FEATURE ARTICLE RESEARCH A comprehensive and quantitative exploration of thousands of viral genomes Abstract The complete assembly of viral genomes from metagenomic datasets (short genomic sequences gathered from environmental samples) has proven to be challenging, so there are significant blind spots when we view viral genomes through the lens of metagenomics. One approach to overcoming this problem is to leverage the thousands of complete viral genomes that are publicly available. Here we describe our efforts to assemble a comprehensive resource that provides a quantitative snapshot of viral genomic trends – such as gene density, noncoding percentage, and abundances of functional gene categories – across thousands of viral genomes. We have also developed a coarse-grained method for visualizing viral genome organization for hundreds of genomes at once, and have explored the extent of the overlap between bacterial and bacteriophage gene pools. Existing viral classification systems were developed prior to the sequencing era, so we present our analysis in a way that allows us to assess the utility of the different classification systems for capturing genomic trends. DOI: https://doi.org/10.7554/eLife.31955.001 GITA MAHMOUDABADI AND ROB PHILLIPS* Introduction numerous natural habitats, untethering us from There are an estimated 1031 virus-like particles the organisms we know and love and giving us inhabiting our planet, outnumbering all cellular access to a sea of genomic data from novel life forms (Suttle, 2005; Wigington et al., organisms (Paez-Espino et al., 2016). Such 2016). Despite their presence in astonishing advances allow us to appreciate the genomic numbers and their impact on the population *For correspondence: phillips@ diversity that is a hallmark of viral genomes pboc.caltech.edu dynamics and evolutionary trajectories of their (Paez-Espino et al., 2016; Edwards and hosts, our quantitative knowledge of trends in Rohwer, 2005; Rohwer and Thurber, 2009; Competing interests: The the genomic properties of viruses remains Simmonds et al., 2017; Simmonds, 2015; authors declare that no largely limited with many of the key quantities Mokili et al., 2012) and now make it possible to competing interests exist. used to characterize these genomes either scat- assemble some of the key numbers of virology. Funding: See page 23 tered across the literature or unavailable alto- In contrast to cellular genomes, which are uni- Reviewing editor: Arup K gether. This is in contrast to the growing ability versally coded in the language of double- Chakraborty, Massachusetts exhibited in resources such as the BioNumbers stranded DNA (dsDNA), genomes of viruses are Institute of Technology, United database (Milo et al., 2010) to assemble in one remarkably versatile. Viral genomes can be States curated collection the key numbers that charac- found as single or double-stranded versions of Copyright Mahmoudabadi terize cellular life forms. Our goal has been to DNA and RNA, packaged in segments or as one and Phillips. This article is complement these databases of key numbers of piece, and present in both linear and circular distributed under the terms of cell biology (Milo et al., 2010; Phillips et al., forms. Additionally, based on their rapid infec- the Creative Commons 2012; Milo and Phillips, 2015; Phillips and tious cycles, large burst sizes, and often highly Attribution License, which permits unrestricted use and Milo, 2009) with corresponding data from error-prone replication, viruses collectively sur- redistribution provided that the viruses. With the advent of high-throughput vey a large genomic sequence space, and com- original author and source are sequencing technologies, recent studies have prise a great portion of the total genomic credited. enabled genomic and metagenomic surveys of diversity hosted by our planet (Kristensen et al., Mahmoudabadi and Phillips. eLife 2018;7:e31955. DOI: https://doi.org/10.7554/eLife.31955 1 of 26 Feature article Research A comprehensive and quantitative exploration of thousands of viral genomes 2010; Hendrix, 2003). Recently, through a large commonly known as retroviruses; and, the dou- study of metagenomic sequences, the known ble-stranded DNA retroviruses (Group VII). viral sequence space was increased by an order Given the prevalence of these viral classifica- of magnitude (Paez-Espino et al., 2016), and tion systems in the categorization of viruses much more of the viral “dark matter” likely today, it is worth remembering that their incep- remains unexplored (Youle et al., 2012). tion predates the sequencing of the first In analyzing an increasing spectrum of genome in 1976. With the fastest and cheapest sequence data, we are faced with a considerable rates of sequencing available to date, we live at challenge that is unique to viruses, namely, how an opportune moment to explore viral genomic to find those features within viral genomes that properties and evaluate these existing classifica- might reveal hidden aspects of their evolutionary tion systems in light of the growing body of history. To put this challenge in perspective, sequence information. when analyzing non-viral data, universal markers In addition to the ICTV and the Baltimore from the ribosomal RNA such as 16S sequences classifications we used a simple classification sys- are used to classify newly discovered organisms tem based on the host domain information, and and to locate them on the evolutionary tree of divided viruses into bacterial, archaeal and life (Hug et al., 2016). Virus genomes on the eukaryotic viruses (Figure 1). The underpinning other hand are highly divergent and possess motivation behind this kind of classification is no such universally shared sequences the Coevolution Hypothesis (Mahy and Van (Kristensen et al., 2011). Regenmortel, 2010; Forterre, 2010). Viruses In the absence of universal genomic markers, are obligate organisms unable to survive without viruses have historically been classified based on their host, and as a corollary it is hypothesized a variety of attributes, perhaps most notably that they have coevolved with their hosts as the hosts diverged over billions of years to form the morphological characteristics, proposed in 1962 three domains of life (Mahy and Van Regen- by the International Committee on Taxonomy of mortel, 2010; Forterre, 2010). A possible piece Viruses or ICTV (King et al., 2011), or based on of supporting evidence for this hypothesis is that the different ways by which they produce there are to date no reported infections of hosts mRNA, proposed by David Baltimore in 1971 from one domain by viruses of another (Baltimore, 1971; Figure 1). The ICTV classifies observed. We also explored a minimal classifica- viruses into seven orders: Herpesvirales, large tion system that divides the virus world into two eukaryotic double-stranded DNA viruses; Cau- groups based on their nucleotide type (RNA and dovirales, tailed double-stranded DNA viruses DNA), here termed “Nucleotide Type” classifica- typically infecting bacteria; Ligamenvirales, linear tion (Figure 1). This classification is introduced double-stranded viruses infecting archaea; as a simplified version of the Baltimore classifica- Mononegavirales, nonsegmented negative (or tion system. In practice, we have assigned Balti- antisense) strand single-stranded RNA viruses of more groups 1, 2 and 7 to the DNA viral plants and animals; Nidovirales, positive (or category, and the remaining Baltimore groups sense) strand single-stranded RNA viruses of to the RNA viral category. vertebrates; Picornavirales, small positive strand Although many viruses are uncharacterized, single-stranded RNA viruses infecting plants, at the time of the analysis of the data presented insects, and animals; and finally, the Tymovirales, here, there were 4,378 completed genomes monopartite positive single-stranded RNA available from the NCBI viral genomes resource viruses of plants. In addition to these orders, (Brister et al., 2015) (data acquired in August, there are ICTV families, some of which have not 2015). However, large-scale analyses of genomic been assigned to an ICTV order. Only those properties for these viruses are generally ICTV viral families with more than a few mem- unavailable. This stands in stark contrast to the bers present in our dataset are explored. in-depth analyses performed on partially assem- The Baltimore classification groups viruses bled viral genomes or viral contigs derived from into seven categories (Figure 1): double- metagenomic studies (Paez-Espino et al., 2016; stranded DNA viruses (Group I); single-stranded Roux et al., 2016). Although these studies have DNA viruses (Group II); double-stranded RNA uncovered many important aspects of viral ecol- viruses (Group III); positive single-stranded RNA ogy with relatively little bias in sampling, they viruses (Group IV); negative single-stranded are limited by the fact that metagenomic studies RNA viruses (Group V); positive single-stranded typically do not result in the full assembly of RNA viruses with DNA intermediates (Group VI), genomes. An interesting example that illustrates Mahmoudabadi and Phillips. eLife 2018;7:e31955. DOI: https://doi.org/10.7554/eLife.31955 2 of 26 Feature article Research A comprehensive and quantitative exploration of thousands of viral genomes A. Baltimore Classification 1. dsDNA 2. +ssDNA 3. dsRNA 4. +ssRNA mRNA (+ssRNA) 5. -ssRNA 6. +ssRNA-RT 7. dsDNA-RT B. Nucleotide Type Classification DNA viruses RNA viruses 1. dsDNA 2. +ssDNA 3. dsRNA