Encoding of odor information and reward anticipation in anterior cortical amygdaloid nucleus

Summary

Olfactory information directly reaches the amygdala through the olfactory cortex, without the involvement of thalamic areas, unlike other sensory systems. The anterior cortical amygdaloid nucleus (ACo) is one of the olfactory cortices that receives olfactory sensory input and is part of the olfactory cortical amygdala, which relays olfactory information to the amygdala. To examine its electrophysiological features, we recorded individual ACo neurons during the odor-guided go/no-go task to obtain a water reward. Many ACo neurons exhibited odor-evoked go cue-preferred activity during the late phase of nose-poking supporting the population dynamics that differentiate go/no-go responses before executing the odor-evoked behaviors. We observed two types of neurons with different anticipation signals: one neuron type exhibited gradual increases of activity toward reward delivery, while another type exhibited a phasic go cue-preferred activity during nose-poking as well as another phasic anticipatory activity for rewards. These results suggest that the ACo may be involved in reward-related behavioral learning by associating the olfactory information with reward anticipation.

Graphical abstract

Highlights

• •Late-phase nose-poking activity reveals go cue-preferred responses in ACo neurons
• •Population dynamics in the ACo distinguish go versus no-go actions before behavior
• •ACo neurons comprise two types that show gradual or phasic reward anticipation
• •Phasic-type ACo neurons that anticipate rewards also respond to go cues

Behavioral neuroscience; sensory neuroscience

Introduction

Olfaction is closely related to emotion in attributing positive (attractive) or negative (aversive) valence to the environment more than any other sensory modality.^1,2 The close bidirectional connection and the particular organization of the olfactory cortex to the amygdala distinguishe the olfactory system from other sensory systems.^3,4 Afferent sensory inputs from the main olfactory bulb (OB) project to the olfactory cortex, which in turn sends direct projections to the amygdala. In contrast, afferent inputs from most other sensory systems enter the amygdala via the thalamus and neocortical regions.⁵ OB mitral cells project their axons through the lateral olfactory tract to the olfactory cortex.⁶ Olfactory cortical amygdala, which is a part of the olfactory cortex, relays olfactory information to the amygdala. The anterior cortical amygdaloid nucleus (ACo) is a part of the olfactory cortical amygdala, and has a bidirectional connection with the amygdala.

A study reported that the ACo receives dense projections from the main OB, moderate projections from the piriform cortex, lateral entorhinal cortex, basomedial amygdaloid nucleus (BMA), and medial amygdaloid nucleus (Me), and scarce projections from the ventral tegmental area (VTA) and the ventral tenia tecta (vTT).³ Moreover, the ACo projects densely to BMA.³ These anatomical studies indicate that the ACo is closely related to the amygdala, and it is possible that the ACo is involved in odor-evoked motivational behaviors.

A behavioral study revealed that ACo participates in olfactory fear conditioning in rats as electrical stimulation of the OB induces evoked field potential signals that are persistently potentiated specifically in the ACo after training.⁷ Moreover, electrical stimulation of the VTA showed that the ACo, besides other mesolimbic structures, displays increased Fos expression in rats.⁸ A whole-cell patch clamp study showed that with the activation of sodium conductance, pyramidal neurons of the ACo displayed rhythmic fluctuations of intrinsically generated voltage-dependent membrane potential in the theta-low beta range, suggesting that the ACo was related to synaptic plasticity and learning.⁹ ACo has been poorly investigated, but comprehensive evidence suggests that it may play a prominent role in reward-related behavioral learning by olfactory stimulation. However, little is known about the electrophysiological features of the ACo neurons for reward-related behavioral tasks.

Here, we recorded the neural activity of ACo neurons during odor-guided reward-directed behaviors. Many ACo neurons responded to the go-cue odor stimulus at the late phase of the nose-poking epoch (from the odor valve off to the odor port exit). The ACo neuron population showed profound and persistent transformations in the dynamics of cue encoding over 400 ms after odor onset. Furthermore, we found that the ACo neuron groups each coded a different type of anticipation signal: one neuron group type exhibited gradual increases in the signals to the reward, while the other type showed phasic anticipation signals with the go-cue preference responses during nose-poking. Our results suggest that the ACo neurons may play an important role in odor-guided reward-directed learning.

Results

Go-cue-odor preferred responses of ACo neurons during the late phase of nose-poking epoch

We recorded 276 well-isolated neurons in the ACo of seven mice performing an odor-guided go/no-go task (Figure 1A; an example of a recording position is illustrated in Figure 1B; Figure S1A; Table 1). Briefly, the go trial requires the mice to first sample a go-cue odor presented at an odor port and then to move to a reward port to receive water reward. Conversely, the no-go trial requires the mice to first sample a no-go-cue odor presented at the odor port and then to stay near it to wait for the next trial. It is important to note that the mice were required to keep their nose inserted into the odor port, at least during odor presentation (500 ms). After the mice were well trained, their behavioral accuracy remained above 80% throughout the session. For all mice, the median duration of the nose-poking epoch (the time from odor valve opening until the mouse withdrew its snout from the odor port) was 927 ms (interquartile range: 745–1,123 ms) in the go trials, and 758 ms (interquartile range, 660–910 ms) in the no-go trials (46 sessions from seven mice).

Figure 1.

Cue-odor-preferred responses of ACo neurons during the late phase of nose-poking in the odor-guided go/no-go task

(A) Time course of the odor-guided go/no-go task. Behavioral epoch temporal progression from left to right.

(B) Nissl-stained frontal section (an arrow indicates recording track) and recording tracks (vertical thick lines) of the ACo. NLOT, nucleus of lateral olfactory tract. CxA, cortex-amygdala transition zone. APC, anterior piriform cortex. Scale bars: 500 μm.

(C) Example firing patterns of ACo neurons during nose-poking epoch (the time from odor valve opening to odor port exit) in the odor-guided go/no-go task. Each row contains spikes (black ticks) for one trial, aligned to the time of odor valve opening (corresponding to odor port entry, orange ticks). Red ticks represent times of odor port exit. Correct trials are grouped by odor, and within each group, are sorted by the duration of the nose-poking epoch (50 selected trials from the end of the session are shown per category). Histograms are averaged across odors, and calculated using a 20 ms bin width and smoothed by convolving spike trains with a 60 msec-wide Gaussian filter (blue, go-cue odor; green, no-go-cue odor). Vertical dashed lines indicate the time of odor valve opening.

(D) Normalized firing rates (auROC values) for go-cue selective neurons (n = 120). auROC values (aligned by odor valve opening) were calculated by go-cue odor presentation versus baseline (left) and no-go-cue odor presentation versus baseline (right) in the sliding bins (width, 100 ms; step, 20 ms). Yellow, increase from baseline; blue, decrease from baseline. Each row corresponds to one neuron, with neurons in the left and right graphs in the same order. Neurons are sorted by the peak time for auROC values calculated by go-cue odor presentation versus baseline.

(E) Cue preference curves (auROC values, go-cue versus no-go-cue odor presentation, aligned by odor valve opening, odor port exit) for go-cue selective neurons. Each row corresponds to one neuron, with neurons in the left and right graphs in the same order of (D). Color scale indicates significant preferences (p < 0.01, permutation test; positive values correspond to the go-cue preferred responses). The black boxes indicate bins with non-significant preferences (p > 0.01, permutation test).

Table 1.

Number of recorded neurons per mouse per session

Session	Mouse#1	#2	#3	#4	#5	#6	#7
1	8	1	9	5	10	14	5
2	5	N/A	8	N/A	6	11	9
3	4	N/A	10	N/A	10	1	4
4	3	N/A	5	N/A	13	5	6
5	3	N/A	6	N/A	10	3	2
6	1	N/A	4	N/A	5	5	N/A
7	N/A	N/A	7	N/A	3	5	N/A
8	N/A	N/A	10	N/A	7	2	N/A
9	N/A	N/A	5	N/A	18	5	N/A
10	N/A	N/A	7	N/A	4	1	N/A
11	N/A	N/A	3	N/A	N/A	5	N/A
12	N/A	N/A	N/A	N/A	N/A	3	N/A
Total	24	1	74	5	86	60	26

Since the ACo receives direct inputs from the mitral cells of the OB, we first focused on whether ACo neurons exhibited cue-odor selective activity during nose-poking epochs (from odor poke-in to odor poke-out). We found that a subset of ACo neurons increased their firing rates during the odor presentation phase (0–500 ms after the odor valve opening) during both go and no-go trials, and then showed a go-cue-odor preferred response 500 ms after the odor onset (an example shown in Figure 1C). To quantify the dynamics of the cue-encoding, we calculated the firing rate changes from baseline (200–0 ms before the end of the inter-trial interval) in the sliding bins during the nose-poking epoch for each neuron. For each correct trial, we calculated the area under the receiver operating characteristic curve (auROC) value at each time bin (width, 100 ms; step, 20 ms), and defined the go-cue selective neurons (n = 120 neurons, 43.5% of the recorded neurons) as those neurons that significantly increased their firing rates from the baseline (p < 0.01, permutation test) for five consecutive bins (100 ms) during the nose-poking epoch in the go correct trials (Figure 1D). Across the go-cue odor-selective population, calculation of go-cue versus no-go-cue preferences during nose-poking epochs clearly showed a go-cue preference manner from 500 ms after the odor onset to the odor poke-out (late phase of the nose-poking epoch) (Figure 1E, p < 0.01, permutation test). In addition to the go-cue-preferring neurons, we also analyzed the no-go-cue-preferring neurons. Their number was comparable to that of the go-cue-preferring population; however, their odor-evoked responses were generally weaker and showed no prominent modulation at the population level (Figures S1B–S1E). Together, these results provide a more complete characterization of the cue-odor selectivity in ACo neurons. To further examine whether these neuronal responses depended on odor identity or task context, we compared data from two groups of mice with reversed odor-reward contingencies and visualized the auROC values for each condition as color maps (Figures S1F–S1I). The results showed that the temporal profiles and response magnitudes of ACo neurons were similar between the two groups, despite the opposite odor-reward contingencies. These results suggest that the ACo received not only a particular odorant profile directly from the OB but rather the complex odor information, including behavioral contexts from other olfactory cortical areas and top-down inputs from higher areas.

Late phase of go-cue odor preferred responses was evoked by the odor onsets and was stable across trials

The go-cue odor-selective population showed cue-odor-preferred responses during the late phase of the nose-poking epoch (Figure 1E). It is possible that the late phase of odor-preferred responses was tuned to the odor port exit behaviors or contained the premotor signals that were observed in many brain regions.^10,11,12 To take these signals into account, and to help isolate signals related to odor presentation and action, we developed an encoding model (generalized linear model, GLM). This model incorporated task-related variables during the nose-poking epoch as predictors of each neuron’s activity (Figure 2A; Figures S2A–S2C).¹³

Figure 2.

Late go-cue odor-preferred responses were evoked by the odor onsets and were stable

(A) Schematic of the encoding model used to quantify the relationship between behavioral variables and the activity of each neuron (see STAR Methods). Behavioral predictors for the odor stimulus-presentation epoch are supported over the window 0 to 500 ms relative to the onset of odor valve, as well as late phase of nose-poking epoch that are supported over the window 0 to 427/258 ms relative to the offset of odor valve in either go/no-go trials (median of the valve offset to odor port exit), and pre-odor-port exit epoch that are supported over the window −300 to 0 ms relative to the odor port exit. Inset, predicted and actual averaged firing rate relative to the odor onset and odor port exit for one neuron.

(B) Top: relative contribution of each behavioral variable to the explained variance of the neural activity, averaged across the go-cue-selective neurons. All error bars represent the standard error of the mean. Bottom: relative contribution significance of the late phase of go-cue nose-poking variable; see Figure S2F for the other variables.

(C) Go-cue odor-preferred responses during correct trials, error trials, and catch (odorless) trials. The auROC values were calculated during the nose-poking epochs and only neurons with a minimum number of five trials for each analyzed condition were included in this analysis. Black horizontal lines and black vertical lines indicate medians and interquartile ranges. The statistical significance among five groups (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001) was assessed by one-way analysis of variance (ANOVA) with Tukey’s post hoc test.

(D) The development of go-cue responses in go-cue-selective neurons during learning. For each go-cue-selective neuron, we calculated the correlation between the firing rate during the go-cue nose-poking epoch relative to the baseline (a mean firing rate during inter trial interval was subtracted for each neuron) and the order of go trial from the start of the session. The correlation coefficient was compared with control values calculated by the 1,000 trial-shuffled data (gray shaded area) and then the statistical significance was determined (<0.5^th percentiles of the control values, negative correlation; >99.5^th percentiles of the control values, positive correlation). Across go-cue-selective neurons, the majority of the go-cue responses were not correlated with trial progression (79.2%, not significant; 10.0%, negative; 10.8%, positive).

Using this encoding model, we quantified the relative contribution of each behavioral variable during the nose-poking epoch to the response of each neuron by determining how much the explained variance declined when that variable was removed from the model (see STAR Methods; Figures S2D–S2F; a relative contribution for an example neuron is shown in images at the bottom part of Figure 2A). Averaged across the go-cue odor selective population, the highest relative contribution during nose-poking epochs was attributed to late go-cue nose-poking (26.1 ± 1.8% of the total variance explained during the nose-poking epoch), followed in descending order by the go-cue odor presentation (24.9 ± 1.6%), the no-go-cue odor presentation (22.6 ± 1.6%), pre-no-go behavior (11.1 ± 1.3%), the late no-go-cue nose-poking (6.2 ± 0.9%), and pre-go-behavior (9.1 ± 1.5%) (bars at the top of Figure 2B). The bottom of Figure 2B shows the breakdown of the late go-cue nose-poking contribution, illustrating the proportions of neurons whose neural responses contributed positively, negatively, or non-significantly to the model prediction. Notably, 60.0% of go-cue odor-selective neurons exhibited significantly positive contributions during this period (a pie chart in the bottom of Figure 2B). The pre-behavior window (300 ms) was defined based on the perievent time histogram, which showed preparatory neural activity approximately 300 ms before the behavioral response, consistent with a previous study.¹⁴ To confirm that our conclusions were not affected by this specific choice, we additionally performed GLM analyses using shorter pre-behavior windows (100 and 200 ms) and compared the relative contributions of behavioral variables. The results revealed no significant differences across the three window lengths (Figure S3A), indicating that our conclusions are robust to the choice of pre-behavior window size. To further assess whether the late-phase activity was influenced by differences in nose-poking duration, we examined the correlation between sampling duration and late-phase firing rate across trials. The correlation coefficients were very small and not significant for both Go (r = −0.007, p = 0.257) and No-Go (r = 0.010, p = 0.108) trials, indicating that the late-phase activity was not simply a consequence of longer odor-port sampling (Figures S3B and S3C). Furthermore, across the population, the go-cue responses during nose-poking epochs in both correct and error trials were higher than those in the no-go-cue correct and odorless trials (Figure 2C), suggesting that the go-cue excitation responses mainly reflected signals of encoding cue-odor information. Notably, the firing rates of the majority of the go-cue responses remained stable across trials (Figure 2D). To ensure that the late-phase activity was not driven by a minority of neurons exhibiting trial-by-trial variability, we separately analyzed the subpopulation (20%) showing significant firing-rate changes across trials using a GLM. The late-phase contributions did not differ from those of the stable neurons (Figure S3D), indicating that the late-phase activity primarily reflected stable cue-odor encoding rather than variability across trials. Taken together, the go cue-preferred responses during the late phase of the nose-poking epoch were considered to reflect the go-cue odor information.

Response dynamics of the ACo neuron population during the late phase of nose-poking epoch

We demonstrated that ACo neurons showed odor-evoked cue-preferred responses during the late phase of the nose-poking epoch (Figures 1 and 2). Were the distinct cue responses reflected in the ACo neuron population dynamics, and how much could the population activity account for animals’ behavioral accuracy? First, to gain insight into the dynamics of the population response, we visualized average population activity using principal component analysis, a dimensionality reduction method. Figure 3A shows trajectories of the mean response of the ACo neuron population to go-cue and no-go-cue odors, represented as projections onto the first three principal components (PCs) during the nose-poking epochs. Throughout the approximately 300 ms interval from the odor onset, trajectories remained converged, showing little difference across conditions. Over the late phase of nose-poking epochs, trajectories in the nose-poking epoch subspace began to spread out and were clearly separated at the population level. To quantify these observations, we measured the instantaneous separation between the population cue responses (Figure 3B). The separation started to increase from 300 ms after odor onset, reaching a maximum at ∼500 ms, and remained above baseline levels until odor port exit. Thus, the ACo neuron population showed profound and persistent transformations in the dynamics of cue-encoding, 300 ms after odor onset.

Figure 3.

Dynamics of ACo neuron population response during the late phase of nose-poking

(A) Visualization of ACo neuron population responses during nose-poking epoch using principal component analysis (276 neurons). The responses to cue odors are projected onto the first three principal components corresponding to nose-poking epoch subspaces. Blue line, go-cue odor; green line, no-go-cue odor. Temporal progression from unfilled blue/green spheres to filled spheres.

(B) Distance between ACo neuron population responses. Gray line and shaded area show mean ± 2SD baseline values during pre-nose-poking epoch. Top dots indicate time bins showing values more than mean + 2SD baseline values.

(C) The time course of odor decoding accuracy. A vector consisting of instantaneous spike counts for 1–276 neurons in a sliding window (width, 100 ms; step, 20 ms) was used as input for the classifier. Training of the classifier and testing were done at every time point. Green horizontal lines indicate the level of animal behavioral performance. Gray horizontal lines indicate chance level (50%). Green vertical dashed lines indicate the first points at which the decoding accuracy reached the level of animal behavioral performance. Shaded areas represent ±SD.

Second, to examine whether the population activity accounted for the animals’ behavioral accuracy, we performed a decoding analysis to determine whether the firing rates of the ACo neuron populations could be used to classify each individual trial as go or no-go. We used SVMs with linear kernels as a decoder. Based on ACo neurons, analyses of the decoding time course, using a sliding time window, revealed that decoding accuracy was first maintained at chance levels until 400 ms after the odor onset, and then increased above the behavioral accuracy level of the animals around 500–600 ms after odor onset (the top-right graph in Figure 3C). In the 500–600 ms period, about 90 recorded neurons were sufficient to account for the behavioral accuracy (the bottom right graph in Figure 3C). Thus, a hundred ACo neurons accounted for animals’ behavioral accuracy in the late phase of nose-poking.

Two types of reward anticipation responses of ACo neurons

We then focused on the ACo activity during odor-evoked behaviors after a nose-poking epoch. A subset of ACo neurons gradually increased their firing rates from the time of water port entry till the reward was received, and another subset of neurons increased their firing rates while waiting for reward (examples shown in Figure 4A). We quantified the data by calculating firing rate changes from baseline (spike data were aligned to the water port entry), and three measures from the values: “time of center of mass,” “onset time,” and “duration” (from water port entry to 1,000 ms after opening the water valve, Figure 4B, see STAR Methods). The time of center of mass corresponds to the weighted average of significant time points, the duration is the time window around the center of mass where significance was sustained for at least five consecutive bins, and the onset time is when this duration first began. The drinking epoch selective neurons (n = 38, 13.8% of the recorded neurons) were defined as neurons that had the time of center of mass during the drinking epoch, and the waiting epoch selective neurons (n = 29, 10.5% of the recorded neurons) were defined as neurons that had the time of center of mass during the waiting epoch. Across the population, the drinking-epoch-selective neurons gradually increased their firing rates −180 ms before the water valve opened for 405 ms, and the waiting-epoch-selective neurons increased their firing rates 10 ms after water port entry for 118 ms (Figures 4C and 4D, p < 0.01, permutation test). Thus, ACo neurons exhibited two distinct types of reward anticipation responses.

Figure 4.

Two types of reward-related responses of ACo neurons

(A) Example firing patterns of reward-related responses. Spike histograms are calculated using a 20 ms bin width and smoothed by convolving spike trains with a 60 ms wide Gaussian filter. A vertical black line indicates the water valve opening.

(B) Evaluation of the reward-related responses. Normalized firing rates (auROC values) were calculated by go-behavior versus baseline in the sliding bins (width, 100 ms; step, 20 ms). Left: red bars show significant excitation (p < 0.01, permutation test). Based on the significant time points, onset time (black triangle), time of center of mass (black circle) and duration (black horizontal line) were calculated. Vertical black lines indicate the water valve opening. Right: each row corresponds to one neuron and neurons are sorted by times of center of mass (white dots) of auROC values. Based on the times of center of mass, drinking-selective neurons and waiting-selective neurons were defined (a horizontal dashed line). Color scale as in Figure 1D. Vertical white lines indicate the water port entry and the water valve opening.

(C) The proportions of neurons that exhibited a significant response, calculated from auROC values (p < 0.01, permutation test) for each neuron group (orange, drinking selective neurons; brown, waiting selective neurons). Vertical black lines indicate the water valve opening.

(D) Distributions of the times of center of mass, onset times and durations (orange, drinking-selective neurons; brown, waiting-selective neurons).

Association of go-cue excitations with excitatory responses for the reward anticipation behavior

We observed that the waiting-epoch-selective neurons showed go cue-preferred activity during the nose-poking epoch; however, the drinking-epoch-selective neurons did not (examples shown in Figure 5A). To examine the relationship between the reward anticipation responses and cue encoding, we quantified the response profiles of each neuron group during odor-evoked behaviors by calculating the firing rate changes from baseline (Figure 5B). Across the population, drinking-epoch-selective neurons showed significant excitatory responses for the waiting and drinking epochs (red histogram at the top in Figure 5B, p < 0.01, permutation test), and significant inhibitory responses for other behavioral epochs (blue histogram at the top in Figure 5B, p < 0.01, permutation test). Alternatively, waiting-epoch-selective neurons showed significant excitatory responses for the late phase of nose-poking in the go-cue and waiting epochs (red histogram at the bottom in Figure 5B, p < 0.01, permutation test), and significant inhibitory responses for the drinking and no-go waiting epochs (blue histogram at the bottom in Figure 5B, p < 0.01, permutation test). The waiting-epoch-selective neurons showed higher responses during the nose-poking phase in the go-cue odor trials than those of other groups (Figure 5C, one-way analysis of variance with Tukey’s post hoc test). Thus, waiting-epoch-selective neurons exhibited associations between the go-cue excitations and excitatory responses for waiting behavior, suggesting that a subset of ACo neurons was involved in cue-outcome associations.

Figure 5.

Waiting-selective neurons also showed go-cue odor-preferred responses during nose-poking

(A) Example firing patterns of cue-outcome responses. Spike histograms are calculated using a 20 ms bin width and smoothed by convolving spike trains with a 60 ms wide Gaussian filter. A vertical black line indicates the water valve opening.

(B) The proportions of neurons that exhibited significant excitatory and inhibitory response calculated from auROC values (p < 0.01, permutation test) for each neuron group. Vertical black lines indicate the water valve opening.

(C) auROC values during nose-poking epoch of go-cue odor-selective responses (top graph) and no-go-cue odor-selective responses (bottom graph) for each neuron group. Black horizontal lines and black vertical lines indicate medians and interquartile ranges. Red dots, significant excitation; blue dots, significant inhibition; gray dots, non-significant (p < 0.01, permutation test). Statistical significance among three groups (∗p < 0.05, ∗∗∗p < 0.001) was assessed by one-way analysis of variance (ANOVA) with Tukey’s post hoc test.

Discussion

The purpose of the study was to understand the electrophysiological features of ACo neurons on odor-evoked reward-related behavioral tasks. We found that many ACo neurons exhibited go-cue odor-preferred responses at the late phase of the nose-poking epoch (Figures 1 and 2). Consequently, the ACo population showed profound and persistent transformations in the dynamics of cue encoding, and provided sufficient information to account for the behavioral performance before executing the odor-evoked behaviors (Figure 3). In addition, we found two types of reward anticipation signals during the odor-evoked behaviors: ramp-like gradual increases in reward-related activity exhibited by drinking-epoch-selective neurons, and phasic anticipation signals exhibited by waiting-epoch-selective neurons (Figure 4). The waiting-epoch-selective neurons exhibited associations between the go-cue excitations and excitatory responses for the waiting behavior (Figure 5). Thus, the ACo showed unique encodings during various behavioral states in the task, suggesting that the ACo neurons play an important role in reward-related learning evoked by olfactory stimulus.

Odor representation of the ACo

The ACo was a previously unexplored area in the olfactory amygdala located caudally to the lateral olfactory tract and rostromedially to the posterolateral cortical nucleus of the amygdala. The projection from the OB terminates in the outer portion of the most superficial layer (layer Ia) of the cortex, and the projection from other olfactory cortex areas terminates in the deep portion of the ACo.^15,16 Unlike other sensory systems, olfactory information from the external world reaches the amygdala without passing through thalamic areas. Therefore, the olfactory amygdala, including the ACo, receives olfactory information from the OB via the olfactory cortex.^17,18 Moreover, previous comprehensive indirect evidence suggests that ACo may play a prominent role in reward-related behavioral learning from olfactory stimuli.¹⁹ However, little is known about the functions of the ACo in olfactory information processing. We performed in vivo recordings from ACo neurons during the odor-guided go/no-go task to obtain a water reward. ACo neurons exhibited cue-odor-preferred responses at the late phase of nose-poking epochs (Figures 1, 2, and 3). The peak firing of ACo neurons during nose-poking epochs occurred later than that observed in other olfactory cortical areas, such as the nucleus of the lateral olfactory tract (NLOT)¹⁴ and the vTT.²⁰ However, this late-phase coding is unlikely to result from a delay in odor delivery. Many ACo neurons exhibited clear firing activity immediately after odor onset during the odor presentation period (Figure 1D), suggesting that the delay between valve opening and odor arrival at the nose was minimal. Consistent with this, our previous studies used the same odor delivery and recording system, as well as the same behavioral apparatus and task, allowing direct comparison with these other olfactory cortical areas. Under identical experimental conditions, odor-evoked responses in the NLOT and vTT appeared within 50–300 ms after valve opening.^14,20 Moreover, a previous behavioral experiment demonstrated that when the duration of odor presentation was shortened without restricting nose-poking, mice showed shorter reaction times while maintaining behavioral accuracy above 80%.¹⁴ These findings indicate that odor arrival occurred within approximately 400 ms after valve opening and that the observed late-phase activity in ACo neurons cannot be explained by delayed odor delivery or prolonged sampling. Therefore, the delayed peak firing in ACo most likely reflects genuine differences in neural processing compared with other olfactory cortical areas.^14,20 Moreover, although premotor influences cannot be completely excluded, several features of the activity pattern suggest that the late-phase coding in ACo neurons is not primarily driven by motor preparation (Figure 2). In particular, the maintenance of cue selectivity and the structured relationship with subsequent reward-expectation activity are not readily explained by generic premotor signals. The population coding of ACo neurons showed profound and persistent transformations in the dynamics of cue-encoding, 400 ms after the odor onset (Figure 3). This may reflect the input of other olfactory cortex from layers Ib, II, and III, and the top-down inputs from other brain areas, rather than the direct sensory inputs from OB in layer Ia. Therefore, we speculate that ACo neurons send task-modulated olfactory information to other amygdala areas.

Role of reward-anticipation response in the waiting epoch

ACo sends axons massively to the BMA.³ The BMA plays a role in learning and storing CS-US associations.¹⁹ Therefore, the projections from the ACo to the BMA may send olfactory information for the CS-US association in the odor-guided learning tasks. We demonstrated that some neurons exhibited reward-anticipation responses in the waiting epoch and also showed a go-cue-odor preference (Figure 5). Since the ACo receives direct inputs from the BMA,³ it may play a role within this reciprocal interaction in linking olfactory information with reward anticipation, thereby contributing to CS-US associations.

A previous study revealed that pyramidal neurons of the ACo displayed rhythmic fluctuations of the intrinsically generated voltage-dependent membrane potential in the theta-low beta range with the activation of sodium conductance.⁹ Synchronized theta oscillations have been found to increase between regions when enhanced communication occurs during memory acquisition^21,22 and goal selection.^23,24 Oscillatory synchronization for CS occurs in the theta band, between the lateral entorhinal cortex (LEC), which is a part of the olfactory cortex, and the medial prefrontal cortex (mPFC), during the learning stage of trace CS-US associative learning tasks.²⁵ The ACo has bidirectional connections with the LEC.³ We speculate that the ACo additionally drives the LEC-mPFC circuit and supports the CS-US association by synchronizing the ACo-LEC-mPFC theta oscillations during the learning stage.

Role of reward-anticipation response in the drinking epoch

In learning, reward signals have very important implications.²⁶ A subset of ACo neurons increased their firing rate during the drinking reward epoch (Figure 4). These neurons started to increase their firing rate before the drinking epoch, and these activities persisted during the epoch. Previous studies reported that similar firing patterns were observed in the dopamine neurons in the VTA. Post-learning, dopamine activity may change phasic responses to cues and rewards, and ramping activity may arise as the agent approaches the reward.²⁷ The ACo receives direct inputs from the VTA.³ It is assumed that the ramping-like response in the ACo may reflect the inputs from the VTA. In addition, the ACo has anatomical connections with other olfactory cortices.³ We speculate that the VTA reward signals may be transmitted to other olfactory cortical areas via ACo, making learning more efficient in the olfactory cortex.

A previous behavioral study revealed that electrical stimulation of the VTA caused the ACo, besides other mesolimbic structures, to display increased Fos expression in rats.⁸ ACo sends excitatory glutamatergic/aspartatergic projections to the nucleus accumbens (NAc).²⁸ Dopamine (DA) projections from the VTA to the NAc, which constitute the mesolimbic DA system,^29,30,31 play an essential role in motivated behaviors, reinforcement learning, and reward processing.^32,33,34 Therefore, the ACo may assist in driving the NAc-VTA circuit for reward-related behavior.

Limitations of the study

A primary limitation of this study is that the odor-evoked responses observed in ACo neurons cannot be clearly attributed to direct feedforward input from the OB. Because the ACo receives substantial associative and modulatory inputs from other olfactory cortical areas, as well as top-down signals from higher-order regions, it remains uncertain whether its responses primarily reflect sensory-driven activity or sensory signals already transformed by contextual or learned factors. Dissecting the pathway-specific contributions to ACo odor encoding will require circuit-level approaches in future work.

Another limitation is that neuronal activity was examined within a single behavioral paradigm. Although the go/no-go task allowed us to identify robust cue-selective and anticipatory responses, it remains unclear how generalizable these response patterns are across tasks with different cognitive or motivational demands. Behavioral manipulations that vary contingency, context, or reward structure will be necessary to determine the flexibility of ACo representations.

Finally, the present study does not establish causal relationships between ACo activity, context-dependent motivated behavior, and the function of its downstream projections. The ACo sends distinct outputs to regions such as the basomedial amygdala and the lateral hypothalamus, yet their respective contributions to odor-guided decision-making cannot be inferred from electrophysiology alone. Projection-specific perturbations and cell-type-resolved recordings, such as optogenetic or imaging-based approaches, will be essential to clarify how ACo circuits interact with downstream structures to support goal-directed odor-guided behaviors.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Hiroyuki Manabe (hmanabe@naramed-u.ac.jp).

Materials availability

This study did not generate unique reagent.

Data and code availability

• •Adjective data reported in this article will be shared by the lead contact upon request.
• •This article does not report original code.
• •Any additional information required to reanalyze the data reported in this article is available from the lead contact upon request.

Acknowledgments

We thank Nozomi Fukui from Doshisha University, Shingo Fukumoto, Kengo Ehara, and Nana Kitamura from Ritsumeikan University for assistance with data collection. We appreciate the Sakurai lab members from Doshisha University for valuable discussions. We also thank Hideki Tanisumi for providing illustrations in the figures. K.S. was supported by JSPS KAKENHI grant numbers 18J21358, 21K06440, 22K15232, 24K18244, Lotte Research Promotion grant, Takeda Science Foundation, Life Science Foundation of Japan, and Kobayashi Foundation. H.M. was supported by the Takeda Science Foundation, the PRESTO, Japan Science and Technology Agency grant number JPMJPR21S9, and JSPS KAKENHI grant numbers 25135708 and 16K14557. Y.S. was supported by JSPS KAKENHI grant numbers 20H00109 and 20H05020.

Author contributions

K.S., Y.T., and H.M. designed the experiments, and K.S., Y.T., and H.M. performed the experiments. K.S., Y.T., J.H., and H.M. performed the data analysis. K.S., Y.T., and H.M. wrote the manuscript. Y.S. supported and advised the project.

Declaration of interests

No conflicts of interest, financial or otherwise, are declared by the authors.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals, peptides, and recombinant proteins
Amyl acetate	Tokyo Chemical Industry Co., Ltd.	N/A
Eugenol	Tokyo Chemical Industry Co., Ltd.	N/A
Experimental models: Organisms/strains
Mouse:C57BL/6NCrSlc	Shimizu Laboratory Supplies Co., LTD.	RRID:MGI:5658686
Software and algorithms
MATLAB	The Mathworks, Inc.	RRID:SCR_001622; http://www.mathworks.com/products/matlab/
Bpod r0.5	Sanworks	RRID:SCR_015943; https://github.com/sanworks
Open Ephys acquisition board	Open Ephys	RRID:SCR_021624
Kilosort2	GitHub (jamesjun)	RRID:SCR_016422; https://github.com/jamesjun/Kilosort2
Phy	cortex-lab	RRID:SCR_016249; https://phy.readthedocs.io/en/latest/

Experimental model and study participant details

Animals

All experiments were performed on adult male C57BL/6 mice purchased from Shimizu Laboratory Supplies Co., Ltd., Kyoto, Japan (9 weeks old; weighing 20–25 g). The mice were individually housed in a temperature-controlled environment with a 13-h light 11-h dark cycle (lights on at 08:00 and off at 21:00). They were provided with water after training and recording sessions so that body weights dipped no lower than 85% of initial levels and food was supplied ad libitum. All experiments were performed in accordance with the guidelines for animal experiments at Doshisha University and with the approval of the Doshisha University Animal Research Committee.

Method details

Apparatus

We used a behavioral apparatus controlled by the Bpod State Machine r0.5 (Sanworks LLC, NY, USA), an open-source control device designed for behavioral tasks. The apparatus was comprised of a custom-designed mouse behavior box with two nose-poke ports on the front wall. The box was contained in another soundproof box (BrainScience Idea. Co., Ltd., Osaka, Japan) equipped with a ventilator fan that provided adequate air circulation and low-level background noise. Each of the two nose-poke ports had a white light-emitting diode (LED) and an infrared photodiode. Interruption of the infrared beam generated a transistor-transistor-logic (TTL) pulse, thus signaling the entry of the mouse head into the port. The odor delivery port was equipped with a stainless-steel tubing connected to a custom-made olfactometer.³⁵ Both eugenol and amyl acetate (Tokyo Chemical Industry Co., Ltd., Tokyo, Japan) were used as odor stimuli, and their assignment as Go or No-Go cues was differing between animals. In four mice, eugenol was used as the go odor associated with reward and amyl acetate as the no-go odor associated with no reward, whereas in three mice, the cue-reward assignment was reversed. These odors were diluted to 10% in mineral oil and further diluted 1:9 by airflow. Water-reward delivery was based on gravitational flow, controlled by a solenoid valve (The Lee Company, CT, USA), and connected via Tygon tubing to stainless steel tubing. The reward amount (6 μL) was determined by the opening duration of the solenoid valve, which was regularly calibrated.

Odor-guided go/no-go task

After a 3 s inter-trial interval, each trial began by illuminating the LED light at the right odor port, which instructed the mouse to poke its nose into that port. This resulted in the delivery of one of the two cue odors for 500 ms. Mice were required to keep their nose poked during odor stimulation to sniff the odor. After odor stimulation, the LED light was turned off and the mice could withdraw their nose from the odor port. If go-cue odor was presented, the mice were required to move to the left water reward port and poke their nose within a timeout period of 2 s. At the water port, the mice were required to maintain their nose poke for 300 ms before water delivery began. Next, 6 μL of water was delivered as a reward. If no-go-cue odor was presented, the mice were required to avoid entering the water port for 2 s following odor stimulation. In both Go and No-Go trials, incorrect responses were penalized by imposing a punish inter-trial interval (ITI). Specifically, for Go trials, failure to enter the water port within the 2 s window was followed by a delay before the start of the next trial. For No-Go trials, false alarms—entering the water port within the 2 s period—also triggered the same punish ITI. Once in 10 trials, we introduced catch trials in which the air stream was delivered through a filter containing no odorants during which the mice were not rewarded regardless of their choice (go or no-go behavior). During the training sessions, mice learned to obtain water rewards at the left water port, move from the right odor port to the left odor port, and associate odor cues with the correct action. The accuracy rate was calculated as the total percentage of successes in the go and no-go trials in a session. The mice performed up to 524 trials (go error: up to 70 trials, no-go error: up to 53 trials, catch: up to 48 trials) in each session per day.

Electrophysiology

Mice were anesthetized with medetomidine (0.75 mg/kg i.p.), midazolam (4.0 mg/kg i.p.), and butorphanol (5.0 mg/kg i.p.), and implanted with a custom-built microdrive of four tetrodes in the ACo (0.1 mm anterior to the bregma, 2.2 mm lateral to the midline). Individual tetrodes consisted of four twisted polyimide-coated tungsten wires (California Fine Wire, single wire diameter 12.5 μm, gold plated to less than 500 kΩ). Two additional screws were threaded into the bone above the cerebellum for reference. The electrodes were connected to an electrode interface board (EIB-18, Neuralynx, MT, USA) on a microdrive. The microdrive array was fixed to the skull using LOCTITE 454 (Henkel Corporation, Düsseldorf, Germany). After the completion of surgery, the mice received atipamezole (0.75 mg/kg i.p.) to reverse the effects of medetomidine and to allow for a shorter recovery period. The mice also received analgesics (ketoprofen, 5 mg/kg, i.p.). Behavioral training resumed at least 1 week after surgery. Electrical signals were obtained using open-source hardware (Open Ephys). For unit recordings, signals were sampled at 30 kHz in Open Ephys and band-pass filtered at 600–6,000 Hz. After each recording, tetrodes were lowered by 40 or 75 μm more by turning a screw of the microdrive to obtain new units. Tetrodes were advanced each day, ensuring that the same units were not recorded across different sessions.

Histology

After recording, the mice were deeply anesthetized by intraperitoneal injection of sodium pentobarbital. Electric lesions were made using 10–20 μA direct current stimulation for 5 s of one of the four tetrode leads. Mice were perfused transcardially with phosphate-buffered saline (PBS) and 4% paraformaldehyde (PFA). Brains were removed from the skull and post-fixed in PFA. Brains were then cut into 50-μm-thick coronal sections and stained with cresyl violet. Electrode track positions were determined in reference to the atlas developed by Paxinos and Watson.³⁶

Quantification and statistical analysis

Data analyses

All data analyses were performed using the built-in software in MATLAB 2019a (The Mathworks, Inc., MA, USA).

Spike sorting

Spikes were sorted into clusters offline using Kilosort2 (https://github.com/jamesjun/Kilosort2), with default parameters. As a parameter, thresholds on spike detection used during the optimization Th(1) was 10 or during the final pass Th(2) was 4. Threshold on the area under the curve (AUC) criterion for performing a split in the final step was 0.9. Lambda is a parameter that controls the bias of individual spike amplitudes toward the cluster mean. In this study, the default value of 10 was used. Kilosort2 sorted spikes on the basis of spike waveform similarity, the bimodality of the distribution of waveform features, and the spike auto- and cross-correlograms. A unit was considered a single unit if Kilosort2 categorized that unit as “good”. The spike clusters produced by KiloSort2 were visualized in Phy2 (https://phy.readthedocs.io/en/latest/). Focusing on the “good” label obtained from Kilosort2 for each unit, the waveform and inter-spike interval (ISI) of the spikes were used to eliminate them from the analysis if they were noise. If there were similar waveform spikes, they were merged based on waveform, ISI, cross correlograms, and amplitude.

Spike train analyses

Neural and behavioral data were synchronized by inputting each event timestamp from the Bpod behavioral control system into the electric signal recording system. For calculation of firing rates during tasks, peri-event time histograms (PETHs) were calculated using a 20 ms bin width and smoothed by convolving spike trains with a 60 ms wide Gaussian filter.

ROC analyses

To quantify the firing rate changes, we used an algorithm, based on ROC analyses, that calculates the ability of an ideal observer to classify whether a given spike rate was recorded compared to baseline or under one of two conditions (e.g., during go-cue or no-go-cue odor presentation).³⁷ We defined an auROC equal to 2 (ROCarea – 0.5), with the measure ranging from −1 to 1, where −1 signifies the strongest possible value for one alternative and 1 signifies the strongest possible value for the other.

The statistical significance of these ROC analyses was determined using a permutation test. For this test, we recalculated ROC curves after randomly reassigning all firing rates to either of the two groups arbitrarily, repeated this procedure a large number of times (500 repeats for analyses of dynamics [Figures 1E, 4B, 4C, and 5B], 1000 repeats for all other analyses [Figure 5C]) to obtain a distribution of values. We then calculated the fraction of random values exceeding the actual value. For all analyses, we tested for significance at α = 0.01. Only neurons with a minimum number of five trials for each analyzed condition were included in the analyses.

For analysis of dynamics (width: 100 ms, step: 20 ms), we calculated three measures from the auROC values of correct trials (Figures 4B and 4D).

• (1)Time of center of mass: the time corresponding to the center of mass of the significant points of the auROC values (p < 0.01, permutation test). Only neurons with significant points for each analyzed condition were included in this analysis.
• (2)Duration: The duration to the time of center of mass over which the auROC values were significant (p < 0.01, permutation test) for five or more consecutive bins, containing the time of center of mass. Only neurons with consecutive bins for each analyzed condition were included in this analysis.
• (3)Onset time: The time at which the duration was first evident.

Generalized linear models

To quantify the contribution of behavioral variables to neural activity, we used a generalized linear model (GLM), which was a multiple linear regression with the firing rate of each neuron as the dependent variable, and predictors derived from the behavioral variables as the independent variables (Figure 2A; Figures S2A–S2C).^10,13,38 In this analysis, the firing rate (20 ms bin width and smoothed by convolving spike trains with a 60 ms wide Gaussian filter) of each neuron is described as a linear sum of temporal filters aligned to task events. For the current study, only odor stimulus onset, offset, and pre-odor port exit events were required, since we considered only the period between odor stimulus onset and 500 ms after the odor port exit. In the model, the predicted firing rate is given as:

$$\widehat{y_t}={\beta }_0+\sum _c\sum _i{\beta }_i^{Eearly}{x}_{t-i}^{Eearly}+\sum _c\sum _i{\beta }_i^{Elate}{x}_{t-i}^{Elate}++\sum _c\sum _i{\beta }_i^{Epre-exit}{x}_{t-i}^{Epre-exit}$$

The response of a neuron at bin t is modeled ($\widehat{y_t}$) by the sum of a bias term (β₀) and the weighted (β_i) sum of various additional binary predictors at different lags (i), and c represents the two conditions (go or no-go trials). Binary predictors for the odor presentation ($x_t^{E_{early}}$) are supported over the window 0 to 500 ms relative to the onset of odor valve in either go or no-go trials (lags corresponding to odor presentation period, 25 time bins) as well as late of nose-poking predictors ($x_t^{E_{late}}$) that are supported over the window 0 to 427/258 ms relative to the offset of the odor valve in either go or no-go trials (lags corresponding to the median durations between the odor valve offset and odor port exit, 22/13 time bins). Binary predictors for pre-odor port exit predictors ($x_t^{E_{pre-exit}}$) are supported over the window −300 to 0 ms relative to the pre-Go behavior or pre-No-Go behavior (15 time bins). The β values were calculated using the glmfit MATLAB function. Trials in which animals exited the odor port before the end of the late nose-poking window were included in the analysis; however, such early-exit trials were extremely rare, accounting for only 44 out of 11,143 total trials (46 sessions; 0.40%), and occurred in only four sessions (8/259, 8/240, 17/207, and 11/210 early-exit/total trials). Given their very low frequency, their inclusion did not materially affect the results.

The lengths of the late nose-poking windows (427 ms for go trials and 258 ms for no-go trials) were defined to match the mean durations of the late nose-poking periods in each condition. This condition-specific definition ensured that predictors captured neural activity during the actual nose-poking phase for each trial type. All predictors were normalized across trials before GLM fitting, ensuring that differences in window length did not bias the estimation of explained variance between go and no-go conditions.

Calculation of the relative contributions of behavioral variables to neural activity

We quantified the relative contribution of each behavioral variable to neural activity (Figure 2B; Figures S2D–S2F) by determining how the performance of the encoding model declined when each variable was excluded from the model.^13,39 We predicted the firing rate of each neuron in either case with all variables (full model), or by excluding one of the variables (partial model), with 5-fold cross-validation (over trials; meaning that in each fold, 80% of trials were used for training the model and the remaining trials were used for testing the model performance). The relative contribution of each behavioral variable was calculated by comparing the variance explained by the partial model to the variance explained by the full model. For the current study, which included six behavioral variables, the relative contribution of each variable was defined as

$$\left(1-\frac{R_{p,i}^2}{R_f^2}\right)/\sum _{j=1}^6\left(1-\frac{R_{p,j}^2}{R_f^2}\right)$$

Here, $R_{p,i}^2$ is the variance explained by the partial model that excludes the ith variable, and $R_f^2$ is variance explained by the full model. Negative relative contributions were set to zero (this occurs when the R² of the full model is lower than that of the partial model, owing to the introduction of noise by the excluded variable).

We used two approaches to exclude variables from the full model and calculated the variance explained by the partial model.¹³ In the first approach, the partial model was equivalent to the full model, except that the β values of the predictors of the excluded variable were set to zero (‘no refitting’). In the second approach, we calculated new β values by re-running the regression without the predictors of the excluded variable (refitting). Both approaches to exclude variables produced comparable results; the no-refitting approach was used to generate the main Figures, and comparison with the refitting approach is shown in Figure S2D.

Moreover, we compared relative contributions as assessed separately using three different approaches: no refitting (NR; used in the paper), no refitting + Lasso regularization (NR + L), and refitting (R). Lasso regularization was applied using the lasso function in MATLAB; the mean square error (MSE) of the model was estimated using 5-fold cross-validation, and we chose the lambda value that minimized the MSE. The results obtained with lasso regularization were almost identical to those obtained without regularization (Figure S2D), which suggested that there was no significant overfitting in our model.

Finally, to evaluate the significance of relative contributions assessed by the no-refitting approach, we calculated the control values. In this approach, the partial model was equivalent to the full model, except that the randomly selected β values of the predictors of the excluded variable (10% of predictors, mostly corresponding to the sum of time bins of each behavioral predictor) were set to zero, in which case, processing was performed 1,000 times (Figures S2E and S2F). Using the control mean ±2 standard deviation (SD), the statistical significance was determined (< mean – 2SD, negative relative contribution; >mean + 2SD, positive relative contribution).

Population vector construction and analyses

We constructed 2 conditions (91 time bins) × 276 neurons matrix^40,41,42 during the nose-poking epoch, in which columns contained the auROC values of the correct trials, corresponding to the trial-averaged firing rate changes from baseline (Figure S4A). By performing principal component analysis (PCA) on the dataset, we reduced the dimensionality of the ACo population from 276 neurons to three principal components (PCs), and obtained the nose-poking epoch subspaces. Note that we used the three subspaces because they explained 67.9% of the total variance (Figure S4B). To visualize the ACo population responses, we projected the dataset onto three-dimensional subspaces (Figure 3A). This allowed us to obtain a point reflecting the entire population response for each of the two conditions at a given instance. The distance between cue responses was computed as the Euclidean distance between pairs of activity vectors of all subspaces at a given instant (Figure 3B).^43,44 This value was compared with the values during the baseline epoch (1200–1000 ms before the odor port entry).

SVM decoding analyses

We used a support vector machine (SVM) algorithm with a linear kernel as a classifier^43,45 and a MATLAB function (fitcsvm) for analyses. All analyses were conducted on trial data pooled across animals. A matrix containing concatenated firing rates for each trial was used, and each neuron was the input to the classifier. The matrix dimensions were the number of neurons by the number of trials. To avoid over-fitting, k-fold cross-validation (k = 10) was used to calculate the decoding accuracy of trial type discrimination. To compute decoding accuracy, forty trials for each trial type (from the start of the session) were chosen as the dataset. Next, the dataset was partitioned into ten equal parts; one part was used for testing, and the remaining parts were used for training the classifier. This process was repeated ten times to test each individual part, and the mean value of the accuracy was used for decoding accuracy. To compute the decoding accuracy of a 100 ms bin window (step: 20 ms), the classifier was trained and tested with a 100 ms bin window (step: 20 ms).

Statistical analyses

Data were analyzed using MATLAB 2021a. Statistical methods in each analysis are described above, in the results section, or in the figure legends. The Tukey-Kramer method was applied for significance tests with multiple comparisons. Sample sizes in this study were not pre-determined by calculation; they were based on previous research in the olfactory cortex fields.^45,46 Randomization and blinding were not employed. Biological replicates for the histological studies are described in figure legends.

Published: January 7, 2026