Neural Activity in the Anterior Insula at Drinking Onset and Licking Relates to Compulsion-Like Alcohol Consumption

Abstract

Much remains unknown about the etiology of compulsion-like alcohol drinking, where consumption persists despite adverse consequences. The role of the anterior insula (AIC) in emotion, motivation, and interoception makes this brain region a likely candidate to drive challenge-resistant behavior, including compulsive drinking. Indeed, subcortical projections from the AIC promote compulsion-like intake in rats and are recruited in heavy-drinking humans during compulsion for alcohol, highlighting the importance of and need for more information about AIC activity patterns that support aversion-resistant responding. Single-unit activity was recorded in the AIC from 15 male rats during alcohol-only and compulsion-like consumption. We found three sustained firing phenotypes, sustained-increase, sustained-decrease, and drinking-onset cells, as well as several firing patterns synchronized with licking. While many AIC neurons had session-long activity changes, only neurons with firing increases at drinking onset had greater activity under compulsion-like conditions. Further, only cells with persistent firing increases maintained activity during pauses in licking, suggesting roles in maintaining drive for alcohol during breaks. AIC firing was not elevated during saccharin drinking, similar to lack of effect of AIC inhibition on sweet fluid intake in many studies. In addition, we observed subsecond changes in AIC neural activity tightly entrained to licking. One lick-synched firing pattern (determined for all licks in a session) predicted compulsion-like drinking, while a separate lick-associated pattern correlated with greater consumption across alcohol intake conditions. Collectively, these data provide a more integrated model for the role of AIC firing in compulsion-like drinking, with important relevance for how the AIC promotes sustained motivated responding more generally.

Significance Statement

The anterior insula cortex is known to drive many motivated behaviors, especially those involving challenge and emotion regulation, but insula activity patterns that promote and sustain such behaviors remain incompletely understood. Here we examined insula firing related to compulsion-like alcohol consumption, where intake persists despite adverse consequences. Compulsion can strongly contribute to alcohol problems and also represents a useful test case for understanding how insula activity could sustain behavior in the face of challenge. Insula cells showed several session-long and lick-synched activity patterns. Compulsion-like drinking exhibited specific firing changes that led to the suggestion that the insula helps evaluate the intake condition at drinking onset and then provides session-long plateau and lick-related activity to maintain responding despite challenge.

Introduction

Compulsion-like alcohol consumption, continued intake despite negative consequences, can be a major obstacle to treatment (Larimer et al., 1999; Tiffany and Conklin, 2000; Epstein and Kowalczyk, 2018; Giuliano et al., 2018). Compulsion features prominently in the DSM-V criterion for Alcohol Use Disorder, and greater alcohol problems predict higher intake (Grant et al., 2015; Takahashi et al., 2017; Patrick and Azar, 2018). Treatment of problem drinking could be greatly aided by identifying underlying brain mechanisms driving individuals toward alcohol. However, despite recent advances (De Oliveira Sergio et al., 2021; Domi et al., 2021; Patton et al., 2021), much remains uncertain.

When seeking critical circuits for alcohol drinking, it is likely that consumption is driven by alcohol's salient motivational properties. The anterior insula (AIC) is a key regulator of the salience network, which mediates rapid identification and responding to important situations (Sridharan et al., 2008; Craig, 2009; Menon and Uddin, 2010). Further, AIC activity is linked to many aspects of human drinking (Centanni et al., 2021), and we previously showed that specific AIC projections promote compulsion-like intake in male rats with little impact on alcohol-only consumption (drinking without overt adverse consequences), including AIC–nucleus accumbens (Seif et al., 2013; De Oliveira Sergio et al., 2021). Other rodent studies also implicate the AIC in compulsion-like drinking (Campbell et al., 2019; Chen and Lasek, 2019). Interestingly, heavy-drinking humans activate a similar AIC–striatal circuit during compulsion-like responding for alcohol (Grodin et al., 2018) and when imagining high-risk intake (Arcurio et al., 2015). In addition, global AIC inhibition in rats reduces both compulsion-like and alcohol-only intake (Jaramillo et al., 2018b; De Oliveira Sergio et al., 2021), suggesting that different alcohol-related AIC signals may promote differing aspects of drinking. However, one major obstacle is the lack of information on in vivo AIC neural firing related to alcohol drinking, and identifying such patterns could reveal critical and foundational insights into how AIC activity drives pathological behavior.

Thus, we used custom-built, 32-microwire arrays (Linsenbardt et al., 2019; Morningstar et al., 2020) to directly examine AIC neuronal activity patterns. In particular, we compared in vivo AIC activity across sessions involving different alcohol-drinking conditions, including alcohol-only consumption, moderate-challenge compulsion-like drinking (10 mg/L quinine, where alcohol intake is maintained), and higher-challenge compulsion-like consumption (60 mg/L quinine, where alcohol intake decreases; Seif et al., 2013; De Oliveira Sergio et al., 2021). We focused on sustained AIC activity patterns across a given drinking session, since our behavioral work suggests that session-long strategies are likely used to maintain compulsion-like drinking (Darevsky et al., 2019; Darevsky and Hopf, 2020). Also, persistent AIC signals reflect main task goals in humans (Dosenbach et al., 2006) and long-term reward value in nonhuman primates (Wittmann et al., 2020). In rats, long-term AIC firing changes may reflect persistent drive for cocaine (Guillem et al., 2010; Pribut et al., 2021). In addition, several studies find increased AIC c-Fos with compulsive alcohol (Campbell et al., 2019, Chen and Lasek, 2019), suggesting importance of cells with greater sustained activity. However, AIC cells can also fire on shorter time scales, including in synchrony with licking (Stapleton et al., 2006; Sadacca et al., 2016; Mukherjee et al., 2019; Neese et al., 2022). Importantly, we examined AIC firing synchronized with licking, but we examined AIC patterns across all licks in a session, suggesting that such lick-related activity also reflects a more sustained, long-term action strategy.

Importantly, we discovered multiple AIC activity patterns that together provide evidence for an integrated model for how different aspects of AIC signaling could promote compulsion-like consumption specifically versus alcohol drinking more generally. Indeed, compulsion-like intake involved higher lick-synched activity and greater plateau firing in one type of sustained activity cell. In addition, we propose a novel model where the AIC provides sustained commitment to act for high-value reward, regardless of challenge level. Our findings likely have important relevance for AIC promotion of motivated responding more generally, especially under challenge, in addition to specifically enhancing alcohol drinking.

Materials and Methods

Animals

Fifteen adult, male Wistar rats (Envigo) arrived at postnatal day (P) ∼55 and were single housed in clear plastic cages, with lights off 11 A.M.–11 P.M. All experiments were carried out in accordance with procedures approved by Indiana University Institutional Animal Care and Use Committee. After 2 weeks habituation to vivarium, alcohol drinking began. Rats were ∼450–750 g at the time of experimental studies.

Alcohol drinking

All consumption occurred in the home cage, with the spout of drinking bottles inserted through holes in the front of the home cage ∼7 cm above the cage floor. In Figure 1A, rats initially drank alcohol under an intermittent-access two-bottle-choice paradigm (20% alcohol in water v/v, vs water-only in a second bottle). For this, rats had three 16–24 h intake sessions per week, starting Monday, Wednesday, and Friday at ∼30 min into the dark cycle. After at least 3 months of intermittent-access drinking, rats were switched to drink under limited daily access, which involved 20 min/d, 5 d/week two-bottle choice drinking (alcohol 20% v/v, and water).

Figure 1.

Alcohol drinking. A, Rats had ∼3 months of intermittent access to alcohol: two-bottle choice of 20% alcohol (v/v) in water versus water-only in a second bottle, with 16–24-h intake starting Monday, Wednesday, and Friday at ∼30 min into the dark cycle. Rats then drank ∼1 month under limited daily access (20 min/d, 5 d/week two-bottle choice). B, After drinking described in A, rats first had limited daily access with 2–3 d/week each of AlcQuinine10 and AlcQuinine60 to habituate to novelty of quinine. After additional habituation (see Materials and Methods), in vivo recordings under different drinking conditions began. C, D, Higher challenge decreased (C) alcohol drinking (p < 0.0001 alcohol-only vs AlcQuinine60, p = 0.0181 AlcQuinine10 vs AlcQuinine60) and (D) total licks (p = 0.0242 alcohol-only vs AlcQuinine60). E, Similar shifts to less variable lick volume (mg/kg alcohol per lick within a session) in both AlcQuinine10 and AlcQuinine60 versus alcohol-only, as well as decreased average lick volume (alcohol-only vs AlcQuinine60 p = 0.0003; alcohol-only vs AlcQuinine10 p = 0.0690). F, Average bout length in a session and (G) licking rate were similar across drinking conditions. *p < 0.05, ***p < 0.001; ^##p < 0.001 differences in standard deviation.

Pre-experiment training and habituation

After the 3–4 weeks of limited daily access in the housing room, rats began drinking under limited daily access in the behavioral room in the dark cycle under red light. In addition, wire cage lids were replaced by a cage extender, an in-house creation where the bottom of a home cage is cut off, and this is placed upside down on top of the rat's home cage and fixed in place. This “tower” allows access for head-attached cables but largely prevented rats from jumping out. Also, custom-created two-bottle holders were attached to the front of their home cage to accommodate angled sippers required for lickometry during recording. Rats were trained for ∼3 weeks drinking in this tower with daily handling to acclimate experimenters prior to intake. During this period, rats also underwent brief alcohol-quinine training (Fig. 1B). For this, 2–3 d/week of limited daily access involved drinking alcohol with quinine added, usually two sessions each of alcohol with 10 mg/L quinine (AlcQuinine10) or 60 mg/L quinine (AlcQuinine60) across 2 weeks. This was done to habituate to quinine and assure that alcohol-quinine was not novel during experimental sessions. We consider AlcQuinine10 a moderate-challenge and AlcQuinine60 a higher-challenge condition, which we previously used (Seif et al., 2013; Darevsky et al., 2019; Darevsky and Hopf, 2020; De Oliveira Sergio et al., 2021). Finally, rats underwent surgery to implant recording devices (below), followed by 1 week recovery. Then rats drank ∼1 week more and then ∼1 week of drinking with cable attached and daily handling, both with tower system (as in Seif et al., 2013; De Oliveira Sergio et al., 2021), before recording studies began.

Surgery

Animals underwent surgery with isoflurane anesthesia and analgesia (buprenorphine and carprofen for 2 d). Four stainless steel anchoring screws were strategically placed rostral and caudal to bregma; the two caudal screws were in the occipital bone to attach grounding wires. A rectangular craniotomy was performed over the AIC, the dura excised, and the site cleared of debris. The probe was descended into the target location (centered at AP, +2.9; ML, +4.8; DV, −6.0 from bregma), cemented into place, and connected by wire to two ground screws in occipital bone. Implants targeted the right AIC, since the right AIC activates for intoxicant cues, drive for intake, and negative affect in humans (Craig, 2009; Centanni et al., 2021), and the right but not left AIC mediates mouse aversion responses (Wu et al., 2020). Animals recovered in a heated home cage until alert and mobile.

Experimental alcohol drinking sessions

For experimental sessions, drinking time was reduced to 4 min per session (Seif et al., 2013), since rats strongly “front-load” and have the vast majority of their intake in the first minutes of alcohol access (Seif et al., 2013; Darevsky et al., 2019; Darevsky and Hopf, 2020; De Oliveira Sergio et al., 2021; Starski et al., 2023). All experimental sessions occurred in the home cage with tower, as described above. We recorded in vivo AIC firing during alcohol-only, AlcQuinine10, and AlcQuinine60 sessions, tested within-rat (Fig. 1B). Drinking conditions were examined in randomized order, and each drinking condition was tested at least twice within each rat (Seif et al., 2013, 2015; Darevsky et al., 2019; Darevsky and Hopf, 2020; De Oliveira Sergio et al., 2021). This was done to have two sessions from each rat per drinking condition that had a sufficient number of recorded cells in each session, balanced against having the most randomized order of drinking sessions. Rats continued 20 min limited access drinking when not undergoing recording sessions.

Saccharin drinking sessions

For saccharin consumption, a separate cohort of rats (n = 4) first had intermittent access and limited daily alcohol drinking. Following our previous methods (Seif et al., 2013, 2015; De Oliveira Sergio et al., 2021), rats were then trained to consume saccharin (0.05%, vs water) under 20 min/d, 5 d/week limited access, which occurred ∼2 h before alcohol intake each day in the home cage with “tower.” After 3–4 weeks, rats had surgery and recovery (as above), and then experiments began, with in vivo recording during at least two saccharin and two alcohol-only 4 min sessions per rat. We note that future studies should examine AIC firing in relation to different levels of saccharin adulterated with quinine. This would help to better understand AIC encoding for different levels of quinine-adulterated saccharin and more directly determine AIC coding of primary reward–cost balance activity; we have previously tested different concentrations of saccharin-quinine and saccharin alone (Seif et al., 2013, 2015).

In vivo recording

In vivo recording methods follow those previously published (Linsenbardt et al., 2019; Morningstar et al., 2020). Thirty-two–channel Multi Wire Arrays were home-made (detailed in Morningstar et al., 2020), with 23 µm tungsten wires (California Fine Wire) 100 µm apart threaded through a custom-made 16 × 2 silicon tube array (100 µm diameter silicon tubing, Polymicro Technologies). Wires were then attached to a custom-fabricated electronic interface board (EIB, developed by Likhtik Lab) with gold pins. A male Omnetics connector (Intan) was then soldered onto the EIB, which the headstage was attached to during experiments. Before each experiment, an Omnetics headstage was connected to the recording implant. The headstage was connected to a standard serial peripheral interface (SPI) cable (Intan) which was attached to a commutator (Doric Lenses). An additional SPI cable connected the commutator to the Open Ephys acquisition box connected to a PC. Rats were then given 10 min to acclimate before the experiment would begin. Data was recorded at 30 kHz sampling rate using an Open Ephys system (acquisition board v2.2), run by Open Ephys software (Siegle et al., 2017). Since rats were well trained in the drinking paradigm, we recorded basal AIC firing for ∼3–4 min before placing one alcohol bottle on the home cage for 4 min. Following removal of the bottle, data was recorded for an additional 2 min.

Extraction and identification of single-unit waveforms

Raw, extracellular data were median referenced and subsequently spike-sorted using Kilosort2 with high-pass filter at 600 Hz. This was followed by exporting clustered units into Phy2 (https://github.com/cortex-lab/phy) for additional manual curation (Chen et al., 2021). Kilosort2-identified clusters were split into “good,” “mua,” or “noise.” “Good” clusters were further verified according to the following features: (1) distinct waveforms, (2) clear refractory period in cross-correlograms, and (3) spikes forming a clear circular cluster grouping. For manual curation, clusters were split into two or more cells if they displayed two clear groups. Spikes without clear waveforms, or with no refractory period in cross-correlogram, were classified as noise and removed. Firing rates <0.33 Hz were not considered. We also calculated % inter-spike interval violations (spikes closer than 2 ms) and excluded cells with >5% ISI violations. Clusters representing putative single units were then exported to custom-written MATLAB routines for further data analysis.

We note that there was some range in action potential waveforms, and thinner waveforms were moderately but significantly correlated with higher baseline firing activity (p = 0.0038; R² = 0.0069), but there were not clear delineations to indicate separate clusters of cells (data not shown). This is consistent with cells with thinner waveforms having higher firing rates, although the effects were modest, and further studies would be required to characterize potential firing patterns of putative AIC interneurons.

Identification of sustained firing cell phenotypes

Analysis methods largely followed those previously used (McCane et al., 2018; Linsenbardt et al., 2019; Morningstar et al., 2020). Analyses used custom routines in MATLAB. Firing rate data was extracted from separate stages of the recording session in 500 ms bins. Basal firing was determined as the average across the 65–5 s before bottles were placed on the home cage. We examine baseline up to 5 s before, rather than to the time of bottles on, because of possible anticipatory firing while watching the experimenter move toward the homecage before putting on the bottle. In addition, basal firing in a given AIC cell could show strong variability across time, with firing interspersed with gaps in firing (see examples in Fig. 2). For this reason, standard deviation in the baseline was quite high, making standard Z-score methods impractical.

Figure 2.

AIC plateau firing patterns: only cells with firing increases at drinking onset had higher plateau firing during compulsion-like intake. A, Examples of AIC cell firing with sustained-increase firing phenotype. Red marks indicate licks (e.g., with ∼1,000 licks in A, top). B, C, No differences across drinking conditions in sustained-increase cells in (B) preintake (baseline) or (C) plateau firing levels (shown for the first half of the drinking period). D, Examples of AIC sustained-decrease cells. E, F, No differences across drinking conditions in sustained-decrease cells in (E) preintake or (F) plateau firing levels. G, Examples of AIC cells with drinking-onset phenotype, with a large increase in firing at onset (first 2 s) of alcohol consumption. H–J, No differences across drinking conditions in drinking-onset cell firing during (H) preintake or (I) initial response, but (J) significantly greater plateau firing in drinking-onset cells in both AlcQuinine10 and AlcQuinine60 versus alcohol-only (p = 0.0372 alcohol-only vs AlcQuinine10, p = 0.0453 alcohol-only vs AlcQuinine60, p > 0.9 AlcQuinine10 vs AlcQuinine60). Postlick period: after a rat stopped licking, postlick firing was the average firing across the remainder of the 4 min access period (and thus a different length for each rat). After drinking, firing returned to near baseline, with no differences across drinking groups (sustained-increase: alcohol-only 7.06 ± 0.74 Hz, AlcQuinine10 6.82 ± 0.76 Hz, AlcQuinine60 7.18 ± 0.68 Hz, KW H₍₂₎= 0.685, p = 0.7101 across groups; sustained-decrease: alcohol-only 7.09 ± 0.47 Hz, AlcQuinine10 8.62 ± 0.73 Hz, AlcQuinine60 8.05 ± 0.61 Hz, KW H₍₂₎= 0.283, p = 0.8682 across groups; drinking-onset cells: alcohol-only 6.43 ± 0.52 Hz, AlcQuinine10 7.21 ± 0.60 Hz, AlcQuinine60 7.20 ± 0.64 Hz, KW H₍₂₎ = 1.857, p = 0.3952 across groups). K, No differences across drinking conditions in abundance of plateau cell types. Alcohol-only had 448 total cells, 174 drinking-onset, 116 sustained-increase, and 172 sustained-decrease. AlcQuinine10 had 378 total cells, 138 drinking-onset, 101 sustained-increase, and 152 sustained-decrease. AlcQuinine60 had 393 total cells, 127 drinking-onset, 110 sustained-increase, and 161 sustained-decrease. L, PCA of activity across all AIC cells during drinking, with ∼30% of variance explained by plateau firing changes; details in Materials and Methods. M, Nissl stains of eight examples of microwire placement, all from different rats. *p < 0.05, ***p < 0.001.

In order to better identify different firing changes in AIC cells, we utilized a criterion using percent change in firing for cells with >2 Hz basal firing and actual change for cells with <2 Hz basal firing (where percent values were less meaningful). In particular, cells were considered to show increased firing in a given period of interest (drinking-onset, or sustained across the session) if they had 30% greater than basal (for >2 Hz basal firing cells) or increased by 0.75 Hz or more (for <2 Hz basal firing cells). “Drinking-onset” cells were those that met criterion for firing increase during the first 2 s of licking. “Sustained-increase” cells were those that met criterion for firing increase across the whole drinking period (except for the first 2 s). Similar criteria were used for “Sustained-decrease” cells, with a 30% or 0.75 Hz decrease relative to baseline averaged across the whole drinking period (except first 2 s). Other criterion showed similar patterns, although, with fewer cells at higher change criterion, there were trends (data not shown). We note that there was some overlap of cell phenotypes, since cells with strong firing increase at drinking onset could also have a sustained increase, sustained decrease, or no change from baseline for the rest of the session (detailed further in Results).

Identification of lick-synchronized cells

For lick-related firing, peri-event histograms were constructed 400 ms before and after each lick and then summed over 1 ms bins for all licks. We examined the autocorrelation of spike counts (not smoothed) in close proximity to each lick. In addition, the resulting histograms were concatenated over all neurons and sessions, resulting in a neuron (row) × time (column) matrix upon which a principal component analysis (PCA) was performed. The first two principal components were considered, as they exhibited clear oscillatory entrainment to licking, and the first and second lick-related PC explained ∼28% and ∼21% of lick-related variance; the third PC explained ∼5% of variance and was similar to the second lick-synched PC (data not shown).

Analysis of sustained firing patterns across cells from different drinking conditions

For cells identified as drinking-onset, sustained-increase, and sustained-decrease, we performed several analyses. First, for each cell, we determined the average firing in each cell during several behavior periods: (1) the baseline (across 65–5 s before placing bottle on); (2) firing at drinking onset (first 2 s of licking); (3) firing across the first half, second half, and whole drinking period (where analyses here are shown for first half of drinking); (4) average firing after the end of licking; and (5) average firing during drinking bouts versus the interbout intervals, the pauses in licking. Average firing levels during the different behavior periods, and the percent of cells with a given plateau firing pattern, were compared across intake conditions separate from session or animal identity (see Statistical analysis below).

Also, to further assess the prominence of plateau-like firing patterns, a PCA was performed on the neural activity patterns across the entire drinking session. As there were differences across sessions in the timing of the stop of drinking, an analysis was required that aligned these timepoints. To accomplish this, each session was separated in 100 bins with the first and last 10 bins corresponding to the pre- and postdrinking periods. Spike counts were normalized to the width of each bin to ensure 1:1 comparison across sessions in the spike counts. Time-normalized spike count matrices were then concatenated across each session, and PCA was performed on this matrix. To ensure that the binning procedure did not artificially induce structure in the principal components, we shuffled the time normalized spike count bins within a neuron and then PCA as performed, which eliminated changes in neural activity associated with consumption (data not shown), confirming that the binning procedure did not artificially introduce structure. The PC shown in Figure 2L explained ∼30% of variance, while all other PCs explained <5% of variance (data not shown), for this particular assessment.

Analysis of lick-related firing patterns across cells from different drinking conditions

To analyze across drinking groups for each lick-related PC, the distribution of loading coefficients for each fluid type was examined for each PC by binning coefficients less than −0.1, between −0.1 and 0.1, and <0.1, resulting in a bin (rows) × fluid (column) matrix for each PC. A chi-square test of independence was used to determine if there were differences in the distribution across fluid types. Projections of each PC based on fluid type were assessed by multiplying row entries by the vector of coefficients for that principal component; cells for each drinking condition were projected on the PC separately (shown in Fig. 6B,F).

Figure 6.

Different lick-synchronized AIC firing patterns predicted compulsion-like versus overall alcohol drinking levels. A, Example raster and (B) combined data for neurons exhibiting strong loading on the principal component for the most abundant lick-synched pattern (∼28% of variance), which we called “alcohol-licking” patterns. B, No differences across drinking conditions for activity of alcohol-licking cells. C, D, Relation of drinking level to % of cells with alcohol-licking pattern within a session, for (C) compulsion-like and (D) alcohol-only conditions. Intake level and abundance of alcohol-licking cells were significantly correlated for AlcQuinine10 and AlcQuinine60, but with no differences in slope between alcohol-only and compulsion-like conditions. E, Example raster and (F) combined data for neurons with strong loading on the principal component for the second most abundant lick-synched pattern (∼21% of variance), which we called “compulsive-licking” patterns. F, Significant differences across drinking conditions for activity of compulsive-licking cells. G, H, Relation between drinking level and % of cells with compulsive-licking loading for (G) compulsion-like and (H) alcohol-only conditions. Intake level and abundance of compulsive-licking cells were significantly correlated for AlcQuinine10 and AlcQuinine60 and with significant differences in slope between alcohol-only and compulsion-like drinking. In A and E, spikes from strong positive loading neurons are in black, strong negative loading neurons are in red. *p < 0.05, **p < 0.01, ***p < 0.001; ^@ in H indicates p < 0.05 differences in slope relation across AlcQuinine60, AlcQuinine10, and alcohol-only.

Analysis of plateau and lick-related firing patterns across sessions

Importantly, in order to compare different firing measures to alcohol drinking level (and other alcohol-related behaviors), we also averaged firing parameters within each session. However, to compare firing patterns with consumption levels (and other alcohol-related behaviors, such as bout length and lick volume), inclusion of sessions with too few firing neurons could give imprecise estimates of the abundance and activity of AIC firing patterns within a given drinking session. Thus, for analyses relating firing to alcohol behaviors, we restricted analyses to sessions with 6 or more recorded cells, which yielded 24 alcohol-only, 24 AlcQuinine10, and 23 AlcQuinine60 sessions (reflecting a balance between number of cells recorded and having more sessions included). There was one outlier each for lick volume, average bout length, and sustained-increase cell firing increase (identified by Grubbs’ test), which were removed from analyses.

For plateau firing phenotypes (described above), we first determined the magnitude of firing change (vs preintake firing) in each cell of a given plateau phenotype. We then calculated the average firing change for each plateau phenotype within a given session (noting that some sessions did not have any cells with such plateau phenotypes). We also determined the percent of cells with a particular firing phenotype within a given session.

For lick-synched firing, specifically to assess whether a given neuron contributed to lick-synchronized firing, we considered a cell significantly loading on a given lick-related PC if the coefficient for that neuron was greater than ±1 standard deviation. Then, for each session, we determined the percent of cells within the session that had significant loading on the particular lick-related principal component.

Lickometry analysis

Licking patterns during experiments were assessed with custom-made capacitive lickometers, with custom-written C++ code running on an Arduino Uno with an Adafruit MPR121 capacitive touch sensing breakout board. A very small capacitive current was run through the 7.6 cm metal licking tube, and the discharge was detected when the rat's tongue contacted the metal tube. We note that this very small change in electrical charge did not disrupt ability to record neurons. Several groups have recorded in vivo firing during licking using very similar lickometers (Ottenheimer et al., 2018; Bari et al., 2019), including AIC firing during licking for tastants (Sadacca et al., 2016; Mukherjee et al., 2019). Even so, to ensure that spike trains were not contaminated by electrical artifacts associated with licking, an analysis of the relationship of spike times to lick times was performed. In this analysis, peristimulus histograms were constructed for 100 ms prior to and after each lick and binned at 1 ms. The histograms were then examined for strong peaks at time zero (e.g., when the animal licked). Seven of 1,218 neurons exhibited a peak at zero lag, and the number of spikes in this bin did not exceed 20 for any of these seven neurons. This indicated that lick-related artifacts could be detected in very few cells and, even when detected, accounted for a small fraction of the spikes observed in the cell. Finally, spike waveforms during baseline versus licking periods had R² > 0.9 correlation, further confirming that licking did not distort single-unit waveforms. Bout length and other lickometry measures were determined using custom-written programs in Python and Matlab.

Statistical analysis

Behavioral, neural firing, and statistical analyses used GraphPad Prism or custom-written programs in Python, Matlab, or R. Electrophysiology analyses largely followed previous methods (McCane et al., 2018; Linsenbardt et al., 2019; Morningstar et al., 2020). Nearly all data were non-normal (Shapiro–Wilk), with Kruskal–Wallis test (KW) and Dunn's post hoc to compare across drinking conditions, and Mann–Whitney (MW) and Wilcoxon for unpaired and paired comparison. Pearson's correlation examined relations between two variables. X² test assessed differences in abundance. We report uncorrected p values, and multiple corrections would apply for comparisons across equivalent conditions.

Analyses were performed separate from rat identity, as we (Darevsky et al., 2019; Darevsky and Hopf, 2020) and others (Griffin et al., 2007; Haggerty et al., 2022) have done, including for AIC firing (Pribut et al., 2021) (detailed validation in Darevsky and Hopf, 2020). In particular, we compared various firing patterns across drinking conditions, at the level of individual cells (separate from session). For some analyses (especially to compare intake level with firing measures), we averaged a firing measure across all cells in that session. Also, with multiwire recording, one cannot verify that the same neurons are recorded across days; thus, cells were treated as separate cells.

Post hoc values shown in figures are given in figure legends.

Histology

At the end of each study, rats underwent perfusion (4% paraformaldehyde, Santa Cruz) to determine electrode placement. Prior to brain extraction, a stimulator (World Precision Instruments) was used to run a small current through each electrode to lesion the location of the microwire endpoints. This led a small circular burn denoting the electrode tip. Placement of the microwires was determined by comparison of cresyl violet stained tissue against our previous AIC studies (Seif et al., 2013; De Oliveira Sergio et al., 2021), with examples shown in Figure 2M.

Results

Alcohol consumption was altered during quinine-challenged drinking

To examine AIC firing patterns related to alcohol intake, adult male Wistar rats first underwent ∼3 months of intermittent access (three 16–24 h intake sessions per week) and then ∼1 month of 20 min/d 5 d/week drinking (Fig. 1A), similar to our previous work (Seif et al., 2013; Darevsky et al., 2019; Darevsky and Hopf, 2020; De Oliveira Sergio et al., 2021). Also, before experimental tests, rats underwent several intake sessions to habituate to the novelty of alcohol with added quinine, including alcohol with 10 mg/L quinine (AlcQuinine10), which we term “moderate-challenge compulsion,” and alcohol with 60 mg/L quinine (AlcQuinine60), which we term “higher-challenge compulsion” (Darevsky et al., 2019; Darevsky and Hopf, 2020; Starski et al., 2023; Fig. 1B). Then, to identify AIC activity changes related to alcohol consumption, we recorded during several sessions of each drinking condition in 15 rats (in randomized order). Further, to have the most consistent comparison across drinking conditions, we selected two sessions from each drinking condition per rat, where we balanced having higher number of recorded cells and randomized order of drinking sessions. Finally, experimental intake sessions were 4 min long, as our rats typically exhibit strong initial consumption (“front-loading”; Seif et al., 2013; Darevsky et al., 2019; Darevsky and Hopf, 2020), as others observe (Jeanblanc et al., 2019; Flores-Bonilla et al., 2021).

In experimental sessions, alcohol drinking level (g/kg) was significantly reduced in AlcQuinine60 versus alcohol-only or AlcQuinine10 (Fig. 1C; KW H₍₂₎= 19.64; p < 0.0001), consistent with our previous findings (Hopf et al., 2010; Darevsky et al., 2019; Darevsky and Hopf, 2020), and we found a related pattern for total licks (Fig. 1D; KW H₍₂₎= 7.887; p = 0.0194). In addition, we previously found that a measure of lick volume was significantly less variable under both moderate-challenge and higher-challenge compulsion-like drinking, compared with alcohol-only (Darevsky et al., 2019; Darevsky and Hopf, 2020). Here, using mg/kg-per-lick (total mg/kg divided by total licks in a session) as a lick volume measure, we observed less variability in the lick volume measure across AlcQuinine10 and AlcQuinine60 sessions compared with alcohol-only sessions (Fig. 1E; p < 0.0001 Bartlett's test metric = 31.19; F test: alcohol-only vs AlcQuinine10: F_(1,29)= 5.215, p < 0.0001; alcohol-only vs AlcQuinine60: F_(1,29)= 6.246, p < 0.0001), in addition to reduced average lick volume under AlcQuinine60 (KW H₍₂₎= 15.20; p = 0.0005). Further, another behavioral variable of interest, average bout length, was not different across drinking conditions (Fig. 1F; KW H₍₂₎= 4.559; p = 0.1023). Finally, once the bottle was in place, rats drank quickly and steadily for several minutes (ranging from ∼30 to ∼330 s of relatively sustained drinking, with some brief, ∼2–3 s, pauses). However, there were no drinking condition differences in the licking speed across drinking conditions, examined as the average time between licks for the first 60 licks (Fig. 1G; KW H₍₂₎= 5.037; p = 0.0806). Thus, animals drank less under AlcQuinine60 challenge and had less variable lick volume under both compulsion-like conditions versus alcohol-only.

Sustained AIC firing patterns across alcohol intake sessions, with differences only in cells with firing increases at drinking onset

Recordings used custom 32-channel multiwire probes (Morningstar et al., 2020), yielding 448 alcohol-only, 378 AlcQuinine10, and 393 AlcQuinine60 recorded AIC neurons. As noted in Introduction, AIC neurons can have sustained firing patterns across a task, perhaps serving to maintain internal attention and/or motivation for the main task goal. Here, many AIC neurons showed sustained changes in firing across the entire drinking period, whether increased or decreased. Criteria for different plateau cell phenotypes are detailed in Materials and Methods and were determined relative to baseline, which was the averaged firing 65–5 s before the alcohol bottle was put on. In the examples of AIC firing shown in Figure 2A,D,G, we show firing across the entire drinking session. A red tick mark is shown for each lick, and the red marks run together, for example, where the top example trace in Figure 2A has ∼1,000 licks in that session. As noted in Discussion, further studies could examine firing at a more granular level, for example, in relation to the onset of drinking bouts within a session (since most rats licked in sustained bouts with some pauses, interbout intervals, across the session).

Some AIC cells exhibited a sustained-increase phenotype (examples in Fig. 2A), but there were no differences across alcohol drinking groups in these cells for basal (preintake) firing (Fig. 2B; KW H₍₂₎= 0.075; p = 0.9632) or plateau firing levels (Fig. 2C; KW H₍₂₎= 0.167; p = 0.9199; data for firing in first half of the consumption period). We also observed cells with a sustained-decrease phenotype (examples in Fig. 2D), which also showed no differences across drinking conditions in preintake (Fig. 2E; KW H₍₂₎= 0.550; p = 0.7597) or plateau (Fig. 2F; KW H₍₂₎= 0.241; p = 0.8865) firing levels. Once drinking overall ceased, firing in both cell types returned to near baseline, with no differences across drinking groups (detailed in figure legend). Thus, while many cells had sustained-increase and sustained-decrease phenotypes (Fig. 2K), they did not demonstrate differences in firing change magnitude across alcohol consumption conditions.

In strong contrast, a third AIC phenotype did have significantly higher firing under compulsion-like versus alcohol-only conditions. These neurons were identified by a strong firing increase at drinking onset (the first 2 s of intake), which we called the drinking-onset phenotype (DOP, examples in Fig. 2G). We note that there was some overlap of drinking-onset with sustained-increase and sustained-decrease phenotypes: drinking-onset cells had a strong initial firing increase and then could have increased, decreased, or no change in activity across the rest of the intake period (see below). While drinking-onset cells showed no differences across drinking groups in basal firing (Fig. 2H; KW H₍₂₎= 1.745; p = 0.4179) or firing during onset of consumption (average of first 2 s; Fig. 2I; KW H₍₂₎= 2.978; p = 0.2256), these drinking-onset cells showed significantly greater sustained, plateau activity under both AlcQuinine10 and AlcQuinine60 compared with alcohol-only, with no differences between AlcQuinine10 and AlcQuinine60 (Fig. 2J; KW H₍₂₎= 8.508; p = 0.0142; data shown for the first half of consumption). Firing in drinking-onset cells returned to near baseline once intake ceased (detailed in figure legend). Thus, cells with strong firing increases at drinking onset had greater sustained plateau activity under both moderate-challenge and higher-challenge compulsion-like conditions. Significant differences in drinking-onset cells concur with our behavioral evidence that, at the beginning of an intake session, rats quickly determine whether they are drinking under alcohol-only or compulsion-like conditions and then rapidly adjust to a different response strategy under compulsion relative to the one used under alcohol-only (Darevsky et al., 2019; Darevsky and Hopf, 2020; further described in Discussion). Our findings also agree with the observation of strong initial activity before sustained AIC increases that both relate to main task plan encoding in humans (Dosenbach et al., 2006).

Other studies find changes in the abundance of AIC firing phenotypes across different behavioral conditions (Guillem et al., 2010; Pribut et al., 2021). Thus, while there were no differences in sustained-increase or sustained-decrease cells in sustained, intake-related firing across drinking groups, these cells could influence consumption level if there were more or fewer of such neurons. However, ∼25% of cells were sustained-increase cells, ∼40% of cells sustained-decrease, and ∼35% of cells drinking-onset, with no differences in abundance across drinking condition (Fig. 2K; drinking-onset: X² = 4.016, df = 2, p = 0.1343; sustained-increase: X² = 0.443, df = 2, p = 0.8014; sustained-decrease: X² = 0.972, df = 2, p = 0.6152). Thus, the AIC had a similar number of sustained-increase, sustained-decrease, and drinking-onset cells, regardless of intake condition.

To better understand sustained firing changes, we ran a PCA to identify patterns of firing across AIC neurons during drinking. By far the strongest PC (∼30% of variance) was a sustained plateau (Fig. 2L), with no differences in the prevalence of this activity pattern across drinking conditions (Kolmogorov–Smirnov test; Alc vs AQ10: D₍₈₂₅₎= 0.0304, p = 0.9904; Alc vs AQ60: D₍₈₄₀₎= 0.0426, p = 0.8344; AQ10 vs AQ60: D₍₇₇₁₎= 0.0375, p = 0.9453). This is consistent with our analyses showing many AIC plateau firing cells, and together these data suggest that a majority of AIC cells showed sustained firing changes across alcohol intake. However, only cells with strong firing increases at the onset of consumption (i.e., drinking-onset cells) had significantly greater sustained firing during compulsion-like conditions.

Also, in addition to analyses described, we did some exploratory examination of firing patterns among different sustained-phenotype cells. First, to compare with firing changes during drinking onset period (first 2 s of access to drink alcohol) in DOP cells (Fig. 2I), we examined firing in the same early time interval for sustained-increase and sustained-decrease cells. For sustained-increase cells, overall firing was higher than baseline in all three drinking conditions (all Wilcoxon p < 0.0001), with no differences across intake conditions (KW = 0.011; p = 0.9947). Further, for sustained-decrease cells, overall firing at drinking onset was lower than that at baseline in all three drinking conditions (all Wilcoxon p < 0.0001), with no differences across intake conditions (KW = 2.928; p = 0.2313).

In addition, about half of DOP cells also exhibited the sustained-increase phenotype, with 47.1% of alcohol-only DOP cells, 52.2% of AlcQuinine10 DOP cells, and 61.4% of AlcQuinine60 DOP cells; this was marginally significant across drinking conditions (X² = 6.054; df = 2; p = 0.0485). In contrast, fewer DOP cells also exhibited sustained-decrease, with 19.0% of alcohol-only, 13.8% of AlcQuinine10, and 13.4% of AlcQuinine60 DOP cells also sustained-decrease, and with no differences across conditions (X² = 2.303; df = 2; p = 0.3162).

Example microwire placements from eight different rats are shown in Figure 2M. Also, while recording implants damaged overlying tissue, implants were only implanted unilaterally (in the right AIC), and alcohol drinking levels were not different before and after surgery (before, 1.02 ± 0.09 g/kg; after, 0.95 ± 0.09 g/kg; paired t test t₍₁₄₎= 1.783; p = 0.0963). Similarly, previous studies using the same implant methods (Linsenbardt et al., 2019) did not observe nonspecific behavioral effects of implantation.

AIC firing during alcohol-only intake was greater than that during saccharin

One important question is whether AIC firing might occur for any form of consumption, or be more specific to alcohol versus another consummant, such as sweet fluid. Indeed, AIC cells can show activity changes for tastes and other oral conditions (reviewed in Centanni et al., 2021). However, we (Seif et al., 2013; De Oliveira Sergio et al., 2021) and others (Kesner and Gilbert, 2007; Jaramillo et al., 2018a) found that AIC inhibition does not reduce intake of moderately sweet fluid. This is in strong contrast to substantial reductions in alcohol consumption during AIC inhibition (Seif et al., 2013; Jaramillo et al., 2018a; De Oliveira Sergio et al., 2021). We utilize 0.05% saccharin, chosen to give approximately the same intake volume for alcohol and sweet fluid (Seif et al., 2015; De Oliveira Sergio et al., 2021). Rats drank alcohol for 4+ months and then were trained to drink saccharin (see Materials and Methods). In four rats (with at least two sessions per solution per rat, yielding 149 alcohol-only cells and 96 saccharin cells), we compared AIC firing during alcohol-only versus saccharin drinking. With the smaller number of neurons for these data, we examined firing patterns across all neurons recorded (i.e., not related to the plateau phenotypes described above).

While total consumption was not significantly different for alcohol-only and saccharin sessions (alcohol-only, 1.78 ± 0.28 g/kg; sacc, 1.32 ± 0.46 g/kg; MW U₍₁₎= 38.0; p = 0.3363), AIC firing during saccharin drinking was significantly different from AIC activity during alcohol-only intake. We note that preintake firing was slightly but significantly lower in saccharin versus alcohol-only (Fig. 3A; alcohol-only, 7.84 ± 0.90 Hz; sacc, 6.62 ± 0.98 Hz; MW U₍₁₎= 5,850; p = 0.0160; although n.s. with multiple correction). We note that saccharin tests were conducted several hours earlier than alcohol, in the morning at end of the light cycle, while alcohol consumption was in the first hour of the dark cycle. Nonetheless, overall AIC firing activity was significantly and substantially greater during alcohol-only versus saccharin during initial firing (first 2 s of intake; Fig. 3B; alcohol-only, 7.50 ± 1.01 Hz; sacc, 4.51 ± 0.73 Hz; MW U₍₁₎= 5,326; p = 0.0007) and the first half of the drinking period (Fig. 3C; alcohol-only, 10.05 ± 1.18 Hz; sacc, 4.78 ± 0.69 Hz; MW U₍₁₎= 4,958; p = 0.0002). Also, when we subtracted drinking-related firing (in Fig. 3C) from preintake firing in each cell, we found significant differences in AIC firing during alcohol-only and saccharin (Fig. 3D; alcohol-only, 2.20 ± 0.62 Hz; sacc, −1.84 ± 0.39 Hz; MW U₍₁₎= 4,297; p < 0.0001). Thus, in concert with the AIC being unnecessary for promoting saccharin intake (at the concentration tested; De Oliveira Sergio et al., 2021), the AIC predominantly showed firing decreases during saccharin, different from alcohol-only. Similar AIC alcohol–sweet fluid differences were observed using c-Fos (Wandres et al., 2021). Also, while we did not determine specific cell phenotypes for these saccharin firing studies, it is likely that many more cells would be identified as sustained-decrease. As noted in Discussion, future studies using higher-density recording probes should examine these possibilities, including across a range of saccharin plus quinine doses.

Figure 3.

AIC firing was reduced during saccharin versus alcohol-only intake. Data shown as cumulative plots: all firing values in a given condition were ordered from smallest to largest and then were ranked up to 100%; the x-axis gives the firing value, and the y-axis describes what % of cells had this firing value or less. A, Saccharin preintake (basal) firing was slightly lower than alcohol-only, which may reflect saccharin intake occurring at the end of the light cycle (where alcohol-only occurred early in the dark cycle). B, Lick onset firing (first 2 s of drinking) and (C) firing during the first half of drinking were significantly lower during saccharin versus alcohol-only consumption. D, Changes in firing during first half of drinking minus basal (preintake) were significantly different between saccharin and alcohol-only. *p < 0.05, ***p < 0.001.

Only plateau cells with firing increases exhibited sustained firing during pauses in licking

The AIC has been linked to oral control (Whishaw and Kolb, 1983; Gutierrez et al., 2006; Stapleton et al., 2006; Sadacca et al., 2016; Mukherjee et al., 2019; Bouaichi and Vincis, 2020; Darevsky and Hopf, 2020; Neese et al., 2022; Starski et al., 2023), and several brain regions show decreases in firing during consummatory behavior (Taha and Fields, 2006; London et al., 2018). Thus, one possibility is that sustained-decrease cells have firing related to the act of consuming, while sustained-increase and drinking-onset have encoding related more to sustained action plans. To examine this possibility, we determined firing levels during pauses in licking, also called interbout intervals. Interestingly, firing in sustained-decrease cells was significantly higher during interbout intervals, compared with the decreased firing during actual intake (Fig. 4A; Wilcoxon ranked sum; comparing firing during drinking and interbout periods: alcohol-only W₍₁₎= 15,759, p < 0.0001; AlcQuinine10 W₍₁₎= 10,016, p < 0.0001; AlcQuinine60 W₍₁₎= 11,238, p < 0.0001). In other words, firing in sustained-decrease cells was significantly elevated during pauses in licking (moving up toward baseline). In contrast, drinking-onset cells maintained plateau firing levels during pauses between intake bouts (Fig. 4B; Wilcoxon: alcohol-only W₍₁₎= 2,402, p = 0.0786; AlcQuinine10 W₍₁₎= 718, p = 0.4519; AlcQuinine60 W₍₁₎=−1,450, p = 0.0776). Sustained-increase cells also maintained elevated firing during pauses in alcohol-only and AlcQuinine10 consumption, although with a small but significant firing drop during interbout intervals under AlcQuinine60 (Fig. 4C; Wilcoxon: alcohol-only W₍₁₎=−1,162, p = 0.1098; AlcQuinine10 W₍₁₎=−806, p = 0.1796; AlcQuinine60 W₍₁₎=−1,997, p = 0.0023; see Discussion). Thus, while activity changes in sustained-decrease cells reflected actual licking behavior, drinking-onset and sustained-increase cells maintained firing increases during pauses in intake, making them a better candidate for sustaining representation of the consumption-directed action strategy during pauses in action.

Figure 4.

Only cells with plateau increases maintained activity changes during pauses in drinking. A, Sustained-decrease cell firing increased during pauses in licking. B, C, Both (B) drinking-onset and (C) sustained-increase cells maintained plateau firing changes across pauses in licking, although with a small but significant drop under AlcQuinine60 in sustained-increase cells. **p < 0.01, ***p < 0.001.

Changes in AIC firing were related to bout length and lick volume

We next examined whether plateau firing measures might relate to drinking level and/or alcohol-related behaviors, especially average bout length and lick volume. For example, as noted above, lick volume was behaviorally less variable for both challenge conditions. As we will see, some aspects of AIC firing did predict alcohol-directed behavior but might play a more permissive, indirect role for consumption level. Also, our previous work (Darevsky et al., 2019; Darevsky and Hopf, 2020) suggests that the AIC may maintain commitment to respond for alcohol despite challenge and similarly so under moderate and higher challenge (detailed further in Discussion), for example, as in Figure 2J. Indeed, since one main interest is the potential difference(s) in firing–intake relationships between alcohol-only and compulsion more generally, for some correlational analyses we combined moderate- and higher-challenge intake conditions. In addition, with variability across sessions, even with the larger sample sizes we have, for some analyses we combined sessions from all drinking conditions. Also, for these more exploratory investigations of AIC firing patterns related to drinking level and associated behaviors, we report statistics for number of conditions, while we only show a subset of the data.

We first examined average bout length in a session, as many studies observe that longer bouts predict greater consumption (reviewed in Starski et al., 2023). Here, longer bouts predicted greater drinking levels for all sessions (F_(1,68)= 15.08; p = 0.0002), compulsion-like conditions combined (Fig. 5A; F_(1,45)= 12.87; p = 0.0008), and AlcQuinine60 (Fig. 5B; F_(1,21)= 7.375; p = 0.0129), but not alcohol-only (Fig. 5A; F_(1,21)= 2.600; p = 0.1218) and a trend for AlcQuinine10 (Fig. 5B; F_(1,22)= 3.979; p = 0.0586). However, with no difference in slope relationship between compulsion and alcohol-only (F_(1,66)= 0.8716; p = 0.3542), we concluded only that longer bout lengths overall predicted greater intake.

Figure 5.

Behavioral measures (average bout length, lick volume) related to specific aspects of plateau firing. A, B, Relation of average bout length and drinking level, for (A) compulsion-like conditions combined and alcohol-only and (B) AlcQuinine10 and AlcQuinine60. C, D, Bout length overall correlated with greater firing increases in (C) drinking-onset and (D) sustained-increase cells. For drinking-onset cells, this was not significant for alcohol-only (F_(1,16)= 2.624; p = 0.1248) or compulsion-like drinking (F_(1,36)= 3.886; p = 0.0564) tested separately. For sustained-increase, this was significant for compulsion-like conditions combined (F_(1,28)= 8.301; p = 0.0075), but not alcohol-only (F_(1,13)= 0.970; p = 0.3428). E–G, Relation of lick volume (mg/kg intake per lick) and drinking levels. H, I, The magnitude of firing changes in sustained-decrease cells correlated with lick volume for (H) compulsion-like drinking combined but not (I) alcohol-only. J, Firing changes in drinking-onset cells did not relate to lick volume. K, Average firing changes in sustained-decrease cells (post hoc p = 0.1808 alcohol-only vs AlcQuinine10). L, Firing changes in sustained-decrease cells did not correlate with intake levels. *p < 0.05, **p < 0.01, ***p < 0.001; ^@p < 0.05 differences in slope relation across AlcQuinine60, AlcQuinine10, and alcohol-only.

Since bout length could predict consumption levels, we compared average bout length in a session with changes in firing in different AIC plateau phenotypes. Across all sessions, longer bouts correlated with greater firing increases in drinking-onset cells (Fig. 5C; F_(1,54)= 6.481; p = 0.0138) and sustained-increase cells (Fig. 5D; F_(1,43)= 8.454; p = 0.0057) but not sustained-decrease cells (F_(1,66)= 0.280; p = 0.5988). However, alcohol consumption levels were not related, across all sessions, to firing changes in drinking-onset cells (F_(1,55)= 0.1757; p = 0.6767) or sustained-increase cells (F_(1,44)= 2.300; p = 0.1365). Bout length also did not correlate with the percent of cells in a given session with drinking-onset (F_(1,68)= 2.348; p = 0.1301) or sustained-increase phenotype (F_(1,68)= 2.057; p = 0.1561). Thus, greater firing increases in drinking-onset and sustained-increase cells predicted longer bouts, but any relation to actual intake would be indirect.

We next examined lick volume (mg/kg-per-lick within a given session), since oral control has been related to AIC/ventral frontal cortex (see above). Larger lick volume correlated with significantly greater consumption for AlcQuinine60 (Fig. 5E; F_(1,21)= 16.80; p = 0.0005) but not AlcQuinine10 (Fig. 5F; F_(1,21)= 1.787; p = 0.1949) or alcohol-only (Fig. 5G; F_(1,21)= 0.568; p = 0.4593) but did correlate for all sessions combined (F_(1,68)= 17.53; p < 0.0001) and compulsion-like conditions combined (F_(1,68)= 17.53; p < 0.0001). The lick volume–intake slope relationship was significantly different across the three drinking conditions (F_(2,64)= 3.499; p = 0.0361) and was nearly significant for alcohol-only versus compulsion combined (F_(1,66)= 3.777; p = 0.0562). Thus, lick volume correlated with consumption level for compulsion-like intake under higher-challenge conditions.

Lick volume showed some relationships with plateau firing, but with a different pattern from bout length. Specifically, the firing change magnitude in sustained-decrease cells correlated with lick volume for compulsion-like conditions combined (Fig. 5H; F_(1,44)= 5.388; p = 0.0250) but not alcohol-only (Fig. 5I; F_(1,20)= 0.058; p = 0.8144), although the slope relationships were not different across intake conditions (F_(1,64)= 0.366; p = 0.5473). In addition, lick volume was not related to firing changes in drinking-onset (Fig. 5J; F_(1,55)= 0.177; p = 0.6759) or sustained-increase cells (F_(1,44)= 0.042; p = 0.8383). Further, the average firing change in sustained-decrease cells differed significantly across drinking groups (Fig. 5J; KW H₍₂₎= 7.137; p = 0.0282; post hoc AlcQuinine10 vs AlcQuinine60 p = 0.0304), but with no differences in firing changes for drinking-onset (KW H₍₂₎= 1.987; p = 0.3704) or sustained-increase cells (KW H₍₂₎= 0.417; p = 0.8119). Taken together, these findings, along with those in Figure 4A, suggest a more specific relationship between sustained-decrease cells and licking-related measures. However, firing changes in sustained-decrease cells were not themselves correlated with alcohol intake level (Fig. 5L; F_(1,67)= 0.504; p = 0.4802), suggesting at best an indirect or permissive relationship between sustained-decrease firing and drinking level under compulsion-like drinking.

Lick-synchronized firing predicted alcohol consumption levels, with one lick-related pattern selective for compulsion-like drinking

In addition to sustained firing patterns, many groups have observed AIC firing synchronized to the act of licking (Stapleton et al., 2006; Sadacca et al., 2016; Mukherjee et al., 2019; Bouaichi and Vincis, 2020; Neese et al., 2022). To explore this possibility, we examined the autocorrelation of spike counts (not smoothed) in close proximity to each lick (±400 ms), determined across the entire drinking sessions. PCA across licks, cells, and drinking groups was then used to identify autocorrelation patterns that might reflect licking. Robust oscillatory relationships were observed and demonstrated significant AIC activity changes synchronized to licking. One lick-synched PC explained ∼28.8% of variance, while the second lick-synched PC explained ∼21.1% of variance (example raster plots in Fig. 6A,E; projection of firing, stratified by drinking condition, in Fig. 6B,F; see Materials and Methods); as described below, we call cells of the first PC “alcohol-licking” and cells in the second PC “compulsive-licking.” Then, to assess whether a given neuron contributed to lick-synchronized firing, we considered a cell significantly loading on a given lick-synched PC if the coefficient for that neuron was greater than ± 1 standard deviation (see Materials and Methods). Overall, 124 cells loaded on the alcohol-licking phenotype, and 119 cells loaded on the compulsive-licking phenotype (∼10.2% and ∼9.8% of recorded cells, respectively). Also, 65/119 of alcohol-licking cells (∼52.4%) also loaded on compulsive-licking, significantly different from chance (X² = 284.9; df = 1; p < 0.0001); thus, ∼14.6% of all recorded AIC cells loaded on alcohol-licking and/or compulsive-licking phenotypes.

We first determined whether there might be overall differences in the distribution of loading of lick-synched firing across drinking conditions. Alcohol-licking cells had a trend for differences across intake conditions (X² = 8.432; df = 4; p = 0.0770), while compulsive-licking showed significant differences across drinking conditions (X² = 162.1; df = 4; p < 0.0001). Thus, the number of cells expressing the compulsive-licking phenotype differed across the different drinking conditions.

We then examined whether the alcohol consumption level in a given session correlated with the percentage of cells that loaded on alcohol-licking or compulsive-licking (% of cells with alcohol-licking or compulsive-licking phenotype in that session). Sessions with greater % of alcohol-licking cells had significantly higher intake across all sessions (F_(1,69)= 11.92; p = 0.0010), AlcQuinine10 (Fig. 6C; F_(1,22)= 4.929; p = 0.0370), AlcQuinine60 (Fig. 6C; F_(1,21)= 7.990; p = 0.0101), but not alcohol-only (Fig. 6D; F_(1,22)= 1.321; p = 0.2781). However, there were no differences between the firing–intake slope relationship when comparing alcohol-only versus compulsion-like conditions combined (F_(1,67)= 1.224; p = 0.2724), or with alcohol-only, AlcQuinine10, and AlcQuinine60 separated (F_(2,65)= 0.987; p = 0.3815). Thus, sessions with more alcohol-licking cells overall had significantly higher drinking levels, with trends but no significant differences across alcohol intake conditions (which is why we termed them “alcohol-licking” cells).

Interestingly, for compulsive-licking cells, the firing–intake relationship was significantly different between alcohol-only and compulsion-like conditions (which is why we termed them “compulsive-licking” cells). Sessions with a greater % of cells with compulsive-licking phenotype had significantly higher drinking levels for AlcQuinine10 (Fig. 6G; F_(1,22)= 11.17; p = 0.0029) and AlcQuinine60 (Fig. 6G; F_(1,21)= 7.868; p = 0.0106), but not alcohol-only (Fig. 6H; F_(1,22)= 0.231; p = 0.6356). Importantly, there was a significant difference in the firing–intake slope relationships across alcohol-only, AlcQuinine10, and AlcQuinine60 (F_(2,65)= 3.313; p = 0.0427). This was also observed when compulsion-like conditions were combined and the firing–intake slope was compared with alcohol-only (F_(1,67)= 6.129; p = 0.0158). Together, these results indicate that having more compulsive-licking cells in a session predicted significantly higher drinking for both moderate- and higher-challenge compulsion for alcohol but did not relate to alcohol-only level. It is also interesting that the “alcohol-licking” firing was better aligned to the action of licking (Fig. 6B), while the “compulsive-licking” activity peak was somewhat offset from lick event (Fig. 6F). In previous studies, there are AIC cells with firing related to rapid sensing of tastants, and the peak activity of these cells is offset from licks themselves; in contrast, other AIC cells show peak firing at lick events (Stapleton et al., 2006; Bouaichi and Vincis, 2020; Neese et al., 2022). Thus, what we call compulsive-licking patterns may be more related to modulation of taste input (perhaps related to “active sensing”; Graham et al., 2014; Neese et al., 2022), while alcohol-licking patterns might be more related to the act of licking itself.

While these findings link lick-related cell firing to total consumption level, another lick measure, lick volume, did not correlate with % of cells with alcohol-licking (F_(1,68)= 0.011; p = 0.9152) or compulsive-licking phenotype (F_(1,68)= 0.006; p = 0.9374), across all sessions. However, bout length did correlate with compulsive-licking cell abundance for compulsion-like intake (F_(1,45)= 5.233; R² = 0.104; p = 0.0269) but not alcohol-only (F_(1,21)= 1.170; R² = 0.053; p = 0.2916). This was specific for compulsive-licking, since bout length did not relate to abundance of alcohol-licking cell during compulsion-like drinking (F_(1,45)= 2.347; R² = 0.050; p = 0.1326) or alcohol-only (F_(1,21)= 0.279; R² = 0.013; p = 0.6032).

Some overlap of plateau and lick firing phenotypes

One additional question is whether individual lick-encoding cells also exhibited plateau firing patterns. As shown in Table 1, both alcohol-licking and compulsive-licking cells had a greater than chance overlap for expressing drinking-onset and sustained-increase phenotypes and a lower than chance overlap with sustained-decrease phenotype. Even so, the overlap was partial (and see Bouaichi and Vincis, 2020), suggesting that different AIC cell populations still had the potential to differentially encode plateau and lick firing phenotypes.

Table 1.

Overlap of lick-synched and sustained plateau firing patterns

Plateau type	% cells alcohol-licking with a given plateau type	% cells not alcohol-licking with a given plateau type	X 2 (df = 1)	p value
Drinking-onset	46.8%	34.8%	6.897	p = 0.0086
Sustained-increase	50.0%	24.2%	37.68	p < 0.0001
Sustained-decrease	23.4%	41.4%	15.27	p < 0.0001
	% cells compulsive-licking with a given plateau type	% cells not compulsive-licking with a given plateau type
Drinking-onset	46.2%	34.9%	5.924	p = 0.0149
Sustained-increase	54.6%	23.8%	51.80	p < 0.0001
Sustained-decrease	21.8%	41.5%	17.47	p < 0.0001

One important question is whether cells with lick-synched firing also exhibited plateau firing patterns. Both alcohol-licking and compulsive-licking cells had greater than chance overlap with drinking-onset and sustained-increase phenotypes (compared with cells without lick-synched firing) and lower than chance overlap with sustained-decrease phenotype.

Discussion

The AIC is implicated in alcohol compulsion in humans and rodents and motivated responding more generally. Thus, we sought to fill a major gap in understanding AIC activity patterns that promote pathological alcohol intake. Lick-synchronized AIC cells had the strongest relationship to consumption levels, and one lick-related pattern predicted greater compulsion-like drinking, but not alcohol-only (called “compulsive-licking” cells). A different lick-synched pattern correlated with higher alcohol intake overall (“alcohol-licking” cells). Importantly, lick-related firing was determined for all licks in each session, suggesting that the AIC mediated a session-long strategy where greater lick-synched firing drove greater compulsion-like intake. Importantly, other firing phenotypes exhibited sustained firing changes across the several-minute consumption period. However, only drinking-onset (DOP) AIC neurons, with strong activity increases at onset of alcohol consumption, had greater protracted activity across compulsion-like drinking sessions. This concurs with our previous findings that rats quickly evaluate and adjust intake strategy under compulsion-like conditions (Darevsky et al., 2019; Darevsky and Hopf, 2020) and to our proposal that compulsion involves an action strategy with attention focused internally on generating more stereotypical action. This is also consistent with the suggested importance of automaticity for habit and compulsion (Koob and Volkow, 2010; Voon et al., 2015; Ersche et al., 2017), and greater attention on action would also help decrease attention on (and impact of) negative consequences (Starski et al., 2023). Similarly, “intake defense” proposes that, for example, eating rotten food when starving, involves an overall goal to consume a “sufficient” amount, not to finely titrate amount of intake (Kaplan et al., 2001; Darevsky et al., 2019; Darevsky and Hopf, 2020). Thus, while the level of plateau increase in DOP cells did not predict alcohol consumption level, the AIC and related regions have been related to maintaining the overall reward-directed strategy (see Introduction). Along with trans-species studies showing AIC importance for alcohol drinking, our findings here support our overarching hypothesis that different cell phenotypes impact and promote various aspects of consumption, including the likely central role of sustained AIC firing changes. Figure 7 summarizes our main findings and how each different firing phenotypes might promote compulsion-like (Fig. 7, green/orange lines) or overall alcohol (Fig. 7, black lines) drinking.

Figure 7.

Main relationships between AIC firing patterns and alcohol intake, average bout length, and lick volume. Thick and thin lines indicate p < 0.001 and p < 0.05, respectively, for a significant relationship between the two measures shown (from data reported in other figures or text). Black line, significant for all alcohol drinking conditions (alcohol-only and compulsion-like); green line, significant for compulsion-like conditions combined; orange line, significant for higher-challenge drinking.

We focus on the AIC with its link to many aspects of alcohol drinking (Centanni et al., 2021), including how alcohol cues (Myrick et al., 2004; Claus et al., 2011) and negative affect (Chester et al., 2016) relate to real-world consumption. In rats, AIC–nucleus accumbens (Seif et al., 2013) and AIC–locus ceruleus projections (De Oliveira Sergio et al., 2021) strongly promote compulsion-like intake, while neither regulated alcohol-only. In concurrence, AIC–striatal circuit activity in heavy-drinking humans relates to punishment-resistant responding and subjective compulsivity for alcohol (Grodin et al., 2018) and imagining high-risk drinking (Arcurio et al., 2015). This agrees with clinical theories that compulsion-like responding is, at core, about conflict (intoxicant desire vs negative consequences) and the presence of conflict recruits cortical conflict-processing systems including AIC circuits (Tiffany and Conklin, 2000; Naqvi and Bechara, 2010). However, global AIC inhibition also suppresses rat alcohol-only and compulsion-like consumption (De Oliveira Sergio et al., 2021).

Thus, we sought to provide an integrated model for how different AIC contributions impact drinking, with some driving general intake, and others selectively promoting compulsion. For example, neurons with high activity at drinking onset (DOP cells) were the phenotype that showed differences with compulsion, especially a sustained increase in activity across the intake period. Another session-long AIC pattern, synchronized to licking, correlated with greater compulsion-like but not alcohol-only drinking. However, many important questions remain and would likely be better answered by future studies using costly high-density silicon probes, which provide greatly increased cell yield. In this way, we can better understand population and individual cell activity related to the level of drinking (and other behaviors). For example, cells with strong, transient increases in AIC firing at drinking onset (DOP cells) are of particular interest, including where more prolonged early firing (initial 10–15 s firing) in DOP cells might promote greater drinking (vs our 2 s assessment period). Also, DOP and sustained-increase phenotypes overlap (∼50%) and sustained-increase cells that do or do not overlap with DOP might have differing phenotypes. In addition, DOP and sustained-increase cells maintained high firing across pauses in drinking (2–4 s interbout intervals), perhaps indicating sustained motivation for alcohol across breaks. However, unlike DOP cells, sustained-increase cells did show a small but significant decrease in firing under higher-challenge, perhaps contributing to premature termination of drinking. Future studies should also better examine AIC encoding for intake of different levels of quinine-adulterated saccharin to help assess primary reward–cost balance activity (Seif et al., 2013, 2015), especially the behavioral importance of interbout AIC firing. Thus, our findings created an important foundation for critical future questions about how AIC firing phenotypes interact to regulate aspects of alcohol drinking (and motivation- and affect-related behavior more generally).

Our strongest finding is that one phenotype of lick-synched firing related to higher compulsion-like intake, while another lick-synched phenotype correlated with alcohol drinking more generally. Along with AIC necessity for alcohol-only and compulsion like intake, it is interesting that AIC/ventral frontal cortex is associated with tongue control (Brimley and Mogenson, 1979; Shipley et al., 1980; Whishaw and Kolb, 1983; Whishaw and Tompkins, 1988; Gutierrez et al., 2006). Indeed, our previous behavioral work found that tongue control, assessed by lick volume, is significantly less variable under both moderate- and higher-challenge when compared with alcohol-only (Darevsky et al., 2019; Darevsky and Hopf, 2020). Thus, even though higher-challenge reduced intake (here and Darevsky and Hopf, 2020) and disrupts lick timing and bout organization (Darevsky and Hopf, 2020), some aspects of behavior and firing still showed similar patterns under both moderate and higher challenge. Thus, we propose a novel AIC role, providing sustained commitment to respond for high-value reward, regardless of challenge level; we speculate that disorganized responding under higher challenge reflects mPFC disruption (Darevsky and Hopf, 2020; Starski et al., 2023). This model is also consistent with the AIC as input/integrator and mPFC as motor output of the salience network (Craig, 2009; Menon and Uddin, 2010).

Our findings may also inform recent studies of larger-scale alcohol-related changes in AIC connectivity (Sommer et al., 2022). Lesions in insula-connected regions improved remission from smoking and drinking (Joutsa et al., 2022). Also, heavy-drinking humans early in withdrawal show overall decreases in connectivity, leaving the AIC with greater centrality and influence, an “exaggerated integration of interoceptive states” that drives relapse (Bordier et al., 2022). Rats show related alcohol changes, with more general disrupted or altered connectivity, but increased connectivity of salience and reward areas, including AIC–nucleus accumbens (Scuppa et al., 2020; Perez-Ramirez et al., 2022), a circuit which promotes compulsion-like drinking in rats (Seif et al., 2013). With these and other AIC connections (e.g., to bed nucleus of the stria terminalis; Flook et al., 2021), future work should examine projection patterns of AIC cells with different alcohol-related firing patterns and how this may change across withdrawal. Also, with such small differences in plateau firing during compulsion-like versus alcohol-only consumption, the use of c-Fos activation to identify compulsion-related brain circuits (Domi et al., 2021) might easily miss the AIC.

We are early in understanding how insula function may differ in females and males, including for alcohol (Radke et al., 2021), and the lack of females is an important limitation of this study. Limited human studies link a similar AIC circuitry to alcohol compulsion in both sexes (Arcurio et al., 2015; Grodin et al., 2018). However, sex differences in alcohol conditions are observed in important insula areas. For example, there are notable sex differences in alcohol-related plasticity in posterior insula connections to the bed nucleus of the stria terminalis in mice (Marino et al., 2021). In addition, for AIC-alcohol mechanisms in mice, disruption of AIC perineuronal nets reduces male but not female compulsion-like intake and sex-dependent alcohol-related AIC c-Fos (Chen and Lasek, 2019; Martins de Carvalho et al., 2023). However, GABA agonist infusion in the AIC in female rats reduced both alcohol-only and compulsion-like drinking (unpublished), as in males (De Oliveira Sergio et al., 2021). Thus, there may be some species differences, especially with AIC's relation to emotional regulation (Centanni et al., 2021) and different basal temperaments in mice and rats, as well as differences in drinking history (several days in Chen and Lasek, 2019; Martins de Carvalho et al., 2023, <5 months here). Thus, future work should record female AIC to better understand sex-different and sex-similar brain mechanisms for alcohol.

Problematic alcohol drinking causes major social, physical, and financial harms, and the AIC is strongly linked to addiction, making it critical to understand AIC firing that might promote pathological intake. We discovered multiple AIC activity patterns, some that could promote intake more generally, while compulsion-like consumption was associated with selective lick-synched firing activities and greater sustained activity in cells with strong firing increases at onset of intake. Together, we provide valuable, novel information, and an integrated model for how the AIC can drive pathological intake, which is likely relevant for AIC promotion of cost- and challenge-resistant behavior more generally (Starski et al., 2023).

Data Availability

Experimental data and analysis scripts are available from the communicating author upon request.