Novel approach to visualize the inter-dependencies between maternal sensitization, breast milk immune components and human milk oligosaccharides in the LIFE Child cohort

Abstract

Background: Numerous studies have shown that specific components of breast milk, considered separately, are associated with disease status in the mother or the child using univariate analyses. However, very few studies have considered multivariate analysis approaches to evaluate the relationship between multiple breast milk components simultaneously.

Aim: Here we aimed at visualizing breast milk component complex interactions in the context of the allergy status of the mother or the child.

Methods: Milk samples were collected from lactating mothers participating in the Leipziger Forschungszentrum für Zivilisationskrankheiten (LIFE) Child cohort in Leipzig, Germany. A total of 156 breast milk samples, collected at 3 months after birth from mother/infant pairs, were analyzed for 51 breast milk components. Correlation, principal component analysis (PCA) and graphical discovery analysis were used.

Result: Correlations ranging from 0.40 to 0.96 were observed between breast milk fatty acid and breast milk phospholipids levels and correlations ranging from 0 to 0.76 between specific human milk oligosaccharides (HMO) were observed. No separation of the data based on the risk of allergy in the infants was identified using PCA. When graphical discovery analysis was used, dependencies between maternal plasma immunoglobulin E (IgE) level and the breast milk immune marker transforming growth factor-beta 2 (TGF-ß2), between TGF-ß2, breast milk immunoglobulin A (IgA) and TGF-ß1 as well as between breast milk total protein and birth weight were observed. Graphical discovery analysis also exemplifies a possible competition for the fucosyl group between 2'FL, LNFP-I and 3'FL in the HMO group. Additionally, dependencies between immune component IgA and specific HMO (6'SL and blood group A antigen tetraose type 5 or PI-HMO) were identified.

Conclusion: Graphical discovery analysis applied to complex matrices such as breast milk composition can aid in understanding the complexity of interactions between breast milk components and possible relations to health parameters in the mother or the infant. This approach can lead to novel discoveries in the context of health and diseases such as allergy. Our study thus represents the first attempt to visualize the complexity and the inter-dependency of breast milk components.

Trial registration: ClinicalTrials.gov NCT02550236.

Conflict of interest statement

WK received an unconditional grant by Societé des Produits Nestlé S.A. to carry out this analysis. MS, ENC, EE, SH, SN, NS, CN and CB were employed by Societé des Produits Nestlé S.A. during the conduct of the study. This does not alter the adherence to PLOS ONE policies on sharing data and materials. AK, LM, and MV have no conflict of interest to disclose for this work. Societé des Produits Nestlé S.A. and authors of this publication hold patents in the context of allergy management: EE, SH, SN, NS, CN, CB. Results of the current work are not used in currently filed or published patents.

Figures

Fig 1. Correlation matrix of breast milk components.

Pearson correlations represented by a color gradient based on the r coefficient are shown for the 24 selected breast milk components.

Fig 2. Principal components analysis (PCA).

Amount and cumulative percentage of variance in each component in the PCA (A). PCA analysis with the allergic status in the infant (B) and the mother (C) is shown. The length of the arrow corresponds to the strength of the correlation between a variable and the PC axes. The direction of the arrow shows which direction and with which axis the variable (breast milk component) is correlated.

Fig 3. Graphical discovery analysis.

An arrow pointing from a variable x to a variable y indicates the influence of the first variable on the second one. The length of the arrow and the position on the graph was generated automatically. The colouring of the components is similar to the one used in PCA. Possible confounders are in grey.