ADVANTAGES OF METAGENOMICS TO STUDY ENVIRONMENTAL SAMPLES

Historically, investigation of microbial communities has been performed using culture-based methodologies. However, less than 1% of bacterial species in environmental communities are thought to be culturable on standard laboratory growth media (Amann et al., Reference Amann, Ludwig and Schleifer1995). To overcome these limitations, a metagenomic approach was suggested to characterize total microbial community DNA (including viruses, prokaryotes and eukaryotes) (Handelsman et al., Reference Handelsman, Rondon, Brady, Clardy and Goodman1998). Such techniques have revealed unprecedented taxonomic and functional diversity in aquatic and terrestrial habitats (Rondon et al., Reference Rondon, August, Bettermann, Brady, Grossman, Liles, Loiacono, Lynch, MacNeil, Minor, Tiong, Gilman, Osburne, Clardy, Handelsman and Goodman2000; Venter et al., Reference Venter, Remington, Heidelberg, Halpern, Rusch, Eisen, Wu, Paulsen, Nelson, Nelson, Fouts, Levy, Knap, Lomas, Nealson, White, Peterson, Hoffman, Parsons, Baden-Tillson, Pfannkoch, Rogers and Smith2004; Sogin et al., Reference Sogin, Morrison, Huber, Mark Welch, Huse, Neal, Arrieta and Herndl2006). Metagenomics encompasses two types of study – whole genome shotgun (WGS) sequencing of all the genes in the microbial community or those targeting a single, taxonomically important gene (e.g. 16S rDNA for bacteria) (Gilbert & Dupont, Reference Gilbert and Dupont2011). Whole genome shotgun sequencing studies generally fall into one of three categories: (1) vector cloning and sequencing studies in which community DNA is cloned into a fosmid, cosmid, or bacterial artificial chromosome (BAC) and the library is screened for particular genes of interest, usually for a particular species or group; (2) Sanger sequencing-based shotgun metagenomic studies in which community DNA is cloned into a vector and randomly sequenced for assembly and/or annotation; and (3) next-generation shotgun sequencing studies that utilize sequencing-by-synthesis technology to generate millions of sequence reads without the need for cloning (Gilbert & Dupont, Reference Gilbert and Dupont2011). In contrast, amplicon sequencing does not encompass the sequencing of total environmental DNA, and is not viewed by many as a metagenomic method in the strict sense. However, the ability to exploit millions of sequence reads to characterize thousands of species has revealed unprecedented diversity in marine samples and may represent a more promising alternative to assess ecosystem health and public health risks than previous culture-based or molecular methods (Unno et al., Reference Unno, Jang, Han, Kim, Sadowsky, Kim, Chun and Hur2010; Gibbons et al., Reference Gibbons, Caporaso, Pirrung, Field, Knight and Gilbert2013; Staley et al., Reference Staley, Gould, Wang, Phillips, Cotner and Sadowsky2014a).

NEXT-GENERATION SEQUENCING INVESTIGATION OF AQUATIC ECOSYSTEMS

In response to the success of early metagenomics studies, several next-generation sequencing (NGS) platforms have been developed that are able to produce 10⁵–10⁷ sequence reads of short-to-intermediate length (approximately 30–500 nt) using massively parallel sequencing approaches (Margulies et al., Reference Margulies, Egholm, Altman, Attiya, Bader, Bemben, Berka, Braverman, Chen, Chen, Dewell, Du, Fierro, Gomes, Godwin, He, Helgesen, Ho, Irzyk, Jando, Alenquer, Jarvie, Jirage, Kim, Knight, Lanza, Leamon, Lefkowitz, Lei, Li, Lohman, Lu, Makhijani, McDade, McKenna, Myers, Nickerson, Nobile, Plant, Puc, Ronan, Roth, Sarkis, Simons, Simpson, Srinivasan, Tartaro, Tomasz, Vogt, Volkmer, Wang, Wang, Weiner, Yu, Begley and Rothberg2005). The most popular of these platforms are the 454-FLX (Roche), Genome Analyzer (Illumina) and SOLiD (Applied Biosystems) systems. The primary differences among these platforms are the sequence length and number of sequence reads achieved, although more detailed comparisons of these systems, including reaction chemistries and costs per sample, have been reviewed (Mardis, Reference Mardis2008). These approaches have greatly reduced the cost and increased the speed at which metagenomics-based approaches can be applied, without the need for the construction of extensive clone libraries, as can be seen by the exponential increase in the amount of sequence data uploaded to public repositories, such as the Sequence Read Archive (SRA) at the National Center for Biotechnology Information (Figure 1).

Fig. 1.

Recent accumulation of sequence data in the Sequence Read Archive at the National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov/sra/).

COMPUTATIONAL ADVANCES FOR METAGENOMIC ANALYSIS

The emergence of NGS technologies brought about a requirement to develop computational approaches to process and analyse the massive volumes of data generated. Advantages and limitations of these tools are outlined in Table 1. Preliminary computational tools, originally designed to handle <10⁴ sequence reads, such as LIBSHUFF (Singleton et al., Reference Singleton, Furlong, Rathbun and Whitman2001) and ARB (Ludwig et al., Reference Ludwig, Strunk, Westram, Richter, Meier, Buchner, Lai, Steppi, Jobb, Förster, Brettske, Gerber, Ginhart, Gross, Grumann, Hermann, Jost, König, Liss, Lüssmann, May, Nonhoff, Reichel, Strehlow, Stamatakis, Stuckmann, Vilbig, Lenke, Ludwig, Bode and Schleifer2004) were encumbered by the computational demands of the emerging datasets, and their limited functionality. Moreover, performing one or a few functions, made them difficult to integrate. As a result, software programs such as mothur and QIIME (Quantitative Insights into Microbial Ecology) have become popular applications to process and analyse taxonomic marker genes (Schloss et al., Reference Schloss, Westcott, Ryabin, Hall, Hartmann, Hollister, Lesniewski, Oakley, Parks, Robinson, Sahl, Stres, Thallinger, Van Horn and Weber2009; Caporaso et al., Reference Caporaso, Kuczynski, Stombaugh, Bittinger, Bushman, Costello, Fierer, Peña, Goodrich, Gordon, Huttley, Kelley, Knights, Koenig, Ley, Lozupone, McDonald, Muegge, Pirrung, Reeder, Sevinsky, Turnbaugh, Walters, Widmann, Yatsunenko, Zaneveld and Knight2010). These programs employ a pipeline of subroutines – a series of computational steps to perform multiple functions – for joining paired-end sequence reads (i.e. forward and reverse sequences), quality control procedures (e.g. quality trimming and chimera removal), sequence alignment, OTU clustering, and taxonomic assignment using standard reference databases such as GreenGenes, SILVA or the Ribosomal Database Project (RDP) (DeSantis et al., Reference DeSantis, Hugenholtz, Larsen, Rojas, Brodie, Keller, Huber, Dalevi, Hu and Andersen2006; Pruesse et al., Reference Pruesse, Quast, Knittel, Fuchs, Ludwig, Peplies and Glockner2007; Cole et al., Reference Cole, Wang, Cardenas, Fish, Chai, Farris, Kulam-Syed-Mohideen, McGarrell, Marsh, Garrity and Tiedje2009). Statistical tools for analysis of complex datasets have also been incorporated into these programs allowing more thorough and complex analyses of ecological datasets.

Table 1.

Computational advantages and limitations of software and subroutines developed for analysis of next-generation sequencing data.

Similar to NGS of taxonomic marker genes, results of WGS sequencing studies were also initially difficult to interpret due to the increased computational requirements. Metagenomic Analyzer (MEGAN) software was developed specifically to deal with the computation of NGS shotgun sequences, bypassing the limitations of extensive sequence assembly from environmental sequence data and the lower abundance of phylogenetically relevant marker genes (Huson et al., Reference Huson, Auch, Qi and Schuster2007). The Community Cyberinfrastructure for Advanced Marine Microbial Research and Analysis (CAMERA) was developed as an online repository for sequence and metadata, and integrates existing and emerging bioinformatics tools for the analysis of metagenomic data originally incorporating analytical packages and workflows used for the Global Ocean Sampling (GOS) expedition (Seshadri et al., Reference Seshadri, Kravitz, Smarr, Gilna and Frazier2007). The program compares sequence reads against a reference database and outputs the results for exploration using a graphical interface. Similarly, web servers like the Metagenomics RAST (MG-RAST) have also been developed where data can be uploaded and stored, and taxonomic and functional annotations are performed automatically (Meyer et al., Reference Meyer, Paarmann, D'Souza, Olson, Glass, Kubal, Paczian, Rodriguez, Stevens, Wilke, Wilkening and Edwards2008). While these tools have facilitated analysis of datasets of previously unprecedented depth and coverage, they each have their unique advantages and disadvantages.

Metagenomic studies to date have been carried out primarily from a descriptive, data discovery perspective, with a focus on what is in an environmental sample and how diversity, as well as the presence and abundance of community members or functional genes, change between samples and habitats. Ordination techniques such as principal coordinate analysis (PCoA) and non-metric multidimensional scaling (NMDS) have enabled visualization of these complex datasets to facilitate these studies. In fact, one of the main drawbacks of these studies is the enormous datasets generated and the inability to view all the data in an easily interpretable format. Network-based approaches have also been developed to allow visual comparisons from complex metagenomic shotgun and taxonomic marker datasets (Mitra et al., Reference Mitra, Gilbert, Field and Huson2010; Larsen et al., Reference Larsen, Field and Gilbert2012). Statistical software packages such as STAMP (Statistical Analysis of Metagenomic Profiles) have been developed to clearly summarize statistical trends allowing meaningful biological inferences to be made from these complex data (Parks & Beiko, Reference Parks and Beiko2010). The recent and continuing decline in the cost of NGS will soon allow for better sample replication, enabling more powerful statistical comparisons, and the generation of terabase datasets that will allow for a better quantitative assessment of taxa and genes present in the environment (Gilbert & Dupont, Reference Gilbert and Dupont2011).

COMPUTATIONAL LIMITATIONS TO METAGENOMIC ANALYSES

Prior to 2006, metagenomics studies were limited to clone-based studies and Sanger sequencing, and the relatively high cost of sequencing (~$500 per Mb) restricted the size of these early datasets, as well as replication efforts for robust statistical comparisons (Kircher & Kelso, Reference Kircher and Kelso2010; Temperton & Giovannoni, Reference Temperton and Giovannoni2012). These methods were subject to PCR amplification and cloning biases. Next-generation sequencing methods have greatly reduced the costs associated with generating larger volumes of sequence data, and have, to some extent, alleviated bias associated with cloning (Wooley et al., Reference Wooley, Godzik and Friedberg2010). However, PCR primer bias remains an intrinsic limitation and this issue, coupled with shorter sequence read lengths, can significantly affect the diversity inferred from NGS data (Youssef et al., Reference Youssef, Sheik, Krumholz, Najar, Roe and Elshahed2009). In addition, next-generation sequences are subject to error due to DNA polymerases, chimera formation and sequencing errors (Kircher & Kelso, Reference Kircher and Kelso2010; Patin et al., Reference Patin, Kunin, Lidstrom and Ashby2013). Bias and error can be reduced by improving reaction chemistries, reducing PCR cycle numbers, using well-designed primer sets, and refining the quality of reagents. However, computational approaches must also be considered to account for these errors as well as difficulties arising from the massive volumes of data generated.

The intrinsic problem of sequencing error is mitigated in single organism genomic sequencing by sequence assembly and high coverage depth (Goldberg et al., Reference Goldberg, Johnson, Busam, Feldblyum, Ferriera, Friedman, Halpern, Khouri, Kravitz, Lauro, Li, Rogers, Strausberg, Sutton, Tallon, Thomas, Venter, Frazier and Venter2006); however, taxonomic marker surveys using NGS methods are prone to overestimation of diversity resulting from sequence error (Kunin et al., Reference Kunin, Engelbrektson, Ochman and Hugenholtz2010). Early analysis of sequence errors in a NGS dataset has shown that exclusion of sequences containing ambiguous bases (Ns) and primer or barcode mismatches reduces the sequence error rate in the dataset to less than that of Sanger sequencing while retaining >90% of sequence reads (Huse et al., Reference Huse, Huber, Morrison, Sogin and Welch2007). Subsequent studies have shown that a 2% pre-clustering step and OTU binning at ≤97% similarity greatly reduced inflated diversity estimates associated with sequencing error (Huse et al., Reference Huse, Welch, Morrison and Sogin2010; Kunin et al., Reference Kunin, Engelbrektson, Ochman and Hugenholtz2010). In addition, several computational approaches, such as UCHIME (Edgar et al., Reference Edgar, Haas, Clemente, Quince and Knight2011), have been developed to identify and remove chimeric sequence reads (Wooley & Ye, Reference Wooley and Ye2009).

Despite these processing steps to improve sequence quality, inherent bias and limitations still exist when using small subunit rRNA genes for taxonomic surveys. Due to fundamental differences in rDNA sequences, prokaryotes and eukaryotes must be sequenced separately. Among prokaryotes, utilization of the 16S rDNA presents unique challenges in that species resolution can be difficult as a result of the highly conserved nature of this gene (Gürtler & Stanisich, Reference Gürtler and Stanisich1996). Furthermore, a single cell can contain up to 15 copies of rDNA and copies may be heterogeneous within the same genome, further complicating species identification and accurate quantification of taxonomic abundances (Klappenbach et al., Reference Klappenbach, Saxman, Cole and Schmidt2001). Finally, even when universal primers are used for amplification, it is unlikely that all members of a certain group (e.g. bacteria or archaea) will be amplified due to the higher diversity of these domains (Davenport & Tümmler, Reference Davenport and Tümmler2013).

For WGS studies of environmental samples, sequencing coverage is often extremely low, thus it is difficult to discern individual genomes within a community due to differences in relative abundances of highly diverse organisms. The first, Sanger-sequencing-based, metagenomic studies had some success using single genome sequence assemblers to build contigs, contiguous sequence fragments, from metagenomic data, owing to relatively small numbers of sequence reads and longer read lengths compared with NGS methods (Venter et al., Reference Venter, Remington, Heidelberg, Halpern, Rusch, Eisen, Wu, Paulsen, Nelson, Nelson, Fouts, Levy, Knap, Lomas, Nealson, White, Peterson, Hoffman, Parsons, Baden-Tillson, Pfannkoch, Rogers and Smith2004; Rusch et al., Reference Rusch, Halpern, Sutton, Heidelberg, Williamson, Yooseph, Wu, Eisen, Hoffman, Remington, Beeson, Tran, Smith, Baden-Tillson, Stewart, Thorpe, Freeman, Andrews-Pfannkoch, Venter, Li, Kravitz, Heidelberg, Utterback, Rogers, Falcón, Souza, Bonilla-Rosso, Eguiarte, Karl, Sathyendranath, Platt, Bermingham, Gallardo, Tamayo-Castillo, Ferrari, Strausberg, Nealson, Friedman, Frazier and Venter2007). However, the large number and small read length of NGS reads limit the effectiveness of sequence assembly by these methods (Wooley et al., Reference Wooley, Godzik and Friedberg2010). Large amounts of nearly identical sequences and the possibility of assembling sequences from different OTUs require a different assembly strategy that has been met by several new programs such as IDBA-UD and meta-VELVET (Namiki et al., Reference Namiki, Hachiya, Tanaka and Sakakibara2012; Peng et al., Reference Peng, Leung, Yiu and Chin2012), which take into account short sequence reads and uneven coverage due to differences in OTU abundance. Nevertheless, there is still a trade-off between assembly, which can reveal novel genes or species not previously described, and read mapping without assembly, which allows semi-quantitative inferences (Davenport & Tümmler, Reference Davenport and Tümmler2013).

While NGS costs have declined compared with Sanger sequencing, complete metagenomic characterization of highly diverse environmental samples remains limited by the prohibitive cost required (Gilbert & Dupont, Reference Gilbert and Dupont2011; Knight et al., Reference Knight, Jansson, Field, Fierer, Desai, Fuhrman, Hugenholtz, van der Lelie, Meyer, Stevens, Bailey, Gordon, Kowalchuk and Gilbert2012). Even with the recent advances in sequencing technology, <0.000001% of the metagenome in seawater is estimated to have been sequenced based on average genome sizes and bacterial density in a one litre sample (Gilbert & Dupont, Reference Gilbert and Dupont2011). Furthermore, 4–5 × 10¹⁴ bp and 3 × 10¹⁵ bp of sequence data are estimated as the requirement for 1× coverage of a one litre seawater and one gram soil sample, respectively (Gilbert & Dupont, Reference Gilbert and Dupont2011; Knight et al., Reference Knight, Jansson, Field, Fierer, Desai, Fuhrman, Hugenholtz, van der Lelie, Meyer, Stevens, Bailey, Gordon, Kowalchuk and Gilbert2012). To accomplish this sequencing depth, >800 and approximately 5000 full runs would be required on an Illumina HiSeq2000 platform, at a cost of tens of millions of dollars currently (Caporaso et al., Reference Caporaso, Lauber, Walters, Berg-Lyons, Huntley, Fierer, Owens, Betley, Fraser, Bauer, Gormley, Gilbert, Smith and Knight2012). However, 6× to 8× coverage is considered the standard to ensure adequate representation of all of the genomes in the community (Akondi & Lakshmi, Reference Akondi and Lakshmi2013), further increasing costs and effort. Thus, analyses of complex environments at the degree of coverage needed may need to wait for the development of new, even less expensive, sequencing technology.

Regardless of the type of sequencing study performed, the taxonomic and functional annotations that result are dependent on the databases used. Databases are known to have a compositional bias, favouring sequences from easily culturable and accessible organisms (Pignatelli et al., Reference Pignatelli, Aparicio, Blanquer, Hernández, Moya and Tamames2008). The quality of the assemblies and the application used for gene or taxonomic annotation can have significant effects on the percentage of reads annotated as well as the accuracy of the prediction (Mavromatis et al., Reference Mavromatis, Ivanova, Barry, Shapiro, Goltsman, Mchardy, Rigoutsos, Salamov, Korzeniewski, Land, Lapidus, Grigoriev, Richardson, Hugenholtz and Kyrpides2007). Furthermore, the completeness of the database dramatically influences the classification of sequences, and taxonomic and functional composition of samples may change depending on the version of the database used, even among recently updated databases (Pignatelli et al., Reference Pignatelli, Aparicio, Blanquer, Hernández, Moya and Tamames2008). These changes primarily result from the assignment of previously unclassified reads to newly sequenced taxa. However, sequencing of closely related species can also result in shifts in assignment (Pignatelli et al., Reference Pignatelli, Aparicio, Blanquer, Hernández, Moya and Tamames2008).

PERSPECTIVES

The studies reviewed here, highlight the efficacy of employing metagenomics approaches in the study of environmental samples, specifically water samples, to better characterize biodiversity, biogeography, the effects of anthropogenic impacts, and potential public health risk. Development of new computational tools to process and analyse NGS data has facilitated the identification of previously unidentified microorganisms, some of which may have important public health implications. Furthermore, more thorough characterization of microbial communities is facilitating better interpretations regarding which practices are important in shaping bacterial community structure. Despite recent advances, drawbacks to these methods still exist, including sequencing error as well as biases and gaps in reference databases favouring easily culturable microorganisms. Further advances in technology and metagenomics studies will allow microbial ecologists and physiologists to fill these knowledge gaps and thus provide a more complete understanding of the interaction between anthropogenic practices, the environment and microbial communities.