Inflammatory endotoxin challenge in individuals with alcohol use disorder and controls

Abstract

Background:

Preclinical and clinical research reveal associations between chronic alcohol use, increases in proinflammatory cytokines (interleukin (IL)-6, IL-8, tumor necrosis factor alpha (TNF-α)), and increases in alcohol consumption, alcohol craving, and negative mood. However, these findings remain largely correlational in clinical samples. Therefore, we conducted a preliminary inflammatory challenge using endotoxin in individuals with alcohol use disorder (AUD) to investigate the immune, behavioral, and brain responses to inflammatory challenge.

Methods:

Participants were randomly assigned to receive a bolus intravenous injection of either low-dose endotoxin (0.8 ng/kg of body weight) or placebo (same volume of 0.9% saline). Blood samples, sickness symptoms, physiology, mood, and alcohol craving were collected at baseline and hourly for 4-hours post-baseline, with a neuroimaging scan occurring at 3 hours post-baseline. Matched control data was used to validate the endotoxin challenge in comparison to the AUD sample.

Results:

Endotoxin led to an acute blunted pro-inflammatory (i.e., TNF-α, IL-6, and IL-8) response in individuals with AUD compared to controls (all p’s<0.039). Endotoxin led to decreased cue-induced craving in both the behavioral human laboratory (p=0.03) and neuroimaging (p’s<0.01) assays. Moreover, higher levels of endotoxin-induced IL-6 were most negatively associated with decreased self-reported craving following baseline (p<0.05) in comparison to lower levels of endotoxin-induced IL-6.

Conclusions:

This preliminary study provides an acute experimental manipulation of inflammatory processes associated with AUD and suggests that the short-term effects of inflammation in AUD phenomenology are multifaced and dose dependent.

INTRODUCTION

A comprehensive understanding of alcohol use disorder (AUD) is dependent upon identifying pathways underlying alcohol use in humans. One such pathway is inflammation, as indicated by the neuroimmune hypothesis of AUD, which posits that alcohol use activates innate immune signaling in the brain to further drive alcohol seeking (Cui et al., 2014; Mayfield and Harris, 2017). The inflammatory hypothesis is supported by both preclinical and clinical research indicating associations between chronic alcohol use and inflammatory cytokines (specifically interleukin (IL)-6, IL-8, tumor necrosis factor alpha (TNF-α)) (Achur et al., 2010; Heberlein et al., 2014; Hillmer et al., 2020). However, while preclinical research utilizes experimental methods such as endotoxin challenges to understand the inflammatory pathway in alcohol use (Afshar et al., 2015; Burnette et al., 2023), clinical research remains largely correlational (Burnette et al., 2021). Therefore, the present study sought to advance the clinical translation of the role of inflammation in alcohol use by conducting a preliminary novel endotoxin challenge in humans with AUD.

The molecular impact of alcohol use on the innate immune system occurs through the effect of alcohol on pattern recognition receptors, specifically toll-like receptors (TLRs), which bind to pathogen-associated molecular patterns (PAMPs) and modulate the innate immune response. TLR4 in particular is bound by bacterial endotoxin or lipopolysaccharide (LPS), a PAMP in the gut (Leclercq et al., 2014; Wang et al., 2010). Alcohol administration in healthy humans modulates endogenous endotoxin levels (Afshar et al., 2015; Monnig et al., 2020). Chronic alcohol use is related to higher levels of serum endotoxin (Liangpunsakul et al., 2017) and young adults who drink at binge-levels have increased plasma endotoxin levels (Orio et al., 2018). This elevation in endotoxin may be due to higher levels of gut permeability and subsequent translocation of endotoxin into the blood (Bala et al., 2014; de Timary et al., 2017; Sturm et al., 2022). Endotoxin binds TLR4, leading to the activation of transcription factors (Crews et al., 2017; Richez et al., 2009) that regulate the expression of pro-inflammatory cytokines, or signaling protein molecules, released by immune cells. These cytokines include TNF-α, IL-6, and IL-8 (Leclercq et al., 2014; Lu et al., 2008), which are cytokines consistently implicated in alcohol-use research (Adams et al., 2020) . Importantly, cytokines typically thought of as being “pro-inflammatory,” such as IL-6, may additionally regulate anti-inflammatory immune responses (Scheller et al., 2011). Moreover, the binding of endotoxin leads to a robust inflammatory response that induces the release of inflammatory cells including TNF-α, IL-6, chemokines (C-X-C-motif) ligand (CXCL)8, IL-10, IL-1β, Macrophage Inflammatory Protein (MIP)-1α, MIP-1β, and Monocyte Chemoattractant Protein (MCP)-1. This inflammatory cascade includes both pro- and anti-inflammatory responses (Kiers et al., 2017). As such, cytokines and the inflammatory cascade resulting from endotoxins are complex.

Chronic alcohol consumption in mice and chronic alcohol application to human blood monocytes increases the sensitivity of TLRs, leading to increased expression of pro-inflammatory cytokines, including TNF-α (Gustot et al., 2006; Kany et al., 2019; Pang et al., 2011). Clinically, adults without AUD who drink at binge-levels have increased expression of TLR4 and higher levels of pro-inflammatory cytokine expression (TNF-α, IL-6) (Bala et al., 2014; Orio et al., 2018). Moreover, individuals who drink alcohol heavily experience increases in cytokine IL-6 over time (Bell et al., 2017). Taken together, manipulating alcohol exposure alters inflammation, and heavy alcohol use may increase peripheral inflammation (Albert et al., 2003; Bell et al., 2017).

This relationship between chronic alcohol use and inflammation is bidirectional, as reciprocal effects may emerge where inflammation alters alcohol-associated behaviors (Blednov et al., 2011). For example, plasma levels of cytokines typically associated with pro-inflammatory responses (TNF-α, IL-6, IL-8) are associated with severity of alcohol problems (Achur et al., 2010; Heberlein et al., 2014). Males with AUD experiencing withdrawal from alcohol have higher serum levels of TNF-α and IL-6 that are associated with continued active drinking and craving for alcohol (Heberlein et al., 2014). Further, adults with AUD experience higher alcohol craving in association with higher levels of IL-8 (Hillmer et al., 2020).

Experimental data using endotoxin in adults without AUD supports these clinical observations. Increased levels of peripheral cytokines following an inflammatory challenge are associated with subsequent increases in negative mood (Reichenberg et al., 2001; Wright et al., 2005) and social disconnection (Eisenberger et al., 2010; Moieni et al., 2015a), both of which are features of depression, which in turn are associated with greater severity of alcohol use disorders (McHugh and Weiss, 2019). Behavioral alterations from inflammation are corroborated by inflammatory-driven neural modulations. Preclinically, activation of proinflammatory cytokines may impair the blood-brain barrier (Friske et al., 2025), resulting in neural damage that leads to neuroinflammation driving increased alcohol consumption, binge drinking, and withdrawal-related anxiety (Leko et al., 2023). Clinically, elevated peripheral inflammation may be associated with alterations in resting state functional connectivity (Kang et al., 2025). Disruptions in resting state brain connectivity may increase susceptibility to drug craving and negative mood (Zhang and Volkow, 2019). Individuals who drink alcohol at binge-levels have increased IL-6, but not TNF-α levels, compared to individuals that drink alcohol socially, and this IL-6 increase is positively associated with alcohol craving and increased neural activation in the ventromedial prefrontal cortex during a neural alcohol cue reactivity task (Blaine et al., 2023). The bidirectional relationship between inflammation and alcohol may affect both brain and behavior.

Despite evidence linking inflammation and alcohol use, findings in samples with AUD are largely correlational (Burnette et al., 2021). The endotoxin challenge can be leveraged to translate preclinical findings by experimentally inducing a dose-respondent, inflammatory response (Burnette et al., 2021; Suffredini and Noveck, 2014). Preclinically, endotoxin stimulation of blood samples taken from intoxicated binge-drinking healthy humans leads to a brief phasic pro-inflammatory cytokine response as shown by increases in TNF-α, followed by a delayed anti-inflammatory response as shown by attenuations in IL-1β levels (Afshar et al., 2015). Endotoxin stimulation of monocytes from individuals with AUD leads to increased production of TNF-α and co-production of TNF-α/IL-6 compared to controls (Burnette et al., 2023). In healthy young adults, the utility of endotoxin as an experimental probe of inflammation has been convincingly demonstrated. At low doses (between 0.4 and 1.0 ng/kg body weight), endotoxin produces a safe, dose-respondent inflammatory response that increases peripheral pro-inflammatory cytokine levels (Suffredini and Noveck, 2014) where the intensity of symptoms and cytokine response peak 1.5-2 hours post-infusion (Fullerton et al., 2016). Endotoxin administration increases temperature, heart rate, and blood pressure in association with increased levels of TNF-α and IL-6 (Moieni et al., 2015c). Despite the interest in a neuroimmune hypothesis for AUD, experimental administration of endotoxin in clinical samples is lacking. To date, only one trial has attempted to conduct an endotoxin challenge with individuals with AUD, but the trial was prematurely terminated due to adverse events related to the combination of endotoxin with the study medication, pioglitazone (Schwandt et al., 2020).

To advance the translation of the role of inflammation in alcohol use, this experimental study conducted a novel endotoxin challenge in individuals with AUD and leveraged data from matched controls who also underwent an endotoxin challenge. Previously stored inflammatory marker data from the matched controls was assayed in batch with data from the AUD sample. Specific aims of the study were to (1) establish the endotoxin challenge paradigm in a clinical sample of individuals with AUD by measuring sickness symptoms throughout the inflammatory challenge, (2) evaluate group differences in (a) the peripheral cytokine inflammatory response (TNF-α, IL-6, IL-8) and (b) the behavioral mood and social disconnection responses from the endotoxin challenge between AUD and matched controls, and (3) evaluate alcohol-associated outcomes (self-reported and cue-induced craving) during the endotoxin challenge within individuals with AUD both through behavioral human laboratory paradigms and through functional neuroimaging. We hypothesized that individuals with AUD would have a higher inflammatory response to endotoxin compared to control and that this inflammatory response would be associated with increased negative mood and increased alcohol craving.

MATERIALS AND METHODS

Study design, intervention, and participants

This was a randomized, double-blind, placebo-controlled trial of low dose endotoxin. The endotoxin was derived from Escherichia coli (E. coli group O:113: BB-IND 12948 to M.R.I.) and was provided by the National Institutes of Health Clinical Center as a reference endotoxin for studies of experimental inflammation in humans. Safe administration of endotoxin has been demonstrated in previous studies (Moieni et al., 2015c; Suffredini and O’Grady, 1999). Participants were randomly assigned to receive either low-dose endotoxin (0.8 ng/kg of body weight) or placebo (same volume of 0.9% saline), which was administered via intravenous bolus infusion. The study protocol was approved by the UCLA Institutional Review Board and registered at ClinicalTrials.gov (NCT04310423). All procedures for participants in the AUD group were conducted from 2021 – 2023. Specific inclusion/exclusion criteria for the AUD group and information about randomization/blinding are presented in the Supplementary Materials.

Participants in the alcohol study were a community sample of non-treatment seeking individuals with AUD symptoms who drink alcohol heavily. Heavy drinking was defined as: (1) Alcohol Use Disorders Identification Test (AUDIT) (Saunders et al., 1993) score between 8 – 15, suggesting hazardous or harmful alcohol use, and (2) report drinking at binge levels at least 1 time in the past month (5+ drinks/day for males, 4+ drinks/day for females) (“Drinking Levels Defined | National Institute on Alcohol Abuse and Alcoholism (NIAAA),” n.d.). Additional matched control participants were blindly selected from a pool of 115 healthy participants (69 female) that completed a randomized, double-blind, placebo-controlled trial assessing the effects of endotoxin on inflammation and social cognition between 2011-2013 (NCT01671150) (Moieni et al., 2015c). Controls were blindly matched with AUD participants on sex, age, and treatment condition (i.e., placebo vs. endotoxin). Importantly, matched controls were considered ineligible for the following: (1) met DSM-IV (First et al., 2012) criteria for an Axis I psychiatric disorder; (2) tested positive for any drug use as shown by a positive urine test. The endotoxin administered was identical to the endotoxin administered in the present study in terms of bacteria (E.coli O:113) and potency (1 μg = 10,000 endotoxin units).

Study procedures and assessments

Specific screening procedures for individuals with AUD are presented in the Supplementary Materials. A CONSORT diagram depicting the sample size maintained through each screening step is presented in Figure 1.

Figure 1.

CONSORT flow diagram.

Endotoxin administration.

Endotoxin administration procedures were completed at the UCLA Clinical and Translational Research Center (CTRC). The endotoxin administration experimental visit began between 8 AM – 10 AM for all participants in both the AUD and matched control samples. Participants were randomized to an experimental condition (i.e., endotoxin vs. placebo). Approximately 90 minutes after the start of the visit, a nurse who was blind to the participant’s condition inserted a catheter with a heparin lock into the non-dominant forearm for IV drug administration and hourly blood draws. One standard meal (i.e., sandwich of participant’s choice) and snack (i.e., flavored yogurt, breakfast bar, muffin, fruit) were provided to participants, and participants were provided oral hydration throughout the visit. At the end of the experimental period, the catheter was removed from participants and individuals in the AUD group were discharged with instructions to abstain from consuming alcohol for 24 hours.

Human laboratory procedures.

Participants in the AUD group completed procedures (i.e., vital sign collection, blood collection, sickness symptom, alcohol craving (alcohol use questionnaire (AUQ); Drummond and Phillips, 2002), and mood (Profile of Mood States (POMS); Curran et al., 1995) assessments) at baseline (T0) and every hour for 4 hours (T1-T4) post-infusion, with additional procedures (i.e., social disconnection (Moieni et al., 2015c) assessment, alcohol cue-reactivity (CR) paradigm) at baseline and at the expected time of peak cytokine response (T2). Importantly, matched controls completed a neuroimaging scan three hours post-infusion (T3) and therefore assessment collection timepoints were at baseline (T0), one-, two-, and four-hours post-infusion (T1-T2, T4). Moreover, previously un-assayed plasma aliquots from blood samples collected from matched controls were re-analyzed with blood samples from the AUD group in a single batch to ensure data comparison and compatibility. Specific assessment information is detailed in Figure 2 and the Supplementary Information.

Figure 2.

Schematic outline of the study design for (A) combined overlapping procedures between AUD sample and matched controls and (B) AUD-only sample alone. In both experiments, T0 is the start of bolus infusion of either endotoxin or saline solution. In the combined samples, time points at baseline, one-hour, two-hours, and four-hours post-infusion were compared given a neuroimaging scan conducted at three-hours post infusion in the matched controls experiment. For additional information about the study assessments, see Supplementary Materials. Additional baseline assessments included breath alcohol concentration (BrAC), blood draw, urine tests (urinalysis, test for drugs of misuse, pregnancy test for females), height, weight, Timeline Follow-Back, Clinical Institute Withdrawal Assessment for Alcohol-revised (CIWA- Ar), and Columbia Suicide Severity Rating Scale (C-SSRS).

Human laboratory alcohol cue-reactivity paradigm:

Phasic craving for alcohol following water and alcohol cue exposure was assessed using the first and last items from the AUQ during an alcohol CR paradigm at baseline and at time of expected peak cytokine response (T2). The alcohol CR followed well-established procedures (Monti et al., 1987). Additional paradigm procedures are detailed in Supplementary Information.

Blood sample collection:

Blood samples were collected at baseline and four hourly time points post-infusion (T0-T4) for the AUD group. Blood samples were collected at baseline (T0), one-, two-, and four-hours post-infusion (T1-T2, T4), for matched controls. Previously un-assayed blood samples for matched controls that were stored identically to samples collected for the AUD group were analyzed with the present AUD group analyses in a single batch using a multiplex assay. Specifically, whole blood samples were collected in ethylenediaminetetraacetic acid (EDTA) tubes. They were centrifuged in a refrigerated centrifuge at 4°C to obtain plasma. Plasma was aliquoted and stored in a −80°C freezer until the completion of the study. Peripheral markers of inflammation (TNF-α, IL-6, IL-8) were assessed using a multiplex assay. Blood samples from both AUD and matched control participant groups were assayed together in a single batch for the present analyses. Plasma levels of TNF-α, IL-6, and IL-8 were determined utilizing the Meso Scale Discovery MULTI-SPOT Assay System (Rockville, MD). A custom 5-plex from the Pro-inflammatory Panel 1 Human Kit was utilized to assay samples in duplicate according to the manufacturer’s protocol. Electrochemiluminescence (ECL) signals were measured on the MESO QuickPlexSQ 120 (Rockville, MD). The average intra-assay CV% for TNF-α, IL-6, and IL-8 standards were 8.92, 2.37, and 3.81, respectively. The inter-assay CV% of an internal laboratory quality control sample was 16.5%, 3.16%, and 3.79% for TNF-α, IL-6, and IL-8, respectively. A lower limit of quantification of 0.21 pg/mL was utilized for IL-6, and this lower limit value was assigned for 10% (n=10) of AUD participant sample timepoints. Levels of TNF-α, and IL-8 were detectable in all subjects at all sample timepoints.

Brain functional magnetic resonance imaging procedures.

Approximately three hours after receiving the infusion of either endotoxin or saline, individuals in the AUD group completed a functional magnetic resonance imaging (fMRI) scan at the UCLA Center for Cognitive Neuroscience (CCN) on a 3.0T Siemens Magneton Prisma scanner using a 32-channel head coil. Specific imaging parameters are detailed in Supplementary Information.

Neuroimaging alcohol cue-reactivity paradigm:

The visual fMRI alcohol-cue reactivity paradigm is well-validated and elicits blood oxygen level dependent (BOLD) response in mesocorticolimbic reward circuitry (Schacht et al., 2013). Participants viewed visual cues and rated their craving for alcohol immediately following each cue block using an optically isolated universal serial bus (USB) interface consisting of a four-button response box. Additional paradigm procedures are detailed in Supplementary Information.

Statistical Methods

Analyses were conducted in SAS version 9.4. All outcomes were assessed for normal distribution by measuring skew and kurtosis at each time point of data collection. Physiological and behavioral outcomes that included diastolic blood pressure, sickness symptoms, and negative mood were not normally distributed and were subsequently square root transformed. Biological cytokine outcomes were skewed as expected and were subsequently log-transformed. No preliminary data are available for effect size estimation; a power analysis for group effect indicated that the combined sample size of 20 participants per group (AUD vs. control) resulted in 0.80 power to detect a large effect size (f=0.45) at alpha level of 0.05.

Human laboratory analyses:

Aims 1 and 2 were assessed using the combined AUD and matched controls samples. Repeated measures of mixed-model analyses of variance were run with group (AUD vs. control) and treatment (endotoxin vs. placebo) as between-subjects fixed factors and time (T0-T2, T4) as a repeated measures fixed factor, while individual subjects were a random effect. Outcomes included physiological symptoms, sickness symptoms, levels of peripheral cytokines (TNF-α, IL-6, IL-8), mood, and social disconnection. Aim 3 was assessed using only the AUD sample. Repeated measures of mixed-model analyses of variance were run with treatment as a between-subjects fixed factor and time (T0-T4) as a repeated measures fixed factor, while individual subjects were a random effect. Outcomes included self-reported craving, laboratory cue-induced craving (assessed at T0 and T2 only), mood, and social disconnection (assessed at T0 and T2 only). Body mass index (BMI) and baseline depression (Beck Depression Inventory-II (BDI-II) (Beck et al., 1996)) severity were used as covariates in all models given their known correlation with inflammation (Liqiang et al., 2023; Simpson et al., 2021). An autoregressive (AR(1)) covariance structure was used to model the repeated time factor for all models given the small sample size and time intervals. Pairwise comparisons between estimated marginal means of significant interactions and main effects in fitted models are reported. To probe group (AUD vs. control) differences in the effect of endotoxin on changes in cytokines over time, post-hoc analyses at individual timepoints that controlled for baseline endotoxin levels were conducted within individuals who received endotoxin. Given the exploratory and novel nature of the present study, multiple comparison corrections were not performed.

Secondary analyses were run to test associations between endotoxin-induced cytokines and behavioral outcomes across time points with BMI and BDI used as covariates in individuals who received the endotoxin infusion only. Significant interactions with time were probed at individual time points to examine the effect of cytokine levels on craving while controlling for baseline craving and cytokine levels. When conducting follow-up analyses with the combined sample, group (AUD vs. control) was treated as a fixed factor. Only individuals who were randomized to receive endotoxin were included in secondary analyses.

Neuroimaging analyses:

Preprocessing steps are detailed in the Supplementary Materials. First-level analyses of the neural alcohol cue-reactivity task were completed using FSL FEAT (Woolrich et al., 2001) to perform general linear models. An alcohol beverage > non-alcohol beverage contrast was specified in the first-level model for each subject. FSL’s FLAME 1 (Woolrich et al., 2004) was used to conduct group-level analyses (endotoxin vs. placebo). Following, in the endotoxin group only, a series of multiple regressions were run to examine how the change in peripheral cytokines from baseline to expected peak inflammatory response (T2) predicted neural alcohol-cue induced activation in significant clusters at the time of fMRI alcohol cue-reactivity task (T3). For all analyses, Z statistic images were cluster thresholded as determined by Z > 2.3 and a corrected cluster significance of p<0.05 (Jezzard et al., 2003). BMI and baseline BDI were used as covariates.

RESULTS

Study sample

The final combined sample contained N=40 participants. The sample consisted of 20 individuals with AUD symptoms (n=10 randomized to endotoxin) and 20 matched controls (n=10 randomized to endotoxin). Sample demographic characteristics are reported in Table 1.

Table 1.

Sample Demographics

Variable	Placebo (n=20)		Endotoxin (n=20)		Statistic	P-Value
	Matched Controls (n=10)	AUD (n=10)	Matched Controls (n=10)	AUD (n=10)
Demographic Characteristics
Sex (no., %)					X2=0.00	p=1.00
Male	6 (60.00)	6 (60.00)	6 (60.00)	6 (60.00)
Female	4 (40.00)	4 (40.00)	4 (40.00)	4 (40.00)
Mean Age (SD)	26.70 (7.72)	27.80 (7.58)	29.50 (7.75)	30.30 (6.53)	F=0.00	p=0.949
Race (no., %)					X2=5.75	p=0.451
White	5 (50.00)	5 (50.00)	3 (30.00)	6 (60.00)
Black/African American	2 (20.00)	1 (10.00)	1 (10.00)	0 (0.00)
American Indian/Alaskan Native	0 (0.00)	1 (10.00)	0 (0.00)	0 (0.00)
Asian/Asian American/Pacific Islander	2 (20.00)	0 (0.00)	2 (20.00)	2 (20.00)
Mixed Race	0 (0.00)	1 (10.00)	1 (10.00)	1 (10.00)
Other	1 (10.00)	2 (20.00)	3 (30.00)	1 (10.00)
Hispanic/Latinx (no., %)					X2=2.85	p=0.091
Yes	1 (10.00)	5(50.00)	3 (30.70)	4 (40.00)
No	9 (90.00)	5 (50.00)	7 (70.00)	6 (60.00)
Mean Yrs of Education (SD)	16.00 (1.00)	16.00 (2.45)	14.81 (1.60)	16.90 (2.13)	F=2.73	p=0.108
Mean BMI (SD)	23.82 (3.41)	24.99 (4.07)	24.62 (2.49)	27.25 (3.72)	F=2.99a	p=0.092
Mean BDI (SD)	1.50 (2.46)	7.40 (6.28)	1.40 (2.37)	7.80 (7.90)	F=13.34 a	p<0.001
Mean AUDIT (SD)	N/A	12.00 (2.66)	N/A	13.30 (6.18)	t=−0.61	p=0.549
Mean DPDD (SD)	N/A	4.58 (1.95)	N/A	4.12 (1.40)	t=0.60	p=0.559
Mean SCID Symptom (SD)1	N/A	4.20 (1.93)	N/A	4.10 (1.85)	t=0.33	p=0.749
RRHDS (no., %)					X2=0.00	p=1.00
Reward	N/A	9 (90.00)	N/A	9 (90.00)
Relief	N/A	1 (10.00)	N/A	1 (10.00)
Habit	N/A	0 (0.00)	N/A	0 (0.00)
Baseline Physiological
Mean Systolic (SD)	112.90 (4.79)	122.10 (13.49)	122.60 (13.29)	124.80 (11.04)	F=3.05b	p=0.089
Mean Diastolic (SD)	61.20 (7.52)	77.00 (4.67)	73.50 (9.43)	80.00 (9.29)	F=19.61 a	p<0.001
Mean Temperature (SD)	97.77 (0.64)	98.12 (0.75)	97.50 (0.78)	98.38 (0.81)	F=6.66 a	p=0.014
Mean Heart Rate (SD)	66.10 (9.76)	68.50 (12.96)	64.30 (7.92)	66.90 (7.40)	F=0.00	p=0.974
Baseline Cytokines
Mean TNF-α (SD)	1.46 (0.38)	1.29 (0.38)	1.34 (0.39)	1.61 (0.78)	F=1.05	p=0.313
Mean IL-6 (SD)	0.97 (1.27)	0.78 (0.96)	0.52 (0.23)	0.54 (0.30)	F=0.16	p=0.689
Mean IL-8 (SD)	2.90 (0.71)	2.81 (0.92)	3.39 (0.88)	3.93 (2.61)	F=0.46	p=0.504
24 Hour Before Experimental Visit Characteristics
Alcohol use (% Yes)	0 (0.00)	1 (10.00)	0 (0.00)	0 (0.00)	X2=1.03	p=0.311
Smoking (% Yes)	0 (0.00)	1 (10.00)	0 (0.00)	1 (10.00)	X2=2.11	p=0.147
Exercise activity (% Yes)	3 (30.00)	6 (60.00)	7 (70.00)	3 (30.00)	X2=0.23	p=0.634
THC Positive (no., %)	N/A	3 (30.00)	N/A	0 (0.00)	X2=1.57	p=0.210

Note. F and p statistic values represent the interaction effect between group and treatment variables unless otherwise indicated. X² and p statistics represent group differences unless otherwise indicated. BMI = Body Mass Index; BDI = Beck’s Depression Inventory; AUDIT = Alcohol Use Disorders Identification Test; DPDD = Drinks per Drinking Day, SCID = Structured Clinical Interview for DSM 5; RRHDS = Reward Relief Habit Drinking Scale; TNF-α = Tumor Necrosis Factor-alpha; IL-6 = interleukin-6; IL-8 = interleukin-8.

¹n=1 participant met for one symptom of alcohol use disorder (AUD) and therefore did not meet criteria for AUD. All other participants in the AUD sample (n=19) met criteria for mild-severe AUD

^aMain effect of group reported; AUD > matched control

^bMain effect of treatment reported; endotoxin > placebo

Validation of endotoxin challenge

We first examined if endotoxin induced increases in sickness symptom intensity (square root-transformed total intensity of sickness symptoms) across time in the AUD and matched control samples. No significant three-way interaction with Treatment x Time x Group was detected (p>0.05). However, a significant Treatment x Time interaction for intensity of sickness symptoms emerged (F_3,108=7.60, p<0.001; see Figure 3A). Follow-up tests of simple effects revealed that the effect of Treatment on intensity of sickness symptoms was significant at T2 (F_1,108=15.95, p<0.001), where individuals who received endotoxin had more intense sickness symptoms compared to individuals who received placebo at T2 regardless of Group (t₁₀₈=3.99, p<0.001). Specific sickness symptoms, along with increases in physiological symptoms, are detailed in the Supplementary Materials (see Figure S1, Figure S2, and Table S1).

Figure 3.

Validation of endotoxin challenge through sickness symptom changes in AUD sample and matched controls split by endotoxin and placebo treatments (A). Group differences in plasma level cytokine response during endotoxin challenge in AUD and matched controls split by endotoxin and placebo treatments (B-D). Time points of assessment include baseline (T0), 1-hour post baseline (T1), 2-hours post baseline (T2), and 4-hours post baseline (T4). For each mixed model, BMI and BDI were entered as covariates. Total intensity endorsed across five sickness symptoms have been square root transformed. (A) Sickness symptoms intensity. Raw plasma levels of cytokines have been log-transformed. (B) Tumor necrosis factor-a (TNF-a), (C) interleukin-6 (IL-6), (D) interleukin-8 (IL-8).

Group and endotoxin-induced differences in peripheral cytokines

We next examined if endotoxin induced increases in peripheral cytokines (log-transformed TNF-α, IL-6, IL-8) across time between the AUD and matched control samples. Significant Group x Treatment x Time interactions for log TNF-α (F_3,108=2.91, p=0.038), log IL-6 (F_3,108=3.82, p=0.012), and log IL-8 (F_3,108=6.48, p=0.001) were detected (see Figure 3B-D). Follow-up tests of simple effects revealed that there were no significant differences in any peripheral cytokines between the groups at baseline (all ps>0.05), but that differences in cytokines emerged at subsequent timepoints. Within individuals who received endotoxin, post-hoc analyses at individual time points controlling for baseline levels of cytokines revealed a non-significant trending effect of group at T2 (p=0.090) and T4 (p=0.059) where the control group who received endotoxin had higher levels of log TNF-α than the AUD group. Group differences emerged at T4 for log IL-6 and log IL-8, where AUD had lower values compared to matched controls at T4 (log IL-6: p=0.009; log IL-8: p<0.010). No additional endotoxin-induced group differences emerged at other individual timepoints (all ps>0.05).

Group and endotoxin-induced differences in mood and social disconnection

Following, we sought to examine if endotoxin induced changes in mood (square root-transformed POMS negative and POMS energetic total scores) and social disconnection between AUD and matched control samples. No significant three-way Treatment x Time x Group interactions, two-way Treatment x Time or Treatment x Group interactions, or significant treatment main effects were detected (all ps>0.05; see Figure S3). Additional mood and social disconnection analyses are detailed in the Supplementary Materials (see Figure S4).

Endotoxin-induced differences in human laboratory alcohol craving

Self-reported alcohol craving:

Within the AUD sample only, we sought to examine endotoxin-induced changes in self-reported alcohol craving (AUQ total sum score) across time. No significant two-way interactions with Treatment x Time were detected (all ps>0.05). However, a significant main effect of Time on self-reported alcohol craving was observed (F_4,72=2.94, p=0.026), such that there was higher self-reported craving at T0 compared to T1 (t₇₂=3.01, p=0.004) and T3 (t₇₂=2.54, p=0.013; see Figure 4A). A non-significant trending effect of Treatment on self-reported alcohol craving was additionally observed (F_1,72=3.05, p=0.085). We additionally added sickness severity as a covariate to the model to see if variability in craving was due to sickness symptoms and found that sickness symptom intensity was a significant covariate (F_1,71=4.71, p=0.034) but did not alter the overall model.

Figure 4.

Human laboratory self-reported (A) and human laboratory cue-induced (B) craving during endotoxin challenge in alcohol use disorder (AUD) split by endotoxin and placebo treatments. Full alcohol urge questionnaire (AUQ) was used to measure self-reported craving and shortened AUQ was used to measure laboratory cue-induced craving. Time points of human laboratory self-reported craving assessment include baseline (T0), 1-hour post baseline (T1), 2-hours post baseline (T2), 3-hours post-baseline (T3), and 4-hours post baseline (T4). Time points of human laboratory cue-induced craving assessment include baseline (T0) and 2-hours post baseline (T2). Follow-up analyses for human laboratory self-reported and cue-induced craving assessed associations between cytokines and craving. Significant interactions in associations were probed at mean level of cytokines, one standard deviation below mean level of cytokines, and one standard deviation above mean level of cytokines. For each model, BMI and BDI were entered as covariates. (A) Human laboratory self-reported craving. (A1) Post-hoc analyses for main effct of tumor necrosis factor-α (TNF-α) on self-reported craving. Sub-panels depict the association between TNF-α and craving at 1) one standard deviation below the mean TNF-α with craving across time (top left panel), 2) the mean level of TNF-α with craving across time (top right panel), and 3) one standard deviation above the mean TNF-α with craving across time (bottom left panel) (A2) Post-hoc analyses for significant interaction between log interleukin-6 (IL-6) and time on self-reported craving. Sub-panels depict association between IL-6 and craving at T2 and T3 where higher levels of IL-6 (i.e., 1 SD above mean IL-6 at each timepoint) were shown to be negatively associated with alcohol craving, with this association reaching significance at T2. IL-6 levels were not significantly associated at other timepoints and are therefore not displayed. (B) Human laboratory cue-induced craving.

Following, we conducted secondary analyses testing the associations between changing levels of cytokines (log-transformed TNF-α, IL-6, IL-8) and self-reported craving (AUQ total sum score) across time. Within the individuals that received endotoxin, a significant log IL-6 x Timepoint (F_4,31=8.67, p=<0.001) interaction on self-reported alcohol craving emerged (see Figure 4A2). We probed the interaction by examining the association of log IL-6 with craving at each individual timepoint while controlling for baseline IL-6 and baseline self-reported craving to assess the change since baseline. We found that log IL-6 was negatively associated with self-reported craving at T2 (p=0.006), the time of peak inflammatory response, and a non-significant trending negative association between log IL-6 and self-reported craving emerged at T3 (p=0.063). We probed the significant T2 association at mean, and one standard deviation above/below mean, log IL-6 values for T2. Specifically, at T2, we found that log IL-6 had the largest negative association with self-reported craving at one standard deviation above T2 mean log IL-6 (β=−3.66) and the smallest negative association with self-reported craving at one standard deviation below T2 mean log IL-6 (β=−1.54). Therefore, higher levels of endotoxin-induced IL-6 at T2, the time of peak inflammatory response, were associated with less craving for alcohol. No additional significant associations between log IL-6 and self-reported craving emerged for remaining timepoints (all ps>0.05). Of note, these secondary analyses highlight the associations between individual endotoxin-induced increases in log-IL6 with self-reported craving at each timepoint of the experiment and reveal negative associations between log-IL6 and self-reported craving at T2, the time of peak inflammatory response. Notably, there was no significant overall change in self-reported craving from baseline across either treatment. However, these secondary analyses suggest that individual differences in cytokine response rather than the endotoxin treatment itself were predictive of changes in self-reported craving.

No significant interactions with time emerged for log TNF-α or IL-8 (all ps>0.05). However, a significant main effect of log TNF-α on self-reported alcohol craving was observed (F_1,31=7.31, p=0.010; see Figure 4A1). We probed the significant main effect by conducting a marginal means comparison where we examined the association of log TNF-α with craving at specific log TNF-α values. Log TNF-α was significantly positively associated with craving at one standard deviation below mean log TNF-α (p=0.010), but significantly negatively associated with craving at mean log TNF-α and at one standard deviation above mean log TNF-α (ps=0.010). Therefore, lower levels of TNF-α were associated with higher craving but higher levels of TNF-α were associated with less craving.

Cue-induced alcohol craving:

Within the AUD sample only, we examined endotoxin-induced changes in cue-induced alcohol craving (abbreviated AUQ sum score) across time. There was a significant Treatment x Time interaction for alcohol cue-induced alcohol craving amongst individuals with an AUD (F_1,17=5.21, p=0.036; see Figure 4B). Follow-up tests of simple effects revealed that individuals who received endotoxin had significantly decreased alcohol cue-induced craving at T2 compared to T0 (t₁₇=3.16, p=0.006), while individuals who received placebo did not experience a significant change in alcohol cue-induced craving between T2 and T0 (t₁₇=−0.00, p=1.000). We additionally added sickness severity as a covariate to the model to see if variability in craving was explained by sickness symptoms and found that sickness symptom intensity was not a significant covariate (p>0.05).

We conducted secondary analyses within individuals who received endotoxin to test the associations between changing levels of cytokines (log-transformed TNF-α, IL-6, IL-8) and cue-induced craving (abbreviated AUQ sum score) across time. Within the individuals that received endotoxin, a significant log IL-6 x Timepoint (F_1,6=6.73, p=0.041) interaction on alcohol cue-induced craving emerged. We probed the interaction by examining the association of log IL-6 with craving at each individual timepoint. We found that log IL-6 was associated with cue-induced craving at T0 (p=0.041), before endotoxin infusion occurred. Given that this timepoint was before the endotoxin infusion occurred, we additionally looked at the T0 timepoint with individuals who received both endotoxin and placebo. This association between log IL-6 and cue-induced craving was not significant when using a larger sample (N=20) and we therefore did not probe it further. Moreover, within the individuals who received endotoxin, log IL-6 was not significantly associated with cue-induced craving at T2 (p=0.886) while controlling for baseline IL-6 and baseline cue-induced craving to assess the change since baseline. Log TNF-α and IL-8 were not significantly associated with alcohol cue-induced craving (all ps>0.05). Therefore, while primary analyses revealed a significant decrease in cue-induced alcohol craving across time within individuals who received endotoxin, this decrease was not significantly associated with individual endotoxin-induced changes in cytokines.

Endotoxin-induced differences in brain functional magnetic resonance imaging

Within the AUD sample only, we conducted a whole-brain analysis to examine endotoxin-induced differences in brain activation during a neural alcohol cue-reactivity task. There was a main effect of treatment on alcohol cue-elicited brain activation, such that individuals who received placebo had greater activation in clusters located within the putamen, caudate, thalamus, insula, precuneus, and posterior cingulate, as compared to those who received endotoxin (see Supplementary Materials (Table S2) for full list of regions). In those who received endotoxin, increased log TNF-α, log IL-6, and log IL-8 levels at T2 significantly predicted decreased alcohol cue-elicited activation in clusters located within the bilateral precuneus, posterior cingulate, precentral gyrus, central opercular cortex, and postcentral gyrus (all ps<0.01; see Figure 5). Increased log IL-6 at T2 additionally significantly predicted decreased alcohol cue-elicited activation in a cluster located within the right precentral gyrus and superior frontal cortex (p=0.03; see Figure 5A2). Increased log IL-8 at T2 additionally significantly predicted decreased alcohol cue-elicited activation in a cluster located within the right nucleus accumbens, caudate, putamen, and left nucleus accumbens and left caudate (p=0.03; see Figure 5A3).

Figure 5.

Brain scan neural cue-induced (A) craving during endotoxin challenge in alcohol use disorder (AUD) split by endotoxin and placebo treatments. Brain scan occurred at 3-hours post-baseline (T3). Associations between change in plasma-level cytokines from baseline (T0) to 2-hours post baseline (T2) and neural activation during fMRI alcohol cue reactivity task at T3 were investigated (A). For each model, BMI and BDI were entered as covariates. (A) Individuals who received placebo had significantly increased neural activation in response to alcohol cues compared to those that received endotoxin. (A1) Association between T2-T0 log tumor necrosis factor-α (TNF-α) levels and brain clusters with significant decreased activation during fMRI cue reactivity task, (A2) association between T2-T0 log interleukin-6 (IL-6) levels and brain clusters with significant decreased activation during fMRI cue reactivity task, (A3) association between T2-T0 log IL-8 levels and brain clusters with significant decreased activation during fMRI cue reactivity task.

DISCUSSION

This is the first study to successfully conduct an endotoxin challenge in individuals with AUD. Results indicate that the endotoxin challenge was safely tolerated and that individuals with AUD experienced a comparable physical response to the endotoxin challenge to matched controls, as both groups had increased sickness symptoms. However, those in the AUD group had a blunted inflammatory response from endotoxin compared to controls. Importantly, baseline inflammatory cytokine levels did not differ significantly between AUD and controls. Individuals in the AUD group experienced lower peak increases in TNF-α and a faster recovery in IL-6 and IL-8 compared to controls. However, endotoxin did not impact mood or social disconnection within either group. Within those in the AUD group, endotoxin led to decreased self-reported alcohol craving and alcohol cue-induced craving, and these behavioral findings were corroborated by lower endotoxin-induced neural activation in response to alcohol cues. Indeed, the magnitude of inflammation induced by endotoxin corresponded to a lower level of craving, subjectively and neurally. We validated the utility of an endotoxin challenge at invoking a safe inflammatory response in AUD while elucidating the effects of inflammation on alcohol craving.

A low dose of endotoxin led to increases in sickness symptoms and inflammatory cytokines, indicating that we successfully elicited an inflammatory response in those with AUD. However, our hypothesis that those with AUD would have an increased inflammatory response compared to controls was not supported. In fact, the effects of endotoxin were blunted in the AUD group compared to control. The neuroimmune hypothesis of addiction predicts a bidirectional relationship between alcohol and inflammation (Cui et al., 2014; Mayfield and Harris, 2017), where cytokine activation results in alcohol consumption (Blednov et al., 2011), but alcohol use additionally increases cytokines (Hillmer et al., 2020). Therefore, we speculate that individuals with AUD may experience consistent ramping up of their immune system from frequent drinking episodes, leading to the appearance of a habituated inflammatory response following the endotoxin challenge compared to controls. Indeed, in individuals with chronic AUD, a decrease in lymphocytes and an increase in infections have been observed, suggested impaired immune function (Barr et al., 2016). Preclinical research reveals that binge alcohol exposure to human monocytes reduces endotoxin-produced inflammatory cytokines, suggesting tolerance to endotoxin following recurrent binge alcohol exposure (Bala et al., 2012; Muralidharan et al., 2014). This dose-dependent decrease in immune function (Barr et al., 2016) could be attributed to an attenuation of immune responsiveness following chronic alcohol use.

Behaviorally, endotoxin-induced inflammatory activation led to decreased self-reported and cue-induced craving. These decreases in craving may be due to the sickness symptoms associated with inflammatory activation. Sickness symptoms and behavior following endotoxin administration include decreased food and water intake, lethargy, (Dantzer, 2001), headaches, shivering, and nausea (Moieni et al., 2015b). Preclinical studies of endotoxin and alcohol use have utilized delay periods to accommodate for the likely effects of sickness behaviors decreasing alcohol consumption (Blednov et al., 2011; Decker Ramirez et al., 2023). Studies using these delay periods have found that endotoxin administration in alcohol-naïve mice leads to prolonged increases in alcohol consumption one week to 30 days after endotoxin administration (Blednov et al., 2011). In alcohol-dependent mice, endotoxin administration leads to increases in alcohol consumption three days following endotoxin injection (Decker Ramirez et al., 2023). Therefore, preclinical evidence suggests a rebounding effect of inflammatory activation increasing drinking consumption following a speculated initial immediate avoidance of alcohol due to sickness behaviors. The present study only captured the window of time of peak inflammatory and sickness symptom response, which may explain the observed decrease in craving. To further translate this preclinical hypothesis of a rebounding inflammatory effect on drinking to human clinical trials, clinical investigations into the longer-term effects of endotoxin on alcohol craving are warranted.

Previous evidence suggests that plasma levels of TNF-α and IL-6 are positively correlated with alcohol craving in individuals with AUD at the onset of acute withdrawal (Heberlein et al., 2014; Leclercq et al., 2014). However, mean plasma concentrations of TNF-α and IL-6 correlated with alcohol craving in this population were approximately 6-8pg/mL and 10-23pg/mL, respectively (Heberlein et al., 2014; Leclercq et al., 2014). Inflammatory activation from our endotoxin challenge raised TNF-α, IL-6, and IL-8 concentrations in AUD well above those levels as peak mean concentrations were 43.45pg/mL, 52.94pg/mL, and 93.98pg/mL, respectively. Therefore, the association between inflammatory cytokine concentrations and alcohol craving may be non-linear, as low-level phasic increases in cytokines may drive a desire to consume alcohol, but elevated increases in cytokines may decrease the desire to consume alcohol. Our secondary analyses support this conclusion, as TNF-α and IL-6 concentrations one standard deviation above their mean values were the most negatively associated with self-reported alcohol craving compared to lower levels of TNF-α and IL-6 concentrations. Aligning with the allostatic model of addiction that characterizes the motivation to use substances through positive and negative reinforcement (Koob and Schulkin, 2019), chronic alcohol use is thought to be driven by the desire for relief from negative emotional states associated with substance use (i.e., negative reinforcement). This relationship, in turn, may be non-linear, as low-level negative states associated with inflammatory cytokines may drive continued alcohol use for relief (Leclercq et al., 2014), but higher levels of inflammatory cytokines associated with sickness behavior may decrease the desire to drink. The recovery period following inflammation and sickness behavior should be examined to understand if the desire to drink resurges, indicating a possible incubation of craving. Endotoxin may have caused a disruption in homeostasis, which, as reviewed in Venniro et al., can be associated with acute stressors leading to the incubation of craving (Venniro et al., 2021). In individuals with more severe AUD who experience acute withdrawal sickness symptoms from alcohol, negative reinforcement may drive them to seek out alcohol despite the occurrence of sickness withdrawal symptoms (i.e., relief drinking). However, our AUD sample consisted of individuals who had an average of 4 AUD symptoms, no clinically significant levels of withdrawal, and mostly reward drinking motives rather than relief (see Table 1). Thus, this inflammatory challenge should be repeated in a sample with more severe AUD where endotoxin-induced sickness behavior might mimic withdrawal from alcohol, thereby inducing alcohol craving.

The observed decreases in craving for alcohol in the laboratory following inflammatory activation were corroborated by neural brain activations during an fMRI alcohol cue reactivity task. Endotoxin-induced increases in log TNF-α, IL-6, and IL-8 at the time of expected peak inflammatory response were correlated with decreased alcohol cue-elicited activation in clusters associated with craving and reward. Specifically, all three cytokines were negatively correlated with alcohol cue-elicited activation in regions that included the bilateral precuneus, posterior cingulate, precentral gyrus, and central opercular cortex. Neural alcohol cue-reactivity paradigms reliably result in activation of these brain regions (Courtney et al., 2016; Schacht et al., 2013). Log IL-8 alone was negatively correlated with activation in the right and left nucleus accumbens, caudate, and putamen. Activation in these regions is specifically associated with increased compulsive alcohol-seeking behaviors (Grodin et al., 2018). Therefore, the endotoxin-induced inflammatory activation in individuals with an AUD may have led to an altered response to alcohol cues and dampened the compulsive response to seek out alcohol that is a hallmark of AUD (Koob and Volkow, 2010). These results may be interpreted through the conclusion that higher levels of inflammatory cytokines associated with sickness behaviors decrease the desire to drink. The decrease in self-reported and cue-induced craving is corroborated by this functional brain response where brain regions typically positively associated with alcohol cues and alcohol seeking have less activation in response to the acute inflammatory challenge.

Study strengths the multi-modal approach utilized which provides robust evidence of the effects of the inflammatory challenge on biology and behavior. The double-blind, placebo-controlled, and stratified randomization design were study strengths to enhance scientific rigor. Additionally, the stringent eligibility criteria for the endotoxin challenge accounted for the possibility of recent immune events that could have confounded the results. Specifically, who reported vaccinations or illnesses in the two weeks preceding the endotoxin challenge were not eligible to complete the challenge. Moreover, participants had to have a breath alcohol concentration of 0.000g/dl at the endotoxin challenge visit, and only one individual in the AUD group who received placebo reported past 24-hour alcohol and cigarette use. As such, we are confident that recent substance use did not confound inflammatory results. Limitations of the study include that matched control data were leveraged from pre-existing data collected over one decade prior to AUD data. Thus, this time lapse may have impacted results. However, we ensured data comparability by assaying previously unused, stored plasma aliquots from matched controls with samples from the AUD group in a single batch. Given that the merging of datasets was retroactive, few identical questionnaires overlapped between the two datasets, limiting the ability to assess negative and positive mood and social disconnection fully across both samples. Indeed, these measures were likely underpowered, leading to our observed null mood and social disconnection results despite previous evidence suggesting that inflammation induces negative affect and social disconnection (Moieni et al., 2015c, 2015a). We investigated negative mood in the AUD sample alone and utilized the full POMS-SF scale administered to the AUD group. However, given that the short form was utilized, only four items assessed negative mood, again potentially contributing to null observed results. Future studies should also incorporate a longer assessment of negative mood and social disconnection to fully explore the effect of acute inflammation on these measures in AUD. Moreover, we were unable to assess alcohol craving in the control sample, although it is likely that floor effects of craving in the control sample would have prevented us from identifying Group x Treatment interactions. The control data additionally did not include granular assessments of drinking levels, though participants were excluded if they endorsed AUD during a substance history review or if they met diagnostic criteria for any psychiatric disorders. Additionally, given the preliminary and exploratory nature of this project, the total combined sample size was N=40, with 20 individuals in the AUD group, and 10 individuals receiving endotoxin or placebo within each group. Therefore, the interpretability of analyses are limited due to the small sample sizes, but the preliminary findings are promising for continuing the use of endotoxin administration to study the role of inflammation in AUD. Moreover, a parallel group design was utilized in place of a crossover design. Though a crossover design would have removed inter-individual variability, we chose to employ a parallel group design to ensure the blind could be upheld given the experience of sickness symptoms following endotoxin administration. Lastly, we interpreted the IL-6, IL-8, and TNF-α cytokines as pro-inflammatory responses to endotoxin, though the inflammatory cascade is complex, and cytokines regulate both pro- and anti-inflammatory responses. Though these cytokines are conceptualized as pro-inflammatory cytokines, it is important for future work to include more granular time measurements to observe the acute fluctuations in cytokines during the endotoxin response and to measure the inflammatory mediators activated from the cytokine response to better understand the pro- and anti-inflammatory patterns.

In conclusion, these preliminary findings reveal that the acute endotoxin challenge successfully induced a safe inflammatory response in AUD that resulted in similar experiences of sickness symptoms compared with matched controls. In comparison to controls, individuals with AUD had a blunted pro-inflammatory response and faster recovery from the inflammatory challenge. Individuals with AUD may experience consistent activation of their immune system from frequent heavy drinking episodes, leading to the appearance of a habituated inflammatory response following the endotoxin challenge compared to matched controls. Moreover, within individuals with AUD, endotoxin led to decreased alcohol cue-induced craving in line with the experience of sickness symptoms, with higher levels of cytokines having more negative associations with self-reported craving. This preliminary validation of the endotoxin challenge to safely study the role of inflammation in individuals with AUD is important for the continued clinical translation of preclinical work investigating inflammation as a mechanism driving alcohol use. Future endotoxin challenge experiments should consider the rebounding effect of inflammation on drinking behaviors and subjective craving following the experience of sickness behaviors. Clarifying the mechanistic role of inflammation in alcohol use can inform the development of novel immune modulatory treatment targets for AUD. Together, our preliminary findings suggest that the bidirectional relationship between inflammation and alcohol-associated behaviors is multi-faceted and sensitive to inflammatory intensity and alcohol use severity.