A guide to enterotypes across the human body: meta-analysis of microbial community structures in human microbiome datasets.
Koren O, Knights D, Gonzalez A, Waldron L, Segata N, Knight R, Huttenhower C, Ley RE. · 2013
Read A guide to enterotypes across the human body: meta-analysis of microbial community structures in human microbiome datasets. by Koren O, Knights D, Gonzalez A, Waldron L, Segata N, Knight R, Huttenhower C, Ley RE. (2013) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc3542080.txt.
Introduction
Gut microbiome types may be an artifact of how you cluster, and Omry Koren leads a nine-author team stress-testing the enterotype idea in 2013. They re-analyze Human Microbiome Project data plus sixteen other studies, varying clustering methods, distance measures, marker picking, sequencing depth, and data type.
At most body sites, key bacterial genera form smooth abundance gradients, not discrete camps. The gut shows a two-humped distribution and the vagina several humps, but not every workflow reveals even those. Enterotype detection, they conclude, depends as much on analytic choices as on biology. The guide warns microbiome researchers against carving continua into types and then believing the carving.
Catalog record
| CREATOR | Koren O, Knights D, Gonzalez A, Waldron L, Segata N, Knight R, Huttenhower C, Ley RE. |
|---|---|
| YEAR | 2013 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 65 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Guide to Enterotypes across the Human Body: Meta-Analysis of Microbial Community Structures in Human Microbiome Datasets By Omry Koren, Dan Knights, Antonio Gonzalez, Levi Waldron, Nicola Segata, Rob Knight, Curtis Huttenhower, Ruth E Ley, Jonathan A Eisen Published in: PLoS Computational Biology DOI: 10.1371/journal.pcbi.1002863 ABSTRACT Abstract Recent analyses of human-associated bacterial diversity have categorized individuals into ‘enterotypes’ or clusters based on the abundances of key bacterial genera in the gut microbiota. There is a lack of consensus, however, on the analytical basis for enterotypes and on the interpretation of these results. We tested how the following factors influenced the detection of enterotypes: clustering methodology, distance metrics, OTU-picking approaches, sequencing depth, data type (whole genome shotgun (WGS) vs.16S rRNA gene sequence data), and 16S rRNA region. We included 16S rRNA gene sequences from the Human Microbiome Project (HMP) and from 16 additional studies and WGS sequences from the HMP and MetaHIT. In most body sites, we observed smooth abundance gradients of key genera without discrete clustering of samples. Some body habitats displayed bimodal (e.g., gut) or multimodal (e.g., vagina) distributions of sample abundances, but not all clustering methods and workflows accurately highlight such clusters. Because identifying enterotypes in datasets depends not only on the structure of the data but is also sensitive to the methods applied to identifying clustering strength, we recommend that multiple approaches be used and compared when testing for enterotypes. INTRODUCTION Together with the MetaHIT consortium [1], the Human Microbiome Project (HMP) represents one of the first major attempts to define the microbial diversity comprising the “normal healthy” human microbiome [2]. The HMP dataset includes 16S rRNA gene sequence data of roughly twice the size of all similarly derived data in previously published studies, effectively tripling the size of combined data available for comparative studies (Table S1). In addition, the HMP generated whole-genome shotgun (WGS) metagenomic data for a subset of individuals. These data allowed for the characterization of patterns of microbial diversity across body sites and between individuals [2]. The HMP data also provides an opportunity to test the generality of the concept of enterotypes in the human microbiome. Arumugam et al. first articulated the concept of enterotypes as robust clustering of human gut samples based on microbial community composition, and largely driven by the abundances of key bacterial genera [3]. Although the term ‘enterotype’ refers to microbiota types within the gut, the concept can be applied generally, and here, for convenience, we use the term ‘enterotype’ to refer to microbiota types across different body sites. The HMP data are ideally suited to test the robustness of the enterotype concept in multiple body sites, and together with recently published community-generated datasets, across multiple populations. In this report, we combined 16S rRNA gene sequence data generated using next-generation sequencing by the scientific community (hereafter, ‘community data’) together with the MetaHIT WGS data [3] and the recently released HMP 16S rRNA gene sequence data and WGS data [2]. Because there is currently no community standard for testing for enterotypes, we explore how the detection of enterotypes is affected by the following: clustering methodology, distance metrics, OTU-picking approaches, sequencing depth (i.e., rarefaction), data type (16S rRNA vs. WGS), and the specific region of the 16S rRNA gene sequenced. We find that the emergence of enterotypes is sensitive to the community structure of communities within each body site, and importantly also to the analysis methods employed. Our comparative analysis of various approaches across datasets informs the discussion on the technical basis for enterotyping and on how to interpret enterotype results. MATERIALS AND METHODS 16S rRNA Gene Sequence Processing We constructed a database containing the recently released HMP 16S rRNA gene sequence data [4] and publically available (published) human microbiome datasets (community data). For inclusion, community datasets were required to contain a minimum of 25 samples per study and sequences generated using the Roche 454 platform (Table S1). The majority of samples were from healthy controls; however, a small subset of samples was derived from subjects that differed from adult healthy subjects due to age (i.e., infants and the elderly), use of antibiotics, or possible presence of disease (Fig. S1). We acquired raw SFF files and metadata files containing the unique identifiers for each sample within a study (barcodes) from the authors and re-processed the data using the default settings in the Quantitative Insights Into Microbial Ecology (QIIME) analysis pipeline [5]. For the majority of samples, quality filtering consisted of rejecting reads <200 nt and >1000 nt, excluding homopolymer runs >6 nt, accepting 0 barcode corrections and 0 primer mismatches; two datasets were processed with slightly different screening parameters, as described in their respective publications [6], [7]. When picking operational taxonomic units (OTUs) we used the OTU tables generated by the HMP, which were created de novo. Because the regions of the 16S rRNA gene differed between studies (and within: the HMP sequenced both V1–V3 and V3–V5 regions), we used a reference-based approach (hence, we did not denoise the data) to pick OTUs at 97% pairwise identity using as a reference the latest release of the GreenGenes (GG) taxonomy [8]. We also used the phylogenetic tree from GG to calculate weighted (abundance based) and unweighted (presence/absence based) UniFrac distances between communities [9], after applying two rarefactions (1,000 and 2,000 sequences/sample) to standardize sequence counts. Principal coordinates analysis (PCoA) was applied to the distance matrices for visualization. Metagenomic Data Processing The HMP and MetaHIT shotgun metagenomic datasets were taxonomically profiled using MetaPhlAn [10] (version 1.1, default parameter settings), which infers relative abundances for all taxonomic levels (from phyla to species) for Bacteria and Archaea. We performed standard quality control on the HMP and MetaHIT samples as reported in the original studies [2], [3] - other metagenomic pre-processing steps (e.g., error detection, assembly, or gene annotation) are not required by MetaPhlAn. The taxonomic profiles of HMP metagenomes are available at http://www.hmpdacc.org/HMSMCP/, and the MetaHIT profiles can be downloaded from http://huttenhower.sph.harvard.edu/metaphlan/. The 690 HMP metagenomic samples from 7 different body sites can be accessed at http://hmpdacc.org/HMASM/, from which we used the ‘WGS’ reads (i.e., we did not use the ‘PGA’ assemblies), collapsing multiple visits from the same individual into one sample. The 124 fecal samples from MetaHIT were downloaded from the European Nucleotide Archive (http://www.ebi.ac.uk/ena/, study accession number ERP000108). Enterotyping To evaluate the clustering results in the context of previously published results reporting enterotypes, we merged publicly available data (genus relative abundance tables) from MetaHIT [3] with data for the 16S rRNA-based HMP and non-HMP samples, based on genus-level taxonomy assignments. We performed enterotype testing using the relative abundances of OTUs (rarified at 1,000 sequences/sample for the majority of analyses, except where effect of rarefaction was tested specifically), to which we applied five distance metrics: Jensen-Shannon divergence (JSD), Root Jensen-Shannon divergence (rJSD), Bray-Curtis (BC), and weighted/unweighted UniFrac distances. For the calculation of JSD and BC distances, we first binned the counts of OTUs at the desired level (95% and 97% ID for genus and species level OTUs, respectively). We used the R “vegan” package [11] for calculating the Bray-Curtis distance according to this formula for the distance between samples j and k, with taxa/OTUs indexed by i: Clustering was performed via partitioning around medoids