DNA sequencing and analysis approaches vary widely, with opportunities to serve a variety of marine monitoring applications. Terminology usage shifts as methods change (e.g., the evolution of long-read Sanger to short-read Illumina sequencing) and varies among applications, particularly between microbial/macrobial and aquatic/benthic specialties. For example, the term “metagenomics” is defined simply and broadly as the analysis of genetic materials obtained directly from environmental samples, including analysis of sequenced amplicon (Xu, 2015). Alternatively, this broad definition is rejected in favor of keeping terminology for amplicon sequencing separate (e.g., “metagenetics” or “metaprofiling”) rather than as a subset of metagenomics (Esposito and Kirschberg, 2014; Escobar-Zepeda et al., 2015; Mendoza et al., 2015; Creer et al., 2016). A similar situation exists for “environmental DNA” (eDNA) which is broadly defined in some cases (Pedersen et al., 2015). However, the term eDNA is increasingly used to connote animal DNA obtained indirectly from environmental samples rather than tissues per se (Bohmann et al., 2014; Kelly et al., 2014; Kelly, 2016), and that connotation is generally used here. Additional ambiguities arise from advances in sequencing technologies. For example, both Sanger (first generation) and later platforms such as Roche 454, SOLiD, and Illumina (second generation) may be referred to as next generation sequencing (NGS) in the literature, and both second and third-generation (PacBio, Ion Torrent, Oxford Nanopore) may be referred to as massively parallel or high-throughput sequencing (HTS). Other examples of terms with ambiguous usage include: conservation genomics (Garner et al., 2016), genetic fingerprinting, tag sequencing, targeted metagenomics, metabarcoding (Mendoza et al., 2015), and community analysis. With continued modifications to technology, expanding applications, and development of bioinformatic tools, terminology is expected to remain a challenge.

In generic terms, it is anticipated that DNA sequencing as a tool in routine marine monitoring will employ high-throughput sequencing coupled with a bioinformatic analysis pipeline. Sequencing may target whole genomes or specific regions that may be amplified by PCR prior to sequencing. DNA may be obtained from an individual, an assemblage of organisms (“meta” approach), or an assemblage of partial organisms (eDNA). For the purposes here, the focus is on high throughput, short-read sequencing of amplicons from macroeukaryotic assemblages (metabarcoding) or individuals (barcoding); amplicons from assemblages of maroeukaryotes obtained indirectly from the environment rather than from whole organisms or tissues (eDNA); amplicons from prokaryotic/microeukaryotic assemblages (metagenetics); and shotgun sequencing of prokaryotic/microeukaryotic assemblages (metagenomics). A variety of references regarding extraction (Cox and Goodwin, 2013; Hazen et al., 2013), amplification, and sequencing strategies for water, sediments, and tissues are available herein, and eDNA is also reviewed extensively elsewhere (Bohmann et al., 2014; Rees et al., 2014; Pedersen et al., 2015; Thomsen and Willerslev, 2015; Shelton et al., 2016); therefore, only a brief technical overview is provided here.

Sequencing of PCR-amplified marker genes is often employed to identify organisms (Patwardhan et al., 2014). This targeted sequencing approach can provide detection of species in relatively low abundance (DeSantis et al., 2006; Sogin et al., 2006; Hamady and Knight, 2009; Quast et al., 2013) and better depth of coverage compared to shotgun whole-genome approaches (Zhou et al., 2015). For example, amplification of a nuclear 16S rRNA gene segment is typical for taxonomic classification of marine prokaryotic assemblages (Klindworth et al., 2013). Amplification of 18S rRNA regions is used for a variety of communities, including phytoplankton and other microeukaryotes (Dunthorn et al., 2012; Hugerth et al., 2014; Johnson and Martiny, 2015; del Campo et al., 2016; Giner et al., 2016), and amplification of the 23S rRNA gene is used to classify organisms such as zooplankton (Hirai et al., 2015a; Bucklin et al., 2016). Amplification of mitochondrial DNA is used to identify an assortment of organisms, with cytochrome oxidase I (COI), cytochrome b, and mitochondrial 16S as examples of popularly employed target regions (Dauble et al., 2012; Pawlowski et al., 2014; Cowart et al., 2015; Guo et al., 2015; Harada et al., 2015; Johnson and Martiny, 2015; Aylagas et al., 2016; Bucklin et al., 2016; Creer et al., 2016; Leray and Knowlton, 2016; Thompson et al., 2016; Trivedi et al., 2016). Newer approaches are being used to identify fish (Miya et al., 2015) and marine mammals (Foote et al., 2012; Ma et al., 2016) from seawater samples.

In comparison to amplifying and sequencing a specific genomic region, shotgun metagenomic methods sequence broadly across entire genomes (Thomas et al., 2012; Escobar-Zepeda et al., 2015; Sharpton, 2014; Guo et al., 2016). In general, DNA extracted from a sample is fragmented, a DNA library is prepared (Head et al., 2014) using a limited number of PCR cycles or via PCR-free methods (Mendoza et al., 2015), and the assortment of random DNA fragments is sequenced. The resulting sequences can be compared directly against genomes of known taxa or can be assembled into longer segments (contigs) which can be further annotated for gene content, function, and taxonomic assignment (Segata et al., 2012; Teeling and Glöckner, 2012; Wood and Salzberg, 2014; Bengtsson-Palme et al., 2015). This approach provides a broad view because a variety of genes are detected, including those that encode metabolic enzymes and pathways. If the goal is to characterize the functional potential of an ecosystem, shotgun sequencing is desirable because whole-genome data provides information on potential metabolic function, in addition to taxonomic information (Handelsman, 2004; Thomas et al., 2012). Furthermore, shotgun sequencing is less affected by primer bias and chimeras compared to amplicon sequencing (Guo et al., 2016). However, obtaining adequate depth of coverage can be problematic (Zhou et al., 2015), motivating some studies to utilize both amplicon and whole-genome approaches (e.g., Ocean Sampling Day, Table 2).

Regardless of the DNA sequencing approach used, bioinformatics is critical to understand the information delivered by DNA sequencing efforts. A plethora of bioinformatic tools are available, with a complete review beyond the scope of this article. Many bioinformatic tools and platforms are now tailored to gene sequences derived from environmental samples, particularly for prokaryotic assemblages (Kanehisa et al., 2008; Meyer et al., 2008; Caporaso et al., 2010; Abubucker et al., 2012; Thomas et al., 2012; Kozich et al., 2013; Hunter et al., 2014). Furthermore, de-novo assembly is increasingly being used to derive nearly complete genomes of previously unknown ecologically important organisms from metagenomes (Garza and Dutilh, 2015; Hugerth et al., 2015). A growing number of assemblers (Boisvert et al., 2012; Peng et al., 2012) and binning tools (Alneberg et al., 2014; Imelfort et al., 2014; Nurk et al., 2017) are available that assemble short sequencing reads into larger contigs and subsequently combine them into bins based on, for example, sequence similarity (Teeling and Glöckner, 2012). In addition to standard biodiversity metrics such as alpha, beta, and gamma diversity (Escobar-Zepeda et al., 2015; Zhou et al., 2015), network analysis tools aid visualization of complex ecological relationships (Mitra et al., 2010; Hurwitz et al., 2014).

Marine microbes were among the first communities to be explored by DNA sequencing (Giovannoni et al., 1990; Venter et al., 2004). Sequencing was initially applied to prokaryotic and microeukaryotic communities in seawater with techniques developed for these whole organisms because traditional culture methods were inadequate to describe microbial biodiversity in water and soil (Rappé and Giovannoni, 2003; Handelsman, 2004). The Global Ocean Sampling expedition helped establish DNA sequencing as a fundamental tool in marine ecosystem research (Rusch et al., 2007; Dupont et al., 2015). Potential benefits of DNA sequencing of microbial communities to marine assessment programs include characterization of food webs, assessing responses to disruption and stress, and detection of sensitive, rare, threatened, toxic, or invasive taxa (Hulme, 2006; Kudela et al., 2010; Mock and Kirkham, 2012; Lindeque et al., 2013; Muller-Karger et al., 2014; Pawlowski et al., 2014; Chown et al., 2015; Mock et al., 2015; Tan et al., 2015; Zaiko et al., 2015; Bowers et al., 2016; Bucklin et al., 2016).

Sequencing of environmental samples such as seawater is now applied in a macrobial sense across a range of trophic levels (“microbes to mammals”) (Trivedi et al., 2016; Valentini et al., 2016), and approaches used to inventory individuals (Ardura et al., 2013; Harada et al., 2015; Thompson et al., 2016) are being applied to assemblages (Carugati et al., 2015; Zimmermann et al., 2015; Aylagas et al., 2016). Studies demonstrate DNA metabarcoding to be a reliable method for biodiversity assessment with potential for inferring biotic indices for marine ecosystem quality assessment (Aylagas et al., 2014; Pawlowski et al., 2014; Cowart et al., 2015; Elbrecht and Leese, 2015; Visco et al., 2015; Ferrera et al., 2016). Comparable results between molecular and traditional approaches are reported in a number of studies (Hirai et al., 2015a; Lejzerowicz et al., 2015; Aylagas et al., 2016, 2017; Valentini et al., 2016), suggesting that assessment programs that require manually intensive sorting and visual inspection will be among the first to formally integrate molecular techniques. For example, larval rockfish time-series are currently used to quantify spawning stock biomass off California, USA and are integral components of rockfish stock assessment (He et al., 2015). However, the utility of the traditional approach for counting rockfish larvae based on morphology is limited because only a few species can be visually identified to species. Recent research to improve stock assessment has genetically identified rockfish larvae collected off southern California and increased from 6 to 36 the number of species with time-series (Chen, 2017). The genetically-derived data is now being considered for integration into formal stock assessments.

Newer eDNA methods allow detection of higher trophic levels, including fish and mammals used to evaluate ecological status, from DNA extracted from sediment or filtered seawater. The eDNA approach promises information about bioindicators and commercially important or protected species without the need to collect tissue or trawl through sensitive habitats (Foote et al., 2012; Bohmann et al., 2014; Thomsen and Willerslev, 2015; Evans et al., 2016; Lacoursière-Roussel et al., 2016; Shelton et al., 2016). Some studies show promising results with regard to abundance estimates (Hänfling et al., 2016; Port et al., 2016), but others show significant differences between eDNA results and traditional tows for epibenthic macroinvertebrates (Kelly et al., 2017) and zooplankton (Hirai et al., 2015b), suggesting that these methods require further research.

DNA sequencing offers insight into ecosystem stress via changes in relative abundance or completeness of enzymatic pathways associated with chemical biodegradation or resistance. For example, a number of microorganisms carry genes that confer resistance to antibiotics or heavy metals. Increased exposure to anthropogenic chemicals promotes production and dissemination of these genes. Increases in the abundance of resistance genes and of integrons, genetic elements involved in horizontal gene transfer, are documented in polluted sediments (Thureborn et al., 2013). Resistance to multiple antibiotics is more frequent at polluted sites, and resistance to antibiotics and heavy metals can co-occur (Henriques et al., 2016). A number of studies demonstrate environmental gradients showing co-occurrence of anthropogenic inputs and gene occurrence in marine environments (Tacão et al., 2012; Chen et al., 2013; Thureborn et al., 2013), providing an opportunity to use individual gene markers (e.g., blaCTX-M, merA) or a full metagenomic profile to indicate pollution in ecological status assessments.

Disrupted ecosystems can also present shifts in particular taxa or overall biodiversity (Nogales et al., 2011; Mason et al., 2014; Mukherjee et al., 2017), opening the possibility to evaluate ecological status according to such metrics. Toward this goal, Aylagas et al. (2017) developed two composite indicators: (1) a sediment quality index that included organic matter content, redox potential, and metal, PAH and PCB concentrations; and (2) a bacterial index of environmental quality adapted from the AMBI. The indices were correlated, a promising result with regard to inclusion of prokaryotic assemblages into future assessment programs (Caruso et al., 2016). Furthermore, DNA sequencing approaches are used for public health applications related to human pathogens derived from fecal contamination (Tan et al., 2015; Staley and Sadowsky, 2016) and to investigate the impacts of anthropogenic stressors on the coral microbiome (Ziegler et al., 2016; Staley et al., 2017). Kelly et al. (2016) demonstrate that eDNA in seawater can be used not only for biodiversity measurement but also for assessing anthropogenic disturbance in coastal environments. In that study, sequencing of metazoan mitochondrial 16S DNA show greater diversity of sessile taxa in more urbanized sites. Higher diversity in a disturbed site also is reported for estuarine benthic eukaryotes using 18S rRNA metabarcoding (Chariton et al., 2015). In contrast, lower diversity of benthic foraminifera is reported for sites impacted by fish farming (Pawlowski et al., 2014). Such a combination of results suggests that a sole biodiversity index may be inadequate to evaluate ecosystem status and indicates that an AMBI-like approach, which weights taxa abundance by pollution tolerance, may be needed to convert taxonomic measures into indices of ecosystem function.

While DNA sequencing holds promise for marine assessment programs (Table 1), improvements are required to adapt these powerful tools to formal monitoring frameworks. One of the greatest needs, cutting across all aspects of the endeavor, is the need for standardization. Standardization is particularly critical for marine monitoring applications because long-term data set generation is a hallmark of such efforts. Standard methods and practices can be applied at every stage: study design; metadata collection and curation; sample collection and preservation; DNA extraction; inclusion of control samples; approaches to amplification, library preparation and sequencing; sequence quality control and other downstream processing; bioinformatic method documentation; and depositing of sequence results with appropriate metadata.

Standardization of metadata, a minimum common set of measurements to be recorded at the time of sample collection, is critical. Adequate metadata is required to integrate DNA sequencing information into marine ecosystem assessment programs, provide environmental context to sequence data, simplify annotation, facilitate data mining, and to allow comparisons between studies. Progress is being made to create standardized protocols (Field et al., 2008; Chain et al., 2009; Gilbert, 2015; Droege et al., 2016). Efforts that promote community collaboration and participation to produce standard practices include the MIxS standard from the Genomic Standards Consortium (GSC) (Yilmaz et al., 2011; ten Hoopen et al., 2015), the Environmental Ontology (Buttigieg et al., 2013, 2016), and data standards of the Global Genome Biodiversity Network (GGBN) (Droege et al., 2016). Despite widespread recognition of its importance and the realization that subsequent data curation is costly and time-consuming (ten Hoopen et al., 2016), compliance with metadata collection and curation standards is lacking. Gold standards with regard to data analysis, applied algorithms, and reference data sets comparable to those sought for the human medical space will improve the reliability and usability of results gained from marine sequencing studies. Recent sampling efforts such as the Ocean Sampling Day (Kopf et al., 2015), TARA Oceans project (Karsenti et al., 2011; Bork et al., 2015) and the Australian Marine Microbes Project (NCRIS, 2016) mark first steps to standardize sample preparation and sampling procedures for marine microbial ecology.

An adequate set of negative and positive control samples (e.g., mock communities, both DNA and whole-cell) are needed to track DNA preservation, extraction efficiency, contamination, and sequence errors, which can significantly affect results (Zhou et al., 2015; Allen et al., 2016; Mulcahy et al., 2016; Siegwald et al., 2017). Work initiated in the area of standardizing laboratory procedures includes much-needed development of reference materials and improved DNA extraction methods, as exemplified by establishment of the International Metagenomics and Microbiome Standards Alliance (IMMSA) (NIST, 2016). Controls used to assess artifacts and bias introduced by sample preparation, PCR, and sequencing are reviewed elsewhere (Pinto and Raskin, 2012; Elbrecht and Leese, 2015; Pedersen et al., 2015; Tan et al., 2015; Zhou et al., 2015; Aylagas et al., 2016; Danovaro et al., 2016) and should be used in conjunction with standard quality control pipelines for sequence quality, including chimeric removal (Smyth et al., 2010; Teeling and Glöckner, 2012; Zhou et al., 2014; Escobar-Zepeda et al., 2015; Jeon et al., 2015). Including mock communities in DNA sequencing efforts can assess technical issues such as incomplete DNA extraction or library preparation, PCR, and sequencing errors (Schirmer et al., 2015) and provide correction factors (Tan et al., 2015; Aylagas et al., 2016). Furthermore, simulated datasets can test the performance of both sequencing technology and analysis pipelines (Bonilla-Rosso, 2015), and bioinformatic tools can be used to mask technical variability (Leek et al., 2012).

The technical variation in sequence data is compounded by the considerable temporal and spatial biological variability in marine systems. Environmental heterogeneity is both natural (e.g., salinity gradients in transitional waters) and driven by proximity to anthropogenic influences. Study design should account for expected variability and include appropriate types and amounts of biological and technical replicates to ensure that meaningful comparisons and statistical analysis, suited to the question at hand, can be provided rather than ad-hoc sequencing based on available budget (Prosser, 2010; Knight et al., 2012; Pinto and Raskin, 2012; Thomas et al., 2012; Zhou et al., 2015). Greater sequencing coverage can aid the ability to differentiate environmental variability, with the balance of biological and technical replicates dependent on the study questions and community complexity (Teeling and Glöckner, 2012; Zhou et al., 2015).

A challenge to DNA sequencing of environmental samples is obtaining accurate taxonomic classification, particularly from short DNA sequence reads. The issue is exacerbated when reference genomes are lacking, such as for marine ecosystems and particularly for eukaryotic and viral components (Leray and Knowlton, 2016; Roux et al., 2016). Taxonomic mis-classification of reads impacts the accuracy of abundance estimates (Aylagas et al., 2016). Lack of intercalibration between morphological and molecular methods presents an additional issue, and transferability across taxa is a concern, particularly if the initial assessment used to establish the database is limited to one or two taxonomic groups (Kelly, 2016). Although comparable molecular and traditional approaches are reported (Lejzerowicz et al., 2015; Aylagas et al., 2016, 2017; Hänfling et al., 2016; Valentini et al., 2016), other studies show discrepancies (Cowart et al., 2015; Hirai et al., 2015b; Mohrbeck et al., 2015; Giner et al., 2016; Kelly et al., 2017); therefore, it has been advised to use both approaches in a complementary fashion rather than outright substitution of morphotaxonomic approaches (Borja et al., 2008; Pedersen et al., 2015; Thomsen and Willerslev, 2015). Although such an approach may deliver a holistic perspective of marine ecological status, it would not produce the desired effect of streamlining monitoring efforts, which is needed to transition research into practice (de Jonge et al., 2006).

Bioinformatic tools to aid microbial classification include similarity based approaches (Liu et al., 2011; Menzel et al., 2016), phylogenetic placement of sequences to known reference trees (Matsen et al., 2010), and using a combination of methods (Darling et al., 2014; Dröge et al., 2015). The reliance of current methods on known sequences, however, is a major drawback, as only a fraction of marine genomes (whole genomes or marker gene regions) are sequenced to date. It is worth noting that sequences obtained from metabarcoding efforts may provide value to ecosystem assessments even if reliable taxonomy cannot be assigned to them. A biodiversity index that does not require taxonomic assignment—a reference-free approach (Mendoza et al., 2015)—could in principle substitute for existing indices targeting specific species. It is noted that diversity in terms of operational taxonomic units (OTUs) vs. strict taxonomic assignment of sequences could satisfy requirements of the EU MSFD (Danovaro et al., 2016). Several bioinformatic tools are available for microbial sequences that could aid this goal (Eren et al., 2013; Callahan et al., 2016; Amir et al., 2017). For example, the Earth Microbiome Project (earthmicrobiome.org) employed Deblur (Amir et al., 2017) to deliver a set of unique sequences with single-nucleotide resolution, rather than representative sequences delivered by OTU clustering. This strategy enabled sequences to be tracked across a variety of studies and habitats and, coupled with environmental metadata, allowed global inferences regarding microbial community structure. Patterns were not dependent on taxonomic assignments, but because exact sequences were provided, taxonomy can later be assigned as reference databases improve. Such approaches can be extended from metagenetics to metagenomics and should continue to improve as longer sequence reads delivered from high-throughput platforms become more routine (Teeling and Glöckner, 2012).

Presence/absence detection appears sufficient for certain ecological quality assessment applications (Aylagas et al., 2016), although improvements in quantification would surely extend the utility of DNA sequencing for marine monitoring purposes. Some studies report adequate estimates of relative abundance (Aylagas et al., 2014, 2016; Elbrecht and Leese, 2015; Tan et al., 2015; Thomsen and Willerslev, 2015) even though read abundance does not necessarily provide a direct correlation to organism abundance because of issues that include variations in copy number, genome size, and growth condition. Bioinformatic solutions to the issue of quantification include estimation of organism abundance based on normalized read counts of clade-specific marker genes (Segata et al., 2012; Sohn et al., 2014), polymorphisms in universal markers (Luo et al., 2015), or accounting for incorrect taxonomic classification through probabilistic models (Lu et al., 2017). Differences in sequencing depth, i.e., the number of reads obtained from an environmental sample, significantly affects results (Rodriguez and Konstantinides, 2014) and can potentially render a study unable to detect low-abundance strains. In cases in which detection or abundance estimation is critical, amplicon sequencing with deep sequence coverage coupled with quantitative PCR (qPCR) may be a workable strategy.

The challenge that residual or ancient DNA poses to quantitation, particularly in benthic environments where DNA turnover is relatively slow, is reviewed extensively elsewhere (Bohmann et al., 2014; Pedersen et al., 2015; Thomsen and Willerslev, 2015). Recent studies in aquatic systems focus on DNA decay rates to better understand the potential of, and limits to, applying eDNA to marine monitoring and stock assessment (Sassoubre et al., 2016). In some cases, qPCR of eDNA is used to address the need for quantification (Laramie et al., 2015). In other cases, qPCR alone is suggested for abundance measurements of harmful algae in marine monitoring programs (Zamor et al., 2012). Overall, overcoming and/or circumventing issues that compromise abundance estimates is a research need that requires attention on all fronts—from the first steps of sample collection and DNA extraction, to routine generation of longer sequence reads, and development of better bioinformatic tools.

DNA sequences can be obtained cheaply and quickly; a decade ago a standard sequencing run returned a few thousand DNA base pairs. Now for the same cost, sequencers routinely generate hundreds of billions of base pairs in a single run. Despite the enormous opportunities posed by this technological revolution, the vast amount of data generated requires new solutions for data handling, storage, processing, documentation, visualization, and dissemination (Desai et al., 2012; Muir et al., 2016). This, in turn, drives a critical need (and current gap) for bioinformatic infrastructure and expertise. Adequate investment must be paid to obtain and maintain the tailored IT infrastructure needed to handle the increasing data volume and sophistication of processing. New developments such as cloud computing provide cost efficient solutions for decentralized storage and analysis of Big Data (Stein, 2010) while facilitating research collaborations. However, a common problem for sequencing projects is a dearth of experienced and skilled bioinformaticians. This is partly due to the rapid expansion of the field; the need has expanded faster than the ability to train scientists (Hughey and Karplus, 2003). In addition, the skill sets needed are diverse and project-dependent. The result is that many bioinformatic tools are executed by researchers without sufficient training and knowledge of the underlying algorithms. Choosing the correct pipeline is important (Siegwald et al., 2017), and competence in bioinformatics is required to choose the appropriate analysis software from the vast number of available tools. Despite the critical importance of bioinformatics expertise to sequencing studies, bioinformaticians can lack appropriate career and development opportunities (Chang, 2015). There is a need for institutions and funding bodies to address this gap and to create attractive career paths and salary models for current and future bioinformaticians. Meanwhile, education and training opportunities are being created to close the bioinformatic workforce gap (Edwards et al., 2013; Atwood et al., 2015), creating a fundamental difference between the expertise gap in bioinformatics and the one for traditional taxonomic identity (Table 1).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer KV and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.