Changes in body mass and body composition
All animals remained healthy throughout the study. One animal lost a considerably greater amount of BM compared to the other animals (3.68% of outset BM) during the ‘pre-diet’ phase of the study (Week -4 to 0). During this time, individual animals were fed 2% BM as hay DM daily, designed to approximate maintenance levels. Therefore, due to suspected underlying dental issues, data from this animal were removed from all analysis. The remaining 15 animals generally held a constant BM during the ‘pre-diet’ phase of the study (overall mean gain of 0.11% ± 1.93).
The rate of weight-loss during the dietary restriction phase of the study (Week 0 to 7; animals fed 1% BM as hay DM daily) was greatest during the first week (overall mean loss 2.82% ± 1.64). During the remaining 6 weeks, animals lost an average of 1.01% ± 0.21 Week 1 BM weekly, leading to overall losses of 6.04% ± 1.23 over the 6 weeks.
The final regression model (Table 1) for investigating changes in body mass over the weight-loss phase of the study included week as a cubed polynomial term. Predicted marginal mean bodyweight was estimated from the regression model and is presented graphically (Figure 1A) and the predicted cumulative proportional weight-loss (adjusted to Week 0) were plotted for individual animals (Figure 1B). Using the predicted model outputs, animals were ranked according to the degree of weight-loss achieved (Table 2).
Table 1. Change in body mass of obese ponies during 7-weeks of dietary restriction.
Explanatory variable
|
Coefficient
|
95% CI
|
P value
|
Week
|
-9.78
|
-11.31 to -8.25
|
< 0.01
|
Week2
|
1.56
|
1.03 to 2.09
|
< 0.01
|
Week3
|
-1.00
|
-0.15 to -0.05
|
< 0.01
|
Start weight
|
1.00
|
0.96 to 1.03
|
< 0.01
|
Baseline
|
0.69
|
-10.36 to 11.75
|
0.90
|
Random effects parameter
|
|
|
|
Pony ID: Unstructured
|
|
|
|
Variance (week)
|
0.25
|
0.09 to 0.70
|
|
Variance (baseline)
|
4.32
|
1.47 to 12.67
|
|
Covariance (week, baseline)
|
1.03
|
0.24 to 1.82
|
|
Variance (residual)
|
5.63
|
4.29 to 7.39
|
|
A mixed-effects linear regression model was built (random effects: Pony ID, random slope: week) with body mass (BM) as the outcome variable and week as a cubic polynomial as the explanatory variable. An unstructured covariance matrix was employed for the random effects. Coefficients are presented ± 95% confidence intervals (CI’s).
Table 2. Outset animal phenotype including ranking of animals by weight-loss (WL).
Ranking
|
Pony ID
|
Year
|
% WL
|
Low/mid/high WL
|
Age (years)
|
BM (kg)
|
BCS (/9)
|
Body fat (%)
|
1
|
10
|
2
|
11.59
|
high
|
9
|
244
|
7.7
|
17.19
|
2
|
16
|
2
|
11.56
|
high
|
7
|
282
|
7.6
|
12.30
|
3
|
11
|
2
|
11.32
|
high
|
6
|
260
|
7.6
|
14.60
|
4
|
14
|
2
|
9.17
|
high
|
10
|
310
|
8.2
|
21.59
|
5
|
15
|
2
|
8.77
|
high
|
5
|
315
|
8.1
|
20.54
|
6
|
12
|
2
|
8.62
|
mid
|
12
|
351
|
8.2
|
24.64
|
7
|
8
|
1
|
8.45
|
mid
|
23
|
279
|
7.1
|
20.02
|
8
|
7
|
1
|
8.42
|
mid
|
20
|
261
|
7.3
|
22.03
|
9
|
13
|
2
|
8.15
|
mid
|
13
|
353
|
8.4
|
20.67
|
10
|
2
|
1
|
7.97
|
mid
|
11
|
283
|
8.1
|
27.32
|
11
|
4
|
1
|
7.94
|
low
|
10
|
335
|
7.8
|
16.95
|
12
|
5
|
1
|
7.94
|
low
|
11
|
347
|
8.3
|
21.65
|
13
|
1
|
1
|
7.53
|
low
|
10
|
318
|
8
|
24.61
|
14
|
3
|
1
|
7.22
|
low
|
11
|
291
|
8.2
|
23.33
|
15
|
6
|
1
|
7.11
|
low
|
8
|
289
|
7.3
|
27.78
|
Body mass was predicted from the mixed-effects model and using this, a cumulative proportional weight loss (adjusted to Week 0) was calculated and used to rank animals according to weight-loss.
The proportion of BM occupied by fat mass remained similar before and after dietary-restriction (21.01% ± 4.38 vs. 21.22% ± 4.40; Table 3), however due to reductions in overall BM, fat mass was correspondingly significantly reduced (63.75kg ± 16.13 vs. 58.92kg ± 14.74; p < 0.01). When losses of lean and fat mass were evaluated as a proportion of outset lean/fat masses, reductions in these compartments were similar (7.32% ± 4.74 fat mass loss vs. 8.70% ± 2.32 lean mass loss; Table 3). There were no associations between outset body fat percentage and percentage of weight lost as fat mass or lean mass.
Table 3. Summary data for pre- and post-diet combined glucose/insulin tests (CGIT), digestibility trials and body composition data.
|
Outset (pre-diet)
Mean ± SD
|
End (post-diet)
Mean ± SD
|
Insulin/glucose dynamics
|
|
|
Baseline insulin (mU/L)
|
10.35 ± 6.77
|
7.00 ± 3.27*
|
Insulin 45 minutes post-infusion (mU/L)
|
217.41 ± 104.69
|
256.34 ± 97.77
|
Insulin 75 minutes post-infusion (mU/L)
|
105.62 ± 124.06
|
143.81 ± 77.15
|
Area under curve insulin (mU/L/min)
|
9970.23 ± 5653.16
|
11927.61 ± 4785.72
|
Area under curve glucose (mmol/L/min)
|
896.27 ±138.90
|
1040.49 ± 99.89*
|
Return to baseline glucose (minutes)
|
73.00 ± 40.30
|
123.33 ± 27.69*
|
Digestibility
|
|
|
Gross energy digestibility (%)
|
48.27 ± 5.76
|
49.67 ± 3.11
|
Dry matter digestibility (%)
|
50.20 ± 4.90
|
51.90 ± 3.39
|
Neutral detergent fibre digestibility (%)
|
49.37 ± 7.36
|
49.31 ± 5.35
|
Body composition
|
|
|
Body mass (kg)
|
301.20 ± 34.83
|
275.20 ± 33.59*
|
Lean mass (kg)
|
237.71 ± 27.13
|
217.15 ± 26.36*
|
Fat mass (kg)
|
63.75 ± 16.13
|
58.92 ± 14.74*
|
Body fat (%)
|
21.01 ± 4.38
|
21.22 ± 4.40
|
Fat loss as % of outset fat mass (%)
|
|
7.32 ± 4.74
|
Lean loss as % of outset lean mass (%)
|
|
8.70 ± 2.32
|
Body composition was calculated from deuterium oxide dilution tests. *indicates significant differences (p < 0.05) from pre-diet values. SD, standard deviation.
Digestibility and glucose/insulin dynamics
Mean apparent digestibilities of GE, DM, and NDF were not altered following the dietary restriction (Table 3). Positive associations were found between pre-diet digestibilities (GE, DM,, NDF) and subsequent weight-loss (Additional File 1). Additionally, positive associations were identified between the change in NDF digestibility (pre-diet minus post-diet) and overall weight-loss (Additional File 1), whereby animals whose apparent NDF digestibility reduced following weight-loss had greater overall weight-loss than those animals whose apparent NDF digestibility was increased following weight-loss. Glucose:insulin dynamics as measured by the CGIT revealed some differences following weight loss. Baseline plasma insulin values were significantly reduced following dietary restriction, but all remained within acceptable normoinsulinaemic ranges (p < 0.05; 10.35µIU/ml ± 6.77 pre-diet to 7.00 µIU/ml ± 3.27 post-diet; Table 3). Insulin measured at time 45 and 75 min, and by default, area under the curve for insulin from the CGIT tended to increase following dietary restriction, but did not reach statistical significance (Table 3). Baseline glucose values were not significantly changed following weight-loss, however the area under the curve for glucose response from the CGIT was significantly increased (p<0.01; 896.27mmol/L/min ± 138.90 pre-diet to 1040.49mmol/L/min ± 99.89 post-diet; Table 3), as was the time to return to baseline for glucose (p<0.01; 73.00 minutes ± 40.30 pre-diet to 123.33 minutes ± 27.69 post-diet; Table 3).
Coverage, Diversity and Volatile Fatty Acids
Quality filtering of 16S rDNA amplicon sequences resulted in 16,028,420 high-quality sequences (320 bp long) which clustered in 9,536 different OTUs. A phylogenetic tree was constructed (PRIMER 6 with Bray Curtis dissimilarity, Additional File 2), which indicated that samples from an animal, collected on each of the three successive sampling days pre- and post-diet, tended to cluster together. This observation allowed data arising from these serial samples to be pooled for each animal. Pooled analyses provided 14,500 sequences per animal, per period after normalization. Rarefaction curves (Additional File 3) demonstrated that sample curves had not plateaued; indicating that complete sampling of these environments had not yet been achieved.
There were significant (p < 0.05) reductions in Inverse Simpson, Shannon-Wiener, and S.obs measures of alpha diversity following weight-loss, but no differences were observed for Chao1 (Table 4). Furthermore, there were significant differences in pre-diet (mean of 3 pre-diet days) measures of diversity (Inverse Simpson, Shannon-Wiener, and S.obs) between the three weight-loss groups (Table 4). This was confirmed by univariate regression analysis, whereby significant negative associations were identified between pre-diet diversity measures and subsequent weight-loss (Additional File 4 and Figure 2).
Table 4. Effect of weight-loss on bacterial diversity and difference in pre-diet diversity between weight-loss groups.
|
Pre-diet
|
Post-diet
|
P-value
|
High
|
Medium
|
Low
|
P-value
|
Inverse Simpson
|
62.02 ± 47.66
|
27.03 ± 16.91
|
<0.01
|
28.31 ± 13.14
|
54.10 ± 21.77
|
103.65 ± 60.32
|
0.03
|
Shannon-Weiner
|
5.46 ± 0.52
|
4.92 ± 0.54
|
<0.01
|
4.94 ± 0.28
|
5.55 ± 0.35
|
5.95 ± 0.29
|
<0.01
|
S.Obs
|
1717.33 ± 222.51
|
1576.20 ± 153.13
|
0.04
|
1488.93 ± 106.99
|
1748.27 ± 156.16
|
1914.80 ± 149.37
|
<0.01
|
S.Chao1
|
2871.63 ± 401.75
|
2729.61 ± 324.73
|
0.37
|
2619.25 ± 279.27
|
2986.81 ± 552.08
|
3008.82 ± 249.78
|
0.24
|
Data presented are mean ± SD. Diversity indices were calculated at both pre- and post-diet time-points (n = 15), and pre-diet diversity indices (mean of 3 pre-diet days) are presented for the high (n=5), medium (n=5) and low (n=5) weight-loss groups.
There were significant reductions in the concentrations of acetate, butyrate, propionate and branched chain volatile fatty acids (BCVFA; comprised of isobutyrate, isovalerate, isocaproic acid and heptanoic acid) following weight-loss, however faecal pH remained unchanged (Table 5). There were no differences in the ratio of acetate plus butyrate to propionate between the pre-diet and post-diet samples, however the pre-diet ratio (mean of 3 pre-diet samples) was greatest in the high weight-loss group compared to the low weight-loss group (4.36 ± 0.58 vs. 3.24 ± 0.50; p = 0.04). Additionally, strong positive associations were identified between pre-diet VFA concentrations and subsequent weight-loss (Acetate, R2 = 0.51, p < 0.01; Butyrate, R2 = 0.28, p = 0.02; Propionate, R2 = 0.27, p = 0.03; Additional File 5 and Figure 3).
Table 5. Effect of weight-loss on faecal dry matter content, pH and volatile fatty acid concentrations and differences in pre-diet faecal dry matter content, pH and VFA concentrations between weight-loss groups.
|
Pre-diet
|
Post-diet
|
P-value
|
High
|
Medium
|
Low
|
P-value
|
Faecal dry matter (%)
|
16.78 ± 1.44
|
18.98 ± 1.96
|
<0.01
|
16.74 ± 1.38
|
16.53 ± 1.14
|
17.07 ± 1.97
|
0.85
|
pH
|
6.73 ± 0.24
|
6.72 ± 0.15
|
0.90
|
6.74 ± 0.04
|
6.77 ± 0.10
|
6.69 ± 0.43
|
0.90
|
Acetate (mM)
|
18.04 ± 7.06
|
14.39 ± 8.35
|
<0.01
|
24.04 ± 6.76
|
17.37 ± 6.58
|
12.70 ± 2.04
|
0.02
|
Propionate (mM)
|
5.19 ± 1.33
|
3.66 ± 1.71
|
<0.01
|
5.98 ± 1.31
|
5.16 ± 1.56
|
4.42 ± 0.74
|
0.19
|
Butyrate (mM)
|
1.77 ± 0.55
|
1.45 ± 0.88
|
0.04
|
2.09 ± 0.73
|
1.73 ± 0.45
|
1.48 ± 0.29
|
0.22
|
BCVFA (mM)
|
1.44 ± 0.58
|
1.13 ± 0.85
|
0.05
|
1.56 ± 0.88
|
1.55 ± 0.51
|
1.22 ± 0.18
|
0.59
|
Data presented are mean ± SD. Faecal pH and VFA concentrations were measured at both pre- and post-diet time-points (n = 15), and pre-diet values (mean of 3 pre-diet days) in the high (n=5), medium (n=5) and low (n=5) weight-loss groups. BCVFA: branched-chain volatile fatty acids.
Due to the significant association between pre-diet acetate and subsequent weight-loss, a logistic regression model was fitted with the binary outcome of “weight loss ≥ 8% BM”. This was used to construct a receiver operating characteristic (ROC) curve to determine the ability of pre-diet acetate concentration to predict subsequent weight-loss success (using target weight-loss of 8% pre-diet body mass) and resulted in an area under the curve of 0.89 (Additional File 6). Optimal test performance achieved with the current data provided a cut-off value of 14mM acetate which gave a test sensitivity of 88.9% (95% confidence intervals: 77.4 – 95.8) and specificity of 88.3% (95% confidence intervals: 67.2 – 93.6) for prediction of weight loss greater than 8%. The pre-diet abundance of Sphaerochaeta, Treponema, (both belonging to the Spirochaetes phylum), Anaerovorax and Mobilitalea (both belonging to Firmicutes phylum) were found to be significantly different between pre-diet acetate concentrations (divided into quartiles; Additional File 7) whereby the pre-diet abundance of the individual genera was greatest for those animals in the lowest quartile of pre-diet acetate concentration.
Changes in bacteria abundance following weight-loss and associations with weight-loss
Across all samples/time-points, Bacteroidetes were the most abundant phylum present, followed by the Firmicutes and Fibrobacteres. The relative abundance of Firmicutes, Tenericutes and Elusimicrobia were significantly reduced following weight-loss (Table 6 and Figure 4), and although the relative abundance of Bacteroidetes did not change significantly following weight-loss, the Bacteroidetes:Firmicutes ratio was increased following weight-loss (2.10 ± 0.54 pre-diet to 2.54 ± 0.56 post-diet; p = 0.01). Similarly, the ratio of the fibre-digesting bacteria, Fibrobacteres:Firmicutes was increased following weight-loss (0.74 ± 0.58 to 1.31 ± 0.88; p = 0.03). The pre-diet abundance of Actinobacteria and Spirochaetes were found to be significantly greater in the low weight-loss group compared to the high weight-loss group (Additional File 8).
Table 6. Relative abundance of bacterial phyla before (pre-diet), and after 7 weeks of dietary restriction (post-diet; n = 15).
|
Pre-diet
|
Post-diet
|
SED
|
Benjamini-Hochberg P-value
|
Firmicutes
|
0.2420
|
0.1894
|
0.015
|
0.007
|
Tenericutes
|
0.0062
|
0.0029
|
0.006
|
0.007
|
Elusimicrobia
|
0.0006
|
0.0014
|
0.003
|
0.007
|
Verrucomicrobia
|
0.0003
|
0.0008
|
0.005
|
0.121
|
Fibrobacteres
|
0.1602
|
0.2264
|
0.041
|
0.187
|
Candidatus Saccharibacteria
|
0.0008
|
0.0006
|
0.002
|
0.198
|
Bacteroidetes
|
0.4874
|
0.4740
|
0.017
|
0.697
|
Unclassified
|
0.0468
|
0.0487
|
0.008
|
0.697
|
Proteobacteria
|
0.0149
|
0.0158
|
0.008
|
0.754
|
Spirochaetes
|
0.0389
|
0.0383
|
0.01
|
0.997
|
Actinobacteria
|
0.0018
|
0.0018
|
0.002
|
0.997
|
At the genera level, whilst the largest proportions of samples were unclassified at this level (belonging to 33 different families, Additional File 9), Fibrobacter was the second-most abundant genus across all samples/time-points. A total of 19 individual genera were altered in abundance following weight-loss: 12 were reduced in abundance, whilst the remaining 7 increased in abundance following weight-loss (Table 7 and Figure 5). The pre-diet abundance of Mobilitalea genus (belonging to the Firmicutes phylum) was found to be significantly greater in low compared to high weight-loss groups (Additional File 10).
Table 7. Relative abundance of bacterial genera before (pre-diet), and after 7 weeks of dietary restriction (post-diet; n = 15).
|
Pre-diet
|
Post-diet
|
SED
|
Benjamini-Hochberg P-value
|
Phascolarctobacterium
|
0.019
|
0.009
|
0.009
|
0.005
|
Prevotella
|
0.016
|
0.007
|
0.008
|
0.005
|
Anaeroplasma
|
0.004
|
0.002
|
0.004
|
0.005
|
Coprobacter
|
0.002
|
0.001
|
0.005
|
0.005
|
Roseburia
|
0.001
|
0.000
|
0.002
|
0.005
|
Anaerorhabdus
|
0.001
|
0.006
|
0.008
|
0.005
|
Candidatusendomicrobium
|
0.000
|
0.001
|
0.003
|
0.005
|
Catabacter
|
0.000
|
0.001
|
0.003
|
0.005
|
Clostridium XIVa
|
0.006
|
0.004
|
0.004
|
0.009
|
Alloprevotella
|
0.005
|
0.001
|
0.007
|
0.012
|
Streptococcus
|
0.001
|
0.002
|
0.005
|
0.022
|
Rikenella
|
0.007
|
0.011
|
0.006
|
0.024
|
Pseudoflavonifractor
|
0.005
|
0.000
|
0.012
|
0.035
|
Saccharofermentans
|
0.001
|
0.001
|
0.003
|
0.035
|
Faecalitalea
|
0.002
|
0.001
|
0.006
|
0.066
|
Anaerovorax
|
0.002
|
0.001
|
0.003
|
0.090
|
Lachnospiracea
|
0.005
|
0.004
|
0.005
|
0.099
|
Paraprevotella
|
0.013
|
0.010
|
0.008
|
0.103
|
Faecalicoccus
|
0.001
|
0.000
|
0.003
|
0.111
|
Anaerocella
|
0.001
|
0.001
|
0.003
|
0.111
|
Macellibacteroides
|
0.001
|
0.000
|
0.005
|
0.125
|
Lachnobacterium
|
0.002
|
0.001
|
0.003
|
0.134
|
Fibrobacter
|
0.161
|
0.228
|
0.042
|
0.150
|
Ruminococcus
|
0.010
|
0.008
|
0.009
|
0.172
|
Asteroleplasma
|
0.002
|
0.001
|
0.006
|
0.172
|
Saccharibacteria
|
0.001
|
0.001
|
0.002
|
0.172
|
Mogibacterium
|
0.001
|
0.001
|
0.001
|
0.175
|
Phocaeicola
|
0.009
|
0.006
|
0.009
|
0.185
|
Sphaerochaeta
|
0.001
|
0.002
|
0.003
|
0.185
|
Rreponema
|
0.028
|
0.031
|
0.008
|
0.216
|
Alkalitalea
|
0.020
|
0.016
|
0.028
|
0.287
|
Unclassified
|
0.629
|
0.597
|
0.017
|
0.300
|
Sporobacter
|
0.002
|
0.002
|
0.002
|
0.354
|
Mobilitalea
|
0.001
|
0.001
|
0.003
|
0.740
|
Paludibacter
|
0.016
|
0.019
|
0.019
|
0.994
|
Barnesiella
|
0.014
|
0.011
|
0.018
|
0.994
|
Osscillibacter
|
0.007
|
0.007
|
0.004
|
0.994
|
Intestinimonas
|
0.002
|
0.002
|
0.005
|
0.994
|
Clostridium IV
|
0.002
|
0.002
|
0.004
|
0.994
|
Ethanoligenens
|
0.001
|
0.001
|
0.006
|
0.994
|
Vampirovibrio
|
0.001
|
0.001
|
0.003
|
0.994
|
A total of 61 OTUs belonging to 13 genera were found to be significantly different in abundance following weight-loss: 7 of which were decreased in abundance following weight-loss and the remaining 51 were increased in abundance (Additional File 11). Significant differences in pre-diet OTU abundances were also identified between the weight-loss groups: the greatest numbers of differences were identified between the high and low weight-loss groups (159 OTUs belonging to 18 genera were significantly different, 101 of which were greater in the high compared to low weight-loss group; Additional File 12). A total of 48 OTUs belonging to 9 genera were found to be significantly different in pre-diet abundance between the low and mid weight-loss groups (Additional File 13), whilst 28 OTUs belonging to 6 genera were found to be significantly different between the mid and high weight-loss groups (Additional File 14).
Microbiome structure as a predictor of weight loss
In an attempt to ascertain whether the pre-diet bacterial community structure could predict subsequent weight-loss, PERMANOVA and ANOSIM analysis was employed. There were significant differences in pre-diet bacterial community structure at the OTU level, whereby the greatest divergence was found between the high and low weight-loss groups (R = 0.67, p < 0.01; Table 8). A Principal Co-ordinate (PCO) analysis was performed using Bray-Curtis distance matrices to illustrate the clustering of the pre-diet bacterial community structure by weight-loss groupings (Figure 6A). Whilst there was some overlapping between the groups, animals in the low weight-loss group clustered together and markedly separately from those in the high weight-loss group where there was more variation in pre-diet bacterial structure between animals. An additional PCO analysis was performed to evaluate associations between the post-diet bacterial community structure and weight-loss groups (Additional File 15) and shows similar clustering of animals as for the pre-diet bacterial community structure. Distance-based redundancy analysis (dbRDA) was performed to evaluate the contribution of pre-diet VFA concentrations and pH to pre-diet bacterial community structure (Figure 6B). Pre-diet acetate concentration correlated strongly with differences in pre-diet bacterial community structure between the weight-loss groups.
Table 8. Summary of ANOSIM and PERMANOVA outputs.
|
Group
|
|
|
PERMANOVA (p-value)
|
|
Low
|
Medium
|
|
Medium
|
0.41
|
|
|
High
|
0.008
|
0.06
|
ANOSIM (R-value)
|
|
|
|
|
Medium
|
0.028
|
|
|
High
|
0.668
|
0.316
|
Results matrices for pre-diet faecal bacterial community structure between weight-loss groups. Significant P-values for PERMANOVA (P < 0.05) are highlighted. Anosim R values indicate the degree of separation between samples (0 = very similar; 1 = highly dissimilar), with significant R-values (with P < 0.05) shown in bold.