In a recent study published in the Cell Host & Microbe Journal, researchers used a diverse set of 5,230 gut metagenomes [gut microbiome reference (GMR) dataset] representing human populations from 13 countries, three continents, and different age groups to train a study model.
Study: Enterosignatures define common bacterial guilds in the human gut microbiome. Image Credit: MeekoMedia/Shutterstock.com
Introduction
They used a metabolic model called non-negative matrix factorization (NMF) to investigate the generalizability of the human gut microbiome, a combination of five enterosignatures (ESSs), which represent bacterial guilds complementary in their metabolism.
The study model helped disentangle the inherent complexity of the human gut microbiome and capture its variability.
It confirmed well-recognized ecological characteristics while enabling the detection of any shifts in ESSs community structures. ESS changes with the host’s age; thus, deviations in gut microbiomes assessed by NMF indicated dysbiotic states detrimental to host health.
Background
All humans harbor 200 to 300 bacterial species in their gastrointestinal tract (GIT). At the strain level, this microbiome composition, however, is unique. This ecosystem is generally in a stable but dynamic equilibrium driven by external and internal processes, e.g., diet.
However, for medical gut microbiome research, it is crucial to disentangle its complexity, implying decomposing the gut microbiome into a small number of universally applicable characteristics.
About the study
In the present study, researchers first trained the 5-ES model on the GMR reference dataset, 74.2% of which comprised Western individuals. Since it potentially limited the generalizability of five ESs, they undertook two validation cohort studies relying on the BMIS (European samples) and NW (no-western) cohorts to evaluate generalizability.
The researchers applied the NMF algorithm to identify ESs signatures, a weighted combination of several genera. They noted that one or two genera strongly dominated some ESs, whereas others were combinations of multiple genera with weaker associations.
Next, they applied the pre-trained model to the Metacardis body mass index spectrum (BMIS) cohort comprising 888 European-origin fecal metagenomes. For comparison, they de novo computed ESs, as were partitioning around medoids (PAMs) or Dirichlet multinomial mixture models (DMMs) ETs.
Further, the researchers investigated temporal ES dynamics in the GMR dataset, from infants to elders and longitudinally for 1,239 individuals. It helped them capture intra-individual short- and long-term changes in the fecal microbiome using ESSs.
Finally, they evaluated 1,152 fecal metagenomes from 12 non-western (NW) countries, representing 119 infants, 278 children, and 740 adults.
The NW samples were distinct in their genus-level composition from the GMR cohort. It helped them test the generalizability of our 5-ES model to samples diverse in age and geography.
Results
The NMF approach-based models recovered three signatures among the five ESs likely representing the community types first described as adult ETs, ES-Bacteroides (Bact), ES-Firmicutes (Firm), and ES-Prevotella (Prev).
Indeed, combinations of ESSs precisely described the gut microbiome composition. ES-Bact often occurred in samples from infants below six months but decreased in the elderly.
Like bacterial diversity, ES diversity increased with age: the frequency of samples colonized by a single ES reduced steadily from infants to adults to elders, and the community became more complex.
Three ESs constituted 39% of adult samples termed adult ES, which ES-Firm, ES-Prev, ES-Bact in 44% of cases in combination with the remaining two ESs.
Further, the GMR dataset suggested that ES-Bact played a crucial role in the recovery of gut ecosystems perturbed antibiotic treatment. Accordingly, its relative abundance increased from 15% to 17% in samples from infants (not preterms) and adults.
Concomitantly, ES-Escherichia and ES-Bifidobacterium in samples from infants decreased from 9% to 7%, respectively, and ES-Prev and ES-Firm in adult samples decreased by 9% and 5%, respectively, post-antibiotic treatments.
Overall, ESs effectively captured short- and long-term gut microbiome changes at an interpretable level.
In the 5,230 samples evaluated in this study, there were 394 ES-atypical samples with a lower evenness and more pathobionts, e.g., Staphylococcus epidermidis, Enterococcus spp.
Investigating ES model fit, the authors discovered that samples from China, Madagascar, and Guinea-Bissau had substantially reduced values, likely because feces collection methods biased community compositions. Helminth infections in 47.3% of the individuals from the Guinea-Bissau cohort perturbed microbiomes and likely lowered ES model fits.
Conclusions
In this study, researchers used NMF, a form of multivariate analysis algorithm, to compose human gut microbiomes into five ESs dominated by Bacteroides, Firmicutes, Prevotella, Bifidobacterium, or Escherichia.
ESs served as an interpretable, generic model for intuitive gut microbiome characterization in health and disease. It demonstrated the augmented functional potential of ES combinations and that ES-Bact had a central role in establishing and maintaining core gut functionality.
Furthermore, this metabolic model enabled the detection of atypical fecal microbiomes associated with infant birth mode and antibiotic use in adults, thereby facilitating the detection of dysbiotic gut microbiomes.
Changes in gut microbiome could signal onset of rheumatoid arthritis