28, Wilcoxon rank sum test). The difference between the number of licks between sound 1 and sound 2 is ~2.6 licks. In comparison, the mean number of anticipatory licks for small rewarded odor 4 in the odor-odor task was 2.2 and the number of licks for aversive odor 1 was 0.08, for a smaller difference of 2.1 licks between the two stimuli. Therefore, we are sure that mice are indeed discriminating quite well between sounds 1 and 2. We present these numbers when describing the sound-odor experiments in the Results section.

Figure 5F-5G and Suppl. Figure 5B-5C

It is more informative to show data separately between O and S for the plots of the OS experiment.

We are not sure we understand the reviewer’s point. In the Odor-Sound experiment, there is no “O” valence score or “S” valence score, since there is only one odor and sound for each valence. The valence scores can only be calculated by comparing odors and sounds of the same vs opposite valence.

Suppl. Figure 4

Airpuff-paired sound triggered little or no conditioned response from the D1 neurons. What does this mean?

We thank the Reviewer for this interesting observation. While we can only conjecture, we believe the reason for this may be that the activity in response to the sound stimuli in the “full task” condition as shown in Suppl. Figure 4 may be dominated by the response of the mice to the stimuli and the outcomes rather than the sound stimuli themselves. In the case of the airpuff-paired sound, the mice perform little anticipatory licking and likely the most salient event occurs at the onset of the airpuff. This is also supported by the fact that the airpuff-onset related neuronal activity is similar in the aversive odor and aversive sound trials. In the case of the water-paired sound, the mice begin anticipatory licking after the onset of the sound and continue to do so when the water outcome is delivered. However, we also note the presence of a significant number of sound 2 responsive neurons in Block 2 of the odor-sound task when there was no licking, indicating that the neuronal activity in response to the rewarded sound is not only due to licking. Indeed, there were very few (2.and%) of D1 neurons which were preferentially activated to aversive sound stimulus in Block 2. It remains to be explored whether D1 OT neurons are more likely to respond to reinforced sounds than aversive sounds, but our data indicates that this may be the case. We really don’t have anything informative to add in the manuscript, so have not done so.

Suppl. Figure 5B-5C

The Y-axis is not labeled.

Sorry for the oversight, and thanks for noting this. Corrected.

p. 6

What do water- or airpuff-paired neural signals encode when they are present in the absence of behavior?

We are not sure what exactly the reviewer is asking, but here is an interpretation. Sensory signals are presumably converted/processed to allow motor outputs. In the earliest stages, the signals are likely to be largely sensory, and the latest stages signals will relate to motor acts (one has to be careful, of course, since plenty of recent evidence suggests that sensory areas have strong motor signals as well). In the intermediate regions, signals are likely to be mixed. We suggest that signals in the OT are strongly sensory-related, but modified substantially to allow for categorical predictions. Therefore, while behavior can affect these signals, there will still be significant sensory component even in the absence of behavior (after all the OT gets strong inputs from the OB and piriform cortex).

The last full paragraph on p. 6

The discussion is confusing. The authors mention that D1R neurons and D2R neurons have differential projections and that they have confirmed the fact in Fg. S1A. However, the authors do not discuss how projection patterns of the D1R and D2R neurons in the dorsal striatum and the nucleus accumbens are relevant to those of the OT. OT D1R and D2R neurons primarily project to the ventral pallidum (Heimer et al. 1987, Zhou et al. 2003); it is not clear how these neurons differentially affect downstream circuits.

Heimer L, Zaborszky L, Zahm DS, Alheid GF. 1987. The ventral striatopallidothalamic projection: I. The striatopallidal link originating in the striatal parts of the olfactory tubercle. J Comp Neurol 255: 571-91

Zhou L, Furuta T, Kaneko T. 2003. Chemical organization of projection neurons in the rat accumbens nucleus and olfactory tubercle. Neuroscience 120: 783-98

Once again, we apologize for the confusing presentation. As noted in the response above to the reviewer’s earlier comment, we do not mean to imply that OT neurons project to VTA/SNc etc. We used Supp Figure 1A to validate the mouse lines used, and we realize that this is confusing/misleading. We have removed Supp Figure 1A and its reference – instead we use the references suggested by the reviewer to make our differential projections point in the Discussion (paragraph 5).

The paragraph starting from the bottom of p. 6 and ending at the top of p. and

It is not coherently written. It is hard to follow the arguments.

We have tried to clarify this part, in Discussion (paragraph 6) now.

p. and

The authors state "D1 OT neurons selectively and bidirectionally encode learned odor valence, unlike D2 neurons". This statement sounds too strong. The sub-heading in the result section on p. 3 characterizes the results more sensibly as follows: "Individual D1 neurons are more likely to encode odor valence and D2 neurons more likely to encode odor identity".

We have modified this sentence to “D1 OT neurons are more likely to encode learned odor valence than D2 neurons, and conversely less likely to encode odor identity”.

References:

Barik, S. and de Beaurepaire, R. (1998). Hypothermic effects of dopamine D3 receptor agonists in the island of Calleja Magna. Potentiation by D1 activation. Pharmacol Biochem Behav 60, 313-9.

Cowley, B. R., Kaufman, M. T., Butler, Z. S., Churchland, M. M., Ryu, S. I., Shenoy, K. V. and Yu, B. M. (2013). DataHigh: graphical user interface for visualizing and interacting with high-dimensional neural activity. J Neural Eng 10, 066012.

Mansour, A., Meador-Woodruff, J. H., Bunzow, J. R., Civelli, O., Akil, H. and Watson, S. J. (1990). Localization of dopamine D2 receptor mRNA and D1 and D2 receptor binding in the rat brain and pituitary: an in situ hybridization-receptor autoradiographic analysis. J Neurosci 10, 2587-600.

Mengod, G., Villaro, M. T., Landwehrmeyer, G. B., Martinez-Mir, M. I., Niznik, H. B., Sunahara, R. K., Seeman, P., O'Dowd, B. F., Probst, A. and Palacios, J. M. (1992). Visualization of dopamine D1, D2 and D3 receptor mRNAs in human and rat brain. Neurochem Int 20 Suppl, 33S-43S.

https://doi.org/10.7554/eLife.75463.sa2

Article and author information

Author details

  1. Nuné Martiros

    Department of Molecular & Cellular Biology and Center for Brain Science, Harvard University, Cambridge, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Writing – original draft
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-9197-7711
  2. Vikrant Kapoor

    Department of Molecular & Cellular Biology and Center for Brain Science, Harvard University, Cambridge, United States
    Contribution
    Investigation, Methodology, Visualization, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0185-3836
  3. Spencer E Kim

    Department of Molecular & Cellular Biology and Center for Brain Science, Harvard University, Cambridge, United States
    Contribution
    Investigation, Methodology
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-3007-4417
  4. Venkatesh N Murthy

    Department of Molecular & Cellular Biology and Center for Brain Science, Harvard University, Cambridge, United States
    Contribution
    Conceptualization, Funding acquisition, Project administration, Resources, Supervision, Writing - review and editing
    For correspondence
    vnmurthy@fas.harvard.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-2443-4252

Funding

National Institutes of Health (F32DC017891)

  • Nuné Martiros

National Institutes of Health (R01DC017311)

  • Venkatesh N Murthy

National Institutes of Health (R01NS116593)

  • Venkatesh N Murthy

Harvard Brain Science Initiative (Bipolar Disorder Seed Grant)

  • Venkatesh N Murthy

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

Acknowledgements

We thank Dr Naoshige Uchida and Dr Mitsuko Uchida for their helpful guidance throughout this project and comments on the manuscript. We thank Dr Hao Wu for his help in establishing the Python-based, ScopeFoundry behavioral control system for the experiments. We also thank Selina Qian and Rebecca Fisher, for helping with animal colony maintenance and habituation, and helpful discussions. This work was supported by grants from the NIH (R01DC017311, R01NS116593, F32DC017891) and a Bipolar Disorder Seed Grant from the Harvard Brain Initiative.

Ethics

All experiments were conducted with approved protocols and in accordance with Harvard University Animal Care Guidelines. All of the animals were handled according to approved institutional animal care and use committee (IACUC) protocols (#29-20) of Harvard University.

Senior Editor

  1. Michael J Frank, Brown University, United States

Reviewing Editor

  1. Alicia Izquierdo, University of California, Los Angeles, United States

Reviewer

  1. Daniel W Wesson, University of Florida, United States

Publication history

  1. Preprint posted: November 2, 2021 (view preprint)
  2. Received: November 10, 2021
  3. Accepted: May 19, 2022
  4. Version of Record published: June 16, 2022 (version 1)

Copyright

© 2022, Martiros et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 809
    Page views
  • 151
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Nuné Martiros
  2. Vikrant Kapoor
  3. Spencer E Kim
  4. Venkatesh N Murthy
(2022)
Distinct representation of cue-outcome association by D1 and D2 neurons in the ventral striatum’s olfactory tubercle
eLife 11:e75463.
https://doi.org/10.7554/eLife.75463

Further reading

    1. Neuroscience
    Kevin J Miller et al.
    Research Article

    Humans and animals make predictions about the rewards they expect to receive in different situations. In formal models of behavior, these predictions are known as value representations, and they play two very different roles. Firstly, they drive choice: the expected values of available options are compared to one another, and the best option is selected. Secondly, they support learning: expected values are compared to rewards actually received, and future expectations are updated accordingly. Whether these different functions are mediated by different neural representations remains an open question. Here we employ a recently-developed multi-step task for rats that computationally separates learning from choosing. We investigate the role of value representations in the rodent orbitofrontal cortex, a key structure for value-based cognition. Electrophysiological recordings and optogenetic perturbations indicate that these representations do not directly drive choice. Instead, they signal expected reward information to a learning process elsewhere in the brain that updates choice mechanisms.

    1. Neuroscience
    Bin Wan et al.
    Research Article Updated

    The human cerebral cortex is symmetrically organized along large-scale axes but also presents inter-hemispheric differences in structure and function. The quantified contralateral homologous difference, that is asymmetry, is a key feature of the human brain left-right axis supporting functional processes, such as language. Here, we assessed whether the asymmetry of cortical functional organization is heritable and phylogenetically conserved between humans and macaques. Our findings indicate asymmetric organization along an axis describing a functional trajectory from perceptual/action to abstract cognition. Whereas language network showed leftward asymmetric organization, frontoparietal network showed rightward asymmetric organization in humans. These asymmetries were heritable in humans and showed a similar spatial distribution with macaques, in the case of intra-hemispheric asymmetry of functional hierarchy. This suggests (phylo)genetic conservation. However, both language and frontoparietal networks showed a qualitatively larger asymmetry in humans relative to macaques. Overall, our findings suggest a genetic basis for asymmetry in intrinsic functional organization, linked to higher order cognitive functions uniquely developed in humans.