Long-term diet quality is associated with gut microbiome diversity and composition among urban Chinese adults

ABSTRACT

Background

Few population-based studies have evaluated the influence of long-term diet on the gut microbiome, and data among Asian populations are lacking.

Objective

We examined the association of long-term diet quality, comprising 8 food groups (fruit, vegetables, dairy, fish/seafood, nuts/legumes, refined grains, red meat, and processed meat), with gut microbiome among Chinese adults.

Methods

Included were 1920 men and women, enrolled in 2 prospective cohorts (baseline 1996–2006), who remained free of cardiovascular diseases, diabetes, and cancer at stool collection (2015–2018) and had no diarrhea or antibiotic use in the last 7 d before stool collection. Microbiome was profiled by 16S rRNA sequencing. Long-term diet was assessed by repeated surveys at baseline and follow-ups (1996–2011), with intervals of 5.2 to 20.5 y between dietary surveys and stool collection. Associations of dietary variables with microbiome diversity and composition were evaluated by linear or negative binomial hurdle models, adjusting for potential confounders. False discovery rate (FDR) <0.1 was considered significant.

Results

The mean ± SD age at stool collection was 68 ± 1.5 y. Diet quality was positively associated with microbiome α-diversity (P = 0.03) and abundance of Firmicutes, Actinobacteria, Tenericutes, and genera/species within these phyla, including Coprococcus, Faecalibacterium/Faecalibacterium prausnitzii, Bifidobacterium / Bifidobacterium adolescentis, and order RF39 (all FDRs <0.1). Significant associations were also observed for intakes of dairy, fish/seafood, nuts/legumes, refined grains, and processed meat, including a positive association of dairy with Bifidobacterium and inverse associations of processed meat with Roseburia /Roseburia faecis. Most associations were similar, with or without adjustment for BMI and hypertension status or excluding participants with antibiotic use in the past 6 mo.

Conclusion

Among apparently healthy Chinese adults, long-term diet quality is positively associated with fecal microbiome diversity and abundance of fiber-fermenting bacteria, although magnitudes are generally small. Future studies are needed to examine if these bacteria may mediate or modify diet–disease relations.

Introduction

The gut microbiota has been recognized as a key contributor to many aspects of human physiology, from nutrition and metabolism to immune responses and neurobehavioral traits (1, 2). Unlike the human genome, the gut microbiome is modified throughout the life span by various factors, including host genetics and health status, as well as diet, medication, and geographic location (3–6). Among those factors, diet plays a major role and is modifiable, which, in turn, has the potential to improve human health via changes to the gut microbiota. Intervention studies have shown that dietary changes can alter the gut microbiome within days to weeks, but the effects of short-term dietary changes seem transient, as the microbiome often regresses to baseline during the late stage or after cessation of the intervention (7–9). These findings support a causal role of diet in determining the gut microbiome while highlighting the importance of long-term studies to investigate the impact of habitual diets on the gut microbiome.

How habitual diets influence the gut microbiome is an ongoing hot research topic. Studies have found Western-style diets and high intakes of animal foods, sugary drinks, and saturated fats associated with low microbiome diversity and increased abundance of Bacteroides and production of harmful bacterial metabolites (e.g., trimethylamine-N-oxide) (10–12). In contrast, plant-based diets, such as vegetarian or Mediterranean-style diets, were associated with greater microbiome diversity, the abundance of Prevotella and fiber-fermenting Firmicutes, and the production of beneficial bacterial metabolites (e.g., SCFAs) (13–16). However, many previous studies had a small sample size and focused on individuals with distinct dietary patterns (e.g., Western compared with agrarian diets; omnivorous compared with vegetarian or vegan diets). Large-scale studies (sample size >1000) are emerging but remain limited (12, 15–18); among them, only a few have evaluated overall diet quality or patterns with the gut microbiome and yielded mixed findings. For example, in the TwinsUK (n = 2070) and Multiethnic Cohort studies (n = 1735), the Healthy Eating Index (HEI, based on the Dietary Guidelines for Americans) and Mediterranean Diet Score were both associated with the Shannon α-diversity index (15, 16), whereas in the Hispanic Community Health Study/Study of Latinos (n = 1674) and Malmö Offspring Study (n = 1726), the alternative HEI and a prudent dietary pattern were not associated with α-diversity indices, including the Shannon index (17, 18). Interestingly, the Hispanic Community Health Study/Study of Latinos found a significant association of the Shannon index with adulthood relocation from Latin America to the United States compared with US-born or childhood relocation (17). This is in line with another study showing that a longer residence of Asian immigrants in the United States was associated with lower microbiome diversity (19). These findings underscore the value of microbiome research among populations from different ethnic backgrounds and geographic locations. To our knowledge, no large-scale, population-based studies have evaluated the influence of overall diet quality and habitual food intakes on the gut microbiome among Chinese adults.

Leveraging resources of 2 prospective cohort studies conducted in Shanghai, China, we assessed long-term diet quality and habitual food intakes via repeated dietary surveys over a span of 15 y and evaluated their associations with gut microbiome diversity and taxonomic composition among 1920 men and women, who had no history of major chronic diseases (i.e., cardiovascular disease, cancer, and diabetes) or antibiotic use before stool sample collection.

Methods

Study participants

The Shanghai Women's Health Study (SWHS) and Shanghai Men's Health Study (SMHS) are population-based prospective cohorts, which enrolled 74,940 women and 61,480 men who were 40–74 y old and lived in urban communities in Shanghai, China, during 1996–2000 and 2002–2006, respectively (response rate was 93% for the SWHS and 74% for the SMHS) (20, 21). In-person interviews were conducted at baseline to collect information on sociodemographics, disease history, diets/lifestyles, and anthropometrics; biospecimens were also collected, including blood, urine, and/or oral rinse samples. Cohort participants were followed up for death and chronic disease outcomes via record linkages to Shanghai Vital Statistics and the Shanghai Cancer Registry (completion rates >99%) and follow-up surveys (response rates >92%). Diet, lifestyle, and anthropometric information was also updated during selected follow-up surveys. Both cohorts were approved by the Institutional Review Boards of the Shanghai Cancer Institute and Vanderbilt University Medical Center. All participants provided informed consent.

Stool sample collection

During the fifth follow-up of the SWHS and third follow-up of the SMHS (2015–2018), stool samples were collected from a total of 10,655 living and willing cohort participants, including 5526 women and 5129 men. Participants were provided stool sample collection kits, each of which included a collection paper band, disposable gloves, a tube containing 5 mL 95% ethanol and glass beads, and a cap with a scoop, a larger container tube, a biohazard bag, and a step-by-step instruction sheet. Participants were asked to collect a peanut-sized (1 full scoop) stool sample from the collection paper into the tube and then shake the tube until the sample was well mixed with ethanol. Participants were also asked to fill out sample collection forms to record the date/time of sample collection, use of antibiotics and medications in the past 7 d and 6 mo, usual bowel movement frequency, and whether they had diarrhea in the past 7 d. Stool samples were shipped to the research laboratory within 24 h after collection and divided into aliquots and stored in −80°C freezers until assayed.

Stool samples of 3358 participants were selected for microbiome profiling, including 1804 based on the availability of disease/biomarker data and 1554 randomly selected participants. Among those samples sent for sequencing, 164 were excluded due to low DNA amounts or failure to pass the quality control. For the current study, we further excluded participants lacking long-term dietary data (n = 692; those who only completed 1 dietary survey or had extreme total energy intakes, i.e., <500 or >3500 kcal/d for women and <800 or >4200 kcal/d for men); those who reported a diagnosis of cancer (n = 27), coronary artery disease (n = 53), stroke (n = 324), or diabetes (n = 113, including use of antidiabetic medications) before stool collection; or those who used antibiotics or had diarrhea in the last 7 d before stool collection (n = 65). A total of 1920 participants were included in our final analysis. A participant inclusion/exclusion flowchart is shown in Supplemental Figure 1.

Long-term diet assessment

Usual food intakes for the preceding 12 mo were inquired using semiquantitative FFQs. The FFQs used in both cohorts were validated against 24-h dietary recalls, which were administered for 12 consecutive months in cohort participant subsets (22, 23). Correlation coefficients between intakes assessed using FFQs and 24-h recalls were 0.41–0.66 in the SWHS and 0.42–0.72 in the SMHS for the major food groups (e.g., fruit, vegetables, fish, legumes, red meat, and refined grains), suggesting good validity of the FFQs. A healthy diet score (HDS) was generated based on 8 food groups—fruit, vegetables (excluding potatoes), dairy, fish and seafood, nuts and legumes, refined grains, red meat, and processed meat. The method for calculating HDS has been described elsewhere (24). Briefly, based on energy-adjusted, sex-specific intake quintiles, the first 5 food groups were assigned by ascending values (1 to 5), and the last 3 groups were assigned by descending values (5 to 1). HDS was the sum of those values, ranging from 8 to 40; the higher the score, the healthier the diet. HDS was significantly correlated with other diet quality indices available in our cohorts (25), including Dietary Approaches to Stop Hypertension score (Pearson correlation r = 0.43, adjusted for age, sex, and total energy intake) and Alternative Healthy Eating Index (adjusted Pearson r = 0.74), both P < 0.0001. A cumulative average of HDS was calculated to reflect long-term diet quality using data from repeated FFQs (26), which were administered 3 times in the SWHS, capturing dietary intakes during 1996–2011, and twice in the SMHS, capturing intakes during 2002–2011. Cumulative average intakes of food groups were also calculated using data from repeated FFQs.

16S rRNA metagenomics

DNA was extracted using QIAGEN's DNeasy PowerSoil kit. DNA libraries were prepared using the NEXTFLEX 16S V4 Amplicon-Seq kit. Sequencing was performed on Illumina HiSeq at 2 × 250-bp paired-end reads. Sequence reads were trimmed to remove adapters and filtered to remove low-quality reads using Sickle (27). BayesHammer was used to correct sequencing errors (28). PANDAseq was used to stitch paired-end reads (29). Clean reads were then clustered into operational taxonomic units (OTUs) at 97% identity against the Greengenes database (30) using a closed-reference OTU picking strategy via Quantitative Insights Into Microbial Ecology (QIIME, v1.9) (31). OTUs were rarefied to the minimum sequencing depth of our study samples (17,013 reads) using QIIME, and diversity metrics were calculated using R package vegan (32), including the Shannon index for α-diversity and the Bray–Curtis distance for β-diversity. Taxonomies from phylum to species were assigned; both raw sequence counts and relative abundance of each taxon within a sample were used in the following statistical analyses.

Statistical analysis

Dietary variables were modeled as sex-specific quintiles and each SD increment, including cumulative HDS, HDS assessed using the last FFQs, cumulative intakes for 7 food groups (fruit, vegetables, dairy, fish/seafood, nuts/legumes, refined grains, and red meat), and intake frequency of processed meat (which was assessed only at baseline). Covariates included sociodemographics (age at stool sample collection, sex, and income), lifestyle factors (cigarette smoking, alcohol drinking, and leisure-time exercise), total energy intake, BMI (in kg/m²), and history of hypertension (including use of antihypertensive medications). Covariate distributions by quintiles of cumulative HDS were compared using ANOVAs for continuous variables and χ² tests for categorical variables. Associations between dietary variables and the Shannon index were evaluated using a general linear model with additional adjustment for sample sequencing depth. Associations between dietary variables and the Bray–Curtis distance were evaluated using a permutational multivariate ANOVA with adjustment for covariates and 999 permutations. For individual taxa, we defined their presence as relative abundance ≥1/17,013 = 0.00588% in a sample (i.e., ≥1 read when there were 17,013 reads, the minimum sequencing depth of our samples). Common taxa were defined if present in >50% of participants; rare taxa were defined if present in 10–50% of participants, and taxa present in <10% of participants were excluded from analyses, as the effects of overall diet quality and major food groups on gut microbiome should be relatively common within a population. The general linear model was used to evaluate associations between dietary variables and read counts of common taxa after centered log-ratio transformation. The negative binomial hurdle model that handles zero-inflated data was used to evaluate associations between dietary variables and read counts of rare taxa. False discovery rates (FDRs) were calculated at each taxonomic level to account for multiple testing; FDR <0.1 was considered statistically significant. Stratified analyses were conducted by age (<65 or ≥65 y at stool sample collection), sex, and statuses of cigarette smoking, alcohol drinking, exercise, obesity (BMI <25 or ≥25), and hypertension. An interaction term of the dietary variable with a stratified variable was added to the regression model, and FDR-adjusted P-interaction <0.1 was considered significant. Sensitivity analyses were conducted by excluding 104 participants who reported use of antibiotics within 6 mo before stool collection. Statistical analyses were performed using R (version 3.6.3; R Foundation for Statistical Computing).

Results

Included were 903 women and 1017 men (Table 1), with a mean ± SD age of 68 ± 1.5 y at stool collection and intervals ranging from 5.2 to 20.5 y between stool collection and the last and baseline dietary surveys, respectively. In both men and women, higher HDS was associated with higher income and participation in leisure-time exercise. Almost all women were never smokers, and men with a higher HDS were more likely to be never smokers. Among men, higher HDS was also associated with older age and a history of hypertension. HDS was not significantly associated with BMI in either men or women, which averaged 24.1.

TABLE 1.

Characteristics of 1920 participants of the Shanghai Women's and Men's Health Studies by quintiles of long-term diet quality¹

	Women (n = 903)			Men (n = 1017)
Characteristic	Quintile 1	Quintile 3	Quintile 5	Quintile 1	Quintile 3	Quintile 5
Healthy diet score (average), mean ± SD	18.5 ± 1.7	24.2 ± 0.6	30.4 ± 2.1	18.4 ± 2.0	24.8 ± 0.8	31.3 ± 2.0
Healthy diet score (most recent), mean ± SD	17.8 ± 2.6	23.9 ± 2.5	30.9 ± 3.1	18.6 ± 3.1	24.4 ± 2.5	31.2 ± 2.7
Age at stool sample collection, mean ± SD, y	70.0 ± 8.8	69.1 ± 8.6	69.2 ± 8.8	66.3 ± 8.4	66.2 ± 9.1	69.0 ± 9.5
High income, %	3.3	5.6	10.6	1.5	9.8	14.4
Smoking status, %
Never	99.4	98.9	98.3	22.0	31.4	46.8
Former	0.6	1.1	1.7	16.7	15.2	15.8
Current				61.3	53.4	37.4
Alcohol drinking status (yes), %	1.7	3.9	5.6	31.4	34.3	31.0
Leisure-time physical activity (yes), %	50.3	53.3	62.8	41.2	53.4	62.6
History of hypertension, %	37.0	30.0	38.9	36.8	37.3	41.4
BMI, mean ± SD, kg/m2	24.7 ± 3.9	24.0 ± 3.8	24.4 ± 3.9	24.2 ± 3.2	24.3 ± 3.2	24.2 ± 3.5
Dietary intakes, mean ± SD
Total energy, kcal/d	1619 ± 310	1630 ± 287	1581 ± 269	2060 ± 427	1976 ± 437	1892 ± 343
Fruit, g/d	119 ± 56	186 ± 78	271 ± 99	73 ± 60	160 ± 133	247 ± 142
Vegetables, g/d	214 ± 56	279 ± 76	382 ± 97	236 ± 88	342 ± 122	487 ± 176
Dairy products, g/d	69 ± 91	129 ± 120	208 ± 221	39 ± 79	101 ± 140	199 ± 195
Fish and seafood, g/d	27 ± 14	48 ± 25	68 ± 30	28 ± 16	50 ± 34	77 ± 47
Nuts and legumes (dry weight), g/d	16 ± 6	20 ± 7	27 ± 10	17 ± 8	24 ± 11	31 ± 13
Refined grains, g/d	318 ± 24	284 ± 28	248 ± 31	409 ± 38	364 ± 39	325 ± 44
Red meat, g/d	47 ± 18	45 ± 20	35 ± 15	64 ± 31	65 ± 35	49 ± 26
Processed meat, times/wk	0.7 ± 1.0	0.4 ± 0.6	0.3 ± 0.6	0.5 ± 0.5	0.5 ± 0.8	0.3 ± 0.4

¹Healthy diet score was calculated based on intakes of 8 food groups listed in the table, assessed via repeated semiquantitative FFQs at cohort baseline and follow-ups. Dietary intakes presented were cumulative average intakes, except for processed meat, which was assessed only at baseline. Other characteristic variables presented were assessed at the last follow-up, except for income, which was assessed only at baseline. High income was defined as family income ≥30,000 yuan/y in the Shanghai Women's Health Study (baseline 1996–2000) and personal income ≥24,000 yuan/y in the Shanghai Men's Health Study (baseline 2002–2006). Continuous and categorical variables were compared by ANOVAs and χ² tests, respectively.

The median sequencing depth of our study samples was 132,602 reads (minimum: 17,013; maximum: 246,041). Higher HDS, both the cumulative average and the most recent scores, was associated with higher microbiome α-diversity, as evaluated by the Shannon index (Table 2 and Figure 1A). Each SD increment in HDS was associated with a 0.029 increase in the Shannon index (∼1%; P = 0.03), and the highest compared with the lowest quintile of HDS was associated with a 0.093 increase in the Shannon index (∼3%; P = 0.03). None of the food groups showed significant associations with the Shannon index. We observed significant associations for age, sex, BMI, and hypertension status with the Bray–Curtis distance (all P < 0.01; data not shown); however, with adjustment for those covariates, no associations were found for HDS or food groups with the Bray–Curtis distance, except for dairy intake (P = 0.05; data not shown).

TABLE 2.

Associations of long-term diet quality and intakes of major food groups with gut microbiome α-diversity in the Shanghai Women's and Men's Health Studies¹

	Linear regression coefficient (SE) of dietary variables with the Shannon diversity index
Dietary variables	Quintile 2	Quintile 3	Quintile 4	Quintile 5	1-SD increment	P value
Healthy diet score (average)	0.068 (0.042)	0.044 (0.042)	0.063 (0.043)	0.093 (0.043)2	0.029 (0.014)	0.03
Healthy diet score (most recent)	0.057 (0.042)	0.020 (0.043)	0.064 (0.043)	0.074 (0.043)	0.035 (0.014)	0.01
Fruit	0.042 (0.043)	0.006 (0.043)	0.082 (0.043)	0.043 (0.044)	0.0005 (0.014)	0.97
Vegetables	0.050 (0.042)	0.069 (0.042)	0.063 (0.043)	0.034 (0.043)	0.008 (0.014)	0.55
Dairy products	−0.035 (0.042)	−0.003 (0.043)	0.030 (0.043)	0.024 (0.045)	0.013 (0.014)	0.36
Fish and seafood	0.045 (0.042)	0.098 (0.042)2	0.111 (0.043)2	0.103 (0.043)2	0.022 (0.014)	0.11
Nuts and legumes	0.004 (0.042)	0.012 (0.042)	0.052 (0.043)	−0.030 (0.043)	−0.021 (0.014)	0.12
Refined grains	0.071 (0.042)	0.035 (0.042)	0.007 (0.043)	−0.014 (0.043)	−0.003 (0.014)	0.82
Red meat	0.0001 (0.042)	0.056 (0.042)	0.067 (0.043)	0.006 (0.043)	0.005 (0.014)	0.73
Processed meat	−0.004 (0.038)	−0.064 (0.038)	−0.120 (0.042)2	−0.040 (0.052)	−0.024 (0.022)	0.28

¹Dietary variables were modeled as sex-specific quintiles with the lowest quintile as the referent group and continuous variables as per SD increase. The general linear model was adjusted for age at stool sample collection, sex, income, smoking status, alcohol drinking status, leisure-time physical activity, total energy intake, BMI, history of hypertension, and sequencing depth. P values for the associations between a 1-SD increase in dietary variables and the Shannon index are shown.

² P < 0.05 for each quintile compared with the lowest quintile.

FIGURE 1.

The Shannon α-diversity index and relative abundance of major diet-related gut microbiome taxa by long-term diet quality and food intakes in the Shanghai Women's and Men's Health Studies (n = 1920). Geometric mean (95% CI) of the Shannon index and microbial taxa relative abundance, adjusting for age at stool sample collection, sex, income, smoking status, alcohol drinking status, leisure-time physical activity, total energy intake, BMI, and history of hypertension. Sequencing depth was additionally adjusted for Shannon index analysis. Healthy diet score was calculated based on intakes of 8 food groups (fruit, vegetables, dairy, fish/seafood, nuts/legumes, refined grains, red meat, and processed meat). Healthy diet score and dairy intake were cumulative averages calculated using repeated dietary survey data at cohort baseline and follow-ups, while intake frequency of processed meat was assessed only at baseline and categorized into 5 groups: less than once per 2 mo, less than once per month, less than twice per month, less than once per week, and once or more per week.

Among 149 common taxa (5 phyla, 10 classes, 12 orders, 21 families, 39 genera, and 62 species), higher cumulative HDS was associated with increased abundance of phyla Firmicutes and Actinobacteria, and, particularly, 2 genera within them: Coprococcus and Bifidobacterium (Table 3; all FDRs <0.1). The adjusted mean relative abundance of Coprococcus was 0.45% compared with 0.36% in the highest compared with the lowest quintile of cumulative HDS (Figure 1B; coefficient = 0.297, P = 0.002), and the adjusted mean relative abundance of Bifidobacterium was 0.33% compared with 0.23% (Figure 1C; coefficient = 0.411, P = 0.009). When HDS from the most recent FFQs was evaluated, similar associations were found with Coprococcus and Bifidobacterium (species adolescentis), although slightly weaker than those for cumulative HDS. However, recent HDS showed additional significant results, including positive associations with an unknown species of family Ruminococcaceae, Blautia obeum, Faecalibacterium (prausnitzii), Clostridium (unknown), Butyricimonas, and Bacteroides uniformis. The adjusted mean relative abundance of Faecalibacterium was 2.42% compared with 1.88% in the highest compared with the lowest quintile of HDS (Figure 1D; coefficient = 0.313, P = 0.01).

TABLE 3.

Associations of long-term diet quality and food intakes with common gut microbiome taxa in the Shanghai Women's and Men's Health Studies

Phylum	Class	Order	Family	Genus species	Median relative abundance, %	Prevalence, %	β (SE) for Q5 vs. Q11	P value	β (SE) for 1-SD increase1	P value
Healthy diet score (average)
Firmicutes					28.3	100	0.095 (0.039)	0.01	0.027 (0.012)	0.03
Firmicutes	Clostridia				27.1	100	0.108 (0.040)	0.007	0.030 (0.013)	0.02
Firmicutes	Clostridia	Clostridiales	Lachnospiraceae	Coprococcus	0.63	99.1	0.297 (0.095)	0.002	0.100 (0.030)	0.001
Actinobacteria					0.47	99.8	0.225 (0.114)	0.04	0.081 (0.037)	0.03
Actinobacteria	Actinobacteria				0.33	97.7	0.351 (0.142)	0.01	0.123 (0.045)	0.007
Actinobacteria	Actinobacteria	Bifidobacteriales			0.30	92.3	0.394 (0.158)	0.01	0.139 (0.050)	0.006
Actinobacteria	Actinobacteria	Bifidobacteriales	Bifidobacteriaceae	Bifidobacterium	0.30	92.1	0.411 (0.158)	0.009	0.149 (0.050)	0.003
Healthy diet score (most recent)
Firmicutes	Clostridia	Clostridiales	Lachnospiraceae	Blautia obeum	0.07	88.2	0.256 (0.104)	0.01	0.090 (0.033)	0.007
Firmicutes	Clostridia	Clostridiales	Lachnospiraceae	Coprococcus	0.63	99.1	0.229 (0.094)	0.01	0.087 (0.030)	0.004
Firmicutes	Clostridia	Clostridiales	Ruminococcaceae	Faecalibacterium	5.35	98.7	0.313 (0.126)	0.01	0.105 (0.040)	0.009
Firmicutes	Clostridia	Clostridiales	Ruminococcaceae	Faecalibacterium prausnitzii	5.17	98.7	0.313 (0.126)	0.01	0.104 (0.040)	0.009
Firmicutes	Clostridia	Clostridiales	Ruminococcaceae	Unknown	0.43	98.9	0.215 (0.122)	0.08	0.091 (0.031)	0.004
Actinobacteria	Actinobacteria	Bifidobacteriales	Bifidobacteriaceae	Bifidobacterium adolescentis	0.05	76.6	0.297 (0.145)	0.04	0.120 (0.046)	0.01
Bacteroidetes	Bacteroidia	Bacteroidales	Odoribacteraceae	Butyricimonas	0.08	65.2	0.246 (0.149)	0.09	0.122 (0.047)	0.01
Bacteroidetes	Bacteroidia	Bacteroidales	Bacteroidaceae	Bacteroides uniformis	1.85	98.1	0.257 (0.137)	0.06	0.110 (0.044)	0.01
Dairy2
Actinobacteria					0.47	99.8	0.465 (0.120)	0.0001	0.120 (0.038)	0.002
Actinobacteria	Actinobacteria				0.33	97.7	0.557 (0.149)	0.0002	0.136 (0.048)	0.004
Actinobacteria	Actinobacteria	Bifidobacteriales			0.30	92.3	0.581 (0.165)	0.0004	0.141 (0.053)	0.008
Actinobacteria	Actinobacteria	Bifidobacteriales	Bifidobacteriaceae		0.30	92.3	0.581 (0.165)	0.0004	0.141 (0.053)	0.08
Actinobacteria	Actinobacteria	Bifidobacteriales	Bifidobacteriaceae	Bifidobacterium	0.30	92.1	0.559 (0.165)	0.0007	0.129 (0.053)	0.01
Fish and seafood2
Firmicutes	Clostridia	Clostridiales	Lachnospiraceae	Coprococcus	0.63	99.1	0.269 (0.094)	0.004	0.081 (0.030)	0.007
Firmicutes	Clostridia	Clostridiales	Ruminococcaceae	Unknown	0.43	98.9	0.296 (0.097)	0.002	0.070 (0.031)	0.02
Proteobacteria	Betaproteobacteria				2.81	96.9	0.255 (0.127)	0.04	0.106 (0.041)	0.009
Proteobacteria	Betaproteobacteria	Burkholderiales			2.80	96.5	0.292 (0.131)	0.02	0.114 (0.042)	0.007
Proteobacteria	Betaproteobacteria	Burkholderiales	Alcaligenaceae		2.78	95.7	0.324 (0.135)	0.02	0.115 (0.043)	0.008
Proteobacteria	Deltaproteobacteria				0.23	88.9	0.382 (0.124)	0.002	0.127 (0.040)	0.001
Proteobacteria	Deltaproteobacteria	Desulfovibrionales			0.23	88.8	0.385 (0.124)	0.002	0.125 (0.040)	0.002
Proteobacteria	Deltaproteobacteria	Desulfovibrionales	Desulfovibrionaceae		0.23	88.8	0.387 (0.123)	0.002	0.124 (0.040)	0.002
Nuts and legumes2
Proteobacteria					6.18	100	0.110 (0.061)	0.07	0.046 (0.019)	0.02
Processed meat2
Firmicutes	Clostridia				27.1	100	−0.127 (0.048)	0.008	−0.035 (0.020)	0.08
Firmicutes	Clostridia	Clostridiales			27.1	100	−0.127 (0.048)	0.008	−0.035 (0.020)	0.08
Firmicutes	Clostridia	Clostridiales	Lachnospiraceae		9.87	100	−0.171 (0.061)	0.005	−0.055 (0.026)	0.03
Firmicutes	Clostridia	Clostridiales	Lachnospiraceae	Roseburia	2.36	97.7	−0.522 (0.151)	0.0006	−0.142 (0.064)	0.03
Firmicutes	Clostridia	Clostridiales	Lachnospiraceae	Roseburia faecis	1.14	95.4	−0.569 (0.167)	0.0007	−0.175 (0.071)	0.01
Firmicutes	Clostridia	Clostridiales	Lachnospiraceae	Roseburia other	0.25	92.5	−0.440 (0.141)	0.002	−0.145 (0.060)	0.01

¹General linear regression β coefficient (SE) for the fifth compared with first quintile (Q5 compared with Q1) and each SD increment of the healthy diet score and food intakes. The model was adjusted for age at stool sample collection, sex, income, smoking status, alcohol drinking status, leisure-time physical activity, total energy intake, BMI, and history of hypertension. Results with a false discovery rate <0.10 are shown. Tests were conducted for common taxa, including 5 phyla, 10 classes, 12 orders, 21 families, 39 genera, and 62 species.

²Intakes of 7 food groups (fruit, vegetables, dairy, fish/seafood, nuts/legumes, refined grains, and red meat) were cumulative average intakes. The intake frequency of processed meat was assessed only at baseline.

Among food groups, we found significant positive associations for dairy intake with family Bifidobacteriaceae and genus Bifidobacterium, fish/seafood intake with families Alcaligenaceae and Desulfovibrionaceae, and nuts/legumes intake with phylum Proteobacteria, while inverse associations included processed meat with family Lachnospiraceae and genus Roseburia (Table 3). The adjusted mean relative abundance of Bifidobacterium was 0.36% compared with 0.19% in the highest compared with the lowest quintile of dairy intake (Figure 1E; coefficient = 0.559, P = 0.0007). The adjusted mean relative abundance of Roseburia was 0.76% compared with 0.33% in the highest compared with the lowest quintile of processed meat intake (Figure 1F; coefficient = −0.522, P = 0.0006). No common microbiome taxa were significantly related to intakes of fruit, vegetables, refined grains (mostly white rice), or red meat in our analyses.

Among 147 rare taxa (6 phyla, 7 classes, 8 orders, 21 families, 43 genera, and 62 species), HDS was associated with increased abundances of phylum Tenericutes, class Mollicutes, and order RF39 (Table 4). The abundance of phylum Tenericutes was also positively associated with nuts/legumes intake but inversely associated with refined grains intake. We also observed significant associations for processed meat with increased abundances of Fusobacteriaceae and Acinetobacter from phylum Proteobacteria.

TABLE 4.

Associations of long-term diet quality and food intakes with rare gut microbiome taxa in the Shanghai Women's and Men's Health Studies

Phylum	Class	Order	Family	Genus species	Prevalence, %	β (SE) for Q5 vs. Q11	P value	β (SE) for 1-SD increase1	P value
Healthy diet score (average)
Tenericutes					25.9	0.371 (0.145)	0.01	0.101 (0.046)	0.03
Tenericutes	Mollicutes				24.2	0.362 (0.146)	0.01	0.088 (0.046)	0.06
Tenericutes	Mollicutes	RF39			23.9	0.387 (0.147)	0.008	0.091 (0.046)	0.05
Healthy diet score (most recent)
Tenericutes					25.9	0.330 (0.145)	0.02	0.133 (0.046)	0.004
Tenericutes	Mollicutes				24.2	0.362 (0.146)	0.01	0.113 (0.046)	0.02
Tenericutes	Mollicutes	RF39			23.9	0.387 (0.147)	0.009	0.118 (0.046)	0.01
Fusobacteria	Fusobacteriia	Fusobacteriales	Fusobacteriaceae	Unknown	44.0	−0.561 (0.168)	0.001	−0.139 (0.052)	0.008
Firmicutes	Clostridia	Clostridiales	Lachnospiraceae	Coprococcus other	18.7	0.496 (0.167)	0.003	0.133 (0.052)	0.01
Nuts and legumes2
Tenericutes					25.9	0.408 (0.145)	0.005	0.098 (0.046)	0.03
Tenericutes	Mollicutes				24.2	0.369 (0.146)	0.01	0.089 (0.046)	0.05
Tenericutes	Mollicutes	RF39			23.9	0.381 (0.146)	0.009	0.092 (0.046)	0.05
Refined grains2
Tenericutes					25.9	−0.271 (0.146)	0.06	−0.114 (0.046)	0.01
Proteobacteria	Alphaproteobacteria	RF32			13.3	−0.438 (0.171)	0.010	−0.132 (0.052)	0.01
Processed meat2
Proteobacteria	Fusobacteriia	Fusobacteriales	Fusobacteriaceae	Unknown	44.0	0.677 (0.221)	0.002	0.228 (0.099)	0.02
Proteobacteria	Gammaproteobacteria	Pseudomonadales			23.3	0.578 (0.211)	0.006	0.205 (0.092)	0.03
Proteobacteria	Gammaproteobacteria	Pseudomonadales	Moraxellaceae		21.1	0.595 (0.205)	0.004	0.219 (0.090)	0.02
Proteobacteria	Gammaproteobacteria	Pseudomonadales	Moraxellaceae	Acinetobacter	17.5	0.689 (0.197)	0.0005	0.277 (0.088)	0.002

¹Negative binomial hurdle model β coefficient (SE) for the fifth compared with first quintile (Q5 compared with Q1) and each SD increment of the healthy diet score and food intakes. The model was adjusted for age at stool sample collection, sex, income, smoking status, alcohol drinking status, leisure-time physical activity, total energy intake, BMI, and history of hypertension. Results with a false discovery rate <0.10 are shown. Tests were conducted for rare taxa, including 6 phyla, 7 classes, 8 orders, 21 families, 43 genera, and 62 species.

²Intakes of 7 food groups (fruit, vegetables, dairy, fish/seafood, nuts/legumes, refined grains, and red meat) were cumulative average intakes. The intake frequency of processed meat was only assessed at baseline.

No significant interactions were found between dietary variables and covariates, except that the associations for long-term HDS with Firmicutes and Actinobacteria seemed stronger among participants with BMI ≥25 than those with normal body weights. The respective coefficients for a 1-SD increase in HDS with Firmicutes and Actinobacteria were 0.049 and 0.150 in overweight/obese participants compared with 0.018 and 0.043 in normal-weight participants (both P-interaction = 0.02). Exclusion of participants who used antibiotics in the past 6 mo did not change the observed associations; many became slightly stronger (Supplemental Table 1). Results without adjustment for BMI or hypertension status were similar to the main results (Supplemental Tables 2–4).

Discussion

In this prospective investigation among 1920 apparently healthy Chinese men and women, we evaluated long-term diet quality and intakes of major food groups with the gut microbiome. We found significant associations for higher diet quality with increased α-diversity and abundance of certain taxa, concentrated in 2 groups: 1) from families Lachnospiraceae and Ruminococcaceae within order Clostridiales of phylum Firmicutes, such as Coprococcus, Blautia obeum, Faecalibacterium prausnitzii, and Oscillospira, which are all fiber-fermenting, butyrate-producing bacteria, and 2) Bifidobacterium (Bifidobacterium adolescentis) from phylum Actinobacteria, which are also fiber-fermenting bacteria that can produce SCFAs (mainly acetate and propionate) and also lactic acid, B vitamins, and neurotransmitters. Analyses of food groups further suggest that a high intake of fish/seafood, but a low intake of processed meat, was associated with increased abundances of taxa from group 1 families, including Coprococcus and Roseburia (Roseburia faecis), and high dairy intake was associated with increased abundance of Bifidobacterium. These observed diet–microbiome associations were independent of covariates, including sociodemographic and lifestyle factors, BMI, hypertension, and recent antibiotic use, as the results were similar after adjustment for these factors and in a series of interaction/sensitivity analyses by these factors. However, the magnitudes of the associations were generally small.

To our knowledge, this is the first large-scale, prospective study evaluating the influence of long-term diet quality/patterns and food intakes on the gut microbiome among Chinese adults. Although dietary patterns and microbiome profiles of Chinese could be very different from those of Westerners (19, 25, 33–35), our major findings agree well with recent findings from European and US populations. In the TwinsUK study (2070 British adults; 99% white) and Multiethnic Cohort study (1735 American adults; 5 racial/ethnic groups: white, black, Latino, Asian, and Native Hawaiian) (15, 16), multiple diet quality indices, including HEI and Mediterranean Diet score, were all positively associated with Shannon α-diversity index, and their primary findings were all related to fiber-fermenting, butyrate-producing bacteria from family Lachnospiraceae or Ruminococcaceae, including Coprococcus, Faecalibacterium, Oscillospira, and Ruminococcus. Similarly, in the Malmö Offspring Study (1726 Swedish adults) and Osteoporotic Fractures in Men Study (517 senior American men; 88% white) (18, 36), factor analysis–derived prudent and health-conscious dietary patterns were associated with higher abundances of Faecalibacterium, Roseburia, Lachnospira, and Clostridium, although those 2 studies found no association between diet patterns and the Shannon index and unexpected lower abundances of some other butyrate-producing genera (e.g., Eubacterium, Blautia, Dorea, and Ruminococcus). We noted that the effect size of diet quality score on the Shannon index was small (∼3% across quintiles), which is similar to results from the Multiethnic Cohort study (∼2% across tertiles) and in line with a recent review, suggesting a limited effect of dietary interventions on gut microbiome α-diversity (16, 37). Nevertheless, findings regarding butyrate-producing taxa have been largely consistent from populations with different ethnicities, geographic locations, and habitual diets and studies using different diet quality assessments, providing strong observational evidence that healthy diets may promote butyrate-producing bacteria. These findings are further supported by emerging evidence from long-term, whole-diet interventions that evaluated gut microbial changes. For example, 2 recent studies, including a large-scale, multicenter study (612 participants in 5 European countries), found that a 12-mo Mediterranean diet intervention significantly increased butyrate-producing taxa, including Faecalibacterium (Faecalibacterium prausnitzii), Roseburia, Clostridium, Eubacterium, Blautia, and Oscillospira (38, 39). Also, a 6- or 12-mo intervention with low-fat, high–complex carbohydrate diets increased F. prausnitzii and Bifidobacterium (39, 40).

A higher abundance of Bifidobacterium (B. adolescentis) along with a higher diet quality is another main finding from our study. Bifidobacterium is another group of fiber-fermenting bacteria (41). In a recent meta-analysis of 64 randomized trials involving over 2000 participants, Bifidobacterium species were the most significantly and consistently increased taxa after dietary fiber intervention (42). They can generate not only SCFAs (mainly acetate and propionate) but also lactic acid, B vitamins (e.g., folate), and neurotransmitters (e.g., γ-aminobutyric acid) (41, 43, 44). Given these features, Bifidobacterium strains have long been considered probiotic bacteria, and they are added to yogurt or other fermented foods or used together with prebiotic fibers to promote human health (45, 46). However, none of the aforementioned studies among European or US populations reported significant associations of diet quality/patterns with Bifidobacterium, suggesting potential population variations. In our analyses of food groups, dairy showed a strong positive association with the abundance of Bifidobacterium. Our study population had generally low but wide-ranging dairy intakes, and almost all were nonfermented products, whereas Western populations have a higher and more common dairy consumption, including many fermented products such as yogurt and cheese, which contain Bifidobacterium strains. The differences in the amounts and types of dairy consumption may explain the population-specific findings regarding Bifidobacterium, which warrant further investigations.

Other significant findings of our study include positive associations of diet quality score with the abundance of a rare phylum Tenericutes and its members—class Mollicutes, order RF39. The TwinsUK study also found positive associations of the HEI with phylum Tenericutes and its members within orders Anaeroplasmatales and RF39 (15); the last finding has been replicated in the Malmö Offspring Study, showing increased abundance of a genus within order RF39, related to a healthy dietary pattern (18). We also found novel associations of phylum Tenericutes/order RF39 with a high intake of nuts/legumes and a low intake of refined grains. Evidence on specific foods/nutrients affecting Tenericutes/RF39 remains limited, although some studies have suggested positive associations of Tenericutes/RF39 with olive oil intake and blood concentrations of polyunsaturated fatty acids but negative associations with blood concentrations of saturated fatty acids (47, 48). In terms of their potential health effects, Tenericutes/RF39 have been associated with lower BMI, lower circulating triglycerides, and/or less frailty among older adults in several studies, including the TwinsUK study, LifeLines–DEEP cohort, and Metabolic Syndrome in Men study (47, 49–51). As a unique rare phylum distinguished by the absence of a bacterial cell wall, how Tenericutes may be modulated by dietary factors and whether they play a role in human health are worthy of further research.

Our previous work in the SWHS and SMHS reported that high diet quality, especially maintained in the long term, was associated with a 15–25% reduced risk of type 2 diabetes and 17–52% reduced risk of death from cardiovascular disease, cancer, or diabetes (24, 25). Our current study adds a new piece of information that long-term healthy eating also modulates the gut microbiota, particularly fiber-fermenting, SCFA-producing bacteria. However, to minimize the influence of diseases/treatments on the diet–microbiome associations, we excluded participants with existing cardiovascular disease, cancer, or diabetes from this analysis, and no cross-sectional biomarker data or incident disease data are available in the current study. Thus, we could not evaluate through which mechanisms, or to what extent, gut microbiota may mediate the diet–disease associations. Nonetheless, there has been a substantial amount of evidence from animal studies and increasing evidence from human studies showing the role of gut microbiota in human health. SCFA-producing bacteria, such as Coprococcus, Faecalibacterium, Roseburia, and Bifidobacterium, have been shown to have anti-inflammatory, antidiabetic, and antihypertensive properties and linked to better insulin sensitivity and blood lipids and reduced systemic and local inflammation (e.g., of the gastrointestinal tract and adipose tissue) (52–54). F. prausnitzii has been considered a next-generation probiotic, adding to existing probiotics Lactobacillus and Bifidobacterium (55). Still, well-powered, prospective population studies are needed to examine whether these microbes contribute to the incidence of major chronic diseases and whether these microbes may interact with habitual diets to affect disease risk.

This study has several notable strengths, including a large sample size within 2 well-executed longitudinal cohorts, being the first study, to our knowledge, on overall diet quality and gut microbiome among Chinese adults; the robustness of results regardless of covariate adjustments and sensitivity analyses; and the consistency of our findings with existing literature. Meanwhile, we acknowledge a few limitations of this study. First, like all other large observational studies, measurement errors exist in the assessments of diets and gut microbiome. Repeated surveys would have overcome some of the errors, as we observed stronger associations with cumulative HDS than the most recent HDS for our main findings (e.g., Coprococcus and Bifidobacterium). Also, our primary exposure is overall diet quality, evaluated by HDS incorporating 8 food groups with a defined range, which reflected multiple aspects of participants’ diets and was not influenced by extreme values. Unfortunately, we had only a one-time measurement of the gut microbiome, although microbiota has been considered largely stable during adulthood and resilient to temporary disturbances (9). As discussed previously, we have taken into account health status, recent use of antibiotics, and sociodemographic and lifestyle factors to minimize their influence on the diet–microbiome associations. Second, despite multivariable adjustments, we cannot rule out residual confounding due to imperfectly measured covariates and unmeasured confounders. However, our results were similar, with and without adjustment for or stratification by covariates. Third, the current study included older urban Chinese adults (mean age = 55 y at baseline dietary survey and 68 y at stool collection); thus, the results may not be directly applicable to younger or rural Chinese populations, who may have different dietary patterns and/or microbial profiles. However, our major findings are consistent with the literature, regardless of the study population, suggesting its generalizability.

In summary, in a large sample of generally healthy Chinese men and women, we found that a long-term healthy diet was associated with a slightly more diverse gut microbiome and higher abundances of fiber-fermenting, SCFA-producing bacteria, including members within Lachnospiraceae, Ruminococcaceae, and Bifidobacteriaceae families, such as genera Coprococcus, Faecalibacterium, and Bifidobacterium. Future studies are needed to investigate whether and to what extent these commensal bacteria may mediate or modify the effects of habitual diets on chronic human diseases.

Notes

Funding: The Shanghai Women's Health Study is funded by UM1 CA182910 to WZ, and the Shanghai Men's Health Study is funded by UM1 CA173640 to XOS from the National Cancer Institute (NCI) at the National Institutes of Health (NIH). DY was supported by Vanderbilt University Medical Center Faculty Research Scholars Program. SMN was supported by a Vanderbilt-Emory-Cornell-Duke Global Health Fellowship, funded by the NCI and Fogarty International Center of the NIH (D43 TW009337).

Supplemental Tables 1–4 and Supplemental Figure 1 are available from the “Supplementary data” link in the online posting of the article and from the same link in the online table of contents at https://academic.oup.com/ajcn/.

Abbreviations used: FDR, false discovery rate; HDS, healthy diet score; HEI, Healthy Eating Index; OTU, operational taxonomic unit; QIIME, Quantitative Insights Into Microbial Ecology; SMHS, Shanghai Men's Health Study; SWHS, Shanghai Women's Health Study.

Data Availability

Data described in the manuscript, code book, and analytic code will be made available upon research study application and approval by the cohort committees.