Correlation studies between OTUs and fecal metabolites of breast- and formula-fed infants

In order to determine microbiome-metabolome characteristics of breastfeeding and formula-feeding, metabolite and microbiota data were combined to search for correlations of metabolites and OTUs at an early (month 1), mid (month 7) and late (month 12) time point. For the correlation, metabolites of the positive or negative ionization mode, respectively and the 16S data of the exclusively fed infants of month 1, 7 and 12 separately were merged and analyzed by OPLS-DA. In order to extrapolate the mass signals and correlated OTUs of month 1, 7 and 12 respectively, which are responsible for the discrimination of the investigated groups (B, F+ and F-), the loadings of each month of the OPLS-DA analysis were extracted. The loadings plots are illustrated in Figure 3.4-24 for UHPLC-(+)-ToF-MS mode (top) and UHPLC-(-)-ToF-MS mode (bottom). Further, CV-ANOVA was applied in order to verify the robustness of each model. Indicators, such as the p-value, the goodness-of-fit R²Y(cum) and the goodness-of prediction Q²(cum) were reported (Bazanella et al. 2017) and read as shown in Table 3.4-7.

Table 3.4-7: Orthogonal signal corrected OPLS/O2PLS-DA results from different model comparison.

Mode Models R²Y(cum) Q²(cum) p-value (CV-ANOVA)

(+) ESI

Month 1 0.94 0.48 6.72·10^-7

Month 7 0.53 0.38 1.84·10^-8

Month 12 0.4 0.18 0.0194

(-) ESI

Month 1 0.94 0.53 1.28*10^-9

Month 7 0.54 0.42 1.43*10^-9

Month 12 0.34 0.21 0.003

In order to reveal if both of the formula fed groups of month 12 can be separated, an additional orthogonal component was added to the model. It was possible to confirm that no further separation among the Y-axis could be obtained.

Figure 3.4-24: Correlation studies between metabolites and OTUs at month 1, 7, and 12

Correlations between metabolites and OTUs of month 1, 7 and 12, analyzed in UHPLC-(+)-ToF-MS (top) and UHPLC-(-)-ToF-MS (bottom), obtained through OPLS-DA analyzed and illustrated as loadings plot of all metabolites (yellow) and OTUs (grey) (circles=metabolites, triangle=OTU) Top: month 1, R²(cum)=0.94, Q²(cum)=0.48; p=6.72*10-7 (CV-ANOVA); month 7, R²(cum)=0.53, Q²(cum)=0.38; p=1.84*10-8 (CV-ANOVA) and month 12, R²(cum)=0.47, Q²(cum)=0.106; p=0.305 (CV-ANOVA, orthogonal component). Bottom: month 1, R2(cum)= 0,94, Q2(cum)=0.53; p=1.28*10-9 (CV-ANOVA); month 7, R2(cum)=0.54, Q2(cum)=0.42; p=1.43*10-9 (CV-ANOVA) and month 12, R2(cum)=0.34, Q2(cum)=0.21; p=0.003 (CV-ANOVA, orthogonal component)

Relatively less information about specific metabolites, which differ in fecal samples of healthy breast fed or formula fed infants – except bile acids and SCFA – is available. Therefore, the non-targeted metabolomics approach enabled to evaluate a broader range of discriminating metabolites, containing known and unknown features and their correlated OTUs in fecal samples of the exclusively breast milk, interventional and placebo group.

Moreover, through correlation experiments of all metabolites and OTUs, 6 OTUs were detected, which are involved in the separation of the different feeding groups and contribute in the feeding-specific shaping of the fecal ecosystem at month 1 (Bazanella et al. 2017). Furthermore, the correlation revealed a relation between F+ specific metabolites and two species (OTU 4, Bifidobacterium bifidum; OTU 142,

Month 1 Month 7 Month 12

UHPLC-(+)-ToF-MSUHPLC-(-)-ToF-MS

Month 1 Month 7 Month 12

B breastfed F- Placebo formula F+ Interventional formula

metabolites OTUs

Lactococcus sp.) in month 1. In contrast, some F- specific metabolites were related with Bacteroides sp. (OTU 10, OTU 18) (Bazanella et al. 2017).

In month 7, a relation between Bifidobacterium sp. (OTU 1) and specific metabolites was detected in month 7. On the contrary, through formula feeding, a correlation with Flavonifractor sp. (OTU 29) showed up. Also, in month 12, several correlations among metabolites and OTUs were observed in the formula-fed infants, including Flavonifractor sp. (OTU 29) and Coprobacillus sp. (OTU 125). Further details are given in Table 3.4-8. The most important mass signals and correlated OTUs for each month and within each feeding group (B, F, F+ and F+) were illustrated in Figure 3.4-25 and listed by rank from high to low importance for month 1 in Table 6.2-19, for month 7 in Table 6.2-20 and for month 12 in Table 6.2-21.

Figure 3.4-25: Correlation studies between metabolites and OTUs at month 1, 7, and 12.

Correlations between metabolites and OTUs of month 1, 7 and 12, analyzed in UHPLC-(+)-ToF-MS, obtained through OPLS-DA analyzed and illustrated as scores plot (top) and loadings plot of main discriminating and correlating features (circles=metabolites, triangle=OTU; bottom) of month 1, R²(cum)=0.94, Q²(cum)=0.48;

p=6.72*10-7 (CV-ANOVA); month 7, R²(cum)=0.53, Q²(cum)=0.38; p=1.84*10-8 (CV-ANOVA) and month 12, R²(cum)=0.47, Q²(cum)=0.106; p=0.305 (CV-ANOVA, orthogonal component). Further details are listed in Table 6.2-19 (month 1), Table 6.2-20 (month 7) and Table 6.2-21 (month 12). From Bazanella, M., Maier, T. V., Clavel, T., Lagkouvardos, I., Lucio, M., Maldonado-Gòmez, M. X., Autran, C., Walter, J., Bode, L., Schmitt-Kopplin, P., Haller, D.: Randomized controlled trial on the impact of early-life intervention with bifidobacteria on the healthy infant fecal microbiota and metabolome. Am J Clin Nutr. (2017), 106(5):1274-1286. Reprinted and adapted from (Bazanella et al. 2017) by permission of Oxford University Press. Copyright (2017) American Society for Nutrition.

-20

B breastfed F- Placebo formula F+ Interventional formula F formula fed metabolites OTUs

metabolites

Table 3.4-8: Correlation between feeding cohort specific metabolites and OTUs.

Correlations between the feeding cohort specific metabolites and OTUs of month 1, 7 and 12 were obtained through Pearson correlation, recording the correlation coefficient (r), and the degrees of freedom (df), as well as p-corr values calculated through regression analysis. Table contains m/z of mass signals obtained through UHPLC-(+)-ToF-MS analysis, compound name, if possible and month, the correlation was observed at. From Bazanella, M., Maier, T. V., Clavel, T., Lagkouvardos, I., Lucio, M., Maldonado-Gòmez, M. X., Autran, C., Walter, J., Bode, L., Schmitt-Kopplin, P., Haller, D.: Randomized controlled trial on the impact of early-life intervention with bifidobacteria on the healthy infant fecal microbiota and metabolome. Am J Clin Nutr. (2017), 106(5):1274-1286. Reprinted and adapted from (Bazanella et al. 2017) by permission of Oxford University Press. Copyright (2017) American Society for Nutrition.

Feed m/z retention time [min]

Compound Month OTU ID r p-corr df

F- 440.2844 6.12 LysoPE(15:0/0:0) Month 1 Bacteroides sp. OTU 10 0.75 7.21E-08 36 F- 616.3490 6.36 no metabolite found Month 1 Bacteroides sp. OTU 18 0.57 2.14E-04 36 F+ 813.5680 4.79 no metabolite found Month 1 Bifidobacterium

sp.

OTU 4 0.49 2.00E-03 36 F+ 417.3345 6.72 no metabolite found Month 1 Lactococcus sp OTU 142 0.45 4.16E-03 36

B 206.0830 3.62 no metabolite found Month 7 Bifidobacterium sp.

OTU 1 0.52 3.94E-04 40 B 261.1469 2.80 no metabolite found Month 7 Bifidobacterium

sp.

OTU 1 0.54 1.89E-04 40 F 417.3371 9.53 no metabolite found Month 7 Flavonifractor sp. OTU 29 0.42 5.60E-03 40 F 407.2455 3.05 no metabolite found Month 7 Flavonifractor sp. OTU 29 0.54 1.78E-04 40

F 427.3608 7.53 no metabolite found Month 12 Flavonifractor sp. OTU 29 0.64 2.90E-04 25 F 303.1916 3.16 no metabolite found Month 12 Flavonifractor sp. OTU 29 0.58 1.37E-03 25 F 447.3480 7.64 no metabolite found Month 12 Coprobacillus sp. OTU 125 0.54 3.58E-03 25

Through correlations of metabolites of the negative ionization mode and OTUs, also no correlations up to the top 50 highest ranked metabolites and OTUs affected through breastfeeding were observed in month 1. The most important mass signals and correlated OTUs for each month and within each feeding group (B, F, F+ and F+) were illustrated in Figure 3.4-26.

Figure 3.4-26: Correlation studies between metabolites and OTUs at month 1, 7, and 12.

Correlations between metabolites and OTUs of month 1, 7 and 12, analyzed in UHPLC-(-)-ToF-MS, obtained through OPLS-DA, analyzed and illustrated scores plot (top) and loadings plot of main discriminating and correlating features (circles=metabolites, triangle=OTU; bottom) of month 1, R²(cum)= 0,94, Q²(cum)=0.53;

p=1.28*10-9 (CV-ANOVA); month 7, R²(cum)=0.54, Q²(cum)=0.42; p=1.43*10-9 (CV-ANOVA) and month 12, R²(cum)=0.34, Q²(cum)=0.21; p=0.003 (CV-ANOVA, orthogonal component). Further details are listed in Table 6.2-22 (month 1), Table 6.2-23 (month 7), and Table 6.2-24 (month 12).

On the contrary, a correlation between F+ specific metabolites, especially the unknown m/z 541.3335 and Bifidobacterium sp. (OTU 4) could be revealed. In the placebo formula group, a relationship between metabolites of the F- metabolites and Bacteroides sp. (OTU 10), especially with m/z 438.2606, which was already correlated to OTU 10 evaluating the data of the positive ionization. Through (+) ToF MS/MS experiments and the comparison of the experimental MS/MS spectra with the METLIN database, it was previously classified as LysoPE (15:0/0:0). LysoPE (15:0) recently was detected by Faith et al., who claimed metabolite-microbial community interactions among lysophosphatidylethanolamine (LysoPE15) and Bacteroides species in cecum samples. They observed increased concentrations of LysoPE15 if B. ovatus or B. vulgatus are present and reach its highest concentration levels if both species are present (Faith et al. 2014). Main correlated features are listed by rank from high to low importance for month 1 in Table 6.2-22, for month 7 in Table 6.2-23 and for month 12 in Table 6.2-24.

0

B breastfed F- Placebo formula F+ Interventional formula F formula fed metabolites OTUs

metabolites

In month 7, flavonifractor sp. (OTU 29) showed up as correlated to traumatic acid in the formula-fed group, whereas no correlation was observed between metabolites and OTUs impacted through breastfeeding. Here, in month 12 also a correlation of enterococcus sp. (OTU 14) and metabolites significant for the breastfed infants was detected. In the formula-fed group flavonifractor sp. (OTU 29) and the bile acid 6-lithocholic acid were highly correlated. Further details are listed in Table 3.4-9.

Table 3.4-9: Correlation between feeding cohort specific metabolites and OTUs.

Correlations between the feeding cohort specific metabolites and OTUs of month 1, 7 and 12 were obtained through Pearson correlation, recording the correlation coefficient (r), and the degrees of freedom (df), as well as p-corr values calculated through regression analysis. Table contains m/z of mass signals obtained through UHPLC-(-)-ToF-MS analysis, compound name, if possible and month, the correlation was observed at.

Feed m/z retention time

[min] Compound Month OTU ID r p-corr df

F- 438.2606 6.10 no metabolite found Month 1 Bacteroides sp. OTU 10 36 8.39E-05 0.59 F+ 541.3335 5.47 no metabolite found Month 1 Bifidobacterium

sp. OTU 4 36 4.00E-03 0.45 F 227.1281 5.16 Traumatic acid Month 7 Flavonifractor sp. OTU 29 40 4.00E-03 0.43 F 435.2754 5.46 no metabolite found Month 7 Flavonifractor sp. OTU 29 40 2.41E-05 0.60 B 583.2704 4.87 no metabolite found Month 12 Enterococcus sp. OTU 14 26 3.21E-05 0.70 B 546.1973 0.93 no metabolite found Month 12 Enterococcus sp. OTU 14 26 4.55E-03 0.52 F 389.2686 5.89 6-Lithocholic acid Month 12 Flavonifractor sp. OTU 29 26 4.00E-03 0.61 F 389.2685 6.74 no metabolite found Month 12 Flavonifractor sp. OTU 29 26 9.58E-05 0.67

As in month 1 no correlations among highly significant metabolites impacted through breastfeeding and OTUs were observed. It encourages the assumption that the metabolites are a major driver for the discrimination of breast- from formula-fed groups. It was already observed that the differences between breast and formula groups were maintained until the end of intervention at 1 year, but it’s remarkable that the correlations of metabolites and OTUs were highly present in the formula-fed group, and the OTUs play an important part in contributing to the discrimination of F+ and F-, as seen in month 1. On the contrary, considering the highly significant discriminative features impacted through breastfeeding in month1, 7 and 12, it appeared predominantly driven by metabolites, rather than through OTUs, even if the metabolites are gaining lesser influence on the discrimination of those two groups in later life.

Abdulkadir et al. evaluated the use of probiotics in preterm infants and their impact on the microbiome and metabolome (Abdulkadir et al. 2016). They concluded that metabolite profiles are different between probiotic and control groups. This result strengthens our results on the discrimination of breast fed and formula fed infants over time.