detection of biological agents

• PRIMMERMANDetection of Biological Agents

VOLUME 12, NUMBER 1, 2000 LINCOLN LABORATORY JOURNAL 3

B have seldom been usedthroughout history [1–4]. The Biological andToxin Weapons Convention of 1972 out-

lawed the possession and use of biological weapons;this convention has been ratified by all but a handfulof major nations [5], and no nation has an overt, de-clared biological-weapons capability. Despite thesefacts, it appears that the threat of biological attack isincreasing. Indeed, biological weapons may becomethe weapons of choice for rogue states and terroristorganizations.

Several factors contribute to this trend. First, thevery horror associated with biological weapons servesto make them attractive to modern terrorist groups.Second, proliferation of biotechnology equipmentand expertise has made it relatively easy to producebioagents. Equipment needed to produce these bio-agents can be found in pharmaceutical industries,food-processing plants, and even microbreweries.Such equipment is widely available on the open mar-ket. Likewise, there is a large, worldwide workforcetrained in the basic techniques necessary to make bio-logical weapons. The relative ease of making biologi-cal weapons is illustrated by the Iraqi program. (Seethe sidebar entitled “Iraq’s Biological-Weapons Pro-gram.”) Within only a few years Iraq clandestinelygenerated a substantial arsenal of biological weapons.

Third, small quantities of bioagents can cause hugenumbers of casualties. For example, a lethal dose ofanthrax spores is less than a microgram. Thus, less

than 1 kg of anthrax would, if optimally distributed,be enough to give a lethal dose to every man, woman,and child in the United States. Even if not optimallydistributed, relatively small amounts of bioagent cancause large numbers of casualties.

The World Health Organization estimated that at-tacking a large city with 50 kg of anthrax sporeswould produce 95,000 deaths and an additional125,000 sicknesses [6]. The U.S. Congressional Of-fice of Technology Assessment estimated that an at-tack on Washington, D.C., with 100 kg of anthraxspores would produce one to three million deaths [7].The estimated casualties vary over a wide range be-cause, fortunately, we have no empirical data on suchan attack. Whichever estimate is accepted, casualtiesin a biological attack could be impressively large.

Finally, the combination of the relative ease of pro-duction of biological weapons and the potentiallylarge number of casualties they can cause makes thebiological weapon the “poor-man’s” weapon of massdestruction. A relatively poor, technologically unso-phisticated nation may well believe that developingbiological weapons offers an advantage over an other-wise militarily superior adversary. Likewise, a terroristorganization can believe that the threat or actual useof biological weapons will gain them what more con-ventional acts of violence have not.

Components of Biological-Warfare Defense

Successfully defending against the threat of biological

Detection of Biological AgentsCharles A. Primmerman

■ Biological weapons pose a real and potentially immediate threat. They arerelatively cheap to manufacture and employ, and they have tremendous potentialimpact as terror weapons. These features make biological weapons attractive torogue states and terrorist organizations. In this article we briefly describe thethreat of biological weapons. We then describe Lincoln Laboratory work indeveloping advanced sensors to help combat the threat of biological weapons.These sensors include an early-warning sensor that can sense small quantities ofbiological particles in the air and issue an alarm in less than one minute and abioelectronic identifying sensor that can potentially identify a biological agentfrom a single sensed particle.


4 THE LINCOLN LABORATORY JOURNAL VOLUME 12, NUMBER 1, 2000

’ -weapons pro-gram provides a case study in thethreat posed by a rogue nation. Itshows both how readily biologicalweapons can be developed andhow easily their development canbe hidden.

At the time of the Gulf War in1991 it was widely suspected thatIraq had an offensive biological-warfare capability. Following thewar, in response to U.N. SecurityCouncil resolutions requiringIraq to disclose all weapons ofmass destruction, Iraq maintainedthat it had only a tiny, defensivebiological warfare effort of onlyabout ten persons.

The U.N. Special Commission(UNSCOM) charged with ob-taining information on Iraq’sweapons of mass destruction andmonitoring their destruction hadhints and suspicions about a sig-nificant Iraqi offensive program.In four years of efforts, however,including intrusive inspections ofapproximately eighty potentialdual-use biotechnology facilitiesin Iraq, UNSCOM was not ableto obtain definitive proof of anIraqi offensive biological-weap-ons program. Only after the Au-gust 1995 defection of GeneralHussein Kamel Hassan, who had(among other responsibilities)been in charge of the Iraqi biologi-cal-weapons program, was Iraqforced to reveal the extent of itsprogram. (In revealing the pro-

gram, Iraq claimed “Hassan had,unbeknownst to the senior levelof the Iraqi leadership, hidden in-formation on the prohibitedprogrammes” [1].)

In late 1995 Iraq admitted tohaving conducted a major offen-sive biological-weapons program.This program, spread across manydifferent sites, involved researchinto various bioagents, produc-tion of large quantities of certainbioagents, field testing of bioag-ents, and actual weaponization.Iraq claimed that its program be-gan in 1985 and that productionand weaponization work acceler-ated dramatically in August 1990,coinciding with the invasion ofKuwait. Given the Iraqi claims,UNSCOM concluded that in fiveyears, “the achievements of Iraq’sbiological weapons program wereremarkable” [1].

Iraq admitted to producing8500 liters of anthrax and 19,000liters of concentrated botulinumtoxin. It also claimed to have made7200 liters of aflatoxin, which re-mains intriguing, for althoughaflatoxin is a known carcinogen, itdoes not appear to be an appropri-ate bioagent. These and all Iraqinumbers on biological weaponsshould be regarded as suspect;subsequent UNSCOM investiga-tions concluded that the Iraqimethodologies for calculatingmost numbers were seriouslyflawed and that verification is im-

possible because of destroyeddocuments [2].

Iraq also admitted to workingon ricin (a toxin produced fromcastor beans), clostridium per-fringins (gas gangrene), and threeviruses: hemorrhagic conjunctivi-tis, a rotavirus, and camel pox. Italso developed an anti-crop bioag-ent—wheat smut, which was pro-duced in large quantities and usedin field tests.

Iraq claimed to have made bio-agents into weapons on variousplatforms. One hundred R400(400 lb) bombs were filled withbotulinum toxin, fifty with an-thrax spores, and sixteen with afla-toxin. Twenty-five special AlHussein missile warheads werebuilt after August 1990; thirteenwere filled with botulinum toxin,ten with anthrax, and two withaflatoxin. These weapons were de-ployed at four sites during theGulf War, and authority to usethe weapons was delegated to fieldcommanders.

Iraq performed field trials ofbioagents in 155-mm-artilleryshells and in 122-mm rockets. Italso tested a delivery techniqueusing a 2000-liter tank and asprayer mounted on a remotelypiloted aircraft [3].

Iraq claims that in 1991, fol-lowing the war, it destroyed allfilled biological munitions and allstocks of bioagents. This unilat-eral destruction was, itself, in vio-

I R A Q ’ S B I O L O G I C A L - W E A P O N S P R O G R A M


VOLUME 12, NUMBER 1, 2000 THE LINCOLN LABORATORY JOURNAL 5

lation of U.N. Security Councilresolutions, and the claimed de-struction remains unverified. Af-ter Iraq admitted its offensive bio-logical-warfare program, certainfacilities, including the major pro-duction plant at Al Hakam, weredestroyed under UNSCOMsupervision.

Before being expelled fromIraq, UNSCOM concluded thatIraq had continued to concealfacts regarding its biological-war-fare program [4]. Given this con-cealment, which is even easier inthe absence of UNSCOM inspec-tors, it is reasonable to presumethat components of the program

still exist, if not actual weapons,then certainly seed stocks of bio-agents and manufacturing reci-pes. Because much of the requiredequipment still exists as dual-usebiotechnology equipment and be-cause the trained workforce stillexists, it is likely that Iraq could re-constitute a biological-warfare ca-pability in a relatively short time.

References1. United Nations Security Council,

“Report of the Secretary General onthe Status of the Implementation ofthe Special Commission’s Plan for theOngoing Monitoring and Verificationof Iraq’s Compliance with RelevantParts of Section of Security Council

Resolution 687 (1991),” S/1995/864(11 Oct. 1995), <http://www.un.org/Depts/unscom/sres95-864.htm>.

2. United Nations Security Council,“Report of the Executive Chairmanon the Activities of the Special Com-mission Established by the Secretary-General Pursuant to Paragraph 9 (b)(i) of Resolution 687 (1991),”S/1998/332 (16 Apr. 1998), <http://www.un.org/Depts/unscom/sres98332.htm>.

3. R.A. Zilinskas, “Iraq’s BiologicalWeapons: The Past as Future?” J. Am.Med. A. 278 (5), pp. 418–424.

4. United Nations Security Council,“Report of the Executive Chairmanon the Activities of the Special Com-mission Established by the Secretary-General Pursuant to Paragraph 9 (b)(i) of Security Council Resolution 687(1991), S/1998/920, (6 Oct. 1998),<http://www.un.org/Depts/unscom/sres98-920.htm>.

weapons requires vigorous efforts in many areas. Fig-ure 1 shows the five principal components of biologi-cal defense: (1) deterrence and destruction—stoppingthe enemy before he has used biological weapons, (2)protection—masks and air filters, (3) decontamina-tion after an attack, (4) medical countermeasures—both immunization before an attack and treatmentafter an attack, and (5) detection and warning.

Detection is the keystone of biological defense.Without detection of an attack there will be no warn-ing to don protective masks. Without detection therewill be no knowledge of what to decontaminate orwhen it is fully decontaminated. Without detectionprompt, specific medical treatment will not bepossible.

Before an attack a slightly different kind of detec-tion is essential. Detecting what a potential enemy ismanufacturing enables deterrent activity. It also en-ables appropriate immunization programs.

Lincoln Laboratory efforts in biological-agent de-fense have concentrated on detection and warning.We have been developing sensors and sensor systemsto provide highly sensitive real-time detection, dis-crimination, and identification of bioagents.

A Biological Attack

To detect and defend against a biological attack, wemust understand its general nature. A biological at-tack can take on many forms depending on whatbioagent is used, how it is dispersed, and what isattacked.

Biological weapons can be used to attack humanbeings, livestock, or crops. We have focused our de-tection efforts on defending humans against biologi-cal attacks. We note, however, that attacks on live-stock and crops could have significant strategic value,and thus should not be ignored. Indeed, Iraq madeboth anti-livestock and anti-crop bioagents intoweapons.

The potential scale of a biological attack on humanbeings spans a wide range. A biological weapon canbe used to kill a single individual, or it can be dis-persed over a major city to kill millions of people. Inbetween these extremes, one can imagine biologicalweapons being used to attack concentrations of per-sons in various military and civilian settings: an airbase, a large office building, an aircraft carrier, asports stadium, a military staging area, a subway. In


6 LINCOLN LABORATORY JOURNAL VOLUME 12, NUMBER 1, 2000

our detection efforts we will assume that the attack isagainst some substantial number of persons.

Bioagents can be disseminated in at least three ma-jor ways: by vector (i.e., insects), by contaminatingwater or food supplies, and by distributing them asaerosol particles in the air. Well-known virulent dis-eases are normally transmitted by insects (for in-stance, fleas transmit plague and mosquitoes transmityellow fever), and the use of insects in biological war-fare has apparently been tested [8]. Nevertheless, thedifficulty of dealing with large numbers of insects andtheir unpredictability once released make significantbiological attacks with insects seem unlikely.

Disease and death from contaminated water andfood are also well known in nature. For example, ty-phoid fever is transmitted by contaminated water,and botulism commonly comes from eating improp-erly canned food. Using biological weapons to con-taminate water or food may be an effective way tomurder one person or a few persons, but it is unlikelyto be effective in infecting large numbers of persons.Consider, for instance, the problems of putting bioag-ents in a municipal water supply, which at first glancemight seem an effective method. First, large quanti-ties of bioagent would be needed because the bioag-ent would be diluted in a large water supply. Second,municipal water supplies are almost universallytreated (e.g., with chlorine) to kill harmful bacteriaand viruses.

By far the most effective method for a biologicalattack is to dispense the bioagent as aerosol particlesin the air. The bioagent particles then float in the airuntil they are inhaled. Aerosol attacks have two majoradvantages: (1) the bioagent particles are naturallydispersed in the atmosphere and, driven by the wind,can drift over large areas, and (2) many diseases aremore virulent when spread by the aerosol route. Formaximum effect it is generally agreed that the bioag-ent particles should be in the size range of 1 to 10 µm.Larger particles will precipitate out of the air in tooshort a time; smaller particles will tend to be expelledfrom the lungs.

Aerosol particles can be generated in various ways.Perhaps the simplest way is to employ a sprayer of thekind used with insecticides. Such a sprayer could becarried by an individual or mounted in a truck or air-

craft. Biological weapons can also be dispersed bybombs, missiles, and artillery shells.

Bioagents tend to decay rapidly when exposed toultraviolet light; thus attacks at night are more effec-tive than in the daytime. Ideal meteorological condi-tions for an attack would include an inversion layer,to keep the bioagent cloud trapped near the earth’ssurface, and moderate steady winds of about 5 to 15kt. These are the conditions a logical, rational attackermight choose; but given that the use of biologicalweapons could be considered, ipso facto, proof of irra-tionality, it would be unwise to assume that biologicalweapons will only be used under optimum condi-tions. We should be prepared for biological attacksunder a wide range of conditions.

Bioagents

Possible bioagents fall into five categories: (1) bacte-ria, (2) rickettsiae (which are similar to bacteria), (3)viruses, (4) fungi, and (5) toxins of biological origin.Toxins are, in fact, more like chemical agents in thatthey are not living organisms and do not reproduce inthe human body; nevertheless, they are normally in-cluded as bioagents.

The list of candidate bioagents is actually remark-ably short. For instance, the Australia Group—a con-federation of nations concerned with the proliferationof chemical and biological weapons—has fewer thanfifty possible bioagents on its primary list [8]. Genetic

FIGURE 1. The five principal components of biological de-fense. Detection and warning constitute the keystone.

Deterrence Decontamination

Protection Medical

Intelligence gatheringDiplomacyInspection

EquipmentPersonnel

ImmunizationDiagnosisTreatment

Individual masksAir filters

Detectionand

warning



engineering could, of course, vastly increase the num-ber of potential bioagents.

To be an efficacious bioagent an organism mustsatisfy six broad criteria. It must (1) be relatively easyto manufacture in sufficient quantity, (2) be storablefor months or years without significant degradation,(3) be disseminatable as a 1-to-10-µm aerosol, (4)survive dissemination effects (e.g., from a bomb), (5)survive for hours in the air without degradation, and(6) cause appropriate effects in its intended victims.

Note that there can be considerable variation inwhat an attacker considers appropriate effects. For in-stance, in a wartime tactical use of bioagents an at-tacker may prefer a fast-acting toxin; in a terrorist actthe attacker may prefer a bioagent with a long gesta-tion period. A military user will likely not want a bio-agent that could cause an epidemic among his ownpeople and thus, might prefer an epidemic-causing,contagious bioagent for the additional panic and dis-ruption it could cause.

Darryl P. Greenwood at Lincoln Laboratory con-ducted a study to assess the bioagents most likely tobe employed by a military user. The most likely bio-

agents and some of their characteristics are given inTable 1, which is derived from Greenwood’s study.The bioagent characteristics in Table 1 have beenscored from the point of view of the biological at-tacker: green is good for the attacker; red is bad forthe attacker; yellow is either fair for the attacker orrepresents an emerging threat worth watching. Notethat what is good and bad can change depending onthe attacker’s mission. Thus, we have scored high fa-tality rate as good, but some attackers may prefer abioagent that is only incapacitating. As a natural bio-agent, anthrax is clearly in a class by itself; it leads vir-tually everyone’s bioagent threat list. (See the sidebarentitled “Anthrax as a Bioagent.”)

Detection System Architecture

Figure 2 shows a detection architecture for defendinga military port against a bioagent attack. The de-fended area is a military airfield, but the arrangementwould be similar for a seaport, staging area, or othermilitary installation. There would be a network ofpoint sensors distributed around the airfield. Most ofthese sensors would be at the perimeter fence, but

FIGURE 2. Nominal detection architecture for defense of a military airfield against a biological attack. The sensor network in-cludes stationary point sensors, mobile point sensors, and a standoff lidar sensor. All sensor information is networked to-gether to present an integrated picture to chemical/biological or medical officers.

Networked array,integrated picture,decision making

Chemical/biological or medical officers

Fieldcommander

Command authority

Fieldunits

Defended area (airfield) Wind direction

Enemybiorelease

Communicationlinks

Sensor with communication linkStandoff lidar sensor



Table 1. Likely Biological Agents and Their Characteristics

Bacteria and Rickettsiae

Disease # Cells Incubation Vaccine Intervention Epidemic Fatality Treatability*to Infect (days) Availability (days) Risk Risk

Anthrax 10,000 1–7 Yes 0.5–1 None High Yes†

Pneumonic Plague 3000 1–6 Ineffective 0.5–1 High High Yes†

Tularemia 10–50 1–10 Yes 0.5–1 None Moderate Yes

Glanders, Unknown 3–5 No 3 Low Moderate YesMeloidosis

Brucellosis 1300 7–21+ No 1–7 None Low Yes

Q Fever 1–10 7–28 Yes 1–7 None Very low Yes

Psittacosis Unknown 1–15 No 3–5 Low Low Yes

Rocky Mountain A few 2–10 Yes 2–7 None Low YesSpotted Fever

Viruses

Smallpox A few 7–21 Yes 3–7 Moderate High No

Dengue Unknown 2–7 In development 1–2 High Very low No

Equine 25‡ 2–15 Experimental 2 Low Low to high NoEncephalitis (EE)

Hantaan Unknown 5–42 In development 2–4 Low High No

Congo-Crimean Unknown 1–2 Experimental 1–2 Low High NoHemorrhagic Fever

Ebola Unknown 2–21 No 0.5–2 Moderate Very high No

Lassa Unknown 3–16 Experimental 1–2§ Low Low No

Biological Toxins

Disease/ Fatal Dose Onset Antitoxin Intervention Stability Fatality TreatabilityAgent (µg) (hours) Availability (hours) Risk

Botulinum 0.1 1–72 Yes 0.5–1 Low High No

Staphylococcal 1–10 1–12 No 0.5–1 Moderate Low NoEnterotoxin B

Perfringens 10–500 6–24 No 1–6 Moderate Low to high No

Ricin 100 8–12 In development 0.5–1 Moderate High No

Saxitoxin ~800 0.1–2+ No 0.2 Low Low No

Tetrodotoxin ~800 0.1–8 No 0.1 Low Moderate No

Aflatoxin 100,000 Very long No 6 High Long term No

* With antibiotics; † If detected early; ‡ Refers to Venezuelan Equine Encephalitis; § According to anecdotal evidence.



everyone’slist of diseases likely to be em-ployed in biological warfare [1].Anthrax, in common with severalother biological-warfare diseases,is principally an animal disease; itmainly effects herbivores, such ascattle, sheep, and goats. Anthraxhas been known and describedsince antiquity, dating back to atleast the fifth Egyptian plague (c.1500 B.C.). It remains pandemicin many parts of the world.

Anthrax is caused by the bacte-rium Bacillus anthracis, which wasfirst isolated by Robert Koch inthe 1870s. Koch used this bacte-rium to provide the first experi-mental proof for the germ theoryof disease. The famous “Koch’spostulates” for proving that a spe-cific microorganism causes a spe-cific disease came from his workwith Bacillus anthracis.

A feature that makes Bacillusanthracis particularly attractive asa bioagent is that, when key nutri-ents are removed, the vegetativebacterium spontaneously devel-ops into a dormant state known asa spore. An electron-microscopicpicture of anthrax spores is shownin Figure A. Spores are rugged andresistant to various stresses such asheat, radiation, drying, andchemical disinfectants. In thespore state, bacteria can survivefor at least decades and perhapsmuch longer. For example, during

World War II the British con-ducted experiments with Bacillusanthracis spores on the Scottishisland of Gruinard; the island re-mained contaminated with viableanthrax spores until it was decon-taminated with formaldehydeand seawater in 1986.

Anthrax spores are relativelyeasy to manufacture, store, anddispense. The spores can infecthumans by being inhaled. In fact,inhalation anthrax was onceknown as woolsorter’s disease, be-cause it affected those in the textileindustry who handled wool frominfected sheep.

After inhaling a sufficient doseof anthrax spores (8,000 to50,000) a person typically beginsto show fairly nonspecific cold or

flu-like symptoms in one to fivedays. An untreated patient’s deathnormally ensues in several daysfrom severe respiratory distress.Anthrax, like most bacterial dis-eases, is treatable with antibiotics;to be effective, however, antibiotictreatment for anthrax must beginbefore symptoms occur. Vaccinesare available, and in fact, all U.S.military personnel are being vacci-nated against anthrax. Anthrax isnot contagious from person-to-person; thus, an anthrax attackwould not cause an epidemic.

Reference1. J.C. Pile, J.D. Malone, E.M. Eitzen,

and A.M. Friedlander, “Anthrax as aPotential Biological Warfare Agent,”Arch. Intern. Med. 158 (5), 1998, pp.429–434.

A N T H R A X A S A B I O A G E N T

FIGURE A. Electron microscope picture of Bacillus anthracis spores. This pic-ture was provided by John Ezzell at the U.S. Army Medical Research Institutefor Infectious Diseases (USAMRIID).



FIGURE 3. Block diagram of complete bioagent sensor. The front end is an early-warning trigger that con-stantly samples the air and rapidly issues an alarm (within a minute) if suspicious bioaerosols are de-tected. The back end is an identifier that is turned on by the trigger and precisely identifies the bioagent toconfirm the attack and guide medical response. In between may be several stages of intelligent particlesampling and fluid preparation to enrich and purify the sample to enhance the performance of the identifier.

some would be placed in the center of the airfield.Robert W. Miller at Lincoln Laboratory performed adetailed analysis of this defensive network. His analy-sis showed that about fifty point sensors might be re-quired for effective defense.

In addition to the emplaced point sensors, severalsimilar sensors could be mounted on mobile plat-forms such as ground vehicles or unmanned air ve-hicles (UAVs). These vehicles could either patrol up-wind of the airfield or be on call to respond ifsuspicious events occurred upwind. A lidar systemcould scan the area upwind of the airfield to providestandoff detection of any biological attack.

To provide an integrated picture of a possible bio-logical attack, information from all the sensors wouldbe transmitted over high-speed data links and inte-grated. Information from ancillary sensors, such asmeteorological sensors, would also be included. Allthis information would be presented in a coherentway to the chemical/biological officer and the medi-cal officer so that they could make informed decisionsand alert the field commander. Automatic warning totroops could also be implemented.

The point sensors shown in Figure 2 would eachcomprise several subsystems, as shown in Figure 3. Atthe front end of the sensor system would be an early-warning (or trigger) sensor that continuously samplesthe air for suspicious events. The early-warning sensorwould not identify specific bioagents, but it would becapable of classifying an aerosol cloud as a threatcloud or a nonthreat cloud. When the sensor detecteda threat cloud, it would sound an alarm.

At the back end of the sensor system would be anidentifier. This system would turn on when the early-warning sensor triggered an alarm. The identifierwould precisely identify the bioagent to confirm theattack and guide the medical response.

Between the early-warning sensor and the identi-fier might be several stages of intelligent particle sam-pling and fluid preparation. These stages would takethe airflow from the early-warning sensor and enrichand purify it, so that the identifier would have asample high in suspicious particles and low in back-ground dirt particles.

An architecture similar to that shown in Figure 2could apply in civil defense against a possible terroristattack. Figure 4 shows the architecture for defense ofa building. Point sensors would be distributedthroughout the building, particularly in air-handlingducts. First responders (e.g., firemen) could handcarry additional sensor units into the building. Again,all the sensors would be networked to give an inte-grated picture, in this case to civil authorities andmedical personnel.

The sensor architectures shown in Figures 2 and 4are designed to warn people before they receive infec-tious or toxic exposures so that they can take protec-tive action. In most cases, a fairly simple mask willprotect a person from a bioagent aerosol attack.

To be able to issue a timely warning to don masks,however, a sensor must operate rapidly and with highsensitivity. Sensitivity and time requirements for asensor collocated with the persons it is protecting areshown in Figure 5. Figure 5 shows the bioagent con-

Intelligent particlesampling

TriggerAir flowBio

Nonbio

Exhaust

Downstream analysis(increased specificity)

Upstream biosensing(rapid response)

IdentifierFluidpreparation

Tentativebioagentwarning

Bioagentconfirmation,identification



centration needed for an unprotected person breath-ing at a normal rate to get an infectious dose, plottedagainst time for various potential bioagents. For ex-ample, at a concentration of 100 particles/liter of Ba-cillus anthracis (anthrax) a person would become in-fected in ten minutes. On the other hand, forFrancisella tularensis (tularemia, or rabbit fever) at aconcentration of only 10 particles/liter a personwould become infected in less than a minute.

To protect a person from disease we need to detectand warn in less time than it takes to infect. Thus, oursensor goals are a sensitivity of 1 particle/liter or fewerand a detection time of less than one minute.

As anyone who has had a throat culture performedknows, most current medical lab techniques are in theupper right of the plot in Figure 5; that is, they takemany hours to days and require many particles. Thesetechniques, of course, are not “detect-to-warn” butare “detect-to-treat.” Illness onset after a biological at-tack typically can take several days; thus, precise iden-tification of the disease need not be as rapid as the ini-tial warning to don masks. We note, however, that formany diseases (e.g., anthrax) treatment must beginbefore symptoms occur, or it is too late to prevent

death. Rapid identification is, therefore, important toinitiate appropriate medical treatment and to help re-duce panic after an attack.

FIGURE 4. Nominal detection architecture for defense of a civilian building against a terrorist attack. The architecture elimi-nates the standoff lidar in the military-port architecture of Figure 2, but adds portable sensors hand carried by first re-sponders. Sensor data are networked as in the military-port architecture, but the integrated information is now presented tocivilian first responders, medical personnel, and other civilian officials.

FIGURE 5. Requirements for a bioagent detector to protectpersonnel against infectious or toxic exposure. The requiredresponse time is plotted against the bioagent concentrationfor various bioagents. This plot assumes that the sensor isbeing used to warn a collocated, unprotected (i.e., un-masked) individual who is breathing at a normal rate of 10 to20 liter/min). Note that SEB is staphylococcal enterotoxin B,and VEE is Venezuelan equine encephalitis.

1 min 1 day10 min 1 hr 6 hr

1

10

102

103

104

105

Time after exposure

SEB

The g

oal

AnthraxPlagueBotulism

SmallpoxTularemiaVEE

Q fever

Bio

agen

t con

cent

ratio

n (p

artic

les/

liter

of a

ir)

Illness onset(days)

Clostridial toxicoses(botulism/SEB)

Medical laboratorytechniques

Networked array,integrated picture,decision making

First responders,medical personnel

Local, state, and national officials

Communicationlinks

Sensor with communication link



Biological-Agent Warning Sensor

Thomas H. Jeys at Lincoln Laboratory developed theconcept for a novel biological agent warning sensor(BAWS) [9, 10] based on the principle of laser-in-duced fluorescence. Subsequently, he has led the de-velopment and field testing of several versions of thissensor. The function of this sensor is to provide earlywarning of the presence of suspicious biological par-ticles in the air. Consequently, the BAWS sensor candiscriminate suspicious particles from naturally oc-curring particles, such as pollen and mold spores, butit cannot identify specific bioagents. Identification isthe function of another sensor, to be described later.

The BAWS concept of operation may be describedwith reference to Figure 6. A fan draws air into the

sensor. Individual aerosol particles in the airstreamare illuminated by a pulsed ultraviolet (UV) laser op-erating at 266 nm. Some of the light is scattered fromthe aerosol particle; some of the light is absorbed bythe particle and rapidly reemitted at longer wave-lengths in a process known as laser-induced fluores-cence. Light from the particle is collected in threephotomultiplier detectors. The first detector is spec-trally filtered to collect only the scattered light at 266nm; this detection signal is proportional to theparticle’s size. The second detector is spectrally fil-tered to collect UV light in the 300-to-400-nm range;this signal comes principally from biofluorescence ofthe amino acid tryptophan, which is found in almostall living organisms. The third detector is spectrallyfiltered to collect visible light in the 400-to-600-nm

Detected signal

Pulsedultraviolet

laser

Spectral filter

Time

Wavelength (nm)

Tryptophan

Elastic scattering and fluorescence emission spectrum

NADH

Agent-containing

particle

Fluorescenceemission and

elastic scattering

Flavin compounds

Rel

ativ

e si

gnal

330 450 560266

Elasticscattering

Photodetector

FIGURE 6. Concept for the Lincoln Laboratory biological-agent warning sensor (BAWS). Indi-vidual aerosol particles are illuminated by a pulsed ultraviolet laser operating at 266 nm. Light fromthe aerosol particle is detected in three channels: one at 266 nm, which measures the elastic scat-tering; a second between 300 and 400 nm, which measures the ultraviolet fluorescence from suchsubstances as the amino acid tryptophan; and a third between 400 and 600 nm, which measures thevisible fluorescence from such substances as NADH (the reduced form of nicotinamide adeninedinucleotide) and flavin compounds.



range; this signal comes principally from biofluores-cence of NADH (the reduced form of nicotinamideadenine dinucleotide) and flavin compounds, whichare also found in living organisms.

The technique of using laser-induced fluorescenceto detect bioaerosols is not novel; it has been knownfor more than two decades. Making this techniqueinto a practical, high-performance sensor, however,has been made possible by two Lincoln Laboratorydevelopments: (1) the passively Q-switched microla-ser and (2) sophisticated processing for real-timediscrimination.

The diode-pumped, passively Q-switched microla-ser was invented by John J. Zayhowski [11–13]. Thislaser, shown in Figure 7, consists of an approximately1-mm3 Nd:YAG (neodymium-doped yttrium-alumi-num-garnet) laser chip and a Cr:YAG (chromium-doped YAG) Q-switching chip. The laser is pumpedthrough a fiber by a conventional 808-nm diode laser.The 1064-nm laser output is doubled to 532 nm by aKTP (titanate-phosphate) crystal and doubled againto 266 nm by a BBO (barium-borate) crystal. The fi-nal output at 266 nm consists of about 1-µJ, 500-psecpulses at a 10-KHz repetition rate. As shown in Fig-ure 7, the entire laser assembly fits in a housing aboutas big as a person’s thumb. The laser has no movingparts or adjustments and is extremely rugged. (In an-other implementation the laser was used in a pollu-tion-monitoring system that was pounded into theground in the head of a cone penetrometer [14].)

The small size and low power consumption of themicrolaser enable us to build a compact, low-powerbioaerosol sensor. The high-repetition rate of the laserenables us to interrogate individual aerosol particlesat an effective sampling rate (about 20 liter/min) thatexceeds normal human breathing rate.

The three detection channels permit us to dis-criminate threat aerosols from nonthreat aerosols.The discrimination processing may be described heu-ristically with reference to Figure 8, which shows fourtwo-dimensional histograms for many measurementsmade with the BAWS. In each case the horizontal axisis the normalized difference between the UV and vis-ible channels, and the vertical axis is the normalizeddifference between the UV and elastic channels. Thenumber of particles for particular differences is indi-

cated by color, with the red end of the spectrum de-noting higher concentration and the violet end de-noting lower concentration.

The upper two histograms in Figure 8 are for twobacteria used as simulants for bioagents: Bacillusglobigii, which is a simulant for Bacillus anthracis (an-thrax), and Erwinia herbicola, which is a simulant forYersinia pestis (plague). The lower left histogramshows a natural background measurement made in adesert environment at Dugway, Utah. The lower righthistogram shows a ragweed pollen.

We see from Figure 8 that the bioagent-simulanthistograms fall in the upper half of the plots and the

FIGURE 7. Diode-pumped, passively Q-switched microlaser.This laser, invented at Lincoln Laboratory, has been built invarious configurations. In the BAWS configuration the fun-damental 1064-nm radiation from the Nd:YAG (neodymium-doped yttrium-aluminum-garnet) laser is quadruped to 266nm. Output at 266 nm is approximately 1 µJ per pulse withless than 1-nsec pulse width at a 10-kHz pulse rate. Note thatBBO represents barium borate, Cr:YAG represents chro-mium-doped YAG, and KTP represents titanate phosphate.

3.5 cm

808 nm 1064 nm 532 nm 266 nm

Diodelaser

Nd:YAGlaser

2ndharmonic

generation

4thharmonic

generation

Laser wavelengths

Fiber

Cr:YAG

KTP crystal

BBO crystal

Window

UV laser

Nd:YAG



FIGURE 9. Three generations of the biological-agent warning sensor (BAWS ). (a) The 1996 proof-of-concept sen-sor required a small van full of electronics for operation. (b) The 1997 BAWS II is a 3.4-ft3 stand-alone sensor weigh-ing 96 lb. (c) The 1999 BAWS III, carried by Thomas H. Jeys of Lincoln Laboratory, is less than 0.8 ft3 and weighsabout 19 lb. Detection and discrimination performance has improved with each generation.

background histograms fall in the lower half of theplots. The simulant and background histograms areclearly separated. These features are common to otherbioagent simulants (and actual bioagents) and otherbackgrounds. This separation in features allows us todiscriminate threat particles from nonthreat particles.

The full real-time bioagent discrimination algo-rithm has been developed by Nathan R. Newbury atLincoln Laboratory. It is, of course, more complicatedthan simply looking at the histograms, and we willnot describe it here in mathematical detail. The fol-lowing discussion, however, provides a general out-line of how the sensor functions.

The BAWS operates continuously, storing mea-surements from every particle on which there are sig-nals in all three channels. When the sensor is firstturned on, it spends several minutes “learning” thebackground. The sensor then looks for “events,” thatis, significant changes from the background. If noevents are detected, it continues to update the back-ground. If an event is declared, the sensor classifiesthe event as threat or nonthreat. If the event persistslong enough (e.g., fifteen seconds or more), an alarmis issued. The detection, classification, and alarmfunctions are all built into the BAWS and are com-pletely automatic.

As illustrated in Figure 9, three generations ofBAWS fluorescence sensors have been designed and

FIGURE 8. Histograms of detection events for the BAWS.Histograms are given as functions of the normalized differ-ence between the ultraviolet (UV) and visible channels (hori-zontal axis) and the normalized difference between the UVand elastic channels (vertical axis). Histograms are shownfor Bacillus globigii (a simulant for Bacillus anthracis), forErwinia herbicola (a simulant for vegetative bacterial bioag-ents, such as Yersinia pestis), for ragweed pollen, and for atypical Dugway, Utah, dirt background. Note that bacteriacan easily be distinguished from background dirt and pollenbecause of the separation on the histogram plots.

Bacillus globigii Erwinia herbicola

Dugway dirt background

UV – visibleUV – visible

1

0.5

0

–0.5

–1

UV

– e

last

icU

V –

ela

stic

1

0.5

0

–0.5

–0.5–1 0 0.5 1–1

–0.5–1 0 0.5 1

Ragweed pollen

(a) (c)(b)



built. Gregory S. Rowe was the chief electronics andsoftware engineer for all three designs. Robert P.Brady was the chief mechanical engineer for the lattertwo designs, which required considerable attention topackaging. In 1996 a proof-of-concept sensor was de-signed, fabricated, and field-tested at a major Govern-ment-run trial in only seven months. The sensor hada reasonably compact sensor head, but the electronicswere commercial units housed in a mini-van. Theprototype sensor had only the two fluorescence chan-nels; it did not have an elastic-scattering channel. Thesensor performed very well in the Joint Field Trials(JFT-3) conducted at Dugway Proving Grounds inDugway, Utah, in the fall of 1996.

Following the success of the prototype sensor wedesigned a compact, self-contained BAWS, fabricatedthree sensors, and fielded them at Dugway for theJFT-4 trials in the fall of 1997. This BAWS II mea-sures 3.4 ft3, weighs 96 lb, and consumes 50 W ofpower. It can run on batteries as an autonomous unit,perform detection and discrimination functions withembedded electronics, and issue an alarm. The sensor

was first implemented with only two channels; lateran elastic-scattering channel was retrofitted.

During 1998 and 1999 we developed the BAWSIII. This sensor has a redesigned optical head for in-creased sensitivity and new discrimination algorithmsfor improved performance. The BAWS III is less than0.8 ft3, weighs only 19 lb, and consumes only 40 W.Thus, the BAWS III is truly a portable sensor. TheBAWS III was successfully tested in the spring of1999. A photograph of two BAWS IIIs at DugwayProving Grounds is shown in Figure 10.

The JFT-4 trials, held at Dugway Proving Groundsfrom 6 October to 2 November 1997, provided an in-dependently refereed test of the BAWS II, along withother candidate early-warning sensors. Figure 11shows an outline of the test arrangement. The test sitecomprised five test stations equally spaced along a52-m line. At each test station, independent U.S.

FIGURE 11. Test arrangement for the 1997 Joint Field Trials(JFT-4) at Dugway Proving Grounds. Harmless bacteriawere dispensed upwind of the test grid by agricultural spray-ers. Various sensors were stationed in the test grid, includ-ing three BAWS IIs, as indicated.

FIGURE 10. Two BAWS IIIs at Dugway Proving Grounds,Utah, in the spring of 1999. 1

BAWS II-1BAWS II-2

BAWS II-3

2345

52 m

Truth sensor stations

800 m

Wind

North

Agricultural sprayers fordissemination of bioaerosol



The Dugway test site, shown in Figures 12 and 13,is a large, flat, high-desert environment, at a 4000-ftelevation, with low vegetation (less than 2 ft). As onemight expect, such an environment has rather benignnatural background conditions.

Figure 14 shows comparisons of the three BAWS IIsignals and the truth data for a representative Bacillusglobigii release. The truth data are given in blue; theyshow that the simulated attack lasted about sevenminutes and reached a level of about 50 particles perliter. (In the biological defense community thebioagent concentrations are normally expressed asagent-containing particles per liter of air [ACPLA].)This attack is a fairly modest one: if it had been ananthrax attack, a canonical unprotected personbreathing normally would have not quite gotten a le-thal dose.

The real-time signals from the BAWS II sensors areshown in red. We see that they agree very well withthe truth data. The shaded area shows when BAWSII-1 (the officially scored sensor) was in alarm. We seethat the BAWS II went into alarm when the particleconcentration was only about 2 ACPLA. We also

FIGURE 13. BAWS II mounted on a tripod at Dugway Prov-ing Grounds. The BAWS II is a self-contained, stand-aloneunit powered by batteries in the blue box. Air is drawn inthrough the tube on top, which was extended above the sen-sor to be at the same height as the truth sensors. To the leftof the BAWS II are the truth sensors—rotating slit-to-agarculture plates. To the right of the BAWS II are a simple par-ticle counter and the radio link for the BAWS II alarm signal.

Army personnel operated conventional, slit-to-agarsamplers to acquire so-called “truth” data. Followingthe tests, the agar plates were cultured and the bacte-rial colonies counted. Finally, some months after thetrials the truth data were made available to us and toother test participants.

Participating test sensors were placed on either sideof the test stations for official scoring. Our scored sen-sor was beside test station 1. We had two other sen-sors on the test site, but these were midway betweentest stations. They provided us with additional databut were not officially scored.

During the trials bacterial simulants were gener-ated by agricultural sprayers mounted in the truckshown in Figure 12. The simulants were dispensedapproximately 800 m upwind of the test stations anddrifted across the site. Exact release times were not an-nounced. The various early-warning sensors were re-quired to issue alarm-on and alarm-off times. DuringJFT-4, 23 trials were conducted, 20 with Bacillusglobigii and 3 with Erwina herbicola.

Figure 13 shows our scored BAWS II at the JFT-4trials. The sensor is mounted on a tripod and has avertical 2-ft air-sampling tube, so that the air intake isat the same height as the truth sensors. The slit-to-agar truth sensors are in the box to the left of theBAWS II; the box cover is removed during tests. Tothe right of the BAWS II is a simple particle counterthat counts all particles—biological and nonbiologi-cal—in the 1-to-10-µm size range

FIGURE 12. The bacteria dissemination truck for the Dugwaytrials. Harmless bacteria were dispensed from the agricul-tural sprayers on this truck. Note, however, that similarlyunsophisticated equipment could be used to dispense realbioagents.

Particlecounter and

radio linkTruth

sensors

BAWS II



FIGURE 14. BAWS II and truth signals as a function of time for a Bacillus globigiirelease. Note the excellent agreement between the real-time BAWS II signal and thetruth signal obtained after several days of culturing the collected bacteria. The BAWSII correctly issued an alarm at a particle concentration of less than five agent-contain-ing particles per liter of air (ACPLA). Note also that although the three BAWS IIsgenerally measure the same concentrations, there are significant variations in thecloud concentrations even though the sensors are only fifteen to twenty meters apart.

note that there can be considerable differences in par-ticle concentrations measured by the three BAWS IIs,even though the sensors were only spaced about fif-teen to twenty meters apart.

In Figure 15 we show a trial that demonstrates theability of BAWS II to discriminate a bioagent-simu-lant cloud from a dirt cloud. We again plot the truthdata in blue and the BAWS II signal in red. Now wealso plot, in green, the data from the simple particlecounter. In this trial, before the release of Bacillusglobigii, there were trucks driving around the site stir-ring up large clouds of dirt. As we see from Figure 15there are six dirt events, two of them higher than1000 particles per liter. The BAWS II did not triggeran alarm on any of these events; its discrimination al-gorithm identified each of them as a dirt event. TheBAWS II did, however, correctly trigger an alarm ontwo simulant events, even though the total numbersof particles for these events were much less than forthe dirt events. As seen in the expanded view of Fig-ure 15, the maximum concentration for one of the

simulant clouds was less than 20 ACPLA.The BAWS II performed extraordinarily well at

JFT-4. The scored sensor showed a sensitivity ofabout 5 particles/liter and had a response time of 26sec. It correctly declared alarms on all the most stress-ing trials (defined officially as less than 15 ACPLA)and correctly declared alarms on all but one of theless-stressing trials (above 15 ACPLA). (The onemissed trial seems to have resulted from a softwareglitch; the release was actually one of the larger ones,and we easily triggered an alarm when we played therecorded raw data back through the real-time algo-rithm after the trials.) The BAWS II had no falsealarms. In addition, the sensor was able to discrimi-nate the bioagent simulants from unintentionalinteferents, such as dirt clouds and diesel fumes.

The BAWS II also proved robust in the field envi-ronment. The three sensors operated for 170 hourseach during the test period. They operated autono-mously, under battery power. They even surviveddriving rain and snow.

50

50

0102030405060

Tru

th s

igna

l (A

CP

LA)

0

10

20

30

40

50

0

10

20

30

40

BA

WS

II s

enso

r sig

nal (

coun

ts/s

ec)

020

40

60

80100

21:35 21:40 21:45 21:50 21:55 22:00 22:05 22:100

10

20

30

40

020

40

60

80

100

Time

East end of grid

BAWS II-1

Truth

Center of grid

West end of grid

Scored sensoralarm on attest station

BAWS II-2

Truth

BAWS II-3

Truth



Laser Induced Fluorescence forStand-Off Detection

The BAWS laser-induced-fluorescence sensor is a lo-cal (or point) sensor: it detects only those particlesthat actually flow through the sensor. On the otherhand, the BAWS can be regarded as an extremelyshort-range lidar, where the distance from the lasertransmitter to the sensed bioparticles is only a fewcentimeters.

In principle, laser-induced fluorescence can beused to make a much longer-range biodetection lidarsystem. The U.S. Army is developing the Short-Range Biological Standoff Detection System (SR-BSDS) to explore this possibility [15]. Shen Shey,Robert Kramer, and Michael O’Brien at LincolnLaboratory have analyzed the performance and po-tential of the SR-BSDS.

Compared to the BAWS point sensor there aremany technical challenges in building a standoff la-ser-induced fluorescence lidar. Obviously, the UV la-ser and the transmitter/receiver optics must both be

much larger to achieve long-range capabilities. Fur-thermore, some fundamental physics effects severelylimit the performance of a UV lidar. Chief amongthese is that UV light is strongly absorbed (principallyby ozone) and scattered in the atmosphere. For ex-ample, with a 266-nm laser, the signal in the UVfluorescence channel would, under moderately goodatmospheric conditions, be reduced by a factor of 104

in going from a range of 1 km to 5 km.To somewhat overcome the limitations of atmo-

spheric absorption and scattering, we can have the li-dar system look over a wide range gate to increase thenumber of bioparticles it senses on a given laser pulse.Unfortunately, sensing in a wide range gate makes thediscrimination of threat particles from the back-ground more difficult, since the detected signal comesfrom a mixture of many potentially different kinds ofparticles. For the BAWS, one of the factors thatmakes discrimination possible is that the aerosol par-ticles are individually sensed.

Even with wide range gates, our analysis shows thatthe ranges achievable from a UV lidar system will

FIGURE 15. BAWS II, truth, and particle-counter signals as a function of time for a Bacillus globigii release in the pres-ence of dirt clouds. The upper plot shows a roughly two-hour time period during which vehicles were stirring up dirt.The latter part of the upper plot and the expanded lower plot show the actual simulant cloud. Note that the BAWS IIdid not trigger an alarm on dirt clouds of about 1000 particles/liter but correctly triggered an alarm on a biologicalcloud at a level of less than 10 particles/liter.

0

5

10

15

20

0

20

40

60

80

100

16:50 17:00 17:10 17:20

0

50

100

150

200

250

0

200

400

600

800

1000

1200

15:30 16:00 16:30 17:00

BAWS II-1TruthB

AW

S I

I si

gnal

(co

unts

/sec

)

Time

Tru

th s

igna

l (A

CP

LA)

and

par

ticle

con

cent

ratio

n (p

artic

les/

liter

)

Particle counter

TruthBAWS II-1

Alarm onParticle counter



only be a few kilometers and that the concentrationsdetectable at these ranges will be large. It appears thatthe best way to use a UV lidar would be to interrogatea bioagent cloud at the release point (e.g., at the im-pact point of an enemy missile), where the concentra-tions are high (perhaps 105 to 106 particles/liter). Un-der this operational concept a reasonably sized UVlidar system might be able to detect a bioagent releaseout to five kilometers.

We presume that people support detecting and dis-criminating a bioagent cloud and issuing a warningbefore the cloud reaches them. Thus, even with itsrelatively short range, a standoff lidar could be usefulin an overall detection and warning system; weshowed such a lidar in the military-port defense archi-tecture in Figure 2. We observe, as an example, that ifthe wind speed were 20 km/hour (10 kt), and if thelidar were able to detect a release 2 km upwind, itwould provide the port or air base with an additionalsix minutes of warning time.

Background Measurements

The BAWS described earlier can detect singlebioaerosol particles. Likewise, the bioelectronic sen-sor to be described in the next sections can, in prin-ciple, identify a substance from a single-particlesample. Thus, these sensors have extraordinary sensi-tivity, given a pure sample. Consequently, the ulti-mate performance of these sensors will not be limitedby their fundamental sensitivities, but by the effectsof background particles in the sample volume.

Although we often do not perceive them, the airwe breathe is heavily laden with a wide variety ofaerosol particles. (See the sidebar entitled “The Natu-ral Aerosol Background.”) There are ordinary dustand dirt particles. There are natural bioaerosols, suchas pollen, mold spores, and benign bacteria. There areanthropogenic particles, such as particles in diesel ex-haust and particles from automobile tires. The totalnumber of background aerosol particles per liter of airnormally exceeds the desired sensitivity of a bioagentsensor by many orders of magnitude.

Background aerosols can have two significant del-eterious effects on a bioagent sensor. First, they canmask real biological attacks, causing a sensor to missan attack. Such a missed detection could, of course,

be catastrophic. Second, the background aerosols canerroneously trigger the bioagent sensor, causing a falsealarm. A false alarm is not catastrophic; but if falsealarms are too frequent, people will quickly begin toignore the alarms, and the sensors will becomeuseless.

Thus, to develop a practical bioagent detector, wemust understand the aerosol background. Further, wemust understand how the background affects the de-tection probability and false-alarm rate of a bioagentdetector.

Many measurements have been made and con-tinue to be made of the atmospheric aerosol back-ground. For instance, the aerobiology communityregularly monitors the concentrations of pollen andmold spores in the air. As another example, the Envi-ronmental Protection Agency (EPA) monitors the to-tal number of particles in several size ranges. Unfortu-nately, existing background measurements areinadequate for our purposes because of two majordeficiencies.

First, existing measurements do not have adequatetime resolution. Pollen counts, for instance, may onlybe measured and reported on a daily basis. Othermeasurements may be made on an hourly basis. Incontrast, as can be seen from Figures 14 and 15, bio-agent attacks typically occur on relatively short timescales. As a biological-agent cloud drifts over a site itmay have a rise time of less than a minute. It may re-main over a site for about ten minutes and then driftaway in less than one minute. Thus, the time scaleswe care about for a biological attack are typically inthe range of ten seconds to ten minutes.

Second, existing background measurements do notuse techniques representative of those we expect touse for bioagent detectors. In particular, existing mea-surements do not use fluorescence detection. We careabout the background, after all, only insofar as it af-fects our measurements. For instance, there may bemany background aerosols that are sensed by a totalparticle counter but that do not fluoresce. These par-ticles would be totally unimportant to our BAWSfluorescence sensor.

To obtain relevant data on background aerosolsLincoln Laboratory, under the direction of Nathan R.Newbury and Jae H. Kyung, is conducting a back-


20 THE LINCOLN LABORATORY JOURNAL VOLUME 12, NUMBER 1, 2000

air custom-arily teems with tiny aerosol par-ticles. Except under extreme con-ditions (e.g., a dust storm), we donot normally sense these particles;although almost everyone hasseen floating dust motes illumi-nated by a sun beam, and personsafflicted by hay fever sense pollengrains in the air even if they cannot see them. It is in the presenceof this natural aerosol backgroundthat bioagents must be sensed.

The situation is illustrated inFigure A. We would like to detectbioagent particles at concentra-tions as low as 1 particle per liter ofair. In a liter of air in a remote,

FIGURE A. Illustrations of the number and composition of background aero-sol particles in the air. Concentrations of particles 1 to 10 µm in diameter rangefrom a few hundred per liter in remote semiarid environments to greater than10,000 particles per liter in urban environments.

semiarid environment, however,there may be several hundred to athousand other aerosols in the sizerange of a bioagent particle (1 to10 µm). In a liter of urban air theremay be 10,000 such particles.

Figure B gives more quantita-tive information on the naturalbackground. We plot the numberof particles above a given diameteragainst particle diameter for fourrepresentative conditions [1].Note that these are nominal, aver-age conditions; particle concen-trations may easily fluctuate anorder of magnitude around theseconditions.

We see from Figure B that the

particle concentration is a strongfunction of particle size. There aretypically millions of particlesabove 0.1 µm, thousands of par-ticles above 1 µm, and only a fewparticles above 10 µm. This strongsize dependence occurs becauselarge particles rapidly fall out ofthe air, whereas small particles arebuoyant enough to stay airbornefor long periods of time.

We also see from Figure B thatthere are significant differences inthe particle concentrations in dif-ferent environments. In the 1-to-10-µm size range the remote,semiarid environment has thefewest background particles; theurban environment has the most;and the rural environment falls inbetween.

In addition to differing num-bers of particles, the different en-vironments have different com-positions of aerosol particles. Inarid and semiarid environmentsmost of the natural aerosol par-ticles are soil derived—mainlyclay minerals such as illite, mixed-layer clays, and kaolinite [2].

Rural environments add bio-logical aerosols to the soil-derivedbackground. There are many dif-ferent types of biological aerosols,including fungal spores, bacteria,pollen, plant spores, plant debris,algae, protozoa, insect parts andfeces, skin cells, and viruses [3].

In urban environments anthro-

T H E N A T U R A L A E R O S O L B A C K G R O U N D

Number andcomplexity of

aerosol sources

Soil-derived (100 –10,000/liter)

Clay

Sand

Composite

Bioaerosols (0 –100/Iiter)

Fungi (0 –100/Iiter)

Bacteria (0 –1/Iiter culturable)

Pollen/other (0 –1/Iiter)

Man-made or anthropogenic(0 –10,000/Iiter)

Combustion (fly ash, vehicleexhaust, woodstoves,incinerators, fires)

Mechanical (tire or tar,industrial processes)

Bioagent (0/Iiter)e.g., anthrax (shown at 10/Iiter)

UrbanRemote semiarid



FIGURE B. Representative cumulative concentration of particles above aspecified particle diameter. The concentrations vary significantly from locationto location. For all locations the concentrations decrease sharply as the par-ticle diameter increases.

genic aerosol particles are added tobiological and soil-derived par-ticles. Anthropogenic particlescome from coal and oil-burningplants, diesel and gasoline engineexhaust, road-dust resuspensionand tire-particle generationfrom vehicle movement, wood-burning stoves, and industrialprocesses [4].

Soil-derived aerosol particlesare generally not a significant in-terferent for UV fluorescence-based sensors. They can, however,be a major problem for simple par-ticle counters, and they can clog orotherwise overwhelm identifyingsensors.

Of the biological aerosols, fun-gal spores are by far the most im-portant interferents for a bioagentsensor. Fungal spores are ubiqui-tous: wherever there is decayingorganic matter, there are fungalspores. They are about the samesize as bioagent particles. They areviable biological particles, so theycontain many of the same basicbiological components as bioag-ents. Finally, concentrations of 10particles/liter are common, andconcentrations can easily reachgreater than 100 particles/liter.

Natural bacteria are not par-ticularly important interferentsfor two reasons. First, concentra-tions are usually lower than 1 par-ticle/liter. Second, natural bacte-ria normally are found rafting onlarger airborne particles.

Although it might be hard tobelieve for those who suffer fromhay fever, pollen is an insignificant

interferent for bioagent sensors.Pollen grains are large—20 to 100µm; they are, in fact, this large sothat they will readily fall out of theair onto plants to perform theirpollination function. Thus, air-borne concentrations of pollen arevery low; even during peak pollenseasons average concentrationsrarely exceed 1 particle/liter. Forthose unusual occasions when thepollen concentration exceeds 1particle/liter (e.g., directly underan oak tree in May), the pollengrains may easily be distinguishedfrom bioagent particles on thebasis of their size.

For a UV fluorescence-basedbioagent sensor, the most impor-tant anthropogenic particles arepolycylic aromatic hydrocarbons(PAHs) generated in incompletecombustion of fossil fuels. PAHsfluorescence strongly under UVlight; thus, if enough PAH mol-

ecules are adsorbed on a particle ofthe right size, the signature mightmimic that of a bioagent particle.Other anthropogenic particlesmay spoof particle counters orclog identifying sensors.

References1. R. Jaenicke, “Tropospheric Aerosols,”

in Aerosol–Cloud–Climate Interactions,P.V. Hobbs, ed. (Academic Press, SanDiego, 1993), pp. 1–31.

2. E.M. Patterson, “Size Distribution,Concentrations, and Composition ofContinental and Marine Aerosols,” inAtmospheric Aerosols: Their Formation,Optical Properties, and Effects, A.Deepak, ed. (Spectrum Press, Hamp-ton, Va., 1982), pp. 1–38.

3. H.A. Burge, ed., Bioaerosols (LewisPublishers, Boca Raton, Fla., 1995);C.S. Cox and C.M. Wathes, Bioaero-sols Handbook (Lewis Publishers, BocaRaton, Fla., 1995).

4. J.M Prospero, R.J. Charlson, V.Mohnen, R. Jaenicke, A.C. Delany, J.Moyers, W. Zoller, and K. Rahn, “TheAtmospheric Aerosol System: AnOverview,” Rev. Geophys. & SpacePhys. 21 (7), 1983, 1607–1629; R.M.Harrison and R.E. van Grieken, Atmo-spheric Particles (Wiley, New York,1998).

0.10.01

108

106

104

102

100

10–2

1 10 100

Diameter ( m)

Cum

ulat

ive

part

icle

con

cent

ratio

nab

ove

a gi

ven

part

icle

dia

met

er(p

artic

les/

Iiter

) Dust storm

Urban

Remote semiarid

Rural

µ



FIGURE 17. Locations where background measurementshave been made. Measurements have been made across thecountry in urban, rural, and desert sites.

ground measurement program. We have equipped avan, shown in Figure 16, with various instruments.The principal instrument is a BAWS fluorescencesensor. We have also purchased and installed variouscommercial instruments including a TSI aerody-namic particle sizer to measure the total number ofparticles as a function of size, several slit-to-agar sam-plers to measure culturable bioaerosols, and aBurkard spore/pollen sampler to collect large aerosolson tape for later counting under a microscope. In ad-dition, we have included routine meteorological in-strumentation to measure wind speed and direction,air temperature, and relative humidity.

Figure 17 shows sites from our extensive measure-ment campaign in progress throughout the continen-

tal United States. Data have been taken in urban,desert, rural, and coastal areas. Generally, data aretaken continuously for a week at each location. Wehave returned to some locations during differenttimes of the year to assess seasonal variations. Mea-surements are planned for additional locations in theUnited States and abroad.

Figure 18 shows background data analyzed andcompiled by Ann E. Rundell. The data are from fivedifferent locations representing desert, urban, rural,and coastal environments. We plot the power spectraldensity of the aerosol particle concentrations mea-sured by the BAWS. For this plot a counted particle isone that had a signal in all BAWS channels; the dis-crimination algorithms discussed in connection withFigure 8 have not yet been applied.

We draw a number of interesting conclusions fromFigure 18. First, we observe that there is wide varia-tion in the backgrounds at the five different sites. Theurban and rural sites in Alabama, Georgia, and Mis-souri have the highest background fluctuations; thecoastal and desert locations in Maryland and Utahhave the lowest.

We also see from Figure 18 that the backgroundfluctuations decrease significantly as a function of fre-quency. The fluctuations in the region that we careabout (from ten seconds to ten minutes) are muchless than those at longer time scales. This phenom-enon is further illustrated in Figure 19, in which we

FIGURE 16. The Lincoln Laboratory mobile bioaerosol laboratory. This van contains a BAWS sensor as well as sev-eral commercial sensors, including a particle sizer/counter, a spore and pollen sampler, and a slit-to-agar sampler.Measurements can be made with the van either stationary or moving.

Particlesizer/counter

Bioaerosolfluorescence

sensor

Spore and pollensampler

Weather station

Aerosol samplerintakes

Massachusetts(urban)

Maryland(coastal)

Georgia(urban/light

industry)Florida

(coastal)

Missouri(rural)Utah

(desert)

Alabama(rural)



plot a time history of particles per liter measured bytwo different sensors. The green curve shows the fun-gal spore count as measured by the Burkard sporetrap. The red curve shows particles measured by theBAWS sensor as having fluorescence signatures in theregion expected for mold spores. We see that the twosensors agree well on the general fluctuations in fun-gal spores. More importantly, we see that most of thefluctuations are on a diurnal basis. These long-time-scale variations are easily distinguishable from thoseexpected during a biological attack.

Another feature apparent from Figure 18 is thatthe fluctuations decrease with approximately an in-verse frequency (1/f ) dependence. A 1/f dependenceis characteristic of noise spectra that are driven bymany different distinct events [16]. Indeed, we findthat we can correlate many of the largest backgroundfluctuations with discrete, principally man-madeevents. Figure 20 illustrates this effect. The upper plotshows a temporal history of fluorescent particles perliter for Friday, 9 October 1998, when there was con-siderable human activity at Fort Leonard Wood, Mis-souri. The activities included trucks and backhoesmoving around, and even the detonation of an explo-sive. The plot identifies approximately when these ac-tivities occurred. We see that there are large, short-time-scale increases in the background associatedwith these activities; the exact times do not overlapbecause the activities were at various distances from

the sensor, and it took time for the aerosols kicked upinto the air to drift past the sensor.

The lower plot in Figure 20 shows measurementsof the same site on Sunday, 11 October 1998, duringthe long Columbus Day weekend when there waslittle human activity around the site. We see that thefluctuations are very much lower than those in theupper plot.

As we mentioned earlier in this section, the impor-tant aspect of the background is how it affects a

FIGURE 18. Power spectral density of aerosol particle fluctuations measured by the BAWS at various sites. Each plotis from one week of continuous measurements. Each plot has roughly an inverse frequency (1/f ) dependence, butthere are significant differences in magnitude from site to site.

FIGURE 19. Concentrations of fungal spores versus time atFort Leonard Wood, Missouri. The green curve gives fungalspores that were collected by a Burkard spore trap andmanually counted under a microscope. The red curve givesthe particles measured by a BAWS specifically tuned to de-tect fungal spores. Note the generally good agreement be-tween the two sensing techniques.

1010

10–5 10–4 10–3

Frequency (Hz)

10 min1 min

10 sec

Atlanta, GeorgiaFort McClellan, AlabamaNaval Surface Warfare Center,Maryland

Fort Leonard Wood, Missouri (Oct.)Fort Leonard Wood (Apr.)Dugway, Utah

1 hr1 dayPow

er s

pect

ral d

ensi

ty

(par

ticle

/lite

r)2 /

Hz

10–2 10–1

108

106

104

102

100

10–2

10/7/980

50

100

150

200

250

300

10/9/98

Time

Par

ticle

s/lit

er

10/11/98 10/13/98

BAWS III

Fungal spore count



sensor’s detection probability and false-alarm rate. Inreal field trials at Dugway Proving Grounds we foundthat the BAWS sensor had essentially perfect detec-tion with no false alarms; but Dugway ProvingGrounds is a desert site, with rather benign back-ground, and human activity is restricted during tests.We do not have the capability to release bioagentsimulants at the other sites where we conducted back-ground measurements, but we can combine the back-ground measurements with the Dugway trial data tosimulate sensor performance at sites with more stress-ing background conditions.

Figure 21 shows performance curves generated byNathan R. Newbury by combining background datawith Dugway trial data to simulate a 120-particle/li-ter anthrax attack. We plot probability of detectionversus false-alarm rate for desert, rural, and urbansites. These calculations were done for a fairly simpledetection and discrimination algorithm, as imple-mented in the BAWS three-channel sensor. We see

FIGURE 20. Background measurements with the BAWS at Fort Leonard Wood, showing the variability with human activity. Theupper plot from 9 October 1998 (Friday) shows large fluctuations resulting from various activities near the sensor. The lower plotfrom 11 October 1998 (the Sunday of the Columbus Day weekend) shows a low, flat background in the absence of human activ-ity at the site.

FIGURE 21. Probability of detection versus false-alarm ratecalculated for the BAWS III for three different locations. ABacillus anthracis attack at a concentration of 120 particles/li-ter is assumed. The curves were generated by using mea-sured Bacillus globigii releases at Dugway Proving Groundsand background measurements at each of the sites. Perfor-mance is perfect at the Dugway site (100% detection, nofalse alarms) and excellent at the other sites (100% detectionat less than one false alarm per day). With more backgrounddata collection and more sophisticated alarm algorithms thefalse-alarm rate can be further improved.

11:00

9 October 1998 (Friday)

Fluo

resc

ent p

artic

les/

liter

11:30

Tru

cks

Scr

aper

Expl

osiv

es

8 tr

ucks

12:000

2000

1500

1000

500

12 b

ackh

oes

Tro

op tr

ansp

ort

Tra

ctor

trai

ler

Dum

p tr

uck

10:30

11 October 1998 (Sunday)

11:00 11:30 12:000

500

100

80

60

40

20

043210

False-alarm rate/day

Dugway Proving Grounds, Utah (desert)

Atlanta, Georgia (urban/light industrial)

Fort Leonard Wood, Missouri (army base )

Pro

babi

lity

of d

etec

tion

(%)



that, consistent with the Dugway trial data, perfor-mance is perfect at the desert site—100% detectionwith zero false alarms. At the urban and rural sites,performance is somewhat worse but still good: we get100% detection with a false-alarm rate of less thanone per day at the urban Atlanta and rural FortLeonard Wood sites.

With increased background data collection andfurther refinements in the alarm algorithm we expectto decrease the false-alarm rate for the BAWS sensorto less than one per week with essentially 100% de-tection, in even the most stressing backgroundconditions.

Bioelectronic Identifying Sensor

The BAWS fluorescence sensor is an early-warningsensor: it rapidly detects and discriminates threataerosol particles, but is does not identify the bioagent.Thus, there is need for a downstream identifying sen-sor. This sensor would be dormant until triggered bythe continuously operating BAWS. It would then pre-cisely identify the bioagent to confirm the attack andguide medical response.

Todd H. Rider at Lincoln Laboratory invented abioelectronic identifying sensor that combines livingcells with microelectronics [17]. The genesis of thissensor was his observation that white blood cells,which detect pathogens within human (and otheranimal) bodies, respond more rapidly, sensitively, andspecifically than currently available man-made sen-sors. This observation led to the approach of usingwhite blood cells as the basic sensing elements and in-terpreting them with an optoelectronic system to pro-vide rapid signal readout.

We use a type of white blood cell known as a Blymphocyte; these B cells can be harvested from miceand then grown for many years in an artificial envi-ronment, given the appropriate culture medium andambient conditions. B cells have the property thattheir surfaces are covered with antibodies. In nature,these surface antibodies detect a pathogen by bindingto its antigens (surface features); this binding thentriggers biochemical reactions in the B cell. It is thisnatural antibody/antigen binding and biochemical-reaction sequence that we make use of in our B-cellsensor.

The B-cell sensor requires two critical pieces of ge-netic engineering to make it into a bioagent identifier.First, we engineer the B cells so they have antibodiesthat are specific to certain bioagents. Second, we addthe aequorin gene to the B cells. Aequorin is a biolu-minescent protein found in a species of glowing jelly-fish. Adding it to the B cells makes them glow whenthey detect a bioagent. This work is done in conjunc-tion with Jianzhu Chen of MIT.

Figure 22 illustrates the basic operation of the B-cell sensor. The genetically engineered B cells are at-tached to a transparent substrate; behind the sub-strate is a CCD array. A liquid sample containingpossible bioagents is injected into the cell culture liq-uid surrounding the B cells. If a bioagent particlewith the proper antigen drifts by, it binds to the sur-face antibodies on the B cell. Binding of the bioagentto the B cell’s antibodies triggers a cascade of bio-chemical reactions within the B cell. This cascade am-plifies the single binding event and results in the re-lease of many Ca2+ ions in the B cells. The Ca2+ ionsthen induce the aequorin to emit photons. Thus, a Bcell that detects a bioagent will emit light. This light

FIGURE 22. Principles of a bioelectronic identifying sensorusing living B cells. B cells are genetically engineered sothat their surface antibodies recognize bioagent antigens.The aequorin gene (which is what makes certain jellyfishglow) is also added to the B cells. When a bioagent antigenattaches itself to the engineered B cell, a biochemical ampli-fication process results in the B cell emitting light. This lightcan then be detected by a CCD array.

B cell withadded aequorin

gene

Cellcultureliquid

B-cell antibodiesrecognize bioagentantigens

Transparent substrate

CCD array

Bioagents injectedinto cell culture liquid

(1)

(2)

(3)

(4)

(5)

Biochemical signalamplification inB cell releases Ca2+

Ca2+ makes aequorinemit photons

CCD-array detectsglowing cells



FIGURE 24. Measured response of B cells that were geneti-cally engineered to respond to phosphorylcholine (PC)-ovalbumin. The red curve shows the large, rapid responsewhen the B cells are exposed to the PC-ovalbumin. Theother curves show the lack of response when the B cells areexposed to similar substances and to other possibleinterferents.

can then be detected by an element of the CCD array.Finally, an electric output from the CCD array willsignal a bioagent detection.

Our calculations indicate that a single bioagentparticle detected by a single B cell should induceenough light to be detected by the CCD. Thus, theB-cell sensor is potentially a single-particle detector. Itshould be possible to engineer different B cells to de-tect the entire panoply of bioagents—bacteria, vi-ruses, rickettsiae, fungi, and proteinaceous toxins.

An overview of a possible bioelectronic-sensorimplementation is shown in Figure 23. There mightbe 100 patches of B cells over a 1-cm × 1-cm CCDarray. Each 1-mm × 1-mm patch would containthousands of B cells. Each patch could be engineeredto respond to a particular bioagent. Thus, light on aparticular area of the CCD array would preciselyidentify that a specific bioagent had been detected.

The compact structure shown in Figure 23 will de-pend in large measure on microfluidic technology.(See the sidebar entitled “Microfluidic Technology.”)Microfluidics will be used to flow the liquid samplecontaining potential bioagents through the B-cellsensor. Microfluidics will also flow nutrients to the Bcells to keep them alive and remove B-cell waste prod-

ucts. A microfluidic bioprocessor can be used to filterand purify the sample before it is sent to the B-cellsensor. This filter helps assure that the B-cell sensordoes not get clogged with impurities.

FIGURE 23. Schematic illustration of a bioelectronic identifying sensor. A microfluidicbioprocessor takes in a sample and purifies it for presentation to the measurementsection. The measurement section comprises many patches of B cells, with the cells ineach patch engineered to detect a particular bioagent. The bioagent is identified by ob-serving which CCD element senses light output. Microfluidics in the measurementsection brings nutrients to the B cells and carries waste products away.

CCD array

Microfluidicbiofilter B-cell sensor

2 cm

1.5 cm

B cells

3.5

3.0

2.5

2.0

1.5

1.0

0.5

00

PC-ovalbuminPCOvalbuminOther:

50 100

Time (sec)

105

phot

ons/

sec

150 200

Bovine serum albuminB. subtilisE. coliYeastPhage M13Sand

Stimulantinjected

Peak/bkgd = 130

Peak/bkgd = 4Peak/bkgd = 1



The B-cell identifying sensor is in an early stage ofdevelopment, but feasibility tests conducted thus farhave been encouraging. We have performed tests todemonstrate both the rapid time response and thespecificity of the basic B-cell reaction. We performedthese tests with B cells that have antibodies specificfor phosphorylcholine(PC)-ovalbumin, a nonpatho-genic molecule. Martha S. Petrovick genetically engi-neered these cells by inserting the aequorin gene. Wethen tested the response of the engineered B cells byinjecting them with PC-ovalbumin.

Results obtained by Rider are shown in Figure 24.We see that the response is very rapid: peak light out-put occurred in about thirty seconds. The measuredlight output was more than one hundred times thebackground level—an easily detectable signal.

To test the specificity of the B-cell detection pro-cess, we exposed the cells to various other substances.These results are also shown in Figure 24. There wasvery little response when the B cells were exposed tojust PC or ovalbumin. There was no measurable re-sponse to bovine serum albumin, which is similar toovalbumin. We also exposed the B cells to commoncontaminants that might be present in environmentalsamples: Bacillus subtilus (a common bacterium), E.coli (another common bacterium), yeast, M13 (a vi-rus), and sand. There was no measurable response toany of these substances. We conclude that the B-cell

reaction is highly specific to PC-ovalbumin.We have also performed a proof-of-principle test of

the planned sensor configuration illustrated earlier inFigure 23. Albert M. Young generated a six-site mi-crofluidic test chip, as shown in Figure 25. Each cellsite in the chip is a 2-mm × 2-mm glass surface withinthe microfluidic channel. In the test shown in Figure25, four of the six cell sites were loaded with a mono-layer of B cells engineered to respond to anti-IgM.Fluid containing anti-IgM was then flowed throughthe microfluidic test chip and the test chip was ob-served with a camera. Figure 25 shows that onlytwenty seconds after the fluid was injected, the fourB-cell sites were glowing brightly.

Many questions remain to be answered before apractical B-cell identifying sensor can be fielded. Forinstance, can the patches of B cells be kept stable andisolated on the transparent substrate? Can the B cellsbe stored for extended periods before use? Can the Bcells be kept alive long enough in deployed sensors?We are working to demonstrate positive answers tothese questions and to develop alternative solutions incase some of the initial answers are negative. For in-stance, it appears from early tests that B cells can bemade to stick well enough to glass substrates, but incase they do not Laura T. Bortolin came up with theidea of using fibroblast cells instead of B cells for thebioelectronic sensor [18]. All results to date show that

FIGURE 25. Microfluidic test chip for a bioelectronic B-cell sensor. (a) The microfluidic flow channel and six 2-mm × 2-mm cellsites along the flow channel. The lower four sites were loaded with a monolayer of B cells. (b) The four B-cell sites glowingbrightly 20 sec after the appropriate simulant bioagent was injected.

(a) (b)



, asimplied by its name, is technologyfor producing integrated submil-limeter fluidic networks for thetransport, manipulation, and re-action of liquids. Microfluidics isperhaps best described in analogyto microelectronics. Microelec-tronics involves the integration ofmany electronic components—transistors, diodes, resistors,etc.—in compact, rugged, easilyproducible packages. Likewise,microfluidics involves the integra-tion of many fluidic compo-nents—valves, pipes, mixingchambers, etc.—in compact rug-ged, easily producible packages.Microelectronic technology has

produced a revolution in com-puter and communications tech-nology and, concomitantly, revo-lutionary changes in people’s lives.Microfluidics may not alter theworld as dramatically as micro-electronics has, but we expect thatmicrofluidics will bring revolu-tionary changes to biochemical,biological, and medical process-ing fields.

For bioagent sensors, micro-fluidic structures offer three po-tential advantages over more con-ventional fluid handling. First,they are compact and rugged,making them ideal for field sen-sors. Second, they are inexpensiveto manufacture; thus, they will

not dominate overall sensor costand, indeed, might be consideredas disposable components in sen-sors. Finally, they enable measure-ments to be made with tinyamounts of fluids (perhaps nano-liters or even picoliters); a sensorthis abstemious with consumablesreduces routine maintenance re-quirements and overall life cyclecosts.

Albert M. Young, Theodore M.Bloomstein, and others at LincolnLaboratory have developed anovel microchemical etching pro-cess to produce microfluidic com-ponents [1]. This process is illus-trated in Figure A. Siliconsubstrates are etched in a chlorineenvironment by focusing an ar-gon-ion laser beam to locally meltthe silicon. In this melting processsilicon chlorides are formed; theserapidly diffuse from the surface,minimizing local particulate for-mation and leaving a clean etch.Computer control of the laserpointing and focus and of the sili-con-substrate position allows usto produce precision three-di-mensional laser-micromachinedcomponents.

Once the silicon substrate hasbeen etched, it can be used as amaster mold to replicate the mi-crofluidic structure cheaply andaccurately. Replica componentshave been generated in a three-stage process. First, we fluorinate

FIGURE A. Microfluidics fabrication process. Submicron structures areformed in silicon by using chlorine-assisted laser etching. The silicon pieceserves as a master mold from which a negative elastomer mold is generated.Finally, inexpensive polymer replicas of the originally silicon structure areformed from the negative elastomer mold.

M I C R O F L U I D I C T E C H N O L O G Y

Master mold in Si

Negative elastomer mold

Lasermicrofabrication

Laserbeam

SiCI4CI2

Si Si Si

Si SiSi Si

Si SiSi Si Si SiSi Si

Si SiSi Si Si Si SiSi Si

CI–

CI2 CI2

CI2

SiCI2

SiCI4CI2

CI–



FIGURE B. Microfluidic biofilter. This microfluidic piece is designed to pre-fil-ter biological particles before they are sent to an identifying sensor. The mi-crofluidic biofilter is an extreme miniaturization of a conventional water-treat-ment plant.

the silicon (with trideca-fluoroctaltrichlorosilane). Next,we generate a negative mold bycasting an elastomer (polydi-methylsiloxane) over the etchedsilicon surface. Finally, we gener-ate a positive copy by casting oneof a variety of polymers over thenegative elastomer mold. We havedemonstrated this replicationprocess to an accuracy of less than1 µm. The elastomer negatives canbe reused, and the final polymerstructures are rugged and cheap.Since the polymers are opticallytransparent, it is even possible toobserve the flow dynamics inthese microfluidic structures.

An example of a microfluidicstructure fabricated by A. Youngas part of the bioagent-detectionprogram is shown in Figure B.This microfluidic component isdesigned to take in clumped bio-logical particles that may be mixedwith dirt or other contaminants,and, through a combination ofchemical and mechanical pro-cesses, output declumped, puri-fied biological particles. This mi-crofluidic component featuresintegrated microvalves, a multi-

element turbulent microdis-perser, several filter units, an en-zyme storage tank, and severalchemical reaction chambers. Allthese microcomponents are con-tained in a chip that is roughly5 mm × 8 mm.

The microfluidic structure inFigure B can be regarded as an ex-treme miniaturization of a wastetreatment plant. The microfluidicstructure does not have nearly the

throughput of the waste-treat-ment plant, but it employs thesame basic chemical and mechani-cal processes.

Reference1. A.M. Young, T.M. Bloomstein, S.T.

Palmacci, and M.A. Hollis, “Elastic-Membrane Microvalves for Microflu-idic Network Integration,” Solid StateResearch Report 1997:4, LincolnLaboratory (23 Apr. 1997), DTIC#A342325.

the B-cell concept has the potential to develop into arapid, sensitive, specific identifying sensor.

Future Bioagent Sensor Vision

Based on our technical progress thus far, we can easilyenvisage the sensor system shown in Figure 26. Thesystem would include a network of miniature, low-power sensors. Each sensor would include a BAWSearly-warning sensor and a B-cell identifying sensor,as well as Global Positioning System (GPS) and auxil-

iary sensors. The fully integrated sensor would beabout the size of a large lunch box so it could easily betransported or mounted on a variety of platforms.

Information from the sensor would be instantlytransmitted to a sophisticated data-fusion system andsituational display, all housed in a laptop. With thissystem, a military chemical/biological officer or a ci-vilian first-responder could interact with a network ofstationary and mobile sensors and rapidly form an in-tegrated picture of a bioagent attack. Such capability

Water treatment plant

Microfluidic biofilter

~8 mm

Declumped,purified

bioparticlesexit

Clumpedbioparticlesmixed withdirt enter

~5 mm



Acknowledgments

The research described in this article was conductedwhile I was program manager for Biological WarfareDefense at Lincoln Laboratory. The work was doneby a large number of talented researchers from manydifferent Divisions at Lincoln Laboratory. I thank ev-eryone who contributed to the program, and I par-ticularly acknowledge the following key researchers:

Roshan L. Aggarwal, Theodore M. Bloomstein,Robert P. Brady, Vincenzo Daneu, Salvatore DiCecca, Herbert W. Feinstein, Darryl P. Greenwood,Mark A. Hollis, Thomas H. Jeys, Bernadette John-son, Robert Kramer, Jae H. Kyung, Michael T. Lan-guirand, Robert W. Miller, Frances Nargi, Michael E.O’Brien, Nathan R. Newbury, Lalitha Parameswaran,Ronald R. Parenti, Martha S. Petrovick, Todd H.Rider, Gregory S. Rowe, Ann E. Rundell, AntonioSanchez-Rubio, Shen Y. Shey, Laura T. Bortolin,David P. Tremblay, Albert M. Young, and JohnZayhowski.

I acknowledge Professor Jianzhu Chen and his col-leagues in the MIT Department of Biology for theirsignificant contributions to the B-cell based sensor. Ialso acknowledge Professor Harriet Burge and hercolleagues at the Harvard School of Public Health fortheir significant contributions to the aerosol back-ground measurements.

This research was supported by the U.S. Army, bythe Defense Advanced Research Projects Agency(DARPA), by the Joint Program Office for BiologicalDefense, and by the Office of the Secretary of De-fense. In the government sponsor community I par-ticularly acknowledge Richard Smardzewski, FelixReyes, and David Sickenburger from U.S. ArmySBCCOM; not only did they provide funding for theBAWS sensor development, but they also were activeparticipants in field tests involving the BAWS sensor.

I thank Darryl P. Greenwood, Mark A. Hollis, andShen Y. Shey for providing helpful editorial com-ments on this article. Finally, I thank my secretary,Marjorie Banks, for her invaluable assistance in as-sembling this article.

FIGURE 26. Possible future bioagent sensor system. Thisfigure shows a network of integrated miniature low-powersensors. Each sensor would incorporate both an early-warn-ing sensor and an identifying sensor.

could provide a significant technology advantage tothe defensive side, helping to tip the balance from thecurrent situation, where it appears that even a low-tech attacker has the advantage if he chooses to usebiological weapons.



R E F E R E N C E S1. R.A. Falkennath, R.D. Newman, and B.A. Thayer, America’s

Achilles Heel: Nuclear, Biological and Chemical Terrorism andCovert Attack (MIT Press, Cambridge, Mass., 1998).

2. G.W. Christopher, T.J. Cieslak, J.A. Pavlin, and E.M. Eitzen,“Biological Warfare: A Historical Perspective,” J. Am. Med. A.278 (5), pp. 412–417.

3. P. Williams and D. Wallace, Unit 731: Japan’s Secret BiologicalWarfare in World War II (Free Press, New York, 1989).

4. K. Alibeck with S. Handelman, Biohazard (Random House,New York, 1999).

5. Israel is a significant exception. Lists of signatories and non-signatories are maintained by the SIPRI Chemical and Bio-logical Warfare Project and may be found at the web page<http://projects.sipri.se/cbw/cbw-sipri-bradford.html>.

6. Health Aspects of Chemical and Biological Weapons (WorldHealth Organization, Geneva, Switzerland, 1970).

7. U.S. Congress, Office of Technology Assessment, Proliferationof Weapons of Mass Destruction: Assessing the Risks, OTA-ISC-559 (U.S. Government Printing Office, Washington, Aug.1993).

8. The Australia Group list may be found at the web page<http://www.dtic.mil/mctl>, under section 3—BiologicalTechnology.

9. Solid State Research Quarterly Technical Report 1998:1, Lin-coln Laboratory (19 June 1998), DTIC #A347651.

10. Solid State Research Quarterly Technical Report 1998:2, Lin-coln Laboratory (28 Oct. 1998), DTIC #A355512.

11. J.J. Zayhowski, “Passively Q-Switched Processor Microlaser,”U.S. Patent No. 5,394,413 (28 Feb. 1995).

12. J.J. Zayhowski, “Passively Q-Switched Microchip Lasers andApplications,” Rev. Laser Eng. 26 (12), 1998, pp. 841–846.

13. J.J. Zayhowski, “Q-Switched Microchip Lasers Find Real-World Applications,” Laser Focus World 35, Aug. 1999, pp.129–136.

14. J.C. Bloch, B. Johnson, N. Newbury, J. Germaine, H.Memond, and J. Sinfield, “Field Test of a Novel Microlaser-Based Probe for in Situ Fluorescence Sensing of Soil Con-tamination,” Appl. Spectrosc. 52 (10), 1998, pp. 1299–1304.

15. L.A. Condatore, Jr., R.B. Guthrie, B.J. Bradshaw, K.E. Logan,L.S. Lingvay, T.H. Smith, T.S. Kaffenberger, B.W. Jezek, V.J.Cannalito, W.J. Ginley, and W.S. Hungate, “U.S. Army Sol-dier and Biological Chemical Command Counter Prolifera-tion Long Range–Biological Standoff Detection (CP LR-BSDS), SPIE 3707, 1999, pp. 188–196.

16. B.J. Westand and M.F. Shlesinger, “On the Ubiquity of 1/fNoise,” Int. J. Mod. Phys. B 3 (6), 1989, pp. 795–819.

17. T.H. Rider, “Optoelectronic Sensor,” U.S. Patent ApplicationSerial No. 08/987,410.

18. T.H. Rider and L.T. Smith, “Continuation in Part of Opto-electronic Sensor,” U.S. Patent Application Serial No. 09/169/196.



. received an A.B. degree inphysics from Duke Universityin 1968, and S.M. and Ph.D.degrees in nuclear engineeringfrom MIT in 1970 and 1975,respectively. Immediatelyfollowing graduate school hejoined Lincoln Laboratory towork on adaptive optics, laserpropagation, and high-energy-laser beam control. In 1984when SDIO was formed, Dr.Primmerman was on loan tothe Directed Energy Office,where he helped to establishnew programs involving high-energy lasers for boost-phaseintercept of ballistic missiles.In 1985 Dr. Primmermanreturned to Lincoln Labora-tory to become group leader.Until the fall of 1999 hedirected a group involved inthe development and fieldtesting of advanced adaptive-optics and other electro-opticalsystems for high-energy-laserand laser-radar applications.Dr. Primmerman is particu-larly known for his pioneeringwork in the use of adaptiveoptics to compensate foratmospheric distortions. Forthis work he was corecipient ofthe 1993 SPIE TechnologyAchievement Award and in1998 was named a Fellow ofthe Optical Society of America(OSA). In 1995 Dr.Primmerman helped initiate anew effort at Lincoln Labora-tory involving the real-time

detection of biological agents.He served as overall ProgramManager for biological-warfaredefense activities at LincolnLaboratory until the fall of1999. In the fall of 1999 Dr.Primmerman took a leave ofabsence from Lincoln Labora-tory to take a temporaryposition in Washington withthe Advanced Systems Con-cepts Office of the DefenseThreat Reduction Agency. Inhis position as Special Assis-tant to the Director, Dr.Primmerman helps both toanalyze emerging weapons-of-mass-destruction (WMD)threats and to developcounters to them. As escapesfrom the WMD world Dr.Primmerman enjoys downhillskiing and sailing his J/24.

detection of biological agents

Documents