Dietary carbohydrate rather than protein intake drives colonic microbial fermentation during weight loss

Overweight or obese healthy, male subjects with a BMI > 27 kg/m², stable body weight (< 2 kg change in the past 3 months) and fasted blood glucose < 6 mmol/L were recruited into the study. Health status was confirmed by the subject’s General Practitioner and medical exclusion criteria were: Diabetes, Kidney disease, Hepatic disease, Gout, Food allergy, Psychiatric disorders, Severe gastrointestinal disorders, Thromboembolic or Coagulation disease, Alcohol or Substance abuse, Eating disorders and Unregulated thyroid disease. The average baseline characteristics of all 18 subjects were: age 49 ± 12 years (range 21–70), BMI 36.6 ± 5.8 (range 26.5–51.7) and basic metabolic rate (BMR) 9.0 ± 1.6 MJ/day (range 6.7–12.7). BMR was measured in each fasted volunteer and used to calculate the individual energy requirements. The study is focussed on men to minimize variability as the number of participants was restricted by the complexity of the dietary design. Further details on the study volunteers and study design can be found in [18]. Medication exclusion criteria were: Orlistat, Rimonabant, Sibutramine, Oral antidiabetics including insulin, Digoxin, Anti-arrhythmics, Antidepressants, Anticoagulants and no antibiotics or drugs known to influence the faecal microbiota were taken immediately prior to or during the course of the study. The drugs reported by at least one subject on the study include: Amlodipine, Aspirin, Atenolol, Becotide (inhalers), Bendroflumethiazide, Diclofenac, Diclofex, Enalipril, Felodipine, Ibuprofen, Liptor, Statin, Tamsulosin hydrochloride, Thyroxine, Tramadol, Trazodone, Valstatin and Ventolin (inhalers). Ethical approval was granted by the North of Scotland Research Ethics Committee and all subjects provided informed signed consent. All volunteers were asked not to change their lifestyle for the duration of the study and there were no drop-outs during the study.

There were four periods of dietary intervention, with all food supplied in cooked (breakfast) or ready-to-eat form (lunch and dinner). All food was prepared in-house and weighed by the kitchen research staff at the Human Nutrition Unit (HNU). Any leftover food that was not consumed by study subjects was collected and weighed to the nearest 0.1 g. All subjects (n = 18) first received a body weight maintenance diet (M, fed to 1.5× Basic Metabolic Rate, BMR) for 1 week before being randomised for consumption of three weight loss diets for 10 days each in a cross-over design. The weight loss diets were: a normal protein weight loss diet (NPWL, 15% of energy from protein, 55% from carbohydrate), a normal protein weight loss diet enriched with free amino acids and moderate amounts of carbohydrate (NPAAWL, 15% of energy from protein, 15% from free AA, 40% from carbohydrate) or a high protein weight loss diet containing moderate amounts of carbohydrate (HPWL, 30% of energy from protein, 40% from carbohydrate). The free amino acid profile of the NPAA diet matched that of beef and chicken [18]. All weight loss diets were fed at 1× BMR as three meals per day and prepared on a 5 days rotation menu. All subjects consumed all three weight loss diets in a random order for 10 days. No washout period was observed between the different diets as the faecal microbiome and metabolome have been shown to adapt to dietary changes within 1–2 days [19].

Using the HNU kitchen record and the subjects’ food diaries, daily nutrient intakes for each subject were determined by trained staff using the WinDiets Nutritional Analysis Software Suite version 1.0 (The Robert Gordon University, Aberdeen, UK), a computerised version of McCance and Widdowson’s the composition of foods [20]. These individual daily intakes of macronutrients (carbohydrate, fat and protein), dietary fibre, non-starch polysaccharides, sugar, starch and total energy were then collated for analysis of statistical comparisons between the various diets (summarised in Table 1). For more detailed analysis of nutrient intake, exact food and nutrient intakes 4 days prior to each faecal sample collection were analysed and averages of 4 days were then correlated with faecal metabolites. Specific intakes of individual food items (e.g. meats) were extracted from the recipe sheets and the protein content of meat products was estimated as 25% of meat weight, based on a weighted average of the ingredients from the McCance and Widdowson’s food composition tables for cooked products [20]. The food group of red meat contained fresh and processed products of beef (steak, mince, topside, rump steak, burgers, sausages) and pork (roast, fillet, loin chops, sausages, ham, bacon). White meat and fish included chicken (breast, nuggets, sausages), turkey, tuna (canned in brine), prawns and haddock (in batter). The other main sources of protein were dairy products and eggs. Dietary fibre (expressed as non-starch polysaccharides) derived mainly from wholegrain bread and pasta, legumes and vegetables.

Short chain fatty acids were determined using capillary gas chromatography [21]. The lower limit of reliable detection of each product was 0.2 mmol/L. Ammonia concentrations were analysed in 30-fold diluted faecal waters following reaction with sodium phenate and sodium hypochlorite by measuring absorbance of the indophenol product at 625 nm [22].

Faecal water samples were diluted 10-fold with distilled water and the content of N-nitroso compounds analysed by measuring the chemical release of nitric oxide detected on a thermal energy analyser. Concentrations were calculated by comparing the thermal response of a faecal water sample to the response of an N-nitrosodipropylamine standard (16.6 µg/mL) and values were expressed as apparent total N-nitroso compounds (ATNC) ng/mL faecal water sample.

Faecal water samples were thawed to 4 °C and 80 µl transferred to an Eppendorf tube. Internal standard 1 (¹³C-labelled benzoic acid; 80 µL), internal standard 2 (4,7,8-tri-MelQx; 80 µL) and methanol (160 µL) were added. The samples were mixed, centrifuged (12,500×g; 5 min; 4 °C) and the supernatant analysed by LC–MS. Liquid chromatography separation of the metabolites produced was performed on an Agilent 1100 HPLC system (Agilent Technologies, Wokingham, UK) using a Zorbax Eclipse 5 µm, 150 mm × 4 mm column (Agilent Technologies). Three different gradients were used to separate the different categories of metabolites and the mobile phase solvents in each case were water containing 0.1% acetic acid and acetonitrile containing 0.1% acetic acid. In all cases the flow rate was 300 µL/min with an injection volume of 5 µL. The LC eluent was directed into, without splitting, an ABI 3200 triple quadrupole mass spectrometer (Applied Biosystems, Warrington, UK) fitted with a Turbo Ion Spray™ (TIS) source. For LC methods 1 and 2, the mass spectrometer was run in negative ion mode with the following source settings: ion spray voltage − 4500 V, source temperature 400 °C, Gases 1 and 2 set at 15 and 40, respectively, and the Curtain Gas set to 10. For LC method 3, the mass spectrometer was run in positive ion mode with the following source settings; ion spray voltage 5500, source temperature 400 °C, Gases 1 and 2 set at 14 and 40, respectively, and the Curtain Gas set at 10. All the metabolites were quantified using multiple reaction monitoring. Standard solutions (10 ng/µL) for all analytes were prepared and pumped directly via a syringe pump. The ion transitions for each of the analytes were determined based upon their molecular ion and a strong fragment ion. For several categories of compounds, it was inevitable that their molecular ion and fragment ion would be the same, but this was overcome by their differing elution times. Their voltage parameters, declustering potential, collision energy and cell entrance/exit potentials were optimised individually for each analyte.

Power calculations indicated that with 18 volunteers a treatment effect of at least 20% could be detected with a power of 90% at the 5% significance level, assuming a within-volunteer spread of 25%.

Data were analysed by analysis of variance with volunteer as a random effect and diet as a fixed effect. Initially, the effect of order and its interaction with diet were included in the analyses but these were found to be non-significant and were therefore omitted from subsequent analyses. When the effect of diet was significant (P < 0.05), means were compared with post hoc t test. When the assumption of constant variance (and normality) was not met, data were log-transformed before analysis. This was the case for the phytochemicals, lactate, lactate%, ATNC and other protein metabolites. Zeros were replaced 0.5× the lowest non-zero value. For log-transformed data, the backtransformed (geometric) means and their corresponding 95% confidence intervals are presented. When more than 15% (more than ten samples out of 72 samples) of the samples were zero, non-parametric Friedman test was conducted instead. When the effect of diet was significant (P < 0.05), means were compared with Wilcoxon signed rank test. These data are presented as their mean and the number of zero samples out of the total samples tested. Linear relationships between variables were investigated with random effects linear regression, where volunteer was regarded as random. Log-transformation was applied to the same set as variables as mentioned above and only metabolites for which less than 15% of the samples were zero were included. All data were analysed with Genstat 17th Edn Release 17.1 (VSN International Ltd, Hemel Hempstead, Herts, UK). A P value < 0.05 was regarded as statistically significant.