Fluorescence lifetime detection with particle counting devices

: Fluorescence-based single particle counting devices have become very powerful tools for human health-related applications such as the detection of blood-borne pathogens. Instead of passing the sample fluid through a thin tube or microfluidic chip, as it is commonly practiced in flow cytometers and sorter devices, single particle counters scan the fluid volume by rotation and translation of the sample container. Hence, single particle counters are not limited by the fluid flow friction and can scan a large volume in a short timeframe while maintaining high sensitivity. A single particle can be detected in a milliliter of the fluid sample within minutes, and diagnostics are being developed using this principle. Until now, signal detection with particle counters has been based on signal intensity and signal separation into multiple wavelength bands coupled with multiple detectors, which limits the number of species that can be resolved. In this paper, we applied fluorescence lifetime detection to single particle counting to increase specificity and enable multiplexing with a single detector. We demonstrate how this principle can be used for diagnostic assays based on fluorescence quenching.


Introduction
Single particle counting devices can be very powerful tools for human health centered applications due to their capability of detecting particles at extremely low concentrations from 1 to 10,000 particles per milliliter with exceptional robustness. As an example, it was previously demonstrated that bacteria can be detected with single-cell sensitivity in a singlestep, culture-and amplification-free process within 1.5-4 h from milliliters of raw blood [1]. Rapid diagnostics such as this have great promise in better management of bloodstream infections and antibiotic treatment. In this type of particle counting device, the fluid sample is transferred into a cuvette that is rotated and translated in front of a lens focusing the illumination beam into the cuvette. The signal from a small observation volume is detected in a particular wavelength band with a fast detector such as a photomultiplier [2,3]. Because the sample volume is explored by moving the container rather than passing the fluid through a tube or channel the specimen is not subjected to turbulence or shear stress and a large volume can be processed in a short amount of time. Additionally, since the optical elements are not actuated, this technique is well suited for miniaturized, point-of-care instrumentation. While scattered light can be used to identify particles passing through the observation volume, the particle size convoluted with instrument parameters (observation volume geometry, cuvette rotational velocity and signal integration time). Based on this information the number of events can be counted and quantified as particle concentration. In addition, true "hits" can be separated from false positive "hits" generated by sample impurities if their sizes and/or shapes are sufficiently different. However, particle identification solely by peak width is limited, especially when unique targets are of comparable size. In addition, signal intensity, i.e., peak amplitude, is not a good criterion either since particles of the same brightness can produce very different amplitudes depending on their position within the detection volume which is not known unless the spatial information is encoded in the signal [4]. Hence, the most common approach to detecting different species is to label them with fluorophores of different excitation/emission properties and excite and detect them in multiple wavelength bands. The drawback of detecting different species in different color channels is an increase of the complexity of the particle counting device and the practical limitation to detection in <10 color channels [5]. More importantly, for fluorescence (de)quenching-based assays, commonly used in conventional fluorescence assays such as TaqMan PCR, the quenching molecule usually does not emit any fluorescence. Hence, without an acceptor fluorophore as quencher, target identification by color is not possible. Yet, the ability to detect fluorescence (de)quenching can be vital for the identification of DNA/RNA sequences as this is often the goal in single particle counting experiments. In such an experiment a DNA/RNA strand complementary to the target sequence is tagged with a fluorescent dye quenched by a second tag in close proximity to the fluorophore. Upon binding to the target sequence, the strand is unfolded or digested, leading to a spatial separation of fluorophore and quencher resulting in an increase in fluorescence quantum yield and lifetime. Combined with nucleic acid amplification methods such as PCR this enables the detection of pathogens in nucleic acid tests for the diagnosis of infectious diseases, hereditary/genetic diseases and cancers. Therefore, in this work, we describe the implementation and application of digital frequency domain (DFD) based fluorescence lifetime detection with a particle counting device. DFD lifetime detection uses the heterodyning principle in which the emission is translated to a cross-correlation frequency much lower than the frequency of the modulated illumination source allowing to minimize cost and complexity of the detection electronics. At the same time, high precision and 100% duty cycle ensures best possible use of the signal available. We further demonstrate how lifetime analysis of the data can be used to separate out sample impurities, distinguish populations of particles detected in a single color channel and how to orchestrate a fluorescence quenching-based assay with a single detector.

Lifetime detection with a particle counting device
Our particle counting setup for lifetime detection is based on a Quanta particle counting system (ISS, Champaign, IL, USA) and a FastFLIM data acquisition card (ISS). The individual components and data acquisition and analysis procedure are described in the following subsections.

Optics and mechanics
The system is based on a Quanta particle counter (ISS) with laser diode illumination (472 nm, ISS) and detection with photomultipliers (65816, Hamamatsu) as previously described [1]. Briefly, a motorized stage holds the cylindrical sample cuvette that can be rotated as well as translated at variable speeds. The excitation/detection unit is movable as well to optimize the position of the observation volume in the sample cuvette. The excitation beam is spectrally cleaned with a band pass filter (473/10 nm, Chroma), reflected off a dichroic mirror and focused into the sample cuvette by an objective lens (20x, NA 0.4, Newport). Fluorescence is collected with the same lens, split from the incident light by the dichroic mirror and filtered with a band pass filter (535/40 nm, Chroma) to exclude scattered excitation light. The fluorescence is then focused onto an aperture/pinhole (0.1-1 mm diameter) with a lens of 50 mm focal len An overview

Electroni
A FastFLIM duration at a to an amplitu maximize the PMT was firs then discrimin TTL pulses m allowed us to noise very eff can be proce electronic setu operated and movement an gth to suppres of the optical p

Digital frequency domain principle and data analysis
To measure the fluorescence lifetime we applied the principle of heterodyning in the digital frequency domain (DFD). As in the analog frequency domain approach a pulsed/modulated light source is used to illuminate the sample for a DFD lifetime measurement. However, instead of modulating the detector by, for example, time gating or gain modulation, the entire signal is collected. Instead, a time gate is applied digitally by splitting the detected signal into several windows, covering a certain time span of the excitation pulse period (here: four windows). This way a 100% duty cycle can be maintained. The frequency of those windows, s f , is adjusted to be slightly different from the excitation pulse frequency, ex f , resulting in a shift in window position with respect to the excitation signal as a function of time ( Fig. 2(A)). This cross-correlation frequency, cc , is much lower than the frequency of the excitation signal and thus requires less bandwidth. For each photon counted, a bin number, p , will be assigned depending on window number, n w , and phase offset, cc p , between sampling and excitation clocks, with the number of phase bins, p n , and the number of windows, w n [6]. From the accumulated photon counts, a phase histogram of the lifetime response can be reconstructed ( Fig. 2(B)). This phase histogram is a convolution of the lifetime decay, the square sampling window, the system jitter and the excitation pulse. To account for these parameters, a calibration data set with a fluorophore of known lifetime is measured. From this data, the position of the lifetime phase and modulation can be calculated and presented as a position on the phasor plot ( Fig. 2(C)). The uncertainty of this position scales with the inverse of the square root of the number of photons collected. In a particle counting device the signal is recorded as a function of time. Particles crossing the observation volume create a peak in signal intensity ( Fig. 2(D)). For peak identification, the data stream is analyzed with a crosscorrelation filter as described previously [2]. In the data presented here we used a Gaussian model for the filter but it can be of any shape. For the photons accumulated during each peak extent the phase histogram is constructed ( Fig. 2(E)) and the lifetime phase and amplitude is calculated for each particle ( Fig. 2(F)). The lifetime data can then be used to discern between populations.

Results
To test the ca different popu a fluorescence

System c
To calibrate Coumarin-6 d of 2.5 ns [7]. with an avera [1]. on the millise f all peaks ide umarin 6 refe of the photons 4.0 ± 0.7 ns (f greement with t ained from the and, consequ Fig. 3(C)). T ecay model. t was to demo rophores with his experimen dye FITC at dye Cy2 at a co n the quantum 10 nm with re remely difficu 1:1 mixture of nly at a ratio th the particle c imes after peak intensity for F sible to fully ng Cy2 by mea or these wide i n volume will he confocal vo e to separate d g the lifetime e lifetime is not volume center e number of p t the periphery ured lifetimes. e in Fig. 3

Robust id
The particle c relatively larg such as dust p introducing f prepared in a As mentioned multiple popu produce muc confocal volu means of sign from the part lifetime is an particles. To microbeads (R with fluoresce and measured in Fig. 5(A) Fig. 6(C). It can be seen that the distribution of lifetimes is very different from the distributions of intensity. Although still overlapping, the presence of two populations (quenched/unquenched) is suggested by the dip in frequency of "hits" at the center of the distribution (indicated by arrows). To compare the results obtained by fluorescence lifetime with the results obtained by intensity, the two parameters are plotted against each other in Fig. 6(D). From the events detected in the negative sample (all quenched), the median values and standard deviations (SDs) of the fraction unquenched determined by lifetime (0.502 ± 0.063) as well accumulated counts per event (669 ± 159) were calculated. The median values plus SDs (fraction unquenched: 0.565, accumulated counts: 828) could serve as thresholds to distinguish positive from negative events in the positive sample and are plotted as dotted lines dividing Fig. 6(D) into four quadrants. As both lifetime and intensity increase during a transition from the quenched to the unquenched state, the distribution of events should follow a linear trajectory from quadrant 1 to quadrant 3. This is not the case. Instead, the distribution shows a curvature where events with a lower intensity but longer lifetime (quadrant 2) are much more frequent than events of high intensity but shorter lifetime (quadrant 4). This shows that the intensity value is compromised by the particle position within the observation volume while the lifetime is not. Hence, for this sample, the fluorescence intensity alone is not a reliable criterion to analyze fluorescence quenching. By adding lifetime analysis, on the other hand, it is possible to identify the different states with higher specificity. To quantify this difference in specificity we calculated the number of events residing in quadrants 2 and 4. Assuming that either a higher intensity or a longer lifetime indicate the presence of an unquenched dye droplet (positive "hit"), we counted 66 positive events detected by intensity but missed by lifetime versus 179 positive events detected by lifetime but missed by intensity resulting in a ratio of 2.7 for the detection of otherwise false negative events by intensity versus lifetime. This would be very useful for the design of an assay in which false negative events are important to exclude. On the other hand, for a test demanding low false positives, the events located in quadrant 3 are of interest as a positive event is confirmed by both, high fluorescence intensity and long lifetime lowering the probability of a false positive. We noticed that the phasor plot position of the negative, quenched droplet sample was relatively far from the position expected for a fully quenched probe. While this is in agreement with the frequent occurrence of "hits" representing quenched droplets during particle counting, we acquired fluorescence lifetime microscopy (FLIM) images for further investigation. A small volume of positive sample was loaded onto a counting slide and subjected to FLIM. The intensity image is presented in Fig. 6(E). The corresponding FLIM image with the pixels color coded according to their phasor plot positions is shown in Fig. 6(F). The two cursors in Fig. 6(G) mark the positions of the negative (cyan) and positive (magenta) droplets. Pixels within the perimeter of those cursors were painted accordingly in the FLIM image. It can be seen that the phasor plot position of the negative droplets measured by FLIM is in excellent agreement with the data obtained by particle counting and the relatively high signal from negative droplets is a consequence of the probe design. This underlines the strength of lifetime detection with particle counting devices.

Discussio
Our prototype intend to use 6 properties. By pulsing the lasers in an alternating fashion, crosstalk between detection channels can be avoided [11]. Hence, the presence of two dyes on the same target can be identified. Further, we propose that instead or in addition to lifetime measurements, spectrally resolved detection could be used with particle counting devices including spectral phasorbased analysis. While fluorescent markers of similar emission spectrum are difficult to distinguish by fluorescence intensity or color, they can be clearly separated by lifetime if significantly different. With phasor-based analysis even small differences can be resolved with only little fluorescence signal. Hence, we propose unmixing of a large number of species (3-10) by lifetime. Fluorescence resonance energy transfer (FRET) is a special form of fluorescence quenching in which the energy of one dye, the donor, is transferred to another dye, the acceptor, over short distances (0.1-1 nm). Since FRET affects the fluorescence lifetime, we propose to use FRET dye pairs as sensors for biomedical applications with particle counting devices.