Bacterial Extracellular Vesicle Composition in Human Urine and the 10-Year Risk of Abdominal Obesity

Abstract

Objective:

We aimed to evaluate a causal relationship between commensal bacteria and abdominal obesity.

Methods:

A prospective study, including 2222 adults who provided urine samples at baseline, was performed. These samples were used for assays of genomic DNA from bacterial extracellular vesicles (EVs). During the 10-year period, the incidence rates of obesity (measured as body mass index) and abdominal obesity (measured as waist circumference) were ascertained as outcomes. To evaluate associations of bacterial composition at the phylum and genus levels with the outcomes, the hazard ratio (HR) and its confidence interval (95% CI) were estimated.

Results:

No significant association was observed for the risk of obesity, whereas the risk of abdominal obesity was inversely associated with the composition of Proteobacteria and positively associated with that of Firmicutes (adjusted P value <0.05). In joint analysis for the combination groups of Proteobacteria and Firmicutes composition tertiles, the group with top tertiles of both Proteobacteria and Firmicutes showed a significant HR of 2.59 (95% CI: 1.33 − 5.01) compared with the reference with lower tertiles (adjusted P value <0.05). Some genera of these phyla were associated with the risk of abdominal obesity.

Conclusions:

These findings suggest that bacterial composition in urinary EV samples can predict the 10-year risk of abdominal obesity.

Introduction

Bacteria produce extracellular vesicles (EVs), which are nanometer-sized spherical particles measuring <300 nm in diameter. In particular, EVs produced by gram-negative bacteria are called outer membrane vesicles (OMVs) because they are derived from the outer membrane, whereas those from gram-positive bacteria are generally membrane vesicles. These bacterial EVs are known to contain genetic materials, as well as proteins and lipids, and thus genome assays can identify their origin, such as bacterial species, and quantify species-specific EVs.^1,2

Recently, a small number of studies sequenced DNA from bacterial EVs in animal stool or human urine samples.^3,4 In addition, some studies sequenced extracellular DNA or RNA from bacteria, presumably from bacterial EVs, in human blood.^5
–7 Although it is still controversial whether bacterial EVs are mainly from gut microbiota, given the leaky gut hypothesis, it is possible as bacteria or bacterial substances pass from the gastrointestinal tract into circulation, contributing to health-related outcomes.⁸

In fact, some experimental studies showed the effects of bacterial EVs on host responses and outcomes, such as glucose metabolism, inflammatory response, and colitis progression and regression.^3,9,10 However, data are limited on the association between bacterial EVs assessed in human biospecimens and the health-related outcomes.

There are accumulating data on the association between gut microbiota profiles and obesity.^11

–16 These data are from cross-sectional studies; however, it has not yet been established whether gut microbiota play a role in development of obesity in humans.

One epidemiologic study that included a large sample examined longitudinal weight changes over ∼9 years and evaluated microbiota profiles in fecal samples collected at follow-up.¹⁷ This study seemed to have limited ability to establish a temporal relationship between gut microbiota and weight changes, but did confirm the reverse relationship; it found significant effects of long-term weight changes on gut microbiota diversity or composition.

A number of clinical trials tested the effects of weight loss on microbiota profiles or the effects of probiotics and prebiotics, which are expected to change gut microbiota composition, on weight changes.^18,19 However, their findings were based on short-term effects and are inconsistent partly due to the differences in study populations, sample sizes, intervention durations, or methodologies (weight loss methods or types of probiotic strains used).

In this epidemiologic study, we aimed to evaluate whether the assessment of bacterial EVs in urine can provide information to predict longitudinal changes in body weight (BW) or waist circumference (WC). Furthermore, we aimed to explore microbial signatures at the phylum and genus levels in a temporal relationship with the risk of abdominal obesity.

Materials and Methods

Study design and population

We conducted a prospective study in a population-based cohort of the currently ongoing Korean Genome and Epidemiology Study. Detailed information regarding enrollment of cohort members and study procedures is available elsewhere.^20,21 Briefly, ∼5000 adults aged 40–69 years and living in Ansan, Republic of Korea, were enrolled between June 18, 2001, and January 29, 2003. They visited the Korea University Ansan Hospital to participate in baseline and biennial follow-up interviews and health examinations, which include anthropometric measurement, collection of blood and urine specimens, and clinical evaluation.

Trained personnel administered a questionnaire-based interview to collect information on sociodemographics; medical history; health conditions; and lifestyle factors, including smoking, alcohol consumption, physical activity, and diet. At each visit, participants signed an informed consent form approved by the Human Subjects Review Committee at the Korea University Ansan Hospital (ED0624).

Among 4973 urine samples initially collected at baseline, a part of the samples (n = 3878) were selected due to budget constraints for bacterial EV composition assays by a researcher who was blinded to outcome, exposure, and confounding variables. For statistical analysis, individuals who reported pregnancy (n = 1) or a diagnosis of cancer, renal disease, or urinary tract infection (n = 194) were excluded to minimize potential pathophysiologic alterations in urinary levels of bacterial EVs.

In addition, 1461 samples were further excluded due to missing anthropometric values (n = 10 at baseline and n = 1451 at the 10-year follow-up). After this exclusion, 2222 samples were included in the present study. We observed no significant differences in baseline BW and WC, follow-up BW and WC, and major phylum profiles between the study participants and cohort members not included in the present study.

Urinary EV and metagenomic analyses

The midstream urine samples were collected from participants (collections were avoided during menstruation) in the morning after fasting for at least 8 hr and immediately stored in a deep freezer (−80℃) at baseline. These frozen samples were recently transported to a commercial laboratory (Institute of MD Healthcare, Inc., Republic of Korea) for bacterial EV assays.

Urinary EV and metagenomic analyses were conducted according to the standard protocol of the laboratory. First, EVs in the urine samples were isolated using differential centrifugation at 10,000 g for 10 min at 4℃. After centrifugation, bacteria and foreign particles were thoroughly eliminated by sterilizing the supernatant through a 0.22-μm filter. The supernatant was boiled for 40 min at 100℃ to extract DNA from the EV membrane and then it was centrifuged at 13,000 rpm for 30 min at 4℃ to eliminate the remaining floating particles and waste.

The EV DNA was extracted using a DNA isolation kit (PowerSoil DNA Isolation Kit; MO BIO) and quantified using the QIAxpert system (QIAGEN, Germany). Bacterial genomic DNA was amplified using 16S_V3_F (5′-TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGCCTACGGGNGGCWGCAG-3′) and 16S_V4_R (5′-GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGACTACHV GGGTATCTAATCC-3′) primers, which are specific for V3-V4 hypervariable regions of the 16S rDNA gene.

Libraries were prepared using PCR products according to the manual of the MiSeq System (Illumina) and quantified using QIAxpert (QIAGEN, Germany). Each amplicon was then quantified, set up in an equimolar ratio, pooled, and sequenced on MiSeq. Raw pyrosequencing reads obtained from the sequencer were filtered according to the barcode and primer sequences.

Taxonomic assignment was performed by the profiling program, MDx-Pro, ver.1 (MD Healthcare, Inc., Republic of Korea). Using this program, high-quality sequencing reads were selected after checking the read length (≥300 bp) and quality score (average Phred score ≥20). Operational taxonomic units (OTUs) were clustered using the sequence clustering algorithm, CD-HIT.

Subsequently, taxonomy assignment was carried out using UCLUST and QIIME against the 16S rDNA sequence database in Greengenes 8.15.13. Based on sequence similarities, all 16S rDNA sequences were assigned to taxonomic levels and then bacterial composition at each level was calculated. For clusters that could not be assigned at the genus level due to the lack of sequences or redundant sequences in the database, the taxon was assigned to a higher level and indicated in parentheses.

In this study, we used bacterial composition data constructed at phylum and genus levels. Particularly, observed OTU counts and the Shannon diversity index were calculated using genus-level data.

Anthropometric measurements and outcomes

Trained researchers performed anthropometric measurements using a standardized protocol. Height (cm) and BW (kg) were measured to the nearest 0.1 cm or 0.1 kg without shoes, and body mass index (BMI, kg/m²) was calculated. WC (cm) was measured using a flexible measuring tape at the narrowest part between the lower rib and the iliac crest, to the nearest 0.1 cm, and the average of three repeated measurements was calculated.

Abdominal obesity was defined as WC ≥90 cm for men and ≥80 cm for women, and obesity was defined as BMI ≥27.5 kg/m², as suggested by the World Health Organization for Asian populations.^22,23 For outcomes, we calculated longitudinal changes in BW and WC between baseline and the fifth biennial follow-up examination as well as incident cases of obesity and abdominal obesity, which were ascertained as outcomes at each biennial follow-up examination over 10 years.

Potential confounding variables

Potential confounding variables included age; sex; smoking status; alcohol consumption; physical activity level; the FTO rs9939609 polymorphism; the presence of hypertension or diabetes mellitus (DM); and dietary factors such as calorie, fat, and fiber intake. The FTO SNP was found to be associated with abdominal obesity in our earlier study.²⁴

Lifestyle information and medication history were collected through questionnaire-based interviews. For smoking status, smoking history and the average number of cigarettes smoked per day were queried. For alcohol consumption, the average frequency of alcohol drinking occasions, the amount of alcohol typically consumed per occasion, and the volume of one standard drink for each alcoholic beverage were queried, and accordingly, the amount of alcohol consumed daily (g/day) was calculated.

For physical activity, hours typically spent daily on activities (classified into the following five intensity levels: sleep or sedentary, very light activity, light activity, moderate activity, and vigorous activity) were queried and then a total metabolic equivalent score (MET-h) was calculated. To obtain dietary information, a quantitative food frequency questionnaire developed and validated by the Korea Centers for Disease Control and Prevention (Seoul, Korea) was used. Genotyping of the FTO rs9939609 polymorphism has been done in a previous study,²⁵ which described detailed procedures regarding genomic sample preparation, genotyping methodology, and quality control information.

Briefly, DNA was extracted using whole blood samples and processed using the Affymetrix Genome-wide Human SNP Array 5.0 (Affymetrix, Inc., Santa Clara, CA). The genotyping call rates were greater than 95%, and genotype distribution of the rs9939609 alleles was in Hardy–Weinberg equilibrium (P = 0.31).

The presence of hypertension or DM was confirmed if the following criteria were met: for hypertension, use of antihypertensive medications or systolic blood pressure (BP) ≥140 mmHg or diastolic BP ≥90 mmHg; for DM, use of insulin or oral hypoglycemic agents or fasting plasma glucose level ≥126 mg/dL or postprandial glucose level ≥200 mg/dL.

Statistical analyses

Descriptive statistics for major phyla, including Proteobacteria, Firmicutes, Actinobacteria, Cyanobacteria, and Bacteroidetes; baseline and follow-up anthropometric measures; and baseline confounding variables were calculated according to the categories of WC changes over 10 years. The Cochran–Armitage trend test and the analysis of variance test with trend analysis were used where appropriate. Linear regression analysis was used to evaluate the association between the phylum groups and changes in BW or WC.

A coefficient estimate and its P value, as well as a P value for the trend, were obtained from the output of the analysis for interpreting the association. To analyze the association between the phylum groups and the risk of obesity or abdominal obesity, Cox proportional hazards regression analysis was used. For this analysis, person-years were calculated from the date when participants underwent the baseline examination to the date when they underwent the follow-up examination and had the onset of obesity or abdominal obesity.

The hazard ratio (HR) and its confidence interval (95% CI) were obtained to interpret the risk of obesity or abdominal obesity. In these linear and Cox regression models, covariates, including age (continuous); sex (binary); baseline BMI or WC (continuous); FTO rs9939609 polymorphism (TT vs. TA and AA); smoking status (never smokers, former smokers, or current smokers with 1–10, 11–20, or ≥21 cigarettes/day); alcohol consumption (nondrinkers or alcohol consumers with 1–15, 16–30, or >30 g/day); physical activity levels (quintiles of MET-h); total calorie, fat, and fiber intake (quintiles); and the presence of hypertension or DM at baseline (binary), were considered.

Because of these covariates, adjusted P values were obtained after multiple testing corrections using a false discovery rate. Composition values of major bacterial phyla were classified into tertiles to be fitted as categorical variables in the models. Furthermore, for a significant phylum group associated with the risk of obesity or abdominal obesity, common genera, whose mean composition was greater than 0.1%, were selected and fitted as continuous variables.

All statistical analyses were performed with the SAS, version 9.1.3, software (SAS Institute, Cary, NC).

Results

This prospective study included 1121 women (50.5%) and 1101 men (49.6%). Mean changes (range) in BW and WC were −0.74 kg (range: −17.80 kg to 17.0 kg) and 0.55 cm (range: −27.93 cm to 26.50 cm), respectively. Mean changes in BW were comparable for men and women, whereas the mean change in WC was greater in women (mean: 0.723) than in men (mean: 0.364). After excluding obesity cases confirmed at baseline, 162 incident cases (8.7%) of obesity were observed in 1857 participants. After excluding abdominal obesity cases at baseline, 468 incident cases (29.3%) of abdominal obesity were observed in 1596 participants. The follow-up duration used for analysis was 10.3 years.

Table 1 demonstrates descriptive statistics of microbiota profiles, including bacterial composition of major phyla, observed OTU counts, and Shannon diversity index; baseline and follow-up anthropometric measures; FTO rs9939609 polymorphism; demographic and clinical characteristics; and lifestyle factors, which were calculated according to the four categories of WC changes over 10 years. Contrary to participants with WC losses, those with WC increases were more likely to have lesser proportions of Proteobacteria and Actinobacteria and greater proportions of Firmicutes and Bacteroidetes.

Table 1.

Descriptive Characteristics of 2222 Study Participants According to the Categories of 10-Year Changes in Waist Circumference

Characteristics	Changes in waist circumference				P for trend
Characteristics	Loss >2 inches	Loss or gain ≤2 inches	Gain of 2.1–3 inches	Gain >3 inches	P for trend
No. of subjects (%)	342 (15.4)	1373 (61.8)	256 (11.5)	251 (11.3)
Men, %	46.5	52.0	50.0	39.8	0.06
FTO rs9939609 risk allele
TA and AA, %	21.9	21.8	25.8	17.5	0.47
Microbiota profile from urinary EV assay
Composition of major phyla
Proteobacteria, %	53.3 ± 14.4	50.4 ± 15.9	47.8 ± 15.4	50.3 ± 15.7	<0.01
Firmicutes, %	17.4 ± 10.8	20.2 ± 12.5	23.3 ± 13.9	19.8 ± 12.3	<0.01
Actinobacteria, %	16.4 ± 7.3	15.3 ± 7.0	14.3 ± 6.1	15.4 ± 7.8	0.02
Cyanobacteria, %	4.37 ± 5.69	4.06 ± 5.85	3.75 ± 2.91	4.10 ± 6.54	0.45
Bacteroidetes, %	3.10 ± 2.89	4.05 ± 3.87	4.33 ± 3.70	4.44 ± 4.74	<0.001
Other phyla, %	5.33 ± 4.48	5.97 ± 5.21	6.53 ± 4.82	5.97 ± 4.57	0.05
Observed OTU counts	126.6 ± 76.6	124.1 ± 78.1	125.1 ± 76.7	121.3 ± 76.5	0.45
Shannon diversity index	4.68 ± 0.57	4.65 ± 0.58	4.67 ± 0.55	4.63 ± 0.56	0.43
Baseline information
Age, years	49.0 ± 7.4	48.2 ± 7.0	47.5 ± 6.9	47.8 ± 7.1	0.02
Body weight, kg	65.6 ± 10.1	65.0 ± 10.0	63.7 ± 9.8	63.6 ± 9.5	<0.01
Body mass index, kg/m²	25.4 ± 2.9	24.8 ± 2.8	24.2 ± 2.8	24.6 ± 3.1	<0.001
Waist circumference, cm	87.3 ± 7.7	80.3 ± 8.1	77.7 ± 7.7	77.3 ± 8.0	<0.001
Current smokers, %	18.4	21.1	23.4	23.9	0.08
Current alcohol drinkers, %	43.6	55.4	52.3	50.6	0.31
Physical activity, MET-h/day	22.7 ± 8.8	23.7 ± 8.6	24.6 ± 9.5	25.5 ± 9.2	<0.001
Total calorie intake, kcal/day	1885.6 ± 647.9	1849.4 ± 476.8	1814.2 ± 430.6	1825.4 ± 463.5	0.10
Fat intake, g/day	32.4 ± 18.2	32.5 ± 16.1	31.8 ± 13.8	31.4 ± 13.9	0.38
Fiber intake, g/day	6.88 ± 3.14	6.39 ± 2.39	6.35 ± 2.28	6.22 ± 2.35	<0.01
Presence of hypertension, %	25.7	21.1	15.2	20.3	0.03
Presence of diabetes, %	13.5	10.3	5.1	8.4	<0.01
Ten-year follow-up information
Body weight, kg	61.3 ± 9.1	63.7 ± 9.8	65.6 ± 10.3	68.2 ± 10.5	<0.001
Body mass index, kg/m²	24.0 ± 2.7	24.5 ± 2.8	25.1 ± 2.9	26.6 ± 3.4	<0.001
Waist circumference, cm	77.2 ± 7.4	80.5 ± 8.1	84.1 ± 7.7	88.2 ± 8.1	<0.001
Presence of hypertension, %	34.5	35.0	36.3	43.4	0.02
Presence of diabetes, %	21.1	19.3	11.7	18.7	0.09

Data indicate mean ± standard deviation or proportion.

EV, extracellular vesicle; FTO, fat mass and obesity-associated gene; MET-h, metabolic equivalent score; OTU, operational taxonomic unit.

However, observed OTU counts and the Shannon diversity index did not differ across groups. Unlike participants with WC losses, those with WC increases consumed less fiber and were younger and more active at baseline. In addition, they were leaner at baseline, but seemed to gain more weight over 10 years, and finally they were more likely to be newly diagnosed with hypertension or DM at the 10-year follow-up. Prevalence increases over 10 years were 23.1% for hypertension and 10.3% for DM in the group of gain >3 inches, while they were 8.8% for hypertension and 7.6% for DM in the group of loss >2 inches.

Table 2 shows coefficient estimates and standard errors obtained from linear regression analyses for the association between the bacterial composition of major phyla and 10-year changes in WC. In the multiple model, we found an inverse association between the composition of Proteobacteria or Actinobacteria and WC changes (adjusted P value for trend <0.05); participants in the first tertile for Proteobacteria had greater changes, particularly increases, in mean WC than those in the second or third tertile.

Table 2.

Association Between the Composition of Major Bacterial Phyla in Urinary Extracellular Vesicles and 10-Year Changes in Waist Circumference

Bacterial phyla [median values, %]	Waist circumference changes, cm	Coefficient estimates ± SE ^a (P value)		P for trend ^c
Bacterial phyla [median values, %]	Mean ± SD	Age-adjusted model	Multiple model ^b	P for trend ^c
Proteobacteria
First tertile [36.3]	1.38 ± 6.09	Reference	Reference	<0.01
Second tertile [50.6]	0.02 ± 6.92	−1.32 ± 0.34 (<0.001)	−1.10 ± 0.31 (<0.001)
Third tertile [64.7]	0.24 ± 6.50	−1.07 ± 0.34 (0.002)	−0.82 ± 0.31 (<0.01)
Firmicutes
First tertile [10.0]	−0.24 ± 6.82	Reference	Reference	<0.001
Second tertile [17.6]	0.34 ± 6.88	0.55 ± 0.34 (0.10)	0.34 ± 0.31 (0.28)
Third tertile [29.5]	1.54 ± 5.74	1.71 ± 0.34 (<0.001)	1.23 ± 0.31 (<0.001)
Actinobacteria
First tertile [9.3]	1.00 ± 6.12	Reference	Reference	<0.01
Second tertile [14.8]	0.97 ± 6.15	−0.05 ± 0.34 (0.89)	−0.04 ± 0.31 (0.90)
Third tertile [21.3]	−0.33 ± 7.21	−1.33 ± 0.34 (<0.001)	−0.89 ± 0.31 (<0.01)
Cyanobacteria
First tertile [0.2]	0.92 ± 5.92	Reference	Reference	0.57
Second tertile [3.1]	0.41 ± 6.79	−0.55 ± 0.34 (0.10)	−0.52 ± 0.31 (0.09)
Third tertile [6.2]	0.31 ± 6.86	−0.61 ± 0.34 (0.07)	−0.18 ± 0.31 (0.55)
Bacteroidetes
First tertile [0.9]	−1.59 ± 7.08	Reference	Reference	<0.001
Second tertile [3.2]	0.94 ± 6.42	1.52 ± 0.34 (<0.001)	1.16 ± 0.31 (<0.001)
Third tertile [6.5]	1.29 ± 5.93	1.85 ± 0.34 (<0.001)	1.42 ± 0.31 (<0.001)
Other phyla
First tertile [1.9]	0.08 ± 6.43	Reference	Reference	<0.01
Second tertile [4.9]	0.20 ± 7.26	0.11 ± 0.34 (0.75)	0.40 ± 0.31 (0.20)
Third tertile [9.4]	1.35 ± 5.77	1.25 ± 0.34 (<0.001)	0.96 ± 0.31 (<0.01)

In the linear regression model, the bacterial composition of a specific phylum was fitted as an independent variable and changes in waist circumference were fitted as dependent variables.

Adjusted for age; sex; baseline waist circumference; FTO rs9939609 polymorphism; smoking status; alcohol consumption; physical activity; total calorie, fat, and fiber intake; and the presence of hypertension or diabetes mellitus at baseline.

Obtained when median values of bacterial composition were fitted as continuous variables in multiple models.

SD, standard deviation; SE, standard error.

Notably, those in the third tertile for Actinobacteria showed a mean change of −0.33 cm, indicating WC losses. On the contrary, we observed a positive association between the composition of Firmicutes and Bacteroidetes and WC changes (adjusted P value for trend <0.001); participants in the third tertile had greater changes, particularly increases, in mean WC than those in the first or second tertile.

Especially, changes in mean WC were −0.24 and −1.59 cm for the first tertiles of Firmicutes and Bacteroidetes, respectively. Higher proportions of other phyla were associated with changes in WC, but the composition of Cyanobacteria was not. No significant bacterial phylum was observed to have an association with 10-year changes in BW (data available upon request).

Table 3 shows HRs for the incidence of abdominal obesity in the association with bacterial composition of major phyla, with 95% CIs. Compared with participants in the first tertile of the composition of Proteobacteria, those in the second and third tertiles showed 29% (95% CI: 0.56 − 0.89) and 22% (95% CI: 0.62 − 0.97) lower risk of abdominal obesity, respectively (P value for trend = 0.03). Adjusted P values were 0.03 and 0.13 for the second and third tertiles, respectively.

Table 3.

Association Between the Composition of Major Bacterial Phyla in Urinary Extracellular Vesicles and the Incidence of Abdominal Obesity over a 10-Year Period Among 1596 Participants Who Did Not Have Abdominal Obesity at Baseline

Bacterial phyla [median value, %]	No. of cases/noncases	Hazard ratio ^a (95% CI)		P for trend ^c
Bacterial phyla [median value, %]	No. of cases/noncases	Age-adjusted model	Multivariate model ^b	P for trend ^c
Proteobacteria
First tertile [35.6]	170/362	Reference	Reference	0.028
Second tertile [50.0]	141/391	0.73 (0.58–0.91)	0.71 (0.56–0.89)
Third tertile [64.2]	157/375	0.79 (0.63–0.98)	0.78 (0.62–0.97)
Firmicutes
First tertile [10.3]	151/381	Reference	Reference
Second tertile [18.2]	161/371	1.18 (0.95–1.48)	1.17 (0.93–1.47)	0.001
Third tertile [30.3]	156/376	1.24 (0.98–1.55)	1.48 (1.17–1.86)
Actinobacteria
First tertile [9.2]	156/376	Reference	Reference	0.36
Second tertile [14.6]	153/379	0.98 (0.79–1.23)	1.00 (0.79–1.25)
Third tertile [20.8]	159/373	0.95 (0.77–1.19)	0.90 (0.72–1.13)
Cyanobacteria
First tertile [0.3]	140/392	Reference	Reference	0.40
Second tertile [3.0]	158/374	1.16 (0.93–1.46)	1.11 (0.88–1.40)
Third tertile [6.1]	170/362	1.22 (0.98–1.53)	1.12 (0.89–1.41)
Bacteroidetes
First tertile [1.0]	157/375	Reference	Reference	0.21
Second tertile [3.4]	158/374	1.08 (0.86–1.34)	1.18 (0.94–1.48)
Third tertile [6.7]	153/379	1.09 (0.87–1.37)	1.17 (0.93–1.47)
Other phyla
First tertile [1.8]	146/386	Reference	Reference	0.47
Second tertile [5.1]	159/373	1.08 (0.86–1.35)	1.07 (0.84–1.33)
Third tertile [9.6]	163/369	1.20 (0.96–1.50)	1.09 (0.87–1.38)

Hazard ratios for cases of abdominal obesity (waist circumference ≥90 cm for men and ≥80 cm for women) in the association with bacterial composition of a specific phylum were obtained using Cox proportional hazards regression analysis.

Obtained when median values of bacterial composition were fitted as continuous variables in multiple models.

CI, confidence interval.

On the contrary, participants in the third tertile of the composition of Firmicutes showed a 48% (95% CI: 1.17 − 1.86) higher risk of abdominal obesity compared with those in the first tertile (P value for trend = 0.001). The adjusted P value of the third tertile was still significant at 0.01. However, the composition of Actinobacteria, Cyanobacteria, Bacteroidetes, and other phyla was not associated with the risk of abdominal obesity. No significant bacterial phylum was observed to have an association with the risk of obesity (data available upon request).

Next, we further conducted joint analysis of the composition of Proteobacteria and Firmicutes for the association with the incidence of abdominal obesity. Tertiles of Proteobacteria (<45%, 45% −55%, and >55%) and Firmicutes (<15%, 15% −22%, and >22%) compositions were combined, generating nine groups for joint analysis.

Compared with the reference of 45% −55% and <15% for the composition of Proteobacteria and Firmicutes, respectively, HRs were 1.36 (95% CI: 0.71 − 2.59) for the group with <45% and <15%; 1.19 (95% CI: 0.79 − 1.80) for >55% and <15%; 1.82 (95% CI: 1.13 − 2.94) for <45% and 15% −22%; 1.16 (95% CI: 0.75 − 1.79) for 45% −55% and 15% −22%; 1.43 (95% CI: 0.87 − 2.36) for >55% and 15% −22%; 1.69 (95% CI: 1.11 − 2.56) for <45% and >22%; 1.57 (95% CI: 0.94 − 2.62) for 45% −55% and >22%; and 2.59 (95% CI: 1.33 − 5.01) for >55% and >22% (Fig. 1).

FIG. 1.

Association between the composition of Proteobacteria and Firmicutes in urinary EV samples and the incidence of abdominal obesity over a 10-year period among 1596 participants who did not have abdominal obesity at baseline. Hazard ratios were estimated in the joint analysis for the composition tertiles of combination groups of Proteobacteria (tertile: <45%, 45% −55%, and >55%) and Firmicutes (tertile: <15%, 15% −22%, and >22%) using the reference group. Adjusted P value is indicated. EV, extracellular vesicle.

Only the group with >55% and >22% showed significance after multiple testing adjustment (adjusted P value <0.05). In addition, when we compared serum levels of C-reactive protein, a biomarker of systemic inflammation, across the four groups, no significant difference was observed (data available upon request).

Next, we attempted to find out the genera that mainly contributed to the composition of Proteobacteria and Firmicutes as well as their associations with the incidence of abdominal obesity. Table 4 presents the associations between major genera of Proteobacteria and Firmicutes with mean composition greater than 0.1% and the risk of abdominal obesity.

Table 4.

Association Between the Composition of Major Bacterial Genera in Urinary Extracellular Vesicles and the Incidence of Abdominal Obesity over a 10-Year Period Among 1596 Participants Who Did Not Have Abdominal Obesity at Baseline

Major genera	Taxonomic information			Composition	Hazard ratio ^a
Major genera	Family	Order	Class	Mean ± SD	(95% CI)	P
Phylum: Proteobacteria
Skermanella	Rhodospirillaceae	Rhodospirillales	Alphaproteobacteria	0.12 ± 0.57	1.17 (1.03–1.33)	0.02
Aquicella	Coxiellaceae	Legionellales	Gammaproteobacteria	0.10 ± 0.32	1.35 (1.06–1.72)	0.02
Agrobacterium	Rhizobiaceae	Rhizobiales	Alphaproteobacteria	1.49 ± 2.06	0.95 (0.90–0.99)	0.04
Achromobacter	Alcaligenaceae	Burkholderiales	Betaproteobacteria	0.45 ± 0.97	0.89 (0.80–0.99)	0.04
Erwinia	Enterobacteriaceae	Enterobacteriales	Gammaproteobacteria	0.23 ± 0.85	0.87 (0.74–1.02)	0.09
Kaistobacter	Sphingomonadaceae	Sphingomonadales	Alphaproteobacteria	0.20 ± 0.55	0.85 (0.68–1.05)	0.13
Comamonas	Comamonadaceae	Burkholderiales	Betaproteobacteria	0.21 ± 0.52	0.87 (0.71–1.06)	0.16
Delftia	Comamonadaceae	Burkholderiales	Betaproteobacteria	0.24 ± 0.87	1.07 (0.96–1.19)	0.23
Citrobacter	Enterobacteriaceae	Enterobacteriales	Gammaproteobacteria	0.32 ± 0.83	1.07 (0.95–1.21)	0.27
Lautropia	Burkholderiaceae	Burkholderiales	Betaproteobacteria	0.19 ± 0.69	0.88 (0.69–1.12)	0.30
Enhydrobacter	Moraxellaceae	Pseudomonadales	Gammaproteobacteria	2.47 ± 3.73	1.01 (0.99–1.04)	0.32
Novosphingobium	Sphingomonadaceae	Sphingomonadales	Alphaproteobacteria	0.72 ± 1.27	0.95 (0.87–1.05)	0.32
Sphingobium	Sphingomonadaceae	Sphingomonadales	Alphaproteobacteria	0.26 ± 0.65	0.92 (0.77–1.09)	0.33
Rheinheimera	Chromatiaceae	Chromatiales	Gammaproteobacteria	0.12 ± 0.45	1.11 (0.89–1.38)	0.35
Shewanella	Shewanellaceae	Alteromonadales	Gammaproteobacteria	0.11 ± 0.52	0.89 (0.68–1.15)	0.35
Methylobacterium	Methylobacteriaceae	Rhizobiales	Alphaproteobacteria	1.86 ± 2.52	0.98 (0.94–1.03)	0.41
Bdellovibrio	Bdellovibrionaceae	Bdellovibrionales	Deltaproteobacteria	0.15 ± 0.50	0.93 (0.75–1.15)	0.52
Acinetobacter	Moraxellaceae	Pseudomonadales	Gammaproteobacteria	3.15 ± 3.34	0.99 (0.97–1.02)	0.64
Tepidimonas	Comamonadaceae	Burkholderiales	Betaproteobacteria	0.12 ± 0.52	1.03 (0.89–1.19)	0.68
Haemophilus	Pasteurellaceae	Pasteurellales	Gammaproteobacteria	0.64 ± 1.65	0.99 (0.92–1.06)	0.70
Sphingomonas	Sphingomonadaceae	Sphingomonadales	Alphaproteobacteria	3.60 ± 3.54	0.99 (0.96–1.03)	0.72
Neisseria	Neisseriaceae	Neisseriales	Betaproteobacteria	0.39 ± 0.89	1.02 (0.90–1.17)	0.73
Brevundimonas	Caulobacteraceae	Caulobacterales	Alphaproteobacteria	0.17 ± 0.60	0.98 (0.84–1.15)	0.81
Hydrogenophilus	Hydrogenophilaceae	Hydrogenophilales	Hydrogenophilalia	0.17 ± 0.75	0.98 (0.81–1.18)	0.81
Paracoccus	Rhodobacteraceae	Rhodobacterales	Alphaproteobacteria	1.02 ± 1.75	1.00 (0.94–1.06)	0.89
Pseudomonas	Pseudomonadaceae	Pseudomonadales	Gammaproteobacteria	15.12 ± 10.75	1.00 (0.99–1.01)	0.91
Roseateles	Comamonadaceae	Burkholderiales	Betaproteobacteria	0.35 ± 0.90	1.00 (0.91–1.11)	0.94
Actinobacillus	Pasteurellaceae	Pasteurellales	Gammaproteobacteria	0.12 ± 0.43	1.01 (0.71–1.43)	0.97
Blastomonas	Sphingomonadaceae	Sphingomonadales	Alphaproteobacteria	0.28 ± 0.65	1.00 (0.84–1.19)	0.97
Klebsiella	Enterobacteriaceae	Enterobacteriales	Gammaproteobacteria	0.13 ± 1.93	1.00 (0.97–1.03)	0.98
Phylum: Firmicutes
Megasphaera	Veillonellaceae	Clostridiales	Clostridia	0.18 ± 0.67	1.13 (1.01–1.27)	0.03
Lactobacillus	Lactobacillaceae	Lactobacillales	Bacilli	4.31 ± 8.07	1.01 (1.00–1.02)	0.03
Blautia	Lachnospiraceae	Clostridiales	Clostridia	0.46 ± 1.88	1.07 (1.00–1.13)	0.03
Bacillus	Bacillaceae	Bacillales	Bacilli	0.67 ± 1.71	1.04 (0.99–1.09)	0.10
Veillonella	Veillonellaceae	Clostridiales	Clostridia	0.60 ± 1.30	0.93 (0.85–1.03)	0.17
Weissella	Leuconostocaceae	Lactobacillales	Bacilli	0.41 ± 2.29	0.97 (0.91–1.03)	0.28
Aerococcus	Aerococcaceae	Lactobacillales	Bacilli	0.15 ± 0.54	1.09 (0.90–1.32)	0.38
Granulicatella	Carnobacteriaceae	Lactobacillales	Bacilli	0.10 ± 0.37	0.85 (0.57–1.27)	0.43
Streptococcus	Streptococcaceae	Lactobacillales	Bacilli	3.50 ± 3.95	1.01 (0.98–1.04)	0.50
Shuttleworthia	Lachnospiraceae	Clostridiales	Clostridia	0.19 ± 1.33	0.98 (0.89–1.07)	0.65
Lactococcus	Streptococcaceae	Lactobacillales	Bacilli	0.14 ± 0.70	0.97 (0.83–1.12)	0.66
Finegoldia	Peptoniphilaceae	Clostridiales	Clostridia	0.16 ± 0.53	1.05 (0.85–1.29)	0.67
Faecalibacterium	Clostridiaceae	Clostridiales	Clostridia	0.22 ± 1.05	1.03 (0.88–1.21)	0.69
Anaerococcus	Tissierellaceae	Clostridiales	Clostridia	0.24 ± 0.59	0.97 (0.82–1.16)	0.75
Peptoniphilus	Peptoniphilus	Clostridiales	Clostridia	0.12 ± 0.47	1.03 (0.83–1.27)	0.80
Dialister	Veillonellaceae	Clostridiales	Clostridia	0.22 ± 0.80	1.02 (0.88–1.18)	0.82
Coprococcus	Lachnospiraceae	Clostridiales	Clostridia	0.16 ± 0.59	0.98 (0.81–1.18)	0.82
Ruminococcus	Ruminococcaceae	Clostridiales	Clostridia	0.24 ± 0.96	0.98 (0.82–1.18)	0.85
Enterococcus	Enterococcaceae	Lactobacillales	Bacilli	0.37 ± 1.04	1.00 (0.91–1.11)	0.94
Staphylococcus	Staphylococcaceae	Bacillales	Bacilli	3.61 ± 4.37	1.00 (0.98–1.02)	1.00

Hazard ratios for cases of abdominal obesity (waist circumference ≥90 cm for men and ≥80 cm for women) in the association with the bacterial composition of a specific genus were obtained from Cox proportional hazards regression analysis. Data were adjusted for age; sex; baseline waist circumference; FTO rs9939609 polymorphism; smoking status; alcohol consumption; physical activity; total calorie, fat, and fiber intake; and the presence of hypertension or diabetes mellitus at baseline.

Within the phylum Proteobacteria, Skermanella and Aquicella were positively associated, whereas Agrobacterium and Achromobacter were inversely associated, with the risk of abdominal obesity. Within the phylum Firmicutes, the composition of Megasphaera, Lactobacillus, and Blautia was positively associated with the risk of abdominal obesity. For these genera, however, adjusted P values were >0.05.

Discussion

In this prospective study, we observed a temporal relationship between bacterial composition in the urinary EV samples and the risk of abdominal obesity. In the metagenomic analysis at the phylum level, the composition of Proteobacteria and Firmicutes significantly predicted the 10-year risk of abdominal obesity, but not the risk of obesity as defined by BMI; an inverse association with Proteobacteria composition and a positive association with Firmicutes composition.

Under the condition of increased proportions of Firmicutes, however, higher Proteobacteria proportions elevated the risk of abdominal obesity to 2.6 times. On the basis of findings regarding longitudinal WC changes, Actinobacteria may be a promising phylum to enhance WC losses.

Dysbiosis indicates an imbalance of commensal bacteria in the human body, mainly in the gut, and has been reported to be associated with several diseases.²⁶ In fact, lower diversity or altered composition of bacterial communities in fecal samples was observed in persons with obesity, diabetic or hypertensive patients, or patients with inflammatory bowel disease than in lean or healthy counterparts.^{11

–16,27,28}

Particularly, persons with obesity were more likely to have increased Firmicutes or decreased Bacteroidetes composition or an elevated ratio of these two phyla in some studies,^11
–13 but not in others.^14,15 In addition to such inconsistent observations, a causal relationship between specific bacteria and development of obesity remains unclear in human studies.

One prospective cohort study investigated the association between the quantity of commensal bacteria and development of diabetes and obesity.⁷ In the study including 3280 participants, researchers assessed bacterial 16S rDNA concentrations in blood samples, which were acquired >9 years ago, and evaluated the association of this concentration with the incidence of DM ascertained through follow-up as well as with the incidence of obesity and abdominal obesity confirmed during the last follow-up period.

They found that blood concentrations of bacterial DNA gene were positively associated with the incidence of diabetes or abdominal obesity, but not with obesity as defined by BMI. However, when they compared the proportions of several phyla between diabetic and nondiabetic participants, they found no significant difference.⁷ They observed that Proteobacteria was the bacterial phylum with the highest concentration in blood, as reported previously.^5,6

On the basis of bacterial composition in fecal samples, which reflects the composition in the lumen of the large intestine, Bacteroidetes and Firmicutes are the most prominently present phyla, dominating over 90%, whereas Proteobacteria occupies <5% of the composition.²⁹

In the present study, we also observed Proteobacteria as the largest bacterial community at the phylum level in the urinary EV composition and found its association with the risk of abdominal obesity, as previously suggested.⁷ In addition, the Firmicutes composition in urinary EVs was strongly and positively related to the risk of abdominal obesity, which is an metabolic syndrome (MS) component. However, no significant phyla were found to have an association with the risk of obesity defined by BMI or with changes in BW, and these findings were not in line with earlier reports.^11
–13

Some discrepancies between our study and others (for example, the association between Bacteroidetes and obesity or BW change) may be partly due to differences in the study participants' age and ethnicity, the study design, or biospecimen materials. Because the participants of our study were middle-aged or older adults, they might be susceptible to gain more fat in the waist, increasing the WC, than in other body sites because of aging. In fact, because the incidence of abdominal obesity was three times higher than that of obesity within a similar sample size, the study participants were considered to be vulnerable to MS risk.

Although no associations for BW change and significant associations for WC change were found in the same sample size, we cannot exclude the possibility of limited statistical power for the association with the incidence of obesity. Subsequently, reevaluation of the association with the incidence of obesity once more cases have accrued may be warranted.

Biological mechanisms underlying the association between commensal bacteria and development of obesity are still being explored: some suggestions include increasing energy harvest by transforming indigestible compounds into digestible energy sources; increasing fat deposition by modulating energy metabolism; and making changes in appetite-regulating hormones, immune system, or energy expenditure.³⁰

A meta-analysis on clinical trials and animal experiments demonstrated that some Lactobacillus species contribute to weight gain.³¹ Blautia is one of the genera of Firmicutes and butyrate-producing bacteria that harvest energy from polysaccharides, but data are still limited on the association between Blautia and obesity.³² Because Proteobacteria and Bacteroidetes are gram-negative bacteria, they produce OMVs containing lipopolysaccharide, which is known to activate the innate immune system.

It was reported that lipopolysaccharide causes proinflammation, which is a condition observed in obesity, particularly with overabundance of Proteobacteria.³³ Thus, we postulate that the mixed influence of Firmicutes and Proteobacteria through the energy harvesting mechanism and inflammatory responses may contribute to an elevated risk of abdominal obesity in the joint analysis.

In our results, WC loss was observed in participants with higher composition of Actinobacteria; however, it is unclear whether WC loss was unintentional. The evidence on the association between Actinobacteria and obesity is still conflicting,^10,34 but this phylum includes some notable genera such as Streptomyces, with antibiotic properties,³⁴ and Bifidobacterium, a strain that may have antiobesity effects.³⁵

The strengths of our study include a prospective investigation, a large sample size, measurement of both BW and WC by trained persons, and consideration of a broad range of confounding factors. A study limitation is the limited generalizability of our findings. The findings may be generalizable to middle-aged and older Asians, but not to other age groups or ethnicities. Metagenomic data from urinary EVs may not reflect those from fecal samples.

In this study, nevertheless, we attempted to discover markers that can predict the risk of obesity by assessing bacterial DNA from urinary EVs because urine is more readily obtainable than fecal samples, which have been used in most metagenomic studies previously, and is more convenient in terms of feasibility of an investigation with a large number of study participants.

Our prospective study found that the composition of Proteobacteria and Firmicutes and some of their genera in urinary EV samples was significantly associated with the 10-year risk of abdominal obesity, but not with the risk of obesity as defined by BMI. Further prospective studies are needed to confirm the findings, and in particular, studies including other ethnicities and larger populations with broader age ranges are warranted.

Footnotes

Authors' Contributions

I.B. provided the study inference, conducted the statistical analysis, and wrote the manuscript and C.S. conducted the study. All authors contributed to interpreting the result and finalizing the manuscript.

Author Disclosure Statement

No conflicting financial interests exist.

Funding Information

This study was supported by a fund (2001-347-6111-221, 2002-347-6111-221) for research at Korea Centers for Disease Control and Prevention and by a National Research Foundation of Korea grant funded by the Korean Government (NRF-2019R1A2C2084000).

References

Brown

, Wolf

, Prados-Rosales

, et al. Through the wall: Extracellular vesicles in Gram-positive bacteria, mycobacteria and fungi. Nat Rev Microbiol, 2015; 13(10):620–630; doi: 10.1038/nrmicro3480

Kim

, Lee

, Park

, et al. Gram-negative and Gram-positive bacterial extracellular vesicles. Semin Cell Dev Biol, 2015; 40:97–104; doi: 10.1016/j.semcdb.2015.02.006

Choi

, Kwon

, Kim

, et al. Gut microbe-derived extracellular vesicles induce insulin resistance, thereby impairing glucose metabolism in skeletal muscle. Sci Rep, 2015; 5:15878; doi: 10.1038/srep15878

Yoo

, Rho

, You

, et al. 16S rRNA gene-based metagenomic analysis reveals differences in bacteria-derived extracellular vesicles in the urine of pregnant and non-pregnant women. Exp Mol Med, 2016; 48(2):e208; doi: 10.1038/emm.2015.110

Wang

, Li

, Yuan

, et al. The complex exogenous RNA spectra in human plasma: An interface with human gut biota?. PLoS One, 2012; 7(12):e51009; doi: 10.1371/journal.pone.0051009

Beatty

, Guduric-Fuchs

, Brown

, et al. Small RNAs from plants, bacteria and fungi within the order Hypocreales are ubiquitous in human plasma. BMC Genomics, 2014; 15(1):933; doi: 10.1186/1471-2164-15-933

Amar

, Serino

, Lange

, et al;

D.E.S

.I.R. Study Group. Involvement of tissue bacteria in the onset of diabetes in humans: Evidence for a concept. Diabetologia, 2011; 54(12):3055–3061; doi: 10.1007/s00125-011-2329-8

Muraca

, Putignani

, Fierabracci

, et al. Gut microbiota-derived outer membrane vesicles: under-recognized major players in health and disease?. Discov Med, 2015; 19(106):343–348

Patten

, Hussein

, Davies

, et al. Commensal-derived OMVs elicit a mild proinflammatory response in intestinal epithelial cells. Microbiology, 2017; 163(5):702–711; doi: 10.1099/mic.0.000468

10.

Kang

, Ban

, Choi

, et al. Extracellular vesicles derived from gut microbiota, especially Akkermansia muciniphila, protect the progression of dextran sulfate sodium-induced colitis. PLoS One, 2013; 8(10):e76520; doi: 10.1371/journal.pone.0076520

11.

Turnbaugh

, Hamady

, Yatsunenko

, et al. A core gut microbiome in obese and lean twins. Nature, 2009; 457(7228):480–484; doi: 10.1038/nature07540

12.

Bervoets

, Van Hoorenbeeck

, Kortleven

, et al. Differences in gut microbiota composition between obese and lean children: A cross-sectional study. Gut Pathog, 2013; 5:10; doi: 10.1186/1757-4749-5-10

13.

Ferrer

, Ruiz

, Lanza

, et al. Microbiota from the distal guts of lean and obese adolescents exhibit partial functional redundancy besides clear differences in community structure. Environ Microbiol, 2013; 15(1):211–226; doi: 10.1111/j.1462-2920.2012.02845.x

14.

Schwiertz

, Taras

, Schäfer

, et al. Microbiota and SCFA in lean and overweight healthy subjects. Obesity (Silver Spring), 2010; 18(1):190–195; doi: 10.1038/oby.2009.167

15.

Patil

, Dhotre

, Chavan

, et al. Molecular analysis of gut microbiota in obesity among Indian individuals. J Biosci, 2012; 37(4):647–657; doi: 10.1007/s12038-012-9244-0

16.

Walters

, Xu

, Knight

. Meta-analyses of human gut microbes associated with obesity and IBD. FEBS Lett, 2014; 588(22):4223–4233; doi: 10.1016/j.febslet.2014.09.039

17.

Menni

, Jackson

, Pallister

, et al. Gut microbiome diversity and high-fibre intake are related to lower long-term weight gain. Int J Obes (Lond), 2017; 41(7):1099–1105; doi: 10.1038/ijo.2017.66

18.

Seganfredo

, Blume

, Moehlecke

, et al. Weight-loss interventions and gut microbiota changes in overweight and obese patients: A systematic review. Obes Rev, 2017; 18(8):832–851; doi: 10.1111/obr.12541

19.

Dror

, Dickstein

, Dubourg

, et al. Microbiota manipulation for weight change. Microb Pathog, 2017; 106:146–161; doi: 10.1016/j.micpath.2016.01.002

20.

Baik

, Kim

, Abbott

, et al. Association of snoring with chronic bronchitis. Arch Intern Med, 2008; 168(2):167–173; doi: 10.1001/archinternmed.2007.8

21.

Baik

, Shin

. Prospective study of alcohol consumption and metabolic syndrome. Am J Clin Nutr, 2008; 87(5):1455–1463; doi: 10.1093/ajcn/87.5.1455

22.

The World Health Organization Western Pacific Region. The International Association for the Study of Obesity, and The International Obesity Task Force. The Asia-Pacific Perspective: Redefining Obesity and Its Treatment. Health Communications Australia Pty Limited: Sydney; 2000.

23.

WHO Expert Consultation. Appropriate body-mass index for Asian populations and its implications for policy and intervention strategies. Lancet, 2004; 363(9403):157–163; doi: 10.1016/S0140-6736(03)15268-3

24.

Baik

, Shin

. Interactions between the FTO rs9939609 polymorphism, body mass index, and lifestyle-related factors on metabolic syndrome risk. Nutr Res Pract, 2012; 6(1):78–85; doi: 10.4162/nrp.2012.6.1.78

25.

Cho

, Go

, Kim

, et al. A large-scale genome-wide association study of Asian populations uncovers genetic factors influencing eight quantitative traits. Nat Genet, 2009; 41(5):527–534; doi: 10.1038/ng.357

26.

Petersen

, Round

. Defining dysbiosis and its influence on host immunity and disease. Cell Microbiol, 2014; 16(7):1024–1033; doi: 10.1111/cmi.12308

27.

Larsen

, Vogensen

, van den Berg

, et al. Gut microbiota in human adults with type 2 diabetes differs from non-diabetic adults. PLoS One, 2010; 5(2):e9085; doi: 10.1371/journal.pone.0009085

28.

Yang

, Santisteban

, Rodriguez

, et al. Gut dysbiosis is linked to hypertension. Hypertension, 2015; 65(6):1331–1340.

29.

Eckburg

, Bik

, Bernstein

, et al. Diversity of the human intestinal microbial flora. Science, 2005; 308:1635–1638; doi: 10.1161/HYPERTENSIONAHA.115.05315

30.

Boulangé

, Neves

, Chilloux

, et al. Impact of the gut microbiota on inflammation, obesity, and metabolic disease. Genome Med, 2016; 8(1):42; doi: 10.1186/s13073-016-0303-2

31.

Million

, Angelakis

, Paul

, et al. Comparative meta-analysis of the effect of Lactobacillus species on weight gain in humans and animals. Microb Pathog, 2012; 53(2):100–108; doi: 10.1016/j.micpath.2012.05.007

32.

Org

, Blum

, Kasela

, et al. Relationships between gut microbiota, plasma metabolites, and metabolic syndrome traits in the METSIM cohort. Genome Biol, 2017; 18(1):70; doi: 10.1186/s13059-017-1194-2

33.

Creely

, McTernan

, Kusminski

, et al. Lipopolysaccharide activates an innate immune system response in human adipose tissue in obesity and type 2 diabetes. Am J Physiol Endocrinol Metab, 2007; 292(3):E740–E747; doi: 10.1152/ajpendo.00302.2006

34.

Procópio

, Silva

, Martins

, et al. Antibiotics produced by Streptomyces. Braz J Infect Dis, 2012; 16(5):466–471; doi: 10.1016/j.bjid.2012.08.014

35.

Yin

, Yu

, Fu

, et al. Effects of four Bifidobacteria on obesity in high-fat diet induced rats. World J Gastroenterol, 2010; 16(27):3394–3401; doi: 10.3748/wjg.v16.i27.3394