Finding The Bugs

Ideally, a surveillance system would provide early enough detection and identification that the population could be warned—the detect-to-warn model. Receiving adequate warning would allow the population to protect itself by avoiding exposure simply by going indoors or covering one’s mouth and nose until in a safe place. Detect-to-warn would require near-real-time detection, however. While this may be viewed as the gold standard, detect-to-warn technologies are largely in the development stage and tend to be quite specific. (For example, most can only detect one or at best a small set of pathogens. See the “Future Biosensor Technologies” text boxes at the end of the chapter for a further discussion of work being done in this field.)

The next best thing is to detect-to-treat. In this approach, the notification comes too late to alert all citizens, but soon enough that those who have become ill can be treated. For example, figure 6 indicates that, even with exposure to certain pathogens, there is a window of opportunity within which people can be treated and casualties minimized. This implies the need for a sensing system that can provide positive identification of the pathogen within a timely manner.

Figure 6 – The impact of response delay on the casualties suffered during a biological attack. Source: Dr. Tim Dasey, MIT Lincoln Laboratory. Received December 19, 2002

Additionally, a surveillance system should provide sufficient material for independent verification. Such verification is necessary as an additional check on the system, as well as confirmation that the measures taken by authorities are correct and adequate.
A surveillance system should also be able to detect-to-reassure. That is, it should continue to monitor during and after an incident to reassure citizens that the levels of the agent in question have subsided enough to discontinue any emergency procedures.

Finally, a system should be able to offer some predictive ability for areas downwind of the initial attack. Not only would this have the effect of providing some detect-to-warn for certain areas, it would provide additional data for officials planning the use and distribution of limited emergency resources.

There are numerous ways to monitor the environment for the presence of biological agents. The most common method currently used is sampling the air with monitors, collectors, or sensors. Generically, all of these systems are sometimes—though inaccurately—referred to as sensors. They are all essentially a form of electro-mechanical device—a box—that takes in air. This study will use the generic term sensor when referring to this class of devices. The discussion below looks at these devices and their strengths and weaknesses.

Monitors/Collectors/Sensors

Biological sensors can be categorized into three broad types: environmental monitors, sample collection devices, and rapidly deployable sensors. Environmental monitors operate continuously and draw in air samples that are then filtered and concentrated for analysis. They are relatively inexpensive to operate. However, they function as little more than “change detectors” by testing the environment for the presence of particles of predetermined sizes. They have the disadvantage of not being able to detect a broad range of pathogens. 10

Sample collection devices usually collect samples on filter paper, which must then be further analyzed. Their chief liability is the cost of operation. Reagents typically cost about $1 per test; some of the reagents have to be maintained under controlled storage conditions, as well. In addition, there is considerable labor involved. While the analysis can be highly specific, the tests are not especially good at identifying novel biological agents. 11

Rapidly deployable sensors, such as hand-held assays, are limited by the somewhat narrow band of pathogens that they can identify. Their advantage lies in their ability to be dropped on a target area, or easily carried by emergency personnel.12
There are common components to all of these devices. First, any sensing system must collect samples from the environment. As discussed above, the size of the items of interest—bacteria, toxins, and viruses—is of such minute scale that, in addition to collecting samples, the samples must be concentrated.

When collecting air samples, it is important to ensure that an adequate amount of air is screened for potential pathogens. Figure 7 uses a range of hypothetical cases to illustrate this point. For example, if the ID50 (Infectious Dose 50, i.e., the dose at which 50 percent of the exposed population will be infected) is 10 organisms, then the sensor will have to collect at least 10 organisms in every 100 liters of air that it samples. The significance of figure 7 is that the level of sensitivity required for a sensor is directly related to the pathogens in question. Thus, one size sampler will not necessarily fit all situations.

Figure 7 – Detector sensitivity requirements given that the human must see 10 organisms to retain 6. Source: Dr. James Valdes, USA SBC Command. Received February 2003.

Table 1 provides a review of the four most common collection technologies and their respective advantages and disadvantages. The important point to be made from table 1 is that no technology is a clear winner. They are either hampered by the inability to collect large samples, or by the costly requirements for labor or material.

Once a sample has been collected, it must then be assessed for the presence or absence of suspicious items. Usually, a “trigger” approach is used. A simple trigger will be activated if there is an increase in the background of particulate matter. If that threshold is reached, then a specific trigger will be activated to investigate the sample for more precise information.13 The specific trigger may be no more sophisticated than to determine if the suspect material is biological or non-biological. 14

Table 2 presents five of the most common triggering/detection technologies, with their advantages and disadvantages. As with the collection technologies, no one technology is clearly superior. Note that the disadvantages mostly center on issues involving additional labor, either to perform follow-on tests or to provide maintenance for the equipment in question.

The final step in any sensing system is the identification of the suspect material. Table 3 lists four of the most common technologies used. The first three—mass spectrometry, antibody based, and DNA-based—all have different variations that are used, but the advantages/disadvantages listed are reasonable summaries that capture the technologies in general. The most significant point of table 3 is the high degree of specificity of these technologies. Thus, sensors using these technologies will be limited to searching for a fairly specific list of possible agents. Novel agents could easily go undetected.

In looking across tables 1, 2, and 3, it is clear that no single collection, detection, or identification technology is superior. Nor is any combination of technologies superior. Moreover, the most reliable technologies are also the slowest—keeping them limited to detect-to-treat at best.

On the positive side, the current state of the art can provide systems that will identify the most likely agents and can be used for specific scenarios. Also, in assessing the current technologies it appears that no theoretical barriers restrict future developments. Thus, the gold standard of detect-to-warn eventually may be achievable.

In an ideal world, a biosurveillance system would embody the following characteristics: rapid; sensitive; specific; easy to use; automated; able to use diverse sample types (blood, serum, urine, food, air, water, soil); compact and portable; self-powered; low-cost; rugged. As tables 1, 2, and 3 reveal, this is not an ideal world. But, as stated above, there are no theoretical barriers to eventually achieving a system to meet these criteria.

How Many Boxes Do We Need?

Assume an ideal world and the ability to build a box that meets all of our criteria. Could we afford to build, operate and maintain all that we would need? Before answering that question, let us consider the origin of our ideas on the use of these devices.

Most of our thinking on how to deploy a biosurveillance system for the protection of civilian populations has come from the military use of chemical detectors on the battlefield. The problems are not the same, for several reasons. First, chemical agents act within minutes, if not seconds. Thus, chemical sensors must operate as detect-to-warn, if they are to have any utility. 15 Second, chemical weapons often are accompanied by a distinctly visible cloud and/or odor. Those in the downwind area may actually be able to see the approaching danger, even without benefit of sensors.16 Third, a chemical cloud is effective only in high concentrations and where terrain and meteorological conditions are favorable. Usually, the cloud is not a threat if it is much more than a few meters off the ground.17 Finally, it is anticipated that the opposing force on the battlefield may use chemical weapons. Typically, the opposing force winds up—at least in part—arrayed in a linear fashion to the front of the friendly forces.18 There is little question as to the direction from which an attack will come and where to place the sensors.

Figure 8 – On the battlefield, the likely direction of a chemical attack is easily assumed, as opposing forces typically array themselves in linear fashion. Source: http://www.conmon.com/gallery/albums/2003_04/030403_iraq_tanks.jpg.

Compare this with biological agents. Biological agents, as seen in figure 9, can take days to produce symptoms.

Figure 9 – The characteristics of a generic infectious disease (bioterrorism agent). For example, smallpox generally does not become symptomatic until after an incubation period that averages 12-14 days.

Thus, a surveillance system can function as detect-to-treat and still be useful. Biological agents would be odorless and invisible. Moreover, if the agents were contagious, then a secondary wave could be established by the presence of infected individuals. (For example, on average, in an unprotected population, every person with smallpox will infect another four to seven others.19 ) There need not be high concentrations of agent to create casualties. A biological attack could be effective in any terrain setting. For example, using a contagious agent, an attack could be successful if it were conducted by the simple introduction of infected individuals. Moreover, the geometry of an urban setting is distinctly different from that encountered in a non-urban battlefield. Obstacles abound and microclimate conditions prevail in the canyons formed by tall buildings. (This would argue against the use of line attacks with large clouds of agents.) Finally, the direction of attack by the “opposing force” would be unknown, causing a city to consider ringing itself with a surveillance system—like an army surrounded on all sides.

Figure 10 – The canyons of urban areas create microclimates and geometries distinctly different from traditional battlefields. Source: http://digilander.libero.it/travelphoto/New%20York/skyscrapers.htm

What about the number of boxes required? A recent JASON report concluded that it is not realistic to undertake a nationwide, blanket deployment of biosensors. 20 With current technologies, JASON estimated that approximately one sensor per square kilometer would be required to protect the U.S. population—at an annual cost of $10-15 billion. 21

In one small study designed to assess the number of sensors needed to protect a moderate-size military base, the numbers suggested that as many as 1,000 sensors might be required to detect an anthrax attack disseminated by a cloud over an area of about 180 square kilometers—roughly the size of Washington, DC. 22

The real message from this analysis is that one size does not fit all. Moreover, it reconfirms the earlier assessment that there is no biosurveillance technology—nor group of technologies—currently available that provides a clearly superior approach. There is no denying that biosensors are and will continue to be an important part of biosecurity. But, as seen from the brief discussion of the number of sensors required and their costs, a strategy based solely on the use of sensors could quickly become cost prohibitive; especially for a major metropolitan area. The near-term role of biosensors will most likely be as detect-to-treat devices deployed at specific high-value sites.

WHAT DOES IT MEAN IF THE ALARM GOES OFF?… OR IF IT DOESN’T GO OFF! What is the appropriate response if a sensor tells us of the presence of a bioagent? Can we be certain it is a real event, or is it a false alarm? Is it a true positive, or a false positive? And, if the alarm does not go off, is that a true negative and we don’t need to worry, or is it a false negative and the alarm should have sounded, but the sensor missed the event? Significant consequences arise from responding to false positives —loss of public confidence in the system, financial and health costs associated with evacuations, quarantines, and with too many “crying wolf events” potential failures to respond to an actual attack, etc. Similarly, false negatives may lead to significant losses of life and of public confidence. Detecting the presence of biological agents with sensors can be viewed in a technical sense as a problem in signal detection theory. There is a signal—the presence of certain biological information. There are various sources of “noise” in the background—any number of things that can confuse, mask or distort our ability to detect the signal for which we are searching. For example, in some systems, pollen grains can become “biological noise” that disrupt our ability to detect the signal for which we are searching. The more sensitive a sensor—i.e., the more discriminating it is with respect to finding the signal amongst the background noise—the greater the probability of false alarms. This relationship is characterized by a type of curve known as a receiver operating characteristics curve (ROC curve). Originally developed for assessing radar, ROC curves have applicability to any number of such tasks. Figure 11 shows a generalized ROC curve. Figure 11 - Generalized reciever operating characteristics (ROC) curve Designers face an inherent trade-off when designing sensor systems: the more sensitive they make them, the greater the probability that they will misinterpret the signal-to-noise ratio and give you a false positive—with all of the attendant consequences suggested above. Reducing the false positive rate, however, reduces the ability of the sensor to discriminate the signal from background noise; thus giving you the potential for false negatives. FUTURE BIOSENSOR TECHNOLOGIES Cellular Analysis and Notification of Antigen Risks and Yields (CANARY) – Information taken from Todd H. Rider, et al., “A B Cell-Based Sensor for Rapid Identification of Pathogens,” Science 2003, 301: 213-215. Researchers at MIT’s Lincoln Laboratories began work in 1997 on the CANARY (Cellular Analysis and Notification of Antigen Risks and Yields) project. It involves the use of B-lymphocytes, a type of white blood cell that our bodies use against bacterial and viral invaders. These cells are already designed by nature to search for any bacteria and viruses very rapidly. In the laboratory, they are given the ability to glow in the presence of certain contaminants by adding a luminescence gene from jellyfish. The actual detectors are pathogen specific antibodies within the B cells that trigger a burst of calcium when an agent is detected. Within seconds, the calcium activates a bioluminescent protein that causes the whole cell to glow. A device termed a luminometer is used to analyze the light-emitting cell. Within the luminometer the cells are kept alive in test tubes and their response is displayed on a computer readout. The system has already been tested successfully against a list of biological agents, including anthrax, smallpox, plague, tularemia and encephalitis. Figure 12 - B cells from the human immune system. Source: http://www3.kmu.ac.jplanat/edu/histology/general/blood/bcell.jpg The CANARY bio-agent forensic analysis of body fluids would be useful for monitoring air, water, and contaminated surfaces as well as body fluids. It is expected to detect more rapidly and with greater sensitivity than conventional sensors that are based on chemical reactions. These chemical reactions can take several hours to complete and the sensors can require several thousands of particles for detection. By comparison, CANARY has been able detect as few as 50 colony-forming units of the plague bacterium in less than three minutes. Furthermore, unlike many existing sensors, CANARY would not require advanced training for operation. Consequently, MIT researchers foresee a variety of applications for the system. For instance, in medical diagnostics, it could be used to immediately separate those patients suffering from the symptoms of a cold from those with SARS. In the environment, it could be used to test water and air quality for pathogens both inside and outdoors. In the event of an emergency, suspicious substances on the street, subways, or airports could be tested quickly. FUTURE BIOSENSOR TECHNOLOGIES Light Detection And Ranging (LIDAR) – Information taken from Petter Weibring, et al., “Versatile mobile lidar system for environmental monitoring,” Applied Optics 2003, 42: 3583-3594. Light detection and ranging (LIDAR) is a tool for cloud detection and recognition based on the same physical principles as radar, except instead of bouncing longer wavelength radio waves off a target, higher energy light waves are used. An acronym for "Light Detection And Ranging," LIDAR is occasionally attributed to "Laser Identification and Ranging" by those who want to emphasize the recognition feature. Using lasers that generate light waves in the infrared, the ultraviolet and the visible portion of the electromagnetic spectrum, the multiple energy wavelengths of LIDAR furnish more detailed information, including three-dimensional imaging. Limitations on detection distance and resolution are due to the collection and processing portions of the detector. The more specific the level of data desired, the closer the instruments must be located to the cloud. Under controlled conditions, detection of aerosolized clouds at long distances has been achieved. The drawbacks are primarily financial and the current limited distance capability. LIDAR instruments are not cheap - costing about $4,000 for a simple LIDAR used for speed monitoring. The U.S. Army's Long Range Biological Standoff Detection System (LR-BSDS) uses LIDAR-based technology on an unmodified UH-60 Blackhawk helicopter to detect aerosol clouds from long distances. The Short Range Biological Standoff Detection System (SR-BSDS) combines infrared LIDAR with ultraviolet light reflectance (UV). The latter provides enhanced discrimination capabilities. Biological agents can be distinguished from non-biological material based on the excitation of the intracellular fluorescent compounds. The most commonly targeted compounds are the amino acid tryptophan, the coenzyme nicotinamide adenine dinucleotide (NADH), the cellular energy storage molecule adenosine triphosphate (ATP) and the vitamin riboflavin. Identification of these compounds verifies that the sample is biological in origin. Possible false positives include pollen, molds, organic excreta and certain agricultural fertilizers based on decaying organic matter. Figure 13- US Army's Long Range Biological Detection System mounted on Blackhawk helicopter. Source: http://www.lanl.gov/orgs/dod/images/Heli-Lidar.jpeg FUTURE BIOSENSOR TECHNOLOGIES Better, Cheaper, Faster – Information taken from Olga Kharif, “A Sharper Nose for Danger,” Business Week Online, May 25, 2003. A variety of sensors are beginning to enter the marketplace with the potential to reduce significantly the cost of constant surveillance. The present family of sensors costs about $2 million per city, per year, according to officials at the Department of Homeland Security. Most of the current costs are related to labor associated with collecting the filter papers from air samplers and testing them in a laboratory for the presence of pathogens. Frances Ligler, a senior scientist at the Naval Research Laboratory, has developed a shoebox-size detector that eventually could screen simultaneously for 12 times as many pathogens as today’s devices. Such so-called microarrays operate at the intersection of physics and biology. Using laser light to illuminate the samples, the device can identify specific life forms—bacteria and viruses—and toxic proteins—toxins—by interpreting the resulting fluorescence patterns. Such sensors have several advantages, including lower false-positive/false-negative rates. In addition, bacterial samples are not destroyed in the process, leaving them available for further testing, e.g., to determine whether or not they are resistant to antibiotics. Figure 14 - The most recent version of the NRL array biosensor with integrated optics and fluidics components. Six-reservoir modules for holding samples and fluorescent tracer reagents are shown separately below. Source: Dr. Frances S. Ligler and Chris R. Taitt, "The Array Biosensors." Received June 1, 2004. Several other approaches are also moving towards commercialization. One Boston-area company is developing a reusable tape to which suspect molecules will bind. The molecules can be made to fluoresce, using low-level light. The tape can be rewound and used again. As biosensor development continues, and their purchase and operating costs decline, their use will expand into other sectors. Food safety and drug manufacture are two areas in particular that will benefit from the ability to rapidly screen for pathogens and toxins.

10. National Institute of Justice, “An Introduction to Biological Agent Detection Equipment for Emergency First Responders,” NIJ Guide 101-00, December 2001, 23-25.
11. Ibid., 23.
12. Ibid., 26.
13. Brian M. Sullivan, “Bioterrorism Detection: The Smoke Alarm and the Canary,” Technology Review Journal, Vol. 11, No. 1, Spring/Summer 2003, pp. 135-141. See <http://www.ms.northropgrumman.com/PDFs/TRJ2003/03ssSullivan.pdf>, accessed March 2004.
14 Note that in this case, the presence of a toxin would likely go unnoticed.
15. David Ruppe, “Rapid, Accurate Biological Attack Detection Capability Is Years Away, Experts Say,” Global Security Newswire, October 22, 2003. See <http://www.nti.org/d_newswire/issues/2003/10/22/cc3cd030-62a3-46bf-bfe2-ee3e9c49a2c2.html>.
16. “Fed’s Sniffing Devices Effective?” Associated Press, Philadelphia, July 17, 2003. See <http://www.cbsnews.com/stories/2003/07/17/tech/main563765.shtml>, accessed March 2004.
17. Army Online Training Course, FM 3-6. See <http://155.217.58.58/cgi-bin/atdl.dll/fm/3-6/CH1.PDF>, accessed March 2004.
18. CAPT Sean E. Hynes, USMC, and Neil C. Rowe, “Multi-Agent Simulation for Assessing Massive Sensor Deployment,” MOVES Institute, U.S. Navy Postgraduate School. See <http://www.cs.nps.navy.mil/people/faculty/rowe/oldstudents/hynespap.htm#_ftn1>, accessed March 2004.
19. Martin I. Meltzer, et al., “Modeling Potential Responses to Smallpox as a Bioterrorist Weapon,” Center for Disease Control and Prevention, Atlanta, GA, December 2001. See <http://www.cdc.gov/ncidod/EID/vol7no6/meltzer_appendix1.htm>, accessed March 2004.
20. JASON, “Biodetection Architectures,” The MITRE Corporation, McLean, VA, February 2003. See <http://www.fas.org/irp/agency/dod/jason/biodet.pdf>, accessed December 2003. JASON is a group of distinguished defense consultants.
21. Ibid., 5.
22. Timothy Dasey, PhD, MIT Lincoln Laboratory, personal communication.