Concentrations of oligosaccharides in human milk and child growth

All data were collected within the LIFE Child study at the Research Centre for Civilization Diseases in Leipzig, Germany (www.​ClinicalTrial.​gov: NCT02550236). Children and their parents have been recruited from the 24th week of gestation to 16 years of age to investigate environmental, metabolic and genetic associations with children’s development [31]. The study is described in detail elsewhere [31, 32].

Between 2011 and 2014, 155 milk samples were collected from 153 mothers who nurse at the 3-month baseline visit (Fig. 1). The 3-month-visit took place after their 2nd full month and before the end of the 1st week of their 4th month of life. We excluded 1 sample because of a twin pregnancy and 9 samples because of preterm birth. Two mothers contributing 2 pregnancies were included. Finally, 145 sample-children combinations were included. Further, 132 (91%) follow-up measurements were documented at 6 months, 122 (84%) at 1 year, 106 (73%) at 2 years, 104 (72%) at 3 years, 90 (62%) at 4 years, 87 (60%) at 5 years, 59 (41%) at 6 years, and 37 (26%) at 7 years of age.

Informed written consent was provided by all parents for their children, participation in each procedure were voluntary. The study has been conducted per the Declaration of Helsinki. The study protocol was approved by the Ethics Committee of the Medical Faculty of the University of Leipzig (Reg. No. 264-10-19042010).

Under private conditions, about 20 mL of milk was collected at the 3-month visit using a milk pump (Medela Symphony®). The samples were obtained during the morning before lunchtime (1 pm). The first 20 mL was used. The milk was within at most 20 min stored without processing at − 80 °C at our biobank [31] until its transport to Nestlé Research, Lausanne, on dry ice. The concentrations of 9 HMO were determined by liquid chromatography with fluorescence detection following a previously described validated method [33]. The calibration standards of 2′-fucosyllactose (2’FL), 3-fucosyllactose (3-FL), 3′-sialyllactose (3′SL), 6′-sialyllactose (6′SL), lacto-N-tetraose (LNT), lacto-N-neotetraose (LNnT), lacto-N-fucopentaose-I (LNFP-I), lacto-N-fucopentaose-V (LNFP-V) and lacto-N-neofucopentaose (LNnFP) were purchased from Elicityl (Crolles, France), where the HMO content of the standard powders was determined by quantitative nuclear magnetic resonance spectrometry. For each HMO, 9 point calibration curves were prepared in the ranges 2’FL: 10–4900 mg/L, 3-FL: 10–2800 mg/L, 3′SL: 2–900 mg/L, 6′SL: 2–900 mg/L, LNT: 4–1900 mg/L, LNnT: 2–1000 mg/L, LNFP-I: 4–2000 mg/L, LNFP-V: 2–900 mg/L, LNnFP: 2–900 mg/L. With each batch of analysis (or every 25 samples if batches were larger) a reference pooled human milk sample (Lee Biosolutions, Maryland Heights, USA) was analyzed to ensure the method performance remained consistent between days and between batches of analyses.

Maternal pre-pregnancy weight and height and the child’s birth parameters were taken from the maternity log (“Mutterpass”) [34]. This booklet is updated by medical staff during pregnancy checkups.

Measurements were taken by trained research assistants according to standard procedures. Height/length was measured using the Dr. Keller II infantometer until 1 year of age and the Dr. Keller I stadiometer afterward. Weight was determined by using a scale (the seca 757 or the seca 701). HC was measured using a non-dilatable measuring tape. BMI was calculated. Height/length, weight, and BMI were transformed to standard deviation scores (SDS) according to the guidelines from the German Working Group on Obesity in Childhood and Adolescence [35]. HC measurements were transformed to SDS using German standards from the KiGGS study [36]. As a measure of growth, growth velocity was calculated as the standardized difference between the 3-month height and the 1-year height of the child and afterward between 2 consecutive height measures.

All statistical analyses were carried out using R 4.0 [37]. Descriptive statistics were given as median [Q1;Q3] for HMO and mean (standard deviation) for the other continuous variables (Table 1). For all HMO, values below the limit of quantification (LoQ) were set to the LoQ and marked as left censored. Correlations (r) between specific HMO were investigated using Pearson correlations on the log-transformed values. Secretor status was determined based on the 2’FL concentration, with values below the LoQ of ≤53 mg/L corresponding to mothers who are non-secretors (NSM) and > 53 mg/L corresponding to mothers who are secretors (SM) [30, 33]. For HMO, differences in medians between SM and NSM were tested using Kruskal-Wallis tests or censored regression models when data below the LoQ occurred. For the other continuous variables, differences in means were tested using t-tests. Chi-squared tests were applied to test differences in proportions. The associations between the HMO and the children’s and mothers’ parameters were examined using generalized additive models for location, shape, and scale [38, 39], with HMO as the outcome and the other variables as predictor variables. Modeling was done separately for each age group, assuming a log-linear relationship between predictor and outcome. Due to the HMO values’ considerable skewness, these values were log-transformed. A Box-Cox Cole and Green distribution or its censored equivalent was chosen to describe the outcomes’ distributions. Investigating the models’ error structure revealed variances according to secretor status. Therefore, variance was modeled dependent on secretor status. Skewness was dependent on secretor status only for LNnT. With evidence of an interaction between the predictor and the secretor status, the respective interaction term was included in the model. The models’ appropriateness was checked using different plots (QQ-plot, variance against fitted, variance against covariates, influence vs. cooks distance; plots not shown). Effects are reported as ratios (R = exp(β)) or differences (β), including the 95% confidence interval. As NSM values of LNFP-I and 2’FL were below the LoQ (≤15 mg/L and 53 mg/L, respectively) [33], associations involving LNFP-I and 2’FL were only modeled in the SM subgroup. P-values ≤0.05 were considered to be statistically significant. Because of the complex dependencies between the tested items, p-values were not adjusted for multiple testing. Essentially, we stress the fact that the results should be interpreted in terms of the occurring patterns instead of emphasizing single significant test results.

The cohort consists of 21 (14.7%) NSM and 122 (85.3%) SM. Two of the SM took part with 2 singleton pregnancies (Fig. 1). 2’FL, LNFP-I and LNnT concentrations were significantly lower in NSM, while 3-FL, 3′SL, LNT and LNFP-V concentrations were higher. Further descriptive statistics are given in Table 1.

Correlation was highest between 2’FL and LNFP-I (r = 0.95, p < 0.001). Both were also positively correlated to LNnT (2’FL: r = 0.56, p < 0.001; LNFP-I: r = 0.61, p < 0.001) and negatively correlated to LNFP-V (2’FL: r = − 0.66, p < 0.001; LNFP-I: r = − 0.62, p < 0.001) and 3-FL (2’FL: r = − 0.54, p < 0.001; LNFP-I: r = − 0.64, p < 0.001) (Fig. 2).

The maternal age at birth was positively associated only with 3′SL (R = 1.06 [1.00,1.11] for every 5 years older, p = 0.03). Pre-pregnancy BMI was negatively associated with LNnT in NSM (R = 0.93 [0.90,0.97], p < 0.001); no association was found in SM. GA was positively associated with LNT (R = 1.08 [1.01,1.15], p = 0.03), 6′SL (R = 1.09 [1.03,1.16], p = 0.005) and LNFP-I (R = 1.15 [1.03,1.28], p = 0.012, only children from SM (SC)). LNFP-I (R = 1.2, [1.05,1.37], p = 0.008, only SC) and 3-FL (R = 0.92 [0.86,0.98], p = 0.011) were significantly associated with birth length (Supplementary Table S1).

The interaction between height-SDS and secretor status was significant for LNT. The association of height-SDS with LNT had a negative direction in children of NSM (NSC) (0.78 ≤ R ≤ 1.01) and a positive direction in SC (1.00 ≤ R ≤ 1.09; Supplementary Table S2; Fig. 3). However, the effects reached significance only for NSC at 2Y. LNFP-I was positively associated with height-SDS effects at 3 M, 6 M and 1Y (R ≈ 1.2, p ≤ 0.02); afterward, no further associations were found. Besides, there were consistently positive associations between height SDS and LNnT. However, statistical significance was only reached at 6Y. There was no evidence of associations between height-SDS and 2’FL, 3-FL, 3′SL, 6′SL, LNnFP, or LNFP-V.

The interaction between growth velocity and secretor status was significant for LNnT. In NSC, we found negative effects for 3 M–1Y (R = 0.95 [0.90,0.99], p = 0.01), 1Y–2Y (R = 0.80 [0.72,0.88], p < 0.001) and for 5Y–6Y (R = 0.71 [0.57,0.87], p = 0.002).

LNT and LNFP-V were negatively associated with growth velocity from 3 M–1Y (LNT: R = 0.97 [0.95,1.00], p = 0.02; LNFP-V: R = 0.97 [0.95,1.00], p = 0.04). LNFP-I and 3′SL were negatively associated with growth velocity from LNFP-I: 1Y–2Y (R = 0.90 [0.83,0.97] p = 0.008) and 3′SL: 4Y–5Y (R = 0.95 [0.90,1.00] p = 0.045). No significant associations were found for growth velocity and 2’FL, 3-FL, 6′SL and LNnFP (Supplementary Table S3).

The interaction between BMI-SDS and secretor status was significant for 3′SL, 6′SL, LNT, LNFP-V and LNnFP. For NSC, we found consistently negative associations between BMI-SDS and 3′SL, 6′SL, LNT and LNFP-V at all time points (Supplementary Table S4; Fig. 3). However, statistical significance was reached only for LNT and LNFP-V at 2Y. Besides, we found consistently positive associations between BMI-SDS and LNnFP from 6 M onward. Statistical significance was reached at 02Y, 05Y and 06Y. We did not find evidence of associations between BMI-SDS and the HMO in SC when the models included the interaction term.

2’FL showed consistently negative associations with BMI-SDS (0.92 ≤ R ≤ 1; Supplementary Table S4; Fig. 3); statistical significance was reached at 3 M, 4Y and 5Y. The 3-FL was consistently positively associated with BMI-SDS between 3 M and 6Y; however, the results did not reach statistical significance. For LNnT and LNFP-I, no consistent patterns were found.

The interaction between HC and secretor status was significant for LNFP-V and LNnFP. LNFP-V showed consistent, significantly negative associations with HC between 3 M and 7Y, with effect sizes varying between 0.63 and 0.92 in NSC (Supplementary Table S5; Fig. 3). SC had no notable pattern. LNnFP showed consistently positive effects on HC from 3 M–7Y, with effect sizes between 1.12 and 2.09 in NSC. In general, the effect sizes increased with age. Statistical significance was reached from 2Y–6Y. Again, we found no notable patterns in SC.

LNFP-I was consistently positively related to HC-SDS from 6 M–7Y with effect sizes between 1.02 and 1.20. However, most of the effects did not reach significance. There were no notable pattern or effects for 2’FL, 3-FL, 3′SL, 6′SL or LNT (Supplementary Table S5; Fig. 3).

NSC had a significantly higher BMI-SDS at 3 M (β = 0.8 [0.4,1.2], p < 0.001) and 6 M (β = 0.8 [0.4,1.2], p < 0.001). At 1 year, the direction of the association remained the same but the difference was not significant (Fig. 4).

At 3 M and 6 M, NSC tended to show an approximately + 0.5 higher height-SDS. However, there was no statistically significant effect. HC-SDS was astoundingly higher in NSC at 3 M (β = 1.3 [0.6,2.0], p < 0.001), 6 M (β = 1.0 [0.4,1.6], p = 0.001), and 1Y (β = 0.7 [0.3,1.2], p = 0.002). Even afterwards, HC-SDS stayed higher in NSC with effect sizes between β = 0.4 and β = 1.0, reaching significance at 3Y, 6Y, and 7Y (Fig. 4).