Ethics approval and participants consent
This study complied with all relevant ethical regulations and was conducted in accordance with the principles of the Declaration of Helsinki and its subsequent amendments, were approved by the ethics committee of the Medical Faculty, University of Bonn (Approval Number: 212/20), and prospectively registered at the German Clinical Trials Register (DRKS; http://www.drks.de) under identifier DRKS00022169. Written informed consent was obtained from all the participants.
Study design
This study was a 1:1 randomized, controlled, prospective dietary intervention with a parallel design conducted between September 2020 and June 2021 at the Department of Nutrition and Microbiota, University of Bonn, Germany.
Potential subjects completed an initial telephone pre-screening, followed by a screening visit with anthropometric measurements and a detailed blood analysis to confirm eligibility, during which they were informed of all study procedures and requirements. Eligible participants were invited for two clinical visits, the first before and the second after the two-day intervention period. After the first visit, participants were randomly assigned to the experimental group (OG) or the CG using computer-generated randomization tables in a block format with variable block length generated by a researcher not clinically involved in the study. These tables were concealed from the researchers and participants until the interventions were assigned by study personnel not involved in final data analysis. Blinding after allocation to the interventions was not feasible. At each visit, anthropometric data, fecal samples, and fasting and postprandial blood samples were collected during an oral glucose tolerance test (OGTT; Fig. 1a, Supplementary Methods). Office blood pressure and resting energy expenditure were measured using a semiautomatic blood pressure measurement device (Boso Carat Professional, Bosch + Son GmbH and Co. KG, Juningen, Germany) and indirect calorimetry, respectively (Supplementary Methods). Detailed nutritional and lifestyle assessment were conducted during the first visit (Supplementary Methods). Interim contact with the study coordinator (via telephone or e-mail) was made available to all participants on each intervention day.
Participants
Participants were recruited through newspaper advertisements, flyers, and social media. The study subjects included adults with overweight or obesity (BMI 27 - 39.9 kg/m2) and MetS aged 45 - 70 years. MetS was primarily diagnosed based on the global consensus definition of the International Diabetes Federation1 as central obesity (waist circumference ≥ 94 cm in men and ≥ 80 cm in women for Europeans) coupled with two of the following five criteria: (1) elevated blood pressure (≥120 mmHg systolic and/or ≥80 mmHg diastolic)58; (2) elevated fasting serum triglycerides (≥150 mg/dl); (3) decreased fasting serum HDL-C (<40 mg/dl for men and <50 mg/dl for women); (4) elevated fasting plasma glucose (≥ 100 mg/dl); (5) indication of insulin resistance (HOMA-IR index > 2.5) (adapted to the WHO definition59). The latter criterion was added after the initial study registration in order to attach more importance to impaired glucose metabolism as a characteristic of MetS and to counteract recruitment difficulties due to the coronavirus pandemic (COVID-19). In addition, the subjects' habitual diet corresponded to a Western dietary pattern without (regular) oat consumption. Exclusion criteria were as follows: (1) diagnosed Diabetes mellitus, chronic kidney, liver, inflammatory, or digestive tract diseases, non-medicated thyroid diseases, past cardiovascular events, acute illnesses or recent surgeries; (2) chronic use of medication that influences glucose metabolism; (3) antibiotics treatment within three months prior to study inclusion; (4) pregnancy or lactation; (5) intolerance to oat; (6) history of smoking or alcoholism; (7) chronic intake of dietary-supplements; (8) vegetarian or vegan diet; (9) planned lifestyle changes.
Dietary intervention
The participants followed a two-day oat diet or a macronutrient-adapted control diet, both hypocaloric and high in fiber (1100-1200 kcal/d; carbohydrates (>65 E%, of which fiber > 15%, β-glucan: 13.5 g vs. 0 g), protein (15 E%), and fat (17 E%)), depending on their allocation (OG vs. CG). The nutrient composition of the test meals was calculated using the computer-based nutrient calculation program EBISpro, based on the German nutrient database Bundeslebensmittelschlüssel, version 2016 (Max Rubner-Institut, Karlsruhe, Germany). Meals were prepared independently by the subjects at home using detailed instructions. Participants in the OG consumed three oat meals per day, each comprising 100 g of rolled oat flakes (Demeterhof Schwab GmbH & Co. KG, Windsbach, Germany) and boiled in water. Small amounts of fruits (apples, pears, or berries) and vegetables (spinach or leeks) were used as additives. No salt, sugar, or sweeteners were added. Participants in the control group consumed two meals per day comprising bread and raw vegetables (breakfast and dinner) and one warm meal per day (lunch). There was a time interval of four hours between meals. During this time, participants were required to consume only unsweetened drinks. For standardization, the participants received a recommendation for dinner the evening before the start of the intervention (carbohydrate-rich bread meal with raw vegetables). In addition, participants were instructed to avoid any other oat products (other than the oat flakes provided for the OG) for at least two weeks before and during the study.
Compliance
Adherence to the treatment protocol was assessed using two independent criteria, both of which had to be fulfilled to be considered compliant. On the one hand, the concentration of the oat-specific biomarker avenanthramides (AVAs)60,61 was quantified as an objective blood marker. Plasma AVAs concentration was determined using LC-MS/MS by Metabolon Inc. (Morrisville) (Metabolon Method TAM223: “LC-MS/MS Quantitation of Dihydroferulic Acid and Three Avenanthramide Compounds in Human Plasma”) based on a previously described method61. Compliance was assumed if the AVAs concentration was quantifiable only after oat diet. On the other hand, the subjects received the exact number of oat packages required for the two intervention days and were asked to return all empty and unemptied packages on the second visit. Compliance was measured by comparing the number of packages provided and returned, as well as by detailed checklists filled out by the participants on each intervention day. Participants were considered compliant when they consumed at least five of six packages provided (≥ 500 g oats) or at least five of six test meals according to instructions (checklists).
Anthropometrics
Anthropometric measurements were performed following a previously published standard operative procedure62. Briefly, body weight was measured in light clothing with an empty bladder, using electronic column scales with an accuracy of up to 100 g (seca scale 704, seca GmbH and Co. KG, Hamburg, Germany). Body height was determined to the nearest of 0.1 cm using a stadiometer (seca scale 704, seca GmbH and Co. KG, Hamburg, Germany). The BMI was calculated using the following formula: BMI = weight [kg]/(height [m])2. Waist circumference was measured midway between the lowest rib and iliac crest at maximal exhalation to the nearest of 0.1 cm in duplicates. Body composition (fat mass and fat-free mass) was determined by air-displacement plethysmography using a BOD-POD body composition system (Cosmed, Firdolfingen, Germany).
Collection of blood and stool samples
Fasting blood samples were collected from participants in the morning between 8:00 a.m. and 10:00 a.m. after a 12-h overnight fast. Postprandial blood samples were collected during a 3-h OGTT (Supplementary Methods). All samples were taken under standardized conditions using tubes containing EDTA, fluoride, or a coagulation activator (S-Monovette, Sarstedt, Germany). Plasma and serum supernatants were obtained by centrifugation at 3000 × g for 15 min at 8 °C after complete coagulation (only serum) and immediately frozen in cryovials at −80 °C until further analysis. Fecal samples were collected within 24 h before each clinic visit according to a standard operation procedure and immediately stored at -80 °C until further analysis63.
Routine laboratory blood analyses
Routine laboratory analyses of serum included fasting and postprandial triglycerides, TC, LDL-C, HDL-C, clinical biochemistry, high-sensitivity C-reactive protein (hsCRP), and fasting and postprandial insulin, while routine laboratory analyses of plasma included fasting and postprandial glucose, HbA1c, and hematology parameters. All parameters were measured in a certified medical laboratory (Central Laboratory of the Institute of Clinical Chemistry and Clinical Pharmacology at the University Hospital Bonn, Germany) within 4 h of blood sampling under standardized conditions using the Roche/Hitachi Cobas c system (Roche Diagnostics, Mannheim, Germany). Methods specifications are available online (https://www.ukbonn.de/ikckp/zentrallabor/ leistungsverzeichnis/). The HOMA index was calculated as follows: HOMA-IR = [insulin (mU/L) × glucose (mg/dL)] ÷ 40564. Insulin resistance was assumed if the HOMA index was > 2.565.
Analysis of non-esterified fatty acids
Serum non-esterified fatty acid (NEFA) concentrations were analyzed using an in-vitro enzymatic colorimetric method assay (NEFA-HR(2), Wako Diagnostics, Mountain View, CA, USA) following the manufacturer’s instructions and the recommended quality control procedure (inter-assay/intra-assay variability 5.4%/4.8%) at the Institute of Nutrition and Food Sciences, University Bonn, Germany, as previously described66.
Targeted plasma metabolomic profile
Plasma dihydro ferulic acid concentration and ferulic acid values (peak areas) were generated by Metabolon Inc. (Morrisville) using LC-MS/MS according to Metabolon Method TAM223 (“LC-MS/MS Quantitation of Dihydroferulic Acid and Three Avenanthramide Compounds in Human Plasma”) based on a previously described method18.
Global metabolomic profiling
For both fasting plasma and fecal samples, non-targeted global metabolomic profiles were generated by Metabolon Inc. (Research Triangle) using UPLC-MS/MS, as previously described26. The metabolomic dataset included a total of 1149 and 1082 metabolites in the plasma and feces, respectively, comprising amino-acids, peptides, carbohydrates, energy intermediates, lipids, nucleotides, cofactors and vitamins, xenobiotics, and partially characterized molecules. These include metabolites of an established microbial origin67. After removing drug-associated metabolites, 1128 plasma metabolites and 1054 fecal metabolites were included in the analyses.
Gut microbiome sample processing
Total genomic DNA was extracted from 120 mg fecal material using ZR BashingBead lysis tubes (0.1 and 0.5 mm, Zymo Research, Freiburg, Germany) in combination with the Chemagic DNA stool kit (Perkin Elmer, Rodgau, Germany) following the manufacturer's instructions68–70. After the addition of lysis buffer, mechanical lysis was performed using a Precellys 24 tissue homogeniser (Bertin Instruments, Frankfurt am Main, Germany). After extraction, the DNA was stored at -20 °C until further analysis. High-throughput 16S rRNA amplicon sequencing of the fecal microbiome was performed at Life & Brain GmbH (Bonn, Germany). Briefly, the V3/V4 region of the 16S rRNA gene was amplified using the Bakt_341F and Bakt_805R primer combination. Details of the library preparation have been described previously71. The final pool was quantified using the Qubit dsDNA HS assay kit (Thermo Fisher Scientific, Waltham, MA, USA), and the fragment size was checked using a D1000 ScreenTape (Agilent, Satna Clara, CA, USA).
Sequencing was performed on an Illumina MiSeq system using the MiSeq reagent kit v3 with 2 × 300 cycles. Clustering was performed at 8 pM with a 20 % spike-in of PhiX. Demultiplexing was performed using the MiSeq system. The 16S rRNA sequencing data were processed using QIIME 2 version 2021.472. Sequence quality control and denoising were performed using DADA273. The QC step included the filtering of PhiX reads and chimeric sequences. After denoising, sequences were classified using the SILVA databases to identify amplicon sequencing variants (ASVs) for sequences with > 99% sequence similarity.
Gut microbiome analysis
Diversity analyses were performed in QIIME2 based on a rarefied table with a sampling depth of 28,767 sequences and used for subsequent statistical analysis. The alpha diversity metrics included Shannon entropy, Pielou’s evenness, and Faith-PD. A linear mixed model (LMM) was applied to assess the differences between the two diet groups (n = 31). The model was defined as diversity matrix ~ time*group + age + sex + BMI + (1|Person-ID). The Person-ID was used as random effect to account for repeating measurements. Beta diversity was calculated based on the Jaccard distance, Bray-Curtis dissimilarity, and UniFrac distance matrices. LMM was applied using the CG as the reference (formula: diversity matrix ~ group + age + sex + BMI; n = 28).
At the feature level, regression analysis of gut bacterial taxa was performed using a negative binomial and zero-inflated mixed model (NBZIMM)74 and linear models for differential abundance analysis (LinDA)75, in addition to sPLS-DA. The analyses were performed at genus level with a relative abundance threshold of 0.01%. Data were normalized using a centered-log-ratio (CLR) transformation with an offset of 1 prior to all analyses. Twenty-eight subjects were included in the analyses (n = 4 dropouts, because only one of the two fecal samples was collected) (Supplementary Methods).
The Phylogenetic Investigation of Communities by Reconstruction of Unobserved States 2 (PICRUSt2, version 2.5.1)25 was used to predict the functional potential of the fecal microbiota. Predicted gene products in each bacterium were annotated using the KEGG database and classified into KEGG orthologous (KO) groups24, resulting in the identification of 6,616 KO across all samples (n = 28). The ko2kegg_abundance function of the ggpicrust2 package76 was used for this task. Subsequently, sparse partial least squares-discriminant analysis (sPLS-DA; details see next chapter) was performed to identify differences in microbial function between the diet groups. Additionally, a single-feature analysis was performed using LMM74 with the CG and baseline value as reference. The model was defined as pathway ~ time*group + age + sex + BMI + (1|Person ID). The Person-ID was used as random effect to take repeated sampling of the same person`s microbiota into account. The analysis was performed on CLR-transformed data.
Sparse partial least squares-discriminant analysis
To identify the key features that best differentiate the two intervention groups in terms of diet-induced changes in clinical markers, gut microbiota composition and function, and global metabolomic profiles in plasma and feces, sPLS-DA was performed. sPLS-DA, implemented in the mixOmics v6.8.5 R package, enables the selection of the most discriminative features in the data to classify the samples by projecting the data into a lower dimensional space in a supervised manner19,20. This approach is optimal if the number of features is very high compared to the sample size, as in this study. To account for the repeated-measures design of the intervention study, the logarithmic fold change (log fold change) was used as the input to the models based on Lee et al.77. The performance of the sPLS-DA models was assessed on the BER and the AUC. Detailed information on the workflow can be found in Supplementary Methods.
Integrative Analysis for Biomarker discovery using Latent cOmponents
To determine the relationship between diet-induced changes in metabolic parameters, gut microbiota, and targeted and global metabolomic profiles, DIABLO, an integrative multi-omics method that aims to identify common information across different types of data while distinguishing between different phenotypes30 was applied. Two different DIABLO evaluations (“models”) were performed. For model 1, clinical marker, metabolomic profiles (targeted, global plasma, and fecal), and microbial composition were used as inputs. Model 2 considered microbial function (KEGG pathways) instead of microbial composition data. The logarithmic fold changes were used as inputs for the models77. Detailed information on the workflow can be found in Supplementary Methods.
Statistics and sample size calculation
Statistical analyses were performed using SPSS (version 29.0; IBM Crop., Chicago, IL, USA), R (version 3.6.2; Boston, MA, USA) and Python (version 3.10). The figures were created using GraphPad Prism (version 10; GraphPad Software, San Diego, CA, USA), R, and Python. For all analyses, a two- tailed significance level was set at P < 0.05. For metabolic data analysis, family-wise error rate correction was applied via the block-wise Bonferroni-Holm method78 to correct for multiple testing (Padj. < 0.05). For microbiome and metabolomic data analysis, FDR was applied using Benjamini/Hochberg method79 (q < 0.05).
The distribution of the clinical marker was evaluated using Shapiro Wilk test and visual inspection. If indicated, the data were log10-transformed prior to statistical analysis. Postprandial results were described as the AUC for the 3-h OGTT, calculated using the trapezoidal method and considering complete data sets. Metabolomic data were transformed with the natural logarithm prior to all statistical analyses and expressed as log fold change. DHFA concentrations below the limit of quantitation but above the peak area of the background DHFA peak in the blank samples or concentrations above the limit of quantitation were extrapolated. Missing values in the global metabolomic profiles were imputed with the minimum observed value for each compound.
To assess whether the data provided evidence of the superiority of the oat diet over the control diet, linear regression models were applied adjusted for the baseline value of the end-point and the confounding factors BMI, age, and sex. The CG was used as reference for the diet group. Beta estimates of intervention (β) with 95 % CI were reported to assess the magnitude of the effect size (Supplementary Methods). To determine the changes over time within each diet group, a paired student’s t-test (or Wilcoxon rank test) was performed depending on the data distribution. Pairwise correlations between changes in LDL-C and TC and other variables (clinical data, microbiome, metabolomic profiles) were assessed using Pearson's correlation coefficient and Spearman's rank correlation coefficient. The logarithmic fold change was used and only variables with at least five data points were considered.
Sample size (n = 17 participants per group) was calculated based on data from a previous intervention study that successfully assessed the effect of whole grain intake on blood DHFA concentration80, expecting a 2-fold change in the plasma DHFA concentration between the two diet groups using a two-sided t-test at a 5 % significance level and with 95 % power.