Introduction

The gut microbiota integrates a large variety of microbes, including bacteria, fungi, archaea and viruses that colonise the digestive tract. These microorganisms offer numerous benefits by interacting with the host and establishing a symbiotic relationship.1 This symbiosis is not only essential for peripheral physiological functions, but also affects neurobiological processes. Multiple studies have demonstrated the existence of a crosstalk between several neurobiological processes and the gut microbiota, including those related to mental disorders, such as depression,2 anxiety, autism spectrum disorders and addiction.2–4 Interestingly, the relationships between addiction and dysbiosis in gut microbiota are gaining high relevance, mainly in alcohol abuse.5 Ethanol consumption reduces protective bacteria, increases intestinal permeability and releases inflammatory factors, which finally contribute to the psychopathology of alcoholism.6 Other substance use disorders, such as opioid, cocaine or methamphetamine disorders, have also been related to gut microbiota dysbiosis.7

Food addiction is a controversial concept still under debate,8 consisting of a complex multifactorial behavioural disorder characterised by a loss of control over food intake that has increased in prevalence in recent years.9 It is characterised by the compulsive intake of palatable foods, which can produce adaptive changes in the reward brain network. This behavioural alteration is related to obesity and other eating disorders and lacks effective treatment, leading to high socioeconomical costs worldwide. Despite the early definition of this concept,10 the fifth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) does not include the concept of food addiction.11 However, a widely accepted instrument currently used in the clinic to evaluate food addiction is the Yale Food Addiction Scale (YFAS), updated in 2016 to YFAS 2.0 to apply the DSM-5 criteria for substance use disorder to food addiction.12 The YFAS 2.0 food addiction criteria can be summarised in three hallmarks also used in rodent models to mimic this disorder: persistent food seeking, high motivation to obtain food, and compulsivity-like behaviour.13 14 In previous studies, we have validated the animal model of food addiction in mice,15 and these behavioural findings were replicated in other studies.13 14 We found that the specific prelimbic to the nucleus accumbens brain circuit was involved in the vulnerability to develop food addiction, and we have also described specific epigenetic mechanisms involved in this multifactorial disease.13 14

The study of the gut microbiota signatures related to food addiction has gained attention in recent years. However, most studies have been performed in rodents,16–18 and few human studies have been reported.19 In spite of these findings, there is a lack of translational studies that validate the functional relevance of human findings in animal models, which would be required to design more successful treatments.8 Environmental factors and dietary patterns have a major influence on gut microbiota composition, and the overconsumption of highly palatable food may promote a gut microbiota dysbiosis that has been recently proposed to participate in the loss of eating control.8 19 In agreement, individuals with obesity, which may be promoted by food addiction,20 showed altered gut microbiota with a reduced diversity that facilitated energy absorption capacity and may affect host brain function.21 However, the functional relevance of gut microbiota in the loss of eating control and food addiction has not yet been demonstrated.

In this study, we have obtained extreme subpopulations of food addicted and non-addicted mice to identify the differential gut microbiota signatures associated with vulnerability to addiction. Using parallel food addiction-like criteria, we have applied the YFAS 2.0 score to classify a cohort of patients to assess the possible gut microbiota signatures of this behavioural disorder as potential biomarkers. We functionally validated the role of Blautia, the genus most differentially expressed in addicted mice and humans, by administering the non-digestible carbohydrates,22 lactulose and rhamnose, that increased Blautia abundance and prevented the development of food addiction in mice. A similar result was observed after oral administration of Blautia wexlerae as a beneficial microbe. The strategy of a beneficial microbe and/or dietary supplements to modulate gut microbiota is promising and can be either extracted from non-digestible carbohydrate materials or synthetically produced.23

ResultsCharacterisation of extreme subpopulations of addicted and non-addicted mice

We used the genetically homogeneous inbreed C57Bl/6J strain of mice that underwent an operant protocol of food addiction during six sessions of fixed ratio (FR) 1 schedule of reinforcement, followed by 92 daily sessions of FR5 (figure 1A). The selected subset of mice belongs to a previous publication with a large cohort of male JAX C57BL/6 J mice to evaluate miRNA signatures associated with vulnerability to food addiction.14 In the late period of the food addiction protocol (figure 1A), addicted mice performed higher persistence of response, motivation and compulsivity than non-addicted mice, as expected (figure 1B–D). In contrast, intake of pellets and body weight were similar in addicted and non-addicted mice (figure 1E, F), possibly due to the limited effort (FR5) required to obtain each reward. Online supplemental material and online supplemental table S1 provide a detailed description of the results of figure 1.

Figure 1

Characterisation of extreme subpopulations of addicted and non-addicted mice. (A) Timeline of the procedure of operant behaviour mouse model. Mice were trained during the first 6 days in operant behaviour sessions of 1 hour at a fixed ratio (FR) 1 schedule of reinforcement, followed by 92 daily sessions of FR5. The addiction-like criteria (persistence of response, motivation and compulsivity) were evaluated in the late period to categorise mice as addicted and non-addicted. (B–D). Behavioural tests for the three addiction-like criteria in the late period (individual values with IQR) in the addicted and non-addicted groups. (B) Persistence of response (t test, ***p<0.001). (C) Motivation (Mann–Whitney U test, ***p<0.001). (D) Compulsivity (Mann–Whitney U test, ***p<0.001). (E) Pellet intake and (F) body weight for those mice classified as addicted (A) and non-addicted (NA) (n=11 mice as A and n=13 as NA mice, trained with chocolate pellets). Statistical details are included in online supplemental table S1. The selected subset of mice belongs to a previous publication with a large cohort of male JAX C57BL/6 J mice to evaluate miRNA signatures associated with vulnerability to food addiction.14

Gut microbiota profile of vulnerability to addiction in mice

We carried out the 16S rRNA gene amplicon sequencing of caecum contents to study gut microbiota signatures associated with food addiction-like behaviour using the cohorts of extreme phenotypes described above. As expected, the murine caecal microbiota was dominated by the phyla Firmicutes and Bacteroidetes, reaching almost 90% of the relative abundance, similar in both addicted and non-addicted mice (figure 2A). See online supplemental material for a detailed description of the results of figure 2. Alpha diversity indexes were similar in both groups (figure 2B, C), and there were no differences in beta diversity (online supplemental figure S1). Importantly, analyses of the data at the different taxonomic levels showed significant differences in the phyla, families and genera’s relative abundances of several bacteria. Actinobacteria phylum (figure 2D), Coriobacteriaceae and Erysipelotrichaceae families (figure 2E), and Lachnospiraceae UCG-001 and Enterohabdus genera (figure 2F) had decreased relative abundances in addicted compared with non-addicted mice. Other genera, such as Allobaculum and Blautia (figure 2F) from Bacillota/Firmicutes phylum, showed a similar decrease in addicted mice, altogether supporting the potential beneficial profile of non-addicted gut microbiota signatures.

Figure 2

Gut microbiota profile of vulnerability to addiction in mice. (A) Pie chart at the phylum level shows a high proportion of the phyla Firmicutes and Bacteroidetes, reaching almost 90% of the relative abundance in both groups of addicted and non-addicted mice. (B–C) Results of Chao1 and Shannon alpha diversity indexes of addicted (A) and non-addicted (NA) mice. (D–F) Volcano plots representing the differential bacterial abundance between addicted (A) and non-addicted (NA) mice after a long operant training protocol using the DESeq2 test. Differences were observed in the relative abundances at the (D) phylum, (E) family and (F) genus levels. The volcano plot indicates −log 10 (p value) for bacteria (Y axis) plotted against their respective log 2 (fold change) (X axis). The coloured dots represent significantly downregulated and upregulated bacteria between the addicted and non-addicted groups, respectively (ie, Blautia is downregulated in addicted mice). Significantly different taxa (p<0.05) are coloured according to the phylum. (G) Discriminant analysis effect size method (LEfSe) comparing addicted and non-addicted groups. Linear discriminant analysis (LDA) was performed. Data are expressed as mean±SEM. DESeq2 test wa performed (n=11 non-addicted (NA) mice, n=13 addicted (A) mice).

Gut microbiota correlates with food addiction features in mice

The possible correlations between gut microbiota signatures, addiction-like criteria, and the phenotypic traits of the different mice groups were further investigated (figures 3 and 4, online supplemental figure S2, online supplemental figure S3). At the genus level (figures 3 and 4), the Ruminococcaceae_NK4A214_group and the Gastranaerophilales_uncultured organism genera positively correlated with motivation in the addicted group, whereas the Clostridiales_vadin BB60 group_uncultured, Erysipelatoclostridium and Parabacteroides genera negatively correlated with motivation in these addicted mice. Finally, the genera Ruminococcaceae_UCG009, Lachnospiraceae_FCS020_group, Peptococcus, Candidatus_Arthromitus, Ruminiclostridium_6, Roseburia, Coprococcus_1 and Acetatifactor positively correlated with persistence of response in the addicted group. Online supplemental material has a detailed description of the results of microbiota and food addiction signature correlations at the family level (online supplemental figure S2 and S3).

Figure 3

Caecal microbiota of non-addicted mice. Corrplot showing significant (p<0.05) Spearman correlations coefficients between microbiota relative abundances at genus level and addiction criteria and phenotypic traits in non-addicted mice after a long operant training protocol (n=11 for non-addicted mice).

Figure 4

Caecal microbiota of addicted mice. Corrplot showing significant (p<0.05) Spearman correlations coefficients between microbiota relative abundances at genus level and addiction criteria and phenotypic traits in addicted mice after a long operant training protocol (n=13 for addicted mice).

Use of YFAS 2.0 to characterise addicted and non-addicted humans

The possible gut microbiota signatures associated with food addiction were investigated in a cohort of patients (n=88) classified following the YFAS 2.0 questionnaire. The three addiction-like criteria measured in our food addiction mouse model recapitulates properly the principal features of the food addiction human disease evaluated by the 35 item self-report YFAS 2.0, as reported previously.14 As expected, the sum of the YFAS 2.0 questions, under the criteria of persistence of response, motivation and compulsivity, was much higher in participants diagnosed with food addiction than in non-addicted subjects (figure 5A–C). We also performed a principal component analysis (PCA) with the main variables (persistence of response, motivation, compulsivity, tolerance, withdrawal, craving and distress). The two principal components (PC) accounted for 75.8% and 7.1% of variations, respectively (figure 5D–G), and two clusters of addicted and non-addicted groups were identified (figure 5D). Interestingly, PC1 has strong loadings (>0.7) from distress, withdrawal, tolerance and craving (figure 5E, F) but not from the persistence of response, motivation or compulsivity that had high loads in PC2 (figure 5E, G). Online supplemental material has a detailed description of the results of figure 5D–G.

Figure 5

Gut microbiota signatures in humans. (A–C) Results of the three addiction-like criteria. (A) Persistence of response (Mann–Whitney U test, ***p<0.001), (B) motivation (Mann–Whitney U test, ***p<0.001), and (C) compulsivity (Mann–Whitney U test, ***p<0.001) (median and IQR), comparing non-addicted (NA) and addicted (A) participants. n=88 for human participants (n=79 NA individuals, n=9 A individuals). (D–G) Principal component analysis (PCA) of the three addiction criteria and the four phenotypic traits in humans. (D) Human subjects clustered by addicted or non-addicted classification on the space yielded by two components of the PCA that account for the maximum data variance. (E) Criteria belonging to each component, principal component (PC) 1 (75.8%) and PC2 (7.1%). (F, G) The order of factor loading of the different variables in PC1 and PC2 is represented. The dashed horizontal line marked loadings >0.7, mainly contributing to the component. (H–J) Correlational analyses between addiction criteria of persistence of response, motivation and compulsivity in addicted and non-addicted mice compared with addicted and non-addicted individuals belonging to the human cohort were analysed together. (H) At the correlation between compulsivity and persistence of response, a strong negative correlation for addicted individuals (both human and mice) and a moderate positive correlation for non-addicted individuals were described. (I) At the correlation between compulsivity and motivation, a mild negative correlation for addicted individuals (mice) was found, while positive correlations for non-addicted individuals (mice and humans) and addicted humans were described. (J) At the correlation between motivation and persistence of response, a strong negative correlation for addicted individuals (humans) and mild for mice was found together with a moderate positive correlation for non-addicted individuals (mice and humans) with a similar slope.

Cross-characterisation of mouse and human behaviours

Correlational analyses between addiction criteria in mice and humans were also plotted together (figure 5H–J). The links between correlations found in mice and humans further underlined the translational value of our behavioural results. In non-addicted individuals, a positive correlation was found in both humans and mice between the following addiction criteria: compulsivity and persistence of response (figure 5H), compulsivity and motivation (figure 5I), and motivation and persistence of response (figure 5J). We also performed a correlation matrix to explore the nature of the association between each addiction criterion and each phenotypic trait with the cohorts of mice and humans, respectively (online supplemental figure S4). Online supplemental material has a detailed description of the results of figure 5 and online supplemental figure S4.

Gut microbiota signatures in humans

Differential bacterial abundance using ANCOM-BC, controlling for age, body mass index and sex were observed in our human cohort (n=88) in the volcano plots (figure 6A–C). Microbiota signatures, distinct between addicted and non-addicted individuals, are represented by the differential expression of bacteria species taxa coloured according to the phylum (figure 6A). Multiple similitudes were found to be overlapped between mice and humans. Interestingly, Blautia wexlerae species were increased in individuals with a low score in the YFAS scale in obese humans (figure 6B), in agreement with the decreased Blautia genus abundance in addicted mice (figure 2F). Additionally, there were no differences in beta diversity (online supplemental figure S5). We explored the influence of diet composition on the gut microbiota data and did not observe any effect (online supplemental table S2). Online supplemental material has a detailed description of the results of figure 6A–G.

Figure 6

Volcano and scatter plots of bacterial abundance. (A–C) Volcano plots representing the differential bacterial abundance (pFDR<0.05) using ANCOM-BC, controlling for age, body mass index and sex in humans for (A) all of the population (n=88), and (B) obese (n=36) and (C) non-obese (n=52) individuals. Fold change (FC) associated with a unit change in the YFAS score and log10 Benjamini–Hochberg p values adjusted (pFDR) are plotted for each taxon. Significantly different taxa are coloured according to the phylum. NS, non-significant. (D–G) Scatter plots of the partial Pearson correlation between the centred log ratio (clr) levels of different species of the genus Blautia and the Yale Food Addiction Scale (YFAS) scores in (D) the whole cohort (n=88) controlling for age, body mass index and sex, and in (E–G) patients with obesity (n=36), controlling for age and sex. The residuals are plotted.

Functional validation with the non-digestible carbohydrates lactulose and rhamnose in the mouse food addiction protocol

Both mice and human results suggest that some specific microbiota could be protective in preventing food addiction. The strong similitudes found in the Blautia genus content in both species underlines the potential beneficial effects of this particular gut microbiota. However, Blautia is a strictly anaerobic bacteria, and its possible therapeutic use as a a beneficial microbe to prevent food addiction would be difficult. Interestingly, several well known prebiotics that could be used in humans have been reported to enhance the abundance of the Blautia genus. Therefore, we have investigated the possible protective effects promoted by oral administration of lactulose and rhamnose as non-digestible carbohydrates able to enhance Blautia genus abundance in the gut.24 25

For this purpose, a total of 41 C57BL/6 J mice underwent an operant protocol of 120 sessions (6-FR1 and 114-FR5 sessions (online supplemental figures S6 and S7 and online supplemental table S3). Online supplemental material has a detailed description of the behavioural results of figure 7 and the PCA and correlation matrix of figure 8. Importantly, motivation for chocolate flavoured pellets was significantly reduced in the rhamnose group compared with lactulose (Mann–Whitney U test=31.50, p<0.05, figure 7C). Significantly, mice receiving rhamnose showed decreased compulsivity in seeking palatable food compared with control mice, revealing a beneficial effect of this non-digestible carbohydrate to prevent food addiction (Mann–Whitney’s U=49.50, p<0.05, figure 7D). Notably, 29.41% of control mice achieved 2–3 criteria and were considered addicted, whereas none of the mice receiving lactulose or rhamnose achieved the addiction criteria (χ2=5, control vs lactulose p<0.05 and χ2=5, control vs rhamnose p<0.05, figure 7E).

Figure 7

Characterisation of extreme subpopulations of addicted and non-addicted mice in the experiment with lactulose and rhamnose. (A) Timeline of the procedure of operant behaviour mouse model. Mice were trained during the first 6 days in operant behaviour sessions of 1 hour at a fixed ratio (FR) 1 schedule of reinforcement, followed by 114 daily sessions of FR5. The addiction-like criteria (persistence of response, motivation and compulsivity) were evaluated in the late period (95-114) to categorise mice into addicted and non-addicted. Mice received the non-digestible carbohydrate lactulose, rhamnose or control in drinking water during the whole experimental sequence. Number of reinforcers during 1 hour of operant training sessions maintained by chocolate flavoured pellets in the three groups (mean±SEM, repeated measures ANOVA, session × treatment effect ***p<0.001, post hoc Newman–Keuls, ˆp<0.05 control vs lactulose, &p<0.05 lactulose vs rhamnose, #p<0.05 control vs rhamnose). (B–D) Behavioural tests for the three addiction-like criteria in the late period (individual values with IQR) in the addicted and non-addicted groups. (B) Persistence of response. (C) Motivation (Mann–Whitney U test, *p<0.05). (D) Compulsivity (Mann–Whitney U test, *p<0.05). (E) Percentage of mice classified as addicted and non-addicted in the lactulose, rhamnose and control groups. (F–I) Behavioural tests for the four phenotypic traits associated with vulnerability to food addiction in the late period (individual values with IQR). (F) Impulsivity. (G) Cognitive inflexibility (Mann–Whitney U test, ***p<0.001). (H) Appetitive cue reactivity. (I) Aversive cue reactivity. (J) Food intake. (K) Water intake. (L) Body weight. (M) (Blautia)/g in mice faeces determined by pPCR (Mann–Whitney U test, *p<0.05). The sample size of mice in the lactulose and rhamnose groups was n=12, and n=17 in the control group (total n=41). Statistical details are included in online supplemental table S3.

Figure 8

Principal component analysis (PCA) revealed differential patterns of behavioural factor loadings in food addiction-like behaviour in mice treated with lactulose and rhamnose. (A) Mice subjects clustered by addicted or non-addicted classification on the space yielded by two components of the PCA that account for the maximum data variance (n=5 control addicted mice, n=12 control non-addicted mice, n=12 lactulose non-addicted mice, n=12 rhamnose non-addicted mice). (B) Criteria belonging to each component, principal component (PC) 1 (37.7%) and PC2 (20.9%). (C, D) Order of factor loading of the different variables in PC1 and PC2 is represented. The dashed horizontal line marked loadings >0.7, mainly contributing to the component. (E) Heatmap correlation matrix of the three addiction criteria and the four phenotypic traits. Colours correspond to the magnitude of Pearson correlations between each pair of variables and range from −1 (red) to+1 (blue). Significant Pearson’s correlations: *p<0.05, **p<0.01, ***p<0.001 (n=36 mice).

We next analysed the faecal microbiota composition of mice from the functional validation experiment after oral administration of lactulose and rhamnose. We found that supplementation with rhamnose increased the levels of several species from the SCFA producing family Lachnospiraceae, including several species from the genus Blautia, such as Blautia faecis, Blautia sp, and Blautia_uc (online supplemental figure S7A). Online supplemental material has a detailed description of the microbiota composition results shown in online supplemental figure S7.

Functional validation with the beneficial microbe Blautia wexlerae in the mouse food addiction protocol

To further demonstrate that Blautia has a protective effect against the development of food addiction, oral administration of Blautia wexlerae in mice that underwent the long protocol of food addiction (6-FR1 and 114-FR5 sessions) described in the previous section was performed. We used the same regimen of chronic Blautia oral administration 3 days per week at a concentration (1×109 CFU) similar to what was described before to prevent obesity and type 2 diabetes.26 For this purpose, a total of 37 C57BL/6 J mice underwent an operant protocol of 120 sessions (figure 9). Online supplemental material has a detailed description of the behavioural results shown in figure 9.

Figure 9

Characterisation of extreme subpopulations of addicted and non-addicted mice in the experiment with Blautia wexlerae supplementation. (A, upper part) Timeline of the procedure of operant behaviour mouse model. Mice were trained during the first 6 days in operant behaviour sessions of 1 hour at a fixed ratio (FR) 1 schedule of reinforcement, followed by 114 daily sessions of FR5. The addiction-like criteria (persistence of response, motivation and compulsivity) were evaluated in the late period (98-114) to categorise mice as addicted and non-addicted. Mice received the beneficial microbe Blautia wexlerae or vehicle control, which were administered by oral route (gavage) during the whole experimental sequence. Specifically, 250 µl of Blautia wexlerae were orally administered (intragavage) at a concentration of 1×109 CFU three times per week for the whole experimental protocol, 1 hour before the self-administration session in the operant chambers. (A bottom part) Number of reinforcers during 1 hour of operant training sessions maintained by chocolate flavoured pellets in the two groups (mean±SEM, repeated measures ANOVA, sessions, ***p<0.001). (B–D) Behavioural tests for the three addiction-like criteria in the late period (individual values with IQR) in the addicted and non-addicted groups. (B) Persistence of response. (C) Motivation (Mann–Whitney U test, *p<0.05). (D) Compulsivity (Mann–Whitney U test,*p<0.05). (E) Percentage of mice classified as addicted and non-addicted in the groups of Blautia wexlerae and control. (F–I) Behavioural tests for the four phenotypic traits associated with vulnerability to food addiction in the late period (individual values with IQR). (F) Impulsivity. (G) Cognitive inflexibility. (H) Appetitive cue reactivity. (I) Aversive cue reactivity. (J) Food intake. (K) Body weight. Sample size of mice in the Blautia wexlerae and vehicle control groups was n=18–19 per group (total n=37). Statistical details are included in online supplemental table S4.

Importantly, Blautia treated mice showed similar persistence of response, but significantly reduced motivation, and decreased compulsivity for highly palatable food compared with control mice, revealing a protective effect of Blautia to develop food addiction (Mann–Whitney U test, p<0.05, figure 9B–D, online supplemental table S4). Remarkably, 21.05% of control mice reached 2–3 addiction criteria in the late period (98–114 sessions), whereas none of the mice receiving Blautia wexlerae were classified as addicted (χ2=4.80, control vs Blautia p<0.05, figure 9E). No significant differences between groups were found in impulsivity, cognitive inflexibility, appetitive cue reactivity and aversive cue reactivity (figure 9F–I). In addition, all mice had similar food intake and body weight during the experimental sequence (figure 9J, K).

PCA and correlation heatmap were performed to understand further the correlation between Blautia administration and behavioural phenotypes leading to the prevention of food addiction. Online supplemental material has a detailed description of the results shown in figure 10.

Figure 10

Principal component analysis (PCA) revealed differential patterns of behavioural factor loadings in food addiction-like behaviour in mice treated with Blautia wexlerae or control vehicle. (A) Mice subjects clustered by addicted or non-addicted classification on the space yielded by two components of the PCA that account for the maximum data variance (n=4 control addicted mice, n=15 control non-addict