Appendix H

Invited Paper: Technology-Focused Outlook for Future Biological Detection Systems

Disclaimer:

The findings and conclusions in this paper are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention (CDC). References to specific products or trade names are provided solely for informational purposes and do not imply endorsement by CDC or the Department of Health and Human Services. Furthermore, this paper was prepared for the September 18-19, 2017, workshop on Strategies for Effective Biological Detection Systems hosted by the National Academies of Sciences, Engineering, and Medicine. It does not necessarily represent the views of the National Academies, the Department of Homeland Security, or the U.S. government.

Duncan MacCannell, Ph.D., and Toby Merlin, M.D.

1. BACKGROUND

In the 14 years that have passed since it was established, the BioWatch program has become an established tool for environmental biosurveillance in the United States, designed to reliably detect a set of important biothreat pathogens using air sampling in more than 30 municipalities and jurisdictions around the country. Although the probability of deliberate or terroristic aerosolized release of pathogenic microorganisms remains unclear, the 2001 anthrax attacks underscored the devastating potential of such an event and the need for a biosurveillance system that could provide early detection of airborne microbial threats and help to inform responses from federal, state, and local stakeholders in the event of an exposure. It is also important to put the BioWatch program into context, as part of a broader domestic biosecurity strategy. A host of other laboratory-based public health, veterinary, and agricultural biosurveillance programs, and other sources of actionable information, such as syndromic surveillance, clinical reporting, and intelligence assessments, complement BioWatch.

The current BioWatch Gen-2 detection platform is built on a panel of standardized, real-time polymerase chain reaction (PCR) assays that amplify and detect specific marker sequences (or signatures) for a set of known biothreat agents. In monitored areas, venues, and events, the BioWatch program deploys air collection units at designated indoor and outdoor locations to continuously sample the environment. Local personnel are dispatched to inspect the collection devices at regular intervals to monitor their operation and to recover and exchange the filters for testing and analysis. At a local dedicated laboratory, the filters are processed and nucleic acids are extracted according to standardized BioWatch protocols. The resulting extracts are then submitted to a battery of PCR assays to determine whether DNA from target pathogens is present on the air collection filter. Any positive result is flagged and verified by secondary pathogen-specific PCR testing. A result that is positive in both the first and second rounds of PCR testing is known as a BioWatch Actionable Result (BAR) and indicates the presence of pathogen nucleic acid. This means the potential detection of one or more organisms of interest in the sample and may be indicative of an airborne release.

Despite some methodological updates, the core technology of the BioWatch program has remained fundamentally unchanged since its inception. The system bears a heavy reliance on the manual retrieval of air filters and batched PCR testing, resulting in turnaround times from a potential release to a laboratory detection of 12 to 36 hours. The limited repertoire of the platform may also be of concern, since its PCR testing is focused on signatures for a handful of high-risk pathogens, and in the absence of ongoing target refinement, the system may overlook important emerging or bioengineered threats. The past decade has seen remarkable advances in laboratory and diagnostic technology, in particular next-generation sequencing, PCR, and mass spectrometry. While the BioWatch program has made efforts to explore or develop new technologies for continuous monitoring and laboratory detection, none have met the unique performance and operational re-

quirements of the program, and PCR-based methods remain the recommended approach after multiple reports and expert panels (GAO, 2015, 2016).

This white paper reviews and updates the current technology outlook for biological detection systems and discusses their suitability as a platform for the next generation of BioWatch testing. While efforts were made to thoroughly review emerging and established biological detection technologies for the purposes of this report, the summary and examples herein are not meant to be a complete and comprehensive description of all possible diagnostic strategies for environmental biosurveillance. Rather, the authors took the specific operational requirements of the BioWatch program into account and selectively deprioritized technologies that were unlikely to be practical or cost effective within the next 5 to 10 years. In drafting this report, the authors also took into careful consideration the following questions from the committee’s statement of task (Box H-1).

2. CRITERIA FOR THE EVALUATION OF NEXT-GENERATION BIOWATCH DETECTION TECHNOLOGIES

Before considering the range of biological detection technologies that may be suitable for the next generation of BioWatch, it is important to underscore the unique and complex requirements of the BioWatch program and the limitations these will impose on any technological selection. Implementing cutting-edge laboratory and bioinformatics approaches in a single, well-staffed, and well-funded research laboratory is a relatively straightforward undertaking, and with sufficient care and effort, nearly any advanced technology can be made to work consistently. Implementing new technology platforms and robust, production-

caliber protocols that can be standardized and performed across a distributed network of laboratories with different levels of capacity and staff turnover is a much more challenging proposition.

Consider PulseNet, a laboratory-based surveillance system for enteric infections established in 1997. Although newer strain-typing technologies, such as multiple-locus variable number tandem repeat analysis (MLVA) have been standardized and implemented on the network for secondary strain typing, PulseNet still relies on pulsed-field gel electrophoresis (PFGE) as a core technology, despite it being more than 30 years old. PulseNet is in the process of transitioning to next-generation sequencing (NGS) for bacterial reference identification, strain typing, and characterization. Even so, plans to implement NGS and phase out the older strain-typing technologies must take into consideration the capacity requirements and transition capabilities of 87 participating laboratories across the United States. Because PFGE is such a robust and cost-effective approach, and the requirements for newer NGS-based testing are complex and costly, PFGE is likely to remain a core technology for PulseNet during an extended transition period, coexisting with newer, higher-resolution methods (Ribot and Hise, 2016; Swaminathan et al., 2001).

These same network constraints apply to BioWatch. New technologies must be extensively validated and the protocols and instrumentation must be standardized and hardened before implementation across a broad network of participating laboratories. Technology platforms and methods must undergo extensive testing and validation at pilot sites as well, with the development and refinement of new processes for test performance, data analysis, results reporting, validation, and quality assurance. As such, any new technology must simultaneously meet the desired detection and operational requirements of BioWatch while remaining compatible with existing architecture, laboratory capacity, support, and quality assurance and control processes. This is a significant and underappreciated challenge for networked surveillance systems and, like PulseNet, both the implementation and the transition strategy for any new BioWatch technology platform should be carefully considered.

In addition, the BioWatch program has other key factors that will impact the evaluation, applicability, and selection of next-generation technologies for biological detection, including the following:

1. Sensitivity and specificity. To be considered for BioWatch implementation, biological detection strategies must have demonstrably high and reproducible sensitivity, with the ability to identify the presence of pathogen signatures in low counts against variable microbial backgrounds and under uncertain environmental sampling conditions. Ideally, the detection strategy must have a high positive predictive value with the ability to definitively identify positive samples on rapid, first-tier testing. False-positive BARs have important implications for incident response and, at the very least, can undermine confidence in the utility and reliability of the platform (GAO, 2015).

2. Cost per sample. For any large, laboratory-based surveillance network, the total cost of testing is an important consideration and includes a range of factors, including the cost of sample logistics and processing, reagents and consumables, personnel, capital equipment, analysis costs, facility charges, data transmission and storage, and results reporting. For the BioWatch program, which currently processes almost 250,000 samples per year, the cost multiplier on these charges is significant. Any new or proposed biological detection strategy must include careful consideration to ensure that the entire assay can be performed routinely and at scale, while remaining within existing or anticipated personnel, resource, and budgetary constraints.
3. Turnaround time and throughput. For near-real-time biosurveillance, the turnaround time, throughput, and scalability of the assay are all important performance considerations. The current BioWatch system relies on a network of personnel who monitor the air-sampling systems and recover the filters for laboratory testing. This is not continuous monitoring and it means that, under the most optimal of circumstances, several hours could pass between an airborne release event and the first positive filters arriving in the laboratory for testing. While this is a common issue with most semiautomated and distributed environmental sampling technologies, sample ingest is an important consideration for technology selection. The speed of the laboratory testing and reporting is obviously an important consideration as well; technologies such as NGS or microarrays may offer important advantages over PCR but may take 24 hours or more to run, depending on configuration and output parameters. An optimal next-generation detection system would include distributed, continuous monitoring at all sample sites, eliminating the need to retrieve and test the filters. That said, automated or semiautomated continuous biological monitoring is unlikely to reach this level of practical sophistication within the next 5 to 10 years.
4. Platform flexibility and scalability. An ideal biological detection assay will be both flexible and standardizable, allowing a validated testing protocol and target set to be locked down for routine use, but flexible enough to accommodate new versions of detection assays, including modified components, dual-use applications, or the addition of new targets to cover emerging pathogens or features of interest. The assay should be amenable to automation and scale rapidly from a handful of samples up to hundreds of samples, as required. In addition, it should accommodate both random access (the ability to dynamically add new samples to the testing queue) and flexible batching of priority samples.
5. Breadth of scope. The existing BioWatch system is limited to the identification of a limited number of high-risk biothreat agents selected on the basis of those that are the most likely to be effectively weaponized

6. for aerosolized dispersal and/or those with the highest potential for catastrophic damage. This reliance on a narrow set of marker signatures allows for high sensitivity, specificity, and reproducibility but is also an important weakness of the current platform. Specific details of the BioWatch assays, including primer and probe sequences, must also remain carefully guarded to prevent deliberate evasion. An optimal next-generation biological detection system would either be completely pathogen agnostic, capable of detecting all microbial hazards in an air sample, or sufficiently broad and flexible to reliably detect both Tier 1 biothreat agents and other emerging or dangerous airborne pathogens in a sample.

7. Laboratory and bioinformatic complexity. Another important consideration is the laboratory and bioinformatics complexity of the assay, from sample acceptance through to result reporting. A complex laboratory assay often requires more capital and accessory equipment, additional facilities costs, higher reagent and consumables costs, and specific requirements for workforce development and operator expertise. Complex assays often have sophisticated bioinformatics needs, including large volumes of data that must be stored, transmitted, and processed, with corresponding investments in information technology (IT) infrastructure and workforce capacity. The issue of costs aside, the complexity and technical requirements of an assay have important implications for its ability to be standardized and deployed across a network of participating laboratories. The entire assay and analysis process should be streamlined and robust, able to be performed and interpreted by operators in laboratories all over the country. Increased complexity can also result in decreased reliability and impact the turnaround time and reproducibility of the assay. In some instances, the complexity of an assay may be offset by leveraging existing laboratory equipment, infrastructure, and expertise across other infectious disease surveillance programs concomitantly managed within the laboratory network.
8. Validation, quality management, and workflow integration. A suitable biological detection assay and platform is also amenable to validation, with established quality assurance and quality management that may be performed under ISO and Clinical Laboratory Improvement Amendments (CLIA) regulated conditions. The entire testing process, including laboratory procedures, bioinformatics analysis, result interpretation, and reporting, should be validated and auditable to ensure reliable, reproducible results. Quality management also plays an important role in the control of testing costs, since it enables the early identification and resolution of testing issues, improves confidence in the testing process, and limits the need for repeat testing of failed samples. Finally, the assay workflow should integrate with other laboratory processes, including second-tier BioWatch confirmatory assays and

These operational criteria will be used to guide this review of biological detection technologies and may provide important context as the working group considers candidate technologies, platforms, and strategies for the future of the BioWatch program.

3. OVERVIEW OF BIOLOGICAL DETECTION TECHNOLOGIES

In June 2013, the Department of Homeland Security commissioned the Institute of Medicine and the National Research Council to convene an expert workshop on potential future technologies for the BioWatch Program. This expert review assessed the technical readiness of autonomous and semiautonomous solutions for biological detection of airborne pathogens and considered four principal categories of technological approach: (1) nucleic acid signatures, (2) immunoassays and protein signatures, (3) genomic or next-generation sequencing, and (4) mass spectrometry (Hook-Barnard et al., 2013). The following overview of detection technologies will revisit several of the same general detection strategies, with emphasis on nucleic acid amplification and next-generation sequencing.

Assuming a 5- to 10-year implementation roadmap for advanced technology, we believe the focus must be on detection strategies that are already working in the laboratory for a broad range infectious disease diagnostics, or that have at least demonstrated a functional proof of concept and an aggressive trajectory for development. Scaling, standardizing, hardening, deploying, and validating a complex bioassay across a network of laboratories requires significant investment, planning, logistics, and workforce development. This process alone can take several years, even under optimal conditions, and must be taken into consideration when determining the selection and rollout of next-generation BioWatch detection strategies.

3.1 Nucleic Acid Signature Detection

3.1.1 Real-Time PCR

The current Gen-2 BioWatch system uses a battery of real-time PCR (rtPCR) assays to identify and characterize a range of different high-risk biothreat pathogens based on an initial set of signatures originally developed by biosecurity researchers and Lawrence Livermore National Laboratory, which have undergone several generations of subsequent refinement (Committee on PCR Standards, 2015; Slezak et al., 2003). The assay panel, which is designed with limited multiplexing, has a 10- to 36-hour time to answer which typically includes an approximately 24-hour aerosol collection cycle, 4 hours for filter recovery and transportation, 6

hours for primary rtPCR screening, and 2 hours for confirmatory, pathogen-specific testing (IOM, 2011). While continuous monitoring and onboard sample processing and detection are planned for the next generation of air sampling instruments, it is important to underscore the fact that the existing Gen-2 assays have performed well over the past decade and across a range of different settings and contexts. Therefore, one of the simplest options to improve the detection process, and “bridge” to Generation 3, would be to upgrade existing or expanded rtPCR primers and probe sets.

3.1.2 Highly Multiplexed Real-Time PCR Luminex xMAP/MAGPIX™. The Luminex MAGPIX platform is the smallest of the company’s xMAP instruments and remains a popular choice for high-order multiplexing of protein and nucleic acid–based assays. The current generation of instruments can interrogate up to 50 different multiplexed targets across 96 samples using bead-based affinity or hybridization probes (Luminex Corporation, 2014). The flexibility of bead-based arrays makes them an attractive development platform for broad-based pathogen identification assays. For example, Los Alamos National Laboratory recently described a number of different applications of multiplex oligonucleotide ligation PCR, in which multiplex ligation and amplification are used to interrogate dozens of targets per sample with amplicon capture and resolution using Luminex microbeads (Woods et al., 2016; Wuyts et al., 2016). More conventionally, Luminex has recently received Food and Drug Administration (FDA) approval for xMAP diagnostic panels for gastrointestinal and respiratory pathogens and emergency use authorization for Zika virus.

Of particular relevance, Luminex MAGPIX was a candidate platform for recent research and development efforts that were undertaken to update BioWatch detection technologies. It was chosen, in part, due to the relative technical maturity of the xMAP platform, the reasonably straightforward process of adapting existing BioWatch Gen-2 primer and probe sets, and the inherent flexibility to alter target sets and mix detection strategies based on changing operational requirements. That said, bead-based array technologies have historically proven costly and challenging for many large-scale laboratory deployments (including biodefense), primarily due to difficulties in maintaining the instrumentation and microbead libraries and the level of workforce expertise required for sustained operation, quality management, and troubleshooting. The BioWatch MAGPIX evaluation project was ultimately discontinued in 2016, when it became apparent that the cost and complexity of the assay were not sustainable within the context of the program.

ThermoFisher TaqMan Array and OpenArray™. TaqMan Array Cards (TACs) are a relatively simple and cost-effective approach for high-order multiplexing that patterns individual reactions across a low-density, 384-well array. Using this approach, Liu and colleagues recently described the adaptation of TaqMan primer/probe sets for 26 different pathogens associated with acute febrile illness to probe six to eight samples on a single array card, at relatively low expense and with results for all targets within 2.5 hours on the ViiA7 platform (Liu et al., 2016). Another option from ThermoFisher is the OpenArray, a higher-density TaqMan array technology that it acquired from BioTrove in 2009 and recently implemented on QuantStudio 7/12k Flex real-time instruments (ThermoFisher, n.d.). The OpenArray chips include nanoliter-scale wells and the ability to simultaneously multiplex 12,228 different reactions in as little as 30 minutes. Depending on the configuration of the OpenArray and the desired operational characteristics of the assay, this platform could interrogate dozens of BioWatch samples for upwards of 96 different targets in a single run.

While these technologies are powerful options for arrayed multiplexing, both share significant limitations that impact their suitability for BioWatch implementation. First, both the TaqMan Array and OpenArray platforms require specific, high-end rtPCR instrumentation, so implementing these approaches across the BioWatch network would represent a significant new investment in terms of capital equipment costs, training/proficiency, and maintenance and support. Second, because these instruments partition the sample across thousands of nanoliter-volume wells, the limit of detection for low-abundance pathogen sequences in a sample is likely to be markedly higher than in a single-tube reaction. The sensitivity and specificity of these array-based assays would need to be carefully evaluated against realistic environmental samples, including suitable preamplification or enrichment protocol steps. Finally, because the arrays are fabricated by the vendor, the primer/probe sets and array layout are both relatively inflexible. This inflexibility may actually be an advantage for distributed surveillance applications such as BioWatch, since the arrays would be manufactured and distributed by the manufacturer, limiting potential errors and inconsistency with array setup and configuration. While on-demand manufacturing of the arrays may help to limit user error, it introduces an important dependency on the vendor’s manufacturing and quality management processes and may result in bottlenecks under conditions of surging or unpredictable testing demand.

Wafergen SmartChip™. The Wafergen SmartChip is another option for array-based, highly multiplexed real-time PCR. Each SmartChip includes 5,184 nanoscale wells that can be configured in a variety of different array and sample layouts, ranging from 12 assays × 384 samples to 384 assays × 12 samples. Unlike the ThermoFisher arrays described above, the SmartChip has significantly more flexibility, since the end user can print new array configurations, as required, within 30 to 60 minutes. The chips and other consumables are not inexpensive, but with a well-designed configuration, the Wafergen SmartChip could interrogate hundreds of targets from dozens of filters in a single 2.5-hour instrument run.

Despite these advantages, the platform shares many potential disadvantages with the ThermoFisher array technologies. First, it represents a significant investment in capital instrumentation, requiring a proprietary liquid handling system (MultiSample NanoDispenser) in addition to the SmartChip thermocycler itself. Second, like the ThermoFisher arrays, and depending on the sample extraction and preamplification workflow, low-abundance targets may not be detected with sufficient sensitivity due to the partitioning of the sample across thousands of individual reaction volumes. While Wafergen stresses the extensive dynamic range of the SmartChip, and a bevy of preamplification options, it nonetheless recommends 3 to 10 ng/µL DNA input per 100-nL well, which would prove challenging in practice with environmental samples. Real-world performance characteristics would obviously need to be assessed. Lastly, unlike the ThermoFisher QuantaStudio and ViiA7 instruments, which are both frequently found in academic, clinical, and public health laboratories, the SmartChip system is not currently widely used in microbiology laboratory settings. As such, implementing these systems across a broad network of laboratories would require additional training, maintenance, and vendor support, all of which must be factored into technology selection. Because the SmartChip platform is relatively new to diagnostic microbiology and biosurveillance applications, instrument and application support may be limited or a secondary priority for the vendor.

Multiplexed Droplet Digital PCR (ddPCR) Technologies. Droplet digital PCR-based technologies are a powerful quantitative approach for the detection of genetic targets and have been used for a range of microbial and ecological sample types and applications. Conventional ddPCR instruments, such as the BioRad QX200, use emulsion-based, nanoliter-scale droplet reaction vessels to precisely quantify the presence of target sequences using hybridization or hydrolysis-based rtPCR. In general, most ddPCR instruments on

the market today are designed neither for scale nor for multiplexed signal detection, although instruments that incorporate both features remain under active development.

One example of upcoming instrumentation that incorporates higher-order multiplexing are systems developed by LexaGene. These instruments combine microfluidics and ddPCR to quantitatively query samples for up to 12 different targets. Lexagene promises high sensitivity, specificity, flexibility, and low-cost-per-sample operation; however, they do not currently have a commercially viable product to evaluate these claims. Even so, preliminary results seem promising and continued advancements in both microfluidics and highly multiplexed rtPCR from this group and other companies will likely result in new instrumentation options from a number of different vendors within the next 5- to 10-year timeframe. Whether these systems will be a good fit for distributed environmental surveillance programs, such as BioWatch, remains to be seen.

3.1.3 Commercial rtPCR-Based Diagnostic Panels

While an open and flexible detection strategy is likely going to be the most preferable option for BioWatch in the long term, commercial PCR-based diagnostic panels are worth considering as a supplementary or transitional technology option, particularly if there is engagement and support from instrument manufacturers in the private sector. Most commercial assays for biothreat pathogens are designated for research use only and some, such as the Cepheid™ BA test, which provides identification and limited characterization of Bacillus anthracis alone, are very limited in scope.

Others, such as the BioFire FilmArray™ Biosurveillance panel (bioMérieux™), provide more comprehensive coverage: amplifying 26 targets for the identification of 16 different high-risk biothreat agents (FilmArray BioThreat Panel, n.d.) (B. anthracis, B. melitensis, Burkholderia spp, Clostridium botulinum, Coxiella burnetii, EBV [Zaire], EEEV, Francisella tularensis, Marburg virus, ricin, Rickettsia prowazekii, variola, VEE, WEE, Yersinia pestis, and orthopox virus). The development of this assay was initially funded by the Department of Defense Joint Biological Agent Identification and Diagnostic System (JBAIDS) contract and the commercial-off-the-shelf assay is currently widely used by field and rapid response teams. In many ways, this assay platform has similar performance characteristics to today’s BioWatch rtPCR-based detection, but there are a number of key advantages that make it worth discussing here: (1) the system uses nested PCR, which affords both high sensitivity and high specificity across a narrow range of targets; (2)

the instrument has pushbutton operation, a 1-hour runtime, and requires minimal sample preparation or operator involvement; (3) the hardware is relatively inexpensive, portable, and amenable to use in mobile laboratories or temporary settings; (4) reagents and consumables do not require cold-chain management and are readily available; and (5) because the BioFire platform has already been FDA approved for a number of different syndromic diagnostic panels, the instruments are increasingly common in clinical and public health reference laboratories around the country. This last point could greatly extend the reach and surge capacity of BioWatch and facilitate more flexible deployments of the technology, although low instrument throughput and the relatively high cost of consumables are both important constraints.

Reliance on “black box” commercial platforms for widespread surveillance also has important disadvantages. Namely, these platforms limit flexibility and control over the assay workflow and abdicate some responsibility for ongoing performance monitoring and validation. Perhaps most importantly, use of commercial platforms creates a specific and ongoing dependency on a single vendor, which can introduce both bottlenecks and procurement challenges. While single-vendor monopolies or market dominance are not uncommon in the biological sciences (e.g., BioRad™ [PFGE], Illumina™ [NGS], Applied BioSystems™ [Sanger], Covaris™ [Ultrasonication], etc.), their success in the context of large-scale surveillance networks lies in the openness and flexibility of their products, and in balancing the advantages of standardized instrumentation and turnkey assays with custom requirements or changing surveillance needs.

3.1.4 Isothermal Amplification

While real-time, multiplexed isothermal amplification remains impractical for large-scale biosurveillance use, its utility as a field-deployable diagnostic device merits some brief discussion. Loop-mediated isothermal amplification (LAMP) and other isothermal methods have been widely used in both commercial diagnostics and research applications and are favored due to their ability to amplify genetic targets with high specificity and with minimally sophisticated laboratory equipment. A number of studies have explored the use of isothermal amplification for field-portable diagnostics, including multiplexed identification of viruses, bacteria, and parasitic diseases. One such platform, used for real-time detection and species-level identification of Plasmodium spp., is noteworthy, since it is small, inexpensive, provides real-time LAMP amplification curves, and can be run completely off battery power.

TABLE H-1 Performance Characteristics of Selected Nucleic Acid Amplification Technologies

Advances in microfluidics have also made low-cost, continuous-flow embedded LAMP devices possible, which may provide the technological foundation for on-device integrated monitoring and detection (Rane et al., 2015). As these technologies improve, they may provide the basis for inexpensive onboard biodetection for air sampling devices in a future generation of the BioWatch program. At present, however, their performance characteristics, namely, complexities with multiplexing and lower sensitivity and specificity relative to current rtPCR-based methods, limit further consideration (Patel et al., 2013).

3.1.5 Microarrays

DNA microarrays have historically been an important tool for broad-based pathogen identification and targeted resequencing efforts. Throughout the 2000s, a number of research groups developed and refined large-scale conventional microarrays to identify and characterize pathogens from environmental or clinical samples. These included the ViroChip from the DeRisi lab at the University of California, San Francisco (Chen et al., 2011), the GreeneChip, an international collaborative effort (Palacios et al., 2007), and the Lawrence Livermore Microbial Detection Array from Slezak and colleagues at Lawrence Livermore National Laboratory (Gardner et al., 2010), among others. DNA microarrays offered an important middle ground between narrowly focused PCR-based pathogen assays and larger-scale next-generation sequencing-based methods. While they continue to be a useful and cost-effective strategy for pathogen identification and discovery, they likely are not a suitable candidate for the BioWatch program because of the following limitations.

• Cost: Although the cost of individual microarrays is relatively low, most arrays cannot be easily multiplexed and cost per sample is likely to be unacceptable at production scale.
• Long turnaround time: Hybridization microarrays require extended hybridization, depending on design and probe affinity. A 24- to 48-hour hybridization step is typical, particularly for weak or nonamplified sequences.
• Variable sensitivity and dynamic range: Microarray analysis is typically based on image analysis of the array and comparison of relative intensity. This introduces important subjectivity and requires significant computational and analytic capacity to analyze and interpret microarrays consistently. For distributed surveillance, this analytical complexi

• ty would require the development of new software to normalize, share, and compare microarray image data.

• Market instability: With the departure of Roche Nimblegen from the microarray market several years ago, many microarray-based assays have been transitioning over to Affymetrix arrays or other platforms. This requires new investment in equipment and support, and significant effort to redesign and revalidate chip designs (McLoughlin, 2011).

As the cost and technical barriers to next-generation sequencing continue to decline, a number of these microarray target sets are being adapted to targeted NGS platforms. This is discussed in more detail in Section 3.3.2.

3.2 Immunoassays and Protein Signature Detection

The 2013 workshop and review of potential technologies for the BioWatch program identified several options for autonomous or semiautonomous biological detection using immunoassays and protein signatures, including multiplexed immunoassays, xMAP enzyme-linked immunosorbent assay (ELISA), the PathSensors CANARY sentinel system, BioHawk, Ramen spectroscopy (Battelle REBS), and SIMOA single-molecule arrays (Hook-Barnard et al., 2013). This list of candidate approaches seems relatively complete, based on our review of emerging technologies, and we will not revisit them here, with the exception of the last: SIMOA single-molecule arrays. Since the publication of the previous workshop report, Quanterix has released viable production hardware based on the SIMOA technology concept and has recently released a newer benchtop version of the instrument in late August 2017, which may merit further consideration.

3.2.1 Quanterix SIMOA SR-PLEX13

Quanterix SR-PLEX recently announced a new compact benchtop instrument based on their SIMOA single-molecule digital detection technology. SIMOA uses 2.7-micron paramagnetic, dye-labeled microbeads bound to conventional antibodies or probes and is sensitive enough to detect a single molecule of bound protein. The detector uses a biotin-streptavidin conjugate with a secondary fluorescent reporter to produce localized, detectable signal across an array of individual microwells. While performance data on this new instrument remain sparse, according to manufacturer specifications, the detection technology will have a thousandfold higher sensitivity over standard immunoassays, and, importantly, the same microbead reporter technology and instrumentation can also be used for ultrasensitive detection of nucleic acids without PCR or other amplification. This suggests that a single platform device could be used for the capture and detection of low-abundance nu-

cleic acids, pathogenic microbes, and some airborne toxins. The instrument is capable of multiplexing, detecting up to six simultaneous protein or nucleic acid targets per test, and can complete 96 samples in approximately 3 hours. Pricing and performance data were not available at the time of this writing. However, data from their first commercial instrument, the SIMOA HD-1, supports the feasibility of most performance claims (Quanterix, n.d.).

3.3 Next-Generation Sequencing (NGS)

The first next-generation-sequencing instruments were introduced to the commercial market in 2005 and both rapidly and fundamentally changed the nature and practice of bioscience research. These systems decreased the overall cost of sequencing by five to six orders of magnitude, with a corresponding increase in the volume and complexity of sequence data that they produced. In the decade that followed, the technology space around NGS has evolved rapidly, with the development of a range of new sequencing applications and platforms and iterative improvements to instrumentation, reagents, and consumables. Today, a number of different sequencing technologies are available with different form factors, performance characteristics, error models, and operational requirements. NGS has already proven itself to be a powerful and flexible tool for the study of infectious diseases and is increasingly being used for outbreak investigations and large-scale laboratory-based surveillance programs alike.

An important advantage of NGS is that most sequencing workflows can accommodate DNA and/or RNA from a variety of different sources, including host, vector, pathogen, and the environment. And these may include whole sequences, fragments or selectively amplified targets. In general, these workflows consist of five or six steps, beginning with a clinical or environmental specimen or pure culture isolate: (1) DNA/RNA extraction, isolation, conversion, quantitation, and quality assessment; (2) enzymatic or mechanical shearing, size selection, and NGS library construction; (3) presequencing cleanup and quality assurance; (4) sequencing; and, finally, (5) downstream data processing, bioinformatic analysis, and result reporting. In order to successfully and sustainably implement NGS in the context of large-scale environmental biosurveillance, all aspects of the sample-to-result workflow must be carefully considered.

The following section provides a brief overview of current and imminent sequencing technologies.

3.3.1 Overview of NGS Platforms

Illumina. Illumina short-read sequencers are currently a predominant platform in many next-generation sequencing markets, with a range of instruments based on a proprietary reversible dye terminator sequencing. Illumina instruments range in size and capacity from small benchtop sequencers, such as the MiniSeq or MiSeq, up to large-scale capital instruments, such as the HiSeq, NovaSeq, and

X Ten, which often require a significant commitment of resources and dedicated facilities. Illumina sequencers are general-purpose instruments that can be run in a variety of different configurations to satisfy many different applications. Depending on configuration, these instruments generate millions to billions of short reads (25 to 300 base pairs [bp] in length) with runtimes that range between several hours and 6 to 7 days. Because of their versatility, massive output, and relatively low cost of ownership and operation, short-read sequencing has rapidly become an established workhorse technology in many academic, clinical, and public health laboratories throughout the world. Despite these advantages, short-read sequencing can present important challenges for downstream bioinformatic analysis and interpretation. This is particularly true in the sequencing of genomes with complex structure or extensive repeats, or in specific applications, such as metagenomics, where mixed assembly, recombination, or phasing may be of concern. In these instances, short-read sequencing data are often paired with long-read sequencing for hybrid assembly or optical mapping, or, alternatively, the results from Illumina sequencing are validated using secondary tests and orthogonal methods. Illumina does support a version of the MiSeq, the MiSeqDx, that is developed specifically for regulated IVD testing environments, the platform has only recently opened to custom IVD assay development. (An overview of Illumina sequencers is provided in Table H-2).

In general, sequencing libraries for Illumina instruments are interchangeable, meaning that a sample prepared for sequencing on a MiSeq may be run for similar purposes on a larger instrument, such as the HiSeq or NovaSeq, without modification. The relatively high output of these sequencers also allows for convenient multiplexing: in practice, most conventional microbial genomic sequencing applications can be safely multiplexed to run dozens of samples per flow-cell lane. Multiplexing is also possible for metagenomic sequencing, depending on the nature of the sequencing (amplicon versus shotgun) and the background complexity of the sample.

Despite the market share and flexibility of the Illumina instruments, a number of logistical and practical considerations may impact its use across laboratory networks. First, like most NGS instrumentation, Illumina sequencers require training and experience to ensure consistent results. While the majority of state public health departments, and a growing number of clinical laboratories,

TABLE H-2 An Overview of Commercially Available Sequencers

NOTE: SBS, sequencing by synthesis; SMRT: single molecule real time. Error type: ^asubstitution, ^bIndel, ^cHomopolymer.

SOURCES: Table adapted from MacCannell (2016). See http://www.illumina.com/systems.html; http://www.pacb.com/products-andservices/pacbio-systems; https://www.thermofisher.com/us/en/home/life-science/sequencing/sequencing-technology-solutions.html; https://www.nanoporetech.com/products/specifications.

have MiSeq instruments available and qualified staff, the infrastructure to support a network of standardized NGS instrumentation poses important considerations. A more important obstacle to the use of Illumina sequencing is time required to extract and prepare samples, and the instrument runtime at the bench. For example, the runtime on a MiSeq can range between 4 and 55 hours, depending on the configuration and data output requirements of the experiment, not including the presequencing workflow and library construction. For critical biosurveillance monitoring applications, a 1- to 3-day turnaround time is not an acceptable compromise in testing performance, regardless of the quality and usefulness of the data that may be generated. For this reason alone, the Illumina MiSeq platform may not be suitable for BioWatch needs. In order to ensure an actionable turnaround time, an Illumina system would need to be run in 1×36 bp (4-hour) or 2×25 bp (5.5-hour) configurations, resulting in 540 to 850 Mb of data under optimal circumstances (Illumina, n.d. a). Although the cost of sequencing continues to decrease, most laboratories also need to consider specimen batching and multiplexing to optimize the cost effectiveness of each run. With these limitations, and the relative complexity of an environmental air filter sample, Illumina metagenomic sequencing would likely be limited to amplicon-based approaches. Even then, an optimistic turnaround time from sample to result would be on the order of 10 to 12 hours, and the sequencing workflow would need careful development, optimization, and validation of both target signatures and controls.

In early 2016, Illumina announced the development of a new modular sequencing platform, codenamed Project Firefly (Illumina, n.d. b), which is targeted for commercial release in late 2017. This platform is intended to be small and inexpensive ($30,000 instrument, $100 consumables) and includes an automated library construction module for sample preparation. Unlike other Illumina instrumentation, the Firefly will implement a single channel of Illumina’s sequencing by synthesis (SBS) chemistry, on a new semiconductor-based detector, generating up to 1.2 giga base pairs (Gb) of data per run with 99 percent base accuracy (Illumina, n.d. b). Although the performance specifications of the platform remain sparse as of this writing, it is likely that this modular system will still require 2 to 4 hours for sample preparation and library construction, with an additional 3 to 12 hours for sequencing, even with faster semiconductor-based signal detection (Robison, 2016).

ThermoFisher IonTorrent. IonTorrent semiconductor sequencing relies on an array of highly tuned ion-sensitive field effect transistor complementary metal-oxide semiconductor (CMOS) detectors to pick up hydrogen ions that are released during DNA polymerization. The underlying technology was originally developed by a research team at Imperial College London, who formed a spinoff company called DNAe (DNA Electronics) to develop and license the intellectual property. While the most notable commercial implementation of semiconductor sequencing is the ThermoFisher IonTorrent line of sequencers, the technology has also been licensed to Roche, GenapSys, and others for sequencing or macromolecular detection, and the core technology remains under active development by multiple companies.

In recent years, ThermoFisher has largely shifted product development efforts away from general-purpose sequencing and toward highly multiplexed amplicon sequencing for oncology, infectious diseases, and other clinical markets. Their most recent lineup of instruments includes the Ion S5 and Ion S5XL, which effectively replace and deprecate the Personal Genome Machine and Ion Proton, although both are still available for purchase. Both instruments offer rapid setup, with sample preparation, library construction, and sequencing requiring less than 45 minutes of operator hands-on time and, depending on the configuration, have sequencing runtimes of 2.5 to 4 hours. Both are also benchtop instruments, capable of generating up to 80 million reads (15 Gb) per run, with streamlined reagents and consumables that have been optimized for clinical targeted sequencing diagnostic workflows (ThermoFisher, n.d. c).

The combined high throughput and high output of the S5 and S5XL are likely compatible with current BioWatch laboratory performance expectations. With a suitable validated AmpliSeq sequencing panel, IonTorrent could allow for highly multiplexed sequencing and sample turnaround in under 6 hours, with an hour of operator hands-on time. AmpliSeq and other highly multiplexed amplicon sequencing (HMAS) technologies are discussed in a later section.

Pacific BioSciences (PacBio). Long-read sequencing first rose to common use with the introduction of the Pacific BioSciences RS sequencer in 2010. While early versions of the platform were plagued by relatively high rates of error, the current generation of instruments, sequencing chemistry, and signal processing algorithms have improved the utility and reliability of PacBio long-read sequencing to the point where bacterial genomes are routinely closed as high-quality draft sequences, either with or without ac-

companying short-read sequence data for hybrid assembly. PacBio instruments perform single-molecule, real-time (SMRT) sequencing and generate hundreds of thousands of reads with average read lengths of 3 to 10 kilobases (kb) with runtimes of 30 minutes to 6 hours. Because the instrument sequences individual DNA molecules, PacBio sequencing has become increasingly useful for deep sequencing and metagenomic applications, and furthermore, because the system can also detect methylated bases by differences in reaction kinetics, epigenetic data are collected simultaneously during the run. Despite these advantages, PacBio sequencing has remained largely impractical for many laboratories due to the size, cost ($750,000), and infrastructure requirements for the large PacBio RSII instruments. In 2016, Pacific BioSciences introduced the Sequel, an instrument jointly developed with Roche. The Sequel has a significantly smaller physical footprint, higher sequence output, and lower cost ($350,000), and seems well positioned for both research and routine production sequencing.

Another important advantage of PacBio SMRT sequencing is that it is possible to directly identify certain epigenetic features over the course of normal sequencing operations. Methylation of bases in the template sequence affects the kinetics of the DNA polymerase as it passes over them, and by analyzing differences in the spacing between sequence pulses (the interpulse duration or IPD), it is possible to computationally identify certain base modifications, including 6-methyl adenine, 4-methyl cytosine, and Tet-converted 5-methyl cytosine, directly from the sequence data. The bacterial methylome is believed to play an important role in the restriction and modification of bacterial genomes and the gain or loss of genes related to virulence and antimicrobial resistance. A number of recent studies have begun to leverage PacBio epigenetic data to understand changing patterns in antimicrobial susceptibility.

There are also some important challenges with PacBio sequencing:

1. Relatively high single-pass error rate (~15 percent). While there has been significant improvement to the chemistry and base-calling algorithms in recent iterations and many of the random errors may be addressed with significant depth or CCS, the error rate of the instrument may still present challenges to certain sequencing projects.
2. Sample requirements. Another important challenge involves the DNA sample requirements for library preparation and sequencing. Since the PacBio sequences single molecules of DNA directly, and does not typically employ an amplification

3. step, library construction requires 1 to 10 micrograms (µg) of high-quality, purified DNA as starting material, depending on the intended application and library insert size. (A 10- to 20-kb insert library typically requires close to 10 µg without contaminants or damage.) In many situations, such large amounts of purified, high-quality DNA may not be feasible and, because of the read lengths involved, whole-genome amplification techniques are often not an option.

4. Output limitations. Relative to short-read sequencers, the PacBio RSII has limited value for certain applications, such as the sequencing and analysis of complex metagenomics samples, since the runs are relatively costly and the number of reads produced is far lower. That said, the long read lengths produced by the PacBio are an important advantage in the phasing and assembly of metagenomics samples and may add critical value for less complex metagenomic samples or with the improved output from the newer Sequel instrumentation.
5. Multiplexing. Multiplexing, or the ability to run multiple samples on a single sequencing run, remains challenging on the PacBio RSII and has yet to be fully developed and validated for the Sequel. Currently, most laboratories are running single isolates per SMRT cell, which results in exceptional draft genome assemblies but rapidly becomes cost prohibitive at scale. The lack of established multiplexing methods is an important limitation to microbial sequencing on the newer Sequel instruments, which have significantly higher output that cannot be applied to multiple samples.
6. Cost and infrastructure requirements. With Pacific BioSciences systems costing upwards of $350,000, the implementation and maintenance costs across a large-scale surveillance network are significant. While the newer Sequel is engineered for clinical laboratory deployments, the older RSII systems also required dedicated facilities and an engineering assessment.

Nanopore (ONT, Roche/Genia). Nanopore sequencing is another more recent long-read sequencing technology, which relies on an array of membrane-embedded, protein nanopores to linearize and sequence molecules of nucleic acid. Most of these platforms measure the current across each nanopore as the DNA molecule passes through the channel, converting these signals into base calls. The Oxford Nanopore MinION was first introduced in 2012, with a preview program that launched for early adopters in 2014. Much like the PacBio RS, the first iterations of the MinION platform were beset by high rates of error, upwards of 15 or 20 percent. Recent improvements in the hardware, chemistry, nanopore configu-

ration, and base-calling algorithms have all greatly improved the accuracy and throughput of the MinION with the release of the MinION Mk 1B in 2016. While error rates remain in the range of 5 to 15 percent, even with the most recent R9 flow cells and chemistry, competent MinION users report up to 10 to 20 GB of sequence data from each flow cell (over 48 hours) and often generate sufficient depth to compensate for sequencing error. Unlike PacBio long-read sequencing error, which is relatively random, errors in Nanopore sequence data are more systematic and predominate in homopolymeric regions.

Nanopore sequencing has shown promise for bacterial genome assembly and is an important new tool for real-time metagenomics. The MinION platform is particularly compelling for a number of reasons:

1. Its diminutive size makes NGS in remote field locations a practical option for infectious disease surveillance, diagnostics, and public health research. For example, the MinION has recently been used for sequencing of both Ebola and Zika viruses, with reference and surveillance sequencing applications both in the laboratory and in the field (Greninger et al., 2015; Quick et al., 2015).
2. Its equipment cost, at a fraction of most short- and long-read sequencing platforms, makes nanopore sequencing an increasingly feasible option for smaller laboratories and those that cannot justify or support sustained investment in large capital instrumentation.
3. Nanopore sequencing is also significantly faster than many other sequencing approaches since instrument cycling is not required and signal data can be streamed directly from the instrument. This streaming model enables rapid base calling and real-time analysis, allowing users to “read until” sufficient data have been gathered and to continuously assess the progress and quality of data generation, rather than analyzing the entire data set after the completion of a run. Basic sample preparation is also extremely fast compared to other platforms, with a streamlined 10-minute protocol, and an inexpensive sample preparation module (VolTRAX) under development. This combined rapidity is particularly important for time-sensitive applications, such as the detection and characterization of biothreat agents.
4. Engineered protein nanopores also offer remarkable flexibility in terms of format, capability, and scalability and have been demonstrated to support direct sequencing of other types of

In addition to existing systems from Oxford Nanopore, a number of other nanopore-based platforms are currently under development or in the premarket space, including a semiconductor-based platform from Roche Genia (www.geniachip.com), and several early-phase proof-of-concept platforms from academic laboratories. As nanopore-based sequencing technologies continue to mature, and other systems enter the market, these flexible, cost-effective sequencing instruments will almost certainly play an increasing role in microbiological testing, particularly in small laboratories, clinics, and the field.

3.3.2 Metagenomics: Highly Multiplexed Amplicon Sequencing (HMAS)

Targeted metagenomic sequencing has been a mainstay of microbial ecology for over a decade, with large-scale surveys based on NGS of partial 16S and ITS sequences. While 16S sequencing has been an important tool for understanding environmental biodiversity and seems to offer more sensitivity than shotgun sequencing-based approaches, it often lacks the ability to resolve critical differences between genera, species, and strains (e.g., 16S is unable to reliably distinguish between many Enterobacteriaceae; Tessler et al., 2017). Modern NGS-based approaches allow for high-level multiplexing of species- and strain-specific amplicons, enabling users to simultaneously interrogate dozens or even hundreds of different targets across multiple samples in a single sequencing run. In the sections that follow, three HMAS technologies are evaluated and discussed.

For highly multiplexed amplicon sequencing, the presequencing work-flow is the most critical rate-limiting step. Unless sample extraction, PCR, library construction, and loading can be reliably optimized and accelerated, it is unlikely that adapting highly multiplexed amplicon sequencing to faster sequencing technologies, like nanopore, will significantly improve the throughput and performance of the assay.

3.3.2.1 IonTorrent AmpliSeq

In recent years, ThermoFisher has moved the IonTorrent NGS platform away from general-purpose short-read sequencing, and has instead prioritized the development of AmpliSeq HMAS assays for

precision oncology, forensics, gene expression, inherited diseases, and other diagnostic applications. AmpliSeq panels can amplify hundreds or even thousands of targets in a given sample using highly multiplexed PCR primer pools (up to 24,000 primer pairs). Like most HMAS approaches, these assays are highly sensitive and are ideal for working with low-abundance targets and complex or difficult sample types (ThermoFisher, n.d. a).

At present, ThermoFisher has prebuilt assays for a range of different bacterial, viral, and fungal identification tasks, as well as a new 48-target panel for acquired antimicrobial resistance genes. Several groups, including the development team for the Lawrence Livermore Microbial Detection Array, are also exploring the utility of adapting and expanding target sets for broad-based microbial identification and characterization using microarrays.

With the IonTorrent S5 ($65,000) sequencing workflow and the Ion Chef automation system ($55,000), a single laboratorian could prepare and run custom AmpliSeq panels against 8 to 24 samples and load the products onto one or more chips–all within 7 hours, including only 15 minutes of hands-on time. Depending on the specific configuration, IonTorrent sequencing could be completed in as little as 2.5 hours, resulting in a total turnaround time of 10 hours or less and a cost of $65 to $120 per sample (ThermoFisher, n.d. b). ThermoFisher’s sequencing workflow has been optimized for clinical laboratories, where next-day results for genomic testing are an acceptable trade-off for reliability, workflow simplicity, and operator efficiency. For routine biosurveillance, however, 10 hours is an important constraint, particularly since it is unlikely that the Ion Chef workflow can be shortened or optimized without sacrificing quality or performance of the assay. Even so, because of the nature of semiconductor sequencing, this ~10-hour turnaround time is significantly faster than comparable Illumina-based HMAS.

3.3.2.2 Illumina TruSeq Custom Amplicon

Illumina offers a similar, customizable assay for highly multiplexed amplicon sequencing that enables the amplification and targeted resequencing of up to 1,536 targets in a single reaction, or more than 150,000 targets, across a 96-well sample plate. While this technology benefits from the flexibility and relative ubiquity of Illumina sequencers in most laboratories, its performance characteristics are likely not compatible with BioWatch requirements. According to Illumina, TruSeq Custom Amplicon requires at least 6 hours for amplification and library construction and an additional 27 hours for MiSeq 2×150 bp sequencing, with 2.5 hours of opera-

tor hands-on time. Even if ultrashort (1×36 or 2×25 bp) Illumina sequencing could be substituted and made to work in this context, the entire workflow would still require 10 to 12 hours to complete under optimal circumstances. This is likely unacceptable for response purposes.

3.3.2.3 Fluidigm Targeted Sequencing Library Preparation Systems

Over the past decade, Fluidigm has introduced several generations of microfluidic array systems for high-throughput amplification and library construction, upstream of targeted DNA resequencing. The Access Array and, more recently, the Juno use microfluidic chips to amplify up to 192 samples against thousands of prepooled primers in a single instrument run, with amplicons ranging in size between 150 and 500 bp. The Juno requires 4 hours of setup time, and 8 to 9.5 hours to complete amplification and library construction. Up to 1,536 samples can be multiplexed on a single sequencing run using conventional barcodes; importantly, both Illumina and IonTorrent sequencing can be used.

While the Juno and Access Array are powerful and flexible instruments for research applications, they are likely impractical for large-scale surveillance use. First, turnaround time continues to be an important consideration, with estimated turnaround times of 14 to 36 hours, with 6 hours of setup. Perhaps more importantly, Fluidigm systems are not common in most microbiology laboratories and would require significant new capital investment, as well as ongoing commitments in terms of reagents, maintenance and support, and workforce training to accommodate any new task-specific instrumentation.

3.3.3 Metagenomics: Shotgun Sequencing

Shotgun metagenomic sequencing, or whole-sample sequencing, has increasingly been used for pathogen discovery and diagnostics, with several health care institutions in the United States now offering routine clinical metagenomic testing under CLIA. In shotgun sequencing, nucleic acids are extracted directly from a sample, usually without amplification or modification, reverse transcribed if necessary, and submitted for next-generation sequencing. For most applications, Illumina short-read sequencing is the platform of choice, primarily due to the quality and quantity of data that are produced. For complex clinical and environmental samples, maximizing the volume of high-quality reads greatly increases the likelihood of identifying potential pathogens within a sample. For example, a patient with a fulminant Shiga toxin–producing Escherichia coli (STEC) infection may present with 1×10⁵ colony-forming-units per milliliter

(CFU/mL) of stool, detectable by ELISA, PCR, and other conventional diagnostic methods. For metagenomic sequencing, however, the STEC is a minority population against an extremely complex background that includes upwards of 1 × 10¹¹ other microorganisms, food, human, and other ingested DNA. As a result, even with a dedicated HiSeq flow cell worth of data (6 days, $10,000), average coverage on the STEC genome could very well be below 0.1 or 0.2 and, even with the most thorough of bioinformatic analyses, it may not be possible to accurately distinguish the presence of STEC from other commensal Enterobacteriaceae without signal amplification. While stool metagenomics is arguably a worst-case scenario for sequencing and analysis, the background complexity of BioWatch filters is likely to be similarly high and the same signal-to-noise issues will almost certainly be a factor.

Without specific amplification, the limit of detection for metagenomic sequencing may be a critical issue for biosurveillance applications. In order to optimize sequencing yield in other contexts, metagenomic researchers have explored a number of different strategies to deplete the background of the sample and/or improve the recovery of specific microbes or pathogen types. Collectively, these approaches are often referred to as “clutter mitigation,” and they can be applied throughout the sample processing and sequencing workflow. Differential cell lysis, separation, filtering, concentration, pull-down, and amplification may all be used to optimize pathogen recovery during nucleic acid extraction. During postextraction cleanup and library construction, the use of specific nucleases, selective cDNA conversion, rRNA depletion, CpG-targeted degradation, preferential separation, and amplification may all be used alone or together to optimize sequencing capacity for specific pathogens and applications. It is important to underscore that, while these strategies may help to ensure the recovery of targeted pathogens, they also introduce significant differences in terms of bias, quality, and yield. Metagenomic sequencing results can vary quite markedly depending on the specific workflow that was used, even from identical samples. This variability has led to concerns over the reproducibility of metagenomic sequencing data, and the National Institite of Standards and Technology, FDA, and others are currently developing methodological standards for laboratory and bioinformatics analyses.

The bioinformatic analysis of shotgun metagenomic sequence data is also significantly more complicated than targeted sequencing, due to the much larger data volumes that are generated, and the absence of target specificity. Consequently, in order to compare and match sets

of reads from a query, large databases of full-genome pathogen, near-neighbor, commensal, and environmental sequences must first be established, curated, and maintained. This is a major undertaking and requires ongoing commitment, significant computational resources, and dedicated bioinformatic expertise. Large-scale queries against open-ended sequence databases are computationally demanding, which necessitates additional investments in high-performance computing and specialized IT support. Access to cloud-based metagenomic analysis software (e.g., CosmosID, OneCodex, IDbyDNA) and hardware-accelerated tools (e.g., EdicoGenome) are helping to democratize resource-intensive sequence analysis and reduce bioinformatic and IT costs. Even so, distributed use of standardized NGS-based methods will require significant planning for data management, transit, and retention policies.

There are also important “gaps” among publicly available reference sequences that must be addressed. Many microbial species and strains are underrepresented or lacking in publicly available data. While there have been a number of directed sequencing efforts to expand reference databases, there are still important omissions. These persistent gaps in reference data impede the analysis of complex microbial samples, particularly for atypical or emerging pathogens. For applications such as BioWatch, where pathogens of concern make up a relatively well-defined set, bioinformatic and database challenges may be of lesser concern since the number of comparator sequences may be more easily constrained.

Relative to targeted amplicon sequencing, the overall cost and complexity of shotgun metagenomics are both significantly higher. Barring significant technological advances, reductions in cost, improvements to laboratory infrastructure and informatics, and concerted efforts to build the necessary databases and bioinformatics capacity that are needed for routine large-scale sequencing and analysis, shotgun metagenomics are unlikely to be feasible or cost effective for large-scale surveillance applications within the next 5 to 10 years. A number of critical challenges contribute to this forecast, including the need for improvements to sample enrichment and multiplexing; the development of inexpensive, high-output long-read sequencing with a low, unbiased error rate; standards and organism-specific frameworks for analysis, interpretation, validation, and reporting; continued expansion of reference databases; improvements to public health IT infrastructure and bioinformatics expertise; and the development of portable, reproducible bioinformatic workflows that can incorporate epidemiologic and contextual metadata for interpretation. Rapid advances that alleviate one or

more of these obstacles within the next 3 to 5 years may result in metagenomic detection strategies that meet the operational requirements of the BioWatch program.

The technological outlook for several of the most promising shotgun metagenomic sequencing options are discussed below.

3.3.3.1 Rapid, Ultrashort Illumina Sequencing (1×36 and 2×25 bp)

Although Illumina sequencing is currently the default standard for most shotgun metagenomic studies, most conventional approaches use 2×125 to 2×300 bp paired end sequencing on the MiSeq, NextSeq, or HiSeq (NovaSeq recently introduced). These surveys typically generate between 5Gbases and 1 tera base pair (Tb) of sequence data, incur significant cost, and require turnaround times of 48 hours or longer. One option that could be explored is the use of rapid, ultrashort sequencing on the MiSeq using either the 1×36 or 2×25 bp kits, which generate between 650 and 850 Mb of data in a 4- to 6-hour run. Paired with NexteraXT or a similar rapid protocol for library construction, this sequencer workflow could feasibly allow for rapid turnaround, with results within 9 or 10 hours of sample receipt. While longer Illumina reads are ideal for metagenomics, several studies have demonstrated success in applying ultrashort 2×25 and 1×36 bp sequencing for unbiased pathogen identification in low-complexity samples at both the species and strain levels.

Even with the fastest of sequencing modes, the feasibility of Illumina SBS sequencing for BioWatch is fundamentally unclear. A 9-to 10-hour turnaround time is already in excess of current rtPCR-based BioWatch performance targets. Significant, too, are the instrumentation and reagent costs; the current list price on 1×36 and 2×25 bp sequencing kits remains close to Illumina’s average of $1,000 to $1,500 per kit. Given the limited output of these kits (850 Mb, ~34 million reads), the options for multiplexing shotgun sequencing samples would remain extremely limited. As such, it is likely that each individual sample would need to be run individually, with a cost in excess of $1,000, and fully occupying a sequencer for more than 6 hours. These limitations are likely untenable for the BioWatch program, unless this sequencing strategy was somehow paired with triggered sample collection, reserved for priority samples, or applied instead to targeted metagenomic sequencing, which could be more easily multiplexed.

3.3.3.2 Oxford Nanopore Sequencing

Recent advances by Oxford Nanopore have made the MinION Mark1b an increasingly attractive and feasible platform for rapid metagenomic sequencing. As discussed earlier in this section, the MinION has important advantages over other NGS instrumentation in terms of size, capital costs ($1,000), and deployability. It also presents some critical advantages in terms of sequencing performance, namely, sequencing speed, read length, simple sample prep and library construction, data output up to 20 Gb, and “read until” continuous data streaming that begins nearly instantaneously upon the start of sequencing. As such, a BioWatch laboratory (or mobile laboratory), equipped with one or more low-cost MinION devices, could theoretically (1) run nucleic acid extraction, (2) run targeted sequencing or sample amplification (as necessary), (3) run a rapid library prep and barcoding, and (4) initiate sequencing and begin analyzing streamed data, all within 2 to 3 hours of sample receipt. The streaming model for data acquisition is a critical advantage of this platform, particularly for time-sensitive applications. As sequence data stream off the device, preliminary results may be analyzed and returned on an ongoing basis, allowing the operator to rapidly flag samples or hits of interest, initiate an appropriate response, and determine whether additional data collection is warranted. Several studies have demonstrated real-time metagenomic analysis on the MinION platform that leverages this data streaming, notably MetaPore by Greninger et al. (2015).

Although the MinION hardware and consumables are consistently improving, current error rates remain high at 5 to 15 percent, with specific bias toward homopolymer errors. While these errors may also be addressed in software by data quality monitoring and error correction, they are more likely to cause bioinformatic issues in metagenomic analysis than in high-coverage microbial genomic sequencing. Until recently, MinION sequencing also required network connectivity and access to Oxford Nanopore’s Metrichor (www.metrichor.com) cloud service, since raw sequence data were uploaded to the cloud for signal processing and base calling. This dependency on cloud services for basic sequencing functions presented important technical challenges for field deployments with limited connectivity, and Metrichor’s lack of FedRAMP approval limited its use by state and federal laboratories. Recently, updated analysis software was released, allowing users to pair the MinION with a powerful workstation for offline base calling. The company is also said to be working on a portable hardware acceleration module that will enable local base calling and signal analysis using field-programmable gate arrays (GenomeWeb, n.d.). Oxford Nanopore

has also begun to ship higher-scale instruments, such as the GridION X5 and PromethION, through early access programs. While real-world performance characteristics of these high-throughput platforms are limited, both use the same flow-cell technology as the MinION, scaled to 5 and 48 flow cells, respectively. Both are likely to share similar flow-cell performance and error models, albeit with additional challenges due to high-bandwidth data streaming and acquisition.

Competing nanopore-based sequencing technologies also remain under active development by several companies, including Roche (Genia). Although it seems likely that these platforms will have similar capabilities and form factors and compete in a similar technology space, development has been lengthy, and it seems unlikely that there will be viable commercial competition to ONT within the next 12 to 18 months.

3.4 Mass Spectrometry

Although mass spectrometry has become an increasingly important tool for diagnostic microbiology over the course of the past decade, nonresearch applications have been primarily limited to matrix-assisted laser desorption ionization time of flight (MALDI-TOF) microbial identification and characterization. MALDI-TOF allows for rapid, inexpensive, and definitive identification of microorganisms from culture picks, and has become an important front-line, FDA-approved method for reference identification of bacteria and fungi, with two approved platforms (Bruker BioTyper and bioMerieux Vitek MS) currently on the market. Several recent studies have evaluated MALDI-TOF for high-consequence pathogens and related biomarkers, including the development of expanded databases for identification and characterization (Tracz et al., 2017). For microbial identification, these instruments are typically tuned to a narrow charge to mass (m/z) range to target specific high-abundance proteins or biomolecules (ribosomal proteins, short chain fatty acids) that provide diagnostically useful spectra. Importantly, MALDI-TOF typically cannot distinguish microorganisms beyond the level of genus or species. This limited resolution may be inconsequential for the purposes of BioWatch surveillance, where accurate high-level identification may suffice. While several studies have examined the use of MALDI-TOF for the direct characterization of pathogens from environmental and clinical samples, it remains unclear whether this approach could reliably detect low-abundance pathogens from complex filter backgrounds.

Electrospray ionization PCR (ESI-PCR) is another mass spectrometry (MS) technology that has been widely used for identification and genotypic characterization of biothreat agents, health care–associated pathogens, and other microbial diagnostic tasks. ESI-PCR was notably implemented in Abbott’s PlexID platform, incorporating multiplexed PCR with ESI spectral analysis to

determine the presence and quantitative composition (typically nucleotide counts) of the resulting amplicons. Performance of ESI-PCR is generally linked to the initial multiplex PCR, and PlexID performance suggested sensitivity of 98.5 percent and specificities of 73.6 to 97.6 percent (Doggett et al., 2016; Sampath et al., 2012). Like many early mass spectrometry systems, the PlexID was costly and difficult to maintain and is no longer commercially available, although other compact instrumentation has been introduced to take its place. A similar concept, MassTag PCR, was developed by researchers at Columbia University using multiplexed primer sets of up to 64 targets with photocleavable mass tags that could be recovered and interrogated using atmospheric pressure chemical ionization mass spectrometry (Palacios et al., 2006). Unlike ESI-PCR, this approach did not provide sequence composition information but was plagued by a number of the same challenges, and was ultimately discontinued from use.

Single particle aerosol mass spectrometry (SPAMS) has also been discussed in the context of BioWatch surveillance, and has been successfully demonstrated for indoor bio-aerosol surveillance during a 7-week demonstration in an airport, with no issues or false positives (Hook-Barnard et al., 2013). The instrument accelerates small molecules (<350 Daltons) from an aerosol sample through a continuous-wave laser detector to first determine particle size, which can be important in determining the nature of a bioaerosol threat. The particles are then ionized using a pulsed laser, generating both positive and negative ions, and dual-polarity TOF is used to generate an m/z spectrum. The system is capable of analyzing a range of biomolecules and is able to reliably resolve and quantify clinically important bacterial species based on spectral differences. Like most MS instrumentation, SPAMS systems are costly and require dedicated expertise for configuration and troubleshooting; however, incremental refinements to instrument design have simplified operation and maintenance to an important degree (SPAMS Overview, n.d.).

3.5 Integrated and Semiconductor Biosensors

In concept, arrays of integrated biosensors are an appealing detection strategy for BioWatch, since they could be designed to allow sensitive and continuous remote monitoring, give rapid unambiguous signal when triggered, could be produced at low cost and at scale, and perhaps, most importantly, could be small enough to be embedded directly in future generations of air sampling devices. Although biosensors are increasingly used for chemical, biological, radiological, and nuclear first response, point-of-care diagnostic testing, food safety, and environmental monitoring, most of the current and emerging generations are multichannel lab-on-a-chip devices that implement microfluidic ELISA or DNA amplification (Fronczek and Yoon, 2015; Hindson et al., 2005). For routine continuous monitoring of environmental samples, inexpensive semiconductor-based designs may be preferable to allow for further miniaturization, parallelization, greater reliability, and lower costs. A number of different research engineering groups are

developing semiconductor-based pathogen biodetector chips. While most of these applications remain in early development stages, recent publications have described the real-time detection of bacteria and bacterial protein toxins using inexpensive CMOS chips (Jeon et al., 2014; Nikkhoo et al., 2016), single-stage amplification and detection of pathogen DNA using optical detectors and liquid crystal light polarization (Khan et al., 2016), or the specific detection of bacterial pathogens using graphene or carbon nanotube–based electrochemical biosensors (Yoo et al., 2017), among others.

It seems unlikely that most solid-state biodetectors will emerge from prototype within the next 3 to 5 years and reach a level of technological maturity where they may be actively deployed for autonomous or semiautonomous environmental monitoring. In the meantime, academic and commercial groups have been making significant progress in the development of low-cost, microfluidic, and paper-based circuitry, which may be useful for field deployments and rapid diagnostic testing.

4. REFERENCES

Hindson, B. J., Makarewicz, A. J., Setlur, U.S., Henderer, B.D., McBride, M.T., Dzenitis, J.M. 2005. APDS: The autonomous pathogen detection system. Biosensors & Bioelectronics 20:1925–1931.

Hook-Barnard, I., M. Posey Norris S. 2013. Potential Technologies for the BioWatch Program. Washington, DC: The National Academies Press.

Illumina. No date a. MiSeq Specifications | Key performance parameters. Available at https://www.illumina.com/systems/sequencingplatforms/miseq/specifications.html (accessed August 18, 2017).

Illumina. (No date). Press Release. Available at https://www.illumina.com/company/newscenter/press-releases/press-release-details.html?newsid=2128277 (accessed August 18, 2017).

IOM (Institute of Medicine). 2011. The BioWatch System. Washington, DC: The National Academies Press.

Jeon, J.-W., Kim, J.H., Lee, J.M., Lee, W.H., Lee, D.Y., Paek, S.H. 2014. Rapid immuno-analytical system physically integrated with lens-free CMOS image sensor for food-borne pathogens. Biosensors & Bioelectronics 52:384–390.

Khan, M., Khan, A.R., Shin J-H., and Park S.-Y. 2016. A liquid-crystal-based DNA biosensor for pathogen detection. Scientific Reports 6:22676.

Liu, J., Ochieng, C., Wiersma, S., Ströher, U., Towner, J.S., Whitmer, S., Nichol, S.T., Moore, C.C., Kersh, G.J., Kato, C., Sexton, C., Petersen, J., Massung, R., Hercik, C., Crump, J.A., Kibiki, G., Maro, A., Mujaga, B., Gratz, J., Jacob, S.T., Banura, P., Scheld, W.M., Juma, B., Onyango, C.O., Montgomery, J.M., Houpt, E., Fields, B. 2016. Development of a TaqMan array card for acute-febrile-illness outbreak investigation and surveillance of emerging pathogens, including Ebola virus. Journal of Clinical Microbiology 54:49–58.

Luminex Corporation. 2014. xMAP® Technology for Research & Applied Markets.

McLoughlin, K. S. 2011. Microarrays for pathogen detection and analysis. Briefings in Functional Genomics 10:342–353.

Nikkhoo, N., Cumby, N., Gulak P.G., and Maxwell, K.L. 2016. Rapid bacterial detection via an all-electronic CMOS biosensor. PLoS ONE 11:e0162438.

Palacios, G., Briese, T., Kapoor, V., Jabado, O., Liu, Z., Venter, M., Zhai, J., Renwick N., Grolla, A., Geisbert, T.W., Drosten,C., Towner, J., Ju, J., Paweska, J., Nichol, S.T., Swanepoel, R., Feldmann, H., Jahrling, P.B., and Lipkin, W.I. 2006. MassTag polymerase chain reaction for differential diagnosis of viral hemorrhagic fevers. Emerging Infectious Diseases 12:692–695.

Palacios, G., Quan, P-L., Jabado, O.J., Conlan, S., Hirschberg, D.L., Liu, Y., Zhai, J., Neil Renwick, J., Hui, J., Hegyi, H., Grolla, A., Strong, E.J., Towner, J.S., Geisbert, T.W., Jahrling, P.B., Büchen-Osmond, C., Ellerbrok, H., Sanchez-Seco, M.P., Lussier, Y., Formenty, P., Nichol, S.T., Feldmann, H., Briese, T. and Lipkin W.I. 2007. Panmicrobial oligonucleotide array for diagnosis of infectious diseases. Emerging Infectious Diseases 13:73–81.

Patel, J. C., Oberstaller, J., Xayavong, M., Narayanan, J., DeBarry, J.D., Srinivasamoorthy, G., Villegas, L., Escalante, A.A., DaSilva A., Peterson, D.S., Barnwell, J.W., Kissinger, J.C., Udhayakumar, V., Lucchi, N.W. 2013. Real-time loop-mediated isothermal amplification (RealAmp) for the species-specific identification of Plasmodium vivax. PLoS ONE 8:e54986.

Quanterix SR-Plex Quanterix. No date. Available at http://www.quanterix.com/productstechnology/instruments/srx (accessed August 23, 2017).

Quick, J., Ashton, P., Calus, S., Chatt, C., Gossain, S., Hawker, J., Nair, S., Neal, K., Nye, K., Peters, T., De Pinna, E., Robinson, E., Struthers, K., Webber M., Catto,

A., Dallman, T.J., Hawkey, P., Loman, N.J.2015. Rapid draft sequencing and real-time nanopore sequencing in a hospital outbreak of Salmonella. Genome Biology 16:114.

Rane, T. D., Chen, L., Zec, H.C., and Wang, T-H. 2015. Microfluidic continuous flow digital loop-mediated isothermal amplification (LAMP). Lab on a Chip 15:776–782.

Ribot, E. M., and Hise, K.B. 2016. Future challenges for tracking foodborne diseases: PulseNet, a 20-year-old US surveillance system for foodborne diseases, is expanding both globally and technologically. EMBO Reports 17:1499–1505.

Robison, K. 2016. Omics! Omics!: Illumina’s Unveils Firefly. Omics! Omics! (accessed March 19, 2018).

Sampath, R., Mulholland, N., Blyn, L.B., Massire, C., Whitehouse, C.A., Waybright, N., Harter, C., Bogan, J., Miranda, M.S., Smith, D., Baldwin, C., Wolcott, M., Norwood, D., Kreft, R., Frinder, M., Lovari, R., Yasuda, I., Matthews, H., Toleno, D., Housley, R., Duncan, D., Li, F., Warren, R., Eshoo, M.W., Hall, T.A., Hofstadler, S.A., Ecker, D.J. 2012. Comprehensive biothreat cluster identification by PCR/electrospray-ionization mass spectrometry. PLoS ONE 7:e36528.

Slezak, T., Kuczmarski, T., Ott, L., Torres, C., Medeiros, D., Smith, J., Truitt, B., Mulakken, N., Lam, M., Vitalis, E., Zemla, A., Zhou, C.E., Gardner, S. 2003. Comparative genomics tools applied to bioterrorism defence. Briefings in Bioinformatics 4:133–149.

SPAMS Overview: Single Particle Aerosol Mass Spectrometry. No date. Available at http://www.livermoreinstruments.com/spams-overview.html (accessed March 19, 2018).

Swaminathan, B., Barrett, T.J., Hunter,S.B., and Tauxe, R.V. 2001. PulseNet: The molecular subtyping network for foodborne bacterial disease surveillance, United States. Emerging Infectious Diseases 7:382–389.

Tessler, M., Nuemann, J.S., Afshinnekoo, E., Pineda, M., Hersch, R., Velho, L.F., Segovia, B. T., Lansac-Toha, F., Lemke, M., DeSalle, R., Mason, C.E., Brugler, M.R. 2017. Large-scale differences in microbial biodiversity discovery between 16S amplicon and shotgun sequencing. Scientific Reports 7:6589.

ThermoFisher. No date a. Ion AmpliSeq Panels. Available at https://www.thermofisher.com/us/en/home/life-science/sequencing/next-generation-sequencing/ion-torrent-next-generation-sequencing-workflow/ion-torrent-next-generation-sequencing-select-targets/ampliseq-target-selection.html (accessed March 19, 2018).

ThermoFisher. No date b. IonTorrent Specifications. Available at https://www.thermofisher.com/us/en/home/life-science/sequencing/next-generation-sequencing/ion-torrent-next-generation-sequencing-workflow/ion-torrent-next-generation-sequencing-run-sequence/ion-s5-ngs-targeted-sequencing/ion-s5-specifications.html (accessed March 19, 2018).

ThermoFisher. No date c. Real-Time PCR OpenArray. Available at https://www.thermofisher.com/us/en/home/life-science/pcr/real-time-pcr/real-time-openarray.html (accessed March 19, 2018).

Tracz, D. M., Tyler, A.D., Cunningham, I., Antonation, K.S., and Corbett, C.R. 2017. Custom database development and biomarker discovery methods for MALDI-TOF mass spectrometry-based identification of high-consequence bacterial pathogens. Journal of Microbiological Methods 134:54–57.

Woods, T. A., Mendez, H.M., Ortega, S., Shi, X., Marx, D., Bai, J., Moxley, R.A., Nagaraja, T.G., Graves, S.W., Deshpande, A. 2016. Development of 11-Plex MOL-PCR assay for the rapid screening of samples for Shiga toxin-producing Escherichia coli. Frontiers in Cellular and Infection Microbiology 6:92.

Wuyts, V., Roosens N. H. C., Bertrand, S., Marchal, K. and De Keersmaecker, S. C. J. 2016. Optimized MOL-PCR for characterization of microbial pathogens. Current Protocols in Cytometry 75:13.15.1–13.15.15.

Yoo, M.-S., Shin, M., Kim, Y., Jang, M., Choi, Y.E., Park S.J., Choi J., Lee, J., Park, C. 2017. Development of electrochemical biosensor for detection of pathogenic microorganism in Asian dust events. Chemosphere 175:269–274.