The Gut Microbiome Modifies the Association Between a Mediterranean Diet and Diabetes in USA Hispanic/ Latino Population

Abstract

Context

The interrelationships among the gut microbiome, the Mediterranean diet (MedDiet), and a clinical endpoint of diabetes is unknown.

Objective

To identify gut microbial features of a MedDiet and examine whether the association between MedDiet and diabetes varies across individuals with different gut microbial profiles.

Methods

This study included 543 diabetic, 805 prediabetic, and 394 normoglycemic participants from a cohort study of USA Hispanic/Latino men and women. Fecal samples were profiled using 16S rRNA gene sequencing. Adherence to MedDiet was evaluated by an index based on 2 24-hour dietary recalls.

Results

A greater MedDiet adherence was associated with higher abundances of major dietary fiber metabolizers (e.g., Faecalibacterium prausnitzii, false-discovery-rate–adjusted P [q] = 0.01), and lower abundances of biochemical specialists (e.g., Parabacteroides, q = 0.04). The gut microbiomes of participants with greater MedDiet adherence were enriched for functions involved in dietary fiber degradation but depleted for those related to sulfur reduction and lactose and galactose degradation. The associations between MedDiet adherence and diabetes prevalence were significantly stronger among participants with depleted abundance of Prevotella (p_interaction = 0.03 for diabetes, 0.02 for prediabetes/diabetes, and 0.02 for prediabetes). A 1-SD deviation increment in the MedDiet index was associated with 24% (odds ratio [OR] 0.76; 95% CI, 0.59-0.98) and 7% (OR 0.93; 95% CI, 0.72-1.20) lower odds of diabetes in Prevotella noncarriers and carriers, respectively.

Conclusion

Adherence to MedDiet is associated with diverse gut microorganisms and microbial functions. The inverse association between the MedDiet and diabetes prevalence varies significantly depending on gut microbial composition.

microbiome, diabetes, diet

Diabetes can be largely prevented through adherence to a healthful diet and lifestyle (1). The human gut microbial composition and function are highly adaptive to host diet (2). By promoting energy extraction and metabolizing foods into numerous molecular cues (3), the gut-resident microbes may be involved in the development of diabetes (4). The individualized nature of the gut microbiome offers a possible mechanism to explain between-person variability in a particular diet’s effects on glucose homeostasis (5-7). Previously, Wu and colleagues proposed that the yield of microbial production of short-chain fatty acids (SCFAs), a group of metabolites implicated in host metabolic regulation (8), in consumers of plant-based diets may vary depending on the abundance of genus such as Prevotella in their gut (7). Most recently, we found that the protective associations of a Mediterranean dietary pattern (MedDiet) with glycated hemoglobin A1c (HbA1c) and plasma biomarkers of lipid metabolism and inflammation varied depending on microbial composition; the associations were significantly stronger in those whose gut microbiomes depleted of Prevotella copri in a cohort of 307 USA men (5). These prior studies shed important light on the unique role of Prevotella, a clade among the discrete gut community “enterotypes” consistently identified in the human population (9-11), in modifying the benefits of healthy diets for maintaining cardiometabolic health and provided an early glimpse of the link between individualized effects of diet and the gut microbiome. However, no study has formally tested whether the gut microbial profile can modify the association between diet and a clinical endpoint of diabetes. In addition, it is unknown whether the observed diet-microbiome interactions in predominantly White populations are generalizable in populations with different dietary and ethnogeographic backgrounds.

Diabetes disproportionately affects Hispanic/Latino population in the United States. In 2011-2012, the prevalence of diabetes in Hispanic Americans (22.6%) was twice that of non-Hispanic White Americans (11.3%) (12). It is well-known that geographic origin is a dominant factor in explaining the variation in the gut microbiomes (13). Several studies, including our own Hispanic Community Health Study-Study of Latinos (HCHS-SOL) cohort, found a decreased diversity in the gut microbiome with the immigration from lower-income countries to the United States (14, 15). Furthermore, certain microbes with low prevalence in the gut microbiomes of predominantly White Western populations, such as Prevotella copri, were frequent inhabitants of the gut of individuals from non-White racial/ethnic groups and nonindustrialized areas, and particularly responsive to changes in exogenous environment including diet (10, 11, 14). Therefore, our HCHS-SOL study population of predominantly foreign-born USA residents provided a unique opportunity to investigate the interrelationship between diet and the gut microbiome in this high-risk population.

The MedDiet, characterized by high intake of fruits, vegetables, nuts, legumes, and olive oil, and fewer red meats and refined grains, and low-to-moderate alcohol consumption (16), is associated with lower risk of diabetes in epidemiologic studies (17) and a randomized controlled trial (18). Various authorities recommend the MedDiet for the prevention of major chronic diseases including diabetes (19, 20). The MedDiet, differing from typical Western dietary patterns, was associated with specific gut microbial taxa (21-23). However, it is largely unknown whether a MedDiet interacts with the gut microbiome to exert its benefits for diabetes prevention.

In this study, we examined the interplays between a healthful Mediterranean-style dietary pattern, the gut microbiome, and diabetes in diabetic (n = 543), prediabetic (n = 805), and normoglycemic individuals (n = 394) from the HCHS-SOL. We hypothesized that the MedDiet would favorably affect the composition and function of the gut microbiome. Based on our prior study (5), we also hypothesized that Prevotella carriage modifies the association between the MedDiet and diabetes prevalence with a more pronounced inverse association in Prevotella noncarriers.

Methods

Study Design and Stool Sample Collection

HCHS-SOL is a prospective, population-based cohort study of 16 415 Hispanic/Latino men and women aged 18 to 74 years in the United States. From 2008 to 2011, HCHS-SOL recruited participants with diverse national origins from randomly selected households near the 4 field centers across the United States (24). The baseline in-person survey of the HCHS-SOL included comprehensive anthropometric measurements, urine and fasting blood sample collection, and interviewer-administered questionnaires that collected information on sociodemography, medical history, medications, and lifestyles. The second in-person visit of the HCHS-SOL (2014 to 2017) included an enrollment of participants for the gut microbiome ancillary study of HCHS-SOL. Every participant in the ancillary study was provided a self-collection kit to collect a single stool sample (ABC Medical Enterprises, Inc., Rochester, MN). This study included 1742 participants who underwent both microbial profiling of stool sample and dietary assessments. Details of study design and stool sample collection and processing can be found elsewhere (24) and in the Supplementary Text (25).

Ascertainment of Diabetes and Prediabetes

A fasting blood draw and a 2-hour oral glucose tolerance test (OGTT) were performed during the first and second in-person clinical visits of the HCHS-SOL. Participants were required to fast for at least 8 hours prior to the visit. Plasma glucose was assessed using a hexokinase enzymatic method (Roche Diagnostics Corporation, Indianapolis, IN). Glycosylated hemoglobin (HbA1c) was measured in EDTA whole blood using a Tosoh G7 Automated HPLC Analyzer (Tosoh Bioscience Inc., South San Francisco, CA). We followed the criteria of the American Diabetes Association (26) to define diabetes and prediabetes based on fasting plasma glucose, 2-hour OGTT, HbA1c level, and/or self-reported/documented use of high blood sugar or diabetes medications in the last 4 weeks. Detailed definitions can be found in the Supplementary Text (25). This study included a random sample of both newly diagnosed and prevalent diabetes and prediabetes patients who provided stool samples at the time of the second visit of the HCHS-SOL. Control participants were also randomly sampled from normoglycemic participants who provided stool samples at the second visit of the HCHS-SOL.

Dietary Assessment

Diet was assessed at baseline using 2 separate 24-hour dietary recalls, the first during the in-person visit, and the second via telephone 6 weeks later. Nutrient intake was calculated using the Nutrition Data System for Research software. We applied a MedDiet index to measure the adherence to the traditional dietary pattern consumed in the Mediterranean region. The MedDiet index, initially developed by Willett et al (16), was based on 9 food/nutrient items: vegetables, legumes, fruit, nuts, whole grains, red/processed meat, fish, alcohol, and the ratio of monounsaturated to saturated fat. The total MedDiet index ranged from 5 (nonadherence) to 45 (perfect adherence). The scoring criteria can be found in Supplementary Table 1 (25).

DNA Extraction and Microbiome Profiling

DNA extraction and 16S ribosomal RNA gene amplicon sequencing were carried out using the Earth Microbiome Project standard protocols (27). Amplicon polymerase chain reaction was performed on the V4 region of the 16S rRNA gene using the primer pair 515f and 806r with Golay error-correcting barcodes on the reverse primer. The amplicon pool was purified with the MO BIO UltraClean PCR (Qiagen, Venlo, Netherlands) cleanup kit and sequenced on an Illumina MiSeq sequencing platform. Sequence data were demultiplexed and minimally quality filtered using QIIME (version 1.9.1) (28). Sequence reads were then clustered into operational taxonomic units (OTUs), matched against the Greengenes reference database (version 13_8) and assigned taxonomy by QIIME. We derived predicted functions of the gut microbiome, as expressed as KEGG orthologies (KOs), using PICRUSt (29).

Statistical Analysis

To determine variability in the relative abundance of taxonomy, we calculated the Bray-Curtis (BC) dissimilarity metric for each participant. We performed principal coordinate (PCo) analysis (PCoA) based on the BC dissimilarity to detect potential intrinsic pattern of microbial community structure. We applied permutational multivariate analysis of variance (PERMANOVA; n = 999 permutations) to quantify percentages of variance in the microbial taxonomy explained by dietary variables, diabetes status, and covariables based on the BC dissimilarity metric. We filtered out microbial features with a relative abundance less than a threshold (0.0001 for taxonomic features and 0.00001 for functional features) in more than 20% of all samples. In addition to the filters, KOs with high correlations with others were removed by taking the most abundant feature from each such cluster as its representative. Given that the analytical database had minimal missing values in only 2 covariables, age at relocation to the USA mainland (missingness: 0.1%) and physical activity level (missingness: 0.3%), we assigned median values to replace the missing values. We employed the linear model in MaAsLin2, an R package designed for determining multivariable-adjusted associations between metadata and microbial features, to quantify the associations between dietary variables and microbial features (30, 31). We arcsine-transformed all the taxonomic and functional features before including them in the MaAsLin2 models. The covariables included diabetes status, age, sex, total energy intake, physical activity level, metformin use, antibiotic use, probiotic use, place of birth, age at relocation to the USA mainland, Bristol stool scale, and frequency of bowel movement. All of the P values estimated from MaAsLin2 models were corrected for multiple comparisons using the Benjamini-Hochberg procedure. We constructed phylogenetic associations in relation to the MedDiet index by applying GraPhlAn (32). We used the median of relative abundance of Prevotella as cutoff to define Prevotella carriers and noncarriers and tested whether the associations between the MedDiet index and the prevalence of 3 diabetes outcomes (prediabetes, diabetes/prediabetes, and diabetes) varied between Prevotella carriers and noncarriers. We applied logistic models that included MedDiet index, Prevotella carriage, and their product term, simultaneously adjusted for body mass index in addition to all the aforementioned covariables, and incorporated the sampling strata, clustering, and weights from the complex survey design of the HCHS-SOL. The likelihood ratio test was used for testing statistical significance of the product term. Secondarily, we used the first 2 PCo scores from the PCoA as summary measures of gut microbial profile and tested whether the associations between the MedDiet index and the prevalence of diabetes outcomes varied across levels of PCo scores using the same logistic models. We also quantified the associations between the MedDiet index and the prevalence of diabetes outcomes in stratum defined by Prevotella carriage and PCo scores. All the analyses were performed using SAS, version 9.4 (SAS Institute) or R software version 3.6.1 (R Foundation for Statistical Computing).

Results

The study population included 1119 women and 623 men. At the time of stool sample collection, the age of the study participants ranged from 23 to 83 years with an average age of 57.1 years. Participants who had a higher adherence to MedDiet consumed more beneficial components of the index including whole grains, vegetables, fruit, nuts, legumes, and monounsaturated fats (at the expense of saturated fats), but less red and processed meat, which is a detrimental component of the MedDiet index (Table 1). As expected, a majority of variation in the gut microbial taxonomy was driven by a tradeoff between the Bacteroidetes vs Firmicutes phyla (Supplementary Figure 1 (25)). The first axis of taxonomic variation (PCo1) was predominantly driven by the Bacteroides genus (correlation coefficient = −0.97), while the second axis of taxonomic variation (PCo2) was correlated with the Prevotella genus in a reasonably strong and inverse manner (correlation coefficient = −0.57; Supplementary Figures 2 and 3 (25)). The 12 most abundant genera together accounted for an average of 85% of community abundance.

Table 1.

Characteristics of study population across tertiles of the Mediterranean diet index

	Tertiles of MedDiet indexa			P valueb
	T1	T2	T3
n	569	640	533
Mediterranean diet index	27.4 ± 0.1	31.7 ± 0.1	36.2 ± 0.1	<0.001
Fruits, servings/day	1.0 ± 0.1	1.8 ± 0.1	2.7 ± 0.2	<0.001
Vegetables, servings/day	1.3 ± 0.1	1.9 ± 0.1	2.7 ± 0.1	<0.001
Whole grains, servings/day	0.8 ± 0.1	1.7 ± 0.1	2.5 ± 0.1	<0.001
Fish, servings/day	0.3 ± 0.1	0.7 ± 0.1	1.3 ± 0.2	<0.001
Legume, servings/day	0.1 ± 0.1	0.5 ± 0.1	1.5 ± 0.2	<0.001
Nuts, servings/day	0.1 ± 0.0	0.2 ± 0.1	0.6 ± 0.1	0.003
Monounsaturated/saturated fat ratio	1.1 ± 0.0	1.1 ± 0.0	1.2 ± 0.0	<0.001
Red/processed meat, servings/day	3.7 ± 0.1	2.2 ± 0.1	1.1 ± 0.1	<0.001
Alcohol, drinks/day	0.2 ± 0.2	0.2 ± 0.1	0.2 ± 0.2	0.28
Age, years	52.2 ± 1.0	56.2 ± 0.8	56.9 ± 1.2	<0.001
Bristol stool scale	3.6 ± 0.1	3.7 ± 0.1	3.7 ± 0.1	0.28
Frequency of bowel movement	1.7 ± 0.1	1.7 ± 0.1	1.7 ± 0.1	0.36
Total energy intake, kcal/day	1957 ± 27	1879 ± 37	1832 ± 41	0.02
Physical activity, MET-min/week	472 ± 47	579 ± 45	625 ± 61	0.34
Age at relocation to USA mainland, years	24.5 ± 1.2	23.7 ± 1.4	25.5 ± 1.6	0.56
Body mass index, kg/m2	30.8 ± 0.4	30.1 ± 0.4	29.6 ± 0.5	0.001
Hemoglobin A1C, %	6.2 ± 0.2	6.2 ± 0.2	6.0 ± 0.2	0.04
Fasting plasma insulin, mIU/L	17.2 ± 0.8	15.5 ± 0.6	13.9 ± 0.7	0.005
Fasting plasma glucose, mg/dL	113 ± 2.3	114 ± 2.9	109 ± 2.0	0.34
Sex, male	42.6 (5.0)	36.2 (4.4)	36.7 (4.6)	0.24
Current smoking	19.4 (5.1)	11.5 (4.8)	8.8 (4.9)	<0.001
Antibiotic use	29.6 (7.6)	29.7 (7.9)	28.1 (7.6)	0.29
Probiotic use	8.9 (4.3)	7.5 (4.0)	11.1 (4.7)	0.08
Metformin use	10.7 (4.3)	11.9 (4.0)	8.2 (4.3)	0.47
Place of birth				0.17
Central America	4.5 (2.0)	5.3 (2.1)	4.8 (2.3)
Cuba	29.3 (7.0)	23.0 (6.1)	11.8 (7.0)
South America	5.3 (2.4)	4.5 (2.4)	5.4 (2.4)
Mexico	21.9 (8.2)	34.9 (7.9)	54.8 (8.4)
Puerto Rico	12.0 (7.1)	9.1 (8.1)	7.5 (9.0)
Dominican Republic	7.0 (4.2)	9.2 (4.3)	6.8 (4.5)
Other country	0.4 (0.7)	0.1 (0.5)	0.6 (0.5)
USA mainland	19.6 (5.1)	13.8 (4.7)	8.3 (4.6)

All the variables except age were age standardized. Values are mean ± standard error for continuous variables and percentage (standard error) for categorical variables. All the calculations incorporated the sampling strata, clustering, and weights from the complex survey design of the Hispanic Community Health Study-Study of Latinos.

Abbreviations: MedDiet, Mediterranean diet; MET, metabolic equivalent tasks.

^aEach participant’s adherence to a healthy Mediterranean dietary pattern was evaluated by a 9-dimensional MedDiet index with a higher score indicating a better adherence. Tertiles were calculated based on the distribution of the Mediterranean diet index in normoglycemic participants. The Mediterranean diet index in tertiles 1, 2, and 3 ranged from 22.0 to 29.8, 29.8 to 33.5, and 33.5 to 44.5, respectively.

^bP values for linear trend across the tertiles of the Mediterranean diet index were derived from general linear models for continuous variables and logistic models for categorical variables.

The MedDiet index significantly covaried with the diversity of the gut microbiome (P = 0.001; Fig. 1A), although it explained a modest proportion of variance (0.4%; Fig. 1B). The MedDiet index explained a higher or comparable percentage of taxonomic variation than several covariables previously reported to have strong influence on the gut microbial communities, such as antibiotic use (0.5%) (33) and the Bristol stool scale (0.1%) (34). As expected, variables related to geographic origins of participants, including place of birth and age at relocation to the USA mainland, explained the largest percentages of taxonomic variation in this diverse population (35).

Figure 1.

Association between the adherence to the Mediterranean diet and the gut microbial taxonomy. (a) Association between the adherence to the Mediterranean diet (MedDiet) and overall variation of the gut microbiome based on principal coordinate analysis of genus-level Bray-Curtis dissimilarity. (b) The proportion of variation in taxonomy explained by dietary factors, diabetes status, and covariables based on permutational multivariate analysis of variance (n = 999 permutations). (c) Associations of MedDiet index and its components with microbial genera. The heatmap includes genus-level microbial features associated with the MedDiet index or one of its components (q ≤ 0.25) (d) Phylogenetic associations of MedDiet index and its components with microbial species. Ring colors correspond to the beta coefficients for dietary variables from linear mixed models that included the MedDiet index and its components as exposure variable and microbial features as outcome variables. Bar heights of the outer ring are proportional to mean relative abundances of microbial species. The plot includes species-level microbial features associated with the MedDiet index or one of its components (q ≤ 0.25). The associations between dietary factors and taxonomy in (c) and (d) were estimated by MaAsLin2 models that simultaneously adjusted for age, diabetes status, sex, total energy intake, physical activity level, metformin use, antibiotic use, probiotic use, place of birth, age at relocation to the USA mainland, Bristol stool scale, and frequency of bowel movement.

In the multivariable model, a total of 21 genus-level features from 4 phyla (Fig. 1C) and 45 species-level features from 5 phyla (Fig. 1D and Supplementary Figure 4 (25)) were associated with the MedDiet index or one of its components (false-discovery-rate–adjusted P [q] ≤ 0.25). We observed a general pattern such that the associations for plant foods were in the opposite direction compared the associations for red/processed meat (Fig. 1C). Several abundant dietary fiber metabolizers and SCFA producers, including Faecalibacterium prausnitzii, Coprococcus, and Lachnospira, were enriched in participants with a higher adherence to the MedDiet. The MedDiet index was inversely associated with microbes that include species-level features, such as Collinsella aerofaciens and Bifidobacterium bifidum, and genera-level features, such as Catenibacterium, Megasphaera, and Parabacteroides. Among the components of the MedDiet, whole grains, vegetables, and fruits were the major driving forces of the associations between the overall dietary pattern and the taxonomic features (Fig. 1C and Supplementary Figure 4 (25)).

The MedDiet index and its components were significantly associated with many microbial functions (KOs) involved in the metabolism of diet (q ≤ 0.05, Fig. 2A). As a general pattern, microbial functions involved in carbohydrate metabolism, except those involved in galactose and lactose degradation, were enriched in participants who had a higher adherence to the MedDiet, whereas microbial functions related to amino acid metabolism were depleted in those with a higher adherence to the MedDiet. Similar to our taxonomic findings, the associations of plant foods with functional features were in the opposite direction compared to those for red/processed meat intake. Fig. 2B shows representative associations between the MedDiet index and microbial functions and the major taxa that contributed to each function. We found lower abundances of bacterial functions involved in lactose and galactose degradation (K01819: galactose-6-phosphate isomerase and K01220: 6-phospho-β-galactosidase) in individuals with a higher MedDiet index (Fig. 2B). Participants with a greater adherence to the MedDiet had higher abundances of bacterial functions for the degradation of dietary fiber (K00008: L-iditol 2-dehydrogenase) and the metabolism of SCFAs (K00248: Butyryl-CoA dehydrogenase). The most abundant bacteria that encoded the enzymes included anaerobic fiber metabolizers, such as F. prausnitzii and Lachnospira spp. (Fig. 2B, Supplementary Figure 5 and Supplementary Table 2 (25)). Lower adherence to the MedDiet was associated with the enzymatic action of bacteria in reducing dietary sulfur to H₂S (K01130: arylsulfatase and K08352: thiosulfate reductase), encoded mainly by sulfur-metabolizing bacteria including Bilophila and Parabacteroides spp., and Akkermansia muciniphila (Fig. 2B, Supplementary Figure 5 and Supplementary Table 2 (25)).

Figure 2.

Associations of the adherence to the Mediterranean diet with microbial processes involved in diverse diet-related functions. (a) Associations of the adherence to the Mediterranean diet (MedDiet) with microbial functions (as KEGG orthologies [KOs]) involved in degradation and metabolism of dietary components. Colors of the heatmap are in correspondence to the beta coefficient for dietary variables from MaAsLin2 models that simultaneously adjusted for age, diabetes status, sex, total energy intake, physical activity level, metformin use, antibiotic use, probiotic use, place of birth, age at relocation to the USA mainland, Bristol stool scale, and frequency of bowel movement. (b) Select MedDiet-associated microbial functions are involved in lactose and galactose degradation, metabolism of short-chain fatty acids, and sulfur reduction, and top microbial genera that contributed most to each KO. The order of participants was determined by the MedDiet index, from the lowest to the highest index, in the compositional plots.

We found that the association between adherence to the MedDiet and prevalence of diabetes varied significantly across Prevotella carriers vs noncarriers (p_interaction = 0.03 for diabetes, 0.02 for prediabetes or diabetes, and 0.02 for prediabetes; Fig. 3 and Supplementary Table 3 (25)). In particular, the protective association between the MedDiet index and diabetes prevalence was significantly stronger among participants with decreased abundance of Prevotella. Every 1-SD increment in the MedDiet index was associated with an odds ratio (95% CI) of 0.76 (0.59-0.98) for diabetes, 0.76 (0.62-0.93) for prediabetes or diabetes, and 0.76 (0.61-0.94) for prediabetes in Prevotella noncarriers, whereas the MedDiet index was not associated the diabetes outcomes in Prevotella carriers. In a secondary analysis, the association between adherence to the MedDiet and prevalence of diabetes also varied across levels of the second axis of taxonomic variation (PCo2, p_interaction = 0.16 for diabetes, 0.05 for prediabetes or diabetes, and 0.03 for prediabetes, Supplementary Table 4 (25)). The association of the MedDiet index with the prevalence of diabetes was stronger in participants with a higher PCo2, and weaker in those with a lower PCo2. Notably, the microbial feature that showed the strongest correlation with the PCo2 was Prevotella (correlation coefficient = −0.57, Supplementary Figures 2 and 3 (25)). The associations between the MedDiet index and the prevalence of diabetes did not significantly vary across levels of PCo1 (Supplementary Table 5 (25)).

Figure 3.

Association of adherence to a Mediterranean diet with the prevalence of diabetes and prediabetes in Prevotella noncarriers vs carriers. The inverse association of the Mediterranean diet (MedDiet) index with the prevalence of diabetes and prediabetes was significantly stronger in a subpopulation with gut microbiomes depleted of Prevotella. The odds ratios and their 95% CIs were calculated from logistic models that simultaneously adjusted for age, sex, total energy intake, physical activity level, metformin use, antibiotic use, probiotic use, place of birth, age at relocation to the USA mainland, current smoking status, Bristol stool scale, frequency of bowel movement, and body mass index. p for interaction was calculated from logistic models that included the MedDiet index, Prevotella carriage, and their product term, as well as the covariables above. All the models incorporated the sampling strata and weights from the complex survey design of the Hispanic Community Health Study/ Study of Latinos.

Discussion

In this study population of USA Hispanic/Latino participants who underwent microbial profiling of stool samples, comprehensive assessments of diet, and detailed ascertainments of diabetes including OGTT, we demonstrate that a healthful Mediterranean-style dietary pattern was associated with phylogenetically diverse gut microbes and their biochemical functions. Notably, our study, for the first time, identified a significant interaction between a proven clinically recommended dietary pattern and a gut microbial profile in relation to diabetes endpoints. The protective association between the MedDiet and the prevalence of diabetes was particularly strong in a subgroup of participants with the gut microbiomes depleted of Prevotella. This finding supports the premise that dietary interventions or recommendations for diabetes prevention could be tailored to an individual’s gut microbial profile.

Our study expands the literature on diet and microbiome (3, 7, 14) by linking a healthful dietary pattern to taxonomic features and their functions. Our findings demonstrate that a dietary pattern enriched for plant-based foods may preferentially select for bacteria that process dietary fiber to produce SCFAs, such as F. prausnitzii, and Coprococcus spp. (36). The taxonomic findings were further supported by our observations that many microbial enzymes involved in the carbohydrate metabolism, particularly degradation of dietary fiber and SCFA fermentation, were enriched in participants with a greater adherence to the MedDiet. Our findings on low abundance of B. bifidum, a lactose-fermenting bacterium, and bacterial functions involved in the lactose and galactose degradation in individuals with greater MedDiet adherence were consistent with the observation that low dairy food consumption as a key feature of the MedDiet (16). We linked a higher adherence to the MedDiet to the depletions of several niche-specific biochemical specialists such as Parabacteroides, E. lenta and C. aerofaciens, and microbial functions that mainly contributed by these bacteria, for example, sulfur metabolism. Prior research has linked E. lenta and select Collinsella and Parabacteroides species with Western-style diets and red meat intake (37-40). In addition, diets rich in red meats provide more sulfur-containing amino acids that are more conducive to the proliferation of sulfur-metabolizing bacteria, e.g., Bilophila wadsworthia and A. muciniphila (41), with increased microbial reduction of dietary sulfur to hydrogen sulfide that promotes inflammation at mucosal and systematic levels (42).

Our study revealed that a protective association between the MedDiet and diabetes prevalence was only observed in a subgroup of participants whose gut microbiomes were depleted of the Prevotella genus. This finding corroborates our recent report that a particularly strong inverse association between the MedDiet and HbA1c among a subgroup of participants could be explained by the absence of P. copri in their gut microbiomes (5), but it extends the diet-microbiome interaction to the hard clinical endpoint of diabetes. In addition, these consistent findings from 2 study populations, the HCHS-SOL and the Health Professionals Follow-Up Study (HPFS), with different racial/ethnic, dietary, and socioeconomic backgrounds and age ranges demonstrate the generalizability of the diet-microbiome interaction. Recently, the role of Prevotella, in particular the most representative species within this genus, P. copri, in the human gut microbiome has gained increasing interest because it represents one of the clearest cases of strain-level diversity (10, 11). Furthermore, diet may be a key driver in Prevotella populations and select for subspecies strains with different abilities in degrading complex carbohydrates and amino acids (10, 11), leading to the observed strong interaction between the dietary pattern and Prevotella. Our findings also support the recently proposed “diet-dependent” hypothesis of the Prevotella genus’s functions (4)—namely, that the link between Prevotella and glucose homeostasis (4, 43, 44) and inflammation (45) may only be observed among individuals with certain dietary patterns. However, it is not possible to conclude definitively that Prevotella affects metabolism of diet, vs alternative explanations for our data. For example, altered Prevotella abundance in response to MedDiet may be a marker of some other aspects of metabolism, and it is also possible that another gut microbial constituent correlated with Prevotella is the explanatory microbiome feature.

A unique feature of this study is that all the participants are Hispanics/Latinos, a population has been rarely investigated in previous microbiome studies. Prevotella follows an unusual ecological distribution in predominantly White populations in Western countries, that is, the genus was not detectable in a majority of participants (10, 11, 14), potentially leading to insufficient statistical power to detect an interaction between diet and Prevotella-enriched microbial profiles in most Western populations. Consistent with previous findings that additional clades of Prevotella appeared in residents of and immigrants from lower- and middle-income countries (10, 11, 14), we found that a large proportion of our study participants had relatively high abundance of Prevotella, providing us a unique setting to test the diet-microbiome interaction. Our study focused on a well-studied dietary pattern, which provides a global assessment of the healthfulness of diet, thereby circumventing the challenges of isolating specific dietary factors from highly correlated components of a diet and offering the promise of applying our findings to clinical intervention research. Our study has several limitations. First, because our study was observational in nature, causality cannot be established. Second, 16S amplicon sequencing does not provide high-resolution taxonomic profiling at species-level and does not allow for strain-level profiling. Third, microbial functions were predicted by PICRUSt rather than directly measured from metagenomes. The predictions were limited to the gene contents of existing reference genomes. Fourth, even though we adjusted for many potential confounders in our statistical models, residual confounding could not be ruled out. Fifth, measurement errors are inevitable in estimates of food and nutrient intakes, although the observed diet-microbe correlations appeared to have face validity and the results were robust despite inevitable inaccuracies in diet assessment. Finally, we conducted this study in a Hispanic/Latino population in the United States. The study findings may not be generalizable in other racial/ethnic groups and in other countries. However, as we discussed above, the findings are consistent with those observed in the Health Professionals Follow-Up Study of White male participants.

In summary, our findings support the premise to tailor dietary intervention according to individuals’ gut microbial profiles. This study represents a step forward in the concept of precision nutrition and has the potential to inform more effective and precise dietary approaches for the prevention of diabetes. In addition to addressing the aforementioned limitations, future studies that employ high-resolution profiling of gut microbiome are clearly needed to investigate the functional implication of diet-induced changes at strain-level resolution in human health, which would shed light on the molecule-level mechanisms underlying the role of diet-microbiome interaction in the prevention of diabetes.

Abbreviations

Abbreviations
 
• BC

  Bray-Curtis dissimilarity

 
• HbA1c

  glycosylated hemoglobin A1c

 
• HCHS-SOL

  Hispanic Community Health Study-Study of Latinos

 
• KO

  KEGG orthology

 
• MedDiet

  Mediterranean diet

 
• OGTT

  oral glucose tolerance test

 
• PCo

  principal coordinate

 
• PCoA

  principal coordinate analysis

 
• PERMANOVA

  permutational multivariate analysis of variance

 
• SCFA

  short-chain fatty acid

Acknowledgments

Financial Support: The Hispanic Community Health Study-Study of Latinos (HCHS-SOL) is a collaborative study supported by contracts from the National Heart, Lung, and Blood Institute (NHLBI) to the University of North Carolina (HHSN268201300001I/N01-HC-65233), University of Miami (HHSN268201300004I/N01-HC-65234), Albert Einstein College of Medicine (HHSN268201300002I/N01-HC-65235), University of Illinois at Chicago (HHSN268201300003I/N01-HC-65236 Northwestern Univ), and San Diego State University (HHSN268201300005I/N01-HC-65237). The following Institutes/Centers/Offices have contributed to the HCHS/SOL through a transfer of funds to the NHLBI: National Institute on Minority Health and Health Disparities, National Institute on Deafness and Other Communication Disorders, National Institute of Dental and Craniofacial Research, National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), National Institute of Neurological Disorders and Stroke, NIH Institution-Office of Dietary Supplements. Additional funding for the “Gut Origins of Latino Diabetes” (GOLD) ancillary study to HCHS-SOL was provided by R01MD011389 from the National Institute on Minority Health and Health Disparities. Dr. Dong D. Wang’s research is supported by R01NR019992 from the National Institute of Nursing Research, R00DK119412 from NIDDK and a Pilot and Feasibility award from the Boston Nutrition and Obesity Center funded by the NIDDK (P30DK046200). None of the funding agencies had a role in the design, conduct, interpretation, or reporting of this study.

Author Contributions: Study conception: D.D.W. and R.C.K. Manuscript preparation: D.D.W. and R.C.K. Data analysis: D.D.W., C.H., and R.C.K. Interpretation of data and critical revision of the manuscript for important intellectual content: All authors. Data and specimen collections: Q.Q., D.S.A., M.D.G., M.D., R.D.B. and R.C.K. Fecal sample processing: M.U. and R.D.B. Microbiome profiling: R.K. Obtaining funding: Q.Q., D.S.A., M.D.G., M.D., R.K., R.D.B., and R.C.K. Drs. Dong D. Wang and Robert C. Kaplan take responsibility for the contents of the article.

Additional Information

Disclosures: None declared.

Data Availability

HCHS/SOL data are archived at the National Institutes of Health repositories dbGap and BIOLINCC. Sequence data from the samples described in this study have been deposited in QIITA, ID 11666, and EMBL-EBI ENA, ERP117287. HCHS/SOL has established a process for the scientific community to apply for access to participant data and materials, with such requests reviewed by the project’s Steering Committee. These policies are described at https://sites.cscc.unc.edu/hchs/. The corresponding author will accept reasonable requests for data and specimen access, which will be referred to the Steering Committee of the HCHS/SOL project.