Stratification and Variation of the Human Gut Microbiota

The microbial communities that live inside the human gastrointestinal tract -the human gut microbiome- are important for host health and wellbeing. Characterizing this new “organ”, made up of as many cells as the human body itself, has recently become possible through technological advances. Metagenomics, the high-throughput sequencing of DNA directly from microbial communities, enables us to take genomic snapshots of thousands of microbes living together in this complex ecosystem, without the need for isolating and growing them. Quantifying the composition of the human gut microbiome allows us to investigate its properties and connect it to host physiology and disease. The wealth of such connections was unexpected and is probably still underestimated. Due to the fact that most of our dietary as well as medicinal intake affects the microbiome and that the microbiome itself interacts with our immune system through a multitude of pathways, many mechanisms have been proposed to explain the observed correlations, though most have yet to be understood in depth. An obvious prerequisite to characterizing the microbiome and its interactions with the host is the accurate quantification of its composition, i.e. determining which microbes are present and in what numbers they occur. Historically, standard practices have existed for sample handling, DNA extraction and data analysis for many years. However, these were generally developed for single microbe cultures and it is not always feasible to implement them in large scale metagenomic studies. Partly because of this and partly because of the excitement that new technology brings about, the first metagenomic studies each took the liberty to define their own approach and protocols. From early meta-analysis of these studies it became clear that the differences in sample handling, as well as differences in computational approaches, made comparisons across studies very difficult. This restricts our ability to cross-validate findings of individual studies and to pool samples from larger cohorts. To address the pressing need for standardization, we undertook an extensive comparison of 21 different DNA extraction methods as well as a series of other sample manipulations that affect quantification. We developed a number of criteria for determining the measurement quality in the absence of a mock community and used these to propose best practices for sampling, DNA extraction and library preparation. If these were to be accepted as standards in the field, it would greatly improve comparability across studies, which would dramatically increase the power of our inferences and our ability to draw general conclusions about the microbiome. Most metagenomics studies involve comparisons between microbial communities, for example between fecal samples from cases and controls. A multitude of approaches have been proposed to calculate community dissimilarities (beta diversity) and they are often combined with various preprocessing techniques. Direct metagenomics quantification usually counts sequencing reads mapped to specific taxonomic units, which can be species, genera, etc. Due to technology-inherent differences in sampling depth, normalizing counts is necessary, for instance by dividing each count by the sum of all counts in a sample (i.e. total sum scaling), or by subsampling. To derive a single value for community (dis-)similarity, multiple distance measures have been proposed. Although it is theoretically difficult to benchmark these approaches, we developed a biologically motivated framework in which distance measures can be evaluated. This highlights the importance of data transformations and their impact on the measured distances. Building on our experience with accurate abundance estimation and data preprocessing techniques, we can now try and understand some of the basic properties of microbial communities. In 2011, it was proposed that the space of genus level variation of the human gut microbial community is structured into three basic types, termed enterotypes. These were described in a multi-country cohort, so as to be independent of geography, age and other host properties. Operationally defined through a clustering approach, they are “densely populated areas in a multidimensional space of community composition”(source) and were proposed as a general stratifier for the human population. Later studies that applied this concept to other datasets raised concerns about the optimum number of clusters and robustness of the clustering approach. This heralded a long standing debate about the existence of structure and the best ways to determine and capture it. Here, we reconsider the concept of enterotypes, in the context of the vastly increased amounts of available data. We propose a refined framework in which the different types should be thought of as weak attractors in compositional space and we try to implement an approach to determining which attractor a sample is closest to. To this end, we train a classifier on a reference dataset to assign membership to new samples. This way, enterotypes assignment is no longer dataset dependent and effects due to biased sampling are minimized. Using a model in which we assume the existence of three enterotypes characterized by the same driver genera, as originally postulated, we show the relevance of this stratification and propose it to be used in a clinical setting as a potential marker for disease development. Moreover, we believe that these attractors underline different rules of community assembly and we recommend they be accounted for when analyzing gut microbiome samples. While enterotypes describe structure in the community at genus level, metagenomic sequencing can in principle achieve single-nucleotide resolution, allowing us to identify single nucleotide polymorphisms (SNPs) and other genomic variants in the gut microbiome. Analysis methodology for this level of resolution has only recently been developed and little exploration has been done to date. Assessing SNPs in a large, multinational cohort, we discovered that the landscape of genomic variation seems highly structured even beyond species resolution, indicating that clearly distinguishable subspecies are prevalent among gut microbes. In several cases, these subspecies exhibit geo-stratification, with some subspecies only found in the Chinese population. Generally however, they present only minor dispersion limitations and are seen across most of our study populations. Within one individual, one subspecies is commonly found to dominate and only rarely are several subspecies observed to co-occur in the same ecosystem. Analysis of longitudinal data indicates that the dominant subspecies remains stable over periods of more than three years. When interrogating their functional properties we find many differences, with specific ones appearing relevant to the host. For example, we identify a subspecies of E. rectale that is lacking the flagellum operon and find its presence to be significantly associated with lower body mass index and lower insulin resistance of their hosts; it also correlates with higher microbial community diversity. These associations could not be seen at the species level (where multiple subspecies are convoluted), which illustrates the importance of this increased resolution for a more comprehensive understanding of microbial interactions within the microbiome and with the host. Taken together, our results provide a rigorous basis for performing comparative metagenomics of the human gut, encompassing recommendations for both experimental sample processing and computational analysis. We furthermore refine the concept of community stratification into enterotypes, develop a reference-based approach for enterotype assignment and provide compelling evidence for their relevance. Lastly, by harnessing the full resolution of metagenomics, we discover a highly structured genomic variation landscape below the microbial species level and identify common subspecies of the human gut microbiome. By developing these high-precision metagenomics analysis tools, we thus hope to contribute to a greatly improved understanding of the properties and dynamics of the human gut microbiome