Bifidobacterium longum influences gut microbial composition in atopic dermatitis
A metagenomic analysis of 60 fecal samples, comprising 31 from individuals with AD and 29 from healthy controls, was conducted. This analysis incorporated clinical variables including feeding type, mode of delivery, and family history (Table S1 and S2). Gut microbial diversity was significantly distinct between AD and control samples in α- and β-diversity indices. Microbial species composition (β-diversity) based on unweighted UniFrac distance differed substantially between AD and healthy controls (Fig. 1A, p = 0.023). Notably, there were also significant differences between groups in PC1 values, explaining the 22% variance between samples in taxonomic composition. Furthermore, the richness of α-diversity (the number of observed species) was higher in healthy controls than in infants with AD (Fig. 1B, P < 0.05).
Microbial richness strongly correlated with PC1 values (Fig. 1A, P < 10− 10), while the abundance of B. longum was inversely correlated (Fig. 1A, adjusted P < 0.05). However, there was no discernible difference in the abundance of B. longum between the AD and HC groups. Likewise, no differences were found in evenness (Shannon's H) and weighted UniFrac indices. The absence of differences in these indices may be attributed to their susceptibility to changes in microbial abundance. Given that B. longum is a predominant species in the infant gut, it profoundly influences these diversity indices. However, other metrics like the richness and unweighted UniFrac indices, which are not influenced by microbial abundance, indicated significant changes. These results indicate that B. longum may have a direct influence on microbial community composition.
Importantly, B. longum emerged as the sole species to demonstrate a significant association with both α- and β-diversity indices (Table S3), particularly richness and the PC1 value of unweighted UniFrac. This provides substantial evidence that B. longum might indirectly contribute to AD pathogenesis by mediating alterations in the overall microbial community structure (Fig. 1C, P < 0.05).
Distinct Bifidobacterium longum subspecies infantis subclades are enriched in different skin phenotypes
For strain-level analysis, we selected three dominant bacterial species sequenced considerably from a sufficient sample that included all groups (Fig. S1). Among them, we found that only B. longum showed strain-level stratification between healthy controls and AD infants referred to as subclade-I and -II, respectively, along their phylogeny of marker genes (Fig. 2 and Fig. S2). These two subclades harbored B. longum subsp. infantis group, while the rest were undifferentiated and mixed. We further performed a pangenome analysis to understand disease-specific strain enrichment, revealing distinct gene distribution differences in two clades with differential potential functions (Fig. 2).
We tested 258 total KEGG orthologs (KOs) to determine the genes enriched between these two subcaldes of B. longum subsp. infantis group. Eighteen genes were differentiated between subclade-I and II (adjust P < 0.05; Fisher’s exact test), all of which were enriched differently than other B. longum subsp. infantis which were not subclade-I and -II (Fig. 2). These genes are associated with various functions related to each subclade's colonization strategy. The two subclades had different bacterial defense mechanisms against external DNA invasion, including CRISPR-related protein Cas1 (K15342) and the Restriction-Modification (RM) system (K01154, K03427) in subclade-I and -II, respectively. We also identified distinct transposases (K07496, K07483) and a toxin gene (K06218) associated with nich-specific adaptive advantages in each subclade.
In each subclade, three genes are likely involved in gut–brain–axis-associated neuropeptide metabolism. The subclade-I specific gene encoded a glutamine-hydrolyzing enzyme (K01953), which catalyzes the conversion of aspartate to asparagine with the transition of glutamine to glutamate. On the other hand, the subclade-II pangenomes harbored a branched-chain amino acid (BCAA) transporter (K01997, K01998) that could interfere with glutamate production or indirectly regulate serotonin and dopamine synthesis by taking up BCAAs. In addition, dipeptidyl-peptidase (K01278) was present in most B. longum subsp. infantis, with the exception of subclade-I. This gene can degrade several neuropeptides and hormones like incretin.
Multi-omics reveal potential cross-talk between Bifidobacterium longum subclades and the host
We further explored the colonocyte transcriptome associated with B. longum depending on its subclades. To minimize confounding effects, we adjusted feeding type, mode of delivery, and family history as fixed effects, then calculated the correlation between B. longum abundance and host transcripts. 71 and 91 transcripts were significantly correlated with clade-I and -II, respectively (Fig. 3A, P < 0.01, |r| > 0.6). Enrichment analysis was performed on these transcripts to determine the involved pathways. However, while most transcripts were associated with multiple pathways, the neuroactive ligand-receptor interaction pathway (GABRR2, GPR156, HTR1B, GRM4, ADRA2B) was observed with a distinct positive subclade-I correlation. Additionally, subclade-I showed a negative correlation with MAOB, leading to the breakdown of serotonin into its inactive metabolites. Conversely, subclade-II was negatively correlated with CACNA1F, ADCY5, and ADCY9, which are associated with cellular signaling pathways associated with various cellular responses. Additionally, subclade-I correlated with transcripts involved in multiple cellular processes, including cell growth, mammalian target of rapamycin (mTOR) signaling, and insulin regulation (AKT2, TNF, MAPK9). In contrast, subclade-II negatively correlated with a transcript (RPTOR) associated with these pathways and positively correlated with a receptor (NR3C1) for the stress hormone glucocorticoid.
We conducted additional untargeted metabolomic analyses on fecal samples according to B. longum subclade groups to validate the previous results. Despite the limited number of available samples (subclade-I: 8 and -II: 13), the results were consistent with human transcriptome findings. This analysis identified 21 and 30 metabolites differentially correlated with subclade-I and -II, respectively (P < 0.05), and interestingly, metabolites suggestive of opposing effects depending on the subclade were identified (Fig. 3B and Table S4). A positive correlation between subclade-II and tetrahydrocortisol, a metabolite related to the glucocorticoid receptor, was observed, while subclade-I showed a decrease. Furthermore, subclade-I demonstrated simultaneous expression of multiple excitatory (serotonin, glutamate, adrenaline) and inhibitory (GABA) neurotransmitter receptors in colonocytes as B. longum abundance increased, accompanied by an increase in other neurotransmitter precursors, such as orthomine and betaine/valine. In contrast, subclade-II exhibited a decrease in neurotransmitter precursors, including N-palmitoyl-GABA and Gln-Leu-Phe/ Pro-Lys-Tyr tripeptides. Additionally, subclade-II displayed an increase in N-palmitoyl-serine, an antagonist of lysophosphatidic acid receptors, which could lead to a decrease in second messengers such as calcium or cyclic AMP, possibly explaining the reduced expression of CACNA1F, ADCY5, and ADCY9.