Dietary food patterns as determinants of the gut microbiome–endocannabinoidome axis in humans

The gut microbiota and the endocannabinoidome (eCBome) play important roles in regulating energy homeostasis, and both are closely linked to dietary habits. However, the complex and compositional nature of these variables has limited our understanding of their interrelationship. This study aims to decipher the interrelation between dietary intake and the gut microbiome–eCBome axis using two different approaches for measuring dietary intake: one based on whole food and the other on macronutrient intakes. We reveal that food patterns, rather than macronutrient intakes, were associated with the gut microbiome–eCBome axis in a sample of healthy men and women (n = 195). N-acyl-ethanolamines (NAEs) and gut microbial families were correlated with intakes of vegetables, refined grains, olive oil and meats independently of adiposity and energy intakes. Specifically, higher intakes in vegetables and olive oil were associated with increased relative abundance of Clostridiaceae, Veillonellaceae and Peptostreptococaceae, decreased relative abundance of Acidominococaceae, higher circulating levels of NAEs, and higher HDL and LDL cholesterol levels. Our findings highlight the relative importance of food patterns in determining the gut microbiome–eCBome axis. They emphasize the importance of recognizing the contribution of dietary habits in these systems to develop personalized dietary interventions for preventing and treating metabolic disorders through this axis.

system.Of note, germ-free mice harbor altered circulating and tissue levels of eCBome mediators compared to conventionally reared mice, while fecal matter transplant reverse most of these changes 6 .Interestingly, some evidence also highlights the production of endocannabinoid-like mediators by gut bacteria [7][8][9] .
The presence of eCBome mediators has been shown in virtually all tissues, especially in the brain, immune cells, adipose tissue and gut.These mediators can also be measured in peripheral circulation, where their origin and physiological significance remain uncertain.Circulating levels of eCBome mediators have been strongly associated with BMI [10][11][12] , NAEs being mainly associated with total fat mass while 2-MAGs seems more closely related to visceral fat mass [13][14][15] .Moreover, circulating eCBome mediators have also been shown to be associated with the fatty acid intakes in cross-sectional cohorts using dietary recalls as well as in the context of dietary interventions [15][16][17][18] .Indeed, a 2-day Mediterranean diet full-feeding intervention, modulates the relative levels of circulating eCBome mediators in a manner reflecting the fatty acid composition of the diet, which means rich in monounsaturated (MUFA) and polyunsaturated (PUFA) fatty acid and poor in saturated fatty acids 15,19 .The profile of dietary fatty acids appeared as a key determinant of the circulating lipidomic profile independently of the increased eCBome tone associated to adiposity.Of note, the Mediterranean diet recommendations are not focused on lipid intakes, but mostly on the consumption of a nutrient rich food such as olive oil, vegetables, fruits, whole grains, legumes, nuts and fish 20 .
We have previously published that a short-term (i.e., 2 days) adherence to the Mediterranean diet recommendations is sufficient to modulate the relative abundance of several gut microbiota taxa 19 and the circulating endocannabinoidome 15 .However, in a cross-sectional analysis, the macronutrient intake, except fatty acids intake, per se is not related to the circulating endocannabinoidome 15 .We thus hypothesize that whole-diet food pattern, rather than its macronutrient contents, is associated with gut microbiota and the circulating eCBome.We, therefore, aimed here at deciphering the associations between the dietary intake and the gut microbiome-eCBome axis using two different dietary intake approaches measuring whole food and macronutrient intakes.The results obtained provide novel insight on how the gut microbiome-eCBome axis is modulated by the diet and this new knowledge will pave the way to improving the design of future preventive or therapeutic nutritional strategies for metabolic health.

Study cohort
The environment and Microbiota-EndoCannabinoidome Axis (eMECA) cross-sectional clinical trial (NCT03463304) included 102 women and 93 men covering a large range of adiposity phenotypes (Table 1), as described previously 15 .All recruitment was performed during the winter and spring of the same year.Individuals with enteropathies, alcohol consumption exceeding the Canadian recommendation for men (> 15) and women (> 10 drinks/week), weight change (± 5 kg) in the last 6 months, having taken antibiotics in the last 3 months and pregnant and/or breastfeeding women were not eligible.Participants who consumed cannabis more than once per week or the week preceding the study (n = 5) were excluded, while current tobacco smokers were not excluded (n = 10).Written informed consent was obtained and the project was approved by the Laval University Ethics Committee (2017-328).Detailed data collecting and sampling procedures have been previously described 15 .

Food intakes
Dietary intakes were determined using a self-administered web-based 24 h dietary recall (R24W) validated in this population 21 , as described previously 15 .Food groups were categorized according to the calculation of the Mediterranean food pattern: whole grains, refined grains, fruits, fruit juices, legumes/nuts, vegetables, vegetable juices, olive oil, milk and substitute, fish and sea foods, poultry, eggs, sweets and red meat/processed meat 22 .Coffee and tea consumption were also considered because of purported associations with the gut microbiota and eCBome system.

16S rRNA gene sequencing
Stool bacterial DNA was extracted (QIAGEN, CA, USA), V3-V4 libraries were prepared (Illumina and Axygen Biosciences, CA, USA) and sequences were processed as previously described for this cohort 15 .Diversity indexes were calculated for all samples at an even depth from rarefied OUT table (Vegan R package).For statistical analysis bacterial relative abundances were normalized using Cumulative Sum Scaling (CSS, MetagenomeSeq R package) as specified in the result section and figure legends.

Statistical analyses
Correlation between dietary intakes, gut microbiota composition and eCBome mediators were computed using Spearman's rank correlations.Multiple factor analysis (MFA), a dimensionality reduction method tailored to handle multiple groups of variables, was performed with the FactoMineR R package (version 2.

Ethical approval
All experiments and methods were performed in accordance with relevant guidelines and regulations.

MFA of the gut microbiota-eCBome axis
The direct association between circulating eCBome mediators, gut microbiota composition and dietary fatty acid intakes have been previously described in the present cross-sectional cohort as well as in a short-term dietary intervention 15,19 .To decipher the complex interrelation between the gut microbiome-eCBome axis with the diet, we have computed a multiple factor analysis (MFA) model including dietary intakes as two independent factors: one with whole foods (Food groups) and the other with macronutrient intakes (Macronutrients).(Fig. 1).The first four dimensions explain 26.6% of the total variance of the model.Sex was included as a supplementary variable in the MFA model (Fig. 1c and d).Women and men showed a similar distribution in all dimensions of the model and confidence interval ellipse are overlapping.
The first dimension, which explains 10.2% of the variance, was mostly driven by Adiposity and Metabolic profile variables (Fig. 2a).The circulating levels of 2-MAGs were also associated with the variance of the first dimension, in accordance with previously reported positive correlations with total and visceral fat mass 15 .Concerning the second dimension of the MFA, variables describing food intakes were its only considerable drivers (Fig. 1a  and 2b).Overall, this dimension gathered common variances between Food groups and Macronutrients intakes.
Interestingly, circulating levels of all NAE mediators contribute to the dimensions 3 and 4 of the MFA.Adiposity variables were only contributing to the first two dimensions of the model.These observations thus imply that the variance of NAEs explained in the third and fourth dimension of the model is independent of adiposity.Interestingly, the dimensions 3 and 4 were characterized by the contribution of several variables of Gut microbiota and Food groups (Fig. 1b).It is noteworthy, however, that variables within Food groups, but not within Macronutrients, were significant contributors to these dimensions (Fig. 2c and d).
In summary, dimensions 3 and 4 of the MFA comprised a subset of variables involved in a complex interrelation between the gut microbiota-eCBome axis and dietary intakes.By design, the model highlights the contribution of these variables while taking into consideration key endogenous variables of these systems, namely adiposity and metabolism.It appears that the relative abundance of gut microbiota families and the circulating levels of eCBome mediators were more closely associated to dietary intake measures based on food groups rather than on specific macronutrients.As previously demonstrated, the associations between macronutrients and mediators of the endocannabinoidome showed virtually no statistical significance 15 .Therefore, subsequent analyses were focused on the contribution of food groups on either system.

Food groups interrelation with families and circulating NAEs
Several food groups and microbial families contribute to both dimensions 3 and 4 and these dimensions explain a certain amount of variance in the circulating levels of NAEs.Variations in circulating levels of NAEs among individuals are highlighted using a color gradient in Fig. 3a.The gradient increases in parallel to the vector x* pointing toward the lower right of the dimensions 3 and 4 planes (Fig. 3a).We computed the vector direction and proceeded with an octagonal rotation of the plane so that the rotated plane optimized the variance of circulating NAEs levels on the Dimension x* (Fig. 3b).The bacterial families contributing the most to dimension x*, and therefore to the variance of circulating NAEs, were Clostridiaceae, Peptostreptococaceae and Veillonellaceae.Other families were more closely following Dimension y* of the rotated plane which was not associated with variances of either NAE or 2-MAG eCBome mediators.To highlight the direction of the association among food groups, families and NAE variables, we stratified the cohort in two clusters according to the Dimension x* coordinate of each individual.
In agreement with the MFA model, individuals in the cluster 2 had significantly higher circulating levels of all NAEs than individuals in cluster 1 (Fig. 3c).Both groups had similar adiposity measures but individuals with higher circulating NAEs levels (i.e., cluster 2) were characterized by higher levels of LDL and HDL cholesterol than individuals in cluster 1.This analysis also highlights that this phenotype (representing by cluster 2) was associated with significantly lower relative abundance of Acidaminococcaceae but higher relative abundance of Clostridiaceae, Peptostreptococaceae and Veillonellaceae.However, these differences were not associated with   www.nature.com/scientificreports/global differences in gut microbiota composition.Indeed, there was no difference in any of the diversity indexes (i.e., Chao1, Simpson and Shannon) even though some food group intakes (i.e., refined grains and meats) were associated with diversity indexes (Supplementary Table 1).Individuals with high circulating levels of NAEs (cluster 2) reported lower intakes of refined grains and meats but higher intakes of vegetables, olive oil, fish and tea and coffee.These differences in food intakes corresponded to a slightly reduced intake from carbohydrates in cluster 2. Total protein and fat (i.e., total, SFA, MUFA, PUFA) intakes were similar between clusters (Supplementary Fig. S1).

Network of food groups and the gut microbiota-eCBome axis
To visualize the relationship among the intake of various food groups, gut microbiota and the eCBome, we computed a correlation network (Fig. 4).Dietary intakes of macronutrients were not sufficiently associated with circulating eCBome mediators to enable similar analysis (Fig. 2c,d).The correlation network based on food groups highlighted interactions between the circulating levels of NAEs and food groups and gut microbiota families (Fig. 4 and Supplementary Fig. S2).Moreover, several direct correlations between gut microbiota families and food group intakes were noted.Indeed, the network revealed that the consumption of several food groups was associated with the abundance of specific gut microbiota taxa and with the circulating levels of eCBome mediators.Interestingly, two bacterial families shared an association with the consumption of two or more food groups.Clostridiaceae relative abundance was associated with intakes of refined grains, fruits and vegetables, while Rikenellaceae was associated with refined grains and olive oil.On the other hand, refined grains and fruits were the food groups with the highest number of correlations with bacterial families.

Discussion
Dietary intakes are known as a key determinant of the circulating eCBome mediator profile and gut microbiota composition.However, there is still an important knowledge gap in understanding the intricate relationships between these systems and food intake.In a well-characterized cohort of healthy men and women, we have compared two approaches, one based on whole foods and the other on macronutrients, to decipher the association between the diet and the gut microbiome-eCBome axis.To our knowledge, this study is the first to directly compare how intake of different foods and macronutrients explains the interindividual variations in the gut-microbiome-eCBome axis.Moreover, the analysis enabled us to isolate a subset of variables of these two "omes" and of food intakes that share common variance independently of adiposity and other confounding factors affecting these systems.Overall, our analysis suggests that food patterns are more closely related to gut microbiota composition and circulating levels of eCBome mediators than the macronutrient composition of the diet.We revealed that vegetables, refined grains, and olive oil are the main food contributing to circulating NAEs levels, and this specific food pattern is also associated with the relative abundances of the Clostridiaceae, Peptostreptococaceae and Veillonellaceae families in the gut microbiota.
As could be expected, dietary macronutrients content and food groups are highly related variables which, nevertheless, measure distinct aspects of dietary intakes.Consumption of a specific food group could be considered a proxy of specific macronutrient intake, as in the case of, e.g., a diet rich in fruits, vegetables, legumes and whole grains, which is associated with a high fiber intake.However, we demonstrate that macronutrients could not explain as much variance in the gut microbiome composition and the circulating eCBome profiles.Such observation could arise from the fact that food groups, in addition to their macronutrients content, are also associated with differences in other nutrients and in food matrix.The food matrix influences digestion and absorption as well as nutrient availability to the gut microbiota 26 .Moreover, whole foods may surpass the effect of nutrients itself on microbial metabolism and host signaling pathways.
This study takes advantage of food groups based on the Mediterranean diet pattern.Consumption of a diet rich in vegetables, fruits, legumes, nuts, whole grains, MUFA-rich oil and fish has proved to provide many beneficial cardiometabolic effects 27 .We showed that consuming a Mediterranean diet for 48 h alters circulating bioactive lipids, i.e., eCBome and short chain fatty acids 15,19 .Relatively few studies have related changes in gut microbiota composition with the beneficial metabolic effects of this diet.Garcia-Mantrana et al. 28 showed that the relative abundance of the Christensenellaceae family was associated with a higher adherence to the Mediterranean diet and negatively associated with adiposity.Here, increased prevalence of Christensenellaceae was associated positively with whole-grain foods consumption and negatively with LEA and PEA circulating levels.However, this association was not observed in the MFA and was not observed in correlations adjusted for fat mass (data not shown).PEA and LEA can respectively bind the G-protein-coupled receptors (GPCR) 55 and 119 and thus improve glucose homeostasis 2 .It remains difficult to predict the overall metabolic impact of changes in the circulating levels of these eCBome mediators because their origin and role in circulation are poorly documented, and each mediator binds different types of receptors to generate a complex cellular response.Nevertheless, as mentioned earlier, this association is aligned with the respective roles on metabolic health of these members of the eCBome 29 .
We highlight novel associations between the relative abundance of the Peptostreptococcaceae family with eCBome mediators, such as higher levels of this taxon were observed in individuals with higher circulating levels of all NAEs, apart from OEA.To our knowledge, no potential association has been previously reported between the Peptostreptococcaceae family and metabolic health parameters, macronutrients, or whole food consumption.Interestingly, in our analysis this taxon was not directly associated with macronutrient intakes but shared a significant amount of variance with the intakes of food groups and NAEs in the MFA.Indeed, independently of other variables included in the model (i.e., food intakes, adiposity and metabolic profile), the relative abundance of Peptostreptococcaceae family was higher in individuals with elevated levels of circulating NAEs.
We observed that Ruminoccocaceae relative abundance was associated to dietary intakes of whole-grain foods, from which we can infer a higher intake of fibers.While this taxon was not directly associated with LEA and PEA, these eCBome mediators were negatively associated with the intakes in whole-grain foods.
An association between fruit consumption and Akkermanciaceae (i.e.Akkermansia muciniphila) relative abundance was found and both variables have been linked to improved metabolic health 31,32 .We previously reported that 2-EPG levels are negatively associated with Akkermanciaceae independently of adiposity 15 .However, when including all groups of variables, i.e., dietary intake, adiposity, clinical parameters, Akkermanciaceae was not a significant contributor to any dimension of the model.
Numerous studies have already demonstrated that a significant part of the gut microbiota composition variance is explained by the diet 33 , but also by variations in whole food intakes as well as in food patterns 28,34,35 .As the synthesis of different eCB congeners is partly defined by the profile of fatty acids in their phospholipid precursors, we have previously shown that the fatty acid profile of dietary lipids is associated with the profile of the circulating eCBome mediators 15 .The present analysis fills in the knowledge gap of the most potent measures relating dietary intakes and the gut microbiota-eCBome axis.Indeed, here we show that the consumption of specific food groups related to the Mediterranean diet was more closely related to the gut microbiota composition and to the profile of circulating eCBome mediators than intake of specific macronutrients.It should be kept in mind that intakes of food and of macronutrients are closely interrelated.Indeed, we found that higher consumption of olive oil, an important source of oleic acid, is positively associated with circulating levels of OEA, an oleic acidderived molecule, even though dietary intakes of oleic acid did not significantly correlate with circulating OEA levels 15 .This finding suggests that oleic acid derived eCBome mediators may be more closely related to specific food groups and thus points to the existence of interactions with eating habits related to higher consumption of olive oil.Similarly, circulating levels of AEA and 2-AG were positively associated with consumption of food rich in the eCB precursor arachidonic acid, such as poultry and red meat.However, intake of arachidonic acid was not directly associated with circulating AEA and 2-AG levels 15 .All 2-MAGs congeners were associated with higher consumption of meats, but these associations seem to be secondary to the strong interrelation of these variables with adiposity and the metabolic profile shown in the first dimension of the MFA.Indeed, 2-MAG levels in circulation are strongly associated with visceral adiposity 15 and so are, on average, meat and processed meat intakes (Supplementary Fig. S3).These independent associations with specific eCBome mediators reinforce the importance of the whole food intake in determining the circulating eCBome mediator profile.
Fish consumption has also been positively associated with some omega-3 eCBome mediators such as EPEA, DHEA and 2-EPG.Overall, fish are rich in the omega-3 fatty acids EPA and DHA, which act as ultimate biosynthetic precursors for these eCBome mediators.Previous studies in mice and humans have demonstrated that fish oil consumption can modify the circulating profile of eCBome mediators, in a way that omega-3 derived mediators are increased, and other mediators are decreased 17,18,36 .This data suggests that the latter mediators are sensitive to the diet, including both dietary intakes of these fatty acids and food

Figure 1 .
Figure 1.Visualization of the multiple factor analysis (MFA) modeling the gut microbiota eCBome axis parameters with variables of adiposity, clinical parameters, food groups intakes and macronutrients intakes.Graph of factors contribution to (a) dimensions 1 and 2, and of (b) dimensions 3 and 4 of the MFA model.Graph of individuals of (c) dimensions 1 and 2, and of (d) dimensions 3 and 4 with sex as factor variables (Women: Blue, Men: Red).Ellipses are standard deviation from the mean center of each group of individuals.

Figure 2 .
Figure 2. Contribution of variables of the multiple factor analysis (MFA) to (a) dimension 1, (b) dimension 2, (c) dimension 3 and (d) dimension 4. Each graph includes the contribution of each factor (Left) as well as the contribution of individual variables (Right).The dashed line corresponds to the expected value if the contribution of each variable was uniform.

Figure 3 .
Figure 3. Dimension reduction to optimize the circulating NAEs.(a) Original and (b) rotated graph of individuals of dimensions 3 and 4. The color gradient represents the mean Z-score of NAEs mediators.The doted vectors correspond to the rotated plan which optimized the variance of NAEs through the new rotated Dimension x*.(c) Adiposity, metabolic profile, gut microbiota families, NAEs and dietary intakes (portions for all groups and mL for tea/coffee) of individuals clustered using the dimension x* coordinates.Boxplots include the median, lower/higher quartiles and 1.5 × inter-quartile range whiskers.The means of the distribution is represented by a + sign (n = 97 and n = 98 for clusters 1 and 2 respectively).*Adjusted p < 0.05 (Wilcoxon sign rank test with Holm-Bonferroni correction), # p < 0.05 before adjustment for multiple comparison (Wilcoxon sign rank test).Holm-Bonferroni corrections have been performed within each group of variables: adiposity and metabolic parameters (n = 11), NAEs and 2-MAGs (n = 13), taxa's families (n = 27) and food groups (n = 15).

Figure 4 .
Figure 4. Correlation network of NAEs, gut microbiota families and food groups.Nodes includes all eCBome, microbial families and dietary variables significant contributing to the dimensions 3 and 4 of the MFA.Edges identify significant spearman correlation coefficient between variables (p < 0.05, n = 195).Correlations within each variable groups are illustrated by wide grey edges while correlation between food groups and microbial families are indicated by green edges.Only correlation involving NAEs with either food groups or NAEs were colored according to the direction of the correlation (Blue: positive, Red: negative) as indicated in the legend.

Table 1 .
Characteristics of the study sample.