Page 22 - Read Online

Page 4 of 15 Ladeira et al. Microbiome Res Rep 2023;2:9 https://dx.doi.org/10.20517/mrr.2023.01

DMM models were calculated for different numbers (k) of clusters, k ∈ [1,30], and evaluated with the
Bayesian information criterion (BIC) and five different seeds. These methods are based on minimizing a
penalized criterion, taking into account model fit and complexity. We chose three random seeds and
calculated the minimum k for different model fits, and then selected the most frequently observed. We
determined the contribution of each Bifidobacterium species to each DMM cluster from the calculated
models.

Functional analysis of Bifidobacterium MAGs
We retrieved 3,973 metagenomic-assembled genomes (MAGs) assigned to 15 Bifidobacterium species from
http://opendata.lifebit.ai/table/?project=SGB. The MAGs were previously decontaminated and
[16]
taxonomically assigned by Mash . The study identifier, sample identifier, assigned species, and
completeness were collected for each MAG. Prodigal was used for gene calling for each MAG, and more
than six million genes were called. The computation time required for annotation was decreased by
clustering the MAG gene against a non-redundant gut Bifidobacterium gene catalog, using CD-HIT at 95%
nucleotide identity, with a minimum sequence overlap of 90%. Non-redundant Bifidobacterium genes were
[26]
[25]
annotated with EggNOG 5.0 and dbCAN version 3 to obtain orthologous genes (OGs) and CAZy
families, respectively. Quality was ensured by selecting the MAGs with completeness > 80%. Bifidobacterium
species with at least 30 associated MAGs were selected, giving a total of six species. 820 MAGs were assigned
to B. longum, 700 to B. adolescentis, 339 to B. bifidum, 178 to B. pseudocatenulatum, 54 to B. catenulatum,
and 34 to B. dentium. Pairwise distances were calculated for all MAGS within each species, with Mash
v2.347 and the default sketch size. Hierarchical clustering was then performed for each species with the
“ward.D2” method and the “pheatmap” R package (1.0.12).

Statistical analysis
The associations between Bifidobacterium partitions and quantitative variables (notably α-diversity and
Bifidobacterium abundances) were analyzed with Kruskal-Wallis tests and a post-hoc test (Mann-Whitney
test, adjusted for FDR). Pearson’s chi-squared test was used to determine whether 1) Bifidobacterium
partitions were associated with categorical variables (age, lifestyle, health status) and 2) whether the
prevalence of OGs in MAGs for each species was associated with health-associated partitions, adjusted for
FDR (within species). DESeq2 (v1.28.1) was used to identify bacterial species for which abundance differed
between Bifidobacterium partitions with the “poscounts” normalization option to accommodate the sparsity
of microbiota data. The global effects of the Bifidobacterium were estimated in likelihood ratio tests and
Wald tests for pairwise comparisons of clusters. A FDR correction for multiple testing was applied to each
test to account for the number of species tested. Log fold-changes in expression are expressed as the
2
estimate ± standard error. When specified, FDR corrections were applied with the Benjamini-Hochberg
procedure.

RESULTS
Analysis of pooled metagenomic studies recapitulates major findings of human gut Bifidobacterium
ecology
We used the “curatedMetagenomicData” (cMD, version 3) database (Pasolli et al.) to study the ecology of
the Bifidobacterium community in the human gut microbiome. The cMD provides standardized, curated
human microbiome data with several pieces of metadata per participant. We selected only gut metagenomes
(one per subject) and obtained 9.515 unique samples [Table 1].

This dataset contains predominantly data for adults with a westernized lifestyle. Individuals under the age of
19 years accounted for less than 8% of this dataset [newborns (2.9%), children (3.4%), and school-age
(1.4%)] [Table 1 and Supplementary Figure 1].