Page 81 - Read Online

van Beek et al. Microbiome Res Rep 2025;4:13 https://dx.doi.org/10.20517/mrr.2024.45 Page 7 of 19

ab130945) for the standards. The detector antibody was biotinylated using Biotinylation Kit / Biotin
Conjugation Kit (Fast, Type A) - Lightning-Link® (Abcam, ab201795) according to the manufacturer’s
instructions. The ELISA was performed according to the general matched antibody pair kit protocol in
Human IgA Matched Antibody Pair Kit (Abcam, ab219536). A linear standard curve was used for the
calculation of the results.

Lysozyme
Lysozyme was quantified by sandwich ELISA using Human Human LZM (Lysozyme) ELISA Kit
(Elabscience, E-EL-H1869). A four-parameter logistic curve was used for the calculation of the results.

Zonulin
Zonulin was quantified by sandwich ELISA using Human Zonulin ELISA Kit (elabscience, E-EL-H5560)
according to the manufacturer’s instructions. A four-parameter logistic curve was used for the calculation of

the results.

Statistical analysis
All analyses were performed in R 4.3.1 (2023-06-16) within Rstudio Version 2023.06.2+561 for macOS.

[30]
[32]
[31]
We used R the packages mare , reshape2 , nlme , and gplots . Both microbial absolute abundances
[33]
and biomarker levels were analysed after log transformation to obtain normal distributions. We calculated
the daily changes in both by subtracting the log-transformed abundance on day t from the log-transformed
abundance on day t + 1. Bacteria were analysed at the family and genus levels, including only taxa that were
present at > 0.1% in at least 50% of at least one of the time series (48 genera and 23 families). We used linear
mixed models (function lme) to identify associations between microbial taxa and biomarkers, with the time
series ID, combining information on subject ID and age, as a random factor. The model residuals were
random and normally distributed without temporal patterns. The models were adjusted for total protein
concentration/change and total bacteria change in the sample.

When modelling the associations between biomarker concentrations and bacterial changes, the biomarker
concentrations were normalised by scaling and centring by time series to remove average-level differences
between individuals. Due to the exploratory nature of the study, we chose to define statistical significance as
P < 0.05 without multiple testing adjustment, as it was considered important to find all true associations
(reduce the number of false negatives) even if some false positives may arise. To reduce the number of false
positives, we did not analyse the data at the species level, as we expect related species to have similar
biomarker associations.

RESULTS
We investigated associations between faecal microbiota and immune-related biomarkers in daily time series
of 6 infants sampled at the age of 5-6 and/or 11-12 months. The infants were selected from a larger cohort
for this exploratory study based on diverse microbiota compositions, including Bifidobacterium-dominated,
Bacteroides-dominated, Enterobacteriaceae-dominated, and Clostridia-dominated communities to
maximize the generality of the results despite the small sample size. All infants were breastfed at 5-6 months
and ¾ infants at 11-12 months.

Biomarker concentrations
The average immune-related biomarker concentrations fluctuated during each time series, but typically did
not show strong directional change. They did not differ by birth mode or infant age, apart from Cal, which