Resistant starch intake facilitates weight loss in humans by reshaping the gut microbiota

Study participants

The inclusion criteria were (1) age 18–55 years and (2) overweight or obesity, as defined as BMI ≥ 24 kg m⁻² or waist circumference ≥85 cm in men and ≥80 cm in women²⁰. The exclusion criteria were acute illness, antibiotic or probiotic supplement use within 3 weeks previously, diagnosis of hyperthyroidism or hypothyroidism, diabetes, current systemic corticosteroid treatment or medications affecting glucose metabolism, and participation in other clinical trials within 4 weeks before the study.

This study was approved by the ethics committee of the Shanghai Sixth People’s Hospital and conformed with the Declaration of Helsinki. Written informed consent was obtained from all participants. Complete clinical trial registration was deposited in the Chinese Clinical Trial Registry (ChiCTR-TTRCC 13003333) (Extended Data Fig. 1).

General protocol of the clinical trial

The detailed study protocol was submitted to the Chinese Clinical Trial Management Public Platform and is available in the Supplementary Information. Individuals with excess body weight consumed either RS derived from maize (HAM-RS2, Hi-Maize 260 resistant starch, 22000B00, provided by Ingredion) or energy-matched CS (AMIOCA cornstarch, 04400110, also provided by Ingredion) alternately for 8 weeks and separated by a 4-week washout period. An independent researcher performed the randomization and participant allocation to the CS-Washout-RS or RS-Washout-CS intervention schemes with at a 1:1 allocation ratio. The randomization schedule was produced by SAS PROC PLAN in SAS software. RS and CS were packaged in identical sealed bags with an identical appearance and participants and investigators were blinded to the group allocations during the double-blind period. Only the research designer was aware of the randomization scheme, whereas the participants, investigators, clinical staff and outcome assessors were blinded to it. The blinding was lifted during the bioinformatics analysis to explore the potential mechanism by which the gut microbiota conferred the physiological benefits of RS.

During this feeding study, participants received a uniform background diet, following the Chinese and American guidelines for the prevention and control of adults with overweight and obesity^20,21. All participants were either lightly active or led sedentary lives. The diet provided 25 kcal kg⁻¹ ideal body weight (ideal body weight (kg) = height (cm) − 105) daily, with 50–60% carbohydrate, 25–30% fat and 15–20% protein. Participants were permitted one item of fruit daily and advised against extra-sugary beverages or snacks. They maintained a diary recording three consecutive 24-h dietary intake (two weekdays and one weekend day) at baseline and at the beginning and end of each intervention period, noting any dietary deviations. A trained dietician assessed starch consumption and dietary regimen adherence during visits. Diet data revealed similar average calorie intake and macronutrient percentages in the RS and CS intervention periods (Supplementary Table 1).

Participants attended the institute for visits (V) 1–10 (Fig. 1a). At the start and the end of each intervention (V1, V5, V6 and V10), participants had a hyperinsulinemic–euglycemic clamp, MTT, MRI scans and sample collections³⁷. Anthropometric and biochemical assessments were conducted at each visit. The primary outcome was body weight and secondary outcomes included VFA and SFA, body fat, waist circumference, lipid profiles, insulin sensitivity, metabolome and gut microbiome.

Anthropometric and biochemical assessments

After overnight fasting for at least 10 h, anthropometric test and samples (venous blood, urine and faeces) collection was conducted in the morning following the study protocol.

Serum samples after centrifugation were stored at −80 °C until measuring AST, ALT, GGT, HDL-C, LDL-C and insulin³⁷. Serum A-FABP (AIS, HKU, 31030), adiponectin (AIS, HKU, 31010), inflammatory cytokines (Ebiosciense, TNFα, BMS223HS; MCP-1, BMS281; IL-1β, BMS224HS; IL-6, BMS213HS; IL-10, BMS215HS), ANGPTL4 (BioVendor, RD191073200R) and FGF21 (AIS, HKU, 31180) were quantified by enzyme-linked immunosorbent assay (ELISA). Serum LPS levels were measured through sensitive limulus amoebocyte lysate (LAL) assay (Hycult Biotechnology, MAK109).

MTT experiment

To evaluate glucose metabolism, serial blood samples were collected in both fasting and postprandial states for laboratory tests following a standard meal (315.2 kcal, including 68.4 g carbohydrate and 10.4 g protein) (China Oil & Foodstuffs Corporation).

Hyperinsulinemic–euglycemic clamp

Insulin sensitivity was evaluated through a hyperinsulinemic–euglycemic clamp. Insulin levels were maintained at approximately 100 μU ml⁻¹ through a prime-continuous infusion of insulin. To maintain the concentration of plasma glucose at basal levels, variable glucose infusion following a negative feedback principle was conducted. The glucose clamp technique was performed as previously described⁵⁰.

Faecal DNA extraction and sequencing

Faecal samples were collected through a tube with a DNA stabilizer (STRATEC Molecular) and stored at −80 °C. Faecal genomic DNA from both humans and mice receiving FMT from human donors was extracted using PSP Spin Stool DNA kit (STRATEC Molecular, 1038100) following the manufacturer’s instructions. Faecal DNA extracts from 86 samples (27 samples before and after RS and 16 samples before and after CS) were randomly selected from all participants for shotgun metagenomic sequencing. HiSeq 1500 was used for 100-bp paired-end shotgun metagenomic sequencing for human samples at the Centre for Genomic Sciences, the University of Hong Kong. An Illumina platform Novaseq 6000 was used for 150-bp paired-end sequencing for mice samples at Novogene. Raw sequences were stored in the National Center for Biotechnology (NCBI) Sequence Read Archive (project ID PRJNA414688).

Metagenomics analysis

Quality control for metagenomic data

We used a series of quality control steps to remove low-quality reads/bases as described previously⁵¹. In brief, all Illumina primer/adaptor/linker sequences were removed and low-quality regions (quality score <20) and reads were trimmed. Subsequently, we mapped all reads to the human genome using BWA v.0.7.4 (ref. ⁵²). Reads with >95% identity and 90% coverage were removed as human DNA contamination.

Taxonomy and functional profiling

Taxonomy affiliations for reads were determined by MetaPhlAn⁵³. We distilled the top 50 most-abundant species based on the average abundance across all samples (minimum average abundance > 0.3%), which accounted for 91% of the total relative abundance in each sample on average. The proportion of unclassified reads was 8.06 ± 6.71% (mean ± s.d.) across all samples. To overcome potential bias from baseline differences between individuals within groups and to facilitate cross-group comparisons, we compared community-level variation patterns (based on the fold change in abundance of species), instead of raw community structures between two groups (based on the relative abundances of species). We calculated the log₂FC in abundance after treatment for each species. Fold change values were normalized (within all samples in each group) ranging from 0 to 1 and further rescaled to make the sum of the normalized fold change as one. We further calculated the Bray–Curtis distance based on the profile of the normalized fold change of species abundance.

Differentially varied species after treatment with RS or CS were tested using a Wilcoxon signed-rank test, with an FDR cutoff of 0.2. The NMDS ordination analysis based on the normalized Bray–Curtis dissimilarity of species-level abundance fold change, described above, was performed with R package VEGAN⁵⁴. Network visualization was performed by Cytoscape 3 (ref. ⁵⁵).

IDBA-UD⁵⁶ was adopted de novo assemblies with k-mer size ranging from 20 to 100 bp. The method for functional annotation and abundance calculations for KEGG pathways is described in an online server that we developed⁵⁷. In brief, MetaGeneMark⁵⁸ was used for microbial gene prediction. Metagenomic reads were mapped to assembled contigs to get gene-level DNA abundance levels with reads per kilobase per million (RPKM) mapped reads. We used KOBAS⁵⁹ to annotate KEGG orthologues and pathways based on annotated genes. Scripts from our publish servers⁵⁷ were used to calculate the aggregated RPKM for each pathway.

Metagenomics association analyses

Generalized linear modelling was performed using the R package glmulti. For each phenotype, the response variable was the fold change of the phenotype; the independent variables were from a subset of the fold change of 12 species with FDR < 0.3 in the differentially varied analysis using a Wilcoxon rank-sum test. The linear models started from ‘PHENOTYPE ~’, generating all possible models to reach a best model with feature selections. The importance for a particular species variable was calculated as the proportion of the total weights/probabilities for the models containing the variable during automatic model selection based on corrected AIC. This importance index reflects overall support of all possible weighted linear models. Only the main effects in the generalized linear models were considered. The P value of a particular species variable was calculate using Fisher’s method to combine the P value among the top ten models.

To quantify the association between (1) phenotypes and metabolites, (2) gut microbes and blood metabolites and (3) gut microbial pathways and blood metabolites, we adopted partial least-squares discriminant analysis (PLS-DA) using the mixOmics package⁶⁰ in R. The number of components was set to 2 and number of features for each component was estimated based on hyperparameter optimization.

Metabolomics profiling of human serum samples

The serum untargeted metabolomics data were quantified through a Shimadzu Prominence HPLC system (Shimadzu) coupled to an LTQ Orbitrap Velos instrument (Thermo Fisher Scientific) at the CAS Key Laboratory of Separation Science for Analytical Chemistry. The details of analytical conditions were as described in a previous study⁶¹. The quantitation of bile acid profiles was performed by Metabo-Profile^62,63,64.

Metabolomics profiling of human faecal samples

The metabolomics data of human faecal samples were quantified by GC–TOF–MS (Agilent 6890N GC coupled with a LECO Pegasus HT TOF–MS). The details of analytical condition as described in previous study⁶⁵.

Faecal lipid assessment

Faecal lipids were extracted from the faeces of participants after 8-week RS or CS interventions using published methods^66,67. Approximately 100 mg lyophilized faeces was acidified with 100 μl ethanol and 500 μl 8 N hydrochloric acid, then kept in a water bath at 80 °C for 40 min. Fatty acids were subsequently extracted with 1,250 μl diethyl ether and 1,250 μl petroleum ether. After filtering the ether-fat upper phase, the clear ether solution was collected and kept in a rotary evaporator for 1 h at 40 °C. Fatty acids were resolubilized in ethanol and NEFA, TC and TG were subsequently measured using enzymatic assays (Jiancheng, A042-1-1, A111-1-1 and A110-1-1, respectively).

Mouse model

The Committee on the Use of Live Animals for Teaching and Research of the University of Hong Kong and the Animal Ethics Committee of Shanghai Sixth People’s Hospital approved the animal experimental procedures. Healthy C57BL/6J male mice were conventionally raised in a specific-pathogen-free barrier facility in individually ventilated cages with a maximum of four mice per cage, with free access to food and water under a strict 12-h light–dark cycle at a controlled temperature (23 ± 2°C) and 60–70% humidity. For FMT experiments, 10-week-old male C57BL/6J antibiotic-treated mice (GemParmatech, N000013) were fed a WD (Research Diet, D12451) for 2 weeks before and during the colonization. For administrating of B. adolescentis to conventionally raised mice, eight-week-old males were similarly fed the WD for 8 weeks before B. adolescentis administration. For the intervention of RS with B. adolescentis in mice housed in a GF environment, 5-week-old male C57BL/6J mice were kept in isolators (GemParmatech, N000295) and fed a C35 diet (45% fat, 20% protein and 35% carbohydrate sourced from 20% RS or 20% CS, and the remaining 80% from maltodextrin) for 10 weeks. Meanwhile, the mice were orally inoculated with B. adolescentis or PBS for the remaining 8 weeks.

FMT experiment

Antibiotic-treated mice with microbiota depletion were given daily oral gavage with a 200-µl antibiotics cocktail (ampicillin 1 g l⁻¹, neomycin 1 g l⁻¹, metronidazole 1 g l⁻¹ and vancomycin 0.5 g l⁻¹) for 7 d, followed by a 4-d antibiotic washout period before FMT. For FMT, the preparation of fresh stool samples and the subsequent operation in mice was conducted as previously described⁴³.

Body weight was measured every 3 or 4 d. Stool samples were collected before and after faecal transplantation and instantly stored at −80 °C until further analysis. Body composition was determined with a Minispec LF90 Body Composition Analyzer (Bruker). In vivo gut permeability, GTTs, ITTs and whole-body oxygen consumption were accessed via a comprehensive laboratory animal monitoring system (Columbus Instruments)⁶⁸. Fat mass in inguinal subcutaneous, epididymal, peri-renal and mesenteric white adipose tissue was determined after death by wet weight measurements. Blood and various tissues were collected for further biochemical evaluations. The study did not blind investigators to group allocations and no mice were excluded.

Bacterial strain

B. adolescentis strain E 298 b (Variant c) (DSMZ no. 20086) was purchased from DSMZ. The culture medium was prepared according to the DSMZ website and B. adolescentis was grown in an anaerobic workstation (Gene Science AG300). B. adolescentis was cultured anaerobically in sterilized DSMZ Medium 58 containing casein peptone (tryptic digest, Sigma-Aldrich), yeast extract (Sigma-Aldrich), meat extract (Sigma-Aldrich), bacto soytone (BD), glucose (Sigma-Aldrich), K₂HPO₄, MgSO₄ × 7 H₂O, MnSO₄ × H₂O, Tween 80, NaCl, cysteine-HCl × H₂O (Sigma-Aldrich), salt solution and 0.025% resazurin (Sigma-Aldrich). The purity of cultures was monitored by Gram staining and the c.f.u. was counted by plating serial dilutions on agar plates.

Administration of B.
adolescentis

B. adolescentis was collected through centrifugation at 1,500g for 30 min at 4 °C, followed by a double washing with sterile PBS and subsequently resuspended in 2 ml anaerobic sterile PBS containing 20% glycerol, achieving of 5 × 10¹¹ c.f.u. per ml. This suspension was then preserved by storage at −80 °C. Before supplementing the drinking water of mice, live B. adolescentis stock was thawed at 4 °C and diluted in 100 ml autoclaved water. On average, each mouse received at least 4 × 10¹⁰ c.f.u. per day. B. adolescentis was heat-killed at 121 °C under pressure of 225 kPa for 15 min. Viability confirmation tests were performed by culturing, showing that heat-killed B. adolescentis did not grow, whereas live B. adolescentis grew well. Drinking water was changed daily in the three groups during the experiment. In addition, GF mice were treated with PBS or 1 × 10⁹ c.f.u. of live B. adolescentis in 200 μl sterile anaerobic PBS by gavage three times per week for 8 weeks.

Biochemical and immunological assays in mice

Serum levels of TC and TG as well as intrahepatic, intestinal and faecal TG were measured by enzymatic analysis (Stanbio Laboratory, 2100430 and 1100430). Serum NEFA was measured by enzymatic analysis (Roche Diagnostics). Serum adiponectin (AIS, HKU, 32010), ANGPTL4 (Abcam, ab210577), inflammatory serum molecules (eBioscience, MCP-1, BMS6005; IL-1β, BMS6002; IL-6, BMS603-2; IL-10, BMS614INST) and FGF21 (Immunodiagnostics Limited, 32180) were measured using ELISA. LPS levels in mesenteric white adipose tissue and serum were measured by LAL assay (Hycult Biotechnology, HIT302).

Metabolomics profiling of mice samples

Amio acid profiling

Target amino acids were quantified using Waters AccQ-Tag derivation kit with 20 μl serum by A Nexera X2 ultra HPLC equipped with a triple quadrupole 8050 (Shimadzu). The details of analytical condition as described in previous study⁶¹.

Bile acid profiling

Target bile acids were analysed by A Vanquish UPLC-Q Exactive (Thermo Fisher Scientific) system. Bile acids in THE serum sample (50 μl) was extracted by 200 μl methanol containing internal standards. Peak detection and integration were carried out by using Trancefinder (Thermo Fisher). The results quantified by the standard curve were used for feature analysis. The details of sample preparation and analytical conditions are as described in a previous study⁶⁹.

SCFA profiling

For sample pretreatment, 30 μl serum was mixed with 60 μl acetonitrile containing the internal standard 0.3 μg ml⁻¹ butyric acid-d7. After vortexing and centrifuging, the supernatant was aspirated for subsequent analysis. For the faecal sample, 40 mg faecal with grinding ball was mixed with 200 μl of acetonitrile containing internal standard extractant, internal standard Butyric-d7 acid 20 μmol l⁻¹. After grinding and centrifugation, the upper supernatant was aspirated for subsequent analysis.

For analytical measurement, an Agilent 5977A instrument equipped with an MSD gas chromatography mass spectrometry (GC–MS) system (Agilent) was used for SCFA analysis. A DB-FFAP column (30 m × 250 μm × 0.25 μm, Agilent) was adopted for GC separation. The initial temperature was 50 °C and maintained for 1 min, then increased to 180 °C at 10 °C min⁻¹, immediately raised to 250 °C at 30 °C min⁻¹ and maintained for 2 min. The flow rate of helium carrier gas was kept at 40 cm s⁻¹. The temperatures of the injector, ion source, quadrupole and interface were 250, 230, 150 and 250 °C, respectively. The selected ion monitoring mode was performed by detecting the protonated molecules at m/z 43 for acetic acid and isobutyric acid, m/z 74 for propionic acid and m/z 60 for other SCFAs. Peak detection and integration were performed using Quantitative Analysis (Agilent). The results quantified by the standard curve are used for feature analysis. All the metabolomic profiles in mice were performed at the CAS Key Laboratory of Separation Science for Analytical Chemistry.

FGF21 response test

After an 8-week B. adolescentis intervention period, an FGF21 response test was conducted in conventionally raised mice. Under general anaesthesia, a subcutaneous fat biopsy was taken and flash-frozen in liquid nitrogen. Subsequently, 2 mg kg⁻¹ recombinant mouse FGF21 or a vehicle was intravenously injected via the inferior vena cava. At 15 min later, a second subcutaneous fat biopsy was taken to assess gene expression and Erk1/2 phosphorylation.

Quantitative PCR assays

The total RNA of tissue samples was extracted by TRIzol (Invitrogen) and reverse transcribed into complementary DNA using ImProm-II reverse transcriptase (Promega). Quantitative PCR with reverse transcription (RT–qPCR) reactions were performed using the Quantifast SYBR Green master mix (QIAGEN) on a Light Cycler 480 system (Roche), normalized with mouse GAPDH. Primers for each gene and B. adolescentis⁷⁰ are listed in Supplementary Table 11.

Histological examination

Ileum sections were treated with ZO-1 or occludin antibodies (Abcam, 1:500 dilution, ab96587; 1:100 dilution, ab216327), followed by incubation with Alex Fluor-596 or FITC-conjugated secondary antibodies (Life Technologies) and counterstaining with DAPI. Ileum sections were treated with ANGPTL4 antibody (Proteintech, 1:500 dilution, 18374-1-AP), followed by incubation with horseradish peroxidase-labelled secondary antibodies (DAKO), developed by DAB following the manufacturer’s recommendations. Five random fields were selected from different regions of individual sections for image quantitation. ImageJ software was used to analyse the intensity of positive staining, represented as a percentage of the total area of the lesion or villi in every field.

Luminal lipase activity

The mixture of luminal content from the small intestine and ice-cold PBS was vortexed thoroughly, followed centrifugation at 15,000g at 4 °C for 10 min. Supernatants were collected and diluted 1:100 in PBS for lipase activity measurement via lipase activity assay kit (Sigma, MAK109) following the manufacturer’s protocol.

Western blotting

WAT proteins were extracted using a lysis buffer containing protease inhibitors (ST505, Beyotime Biotechnology) and phosphatase inhibitors (P1082, Beyotime Biotechnology). Protein samples were resolved by SDS–PAGE (8%) and immunoblotted onto polyvinylidene difluoride membranes. Immunoblotting was conducted using extracellular signal-regulated protein kinase 1/2 (Erk1/2) (Cell Signalling Technology, 1:1,000 dilution, 9102) and phosphorylated Erk1/2 (Thr202/Tyr204) (Cell Signalling Technology, 1:1,000 dilution, 9101). Immunoreactivity was detected using enhanced chemiluminescent autoradiography (Millipore). Band quantification was performed using ImageJ software.

Statistical analysis

The statistical analysis plan for this study is available in the Supplementary Information. EpiData v.3.1 facilitated clinical data entry and documentation. Statistical analyses unless otherwise stated, utilized SPSS v.17 and R v.3.3.2 software. Normally distributed clinical data were presented as mean ± s.d. Non-normally distributed data, verified by the Kolmogorov–Smirnov test, were presented as medians and IQRs. The intervention’s effects on outcomes were analysed using a linear mixed model adjusted for age, sex and intervention order, with Bonferroni adjustment for multiple testing correction. In animal studies, data distribution and variance equality were assessed using a Shapiro–Wilk test and Levene’s test. For two-group comparisons, a two-tailed Student’s unpaired t-test (normally distributed) or nonparametric Wilcoxon rank-sum test (non-normally distributed) was applied. Among multiple groups, a one-way ANOVA (normally distributed) followed by Tukey’s post hoc test or Kruskal–Wallis test (non-normally distributed) followed by Dunn’s test was used. In the clinical study, based on literature and preliminary data, the expected change in body weight and its s.d. after RS intervention was estimated as 2 ± 2.8 kg. A two-tailed test with a 0.05 significance level and 80% power required a minimum sample size of 31. Thirty-seven participants were invited to account for an expected 17% dropout post-randomization. For animal studies, sample sizes were estimated based on previous research and assay variability. Two-sided P values < 0.05 were deemed statistically significant.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.