A catalog of tens of thousands of viruses from human metagenomes reveals hidden associations with chronic diseases

Despite remarkable strides in microbiome research, the viral component of the microbiome has generally presented a more challenging target than the bacteriome. This gap persists, even though many thousands of shotgun sequencing runs from human metagenomic samples exist in public databases, and all of them encompass large amounts of viral sequence data. The lack of a comprehensive database for human-associated viruses has historically stymied efforts to interrogate the impact of the virome on human health. This study probes thousands of datasets to uncover sequences from over 45,000 unique virus taxa, with historically high per-genome completeness. Large publicly available case-control studies are reanalyzed, and over 2,200 strong virus–disease associations are found.

The human virome is the sum total of all viruses that are intimately associated with people. This includes viruses that directly infect human cells (1, 2) but mostly consists of viruses infecting resident bacteria (i.e., phages) (3). While the large majority of microbiome studies have focused on the bacteriome, revealing numerous important functions for bacteria in human physiology (4), information about the human virome has lagged. However, a number of recent studies have begun making inroads into characterizing the virome (5–13).Just as human-tropic viruses can have dramatic effects on people, phages are able to dramatically alter bacterial physiology and regulate host population size. A variety of evolutionary dynamics can be at play in the phage/bacterium arena, including Red Queen (11), arms-race (14), and piggyback-the-winner (15) relationships, to name just a few. In the gut, many phages enter a lysogenic or latent state and are retained as integrated or episomal prophages within the host bacterium (16). In some instances, the prophage can buttress host fitness (at least temporarily) rather than destroy the host cell. To this effect, prophages often encode genes that can dramatically alter the phenotype of the bacteria, such as toxins (17), virulence factors (18), antibiotic resistance genes (19), photosystem components (20), other auxiliary metabolic genes (21), and CRISPR-Cas systems (22), along with countless genes of unknown function. Experimental evidence has shown that bacteria infected with particular phages (i.e., “virocells”) are physiologically distinct from cognate bacteria that lack those particular phages (21).There have been a few documented cases in which phages have been shown to be mechanistically involved in human health and disease, sometimes through direct interactions with human cells. This includes roles in increased bacterial virulence (17), response to cancer immunotherapy (23), clearance of bacterial infection (24), and resistance to antibiotics (25). Furthermore, phage therapy, the targeted killing of specific bacteria using live phage particles, has shown increasing promise for treatment of antibiotic-resistant bacterial infections (26). Considering the progress already made, phages represent attractive targets of and tools for microbiome restructuring in the interest of improving health outcomes.In addition, several studies have conducted massively parallel sequencing on virus-enriched samples of human stool, finding differential abundance of some phages in disease conditions (6, 27–29). A major issue encountered by these studies is that there is not yet a comprehensive database of annotated virus genome sequences, and de novo prediction of virus sequences from metagenomic assemblies remains a daunting challenge (3). Further, though some tools are able to predict virus-derived sequences with high specificity (30, 31), these tools have not been applied to human metagenomes at a large scale [with a possible exception (13)], and, regrettably, most uncovered virus genomes do not end up in central repositories. One study suggests that only 31% of the assembled sequence data in virion-enriched virome surveys could be identified as recognizably viral (32). On the other hand, another study of 12 individuals was able to recruit over 80% of reads from virus-enriched samples to putative virus contigs (11). Still, most of the potential viral contigs from this study were unclassifiable sequences, and a large majority of contigs appeared to represent subgenomic fragments under 10 kb.The current study sought to overcome the traditional challenges of sparse viral databases and poor detection of highly divergent viral sequences by using Cenote-Taker 2, a new virus discovery and annotation tool (33). The pipeline was applied to sequencing data from nearly 6,000 human metagenome samples. Strict criteria identified over 180,000 viral contigs representing 45,033 specific taxa. In most cases, 70 to 99% of reads from virus-enriched stool datasets could be back-aligned to the Cenote-Taker 2–compiled Human Virome Database. Furthermore, the curated database allowed read-alignment–based abundance profiling of the virome in human metagenomic datasets, enabling the reanalysis of a panel of existing case-control studies. The reanalysis revealed previously undetected associations between chronic diseases and the abundance of 2,265 specific virus taxa.