Invited Paper
Detection of fluorescence spectra of individual bioaerosol particles Per Jonsson*a, Fredrik Kullandera, Pär Wästerbyb, Mikael Tiihonenc and Mikael Lindgrena,d a
FOI - Swedish Defence Research Agency, Sensor Technology, PO Box 1165, SE-581 11 Linköping, Sweden b FOI - Swedish Defence Research Agency, NBC Defence, Cementv. 20, SE-901 82 Umeå, Sweden c Dept. of Physics, Royal Institute of Technology (KTH), SE-106 91 Stockholm, Sweden d Dept. of Physics, Norwegian University of Science and Technology, NO-7491 Trondheim, Norway ABSTRACT We present initial results of a measurement system designed for detecting the fluorescence spectrum of individual particles of biological warfare agent (BWA). A compact optical parametric oscillator with intracavity sum-frequency mixing and a commercial Nitrogen gas laser was used as excitation sources to generate 293 nm or 337 nm UV laser irradiation. The pulsed lasers and a photomultiplier tube (PMT) array based spectrometer were triggered by a red laserdiode and a PMT detector that sensed the presence of a particle typical of size 5-20 µm in diameter. The spectral detection part of the system consisted of a grating and a PMT array with 32 channels, which measured fluorescence in the wavelength from 280 nm to 800 nm. The detector system was used to demonstrate the measurement of laser induced fluorescence spectra of individual BWA simulant particles by excitation of single UV laser pulses. The spectrum obtained by averaging spectra from several BWA aerosol simulant particles were found generally similar, but not identical, to the fluorescence spectrum obtained from water solutions containing the same particles dissolved. Keywords: point-detection, biosensing, bioaerosol detection, ultraviolet laser excitation, fluorescense spectroscopy
1. INTRODUCTION In general it is difficult to detect hazardous levels of biological warfare agent (BWA) since very low doses of the agents can cause disease for humans. Genetic methods seem to be very promising, especially since they are precise and becoming feasible with the development of micro-labs. Still, this technology is expensive; it requires “wet-chemistry” to some extent and can be time-consuming if one does not know what to look for. To achieve an overall quick and secure response time, many BWA detection systems are based on an initial trigger/warning detector that continuously monitor some inherent property of the dry aerosol particles, e.g. their size, shape, intrinsic fluorescence and elemental composition. Such aerosol particle detectors have a relatively low specificity and can therefore give rise to false positive alarm for natural non-pathogenic bacteria, or other fluorescent compounds such as diesel exhaust, in the air. Examples of such trigger detectors are the Aerosol Size and Shape Analyzer (ASAS) system that uses the aerosol shape as one of the key parameters [1-3], the Fluorescence Aerodynamic Particle Sizer (FLAPS) system that uses particle fluorescence [4] and the Biological Alarm Monitor (MAB) which measure the elemental decomposition by flame spectrophotometry. These detectors can reach close to real-time warning. Multiple excitations (266 and 355 nm) for the detection of bioaerosols have been reported [5] as well as the spectral characteristics of fluorescence from particles excited at 266 nm [6-8]. In order to increase the specificity and lower the false positive alarms, we are developing an early-warning system measuring the fluorescence spectrum of individual BWA aerosol particles excited with dual wavelengths (about 290 nm and 340 nm) ultraviolet (UV) laser pulses. Biogenic chemicals associated with cell metabolism, such as reduced nicotinamide adenine dinucleotide (NADH) and riboflavin have their dominating electronic absorption at around 340 nm and the resulting fluorescence peaks between 450 nm and 560 nm [9]. Aromatic amino acids, such as tryptophan, tyrosine and phenylalanine absorb light at 280290 nm and they emit in a band between 300 nm and 400 nm [9]. Furthermore, dipicolinic acid, present in spores, has *
[email protected], Phone +46-13-37 8578, Fax +46-13-37 8066, www.foi.se Optically Based Biological and Chemical Sensing, and Optically Based Materials for Defence, edited by John C. Carrano, Arturas Zukauskas, Anthony W. Vere, James G. Grote, François Kajzar, Proc. of SPIE Vol. 5990, 59900M, (2005) · 0277-786X/05/$15 · doi: 10.1117/12.630141 Proc. of SPIE Vol. 5990 59900M-1
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2014 Terms of Use: http://spiedl.org/terms
its maximum absorption around 340 nm with an associated emission-band that peaks between 400 and 450 nm [10]. For this reason there is an immense effort to develop novel light sources that can be used for fluorescence spectroscopy and sensing devices in these spectral ranges. Particularly for the UV-B and UV-A bands (280-400 nm) that allow direct excitation of such photoactive molecules pertinent to most biological compounds. The development of photonics detector technologies has matured into a variety of detectors and detector systems with single photon sensitivity and a very low background noise. Avalanche photodiodes as well as photomultiplier tubes are also becoming feasible as array detectors, and together with grating devices it allows spectral detection from one read-out frame with hitherto unprecedented sensitivity. This is particularly relevant for sensing applications in the visible and ultraviolet spectral regions where background is relatively weak. The route we and others are pursuing for detection of biological warfare agents in terms of dried aerosols or dissolved in some water based denaturing reagent is based on laser induced fluorescence at several excitation wavelengths with spectrally resolved detection [11-15]. This method can provide a real-time warning and even classify aerosol particles. We have previously demonstrated the successful use of a novel UV laser at different wavelengths around 290 nm and 340 nm to excite molecular aggregates such as proteins and bacteria in solutions [12-14], as well as to obtain fluorescence spectra of individual bioaerosol spores [15]. Moreover, multivariate data analysis (principal component analysis (PCA)) of spectral fluorescence data has been used to classify different BWA simulants and distinguish them from background compounds in field trials [16]. The ultimate goal of our work was to demonstrate a point detector for biological sensing that measures the UV induced fluorescence from single particles in air. The BWA aerosol is collected, concentrated and arranged into an aerosol beam. A red laser is focused on the aerosol beam and a trigger PMT senses the presence of individual particles by measuring the scattered light. When a particle is present in the detection volume, a laser pulse is triggered from the UV laser and the fluorescence spectrum is acquired with a detection system based on a diffraction grating and a 32 channels PMT array with single-photon sensitivity. In this report we present the initial tests of such a point detector system using our own developed laser at a wavelength of 293 nm and a commercial Nitrogen gas laser for excitation at 337 nm. The purpose of the tests was to determine the detection limit and examine the information in individual spectra compared to the ensemble average.
2. EXPERIMENTAL SETUP 2.1. Aerosol generation The particles used in this study were class 1 organisms, often used as simulants for BWA. Bacillus atrophaeus (also known as Bacillus subtilis var. niger and Bacillus globigii, red strain) and Turex (a non-pure technical preparation of Bacillus thuringiensis) are being used as simulants for pathogenic bacterial spores as for example Anthrax. Ovalbumin from chicken egg (Sigma, A-5253, Grade II) is commonly used as a simulant for toxins. We are using the common abbreviations BG, BT and OA, respectively, for the simulants. The micron sized spores of BG and BT and the molecules of OA tend to cluster into the form of larger particles. The typical diameter of the particles investigated was between 5 and 20 µm. The general setup for aerosol generation is shown in Fig. 1. The simulants were introduced to the system either from an aerosol generator (BG) or directly from the particle containers (BT and OA). Particles are sucked from the generator or containers into the pipes through the suction of the particle counters. The particle generator was used to generate BG aerosols with a controlled and stable size distribution. Details of the aerosol generator can be found elsewhere [15]. Since only one generator was available, BT and OA were injected directly from the particle containers. Therefore, the aerosol distributions with these simulants were fluctuating much more during the experiments. The aerosol injector on top of the sample cell (an open-ended quartz cuvette) directed the aerosols from the pipes into a laminar aerosol beam inside the cell. The aerosol was injected through the inner of two concentric tubes. The suction of the aerodynamic particle counter (TSI 3321 APS) connected to the bottom of the sample cell, yielded a total flow rate of 1 l/min through the sample cell. The fraction of the filtered air through the outer tube of the injector was regulated to enable a stable aerosol beam through the cell. This produced an aerosol beam of approximately 1 mm in diameter inside a protecting sheet flow of filtered air. The resulting particle speed in the aerosol beam centre was approximately 0.5 m/s.
Proc. of SPIE Vol. 5990 59900M-2 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2014 Terms of Use: http://spiedl.org/terms
A second optical particle counter (Climet 500) was connected in parallel with the sample cell. Particles ranging from 0.3 µm to 25 µm. were counted and their size measured by both counters.
Clean air for sheet flow
Particle container
Sample cell
Particle counter (Climet 500)
Aerosol generator
Particle counter (TSI 3321 APS)
Figure 1: Schematics of the particle generation.
2.2. UV lasers Two different types of lasers were used to create UV laser induced fluorescence from the particles in the experiments. The particles were excited with either 293 or 337 nm UV radiation. For excitation at 293 nm, we have developed a compact solid-state laser based on an optical parametric oscillator (OPO) with intracavity sum-frequency mixing (SFM). The OPO is pumped with a diode-pumped frequency-doubled passively Q-switched Nd:YAG laser emitting at 532 nm. The OPO consists of an input mirror, a periodical poled KTiOPO4 (PPKTP) crystal, a β-barium borate (BBO) crystal, a dichroic mirror and an output mirror as shown in Fig. 2. By changing the grating period of the PPKTP crystal it is possible to reach any wavelength between 280 and 350 nm. In our experiments we used a grating period of 12.77 µm, resulting in a signal at 650 nm (idler at 2.9 µm). The nonlinear SFM process between the pump and the signal creates UV radiation at 293 nm, which is coupled out of the cavity by the dichroic mirror. Further details of this laser can be found elsewhere [16, 18]. To facilitate simple switching between the excitation wavelength (without changing the PPKTP crystal and realign the laser), we used a commercially available nitrogen gas laser (Oriel 79111) for excitation at 337 nm.
Ml
4I
U V-output
PPKTP BBO
M2
OPO signal
Figure 2: Schematics of the optical parametric oscillator with intracavity sum-frequency mixing.
The maximum pulse energy of the OPO laser at 293 nm is around 30 µJ, the maximal pulse repetition rate is 100 Hz and the pulse length is approximately 2 ns. Due to the passive Q-switching in the pump laser, there is a time-delay of about 240 µs from trigging a pulse to the UV pulse output. The maximum pulse repetition rate of the nitrogen gas laser is 50 Hz. The maximum pulse energy and the pulse length, which depends of the repetition rate, are approximately 300 µJ and 5 ns, respectively. The time-delay from trigger to UV pulse output is less than 0.5 µs for the nitrogen gas laser. The UV radiation from both lasers was coupled to the aerosol sample cell via a silica fibre (Oriel 77681). The diameter of the fibre core was 0.4 mm, the numerical aperture (NA) was 0.22 and the length 1 m. Due to inefficient fibre coupling
Proc. of SPIE Vol. 5990 59900M-3 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2014 Terms of Use: http://spiedl.org/terms
and other losses, only 1-10% of the laser pulse energy is impinging on the aerosol beam. The value of the impinging pulse energy is stated for each experiment in the result section.
2.3. Optical setup A top view of the optical setup used in the experiments is shown in Fig. 3. The air beam containing the aerosol was centred in the sample cell. Around the sample cell were two laser sources and two collection probes mounted. The foci of the lasers and the collection probes were aligned to overlap the aerosol beam and each other horizontally and vertically. The trigger laser (Flexpoint-76/3TAE) was a laser diode with a wavelength of 670 nm. This laser could be modulated up to 100 kHz with a maximal output power of 3 mW. The focus of the beam was adjusted with the in-build lens set resulting in a diameter being smaller than 0.5 mm in diameter. The purpose of this laser was to generate scattered light when particles passed the focus. A fraction of this scattered light was collected by an optical probe focusing the light into a fibre bundle, which directed the light onto a PMT detector (Hamamatsu H5783-01) used to trigger the other systems. When the signal from this trigger detector reached a preset threshold, the trigger laser was first turned off and sequentially a pulse from the UV laser as well as the data acquisition from the PMT array were triggered. Details of this sequence are found in Sect. 2.4. The UV laser pulse, ignited by the presence of particles in the focal volume of the trigger laser, was focused onto the same spot as the trigger laser with a focusing probe (Oriel 77646, NA=0.22). The diameter of this focus was slightly less than 1 mm. The main part of the UV radiation was scattered by the particles in the focal volume, but some of the energy is absorbed by the aerosol particle to generate fluorescence. A fraction of the scattered radiation and the fluorescence were collected by the identical collection probes (Oriel 77646, NA=0.29) connected to 1 m long fibre bundles connected to the trigger detector and the spectrometer, respectively. The bundle connected to the trigger detector (Oriel 77561) has a diameter of 1.6 mm and NA=0.27 resulting in a diameter of the collection cone in the detection volume of about 2 mm. The bundle connected to the spectrometer (Oriel 77575) has the same diameter but NA=0.22, resulting in collection cone diameter around 1.5 mm. A colour glass filter was placed between this fibre bundle output and the spectrometer to block the scattered UV radiation while transmitting the fluorescence. Spectrograph PMT array Readout module 32 channels To computer
Laser blocking filter
To computer
Fibre bundle (1 m, D=1.6 mm, NA=0.22)
UV Laser Focusing probe (collection) Fibre (1 m, D=0.4 mm, NA=0.22) Focusing probe (excitation)
To computer
Sample cell
Trigger laser
Focusing probe (collection) To computer
Fibre bundle (1 m, D=1.6 mm, NA=0.27)
Trigger PMT
Figure 3: Schematics of the optical setup as seen from above.
The spectrometer was built with a spectrograph (Oriel 77442) and a 32 channel linear PMT array (Hamamatsu H72604), and measured fluorescence spectra in the wavelength band from 280 nm to 800 nm. The spectrograph has a concave diffraction grating (grating density of 405 lines/mm and a blaze wavelength of 350 nm) and has a fibre-coupled input port well suited for the fibre from the collection probe. The centre wavelength λ for every channel number N was calibrated with a xenon lamp and monochromator and was found to approximately follow λ = (264.4+16.6N) nm. The magnification of the spectrometer is unity and the separation of the PMT channels is 1 mm. Since no slit was used on
Proc. of SPIE Vol. 5990 59900M-4 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2014 Terms of Use: http://spiedl.org/terms
the input of the spectrometer, the resolution of the spectrometer was basically set by the fibre bundle diameter in relation to the channel separation, resulting in a resolution of about 30 nm. For acquisition of the signal from the PMT array, a readout module was used (Hamamatsu C8678). The readout module was externally triggered, as mentioned above, allowing for charge collection using integrating capacitors. The capacitance C = 33 pF and the integration time was fixed to 1 µs. The voltage captured by the sample and hold circuit relate to the charge by U = Q/C and represent the charge collected during 1 µs.
2.3.1. Collection efficiency Following the notation from our previous publication [14], the output voltage spectrum U(λ) generated by the PMT array readout module upon trigger is given by U (λ ) = H (λ) E F (λ ) .
(1)
where H(λ) and EF(λ) are the detection system transfer function and the total emitted fluorescence energy respectively. In our system H(λ) is given by H ( λ) = η c Taq8 Tfilt ( λ)Tfib ( λ)η spec (λ ) S (λ ) .
(2)
If the fluorescence radiation is isotropic the collection probe efficiency, ηc, becomes equal to the solid angle fraction of 4π opened up by the collection probe. Hence, ηc = sin2(α/2), where 2α is the acceptance cone angle of the collection probe. When the numerical aperture NA of the collection fibre bundle sets the limit, as is our case, the acceptance cone angle is given by α = arcsin(NA). Taq, Tfilt(λ) and Tfib(λ) are the transmittance for an air to quartz boundary, the colour filter and the fibre, respectively, ηspec is the spectrometer efficiency and the grating efficiency and S(λ) is the overall sensitivity of the PMT receiver in units of V/J. S(λ) is dependent on the adjustable PMT high voltage level. In our experiments a high voltage (HV) of -800 V (gain =1.4×106) was used. The spectral characteristics, apart from the colour filter edges, come mainly from the spectrometer grating efficiency and the PMT cathode radiant sensitivity. The NA of the fibre is 0.22 yielding ηc = 0.012 and Taq = 0.96 is assumed (yields Taq8 = 0.72). For the other parameters the manufacturers specifications have been used. The accumulated transmission factor in the fluorescence collection at the peak wavelength of 370 nm or 400 nm, depending on filter, amounts to approximately 4.5×10-3 of which a factor of 0.012 is due to the collection probe efficiency. Compared to our previous publication [14], we have improved the transmission by a factor of four, which is mainly a consequence of removing the slit on the input of the spectrograph. A drawback of this removal is the lower resolution of the spectrometer. The equivalent RMS voltage noise level noise for the C8678 PMT readout module is sU= 2 mV with the PMT HV being -800 V. This specified value correspond to a noise equivalent level for the fluorescence energy going from 0.3×10-18 J to 1×10-18 J between 300 and 600 nm which is comparable to the single photon energy, around 0.5×10-18 J. The threshold level for the fluorescence emission per wavelength bin, EF(λ), is 2×10-16 J at the peak wavelength and below 1×10-15 J in a wavelength band from 320-600 nm or 370-600 nm depending on the choice of filter. As will be shown in the section with the results, it was not possible to reach down to this detection level due to a high level of background fluorescence throughout the measurements. The background fluorescence from the sample cell was on the order of 5×10-14 J and fluctuated with a standard deviation of about 2×10-14 J. Consequently, the practical detection limit became almost two orders of magnitude higher than achievable with the hardware of the arranged detection system.
2.3.2. Fluorescence energy spectrum The fluorescence energy spectrum as defined in Eq. (1) is related to the excitation energy by the fluorescence cross section. Restricting to situations where the level of irradiation is sufficiently low that non-linear effects can be neglected and that no appreciable depletion of the number of molecules in the ground state is produced we have
Proc. of SPIE Vol. 5990 59900M-5 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2014 Terms of Use: http://spiedl.org/terms
λ + ∆λ 2
E F (λ ) ≅ n
E exc F σ (λ L ) LF ( λ ′) dλ ′ Aexc λ − ∆λ 2
∫
(3)
following the nomenclature of Measures [19]. A more accurate expression than this is generally not required for this purpose. We assume here that only one species contributes to the fluorescence signal. Eexc represents the total energy of the excitation light pulse, approximated to be flowing with uniform spatial distribution through a constant beam cross section denoted Aexc, n represents the number of fluorescing entities (cluster of spores, spores, cells or molecules depending on species) within the probe volume, V, cut out by the intersection of the excitation beam and the field of view of the collection probe. When we treat the sheet flow air beam the aim is to detect single particles for which n = 1 if a particle (possibly clusters of spores) is the entity being connected to the fluorescence cross section. Then σF(λL) represents the spectrally integrated fluorescence cross section attributed with one entity and provides a direct relationship between the irradiance and the total emitted fluorescence power. LF(λ)dλ represents the fraction of the scattered radiation that falls into the wavelength interval (λ ,λ+dλ) where LF(λ) is the normalized fluorescence spectrum associated with the species. Finally, ∆λ is the spectrometer channel bandwidth. 2.4. Data acquisition The data acquisition system was triggered by the signal from the trigger PMT as described above. It was necessary to carefully adjust the timing of the events to assure that the PMT array data were collected in coincidence with the UV laser excitation. External timing circuits were used for this purpose. In practice, the internal time-delay of the lasers required that they were triggered in advance of the PMT array acquisition. However, the 1 µs long integration of PMT array signal started slightly before the UV laser pulse, in order to comprise the whole fluorescence signal which occurs in the ns range, partially within the excitation pulse. It is known that for similar species in solution the decay time of the fluorescence is typically in the order of 3-10 ns [14].
trigger level
Trigger PMT
Amplitude [a.u.]
A schematic of the trigger PMT signal during a data acquisition cycle and the sequential triggering events is shown in Fig. 4. Each cycle for detecting an aerosol particle and record and store its fluorescence spectrum included the following sequential events; • at time t1, the trigger PMT signal reach the trigger level and the acquisition is initiated • at time t2, the trigger laser (670 nm) is turned off • at time t3, the PMT array detector acquisition is triggered • at time t4, the UV laser pulse (293 or 337 nm) is triggered • at time t5, the trigger laser (670 nm) is turned on again
Trigger laser PMT array trigger UV laser trigger t1
t2
t3t4
t5
t
Figure 4: Schematics of the triggering sequence.
The trigger laser was turned off during the acquisition to reduce the background level. The gate time, i.e. the time the trigger laser is turned off (t5- t2), was adjusted to match with the maximum pulse repetition rate of the laser sources. The allowable rate of the data collection systems was above a kHz, i.e. well above the maximum 100 pulses/s of the lasers.
Proc. of SPIE Vol. 5990 59900M-6 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2014 Terms of Use: http://spiedl.org/terms
The minimum cycle time as set by the aerosol injection system was about 2 ms during which an aerosol at 0.5 m/s passes through the probe volume of about 1 mm. A way to prevent that fluorescence from each particle is collected more than once is simply to adjust the gate time. This was not of primary interest during these initial tests. Nevertheless, it was well taken care of by the comparably long gate time of more that 10 ms, as imposed by the laser constraints.
3. RESULTS AND DISCUSSION 3.1. Excitation at 337 nm Data from a measurement series on BG aerosols excited with 337 nm is plotted in Fig. 5 below. The left-hand plot shows the signal recorded on the trigger detector. The initial peak at each index (corresponding to an event) is due to the scattered red light from a particle that is entering the laser beam. The threshold level is on the order of 0.1 V in this case and the acquisition of the trigger detector signal is started once the signal is exceeding this level. The red laser is then turned off after 50 µs, the UV laser is triggered after 100 µs and finally the PMT array acquisition 700 ns after the UV laser pulse trigger, in order to match the trigger delay of the laser driver. As is evident from the graph the pulses have different heights which is mostly due to the different sizes of the particles although a small fraction of the particles can be assumed to be only partly illuminated. The peaks saturates at about 1.6 V due a limited amplification range. While the rising edge is related to the speed with which the particle is moving, the falling slope is solely due the discharge time for the trigger PMT through a shunt resistance of 100 kΩ. The same slope is seen for the second peaks at 100 µs originating from the UV radiation scattered by the particles. The detector is almost always saturated by this signal due to the high pulse energy of about 15 µJ from the Nitrogen gas laser.
0
0
2000
200
0
300 400
index
I-1 (js)
index
X(nm)
Figure 5: Measurement series for BG excited with 337 nm. The plot to the left shows the signal on the trigger detector and to the right the signal on the spectrometer.
In addition to start the events the trigger detector signal can be used to extract information about the particle size and velocity. Both peaks can be used but it requires that their levels are adapted to the sensitivity of the detector. If necessary, a coloured glass filter can be used to equalize the two peaks. It was the intention at these experiments to have the trigger detector signals synchronized with the spectral readout. Unfortunately, for reasons yet not understood, events were lost in the recordings and it is therefore not possible to exactly match the trigger channel data with the fluorescence spectra. However, the general trend is the same. Since the UV laser was turned of twice during this run two gaps appear in the sequences (only for the UV pulse in the left-hand graph). The trace drawn along the 200 nm line in the left graph represents the mean value of the spectra. By comparing the trigger rates with the data from the particle counter it is possible to estimate the size of the detected particles. The thin lines in Fig. 6 show the number of particles per minute in a certain particle size interval collected by the aerodynamical particle counter that was place after the sample cell (see Fig. 1). Within the depicted time frame two
Proc. of SPIE Vol. 5990 59900M-7 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2014 Terms of Use: http://spiedl.org/terms
data series with different threshold levels were acquired. The raw data from the first measurement is shown in Fig. 5 above. The trigger rate (triggers/min) is shown as a thick line to the left in Fig. 6 below. In the second measurement to the right in the figure, the trigger level was increased with a factor of about three and as a consequence the trigger rate decreased. The trigger rate, at most 15 triggers/s, was always well below our maximum rate of 50 triggers/s due to the detection cycle T = t5-t1 set to 20 ms. Hence, the mean time distance between the particles is larger than the detection cycle. If the time distance between the particles is smaller, all particles will not be detected and the trigger rate rt will underestimate the particle rate rp. We assume that the number of particles within the detection volume is following a Poisson distribution, i.e. P (n(t )) = e
− rp t
( rp t ) n
(4)
n!
where P(n(t)) is the probability that n particles are in the detection volume during an elapsed time t. The mean value of n(t) for a Poisson distribution is rpt. Therefore, the mean value of the number of particles in the detection volume during a detection cycle is rpT. Since a detection cycle always starts with a detection of a particle, the relation between the trigger and particle rates is given by rp=(1+ rpT) rt, which can be rewritten as rp =
rt . 1 − rtT
(5)
Note that the denominator cannot be negative since rt cannot be grater than 1/T. This corrected rate is plotted in Fig. 6 as a thick dotted line. Since the scattering signal increases with larger particles, we estimate that particles larger than approximately 6 µm are detectable with the lower threshold, and particles larger than about 8 µm for the higher threshold. 5
Particles and triggers per min
10
5-20 µm 6-20 µm 8-20 µm 10-20 µm Triggers Corr.
4
10
3
10
2
10
1
10
0
10 13:50
13:55
14:00
14:05
14:10
Time (hh:mm)
Figure 6: Time traces of the counted particles and trigger rates during two measurement series with different trigger threshold levels.
As seen in Fig. 5 strong fluorescence signals are detected. Unfortunately, a substantial amount of this signal is due to background fluorescence from particles stuck on the sample cell walls. The left plot in Fig. 7 shows the relation between the mean values of the measurement series described above and the background measurement performed immediately after. The background was simply measured without the particle beam present in the sample cell. The signal-to-noise ratios for the high threshold measurement series are approximately between two and four in the wavelength band 400 nm to 600 nm and it is even worse for the other measurement series. Also shown in the figure are background measurements performed with a clean sample cell (cuvette) and without this sample cell. The results shows that the sample cell itself give rise to some fluorescence but the main contribution to the background comes from contaminating particles. The right graph Fig. 7 shows the same results recalculated to energy using Eq. 1.
Proc. of SPIE Vol. 5990 59900M-8 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2014 Terms of Use: http://spiedl.org/terms
-14
0.50
10
0.45
High threshold Low threshold Background Clean cyvette Without cuvette
0.40
0.30
High threshold Low threshold Background Clean cuvette Without cuvette
9 8 7 6
0.25
EF (J)
Signal (V)
0.35
x10
0.20
5 4
0.15
3
0.10
2
0.05
1
0.00
0 350
300 350 400 450 500 550 600 650 700 750 800
400
450
Wavelength (nm)
500
550
600
650
Wavelength (nm)
Figure 7: The mean spectra from the measurements series of BG excited with 337 nm and the background (with the standard deviation). Also shown are the backgrounds with a clean cuvette and without a cuvette in the sample cell. The graph to the left shows the signal on the PMT array and to the right the calculated corresponding energy.
A fare assumption is that the total fluorescence signal increases with particle size. Therefore, we have calculated the total energy EF,tot in the detection wavelength band (360-650 nm when excited at 337 nm) for each individual spectrum. The rate of events as a function of the total energy can then be plotted in a histogram. We compare the rate from the two series above with the particle counter data (Fig. 8). For clarity the particle size histogram shows the mean value during both the series. No time information was saved with the individual spectra. Therefore, since the trigger detector data and the spectra acquisitions were not well synchronized as describe above, it is not possible to create time traces as in Fig. 6. At low total energies EF,tot, the number of spectra is clearly lower in the high threshold data compared to the low threshold data, while at the number of spectra are about the same at higher EF,tot, as expected. 5
3
10 Low threshold High threashold
2
10
4
Counted particles per min
Number of spectra per min
10
1
10
0
10
10
-1
10
-2
-12
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
x10 4.0
10
3
10
2
10
1
10
0
10
-1
10
2
4
EF,tot (J)
6
8
10
12
14
16
18
20
Aerodynamical particle size (µm)
Figure 8: Left: Histogram of the total energy per spectrum for the two measurement series. Right: Histogram of the particle counted in different bin sizes during both the series.
We divided the spectra into four different energy intervals and calculated the mean spectrum for them. The results are shown in Fig. 9. The shapes of the mean spectra for the different intervals are quite comparable. The largest difference is found in the spectra for the particles with the highest EF,tot, which are red-shifted compared to the others. This can be a consequence of detector saturation for the channels around 450 nm. We also compared the numbers of spectra in the different intervals with the particles counted in different size bins of the particles counter under the assumption that the fluorescence energy increases with particle size. The number of spectra with EF,tot > 1 pJ is approximately equal to the
Proc. of SPIE Vol. 5990 59900M-9 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2014 Terms of Use: http://spiedl.org/terms
number of counted particles larger than 12 µm in both series, with compensation for when the laser was turned off (see Fig. 5) and for the difference in trigger and spectra collection rates during the series. In the same way, the numbers of spectra above 0.5 pJ and 0.2 pJ roughly corresponds to particles larger than 10 µm and 8 µm, respectively. Finally, when all spectra are added, the numbers for the two series correspond to the numbers particles larger than 6 µm (low threshold) and 7.5 µm (high threshold), which is in agreement with the results obtained in Fig. 6. x10
Low Threshold
-14
High Threshold
-14
x10 15
10
10
EF (J)
EF (J)
15
5
5
0
0 350
400
450
500
550
600
650
350
700
400
450
500
550
600
650
700
Wavelength (nm)
Wavelength (nm)
Figure 9: Mean spectra with standard deviation for different intervals of EF,tot. The intervals are EF,tot > 1.0 pJ (squares), 0.5-1.0 pJ (circles), 0.2-0.5 pJ (up triangle) and < 0.2 pJ (down triangle), respectively. The left plot shows the measurements taken with the low threshold and the right plot shows the high threshold results.
Since we are interested in the ability to classify individual particles from there spectra, it is important that the particle specific features can be discerned from each individual spectrum. Fig. 10 shows examples of individual spectra for the two middle intervals from the measurement series with high threshold and the mean spectrum for both intervals. Although noisy, the individual spectra show fluorescence fingerprints that can be used for classification. In a previous report [14], we showed that the spectra become noisy when the light level at the PMT array is low. In this case, we are well above this level. The noise is almost entirely coming from the background fluorescence. -14
8
-14
x10
5
x10
7 4
6 5
3
EF (J)
EF (J)
4 3 2
2 1
1 0
0 -1 350
-1 400
450
500
550
600
650
700
350
400
450
Wavelength (nm)
500
550
600
650
700
Wavelength (nm)
Figure 10: Individual spectra (squares, circles and triangles) from the measurements made with the higher threshold. The plot to the left shows examples of spectra from the energy interval EF,tot between 0.5 and 1.0 pJ, while the right plot shows examples in the interval between 0.2 and 0.5 pJ. The thick line is the mean spectra for respectively interval.
Proc. of SPIE Vol. 5990 59900M-10 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2014 Terms of Use: http://spiedl.org/terms
It is of interest to compare the dry BG particle spectral characteristics with data obtained from measurements with dissolved samples of BG [14]. The wet sample spectrum has a large peak at 380 nm, due to Raman scattering in the water. Apart from this peak the spectra are similar. However, the agreement is better with the background included in the signal. In fact, the shape of the background itself is in well accordance with the wet sample trace. In comparison, the background reduced signal has a lower energy level around its maximum. This difference cannot be well explained. However, since no other particles were used in the aerosol generation system before these measurements it can be assumed that the background is actually coming from BG spores, which makes the comparison relevant. -14
12
x10
50 Total signal BG aerosol Background
11 10
x10
-14
BG (16 µg/ml) 40
9 8
EF (J)
6 5
20
EF (J)
30
7
4 3 10
2 1 0 350
0 400
450
500
550
600
650
700
Wavelength (nm) Figure 11: Mean spectra of BG excited with 337 nm for the measurements series with a high threshold and the background compared with the spectrum from BG dissolved in water. The data denoted total signal is the signal without background correction. The right axis shows the values for the wet sample fluorescence energy.
The fluorescence energy level around is 4×10-14 J at the peak channels is not in disagreement with the level from the wet samples. The concentration of 16 µg/ml of BG was found to yield somewhere between 1000 and 2000 living spores in the probe volume [14]. The wet sample produce about 6 times more fluorescence at an 8 times lower excitation level than the average dry particles during this series which indicates that the dry particles on the average contain less than 50 living spores contributing to the fluorescence. The fluorescence cross section is easily deduced when the intensity of the UV pulse is known. Many sources of errors remain but an estimate can be found. Using Eexc = 15 µJ and Aexc = 5×10-7 m2 is resulting in the numerical relationship, σF(337nm) = 0.03 EF,tot. The total energy of 0.2 pJ, for which the spectral rate was found to be in agreement with particles of about 8 µm, correspond to σF(337nm) = 0.7×10-10 cm2 which is not unreasonable for particles of this size.
3.2. Excitation at 293 nm It was problematic to run the detection system in a triggered acquisition mode due to long time delay from trigger to emission of the OPO laser (approx. 240 µs). The main problem was that the time delay was dependent on the trigger rate. This was not possible to solve during the measurements. Therefore, all our data at 293 nm have been captured in a non-triggered mode where the pulse repetition rate was kept constant at 100 Hz. As a consequence, most recorded events were acquired when no particles were present in the detection volume. Instead, separation of the particle data from the noise events was done in a post-triggering routine where the events with an integrated spectral energy below a certain threshold were rejected. The threshold level was chosen to be above the background level with high probability. Since the aerosol beam velocity was about 0.5 m/s and the diameter of the excitation laser was about 1 mm in focus, the detection probability for each particle was reduced to 0.2.
Proc. of SPIE Vol. 5990 59900M-11 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2014 Terms of Use: http://spiedl.org/terms
The mean spectra from the measurements series of BG and OA aerosol excited with 293 nm are presented in Fig. 12. As was done for BG excited at 337 nm, the spectral characteristics are compared with our previously acquired data with dissolved samples [14]. The profiles from the dissolved samples and the dry particles are similar. But again, the agreement for BG looks better for the background even if the double peak shape with maxima at 350 and 430 nm are seen. On the other hand, the spectra for OA agree very well. It should be noted that the SNR was much higher in this case. The fluorescence energy level around is 1×10-14 J for BG can be compared to the level from the wet samples. The concentration of 160 µg/ml of BG corresponds to somewhere between 10000 and 20000 living spores in the probe volume. The wet sample produced about 20 times more fluorescence with Eexc = 80 nJ which was 2.5 times less than during the current tests (about 200 nJ in the focal volume). Hence, the average BG particle above threshold in this series is found to contain around 300 living spores. For OA which was dissolved on a molecular level in the water the comparison cannot be made in the same way. However, as seen, about 5 times more fluorescence was obtained from the dissolved sample, again with a 2.5 times lower excitation energy. The concentration of OA corresponded to about 5 ng in the probe volume. Hence, this comparison yields that the average detected dry OA particle in the current series contain approximately 0.4 ng of OA. -14
2.5
-14
x10
2.0
35 Total signal BG aerosol Background
-14
x10
-14
x10
BG (160 µg/ml)
10
30
60 OA aerosol aerosol Background
x10
OA (48 µg/ml) 50
8
25
EF (J)
15
1.0
EF (J)
EF (J)
20
6
30
4
20
10 0.5 5 0.0 300
0 350
400
450
500
550
600
2
0 300
EF (J)
40 1.5
10 0 350
400
450
500
550
600
Wavelength (nm)
Wavelength (nm)
Figure 12: Mean spectra of BG (left) and OA (right) excited with 293 nm and the background compared with the spectrum from a wet sample of dissolved BG and OA. The data denoted total signal in the left graph is the signal without background correction. The right axis shows the values for the wet sample fluorescence energy.
To estimate the size of the particles contributing to mean spectra, the time traces of the particle and trigger rates are compared in Fig. 13 (left graphs). The thin lines show the particle rates for different particle sizes. The thick lines are the trigger rate and the dotted thick lines are the trigger rates multiplied with a factor of five to correct for the detection probability of 0.2 described above. During the measurement series with OA, the laser was turned off twice. The laser on/off time is shown as a thin dotted line in the graph. From the graphs we conclude that the used post-trigger threshold corresponds to particles larger than approximately 8 µm. The right graphs in Fig. 13 show the mean spectra for different energy intervals of EF,tot (EF integrated between 300 nm to 600 nm for this excitation wavelength). It is clear that the signals were saturated for the spectra with the highest EF,tot. The saturation results in a reduced standard deviation for the channels around 400 nm in the BG results and around 375 nm in the OA results. The individual spectra (not shown here) show a similar behaviour as the results for BG excited with 337 nm (see Fig. 10), i.e. the signal fluctuates around the mean spectra. The fluorescence cross section can be estimated as before. Using Eexc = 0.2 µJ and Aexc = 5×10-7 m2 is resulting in the numerical relationship, σF(293nm) = 2.5 EF,tot. For BG the total energy of EF,tot = 0.2 pJ, for which the spectral rate was found to be in agreement with particles of about 10 µm, correspond to σF(293nm) = 5×10-9 cm2. For OA we obtain the value σF(293nm) = 1×10-8 cm2 for particles of around 8 µm. Clearly, the fluorescence cross section is larger for excitation at 293 nm than at 337 nm. We have estimated the fluorescence cross sections but we are aware of the many uncertainties in our measured quantities and treat the values as order of magnitude estimates.
Proc. of SPIE Vol. 5990 59900M-12 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2014 Terms of Use: http://spiedl.org/terms
-14
25
5
EF,tot,max
5-20 µm 6-20 µm 8-20 µm 10-20 µm Triggers Corr.
4
10
3
10
2
10
>50% 25%-50% 10%-25% < 10%
20
15
EF (J)
Particles and triggers per min
10
x10
10
5
1
10
0
0
10 16:14
16:16
16:18
16:20
300
16:22
350
x10 6
5-20 µm 6-20 µm 8-20 µm 10-20 µm Triggers Corr. Laser
4
10
3
10
500
550
600
-14
70
EF,tot,max
60
>50% 25%-50% 10%-25% < 10%
50 40
EF (J)
Particles and triggers per min
10
5
450
Wavelength (nm)
Time (hh:mm)
10
400
30 20
2
10
10
1
10
0 0
10 17:40
17:42
17:44
17:46
17:48
300
350
400
450
500
550
600
Wavelength (nm)
Time (hh:mm)
Figure 13: Time traces of the counted particles and trigger rates during measurement series with BG (upper left graph) and OA (lower left graph).The mean spectra with standard deviation of this series are shown to the right; BG (upper right) and OA (lower right). The spectra are grouped in different energy intervals related to the spectrum with highest total energy EF,tot,max. The intervals are > 50% (squares), 25%-50% (circles), 10%-25% (up triangle) and < 10% (down triangle) of EF,tot,max, respectively.
4. CONCLUSION AND FURTHER WORK An early-warning system for detection of biological warfare agents has been demonstrated. It is based on the measurement of the emission spectrum upon excitation with ultraviolet pulses (2-5 ns) at 293 or 337 nm. These lasers had maximum pulse energies of approximately 30 and 300 µJ, respectively, whereof 1%-10% was used for the excitation of particles. A detection system based on a spectrograph and a photomultiplier tube array with the sensitivity near single-photon level was developed. The system is triggered by the presence of an aerosol beam particle in the excitation and detection focal volume. The most severe restriction in the sensitivity level in the work here reported was caused by background signals. These appear to originate from contamination of the sample cell glass walls by the used bioaerosols. By carefully measuring the background signal before and after the particle data acquisition, these background signals could be partially accounted for. Thus, a detailed analysis of the events involving biological warfare agent stimulant particles of sizes 5-20 µm gave quantitative data on both particle size distributions and their associated
Proc. of SPIE Vol. 5990 59900M-13 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2014 Terms of Use: http://spiedl.org/terms
emission signal levels. For example, it was demonstrated that BG spores of 8 µm size could be detected and spectrally resolved. Fluorescence cross sections were estimated and were found to be more than an order of magnitude larger for excitation at 293 nm compared to 337 nm. The successful initial tests showed that the fluorescence spectrum of individual particles of size 6-8 µm of an aerosol beam can be determined in one single shot of the laser pulse. The sensitivity of the system can be improved with orders of magnitude. A reduced background will increase the sensitivity with at least an order of magnitude. Furthermore, the collection efficiency can relatively easy be increased with a factor of 10 with small changes. The total collection efficiency, which includes the efficiency of the spectrometer, is in the present system about 0.5% at the peak wavelength. The last but not least factor of the improvement is the excitation pulse energy. So far we have not observed any saturation in the fluorescence due to the excitation pulse energy. Therefore, orders of magnitude can be gained here, although this limit is not yet established. Our further work will focus on the following issues: Redo the experiments with a clean (or without) cuvette, to reduce the background level. A better characterization of the triggering signals in order to correlate the scattering and the fluorescence signals, which will include investigations of the spectral and the total emission energy dependence of the particle size. A more precise determination of the fluorescence cross section and at what excitation pulse energy levels the fluorescence saturates. Determine the optimal number of channels in order to distinguish different species of harmful bioaerosol from relevant background signals such as diesel exhaust and natural harmless particles.
ACKNOWLEDGEMENTS This work was funded by the Swedish Defence Material Administration and the Swedish Armed Forces.
REFERENCES 1.
P. H. Kaye, J. E. Barton, E. Hirst, and J. M. Clark, Simultaneous light scattering and intrinsic fluorescence measurement for the classification of airborne particles, Appl. Opt. 39 (21), pp. 3738-3745 (2000).
2.
E. Hirst, P. H. Kaye, V. Foot, J. M. Clark, P. B. Withers, An instrument for the simultaneous acquisition of size, shape, and spectral fluorescence data from single aerosol particles, Proc. SPIE 5617 416-423 (2004).
3.
M. J. Shelton, S. P. Evans, P. D. Smith, I. A. Simpson, P. H. Kaye, J. M. Clarke, Real-time biological agent detection using particle size, shape and fluorescence characterisation, Proc. SPIE 5617, pp. 284-290 (2004).
4.
J. Ho, Future of biological aerosol detection, Anal. Chim. Acta 457 (1), pp. 125-148 (2002).
5.
V. Sivaprakasam, A. L. Houston, C. Scotto, and J. D. Eversole, Multiple UV wavelength excitation and fluorescence of bioaerosols, Opt. Express 12 (19), pp. 4457-4466 (2004).
6.
A. L. Houston, V. Sivaprakasam, C. S. Scotto, H.-B.Lin, J. D. Eversole, A. Schultz, J. Willey, Optical classification of bioaerosols using UV fluorescence and IR absorption spectroscopy, Proc. SPIE 5617, pp. 300-311 (2004).
7.
V. J. Foot, J. M. Clark, K. L. Baxter, N. Close, Characterising single airborne particles by fluorescence emission and spatial analysis of elastic scattered light, Proc. SPIE 5617 pp. 292-299 (2004).
8.
R. G. Pinnick, S. C. Hill, Y.-L. Pan, R. K. Chang, Atmospheric Environment 38 (11), pp. 1657–1672 (2004).
9.
J. R. Lakowicz, Principles of Fluorescence Spectroscopy, 2nd Ed. (Kluwer Academic/Plenum Publisher, New York, 1999).
10. A. Alimova, A. Katz, H. E. Savage, M. Shah, G. Minko, D. V. Will, R. B. Rosen, S. A. McCormick, and Alfano, Native fluorescence and excitation spectroscopic changes in Bacillus subtilis and Staphylococcus aureus bacteria subjected to conditions of starvation, Appl. Opt. 42 (19), pp. 4080-4087 (2003).
Proc. of SPIE Vol. 5990 59900M-14 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2014 Terms of Use: http://spiedl.org/terms
11. S. C. Hill, R. G. Pinnick, S. Niles, Y.-L. Pan, S. Holler, R. K. Chang, J. Bottiger B. T. Chen, C.-S. Orr, and G. Feather, Real-Time Measurement of Fluorescence Spectra from Single Airborne Biological Particles, Field Anal. Chem. Technol. 3 (4-5), pp. 221-239 (1999). 12. M. Tiihonen, V. Pasiskevicius, F. Laurell, P. Jonsson, and M. Lindgren, A compact OPO/SFG laser for ultraviolet biological sensing, Proc. SPIE 5332, pp. 134-142 (2004). 13. M. Tiihonen, V. Pasiskevicius, F. Laurell, P. Hammarström, and M. Lindgren, A UV laser source for biological and chemical sensing, Proc. SPIE 5240, p.127-136 (2004). 14. P. Jonsson, F. Kullander, M. Nordstrand, T. Tjärnhage, P. Wästerby, and M. Lindgren, Development of fluorescence-based point detector for biological sensing, Proc. SPIE 5617 pp. 60-74 (2004). 15. P. Jonsson, F. Kullander, M. Tiihonen, M. Nordstrand, T. Tjærnhage, P. Wæsterby, G. Olofsson, and M. Lindgren, Development of fluorescence-based LIDAR technology for biological sensing, Mater. Res. Soc. Symp. Proc. 883, pp. FF1.6.1-11 (2005). 16. T. Tjärnhage, M. Strömqvist, G. Olofsson, D. Squirrell, J. Burke, J. Ho, and M. Spence, Multivariate Data Analysis of Fluorescence Signals from Biological Aerosols, Field Anal. Chem. Technol. 5 (4), pp. 171-176 (2001). 17. M. Tiihonen, V. Pasiskevicius, F. Laurell, and M. Lindgren, A novel UV-laser source for fluorescence excitation of proteins, Proc. SPIE 5617, 261-268 (2004). 18. M. Tiihonen, V. Pasiskevicius and F. Laurell, Tailored UV-laser source for fluorescence spectroscopy of biomolecules, accepted for publication in Special issue of Opt. Lasers Eng. (2005). 19. R. M. Measures, Laser Remote Sensing: Fundamentals and Applications, (John Wiley & Sons, New York, 1984).
Proc. of SPIE Vol. 5990 59900M-15 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/25/2014 Terms of Use: http://spiedl.org/terms
