Selective attention controls olfactory decisions and the neural encoding of odors

Summary

Critical animal behaviors, especially among rodents, are guided by odors in remarkably well-coordinated manners, yet many extramodal sensory cues compete for cognitive resources in these ecological contexts. That rodents can engage in such odor-guided behaviors suggests that they can selectively attend to odors. Indeed, higher order cognitive processes, such as learning and memory, decision making and action selection, rely on the proper filtering of sensory cues based on their relative salience. We developed a behavioral paradigm to reveal that rats are capable of selectively attending to odors in the presence of competing extramodal stimuli. We found that this selective attention facilitates accurate odor-guided decisions, which become further strengthened with experience. Further, we uncovered that selective attention to odors adaptively sharpens their representation among neurons in the olfactory tubercle, an olfactory cortex region of the ventral striatum considered integral for evaluating sensory information in the context of motivated behaviors. Odor-directed selective attention exerts influences during moments of heightened odor anticipation and enhances odorant representation by increasing stimulus contrast in a signal-to-noise type coding scheme. Together, these results reveal that rats engage selective attention to optimize olfactory outcomes. Further, our finding of attention-dependent coding in the olfactory tubercle challenges the notion that a thalamic relay is integral for the attentional control of sensory coding.

Introduction

A fundamental task for our sensory systems is to facilitate behavior towards salient stimuli. In countless cognitive functions, including learning and memory, decision making, and action selection, sensory information must be appropriately filtered in adaptive manners [1–3]. One of the mechanisms the brain employs to filter sensory information is through selective attention – attending towards one aspect of sensory information at the expense of another. The effects of selective attention on behavior and sensory processing have been widely studied in the visual and auditory systems [4–8], though much less is known about the role of attention in the olfactory system [9].

From neonatal attachment and suckling responses [10,11], to selecting mates, finding food sources, and avoiding predators [12–14], rodent behavior is guided by odors in remarkably well-coordinated manners. The fact that these behaviors can be successfully orchestrated lends reason to believe that rodents must be selectively attending to odors at the expense of competing extramodal cues. As a rat forages for food, it must simultaneously ‘filter’ out competing auditory and visual stimuli arising from irrelevant sources. Rodents readily display shifting of attentional sets, including those involving odors [15], and can direct attention towards information from other modalities [8,16], but definitive evidence of selective attention regulating olfactory perception in rodents has yet to be produced and tested in a laboratory setting. This is important given the prevalence of rodents as models for olfactory function and due to the powerful control of olfactory perception by attention in humans [9,17].

Given the aforementioned relevance of olfaction for survival, we reasoned that the olfactory system adaptively encodes odor information in manners dependent upon attentional demands, providing the brain with a mechanism to adjust the processing of incoming odor information based upon behavioral demands and context. We predicted that selective attention would shape the representation of odors within the ventral striatum’s olfactory tubercle (OT). This is likely given that the ventral striatum is important for evaluating sensory information in the context of motivated behaviors [18], a function integral for attention [19]. Offering precedence for this is evidence provided by human functional imaging for increased hemodynamic responses to odors in the ventral striatum during attention [17,20], particularly in the OT, a region extensively innervated by olfactory input [21] that encodes odors in behaviorally-relevant manners [22]. The coding strategy OT neurons engage in, which may underlie this attention-dependent phenomenon [17,20], is unknown.

Defining the control of olfactory processing by selective attention has been hindered by the lack of a behavioral task that precisely manipulates intermodal odor-directed selective attention in rodents. Here we developed such a task, and combined it with single-unit recordings, to uncover fundamental principles of how rats utilize selective attention in manners advantageous for olfactory decision-making. Our results indicate that selective attention to odors facilitates engagement in accurate olfactory decisions and enhances the contrast of odor representation in the OT by amplifying odor signal-to-noise ratios.

Results

We first sought to demonstrate that rats are capable of displaying selective attention to odors by developing a behavioral paradigm to probe intermodal selective attention. In this modified standard two-alternative choice (2AC) task, termed the Carlson olfactory Attention Task (CAT) (Fig 1A&B), rats were shaped to nose-poke into a center port, receiving simultaneously presented tones and odors, and learned to attend to the modality (auditory or olfactory) that signaled reward availability in either of two neighboring side ports.

Figure 1. Odor-directed attention dictates discrimination accuracy.

See also Figures S1–3. (A) CAT trial outline. Example trials show correct 2AC choices during odor-directed attention on ‘tone off’ trials. Dashed line indicates mandatory preparatory hold time. (B) Four possible trials during the final phase of the CAT. Top arrows for each trial indicate reward direction for tone cues; bottom arrows indicate reward direction for odors. Faded icons indicate cues that are present, but should be ignored when attending to the correct modality. (C) Example 2D histograms displaying performance of 7 rats over the course of six sessions of switching their attention during the CAT (20 trials/bin). Solid overlaid lines indicate performance (% correct); dashed horizontal lines indicate criterion performance (80%). Vertical dashed line with arrowheads indicates the uncued experimenter-controlled switch from tone to odor attention. See STAR Methods for additional details. For each rat, the top three rows were from early sessions, the bottom three rows from late sessions, except rat 1 which has 5 sessions. Rats 1–4 were used for the neural recordings.

The CAT was designed with several important features, distinguishing it from other available tasks and affords the ability to rigorously distill influences of intermodal selective attention on behavior and olfactory physiology. Main-stream attentional set-shifting tasks utilizing odors do not provide robust, controlled stimulus presentations nor do they allow for hundreds of trials, throughout which all conditioned stimuli are assigned equal valence, to be completed within a single session (e.g., [15]). With the four possible trial types available in the CAT, odors may be either unattended or attended (Fig 1B, ‘tone attention’ vs ‘odor attention’). Half of these trials do not include a tone (Fig 1B, bottom half of trials), eliminating potential multisensory confounds (unlike the Wisconsin-Card Sorting Task [23] or [8]), particularly important given that multisensory responses are observed in the OT [24,25]. The rewarded value, stimuli presented, and sensorimotor aspects for each trial type are identical, with the only difference being the rat’s intermodally-directed selective attention. Rats also learned to perform both the single-modality 2AC odor discrimination and the more challenging multi-modal CAT in the same session, allowing for questions related to task demand to be addressed. Furthermore, as cue-related anticipation influences neural activity in chemosensory cortices [26], anticipatory cues were not utilized before each trial or as ‘occasion setters’ at the beginning of the switch to indicate which modality should be attended, allowing us to monitor the progress of the attentional shift within and across sessions, allowing for behavioral flexibility and odor coding relative to the attentional switch to be tracked.

Rats selectively attend to odors and this dictates discrimination accuracy.

We shaped 7 water-motivated rats to perform the CAT (see STAR methods for details). Over several phases of behavioral training, across blocks (20 trials/block) and sessions (1–2 hours), rats were shaped to criterion performance (≥85% correct responses/block) on 2AC tone detection (Fig S1A–D, S2) and odor discrimination tasks (Figs S1E–F, S2). Once achieving criterion on these single-modality tasks, cues from both modalities were presented simultaneously and rats were shaped to shift their selective attention from tones to odors in the final CAT (Figs 1B, S1G–H). The first half of a session consisted of auditory attention blocks (‘tone attention,’ Fig S1H, left), wherein rats attended to the presence or absence of a tone, while odors were presented simultaneously. The tone and odor cues were either congruent (non-competing, signaling the same reward-port side) or incongruent (competing, signaling the opposing reward-port side). After reaching criterion performance (≥80% correct responses/block for ≥ 6 blocks), the task was switched to the ‘odor attention’ blocks (Fig S1H, right) within the same session, and rats then had to direct their attention to the conditioned odors, ignoring the competing auditory information to which they had previously been attending. It took the rats 392.6±44.6 blocks, across 24.9±1.3 sessions, to reach the first criterion switch (Tables S1–2). To establish robust behavioral performance, rats were then over-trained on numerous successive sessions. Among the last four sessions of this over-training, they took an average of 10.5±0.8 and 9.7±0.5 blocks to reach session criterion (≥80% correct responses/block for ≥6 blocks) for the tone attention and odor attention tasks, respectively.

Several significant findings emerged from the rats’ CAT performance. First, we found that task accuracy is dependent upon the animal’s attentional strategy. Following shaping, rats performed the ‘tone attention’ task, despite the presence of competing conditioned odors, with an average of 85.48% correct responses per block (±1.14 SEM, inter-animal range: 82.92–91.25%). Directly after the task was changed from ‘tone attention’ to ‘odor attention,’ there was an immediate decrease in performance (t(6)=9.78, p<0.0001, block −1 vs block 1, early blocks; Figs 1C & 2A). The rats initially made perseverative errors on incongruent trials (t(6)=−10.87, p<0.0001; block −1 vs block 1, early blocks; Fig 2B, red dashed line), indicating that they maintained their strategy of continuing to attend to the tone. As they received error feedback (no reward for incorrect trials), the rats modified their strategy, switching their attention to odors, which consequently led to fewer incongruent errors (t(6)=4.5, p<0.01; block 1 vs block 6, early blocks; Fig 2B, red lines) and increased task accuracy (t(6)=−4.88, p=0.0028, block 1 vs block 6, early blocks; 2A, bold line). Rats displayed an average of 89.7% correct responses for ‘odor attention’ (±1.63 SEM, inter-animal range: 85.42–96.67%). Second, we observed that odor-directed selective attention is subject to plasticity with experience. Across sessions, rats improved their ability to shift their attention to odors (Fig 1C & 2A, compare dashed vs. bold lines), with high levels of performance reached sooner in late sessions versus early sessions (t(6)=−2.74, p=0.034, block 6 (early) vs block 6 (late); Fig 2A). Incongruent errors decreased more rapidly across the shift in later sessions, indicating that rats shifted their attention following fewer incongruent error trials (t(6)=2.59, p=0.041, block 6 (early) vs block 6 (late), Fig 2B, red lines). The number of congruent error trials remained relatively stable across the shift (t(6)=0, p=1.0, block −1 vs block 1, early blocks; t(6)=2.25, p=0.066, block 1 vs block 6, early blocks) and across sessions (t(6)=1.55, p=0.172, block 6 (early) vs block 6 (late)); Fig 2B, blue lines), suggesting that these trials were less informative in the attentional shift from tones to odors.

Figure 2. Rats make fewer incongruent errors as they shift their attention, which leads to increased performance accuracy.

See also Figure S4. (A) Average performance of all 7 rats relative to the attentional shift, on their first two (early) and last two sessions (late). Performance dropped to chance levels immediately after the task switch and returned to criterion as the rat shifted its attention to odors. Note that performance improved more quickly during late sessions. ^#p<0.01 (block −1 vs. 1, late), ^†p<0.01 (block 1 vs. 6, late), ^‡p<0.05 (block 6 early vs. block 6 late). (B) Corresponding number of incongruent (red) and congruent (blue) errors per block relative to the attentional shift for early (dashed) and late (bold) sessions. Incongruent errors increase significantly at the task switch and decrease across blocks and over sessions. ^¥p<0.01 (block −1 vs. 1, late), ^¢p<0.01 (block 1 vs. 6, late), ^¤p<0.05 (block 6 early vs. block 6 late). (C) The average number of blocks for each rat to reach criterion after the attentional shift (minimum of 6 blocks ≥80%) for early and late sessions, *p<0.05. (D) Average number of blocks for the rats to reach the first full criterion block of ≥80% performance.

The rats took fewer blocks to reach criterion level (≥80% correct responses/block for ≥6 blocks) in late sessions, averaging 9.07 ±0.55 blocks (including the 6 blocks performed at ≥80%) in comparison to 11.71 ±1.06 blocks in early sessions (t(6)=3.34, p=0.016; Fig 2C). They reached their first high performance block (≥80% correct) within 3.36 ±0.51 blocks relative to the attentional shift (Fig 2D), demonstrating that they are capable of shifting their attention from tones to odors often following less than 30 informative incongruent trials, and that this shifting is enhanced with experience. We also tested a subset of rats for their abilities to direct selective attention to odors when perceptual demands were increased, given that there is interplay between attention, performance accuracy, and perceptual difficulty in other sensory systems (e.g., [4]). Rats required more blocks to shift their attention to odors of lower intensities (Fig S3). Together, these results demonstrate that rats can selectively attend to odors and that odor-directed attention improves with experience.

In agreement with the known influence of attention on dictating subtle, yet critical aspects of behavior [27,28], we also uncovered that trial congruency and multisensory input impact CAT performance (Fig S4). We hypothesized that the higher attentional load of the intermodal CAT would require more time to be invested sampling odors than the single-modality 2AC odor task. We further hypothesized that attending to one cue in the presence of an incongruent competing cue would impinge on the rat’s rapid decision-making, and thus that rats would invest more time directed at stimulus sampling for incongruent versus congruent trials. To test these hypotheses, we analyzed two different behavioral decision epochs: sampling durations (length of odor sampling) and reward latencies (center port withdrawal to reward retrieval). The trial outcomes were ‘correct’ (correct reward port choice), ‘incorrect’ (incorrect reward port choice), and ‘omission’ (no reward port choice), made within 4s of the trial start. To prevent biasing of the data (sampling durations from incorrect trials or omissions could skew latencies), the behavioral analyses were held to strict criterion. Only correct trials were analyzed from ≥80% performance blocks and from sessions in which they successfully switched, reaching criterion (≥80% for ≥6 blocks) on all tasks. These measures were grouped and analyzed across the different task types (‘odor only,’ ‘tone attention,’ and ‘odor attention’), according to congruency, and were further divided among trial type (odor A + tone, odor A + no tone, etc.).

Across the three task types, the sampling and reward latency durations were not significantly different (Fig S4B&C), suggesting that, in the context of the CAT, task demand does not influence decision deliberations overall. Several aspects of odor-guided behavior beyond solely discrimination accuracy, were, however, influenced. First, while errors contributed to a small number of trials overall (0.19±0.07 congruent errors/block, 2.42±0.22 incongruent errors/block) once the rats reached criterion, they committed more incongruent error trials (t(6)=−8.00, p<0.001; Fig S4D). Proportionally, 91.76±3.18% of the errors were made on incongruent trials, significantly greater than the 8.24±3.18% of errors made on congruent trials (t(6)=−13.11, p<0.0001; S4E). The low error rate reflects that for the majority of trials, rats are indeed able to ignore the competing irrelevant modality, and perform within the range of that reported by other groups in odor only 2AC tasks (e.g., [29–31]), but indicates that an incongruent stimulus may be more likely to capture the rat’s attention than a congruent stimulus, leading to more frequent lapses in attention to the opposing modality. Second, as predicted, among correct decision trials, rats invested more time sampling the stimulus if that trial was incongruent (33±8ms difference; t(6)=−4.20, p=0.0057; S4F); that is, they spent more time to make their decision when conflicting cues were present. This difference was small, however, in the context of the mean sampling duration which was ~500ms. Despite these differences, there was no impact of trial congruency upon the latency to retrieve the reward, suggesting that animals did not deliberate upon their decision as they approached the reward port, nor were they less motivated to retrieve a reward (t(6)=−1.45, p=0.197; S4G).

Given that there was an effect of congruency on odor sampling durations, we further separated the data into the four trial types to determine if one was more greatly influenced by different combinations of sensory input. As multisensory input facilitates rapid decision-making [32–34], we hypothesized that rats would need less time to sample the stimulus when both cues were present (tone on + odor) and congruent. In accordance with this, we also hypothesized that rats would take longer to sample odors when one of the cues was absent (tone off + odor) and the cues were incongruent. While sampling and reward latency durations were highly similar within an attentional task (i.e., comparing the four trial types to one another within either tone or odor attention), and across task types (i.e., comparing the ‘odor A + tone’ trial type between tone and odor attention), we did find two significant differences (Fig S4H). During odor attention, affirming that multisensory cues in the CAT can facilitate rapid decision-making, rats sampled shorter for congruent trials in which the tone was on, as compared to incongruent trials in which the tone was off (t(6)=6.05, p=0.0009, Bonferonni critical p=0.0083; S4H). Additionally, when the tone was off and the trial was incongruent, rats invested more time sampling the odor when they were attending to it versus when they were attending to the tone (51±15ms longer; t(6)=3.352, p=0.015; S4H). Importantly, the stimuli in these trials were identical (odor + tone off), but attending to the tone provided a sampling duration advantage, evidence that the rats were indeed attending to the correct modality. Altogether, while odor-directed selective attention controls performance accuracy and rats can learn to switch their attention to the relevant modality across sessions, there are additional influences of enhanced cognitive demand, trial congruency, multisensory input, and attention, on these subtle, yet critical, aspects of sensory driven behaviors.

Attention controls the neural representation of odors.

The finding that selective attention is a prerequisite for accurate olfactory decisions indicates that attention may also sculpt the neural representation of an odor in a manner that facilitates perception. Does the brain represent an odor, of equal intensity and valence, differently dependent upon whether it is attended? To address this question, rats were unilaterally implanted with drivable tetrodes [35] into their OT. We successfully performed OT single-unit recordings from 4 rats (Fig 3A, rats 1–4 in Fig 1). Across multiple recording sessions/rat (range: 6–10), we lowered the tetrodes, and identified 232 cell-odor pairs (116 total single-units × 2 odors) (Table S3).

Figure 3. OT units are bidirectionally modulated by odor-directed attention.

(A) Histological confirmation of recording locations in the OT. The dorsal-ventral and medial-lateral spans of the micro-drivable arrays are shown (red outline) within the grey outlined OT (n = 4 rats). Front panel: +2.20mm; back panel: +1.7mm relative to bregma. (B) A representative odor-excited single unit (blue), which displayed an increase in firing rate (FR) during the preparatory (hold) and stimulus presentation (odor). This FR was enhanced with odor-directed attention (bottom versus top). (C) A representative odor-inhibited unit (red), which displayed a decrease in FR during the preparatory (hold) and stimulus presentation (odor). This decrease in FR was further suppressed during odor-directed attention. Single-unit activity (SUA) was sorted from the multi-unit activity (MUA); the tetrode waveforms for each of the sorted units are in the upper right hand corner of their respective trace (scale: 100μV, 0.5ms). Peri-stimulus time histograms (PSTHs) are shown below the traces as averages across the tone (top) and odor-directed attention trials (bottom). These averages include only correct trials, from a single trial type (Odor B + tone off), taken from criterion performance blocks (≥80%) in either tone (top) or odor (bottom) attention. Raster plots include the first 30 of these correct trials.

To directly test for attentional modulation of odor coding, we only analyzed ‘tone off’ trials (50% of trials) (Fig S1H, blue box). We identified four epochs relative to stimulus onset, to assess behaviorally-relevant changes in neuron firing: (1) background (−1400 to −800ms), (2) stimulus port approach (−800 to −600ms), (3) preparatory hold (−600 to 0ms), and (4) odor stimulus duration (0 to 400ms). Across the entire population, during odor attention, 36 neurons were modulated by odor (32.03% of 116). We identified 55 odor-modulated cell-odor pairs out of the 232 possibilities (23.71%, 116 × 2 odors); n=27 odor-excited, n=28 odor-inhibited, n=177 unmodulated by odors, with some cells modulated by both odors (see STAR Methods).

We found that odor-directed selective attention bi-directionally sculpts the coding of odors in the OT by increasing the FRs of odor-excited cell-odor pairs (Fig 3B, 4A), while further decreasing the FRs of odor-inhibited cell-odor pairs (Fig 3C, 4B) during the preparatory hold and odor epochs. The normalized average FRs display enhanced contrast for odor-excited (S5A) and odor inhibited (S5B) cell-odor pairs, such that the average FR change with attention (ΔHz_attention=FR_attended-FR_unattended) is significantly greater during the hold and odor epochs for both odor-excited (hold: t(26)=−3.386, p=0.002; odor: t(26)=−3.459, p=0.002; Fig S5D) and odor-inhibited (hold: t(27)= 3.175, p=0.004; odor: t(27)=3.535, p=0.001; Fig S5E) cell-odor pairs. Changes in FRs were not statistically different for unmodulated neurons (S5C&F).

Figure 4. Odor-directed attention controls odor coding.

See also Figure S5. 2D histograms (50ms bins) displaying normalized firing rates (FRs) of odor excited (A) and odor-inhibited (B) cell-odor pairs across the task states. Neurons are arranged from highest to lowest FRs, averaged over the first five bins post-stimulus onset during odor attention. See STAR Methods for normalization details. As indicated by auROC significant bins, odor attention increases the FRs for odor-excited cell-odor pairs (C), and further decreases the FRs for odor-inhibited cell-odor pairs (D) during the preparatory hold and odor epochs. Each row represents the corresponding neuron from the 2D histograms in A and B. Odor-directed attention increases the percentage of odor-excited cell-odor pairs that have significantly excited activity (E) and the percentage of odor-inhibited cell-odor pairs that have significantly inhibited activity (G) relative to background, earlier and for a longer duration. Odor attention thus significantly increases the duration of excitement (F) and inhibition (H) during both hold and odor epochs. *p<0.05, ^**p<0.01, two-tailed, paired t-test. Data from four rats (same as in Fig 1), 2–6 sessions/rat.

To statistically represent significant FR changes for the cell-odor pairs, we classified the data using auROC analyses [22,36] (see STAR Methods), which represents changes in FR within sliding windows of time relative to a shuffled background distribution. Greater significance emerges during odor attention for both populations during the hold and odor epochs (Fig 4C&D). During odor attention, for odor-excited cell-odor pairs, a large proportion of the population was significantly and rapidly excited during the hold and odor epochs (Fig 4E). The duration of this excitement was significantly longer during both the preparatory hold and odor epochs when rats attended to odors versus when they attended to tones (hold: t(26)=−3.20, p=0.0036; odor: t(26)=−3.51, p=0.0016), while 2AC odor only discrimination was not significantly different (hold: t(26)=−2.15, p=0.041; odor: t(26)=−2.37, p=0.026) (Bonferonni critical p=0.0167; Fig 4F). Similarly, for odor-inhibited cell-odor pairs, a large proportion of the population was significantly and rapidly inhibited during the hold and odor epochs (Fig 4G). The duration that odor-inhibited cell-odor pairs were significantly suppressed relative to background during both the preparatory hold and odor epochs was significantly increased during odor attention as compared to tone attention (hold: t(27)=−3.79, p<0.001; odor: t(27)=−4.09, p<0.001) and odor only discrimination (hold: t(27)=−3.87, p<0.0001; odor: t(27)=−4.67, p<0.0001) (Bonferonni critical p=0.0167; Fig 4H).

The above results indicate that selective attention to odors bi-directionally controls both OT ensemble activity and the representation of odors. To define how individual neurons incorporate attentional demands into their representation of odors and the distribution of their changes, we used the cell-odor pairs classified above and calculated their individual changes in FR with attention (ΔHz_attention=FR_attended-FR_unattended) to yield a simple index for the direction of change in firing. Neurons were classified, for each epoch, as shifted negatively or positively if their FR either increased or decreased ≥1Hz. Among those odor-excited cell-odor pairs whose FRs shifted (n=10/27 during background, n=17/27 during hold, n=20/27 during odor), we found that the majority decreased their background FRs (70%, 7/10), while increasing their FRs during the hold (70.6%, 12/17) and odor (60.0%, 12/20) epochs with odor-directed attention (Fig 5A). The proportion of odor-excited cell-odor pairs with decreased background FRs was greater than the proportion with increased background FRs (One sample proportion z=2.8, p<0.01), while the proportion of cell-odor pairs with increased FRs during the preparatory hold was greater than the proportion with decreased firings rates (z=3.7, p<0.001). The proportion of cell-odor pairs with increased FRs during odor did not reach significance (z=1.8, p=0.0679). Furthermore, of the 16 odor-excited neurons whose FRs were shifted positively during the hold or odor epochs, 4/16 (25%) were shifted positively during only the odor epoch, 4/16 (25%) during hold only, and 8/25 (50%) were modulated during both the hold and odor epochs. Therefore, while the attentional effects on these odor-excited neurons were sometimes specific to the odor epoch, increases in FRs also frequently occurred in tandem and in a similar direction during the preparatory hold.

Figure 5. Attention yields enhanced signal-to-noise among odor coding neurons.

See also Figure S5. Changes in FR with odor-directed attention, ΔHz_attention=FR_attended-FR_unattended, for odor-excited (A) and odor-inhibited (B) cell-odor pairs for the three behavioral epochs. Pie chart: The proportion of increased or decreased FRs among neurons that shifted either negatively or positively (≥1Hz). ns above pie charts indicate the number of cell-odor pairs shifted out of the total number of excited or inhibited cell-odor pairs. ΔHz_attention during background plotted against ΔHz_attention during either the hold or odor epochs for excited (C&D) or inhibited (E&F) neuron populations. Excited cell-odor pairs falling above the dotted unity line indicate a greater change in FR relative to background. Inhibited cell-odor pairs falling below the dotted unity line indicate a greater decrease in FR relative to background. Blue (excited) and red (inhibited) arrowheads are shown for illustrative purposes to denote these shifts. Odor-modulated neurons that were not significantly modulated during the hold (only odor modulated) are colored in white. R²: [odor-excited, df=25] (background vs hold: 0.493, p<0.05; background vs odor: 0.641, p<0.05; [odor-inhibited, df=26]: (background vs hold: 0.243, n.s.; background vs odor odor: 0.061, n.s.). (G&H) The percentage of cell-odor pairs classified as either significantly excited or inhibited relative to background during the specified epochs are increased with attention to odor, particularly during the preparatory hold. Data from rats and sessions as in Fig 4. p-values in (H), as denoted, are two-tailed, paired t-tests. ^**p <0.01, ^***p <0.001, one-proportion z-test.

An opposite direction of change was observed among the odor-inhibited cell-odor pairs, where among those whose FR changed (n=13/28 during background, n=16/28 during hold, n=16/28 during odor), the majority decreased their firing during the hold (75.0%, 12/16) and odor (68.8%, 11/16) epochs while the rats were attending to odors (Fig 5B). A greater proportion of odor-inhibited cell-odor pairs decreased their FRs during the preparatory hold (z=4.6, p<0.0001) and odor epochs (z=3.2, p=0.0012) with attention. Furthermore, of the 17 odor-inhibited neurons whose FRs were shifted negatively during the hold or odor epochs, 5/17 (29.41%) were shifted negatively during only the odor epoch, 6/17 (35.29%) during hold only, and 6/17 (35.29%) during both the hold and odor epochs. Similar to the odor-excited neurons, while the effects on the odor-inhibited neurons were sometimes specific to the odor epoch, decreases in FRs also frequently occurred in parallel with decreases during the preparatory hold epoch.

Notably, we determined that these effects were selective to odor-modulated cell-odor pairs, as the majority of FRs for those which were unmodulated were unchanged during background (87.01%, 154/177), hold (88.70%, 157/177), and odor epochs (87.57%, 155/177; Fig S5G). Among those unmodulated cell-odor pairs that were shifted (11.30%, 20/177), we observed that a greater proportion displayed decreased firing during the preparatory hold (z=3.9, p< 0.0001). Overall, with odor-directed attention, odor-modulated cell-odor pairs display shifts in firing rates that occur within these task-critical moments of preparation and odor stimulus sampling. For odor-excited cell-odor pairs, the FR relative to background is shifted positively in preparation for the upcoming stimulus, while odor-inhibited cell-odor pair FRs are shifted negatively relative to background both in preparation for and during the odor stimulus.

It is possible that odor-directed attention controls individual neural FRs by broadly influencing the direction and magnitude of FRs across the background, hold, and odor epochs, indicative of a general ramping up or down of overall activity, and thus little enhancement of the odor signal relative to background. However, Figures 5A and B suggest that odor attention may be controlling the odor signal-to-noise ratios such that an odor-excited neuron’s FR is increased during the preparatory hold and odor epochs while background activity remains either unchanged or is decreased. In contrast, an odor-inhibited neuron’s FR during the preparatory hold and odor epochs may be further suppressed, while background activity remains either unchanged or is increased.

To address the above question and determine how these FRs are changing relative to these critical behavioral epochs for each neuron, in a final series of analyses, we compared the change in FR with attention (ΔHz_attention) of the background to either the hold or odor epochs for both odor-excited and odor-inhibited cell-odor pairs (Figs 5C-F). The unity line (dashed line) illustrates where changes in FR with attention would fall if they were similar in direction and magnitude across the epochs, which would support a general increase/decrease in neural activity within a trial, irrespective of epoch-specific influences. We found, however, for odor-excited cell-odor pairs, that the change in FR during both the preparatory hold and odor epochs was increased relative to the change in FR of the background (hold: t(26)=−2.32, p=0.028, odor: t(26)=−2.54, p=0.017) Fig 5C&D). In many cases, the background FR decreased, while the FR during the hold and odor epochs increased. 37.04% (10/27) of neurons had decreased background FRs, with increased FRs during the hold, while 33.33% (9/27) of neurons had decreased background FRs with increased FRs during the odor. As also predicted, for odor-inhibited cell-odor pairs, the change in FR during the preparatory hold period was more greatly decreased relative to the change in background FR (hold: t(27)=3.227, p=0.003, odor: t(27)=1.93, p=0.064; Fig 5E&F).

Notably, these effects were specific to odor-modulated neurons. During odor attention, unmodulated neurons displayed FR changes that were similar in both their direction and magnitude (hold: t(176)=1.38, p=0.169, odor: t(176)=1.01, p=0.313; Fig S5H). Odor-directed attention recruited more cell-odor pairs to encode the acts of the preparatory hold (odor attention: 21.12%, 49/232 vs tone attention: 13.36%, 3½32; Fig 5G) and odor sampling (odor attention: 21.552%, 50/232 vs tone attention: 23.707%, 55/232; Fig 5H). Therefore, these results indicate that selective attention sculpts odor coding and the preparatory activity that occurs pre-stimulus arrival within these task-critical moments by enhancing the contrast of the signal-to-noise.

Discussion

Olfactory perception and processing are shaped by behavioral state in robust manners. For instance, sleep-like states and behavioral context, which may influence behavioral state, modulate activity among neurons in the olfactory system [37–40]. Here, we expanded upon these reports, and human brain imaging results [17], which described the modulation of olfactory cortex activity by selective attention, in order to define the cellular strategies underlying attention-dependent odor perception.

We demonstrated that rats are capable of selectively attending to odors in the presence of conflicting stimuli. We predict this executive capacity affords rodents the ability to engage in ecologically critical behaviors (e.g., foraging, predator avoidance, mate selection), which are highly multisensory contexts requiring animals to focus at times upon a single modality at the expense of others. Not only does our work show that selective attention enhances odor discrimination capacity, but also that this ability improves with experience. This result highlights an important interplay between attention, olfactory processing, and learning, and indicates that rodents develop a strategy to selectively attend to odors.

Equally important is our finding that selective attention contributes to olfactory processing by enhancing the contrast of odor representation in the OT through amplification of odor signal-to-noise ratios. We observed the sculpting of relevant odor responses in both directions – increased as well as further suppressed responses. Limiting the neurophysiological analyses to odor trials in the absence of tones reflects the influence of selective attention alone, independent of multisensory processes known to occur in the olfactory cortex [25,41]. While in other sensory systems selective attention may suppress stimulus-evoked activity in an engaged task condition [8] as well as non-optimal stimuli [42,43], our findings add to standing literature that it often enhances sensory responses among neurons (e.g., [4–6,44–46]). It is also more likely that odor attention enhanced the odor responses (vs suppression by tone attention), given that odor-evoked responses during the odor only task were similar to those of tone attention. These results add to a growing body of literature on state-dependent cellular coding in chemosensory systems (e.g., [26,39,47]) and highlight that even a primitive sensory system found in a rodent incorporates executive functions to facilitate behavior. How this occurs in a sensory system lacking the canonical thalamic relay through which peripheral sensory information passes before reaching neocortical areas [48–50] is yet to be determined. In no other instances have olfactory cortex neurons been reported to be modulated by attention.

Given that the OT is strongly interconnected with motivational brain systems, it is possible that an enhancement of these responses within this structure plays a particular role in guiding behavioral decisions. This, together with possible attentional modulation in other olfactory structures, is likely responsible for the effect of selective attention on facilitating accurate odor discriminations. Selective attention may thus ‘filter’ available odor information into the entirety of down-stream structures important for emotion, motivation, and memory.

The CAT is not an all-encompassing behavioral task. First, it does not allow for the investigation of intra-modal attention in which subjects attend to stimulus attributes (i.e. spatial location or target in a mixture), like many visual studies (e.g. [51]), and one olfactory study (see [52]). Second, while an odor on/off design would have matched the tone on/off structure, it would have resulted in half the number of trials and cell-odor pairs to analyze. Despite this asymmetry, similar behavioral performance and comparisons of the same trial types should still show any effects on neural activity that are directly related to the attentional state of the rat, and we predict that an odor on/off structure would produce similar effects. Third, we investigated FR differences with attention for high performance blocks, but there are low performance blocks as the rat shifts its attention across a session. These data support the notion that the FRs of neurons may be sculpted across an attentional shift, but the low FRs of OT neurons made a quantitatively significant analysis of these changes difficult. Furthermore, it is unclear if the firing rate changes we observed are causal in odor-directed attentional behavior or for the shift to occur. It is possible that they are not necessary for the shift, but instead may reflect an enhancement of sensory representation that occurs after a shift has been made.

While an animal is engaged in a dynamic sensory task one cannot rule out influences of sensorimotor demands upon sensory processing. That said, our findings that rats sampled odors similarly whether they were attending to odor or not, and the influence of selective attention enhancing odor representation in the attention task going beyond those responses in the odor only task, indicate that these effects surpass those of sensorimotor influences alone (e.g., go left or right). Finally, the CAT was designed with a preparatory hold epoch. Initially, this was to be allocated for pre-stimulus ‘background’ comparisons, but many of the odor-modulated neurons were also modulated during this time. It is known that neurons encode anticipation of port entry and task engagement [26,53,54], but beyond this, we found influences of selective attention, suggesting that the impact of selective attention is not restricted to the overt stimulus period. The influence of behavioral state on the encoding of sensory information does not mandate itself to occur within solely the moment the stimulus is engaged by the brain, but may occur even while one prepares for stimulus arrival. Indeed, increased activity with odor-directed attention has also been observed in the OTs of humans during the epoch preceding stimulus delivery, when subjects were given verbal instructions to attend to an odor [17]. This may be a preparatory phenomenon related to attentional direction (‘I’m anticipating an odor’) that aides in guiding olfactory goal-directed behaviors.

Taken together, a rodent, just like a human [9,17], can employ selective attention to aid in olfactory perceptual goals and this attention enhances the representation of odor information within part of a brain system that is integral for evaluating sensory information in the context of evolving motivational demands. Our results put forward a model whereby an attention-dependent signal-to-noise coding strategy facilitates odor perception.

STAR Methods

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Kaitlin Carlson (ksc46@case.edu).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Animals.

Adult, male Long-Evans rats (n = 7, 2–3 months of age, 250–350g) were obtained from Charles River Laboratories (Wilmington, MA) and maintained within the Case Western Reserve University School of Medicine vivarium. Rats were housed on a 12:12-hour (light:dark) cycle with food and water available ad libitum, until behavioral shaping or electrophysiology experiments. Experiments were performed during the light cycle and conducted in accordance with the guidelines of the National Institutes of Health and approved by both the Case Western Reserve University and the University of Florida Institutional Animal Care and Use Committee.

METHOD DETAILS

Olfactory and auditory stimuli.

The odor pairs used were: 1) isopentyl acetate vs limonene (−) and 2) 2-butanone vs 1,7-octadiene. Molecularly diverse, monomolecular hydrocarbons were selected to generate coarse (perceptually easy) odor pairs to discriminate. Unless otherwise noted, odors were diluted in mineral oil to equal vapor pressure (0.5 Torr, 66.67 Pa; Sigma Aldrich, St. Louis, MO) and presented through a custom air-dilution olfactometer (2.0 L/min) with independent stimulus lines up to point of entry into a custom 3D-printed [polylactic acid (PLA)] nose-poke port. The photoionization detector (miniPID, Aurora Scientific) trace in Figure S4A confirms the precision and stability of the odor presentation and clearance temporal dynamics in our apparatus. The auditory stimulus, positioned immediately above the behavioral apparatus, 60cm above the rat, consisted of a 2.8kHz tone (76dB, piezo speaker, RadioShack). Stimuli were presented pseudorandomly (random, but organized in such a manner that an equivalent number of the four trial types were given across a block) and counterbalanced, such that all four trial types (Fig 1B) were equally possible in each block (20 trials/block).

CAT behavioral shaping.

Rats were mildly water-restricted for several days before shaping. Over the course of training (1–2 months), though water-restricted, the rats did not display any signs of illness or distress, sustained task performance motivation, and even gained weight (Table S4), without exceeding a ≥ 15% drop body weight. Each shaping phase of the CAT will be detailed herein and the corresponding number of blocks and sessions to reach criterion can be found in Supplemental Tables S1–2. In a dimly lit, well-ventilated room (20–22°C), rats were placed into an open-top chamber (ABS plastic, acrylonitrile butadiene, 24cm², 48cm high), with ports/spouts positioned along one wall: left reward spout, center stimulus port, right reward port. Ports/spouts were embedded with 880nm infrared photobeams, the status of which were acquired to detect entry (380Hz).

Rats were first shaped on the single-modality 2AC tasks (odor only and tone only) in blocks of 20 trials (Fig S1A–E, S2, Table S1). 5/7 rats were trained on the tone only 2AC first, and 2/7 on the odor only 2AC first, so that we could inspect for gross differences in CAT learning based upon initial shaping modality, of which none were observed. For these early phases, the rats were shaped to nosepoke in the center port and retrieve rewards to the right or left; they were required to reach two consecutive blocks of ≥85% correct before moving onto each subsequent phase. Rats were initially shaped with a ≥85% correct criterion on the single-modality tasks, which was later decreased to ≥80% in the final intermodal CAT. In phase 1 (Fig S1A, S2A), the left reward port was covered and rats learned to nose-poke into the center port for 200ms and retrieve a reward (25μL of 5mM saccharin) from the right reward spout. Initially, the reward was auto-triggered to the right spout if the rat remained in the center port for ≥200ms to aid in behavioral shaping. Rats eventually associated the center nose-poke with a right reward and within 4s post-stimulus presentation would retrieve it. After two consecutive blocks at criterion (breaking the IR beam within 4s post-stimulus delivery), the auto-triggered reward was removed and the rat was required to break the reward IR beam for retrieval on its own. Many rats displayed a brief dip in performance when this auto-triggered reward was removed (Fig S2A).

In phase 2, (Fig S1B, S2B), with the left reward port still blocked, rats (5/7; tone only 2AC) held for an additional 300ms (‘tone off’), the initial minimum stimulus duration requirement. After two consecutive criterion blocks, rats were advanced to phase 3, (S1C, S2C), wherein the right reward port was covered, while the left was made available. The rats nose-poked into the center port, waited the required 200ms hold period, and then received a ‘tone on’ stimulus during the 300ms stimulus duration, and were shaped to retrieve a left reward. Finally, in phase 4a (S1D, S2D), termed the ‘tone only’ task, both reward ports were made available and rats were shaped to report detection of the tone (‘tone on’ vs ‘tone off’). After reaching criterion (≥6 blocks of ≥85%), the rats were advanced to phase 4b, the ‘odor only’ task, where they were given odor stimuli instead of tone stimuli and shaped to discriminate between the two odors (‘odor A’ vs ‘odor B’) (S1E, S2E). The remaining 2/7 rats learned the ‘odor only’ task first and the ‘tone only’ task second. Regardless of which task they were shaped on first, rats took significantly longer to reach criterion for the ‘tone only’ task than for the ‘odor only’ task (101±6.3 vs 11±2.1 blocks, t(6)=−16.98, p<0.0001; 5.86±0.26 vs 1.29±0.18 sessions, t(6)=−15.37, P<0.0001), but despite these learning differences, all rats were trained to the same level of criterion on each task.

We then shaped the rats to switch between modalities, on ‘tone only’ and ‘odor only’ blocks (Fig S2F, Table S2). The block task type was changed after they reached ≥85% criterion performance across three blocks for a single modality. After they were able to switch between these single modality tasks for at least one session, we began to shape them with cues from both modalities present simultaneously. In the following session, rats needed to complete three blocks of the tone only task at ≥85%, at which point we then presented both tone and odor simultaneously, but rewarded only the choices made towards the odors, in what we termed ‘odor attention’ blocks (Fig S2G). Rats were shaped to complete 6 blocks of ≥80% correct on this odor attention task. After successful completion at this criterion, we began the following sessions with the alternate ‘tone attention’ task, in which cues from both modalities were again presented simultaneously, but only choices made towards the tones were rewarded (Fig S2H, left). That is, while the same (now irrelevant) odors from the odor attention task were presented simultaneously, the rats learned that they no longer predicted reward availability. Once they achieved 6 blocks of ≥80% correct, the rats were switched back to the ‘odor attention’ blocks, in which the odors were then rewarded, while the tones were now irrelevant (Fig S2H, right).

At this point, the rats had learned to switch from ‘tone attention’ to ‘odor attention’ blocks across the course of a session. At this point, the 200ms hold and 300ms stimulus duration times were gradually increased to 600ms and 400ms, respectively, in 100ms increments. For each increment, the rats were required to perform two consecutive blocks at ≥80%. Then, over the course of a session, a 1s ITI was imposed, followed by a second session in which a nose-poke within that 1s ITI period would reset the ITI and trial (viz., a ‘time-out’). We selected this mandatory ITI and time-out to provide a background epoch between trials for analysis of the physiology data (see later). The rats established robust behavioral performance over numerous successive sessions in this final CAT phase (hereafter simply called the ‘CAT’; Figs 1C & 2A, S2H).

Classifying selective attention.

Attentional shifts from tones to odors were determined by an initial decrease in performance at the attentional switch point, which gradually increased over the course of the session as the rat shifted its attention (Fig 1C & 2A). As expected, we found performance to fluctuate as rats determine which modality to attend to, so we focused our behavioral and physiological analyses on only correct trials from those blocks in which the rat performed at ≥80% correct (high performance before the switch [tone attention]; high performance after the switch [odor attention]), well above chance. It is possible that during these blocks, there is an occasional lapse in attention to the opposite modality (e.g., as observed in other systems [55]) resulting in either 1) an incorrect choice (on incongruent trials) or 2) a correct choice for the wrong reason (on congruent trials). However, this would be infrequent and unlikely to bias our data given our strict requirement of ≥80% correct and discrete choice of analyzing only correct trials.

Behavioral data analysis.

Two major epochs relative to odor onset were chosen for behavioral data analyses: Sampling duration: time from stimulus-onset to withdrawal from center port; latency to reward: withdrawal from center port to reward spout.

Task type was separated into three categories: ‘odor only,’ ‘tone attention,’ and ‘odor attention.’ Within these tasks, congruent and incongruent trials were separated and compared. Further, trials were separated based upon type (four different possible combinations, Fig S1H) and whether or not the rat was attending to the odors. This allowed for both the behavioral and neural data to be analyzed for the specific trial types that corresponded to the two possible odors in the absence of tone stimuli. For behavioral and electrophysiological data analyses, we required a minimum of: 1) 3 blocks (60 trials) of 2AC odor only at criterion (≥80%), 2) 6 blocks (120 trials, 60 trials unattended and with no tone) of tone attention at criterion (≥80%), and 6 blocks (120 trials, 60 trials attended with no tone) of odor attention at criterion (≥80%) within a single session. All behavioral data was taken after CAT learning; individual rats contributed a single mean value per behavioral measure, averaged over three sessions each. For final percent correct averages during tone and odor attention as reported in the main text, averages of the six criterion blocks pre and post switch were taken from the last 2 sessions/rat. All error bars are standard error means (SEMs).

Micro-drive and tetrode construction.

Following Voigts et. al [35], we constructed flexDrives. These ultralight Microdrive implants with independently moveable tetrodes were adapted slightly to fit our needs by utilizing only 8 of the independently moveable tetrodes (instead of 16) and increasing the length of the static guide tube to reach just dorsal to the OT. Tetrodes were constructed following Nguyen et al. [56]. We twisted 12.5μm XTC-bonded ni-chrome wires with a tetrode twister (https://open-ephys.atlassian.net/wiki/display/OEW/Twister), and electroplated tetrodes to 200–250 kOhm in a neuralynx gold:PEG solution (Neuralynx non-cyanide gold plating solution in PEG [polyethylene glycol, Sigma-Aldrich, 8000MW, 1g/L in ddH2O]), following Ferguson et. al [57]. We also constructed a 3D-printed cone and cap [polylactic acid (PLA)] to cover the drive body, lined the inside with aluminum foil, and secured the cap top into place with plastic screws.

Surgical procedures.

Rats were provided ad libitum access to both food and water for at least 48 hrs prior to surgery. We then unilaterally implanted all rats (n=7) with the constructed flexDrives just dorsal of the right OT. Only 4 were later utilized for electrophysiology data analyses due to electrode placement errors, poor signals, or inability to perform/sustain motivation during the course of the cognitively demanding CAT following surgery. Surgery was performed as described previously [58], with the following modifications. After anesthesia (isoflurane, 3.5–1% in 1.5L/min O2) and preparation of craniotomies, three 0–80 s/s screws were implanted into the skull for anchoring the cement. The skull was cleaned with 0.9% physiological saline and 0.3% hydrogen peroxide, allowed to dry, and coated with a thin layer of Vetbond. The flexDrive array was lowered (centered at 1.2mm anterior to bregma, 1.75mm lateral to midline, 7.7mm ventral). Wax was applied to seal the craniotomy and a layer of cement was placed around the PI tubing to keep the flexDrive implant initially stable. One of the stripped ground wires (s/s) was wrapped around the ipsilateral skull screw and silver paint was applied to ensure conductivity. The stripped reference electrode was lowered into the contralateral hemisphere, sealed with wax, and cemented in place. Finally, additional layers of cement were used to secure the flexDrive base more rigidly to the skull and the ground skull screw. Before fully fleshing out the cement, the 3D-printed plastic cone was attached, surrounding the flexDrive. More cement was used to hold the drive and cone into place. The second ground wire was threaded through the cone and attached to a screw on the outside of the cone, coated with silver paint, and epoxied into place. The cover of the cone was secured in place with plastic bolts, which allowed for its removal to access the connector during recordings. Rats were then injected with 2.0 mL physiological saline (0.9%, s.c.) to aid in rehydration. After surgeries, rats were returned to their individual fresh home cages and they were given Carprofen daily for 5 days post-surgery (Rimadyl, Pfizer Animal Care, 5mg/kg s.c.).

Data acquisition.

We recorded full-band neural activity with Intan hardware (Intan Technologies, Los Angeles, CA), amplified and digitized at the headstage, and stored at 25kHz through the Intan software GUI. Behavior and odor presentation events were recorded simultaneously using OpenEx (Tucker Davis Technologies,TDT) and an RZ5 BioAmp Processor with a sampling rate of 25kHz. We used an Arduino to ensure synchrony of behavioral and neural data.

Electrophysiology recordings.

After surgery, the rats were allowed to recover for at least three days before being placed on a gradual water-restriction schedule, at which point they were handled and re-acclimated to the chamber. Their body weight was monitored and maintained daily by means of supplemental water, given at the end of each session (Table S4). By day five post-surgery, rats began CAT re-shaping with their newly implanted hardware, but without the headstage/tether attached. Once they began switching, we plugged them into the headstage. Each session began with several blocks of the odor only 2AC, followed by tone attention blocks, and then odor attention blocks. Some tetrodes were independently advanced each day, such that they traversed the OT. If they had units, but the rat didn’t switch, for example, we did not move them and tried recording again the following day. If tetrodes were advanced, they were advanced at least 60 μm to ensure we were capturing novel neurons. Not all implanted rats contributed physiology data due to electrode placement errors, poor signals, or their inability to perform the cognitively demanding CAT following surgery.

Perceptually demanding CAT.

After collecting the standard CAT data, we tested n=2 rats on the perceptually demanding reduced odor intensities. The CAT was identical, except we utilized a descending staircase design of decreased odor intensity over consecutive days (0.5, 0.05, 0.005, 0.0005 Torr). Performance on the single modality 2AC odor discrimination task when odor intensities were reduced: (0.5 Torr: 93±1.7%; 0.05 Torr: 91±0.8%; 0.005 Torr: 95±0%; 0.0005 Torr: 85±4.2%).

Electrode placement verification.

Following recordings, the rats were lethally overdosed with a urethane injection (i.p.). Current was delivered into the tetrode array (Cygnus stimulus isolator; 50 uA, 15s) to aid in post-mortem localization of tetrode tips. The rats were then transcardially perfused with 4°C 0.9% NaCl, followed by 10% formalin (Fisher Scientific, Waltham, MA). After perfusion, brains were removed and stored in 30% sucrose formalin at 4^oC. We sectioned rat brains coronally (40μm), mounting on gelatin-subbed slides, and labeled with 0.1% cresyl violet. Micro-drive tip distances were reconstructed by finding the most ventral tip from the most ventral section, referencing the Paxinos and Watson (1997) [59] rat brain atlas. OT recording sites were found mostly in the anterior OT, spanning medial-lateral and dorsal-ventral about 1mm² (Fig 3A). Any recordings dorsal to the OT were not analyzed.

QUANTIFICATION AND STATISTICAL ANALYSIS

Electrophysiology with behavior data analysis.

Neural data was converted from Intan to Spike2 (Cambridge Electronic Design) and merged with TDT behavioral data. The full-band neural data was filtered (second-order low-pass Butterworth, 0.2–3000Hz). Units were sorted following tetrode-sorting methods in Spike2, where we used a combination of template matching and cluster cutting based on principle component analysis. Single neurons were defined as having <2% of spikes occurring within a 2ms refractory period. Spike times associated with each trial were then extracted, exported, and analyzed with custom MATLAB (MathWorks) scripts, as we have described previously [60].

For analysis of electrophysiology data we categorized behavioral epochs relative to odor onset including: (1) background (−1400 to −800ms): time prior to approaching the center stimulus port during the ITI, (2) approach (−800 to −600ms): time period as rat approaches center port, (3) preparatory hold (−600 to 0ms): center port nose-poke until stimulus onset, and (4) odor (0 to 400ms): stimulus onset until the minimal 400ms required stimulus duration. Of particular interest is the odor epoch, during which information may be mapped onto the neurons and used to guide behavioral choice. Also of interest is the preparatory hold epoch, in which the rat is holding its nose in place in preparation for the arrival of the upcoming stimulus. Since it is not uncommon for neurons to encode anticipation of port entry and/or task engagement [26,53,54], the 200ms approach epoch was categorized to clearly separate out background from the hold and odor epochs. The stimulus duration epoch is referred to as ‘odor’ because we only analyzed ‘tone off’ trials. Importantly, the required 400ms minimum stimulus sampling allowed for an unbiased epoch of time to be consistently compared.

Modulations of firing rate (FR) within a single trial were examined similar to [60]. Mean FRs across trials were measured in 50ms bins. The mean background FR for each neuron was averaged across the background epoch. Neurons were categorized as significantly modulated (excited, inhibited, or unmodulated) by comparing the mean FR during the odor to the background FRs during odor attention utilizing a t-test. Neurons FRs during the odor epochs were classified as odor-excited (significantly greater), odor-inhibited (significantly less), or unmodulated (not significantly different). Across 116 neurons (Table S3), we performed this significance test for each odor pair. Many of the odor-excited and odor-inhibited cells were also modulated during the preparatory hold period. To test for this modulation, we similarly compared the hold FR to background FRs and tested for significance. For the 2D histograms of FR in Fig 4 A&B, we normalized the data to the minimums and maximums [FR_normalized = (FR_x-min)/(max-min)], where FR_x = FR of one 50ms bin, within each neuron across each task type. A single neuron’s lowest FR was then 0, while its highest FR was 1, so that differences in FR could be observed across attentional demand and states.

ROC analysis.

We additionally performed an area under the receiver operating characteristic analysis (auROC), a nonparametric measure of the discriminability of two distributions [61]. This normalized activity across neurons and allowed us to significantly quantify stimulus-related changes in FR relative to the background activity, on a scale from 0–1, following Gadziola et al. [60] (more details see [36]). A value of 0.5 indicates overlapping distributions, while 0 or 1 indicate perfect discriminability. The auROC was calculated at each 50ms time bin over the 3.6s period (−1800ms to 1800ms), centered on odor onset for each neuron. Values >0.5 indicated the probability that FRs were increased relative to background, while values <0.5 indicated the probability that FRs were decreased relative to background. A null distribution of auROC values of ~0.5 was created by utilizing a permutation test, where the “response” and “background” FR labels were randomly reassigned and calculated 1000 times. Significant auROC bins (50ms), as reported, were determined by testing whether or not the actual auROC value was outside the 95% confidence interval of the null distribution [62].

Statistical information.

All statistical analyses were performed in Microsoft Excel or MATLAB (Mathworks, Waltham, MA), and all data are reported as means +/− SEM unless noted otherwise. All t-tests are paired and two-tailed, unless otherwise stated.

DATA AND SOFTWARE AVAILABILITY

Data availability.

The data that support the conclusions of this study are available upon request to the corresponding author.

Code availability.

Custom behavioral and analytical code are available upon request to the corresponding author.