A“quasi”confocaldropletreaderbasedonlaser … · 2019-08-12 · Changchun New Industries Optoelectronics Technology Co., Ltd (Changchun, China). Two dichroic mirrors were purchased

Contents lists available at ScienceDirect

Talanta

journal homepage: www.elsevier.com/locate/talanta

A “quasi” confocal droplet reader based on laser-induced fluorescence (LIF)cytometry for highly-sensitive and contamination-free detectionXiurui Zhua,1, Baoxia Liub,1, Shisheng Sua, Bo Wangb, Yu Baib, Haiwang Huangb, Xiaobin Liub,Xin Chengb, Xianhua Wangb, Lingxiang Zhub,c, Wenjun Yanga,b, Na Gaob, Gaoshan Jingd,**,Yong Guoa,*a Department of Biomedical Engineering, School of Medicine, Collaborative Innovation Center for Diagnosis and Treatment of Infectious Diseases, Tsinghua University,Beijing, Chinab TargetingOne Corporation, Beijing, ChinacNational Research Institute for Family Planning, Beijing, ChinadDepartment of Precision Instrument, School of Mechanical Engineering, State Key Laboratory of Precision Measurement Technology and Instruments, Tsinghua University,Beijing, China

A R T I C L E I N F O

Keywords:Droplet digital PCRFluorescence detectionConfocal opticsContamination-free detection

A B S T R A C T

Highly-sensitive and contamination-free droplet digital PCR (ddPCR) is an enabling technology and widelyneeded for accurate quantification of nucleic acid in clinical applications. In this paper, a novel droplet readerwas developed by combining a “quasi” confocal laser-induced fluorescence (LIF) cytometry with a delicatemicrofluidic chip design. The droplets with a size of 90 μm was illuminated at an out-of-focus position by twoaligned laser beams to generate maximum fluorescent signal. Additionally, the lateral offset position of themicrofluidic chip should be precisely tuned so that the bandwidth of the FAM and VIC channels were configuredat the matching sizes. Then, PMT gain voltages and pneumatic pressures were optimized for better dropletdetection efficiencies. An aerosol adsorption experiment was performed to demonstrate that there was no aerosolcontamination, and detected copy numbers of both mutants and wild types scaled linearly with the expectedinput copy numbers ( >r 0.9982 ) with a LoB of 0.0 copies and LoD of 3.0 copies. The results demonstrated thatthis droplet reader with the delicate chip is a convenient, highly-sensitive and contamination-free to detectfluorescence signals inside droplets after ddPCR, which is highly promising for broad applications of ddPCR inclinical diagnosis.

1. Introduction

Droplet digital PCR (ddPCR) is an enabling technology for accuratequantification of nucleic acid in clinical applications, including liquidbiopsy [1], non-invasive prenatal testing [2], early diagnosis of con-tagious diseases [3], and detection of rejection after organ transplan-tations [4]. A typical strategy of ddPCR is as follows: one specific vo-lume of analyzed sample, e.g., 20 μL–50 μL, is evenly distributed into alarge number of tiny reaction units (“droplets”) and the number of re-action units could vary from less than one thousand to ten million.Then, these tiny reaction units of droplets simultaneously perform PCRreactions to achieve PCR amplification of single-copy or multi-copytarget sequences. After the PCR reaction, these tiny reaction units with/

without amplicons are detected in parallel or in serial, in whichthreshold values are set for the fluorescence signals detected from eachreaction unit. When a droplet's fluorescent signal in a specific channel ishigher than the corresponding threshold value, the reaction unit of thedroplet is designated as “one” (“positive”). Otherwise, the reaction unitof the droplet is designated as “zero” (“negative”). With the dropletspartitioning, an absolute count of target templates can be derived bycounting “ones” and “zeroes” without the help of standard curve as forqPCR, making ddPCR an ideal technique for sensitive detection of raretargets.

Although increasing evidences indicate that ddPCR reaction is asensitive and promising diagnostic method, its implementation in rou-tine diagnostic settings has been limited by concerns over PCR

https://doi.org/10.1016/j.talanta.2019.120200Received 31 May 2019; Received in revised form 29 July 2019; Accepted 30 July 2019

* Corresponding author. Room C215, Department of Biomedical Engineering, School of Medicine, Tsinghua University, Beijing, 100084, China.** Corresponding author. Room 3303, Department of Precision Instrument, School of Mechanical Engineering, Tsinghua University, Beijing, 100084, China.E-mail addresses: [email protected] (G. Jing), [email protected] (Y. Guo).

1 These authors contributed equally to this work.

Talanta 206 (2020) 120200

Available online 01 August 20190039-9140/ © 2019 Elsevier B.V. All rights reserved.

T

contamination and the sensitive detection. Accordingly, the dropletsreader and microfluidic chip are crucial to differentiate droplets with“one” (“positive”) and “zero” (“negative”) fluorescent signals in ahighly sensitive and contamination-free manner.

To achieve highly sensitive quantification of nucleic acids, twotypes of methods are commonly applied for ddPCR detection, fluores-cence imaging and laser-induced fluorescence (LIF) cytometry. Influorescence imaging, all the generated droplets are placed on a flatsubstrate in a monolayer fashion. Once these droplets are illuminatedby a broad-spectrum light source (e.g., LED) with specific excitationfilters, specific excited fluorescent signals of these droplets such as FAM(515–530 nm), VIC (560–580 nm) and ROX (610–650 nm) are taken bya CCD camera with specific emission filters [5]. Indeed, there are sev-eral advantages to the fluorescence imaging method. First, dropletswith fluorescent signals can be detected in a high-throughput manner,and thousands of droplets can be detected simultaneously with fluor-escent signals acquired in one image [6]. With a unique wide-anglelens, even one million droplets can be detected simultaneously [7].Second, design of the optical system is relatively simple and straight-forward: a conventional digital single-lens reflex camera (DSLR) or evena smartphone can be used for fluorescence imaging [8]. Third, multi-color fluorescence imaging can be conveniently achieved by addingmultiple fluorescence filter sets for multiplexed assays. However, de-tection sensitivity and dynamic range of fluorescence imaging arelimited due to variation of the illumination source and fluorescencesignal interference between adjacent droplets [9]. Another detectionmethod is laser-induced fluorescence (LIF) cytometry, which has beenused by two commercialized ddPCR systems from Bio-Rad and Rain-dance Technologies [10,11]. In LIF, droplets are treated like mamma-lian cells, and fluorescence signals can be detected using flow cyt-ometers at a rate of hundreds to thousands of droplets per second [12].Though not much detailed technical information was disclosed for thesetwo commercialized systems, LIF detection can be realized with anoptical confocal detection module which will greatly improve the sen-sitivity of fluorescence detection [13–16]. In confocal optics, pinholesare ubiquitously applied for incident lasers and exited fluorescencesignals to increase the signal to noise ratio by eliminating non-focaloptical noises [17]. First, laser beams are emitted and focused on amicrofluidic detection channel. Then, fluorescence inside the droplets isexcited and detected when droplets pass the detection spot. Finally, thefluorescent signals are acquired in serial by photomultiplier tubes(PMTs) to convert optical signals into electrical signals. In this way, theLIF cytometry method achieves highly sensitive detection of fluorescentsignals with decent throughput. Higher sensitivity and dynamic rangecan be achieved using LIF cytometry compared to fluorescence imaging.However, the design of the LIF optical system is relatively complicatedcompared to a fluorescence imaging system. Additionally, conventionalflow cytometers are not suitable to analyze larger droplets with a dia-meter of 100 μm because these flow cytometers are tuned for fluores-cence detection in a typical diameter of 10–20 μm, a typical size ofmammalian cells. For clinical applications, highly sensitive fluores-cence detection is essential for ddPCR to achieve reliable and accuratediagnostic results for patients’ samples.

In addition to highly sensitive fluorescence detection, con-tamination-free detection is another crucial requirement for clinicalapplications using PCR-based technologies, which is even more cri-tical for ddPCR technology. After ddPCR amplification, amplifiedDNA molecules are highly concentrated in a droplet with a typicalsize of tens of microns. Even the release of such highly concentratedamplicons from several droplets will cause massive aerosol con-tamination. The resulted cross-contamination will result in false-positive diagnosis in quantification [18]. Therefore, the release ofamplicons from droplets should be prevented at all cost to avoidcrossover contaminations during the detection process of ddPCR.Several delicate microfluidic chips have been developed to integratePCR amplification reaction and fluorescence signal detection into an

enclosed space to avoid contamination. Prakash et al. designedmicro-valves inside a ddPCR chip to isolate the PCR chamber and therelease of aerosol through the inlets, as well as to prevent the eva-poration of droplets. After PCR amplification, the droplets weredriven towards an enclosed detection site downstream, and fluor-escent signals were recorded [19]. Dorfman, Frey, and Jiang et al.used a continuous-flow ddPCR method to prevent contamination, inwhich droplets were driven continuously in a long serpentine mi-crofluidic channel crossing different temperature zones. During thefinal stage of the continuous-flow, fluorescent signals inside dropletswere detected by a flow cytometer [20–22]. For a commercial NeicaddPCR system from Stilla Technologies [23], droplet generation, PCRamplification and fluorescence detection can be integrated into onemicrofluidic chip. Highly sensitive detection without contaminationusing ddPCR is essential and crucial to achieve a reliable and accu-rate diagnostic result for patients’ samples, even though the com-plexity of a microfluidic chip for droplet detection would be sig-nificantly increased.

To further increase the application of ddPCR in clinical diagnosis,ddPCR should be seamlessly integrated with essential consumableswidely used in clinical assays, for instance, eight-strip tubes and 96-well plates as standard components for most of the manual and au-tomated instruments. One crucial step in operating eight-strip tubesand 96-well plates is to avoid cross contamination during opening/sealing processes. In this paper, a novel droplet reader was developedfor highly-sensitive and contamination-free detection of fluorescencesignals inside droplets after ddPCR, by combining a “quasi” confocalLIF cytometry with delicate microfluidic chip design. Compared toconventional confocal flow cytometry, a droplet with 90 μm in dia-meter was illuminated at an out-of-focus position by two alignedlaser beams (473 nm and 532 nm) to generate maximum fluorescentsignals. Once the fluorescence was excited, it was detected with apinhole in each fluorescence channel placed in conjugation with thefocal plane of the PMTs only in the near vicinity of the focal plane,which would result in a superior signal to noise ratio by eliminatingout-of-focus noises. Furthermore, the droplet reader managed todetect fluorescent signals inside droplet with high sensitivity througha series of optimization. Driving pressures were adjusted for thehighest detection throughput without signal aliasing, and opticalsetup was tuned for confocal alignment with microfluidic channels toachieve the highest signal-to-noise ratio. Meanwhile, a delicate mi-crofluidic chip to obtain contamination free detection was designedfor conventional eight-strip PCR tubes containing droplets after PCRamplification. The standard eight-strip PCR tubes were tightly sealedwith the microfluidic chip through a well-designed cap-piercing andchannel-sealing process, which will be realized simultaneously. Anaerosol adsorption experiment was performed to demonstrate thatthere was no aerosol contamination in the near vicinity of the dropletreader, and an absolute quantification experiment with gradientconcentrations was carried out to assess the quantification perfor-mance of the droplet reader. The droplet reader utilized a “quasi”confocal LIF cytometry, a highly optimized counterpart for the con-ventional confocal LIF cytometry in terms of ddPCR. Using a delicatemicrofluidic chip design, conventional consumables of eight-striptubes are integrated seamlessly into the droplet reader without crosscontamination. The experimental results demonstrated that thedroplet reader is a convenient, efficient, highly-sensitive and con-tamination-free system to detect fluorescence signals inside dropletsafter ddPCR, which is highly promising for broad application ofddPCR in clinical diagnosis.

2. Experimental

2.1. Materials and reagents

Two lasers (MBL–III–473 and MGL–III–532) were purchased from

X. Zhu, et al. Talanta 206 (2020) 120200

2

Changchun New Industries Optoelectronics Technology Co., Ltd(Changchun, China). Two dichroic mirrors were purchased fromSemrock, IDEX Health & Science LLC (New York, NY). Two photo-multiplier tubes (PMTs) (H9307-2) were purchased from HamamatsuPhotonics (Iwata, Japan). Three multi dichroic beam splitters (ZT473-532-633rpc-UF1, ZT488rdc-UF1, T550lpxr) were purchased fromSemrock, IDEX Health & Science LLC (New York, NY). Other mis-cellaneous optical accessories, such as reflective mirrors, focal lenses,an objective lens, pinholes, and adjustable lens mounts were purchasedfrom Daheng Science and Technology (Beijing, China). Droplet gen-erator, epidermal growth factor receptor (EGFR) T790M and human leu-kocyte antigen B27 (HLA-B27) mutation detection kits were purchasedfrom TargetingOne Corporation (Beijing, China). Mutant and wild typeplasmids were prepared by GENEWIZ (Suzhou, China).

2.2. Optical system of the droplet reader

The optical module consists of three parts: an excitation part, amicroscopic part, and a detection part. The excitation part is composedof two lasers at wavelengths of 473 nm (MBL-S-473) and 532 nm (MGL-S-532), a reflective mirror (GCC-102201) and a dichroic mirror one(Di02-R514-25x36). The microscopic part is composed of a dichroicmirror two (Di01-R405/488/532/635-25×36) and an objective lens(GCO-2132). The detection part is composed of a focal lens (GCL-010137), a dichroic mirror three (Di02-R532-25x36), two pinholes(GCO-01100A) and two PMTs (H9307-2). All the mirror, focal lens anddichroic mirrors are mounted on adjustable lens mounts to flexibly tunethe optical configuration.

2.3. Fabrication of microfluidic chip for the droplet reader

The microfluidic chip for the droplet reader was fabricated bycombining photolithography, electroplating, injection molding, onestep surface modification and bonding, and ultrasonic welding. Briefly,channels with microfluidic channels as small as 70 μm were first de-signed with AutoCAD (Autodesk Inc.). Next, the microfluidic channelswith a depth of 90 μm were fabricated with contact photolithography,electroplating and injection molding using polycarbonate (PC) mate-rial. Then, two PC substrates with microfluidic channels were bondedtogether with hydrophobic surface properties with a one-step methoddescribed in a previous publication [24]. Finally, the sealed micro-fluidic was assembled with a PC adaptor (fabricated with injectionmolding) with ultrasonic welding technique.

2.4. Evaluation of aerosol contamination

Before the evaluation of aerosol contamination, EGFR T790M mu-tation detection assays were performed with the droplet reader forseven consecutive days. To find out whether there was aerosol con-tamination around the droplet reader, seven samples (30 μL each)without templates and one sample (30 μL) with about 400 copies ofmutants and wild types each were prepared with the EGFR T790Mmutation detection kit. Six samples without templates were placed inthe open air near 15 cm around the droplet reader for 15min. The restof the samples served as negative and positive controls. Droplets weregenerated from these eight samples with a droplet generator and thenwent through a PCR program of 95 °C for 10min, followed by 40 cycles

Fig. 1. Schematic diagram of droplet reader. (a, b)Schematic design and development of the droplet reader.The scale bar is 10 cm. (c) The optical setup of the dropletreader. The 473 nm laser is reflected by the mirror, and thenmerges with the 532 nm laser at the dichroic mirror one. Themerged laser is reflected by the dichroic mirror two into theobjective lens. The merged laser is then focused on the de-tection channel by the objective lens to excite the fluores-cence inside droplets. The excited fluoresce is collected bythe objective lens, transmits through the dichroic mirrortwo, and then gets focused by the focal lens. Finally, thefluorescence is split by the dichroic mirror three into twodifferent channels, and the split fluorescence passes throughtheir respective pinholes into different photomultiplier tubes(PMTs). (For interpretation of the references to color in thisfigure legend, the reader is referred to the Web version ofthis article.)

X. Zhu, et al. Talanta 206 (2020) 120200

3

of (94 °C for 30 s and 60 °C for 1min) on a commercialized PCR thermalcycler, as suggested in the kit's application note. Finally, droplets weredetected with the droplet reader to reveal any possible cross-con-tamination from the aerosol around the droplet reader.

2.5. Evaluation of ddPCR quantification performance

To perform the evaluation of quantification performance withddPCR, 32 samples (30 μL each) were prepared and divided into eightgroups. Different amounts of mutants and wild types were added toeach of the four replicates at 0, 10, 30, 100, 300, 1,000, 3,000 and10,000 copies approximately. Droplets were generated from these eightgroups with the droplet generator and then went through a PCR pro-gram of 95 °C for 10min, followed by 40 cycles of (94 °C for 30 s and60 °C for 1min), as suggested in the kit's specifications. Finally, dropletswith fluorescent signals were detected with the droplet reader to derivethe detected copy numbers of mutants and wild types.

3. Results and discussion

3.1. Working principle of droplet reader

As shown in Fig. 1c, the droplet reader (Fig. 1a–b) consists of a“quasi” confocal system and a microfluidic chip to detect fluorescencesignals inside droplets in a highly sensitive manner. For a conventionalconfocal system, two pinholes are used for one channel of fluorescentdetection to acquire an image with superior spatial resolution, as wellas a large signal-to-noise (S/N) ratio for a sample. One pinhole is placedin front of a laser source to generate a point illumination, while anotherpinhole in front of the detector to effectively eliminate out-of-focussignal noise. Thus, a spatial resolution of sub-micrometer can be ob-tained with superb signal-to-noise ratio through 2D or 3D image scan-ning and multi-image reconstruction. In this “quasi” confocal system,there are no pinholes in front of the two lasers, and the two lasers weretreated as parallel light sources. Still, a pinhole is placed in front of eachdetector to eliminate out-of-focus optical noise. The objective of the“quasi” optical system is to illuminate a droplet with a size of 90 μm andgenerate maximum fluorescent signal. To fully illuminate the dropletwith a size of 90 μm, the illumination system is adjusted in an out-of-focus manner. First, a 473 nm laser beam was first reflected by a mirrorand merged with a 532 nm laser beam at the dichroic mirror one. Next,the merged laser beam was reflected by the dichroic mirror two andentered the 20× objective lens, which will illuminate the droplets thatpass by the detection site. Then, the excited fluorescence (FAM: 520 nmand VIC: 553 nm) was collected by an 20× objective lens, transmittedthrough the dichroic mirror two, and focused by an 8× focal lens. Fi-nally, the fluorescence was split into two channels (FAM and VIC) bythe dichroic mirror three and entered two photomultiplier tubes (PMTs)respectively, aided by two pinholes, to generate fluorescence-basedelectric signals. Before entering the objective lens, the two lasers can betreated as parallel light sources with a diameter of 1.0mm. Then, thesetwo lasers were focused by the objective lens to generate a light spotwith a diameter of about 10 μm on the detection microfluidic channel.To fully illuminate the droplet with a size of 90 μm, the excitation op-tical system is adjusted in an out-of-focus manner. Once the fluores-cence signals were excited, they were converted to parallel light by the20× objective lens. The parallel fluorescent lights were focused by the8× focal lens with two pinholes placed in conjugation with the focalplane of the PMTs only in the near vicinity of the focal plane, whichwould result in a superior signal to noise ratio by eliminating non-focaloptical interfering noises.

To avoid the release of aerosol that may cause cross-contamination,an enclosed design was realized in a delicately designed microfluidicchip (Fig. 2a). As shown in Fig. 2b, the eight-tube strip that containeddroplets to be detected was first mounted onto the integrated adaptor ofthe microfluidic chip. Hollow needles on the adaptor pierced through

the rubber cap on the tube strip, and in the meantime, the O-ring on therubber cap was pushed tightly against the adaptor. During this step, atight seal of the space was formed in which droplets were driven toensure that no aerosol was released. Next, driving oil and spacing oilwas driven into the microfluidic channels through pneumatic pressure.Then, the driving oil was injected into one of the tubes, floating dropletsinto the hole at the center of the hollow needle connected to the de-tection channel. Meanwhile, the spacing oil was driven through themicrofluidic channels connected to the nozzle. Finally, the dropletsfloated from the hollow needle were separated by the spacing oil at thenozzle, pushed through the detection channel, and collected by theclosed waste reservoir after detection.

The complete operation process of the droplet reader is further il-lustrated with Fig. S1 in the supplementary data.

3.2. Optimization of optical parameters

To obtain optimal performance of the “quasi” confocal system, threeparameters were precisely adjusted: offset between the laser's focalpoint and droplets inside of the microfluidic channel, lateral offsetposition, and PMT gain.

To facilitate the scope of illumination to the size of droplets, anoptimal out-of-focus position should be used for droplet detection, sincethe droplet diameter (90 μm) is much larger than the spatial resolutionfor conventional confocal microscopy. Indeed, we found the largestfluorescence signals can be obtained at an out-of-focus position, whichis 300 μm lower than the focal point of the lasers. In this position, adroplet with a size of 90 μm was wholy illuminated and generated themaximum fluorescence signal.

Once the optimal out-of-focus position for droplet detection is de-termined, the lateral offset position of the microfluidic chip should alsobe precisely tuned and optimized since the microfluidic chip wasmoving horizontally along the detection plane. As shown in Fig. 3a,when the lateral offsets changed from −70 μm to +70 μm, the laserspots entered and then left the detection channel, causing a rise-and-fallamplitude pattern as fluorescent droplets flowed through the detectionchannel. Here, the bandwidth was defined as the width of the range inwhich signal amplitudes were over 70.7% of the maximum. The dis-tance between the centers of the bandwidth in two fluorescence chan-nels was regarded as the distance between two laser spots. The band-width of the FAM channel was from −30 μm to +35 μm, and that forthe VIC channel was from −25 μm to +30 μm. Since the bandwidths ofthe two fluorescence channels were concentric, the laser spots of thetwo laser beams were accurately aligned. The lateral offset for precisealignment between laser spots and the detection channel was +2.5 μm.With the carefully calibrated lateral offset and the existence of the twofluorescent channels’ bandwidth, the droplet reader system is able todetect fluorescence intensities inside droplets in a robust and accurateway.

After the focal point and lateral offset position of the microfluidicchip were determined, PMT's gain voltage was optimized to fit theirmeasurement ranges. Since the commonly used probe concentrations ofddPCR were 0.05–0.3 μM [25], the amplitude of the fluorescence sig-nals was adjusted to about 10,000 when droplets containing 0.1 μMFAM and VIC each was detected, given that the digitized output of thefluorescence signal was 0–32,767. The amplitude of the fluorescencesignals was adjusted by changing the gain voltage of the PMTs. Asshown in Fig. 3c, the amplitude of the fluorescence signals scaled ex-ponentially with the gain voltage as previously described [26]. With anexponential fitting, the gain voltages that tuned the amplitudes to10,000 for FAM and VIC channels were 2.22 V and 2.19 V, respectively.Such configuration of gain voltage can accommodate most of biologicalassays [27].

X. Zhu, et al. Talanta 206 (2020) 120200

4

3.3. Optimization of pneumatic parameters

To achieve high-throughput and highly-efficient fluorescence de-tection, pneumatic pressures need to be optimized. Pneumatic pressuresaffect the frequency that droplets pass through the detection site, aswell as the spacing between droplets, which is characterized by the dutyratio of the fluorescence signal. For the droplet reader, the pneumaticpressures were optimized to achieve: (1) the highest detection fre-quency, (2) the highest duty ratio, and (3) no aliasing between fluor-escence signal peaks from two successive droplets. To ensure thatdriving oil, spacing oil and droplets should flow in the desired direction;pneumatic pressure was equally exerted onto driving oil and spacingoil.

As shown in Fig. 4a, when the pneumatic pressures increased from400mbar to 1,000mbar, the detection frequency increased with theexerted pneumatic pressure from 319.3 Hz to 1,577.7 Hz. Similarly, asshown in Fig. 4b, the duty ratio also increased with the exerted pneu-matic pressure from 15.61% to 56.88% during the process. However,when 1,000mbar was used onto driving oil and spacing oil, dropletssometimes formed aggregates due to stochasticity when entering themicrofluidic channel, which produced aliasing between fluorescencesignal peaks. As shown in the microscopic image inset, excessively highpneumatic pressure might cause a deficiency in droplet spacing andthus result in aliasing of fluorescence signals. Therefore, the optimalpneumatic pressures for the droplet reader were 900mbar with a dro-plet detection rate of 1,350 droplets/s and a duty cycle of 55%.

Fig. 2. Schematic diagram of droplet reader's microfluidic chip. (a) The microfluidic chip developed for the droplet reader. The scale bar is 1 cm. (b) Thedetection process of the droplet reader's microfluidic chip. First, the microfluidic chip is mounted onto an Eppendorf tube strip with rubber cap. The hollow needle onthe adaptor pierces through the rubber cap to open a path for droplet detection, while simultaneously the O-ring on the rubber cap seals the space in which dropletsare driven to ensure that no aerosol is released. Next, driving oil and spacing oil are added to the designated ports, and driven by pneumatic pressure into themicrofluidic channels. Then, the driving oil is injected into the tube strip through the slot on the hollow needle, floating droplets towards the outlet in the center ofthe hollow needle. Meanwhile, the spacing oil is driven to the nozzle, where droplets are spaced before flowing to the detection site. Finally, droplets flow through thedetection site in the detection channel and are collected in the closed waste reservoir after the detection. (For interpretation of the references to color in this figurelegend, the reader is referred to the Web version of this article.)

Fig. 3. Schematic illustration of an out-of-focusconfocal configuration. (a) First, the laser beamwas in an out-of-focus position when illuminatingdroplets, so that the whole droplet could be excitedto increase fluorescence intensity. Then, the fluor-escence emitting from the excited droplet wentthrough a confocal system, where a pinhole wasplaced in conjugation with the microfluidic channelplane for each fluorescence channel to eliminate theentry of undesired light. (b) The alignment betweenlaser spots and the detection channel. The objectivelens remained stationary, and the microfluidic chipmoved in a direction perpendicular to the detectionchannel in the detection plane with a lateral offsetfrom −70 μm to +70 μm. Amplitudes of fluores-cence signals were monitored and recorded to revealthe bandwidth of each channel. The concentric pat-tern of the bandwidths showed that the laser spotswere well aligned. The lateral offset for the align-ment with the detection channel was +2.5 μm. (c)The compatibility between fluorescence signal am-plitudes and the measurement range of PMTs. Boththe objective lens and the microfluidic chip remainedstationary. Different gain voltage was applied on thePMTs from 1.70 V to 2.25 V. Amplitudes of fluores-

cence signals were monitored and recorded, and then tuned to 10,000 by fitting the amplitude-voltage curve to an exponential model, deriving 2.22 V for the PMTmounted in FAM channel and 2.19 V for the one in VIC channel. (For interpretation of the references to color in this figure legend, the reader is referred to the Webversion of this article.)

X. Zhu, et al. Talanta 206 (2020) 120200

5

Eventually, all the droplets in eight tubes can be completly detected in30min.

3.4. Evaluation of aerosol contamination with epidermal growth factorreceptor (EGFR) T790M mutation detection assays

To evaluate whether there was aerosol contamination released fromthe droplet reader, EGFR T790M mutation detection assays were per-formed for seven consecutive days before the evaluation. For the eva-luation experiment, six tubes of EGFR T790M reaction mix were

exposed to the air in the near vicinity of the droplet reader for 15min,together with a positive control sample (with 400 mutants and wildtypes each) and a negative control sample (without templates or ex-posure to the air). As shown in Fig. 5a–f, no template (0.0 copy ofmutants and wild types each) was detected in all of the six samplesexposed to the air, while the positive control sample yielded expectedresults as 369.8 copies of mutants and 440.6 copies of wild types, andthe negative control sample 0.0 copy of mutants and wild types each.These results showed that no aerosol contamination was detectedaround the droplet reader since the enclosed design of the microfluidic

Fig. 4. Optimization of droplet reader's pneu-matic parameters. When pneumatic pressure wasequally exerted onto driving oil and spacing oil from400mbar to 1000mbar, (a) detection frequency in-creased with the exerted pneumatic pressure from319.3 Hz to 1,577.7 Hz, and (b) duty ratio also in-creased with the exerted pneumatic pressure from15.61% to 56.88%. (c) For pneumatic pressures up to900mbar, droplets were effectively separated by thespacing oil, forming well-defined peaks in the fluor-escence waveform. (d) Excessively high pneumaticpressure might cause a deficiency in droplet spacingand thus result in aliasing of fluorescence signals.When 1,000mbar was exerted onto driving oil andspacing oil, droplets formed aggregates from time totime, producing merged fluorescence signal peakswhich were challenging to be differentiated. (Forinterpretation of the references to color in this figurelegend, the reader is referred to the Web version ofthis article.)

Fig. 5. Evaluation of aerosol contamination. After seven consecutive days of EGFR T790M mutation detection assays, six EGFR T790M samples without templateswere exposed to the air around the droplet reader. (a–f) No aerosol contamination was observed in all of the six samples. (g) An EGFR T790M experiment with 400copies of mutants and wild types each was performed as the positive control. 440.6 copies of mutants and 369.8 copies of wild types were detected. (h) An EGFRT790M experiment without templates or air exposure was performed as negative control. 0.0 copies of mutants and 0.0 copies of wild types were detected. (Forinterpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

X. Zhu, et al. Talanta 206 (2020) 120200

6

chip did not allow any release of aerosol contamination from the eight-tube strips.

3.5. Evaluation of quantification performance with human leukocyteantigen B27 (HLA-B27) mutation detection assays

To evaluate the performance of the droplet reader, a series of HLA-B27 mutation detection assays were performed with gradient con-centrations of template inputs. When these ddPCR samples wentthrough the droplet reader, the results underwent the following processto achieve statistical analysis. First, classification was performed oneach ddPCR data with an automatic density-watershed algorithm(DWA) method as previously described [29]. Then, the classificationresults were analyzed with Poisson statistics to calculate the copynumbers and its 95% confidence interval.

The copy number of templates in a droplet conforms Poisson dis-tribution [11]. If CPD is defined as , the probability that a dropletcontains k copies of templates can be calculated as Equation (1).

=p kk

exp( )!

( )k

(1)

By definition, the proportion of positive droplet count=Np N p/ 1 (0), where N is total droplet count and Np is the number

of droplets that contain templates. According to Equation (1), CPD ( )can be calculated with Equation (2).

= NpN

ln 1(2)

The copy number is calculated with Equation (3), where =N 750000is the number of droplets generated, according to the droplet gen-erator's specification.

=K N0 (3)

As mentioned above, the number of droplets that contain templates(positive droplet count) can be derived from ddPCR classification re-sults. Theoretically, since the copy number of templates in a dropletconforms a stochastic distribution, the number of template-containingdroplets Np( ) is also random. With interval estimation, the lower andupper bounds of 95% the confidence interval (Np and +Np ) is given bythe inverse 2 or inverse normal distribution as Equation (4) [28].

=<

+

=+ <

++

NpNp Np

Np z Np p Np

NpNp Np

Np z Np p Np

Inv (0.025, 2 ), if 20

(0.025) (1 ) , if 20

Inv (0.975, 2 2), if 20

(0.975) (1 ) , if 20

12

2

12

2

(4)

Fig. 6. Evaluation of quantification performance. In 32 samples grouped in four, different amounts of HLA-B27 mutants and wild types were added, including 0,10, 30, 100, 300, 1,000, 3,000 and 10,000 copies each. Up to four clusters formed clearly in the detection results. Statistical analysis showed that the detected mutantcopy numbers were about 96% of the expected ones with an =r 0.99892 , and the detected wild-type copy numbers were about 96% of the expected ones with an

=r 0.99852 . Gray zones corresponded to calculated 95% confidence interval against different input copy numbers when detected droplet counts were 10,000. (Forinterpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

X. Zhu, et al. Talanta 206 (2020) 120200

7

Since copy number ( =K N0 ) increases monotonically with positivedroplet count (Np) in Equation (2) and Equation (3), the lower andupper bound of copy number (K and +K ) can be calculated withEquation (5).

=

=+ +( )( )K N

K N

ln 1

ln 1

NpN

NpN

0

0 (5)

Limit of detection (LoD) was calculated with a limit of blank (LoB)as previously described [11]. First, LoB was calculated with false po-sitive counts per sample FP, as shown in Equation (6).

==

<+ + >

LoB0, if 01, if 0 0.05

1.645 0.8, if 0.05

FP

FP

FP FP FP (6)

Then, LoD was calculated with the LoB calculated above, as shownin Equation (7).

==

<+ + >LoB

LoD3, if 05, if 0 0.05

(1.645 1.645 4 ) , if 0.05

FP

FP

FP14

2 2(7)

As shown in Fig. 6a–h, all of the detection results formed clearclusters. As shown in Table 1 and Table 2, the detected mutant andwild-type copy numbers were highly consistent with the expected inputcopy numbers. As shown in Fig. 6i and j, the detected copy numbers ofboth mutants and wild types scaled linearly with the expected inputcopy numbers (slope =k 0.96 and >r 0.9982 ). With reference to the95% confidence intervals, the detected copy numbers showed goodreproducibility and consistency. Since no false positive droplets weredetected in negative samples, the LoB was 0.0 copies, and the LoD was3.0 copies. No false positives or false negatives were detected inFig. 6a–h.

4. Conclusions

In summary, a highly-sensitive and contamination-free dropletreader was developed for accurate and absolute quantification of

nucleic acids with droplet digital PCR (ddPCR). With the optimizedpneumatic parameters, droplets were driven to the detection site athigh frequency and duty ratio without signal aliasing. With the “quasi”confocal laser-induced-fluorescence (LIF) optical design, fluorescentsignals from the droplets were detected in a high-throughput andhighly-sensitive manner. With the enclosed design of the microfluidicchip, no aerosol contamination could be released from tube strips thatcontained droplets after ddPCR, as validated with the epidermal growthfactor receptor (EGFR) T790M mutation detection assays. With all ofthese optical and microfluidic design, the droplet reader achieved ac-curate and absolute quantification of both human leukocyte antigen B27(HLA-B27) mutants and wild types ( >r 0.9982 ). The experimental re-sults showed that the droplet reader is a highly-sensitive, contamina-tion-free and high-throughput droplet detection apparatus for accurateand absolute quantification of nucleic acids, which will greatly promotethe application of ddPCR in clinical diagnosis.

Acknowledgments

This work was financially supported by the National Natural ScienceFoundation of China (Grant Nos. 81572083 and 041320034), theCentral Public Interest Scientific Institution Basal Research Fund (GrantNo. 2016GJZ01), and the Fund from TargetingOne Corporation (GrantNo. 20162000009).

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.talanta.2019.120200.

Conflicts of interest

The authors declare no competing interests.

Author contribution

Xiurui Zhu, Baoxia Liu, Gaoshan Jing and Yong Guo conceived theidea, discussed the experimental results, performed data analysis, andprepared the manuscript. Xiurui Zhu and Baoxia Liu conducted theexperiments and collected experimental data. Bo Wang, Yu Bai,Gaoshan Jing, Xiurui Zhu and Shisheng Su designed and fabricated thedroplet reader's microfluidic chip. Xianhua Wang, Yu Bai and ShishengSu designed the droplet reader's optical, pneumatic and control mod-ules. Xiurui Zhu, Shisheng Su, Xiaobin Liu, Xin Cheng and HaiwangWang developed and tested the droplet reader's software for dropletreader's workflow, data acquisition and data analysis. Baoxia Liu andLingxiang Zhu provided the ddPCR reagent kits. Lingxiang Zhu, WenjunYang, Na Gao, Gaoshan Jing and Yong Guo provided financial supportto this project.

References

[1] H. Kinugasa, K. Nouso, K. Miyahara, Y. Morimoto, C. Dohi, K. Tsutsumi, H. Kato,T. Matsubara, H. Okada, K. Yamamoto, Detection of K‐ras gene mutation by liquidbiopsy in patients with pancreatic cancer, Cancer 121 (13) (2015) 2271–2280.

[2] M.Y. Chang, S. Ahn, M.Y. Kim, J.H. Han, H.-R. Park, H.K. Seo, J. Yoon, S. Lee, D.-Y. Oh, C. Kang, One-step noninvasive prenatal testing (NIPT) for autosomal re-cessive homozygous point mutations using digital PCR, Sci. Rep. 8 (1) (2018) 2877.

[3] T.F. Pinheiro-de-Oliveira, A.A. Fonseca Jr., M.F. Camargos, M. Laguardia-Nascimento, A.M. de Oliveira, A.C. Cottorello, A. Goes-Neto, E.F. Barbosa-Stancioli,Development of a droplet digital RT-PCR for the quantification of foot-and-mouthvirus RNA, J. Virol. Methods 259 (2018) 129–134.

[4] J. Zou, B. Duffy, M. Slade, A.L. Young, N. Steward, R. Hachem, T. Mohanakumar,Rapid detection of donor cell free DNA in lung transplant recipients with rejectionsusing donor-recipient HLA mismatch, Hum. Immunol. 78 (4) (2017) 342–349.

[5] S. Zhou, T. Gou, J. Hu, W. Wu, X. Ding, W. Fang, Z. Hu, Y. Mu, A Highly IntegratedReal-Time Digital PCR Device for Accurate DNA Quantitative Analysis, BiosensBioelectron, 2019.

[6] F. Schuler, M. Trotter, M. Geltman, F. Schwemmer, S. Wadle, E. Domínguez-Garrido, M. López, C. Cervera-Acedo, P. Santibáñez, F. von Stetten, Digital droplet

Table 1Quantification results of HLA-B27 mutants. (Rep.= replicate; Std.Dev.= standard deviation.)

Expected Rep. 1 Rep. 2 Rep. 3 Rep. 4 Mean Std. Dev.

0.0 0.0 0.0 0.0 0.0 0.0 0.010.0 15.1 19.5 10.5 8.3 13.4 5.030.0 35.8 30.4 27.3 32.7 31.5 3.6100.0 79.1 105.2 116.3 124.3 106.2 19.7300.0 309.4 292.1 286.1 265.7 288.3 18.01000.0 1110.0 975.0 1153.3 1063.4 1075.4 76.43000.0 3239.9 3030.1 3228.8 3107.3 3151.5 100.810000.0 9507.9 9595.3 9888.3 9065.7 9514.3 340.4

Table 2Quantification results of HLA-B27 wild types. (Rep.= replicate; Std.Dev.= standard deviation.)

Expected Rep. 1 Rep. 2 Rep. 3 Rep. 4 Mean Std. Dev.

0.0 0.0 0.0 0.0 0.0 0.0 0.010.0 15.1 7.8 10.5 25.0 14.6 7.530.0 35.8 39.0 50.6 16.4 35.5 14.2100.0 104.3 123.5 95.1 89.4 103.1 14.9300.0 354.7 284.8 293.7 272.1 301.3 36.71000.0 1175.0 1032.3 1002.6 1186.7 1099.2 95.23000.0 3082.1 3043.5 3369.2 3374.4 3217.3 179.110000.0 9131.4 9909.0 10086.1 8979.1 9526.4 552.4

X. Zhu, et al. Talanta 206 (2020) 120200

8

PCR on disk, Lab Chip 16 (1) (2016) 208–216.[7] A.C. Hatch, J.S. Fisher, A.R. Tovar, A.T. Hsieh, R. Lin, S.L. Pentoney, D.L. Yang,

A.P. Lee, 1-million droplet array with wide-field fluorescence imaging for digitalPCR, Lab Chip 11 (22) (2011) 3838–3845.

[8] T. Gou, J. Hu, W. Wu, X. Ding, S. Zhou, W. Fang, Y. Mu, Smartphone-based mobiledigital PCR device for DNA quantitative analysis with high accuracy, Biosens.Bioelectron. 120 (2018) 144–152.

[9] Y. Zhu, Q. Fang, Analytical detection techniques for droplet microfluidics—a re-view, Anal. Chim. Acta 787 (2013) 24–35.

[10] B.J. Hindson, K.D. Ness, D.A. Masquelier, P. Belgrader, N.J. Heredia,A.J. Makarewicz, I.J. Bright, M.Y. Lucero, A.L. Hiddessen, T.C. Legler, High-throughput droplet digital PCR system for absolute quantitation of DNA copynumber, Anal. Chem. 83 (22) (2011) 8604–8610.

[11] C.A. Milbury, Q. Zhong, J. Lin, M. Williams, J. Olson, D.R. Link, B. Hutchison,Determining lower limits of detection of digital PCR assays for cancer-related genemutations, Biomolecular Detection and Quantification 1 (1) (2014) 8–22.

[12] L. Mazutis, J. Gilbert, W.L. Ung, D.A. Weitz, A.D. Griffiths, J.A. Heyman, Single-cellanalysis and sorting using droplet-based microfluidics, Nat. Protoc. 8 (5) (2013)870–891.

[13] T. Rane, C. Puleo, K. Liu, Y. Zhang, A. Lee, T. Wang, Counting single molecules insub-nanolitre droplets, Lab Chip 10 (2) (2010) 161–164.

[14] M. Srisa-Art, A.J. DeMello, J.B. Edel, High-throughput DNA droplet assays usingpicoliter reactor volumes, Anal. Chem. 79 (17) (2007) 6682–6689.

[15] C. Benz, H. Retzbach, S. Nagl, D. Belder, Protein–protein interaction analysis insingle microfluidic droplets using FRET and fluorescence lifetime detection, LabChip 13 (14) (2013) 2808–2814.

[16] K. Leung, H. Zahn, T. Leaver, K.M. Konwar, N.W. Hanson, A.P. Pagé, C.-C. Lo,P.S. Chain, S.J. Hallam, C.L. Hansen, A programmable droplet-based microfluidicdevice applied to multiparameter analysis of single microbes and microbial com-munities, Proc. Natl. Acad. Sci. U. S. A. 109 (20) (2012) 7665–7670.

[17] J.R. Lakowicz, Principles of Fluorescence Spectroscopy, Springer, 2013 US.[18] D.M. Coen, Quantitation of rare DNAs by PCR, Current Protocols in Molecular

Biology 56 (1) (2001) 15.7. 1–15.7. 8.

[19] R. Prakash, K.V. Kaler, An integrated genetic analysis microfluidic platform withvalves and a PCR chip reusability method to avoid contamination, Microfluid.Nanofluidics 3 (2) (2007) 177–187.

[20] K.D. Dorfman, M. Chabert, J.-H. Codarbox, G. Rousseau, P. De Cremoux, J.-L. Viovy, Contamination-free continuous flow microfluidic polymerase chain reac-tion for quantitative and clinical applications, Anal. Chem. 77 (11) (2005)3700–3704.

[21] O. Frey, S. Bonneick, A. Hierlemann, J. Lichtenberg, Autonomous microfluidicmulti-channel chip for real-time PCR with integrated liquid handling, Biomed.Microdevices 9 (5) (2007) 711–718.

[22] X. Jiang, W. Jing, L. Zheng, S. Liu, W. Wu, G. Sui, A continuous-flow high-throughput microfluidic device for airborne bacteria PCR detection, Lab Chip 14 (4)(2014) 671–676.

[23] J. Madic, A. Zocevic, V. Senlis, E. Fradet, B. Andre, S. Muller, R. Dangla,M. Droniou, Three-color crystal digital PCR, Biomolecular Detection andQuantification 10 (2016) 34–46.

[24] S. Su, G. Jing, M. Zhang, B. Liu, X. Zhu, B. Wang, M. Fu, L. Zhu, J. Cheng, Y. Guo,One-step bonding and hydrophobic surface modification method for rapid fabri-cation of polycarbonate-based droplet microfluidic chips, Sens. Actuators B Chem.282 (2019) 60–68.

[25] O. Gangisetty, D.S. Reddy, The optimization of TaqMan real-time RT-PCR assay fortranscriptional profiling of GABA-A receptor subunit plasticity, J. Neurosci.Methods 181 (1) (2009) 58–66.

[26] A. Wright, The Photomultiplier Handbook, Oxford University Press, 2017.[27] G. Karlin-Neumann, F. Bizouarn, Digital PCR: Methods and Protocols, Springer New

York, 2019.[28] K.S. Stephens, Reliability Data Analysis with Excel and Minitab, ASQ Quality Press,

2012.[29] X. Zhu, S. Su, Z. Peng, D. Wang, X. Rui, F. Wang, X. Liu, B. Liu, L. Zhu, W. Yang,

N. Gao, G. Huang, G. Jing, Y. Guo, A density-watershed algorithm (DWA) methodfor robust, accurate and automatic classification of dual-fluorescence and four-cluster droplet digital PCR data, Analyst 144 (16) (2019) 4757–4771.

X. Zhu, et al. Talanta 206 (2020) 120200

9