Rapid evolution of the human gut virome Samuel Minota, Alexandra Brysona, Christel Chehouda, Gary D. Wub, James D. Lewisb,c, and Frederic D. Bushmana,1 aDepartment of Microbiology, bDivision of Gastroenterology, and cCenter for Clinical Epidemiology and Biostatistics, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, PA 19104 Edited by Sankar Adhya, National Institutes of Health, National Cancer Institute, Bethesda, MD, and approved May 31, 2013 (received for review January 15, 2013) Humans are colonized by immense populations of viruses, which sequenced independently to allow estimation of within-time point metagenomic analysis shows are mostly unique to each individual. sample variation. Virus-like particles were extracted by sequential To investigate the origin and evolution of the human gut virome, filtration, Centricon ultrafiltration, nuclease treatment, and sol- we analyzed the viral community of one adult individual over 2.5 y vent extraction. Purified viral DNA was subjected to linear am- by extremely deep metagenomic sequencing (56 billion bases of plification using Φ29 DNA polymerase, after which quantitative purified viral sequence from 24 longitudinal fecal samples). After PCR showed that bacterial 16S sequences were reduced to less assembly, 478 well-determined contigs could be identified, which than 10 copies per nanogram of DNA, and human sequences were are inferred to correspond mostly to previously unstudied bacterio- reduced to below 0.1 copies per nanogram, the limit of detection. phage genomes. Fully 80% of these types persisted throughout the Paired-end reads then were acquired using Illumina HiSeq se- duration of the 2.5-y study, indicating long-term global stability. quencing, yielding more than 573 million reads (Q ≥ 35; mean Mechanisms of base substitution, rates of accumulation, and the read length, 97.5 bp), with 15–39 million reads per sample (Table amount of variation varied among viral types. Temperate phages S1). No attempt was made to study gut RNA viruses, which also showed relatively lower mutation rates, consistent with replication are known to exist, although some samples were dominated by by accurate bacterial DNA polymerases in the integrated prophage abundant plant RNA viruses ingested with food (15). state. In contrast, Microviridae, which are lytic bacteriophages with Sequence reads from each sample were first assembled in- single-stranded circular DNA genomes, showed high substitution dividually using MetaIDBA (16). When reads were aligned back − rates (>10 5 per nucleotide each day), so that sequence divergence onto contigs generated within each sample, only 71% of reads over the 2.5-y period studied approached values sufficient to distin- could be aligned. Improved contigs then were generated using guish new viral species. Longitudinal changes also were associated a hybrid assembly method combining all samples, taking advan- MICROBIOLOGY with diversity-generating retroelements and virus-encoded Clus- tage of the fact that viruses that are rare at one time point may tered Regularly Interspaced Short Palindromic Repeats arrays. We be abundant at another. After this step, 97.6% of the reads could infer that the extreme interpersonal diversity of human gut viruses be aligned to contigs, allowing assessment of within-contig di- derives from two sources, persistence of a small portion of the versity. Rarefaction (collector’s curve) analysis showed that the global virome within the gut of each individual and rapid evolution detection of these contigs was saturated at <107 reads per sample of some long-term virome members. and at 7–10 samples (Fig. 1B), well below our sampling effort. After quality filtering and manual editing, 478 contigs showed metagenomics | microbiome | diversity generating retroelement | CRISPR >20-fold coverage (median, 82-fold); from the purification re- sults, we infer these contigs to be mostly or entirely DNA viruses here are an estimated 1031 viral particles on earth, and human (Fig. 1C). Sixty contigs assembled as closed circles (ranging in – Tfeces contain at least 109 virus-like particles per gram (1–3). size from 4 167 kb), an indication of probable completion of Many of these are identifiable as viruses that infect bacteria (bac- these genome sequences, providing an estimate of the viral pop- teriophages), but the great majority remains unidentified. Even ulation size and composition in unprecedented detail. One circular today, gut virome samples taken from different human individuals genome was sequenced independently using the Sanger method fi still yield mostly novel viruses (4–8), and only a small minority of and was con rmed to have the structure predicted from the Sol- viral ORFs resembles previously studied genes (7). exa/Illumina data (SI Methods). The abundance of each contig at Bacteriophages are of biomedical importance because of their each time point was measured by the proportion of reads that ability to transmit genes to their bacterial hosts, thereby con- aligned to it, normalized to the length of each contig. The corre- fi ferring increased pathogenicity, antibiotic resistance, and per- lation coef cient between replicate samples from the same time haps new metabolic capacity (4, 5, 9, 10). Despite their importance, point was at least 0.99, indicating a high degree of reproducibility the forces diversifying bacteriophage genomes in human hosts (Fig. S1). have not been studied in detail. Humans show considerable in- Viral Groups Detected. Taxonomic analysis of these contigs in- dividual variation in the bacterial lineages present in their guts dicated recovery of Microviridae, Podoviridae, Myoviridae, and (11–13); this variation likely is one reason for the differences in Siphoviridae, but contigs with taxonomic attributions were a mi- their phage predators (5–8, 14). The large differences in phage nority, only 13%, emphasizing the enormous sequence variation populations among individuals also may be influenced by within- present in bacteriophages. Microviridae (the group including individual viral evolution. To investigate the origin and nature of human viral populations, we carried out a detailed study of a single human gut viral com- Author contributions: S.M., A.B., G.D.W., J.D.L., and F.D.B. designed research; S.M. and munity. Ultra-deep longitudinal analysis of DNA sequences from A.B. performed research; S.M., A.B., C.C., and F.D.B. analyzed data; and S.M., A.B., and the viral community, combined with characterization of the host F.D.B. wrote the paper. bacteria, revealed rapid change over time and begins to specify The authors declare no conflict of interest. some of the mechanisms involved. This article is a PNAS Direct Submission. Results Data deposition: The sequences reported in this paper have been deposited in the National Center for Biotechnology (NCBI) database, www.ncbi.nlm.nih.gov (accession Sample Collection, Viral Purification, and DNA Sequencing. Stool no. SRP021107). samples (n = 24) were collected from a healthy male at 16 time 1To whom correspondence should be addressed. E-mail: [email protected]. points spread over 884 days (Fig. 1A). For eight of the time This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. points, two separate samples taken 1 cm apart were purified and 1073/pnas.1300833110/-/DCSupplemental. www.pnas.org/cgi/doi/10.1073/pnas.1300833110 PNAS Early Edition | 1of6 Downloaded by guest on September 29, 2021 A Day 0 180-184 851-855 879-883 Bacterial community membership and taxonomic proportions showed only modest variation over time. Longitudinal Base Substitution in Viral Contigs. Replicates 2 1 1 2 1 1 2 2 2 1 1 2 2 2 1 1 The depth of se- quence information available and the quality of the viral contigs 2 years 5 months allowed a detailed assessment of the rates of accumulation of base substitutions. For each viral contig at each time point, the BCCircular 750 1,000,000 Linear 100,000 A 500 10,000 0.95 950 Contigs 900 1,000 250 Fold coverage 850 0.90 Contigs 800 100 750 0 0 5 10152025 0.85 Samples 0e+00 1e+07 2e+07 3e+07 4e+07 4 4.5 5 10 10 10 Reads Contig Length (bp) 0.80 Jaccard Index Fig. 1. Longitudinal analysis of the human gut virome from a single in- dividual. (A) Timeline of sample collection. Note that at some time points, two 0.75 separate portions of the stool sample, taken approximately 1 cm apart, were processed and sequenced independently to assess reproducibility. (B)Rare- faction analysis of sampling depth by number of reads; detection of each 0.70 contig is scored as positive if 50% of the contig is covered by sequence reads. (Inset) Contig recovery. The x-axis is the number of samples included (black (shared membership between paired time points) line: 2 million reads; blue line: 15 million reads). (C) Contig spectrum, relating 1 10 100 850 the lengths of the contigs assembled in bas pairs (x-axis) to the depth of Interval between time points coverage (y-axis). Circular contigs are shown as blue and linear contigs as red. compared (days) ΦX174) predominated, but this predominance could be a con- B sequence of favored amplification by Φ29 polymerase of the Not temperate Temperate small circular genomes that characterize this group. 10-4.5 The most abundant contigs were mostly retained over the du- ration of the experiment. Because there are many possible pair- wise comparisons between time points, distances between time 10-5 points analyzed (Fig. 2A, x-axis) were compared with Jaccard index values (Fig. 2A, y-axis), which score shared membership, over all of the possible pairwise comparisons of time points. On average, more than 80% of contigs were found in common be- 10 -5.5 tween the time points separated by 850 d (points at the right side of the plot), the longest time intervals compared. No contigs corresponded to known viruses infecting eukaryotic 10-6 cells.