Animals and Husbandry
Sixteen obese (BCS >7/9), Welsh Mountain (Section A) pony mares (n = 8 Year 1; n = 8 Year 2), were studied over the same 11-week period (December-February; Table 2) in both years. The animals were loaned to the study by local pony breeders within a 30-mile radius of the University of Liverpool’s, Ness Heath Farm. All animals were clinically examined at the point of recruitment, had no signs of overt disease and were considered to be in good overall health. Routine foot care, vaccination and anthelmintic protocols were maintained throughout the study. Animals were individually housed in (3m x 5m) loose boxes within the same barn, bedded on wood shavings and had free access to fresh water at all times. When possible, ponies were given access to pasture for 30 minutes daily to permit basal exercise and social contact. To ensure total control over nutrition, which was provided within the stables, pasture intake while at exercise was prevented using closed grazing muzzles (Shires).
All procedures were conducted in accordance with Home Office (ASPA) requirements, approved by the University of Liverpool’s Animal Welfare Ethics Committee (Home Office project license number: PPL 70/8475). Written informed consent was obtained from all owners. Upon completion of the study, all animals were returned to their owners care.
This 11-week study was designed to allow the evaluation of any associations between weight loss responsiveness and changes in the gastro-intestinal microbiome during a 7-week period of dietary restriction. Animals were maintained on a grass hay diet (same batch across both years) throughout.
Prior to the initiation of negative energy balance, animals first underwent a 4-week period of adaptation to the study diet and husbandry conditions. During this 4-week ‘pre-diet’ period, individual animals were offered grass hay at 2% of their outset BM as DM daily, a plane of nutrition estimated to approximate maintenance in terms of digestible energy (DE) and crude protein (CP) content. To mimic standard husbandry practices, daily hay allowances for each animal were equally divided and offered as two daily meals (08.30h and 16.30h). To ensure adequacy of vitamin and mineral intakes, a moistened nutrient balancer product (Spillers Lite) was fed daily (08.30h, 0.1% BM as DM). Mean compositions for key nutrients were: Hay; Gross energy (GE), 18.9 MJ/kg DM, Ash, 4.0%, Crude Protein (CP) 8.1%, Acid detergent fibre (ADF), 41.2%, Neutral detergent fibre (NDF) 64.7%, starch. 0.6%, water soluble carbohydrates (WSC) 15.6%. Vitamin / mineral balancer; GE 16.35 MJ/kg DM, DE, 9.5 MJ/kg DM, Ash, 14.0%, CP 21.1%, ADF, 14.3%, NDF, 31.7%, starch, 10.5, WSC 12.5%. Samples of hay and balancer were analysed prior to the 4-week pre-diet phase in both years and there was no difference in the composition of key nutrients between years. For the remaining seven weeks of the study, hay intake was restricted to 1% BM as DM daily with 0.1% BM as a nutrient balancer meal. This degree of negative energy balance was predicted to induce a mean weight loss of 1% of outset BM weekly. Meals were offered as previously described. Individual feed provisions were recalculated weekly during this period to account for changes in BM.
Each week, BM was recorded (to nearest 500g) between 08.30h and 09.00h (regularly calibrated weighbridge: Lightweight Intermediate; Horseweigh). BCS was recorded for each animal using the Kohnke (1992) modification of the Henneke et al. (1983) system, by the same individual throughout the duration of the study.
Faecal samples were collected across the final 3 consecutive days of the ‘pre-diet’ period and again during the final three consecutive days of dietary restriction. Samples of faeces were collected from each animal within 5 minutes of spontaneous voiding (first defecation spontaneously voided after 09.00) from the bedding/floor area with a gloved hand. They were sampled into a clean steel bowl to minimise environmental contamination, hand-mixed and aliquoted into four, 5 ml sterile vials (Scientific Laboratory Supplies, UK). Vials were snap-frozen in liquid nitrogen and stored at -80°C prior to DNA extraction.
Estimates of total body composition
Total body water (TBW) and total body fat mass were calculated twice for each individual animal; once during the final week of the ‘pre-diet’ phase and again during the final week of the dietary restriction phase, using the deuterium oxide (D2O) dilution method, as previously described and validated for clinical use in the pony . Deuterium enrichments in plasma samples were analysed in duplicate by a commercial laboratory (Iso-analytical, Cheshire, UK) by gas isotope ratio mass spectrometry.
Combined glucose- insulin tolerance test (CGIT)
A dynamic CGIT test was performed twice for each individual animal; once during the final week of the ‘pre-diet’ phase and again during the final week of the dietary restriction . Blood samples were immediately transferred to lithium heparinised tubes (BD vacutainer), mixed and placed on ice prior to centrifugation (2000g for 10 minutes). Plasma was aliquoted in duplicate and stored at -20°C prior to analysis. Plasma glucose samples were analysed using the hexokinase method. Plasma insulin concentrations were measured using a chemiluminescent assay (Siemens Immulite 1000R, Chemiluminescent Assay).
Digestibility trials were conducted twice for individual ponies by total faecal collection over 3 consecutive days (72 hour) once during the final week of the ‘pre-diet’ phase and again during the final week of the dietary restriction. Any refused feed was recorded at the end of each 24-hour period. Total daily faecal collections were weighed, thoroughly mixed and duplicate samples were collected for analysis. Dry matter (DM) contents of faeces were determined by drying (70°C) duplicate samples (~250g) to constant mass. Ash contents were recorded following combustion of duplicate DM samples at 550°C (Carbolite OAF:1; Carbolite Furnaces). The three dry faecal samples collected over a 72-hour period were ground, pooled and homogenized for the evaluation of GE content (MJ/kgDM) by bomb calorimetry at a commercial laboratory (Sciantec, UK). Additionally, the fibre content (ADF and NDF) of the dried faecal samples was measured by wet chemistry at a commercial laboratory (Dairy One, Ithaca, USA).
Genomic DNA was extracted from freeze-dried faecal samples (25 mg DM) which were bead-beaten in 4% SDS lysis buffer for 45 s. DNA was extracted using a CTAB/Chloroform method (adapted from ). Concentrations and qualities of genomic DNA were assessed by spectrophotometry (Nanodrop ND-100, Thermo Scientific, USA).
Ion Torrent Next Generation Sequencing
Bacterial communities were studied using Next Generation Sequencing (NGS) . For bacterial profiling, amplification of the V1–V2 hypervariable regions of the 16S rRNA was carried out using bacterial primers (27F and 357R) followed by Ion Torrent adaptors. Forward primers were barcoded with 10 nucleotides to allow sample identification. PCR was carried out in a 25 μL reaction vessel containing DNA template (1 μL), 0.2 µL reverse primer, 1 µL forward primer, 5 µL buffer (PCR Biosystems Ltd., London, UK), 0.25 µL bio HiFi polymerase (PCR Biosystems) and 17.6 µL molecular grade water. Amplification conditions for bacteria were 95ºC for 1 min, then 22 cycles of 95ºC for 15 s, 55ºC for 15 s and 72oC for 30 s. To assess the quality of amplifications, resultant amplicons were visualized on a 1% agarose gel. PCR products were then purified using Agencourt AMpure XP beads (Beckman Coulter Inc., Fullerton, USA) and DNA concentration was determined using an Epoch Microplate Spectrophotometer fitted with a Take 3 Micro-Volume plate (BioTek, Potton, UK) to enable equimolar pooling of samples with unique barcodes.
Libraries were further purified using the EGel system with 2% agarose gel (Life Technologies Ltd., Paisley, UK). Purified libraries were assessed for quality and quantified on an Agilent 2100 Bioanalyzer with High Sensitivity DNA chip (Agilent Technologies Ltd., Stockport, UK). Library preparation for NGS sequencing was carried out using the Ion Chef system (Life Technologies UK Ltd) and the Ion PGM HiQ Chef kit, and sequencing using the Ion Torrent Personal Genome Machine (PGM) system on an Ion PGM Sequencing 316 Chips v2 BC.
Following sequencing, data were processed as previously described . Briefly, sample identification numbers were assigned to multiplexed reads using the MOTHUR software environment. Data were denoised by removing low-quality sequences, sequencing errors and chimeras (quality parameters: maximum 10 homopolymers, qaverage 13, qwindow 25, for archaea the qwindow was set at 30, and erate = 1; Chimera check, both de novo and database driven using Uchime). Sequences were clustered into OTUSs using the Uparse pipeline at 97% identity. Bacterial taxonomic information on 16S rRNA sequences was obtained by comparing against Ribosomal Database Project-II . The number of reads per sample were normalised to the sample with the lowest number of sequences.
Volatile Fatty Acid Measurement
Faecal samples were defrosted and diluted 1:5 w/v with distilled H2O (2 g sample/8 mL H2O) to make a faecal slurry and pH was recorded prior to the addition of 20% orthophosphoric acid (containing 20 mM 2-ethyl butyric acid as an internal standard) also at 1:5 (1 mL acid/4 mL faecal slurry) to deproteinise the samples. For VFA analyses, slurries were left for 24 h to allow sediments to settle before being syringe-filtered through a glass-fibre pre-filter (0.7 µm pore, Millipore) and a nitrocellulose membrane (0.45 µm pore; Millipore) into a glass vial and capped. VFAs were determined by gas liquid chromatography using ethyl butyric acid as the internal standard as described by (Stewart and Duncan, 1985).
All data was inputted into Excel and exported for statistical analysis into STATA 13.1 (Statacorp, Texas). The pattern of weight change during the weight-loss phase of the study was investigated by fitting a mixed effects regression model with body mass (kg) as the outcome variable. Outset body mass (Week 0) was included to account for differences between individual animals in outset BM. Time (week) was added as an explanatory variable and polynomial terms were fitted if they improved model fit as judged by the likelihood ratio test. Pony identity was included as a random intercept with time as a random slope. An unstructured covariance matrix was employed for the random effects. Body mass was predicted from the model and using this, a proportional weight loss (adjusted to Week 0) was calculated and used to rank animals according to weight-loss. This was then divided into terciles (high, mid and low weight-loss) to enable further interrogation of the data. Wilcoxon sign-rank test was employed to evaluate differences in digestibility, body composition and glucose/insulin dynamics between pre- and post-diet measurements.
Diversity indices (Inverse Simpson’s and Shannon-Wiener), species observed (S.Obs) and estimated species richness (S.Chao1) were calculated for all faecal samples at all time points using normalized data as recommended to reduce over-inflation of true diversity in pyrosequencing data sets. Data were assessed for normality visually adopting the “ladder of powers” approach. Transformations were performed where appropriate. Student t tests were performed to assess mean changes in VFA concentrations and diversity indices between the pre-diet and post-diet period (using arithmetic mean of final 3 days of pre-diet and post-diet samples). One-way ANOVA (Genstat® 12th edition; VSN International ltd.) was employed to assess differences in pre-diet diversity and faecal VFA concentrations between weight-loss groups (high, mid, low), using Bonferroni-adjusted P values to correct for multiple testing. P-values were considered significant if < 0.05. Findings were confirmed by univariate analysis (STATA 13.1, Statcorp, Texas) whereby pre-diet diversity or faecal VFA concentration was the explanatory variable and overall proportional weight-loss (logit transformed) was the outcome variable. In order to investigate further the ability of pre-diet acetate concentration to predict the weight-loss success of an animal, a binary (yes/no) variable was generated whereby weight-loss success was defined as total percentage weight-loss of ≥8%. This value was selected on the basis of data collected within the current study. A receiver operating characteristic (ROC) curve was generated for pre-diet acetate concentration as a predictor of weight-loss (yes/no). This allowed investigation of the optimal cut-off score for pre-diet acetate concentration and generated sensitivity and specificity parameters. To investigate associations between pre-diet acetate concentration and the pre-diet relative abundance of individual bacterial genera, concentrations of pre-diet acetate (mean of 3 pre-diet days) were divided into quartiles as follows: Quartile 1, n = 4, 9.45mM ± 0.27 (mean ± SD); Quartile 2, n = 4, 11.66mM ± 1.04, Quartile 3, n = 3, 16.76mM ± 1.13, Quartile 4, n = 4, 22.63mM ± 4.49.
Phylum and genera level differences in the microbiome following weight-loss and between weight-loss groups were investigated by ANOVA (Genstat® 12th edition; VSN International ltd.). Statistical analyses excluded those for which an abundance of less than 0.05% was recorded. P values were adjusted for multiple testing using the method proposed by Benjamini and Hochberg  to decrease the false discovery rate. Findings with P < 0.10 when applying Benjamini and Hochberg (1995) correction were regarded as statistically significant.
Differential abundances at an OTU level were evaluated using the bioconductor package DESEQ21 in the statistical package R, a methodology appropriate for the interrogation of high-throughput, sequencing count data, allowing models to be built using a negative binomial distribution to account for the distribution of read counts from each OTU .
Permutation multivariate analysis of variance (PERMANOVA) was used to determine overall significant differences in bacterial communities. Analyses were performed in PRIMER 6 & PERMANOVA+ (versions 6.1.18 and 1.0.8 respectively; Primer-E, Ivybridge, UK). Abundance percentage data were subjected to square-root transformation and Bray-Curtis distance matrices were calculated. PERMANOVA was carried out using default settings with 9999 unrestricted permutations and the Monte Carlo P value was calculated. Analyses of Similarity (ANOSIM) were conducted in PRIMER 6 & PERMANOVA+, using the Bray-Curtis distance matrix calculated above. This analysis was used to provide a metric of the degree of divergence between communities, as given by the R statistic.
To calculate the contribution of environmental data on bacterial communities, distance-based linear modelling was used to calculate which environmental variables had a significant correlation with the community data. Significant variables were used in distance-based redundancy analysis (dbRDA)  as implemented in PRIMER 6 and PERMANOVA+.