The Human Microbiome Project (HMP) [1],[2] is a concept that was long in the making. After the Human Genome Project, interest grew in sequencing the “other genome" of microbes carried in and on the human body [3],[4]. Microbial ecologists, realizing that >99% of environmental microbes could not be easily cultured, developed approaches to study microorganisms in situ [5], primarily by sequencing the 16S ribosomal RNA gene (16S) as a phylogenetic and taxonomic marker to identify members of microbial communities [6]. The need to develop corresponding new methods for culture-independent studies [7],[8] in turn precipitated a sea change in the study of microbes and human health, inspiring the new term “metagenomics" [9] both to describe a technological approach—sequencing and analysis of the genes from whole communities rather than from individual genomes—and to emphasize that microbes function within communities rather than as individual species. This shift from a focus on individual organisms to microbial interactions [10] culminated in a National Academy of Science report [11], which outlined challenges and promises for metagenomics as a way of understanding the foundational role of microbial communities both in the environment and in human health.

Any study of human populations must put both subject protection and study design first, and the HMP was no exception. Power calculations for microbiome studies in human cohorts are particularly challenging, as they must simultaneously address assay types (e.g., 16S versus shotgun), depth of sequencing, taxon detection, and fold abundance changes in clades, genes, or pathways of interest [28]–[31]. After study design, as the HMP spanned multiple sequencing centers over a prolonged duration, the group established standardized and benchmarked protocols for sample collection [2], handling, and subsequent 16S profiling [32]. Metagenomic library construction was likewise standardized among centers, and stringent quality control was aided by the optimization of 16S read processing [33] and by improved taxonomic frameworks for classification of microbial sequences prior to biological interpretation [34].

The HMP was designed in part to address a key question about our microbial selves: do all humans have an identifiable “core" microbiome of shared components comparable to our shared genome [50]? Several definitions of “core" have been proposed, recently unified in one conceptual framework [51]. Earlier studies reported that different people shared few microbes in their gut and skin microbiota [17],[52]–[56], a greater fraction of their oral microbiota [56],[57], or might be classifiable into multiple core microbiomes based on vaginal [20] and gut communities [58]. The HMP provides a comprehensive picture of the human microbiome covering multiple body sites and thus an in-depth exploration of these concepts. The study confirmed high inter-individual variation [59] and showed that even rare organisms in these communities are important reservoirs of genetic diversity [60]. Additionally, the large HMP cohort shows that the composition of the gut microbiome rarely clusters subjects into discrete types, as was suggested before on more limited data [61]; although other habitats such as the vagina can exhibit such clustering [20], the gut was most often characterized by smooth abundance gradients of key organisms [2].

Data from individuals without overt signs of disease serve as an excellent reference for disease-associated microbiome studies, while also providing a comprehensive baseline for comparison of Western populations with disparate geographic, ethnic, and genetic cohorts [63]. The adoption of uniform sampling, nucleic acid extraction, sequencing, and analysis protocols is an important step in such integration, with some success already realized in, for example, several aspects of autoimmune disease. The inflammatory bowel diseases have long been linked to the human gut microbiome [22], with integration of host genotype, gene expression, and microbial membership now suggesting mediation of specific host-microbial interactions by human gene products as well as by host environment [64],[65]. Bacteria are of course not the only mediators of dysbiotic disease, and metagenomic approaches can also be used to identify potential viral etiologies (e.g., in pediatric fever of undefined origin [66]). Likewise the “healthy" microbiome provides a baseline not only for integration with disease-related studies, but for broader populations such as a recent comparison using HMP protocols among a cohort of pregnant women [67]. The normal variation of the microbiome within healthy states and its potential misregulation in disease is thus being pursued in earnest, as related laboratory and computational methods continue to be adapted to better characterize the impact of bacteria, archaea, viruses, and fungi throughout human body habitats.