Study population
Twenty-two infants were enrolled in the study, of which 19 completed the study. Three infants discontinued during/after the run-in period because of either diarrhoea (1), disliking the formula (1) or failure to deliver data and samples (1) (Figure 1). Complete datasets were obtained for 16 infants, 7 in the VF-MF group and 9 in the MF-VF group. For one infant in each of the groups, no faecal samples were obtained in the second intervention period, and for that reason, faecal samples of these infants were not analysed. For another infant in the VF-MF group, questionnaires and biochemical analysis of the faecal samples were lacking. Additional file 1, Supplementary Table 1 gives an overview of IF allocation and availability of data per subject.
The baseline and family characteristics of the infants are described in Table 2. Weight at birth, gestational age and infants’ age and weight at inclusion were similar among the groups.
Table 2 Baseline and family characteristics of the study subjects
|
MF-VF
(n=10)
|
VF-MF
(n=9)
|
Gender, n (%)
|
|
|
Female
|
5 (50.0)
|
4 (44.4)
|
Male
|
5 (50.0)
|
5 (55.6)
|
Age at screening, days
|
|
|
Mean (SD)
|
92.0 (11.6)
|
89.4 (14.6)
|
Median
|
89
|
84
|
Min-max
|
70-112
|
67-111
|
Ethnicity, n (%)
|
|
|
Albanian
|
1 (10.0)
|
1 (11.1)
|
Greek
|
9 (90.0)
|
8 (89.9)
|
Length at birth, cm
|
|
|
Mean (SD)
|
49.4 (2.0)
|
50.2 (1.5)
|
Median
|
50.0
|
50.0
|
Min-max
|
44.0-51.0
|
49.0-54.0
|
Weight at birth, g
|
|
|
Mean (SD)
|
3063.0 (372.8)
|
2935.0 (265.0)
|
Median
|
3150.0
|
2910.0
|
Min-max
|
2500-3480
|
2500-3455
|
Gestational age, weeks
|
|
|
Mean (SD)
|
37.7 (0.8)
|
38.6 (0.5)
|
Median
|
37.5
|
39.0
|
Min-max
|
37-39
|
38-39
|
Mode of delivery, n (%)
|
|
|
C-section
|
8 (80.0)
|
6 (66.7)
|
Natural delivery
|
2 (20.0)
|
3 (33.3)
|
Length at screening, cm
|
|
|
Mean (SD)
|
62.05 (2.2)
|
62.1 (3.1)
|
Median
|
61.5
|
61.5
|
Min-max
|
59.0-65.3
|
58-68
|
Weight at screening, g
|
|
|
Mean (SD)
|
5963.7 (601.2)
|
5827.1 (945.9)
|
Median
|
5958.5
|
6180
|
Min-max
|
4860.0-7083.0
|
4130-7200
|
VF: standard formula with 100% vegetable fat source; MF: test formula with 50% milk fat; SD: standard deviation.
MF-VF: crossover group with MF first; VF-MF: crossover group with VF first
Formula consumption and anthropometric data
The average weekly milk intake and the infants’ weight and length development during the study were similar for the MF-VF and VF-MF groups (Additional file 1, Supplementary Table 2).
Microbiota analysis
The faecal microbial profiles for most infants (13 out of 17 infants, 76%) were dominated by the genus Bifidobacterium and in line with previous reports (16,17). The microbiota of the other infants were dominated by Klebsiella (3 children) and Escherichia (1 child). Through the intervention, changes in the microbial profiles were observed. However, the variation in the microbiome between the infants was larger than the effect of the intervention (Figure 2). Furthermore, no specific clades were directly and significantly affected by the interventions performed in this study. We performed a basic univariate analysis on each separate type of the 250 different species, where only Enterococcus faecalis was significant before multiple testing (p<0.02), though not after multiple testing correction, due to the high number of species tested and relatively low sample size.
To determine whether the interventions affected the microbial diversity in the infants, both the Shannon alpha diversities and Bray-Curtis beta diversities were determined; however, neither of these methods showed an intervention-specific effect.
Multidimensional scaling (MDS) analysis identified three beta diverse distinct groups, also found with hierarchical clustering (Figure 3). These sample groups do not correspond to the time of sampling or interventions. With few exceptions, the samples from singular infants localize together. Of the abundant microbes, the relative proportion in the samples of the genera Aeriscardovia, consisting only of the species Aeriscardovia aeriphila that was formerly known as Bifidobacterium aerophilum (adjusted p-value 2.4E-3) and Klebsiella (adjusted p-value 9.4E-6), primarily differentiate group 1 from 2 and 3. In addition, group 2 can be differentiated from groups 1 and 3 by their, albeit very low (<0.08%), relative abundance of Thermoleophilum (adjusted p-value 8.1E-6), a genus of the phylum Actinobacteria. In total, the relative abundances of 16 genera were significantly different between the 3 clusters. Using Pearson's chi square tests, the clusters were compared to infant-specific categorical clinical parameters. Clusters were significantly associated with gender, prior breastfeeding and stool frequency (Table 3). Group 3 was enriched for males, and group 2 was enriched for breastfed infants.
The three groups, determined by hierarchical clustering, are indicated by gray ovals. The shapes indicate the sexes male (circle) or female (triangle), while the colours indicate absence (blue) or presence (red) of prior breastfeeding before the start of the trial.
Table 3 Clinical parameters significantly associated with beta diversity.
Parameter
|
p-value
|
Adjusted p-value (FDR)
|
Gender p
|
0.007
|
0.034
|
Prior Breastfeeding p
|
0.001
|
0.009
|
Mode of delivery p
|
0.852
|
0.852
|
Bowel frequency p
|
0.063
|
0.190
|
Stool consistency p
|
0.191
|
0.344
|
Stool colour p
|
0.305
|
0.457
|
Stool amount p
|
0.486
|
0.546
|
Formula volume consumed k
|
0.433
|
0.546
|
Stool frequency (day 12-14) k
|
0.172
|
0.344
|
Analysis based on Pearson's chi-squared tests (p) or Kruskal-Wallis rank sum tests (k); FDR: false discovery rate
Stool fatty acids, fatty acid soaps and calcium concentration
The results of the main biochemical measurements of stool samples are shown in Table 4. The IF used in the current study strongly influenced the formation of faecal palmitic acid soap and total fatty acid soap concentrations. The MF formula with a higher level of sn2-palmitate resulted in significantly lower faecal levels of palmitic acid soap and total fatty acid soaps than the VF formula (p<0.0002), as illustrated in Figure 4. Concomitantly, calcium excretion and palmitic acid concentration were significantly (p<0.05) reduced in stool samples after the MF intervention. No significant difference was noted for the total free fatty acid concentration. For faecal palmitic acid soap, the concentration as well as the proportion palmitic acid soap of total faecal soaps were significantly different between groups.
Table 4 Mean stool fatty acids, fatty acid soaps and calcium composition (mg/g stool dry weight)
Variable
|
Run-in VF
(n=16)
|
Intervention MF
(n=16)
|
Intervention VF
(n=16)
|
p-value intervention
|
Palmitic acid soap (SD), mg/g
|
214.3 (36.6)
|
133.5 (44.9)
|
206.2 (33.9)
|
0.0002
|
Palmitic acid soap (SD), %
|
74.2 (3.4)
|
58.9 (9.5)
|
73.4 (6.6)
|
0.0003
|
Palmitic acid (SD), mg/g
|
9.7 (13.1)
|
3.8 (4.6)
|
5.2 (6.3)
|
0.0335
|
Palmitic acid (SD), %
|
34.3 (12.4)
|
26.3 (12.1)
|
29.1 (12.7)
|
0.3409
|
Calcium excretion (SD), mg/g
|
47.4 (6.7)
|
44.4 (7.9)
|
47.7 (6.5)
|
0.0335
|
Total fatty acid soaps (SD), mg/g
|
288.6 (47.1)
|
222.7 (47.3)
|
281.9 (43.8)
|
0.0001
|
Total free fatty acids (SD), mg/g
|
24.1 (24.8)
|
14.4 (12.5)
|
16.4 (10.3)
|
0.3016
|
VF: standard formula with 100% vegetable fat source; MF: test formula with 50% milk fat; SD: standard deviation.
Stool consistency and gut comfort
Figure 5 shows the distribution of the ordinal stool consistency scores for the VF and MF interventions. The stool consistency scores, according to AISS, showed a p-value of 0.02 for the intervention fixed effect test, where the odds ratio for intervention difference was 4.0 (95% CL: 1.2-13.2). This implies that stool samples generally have a lower rank on the (ordinal) AISS scale in the MF intervention, i.e. altering the stool composition towards softer stools. AISS assigns one of four categories to an infants’ stool: Watery, Soft, Formed and Hard. The latter category, however, was not assigned to any faecal sample.
There was no significant difference for vomiting, regurgitation, colic, constipation, diarrhoea and crying episodes between groups based on the QPGS-RIII infant/toddler. Additionally, no difference in stool volume or stool colour based on AISS was observed (data not shown).
Metabolite analysis
In total, 870 metabolites were measured in the faecal samples (see Additional file 2), of which 263 were significantly different between the MF and VF interventions determined by a crossover ANOVA (p<0.05). The majority of these 263 metabolites were found to be higher in the faeces following MF intervention, potentially indicative of the complex composition of milk fat (18). The significantly affected metabolites (45) with at least a 2-fold change between interventions are indicated in Figure 6. Most of these are lipid-derived, but some are metabolites from vitamin, amino acid and nucleotide metabolism. Higher levels of faecal metabolites after MF intervention include molecules of known milk fat origin: margarate, 12/13-methylmyristate, pentadecanoate, pristanate, phytanate, docosapentaenoate, 14/15-methylpalmitate, docosadienoate, palmitoleate and carotene diol. Furthermore, several other lipid-derived molecules, fucitol, 3-aminoisobutyrate, 3-indoleglyoxylic acid, N-carboxymethylalanine, N-methylproline and several forms of vitamin E (tocopherols), were found at levels that were at least 2-fold higher after MF intervention than after VF intervention. In contrast, significantly influenced metabolites higher after VF intervention included tocotrienols, which are other forms of vitamin E. The different forms of vitamin E measured in faeces after either MF or VF intervention are consistent with the composition of the IF consumed, as tocotrienols are more abundant in the vegetable fat blend used for the VF formula than tocopherols. Several metabolites related to polyphenol metabolism (2-hydroxyhippurate, hippurate, and cinnamoylglycin) were also higher after VF intervention, which could result from the higher levels of polyphenols in the vegetable fat blend (high in palm oil) included in the VF formula. For other metabolites higher after VF intervention (gamma-amino butyric acid (GABA), phenethylamine, agmatine, N-carbamoyl putrescine, and nicotinate ribonucleoside), no direct link with the compositions of the study products is suspected, and these metabolites might thus reflect a treatment effect.
Coloured dots indicate significantly different (p<0.05) metabolites with at least a two-fold change between MF and VF intervention (upper left corner: higher in VF intervention; upper right corner: higher in MF intervention).
Random forest analysis (Additional file 1, supplementary Figure 1) shows that the two intervention groups can be distinguished from their metabolite profile with a predictive accuracy of 91%. In addition, it provides the metabolites that are most predictive of the type of intervention. These metabolites partly overlap with the metabolites shown in Figure 6, but in addition several carnitines, ergosterol, 7-ketocholesterol, lyxonate, myristate and laurate helped to distinguish both groups.