Page 24 - Read Online

Page 6 of 15 Ladeira et al. Microbiome Res Rep 2023;2:9 https://dx.doi.org/10.20517/mrr.2023.01

Bifidobacterium-based partitioning of the human gut microbiome
We further explored the ecology and variation of Bifidobacterium species between subjects. We used the
[24]
Dirichlet multinomial mixtures (DMM) partitioning method , which is commonly used to identify
partitions of the human gut microbiome; however, in this case, we applied it exclusively to Bifidobacterium
species. This made it possible to focus specifically on the association of the variable within-Bifidobacterium
community distribution with the ecological features of the gut microbiome, lifestyle, and health. We applied
DMM to 5.329 subjects (see methods). On the basis of BIC minimization, k = 6 was chosen for individuals
for whom Bifidobacterium was detected [Figure 1A]. We added a group (k = 7) corresponding to subjects
for whom no Bifidobacterium reads were detected (n = 380). All partitions had a median number of reads
above 30 M [Supplementary Table 1].

Partitions #1 and #2 accounted for 48% of subjects, whereas the partition corresponding to the detection of
no Bifidobacterium species (partition #7) accounted for the smallest number of subjects (17%) [Figure 1B].
We extracted the scaled contribution of each Bifidobacterium species to each partition, which reflects the
relative abundance of these species within Bifidobacterium. Some of the dominant Bifidobacterium species
(shown in yellow) differed in abundance between partitions [Figure 1C].

In partitions #1, #2, and #6, both B. adolescentis and B. longum were abundant, whereas in partitions #3, #4,
and #5, either B. adolescentis or B. longum was the dominant species. Partition #1 consisted mostly of B.
adolescentis and B. longum (ado_lon), whereas partition #2 also included B. bifidum (ado_lon _bif), and
partition #6 included B. catenulatum (lon_ado_cat). Partition #3 composition was dominated by both B.
longum and B. pseudocatneulatum (lon_pse), whereas partition #4 consisted mostly of B. longum (lon).
Partition #5 was dominated by both B. adolescentis and pseudocatneulatum (ado_pse) and contained
multiple sub-dominant species. For the less abundant species, B. breve, B. animalis, and B. dentium, relative
abundance was highest in partitions #4 (lon) and #5 (ado_pse).

Association between Bifidobacterium-based partitions and the gut microbiome and health
We then investigated the distribution of Bifidobacterium-based partitions as a function of age category,
lifestyle (westernized/non-westernized), and health conditions [Figure 2A-C and Supplementary Figure 4].
The associations of Bifidobacterium partitions with health status (healthy or with one of the health
conditions considered), lifestyle (westernized/ non-westernized), and age category were significant (chi-
squared, P-value < 0.05).

The prevalence of seniors and subjects with a non-western lifestyle was higher in the lon and no_Bif
partitions, respectively [Figure 2C]. The Bifidobacterium partitions lon_pse, lon, ado_pse, and no_Bif
included ~70% healthy subjects, whereas more than 90% of the subjects in the Bifidobacterium partitions
ado_lon, ado_lon_bif, and lon_ado_cat were healthy. We also found that 64% of the healthy subjects
belonged to the Bifidobacterium partitions ado_lon, ado_lon_bif, and lon_ado_cat, whereas 74% of the
subjects with health conditions belonged to the Bifidobacterium partitions lon_pse, lon, ado_pse, and no_Bif.

We then investigated the association of the Bifidobacterium partitions with the gut microbiome. We found
that Bifidobacterium partitions were significantly associated with Bifidobacterium abundance [Figure 3A].
and gut microbiome Shannon diversity [Figure 3B] (Kruskal-Wallis, P < 0.001). Bifidobacterium partitions
ado_lon, ado_lon_bif, and lon_ado_cat had a higher relative abundance of Bifidobacterium, and a higher gut
microbiome α-diversity (Shannon index) than the other partitions [Supplementary Table 1] (Mann-
Whitney test, P < 0.001). Given that cMD also includes infants less than one-year-old, which may influence
alpha-diversity results, we further investigated whether partitions could be detected in single adult cohorts
in which samples were processed with the same analytical procedure. We selected three large cohorts from