Are Your Gut Bacteria Sorted Into Types — Or Is It More Complicated Than That?

For a while, the science seemed almost elegantly simple: your gut microbiome belongs to one of three tribes. In 2011, Arumugam and colleagues published a landmark study proposing that human gut microbial communities cluster into three distinct “enterotypes,” each dominated by a different bacterial genus — Bacteroides, Prevotella, or Ruminococcus. The analogy to blood groups was irresistible, and the media ran with it. Your gut had a type, just like your blood.

The trouble is, that may never have been quite right.

The Original Claim

The 2011 paper used principal component analysis (PCA) and clustering algorithms on fecal metagenomic data from subjects across multiple countries. The authors found what appeared to be three densely populated regions in the multidimensional space of microbial community composition — three enterotypes, seemingly independent of age, sex, nationality, or body mass index. The finding suggested that, beneath all the individual variation in gut bacteria, humans sorted into a handful of discrete community types.

The clinical appeal was obvious. If gut microbiomes fell into defined categories — like blood types — you could potentially use enterotype to guide diagnostics, predict drug responses, or tailor dietary interventions. A tidy framework for an otherwise bewildering ecosystem.

The Cracks Appear

Replication was messy from the start. Subsequent studies found two enterotypes, or four, or a gradient — depending on the dataset, the method, and who was doing the analysis. A 2017 perspective by Costea and colleagues (including many of the original authors) revisited the concept with three large datasets and arrived at a more nuanced conclusion: yes, there are preferred regions of microbial community composition, but the boundaries between them are fuzzy, and a substantial fraction of people fall between putative clusters rather than clearly within one.

Jeffery and colleagues had already flagged the conceptual problem in 2012: at the phylum level, gut microbiome variation is clearly continuous — a gradient between Bacteroides-dominated and Prevotella-dominated communities, not a set of discrete bins. The Ruminococcus enterotype, in particular, seemed increasingly hard to pin down.

The Methodological Problem

The sharpest critique came from two directions. Knights and colleagues (2014) pointed out that the same clustering algorithms used to “discover” enterotypes will produce clusters from continuous data too — the method forces groupings whether or not real groupings exist. They showed that an individual’s apparent enterotype can shift substantially over the course of a year, traversing much of the community composition space, which is hard to reconcile with the idea of stable discrete types. They also demonstrated that apparent clusters could be artifacts of who you sampled: include infants and adults together, and you get age-driven clusters; remove Americans from a three-country study, and you get nationality-driven clusters. The clusters followed the sampling frame, not some underlying biological truth.

The most technically rigorous challenge came from Bulygin and colleagues (2021), who applied non-linear dimensionality reduction methods — t-SNE and UMAP — rather than the linear PCA used in the original work. The argument matters: PCA finds the best flat projection of high-dimensional data. If the true structure is curved or complex, PCA can impose apparent clusters where none exist. Using the Human Microbiome Project and American Gut Project datasets with multiple non-linear embedding techniques and clustering algorithms, Bulygin et al. found no stable, distinct clusters that could qualify as enterotypes. What they did find was that gut microbial communities vary continuously along a low-dimensional manifold — a curved surface in a high-dimensional space, not a set of separated islands.

Where Things Stand

The enterotype concept has not been abandoned so much as reframed. Costea et al. argue it still has descriptive utility: Bacteroides and Prevotella remain the dominant axes of variation, functional differences exist across the composition spectrum, and stratifying patients by community type may still be clinically useful even if the boundaries are soft. But the original vision — discrete, stable community types analogous to blood groups — appears to have been an artifact of applying the wrong analytical tools to a continuous distribution.

The gut microbiome is better thought of as a landscape with hills and valleys than as a map divided into labeled territories. Most people live somewhere on the slopes.

References: Arumugam et al., Nature 2011; Costea et al., Nature Microbiology 2018; Knights et al., Cell Host & Microbe 2014; Jeffery et al., Nature Reviews Microbiology 2012; Bulygin et al., bioRxiv 2021.