Digital Nucleic Acid Detection Based on Microfluidic Lab-on-a-Chip Devices

Microfluidic lab-on-a-chip (LOC) technologies have been developed as a promising alternative to traditional central laboratory-based analysis approaches over several decades due to the capability of realizing miniaturized multiphase and multistep reactions. In the field of nucleic acid (NA) diagnosis, digital NA detection (dNAD) as a single-molecular-level detection is greatly attributed to the perfect combination of NA amplification and microfluidic LOC techniques. In this chapter, the principle, classifi‐ cation, advances, and application of dNAD will be involved. In particular, the focus will be on chip-based dNAD for giving a deep interpretation of the analysis and evalua‐ tion of digital detection. The future prospect of dNAD is also anticipated. It is sure that dNAD by means of microfluidic LOC devices as the promising technique will better serve the ambitious plan of precision medicine through absolute quantitation of NA from individuals.


Introduction
It has been clearly investigated that nearly all of the diseases possess a series of biomarkers associated with nucleic acid (NA) molecules during the development of biological researches [1][2][3][4][5]. Determining these NA molecules and their intercellular and extracellular changes is a well-worked strategy for estimating therapy efficacy, monitoring minimal residual diseases, unveiling the mechanisms of cellular signal transduction, and so on [6][7][8]. To reflect individual genetic differences, single-molecule level quantitation of NAs has been increasingly con-cerned recently due to its superiority on analytical sensitivity and accuracy [9,10]. Furthermore, single NA molecule detection is also highly preferred as the calibration strategy for nextgeneration sequencing (NGS) to better serve recently proposed ambitious plans of precision medicine [11][12][13]. Thus, testing NAs, especially in single-molecule level, plays an essential role in modern biological researches and diagnosis fields.
At present, the widely used approaches in detecting NA molecules are quantitative polymerase chain reaction (qPCR) and quantitative reverse transcription-PCR (qRT-PCR). Apart from the capability of real-time monitoring the amplification, they can be applied to quantify the target NA molecules through two common strategies: relative quantification and absolute quantification. The former is based on internal reference genes (namely, housekeeping genes) to normalize and reflect fold differences in expression levels of mRNA, which is commonly interpreted as cDNA [14,15]. The latter can provide the exact number of targeted molecules using an established standard curve of the change in quantification cycles with known molecule number of NA standards [16][17][18][19]. However, qPCR is compromising the ability of single-molecule quantitation analysis [20,21]. Alternatively, when PCR meets microfluidic or nanofluidic chips, a highly sensitive NA quantification technique [digital PCR (dPCR)] emerges, estimating NAs advantageously at a single-molecule analysis level [22].
At the end of 20th century, the first concept of dPCR was proposed by Vogelstein and Kinzler [23]. Since the concept was proposed, many dPCR platforms have been launched for several decades based on differently designed microsystems, including femtoliter array, spinning disk, SlipChip, droplet, microfluidic formats, and so on [24][25][26][27][28][29][30]. Some even have been successfully paced into industrial phase because of the superiority of testing and the promising application. Currently, several vendors in the biological industry, such as Fluidigm, Bio-Rad Laboratories, Life Technologies (ThermoFisher), RainDance Technologies, and Formulatrix, have launch individually their commercialized dPCR devices (Figure 1).
Apart from dPCR, digital isothermal NA amplification (dINAA) devices also arouse great concern. Unlike dPCR, dINAA leans on the isothermal NA amplification, which can be carried out at a consistent temperature, obviating the requirement of highly stable thermocycling devices. Thus, when targeting practical point-of-care testing (POCT) devices, dINAA is superior to dPCR. However, viewed from the principle of realizing digital detection, the concept of dINAA is the same as that of dPCR, just replacing PCR with isothermal amplification. In particular, due to loop-mediated isothermal amplification (LAMP) displaying as the best promising method among a lot of isothermal NA amplifications, digital LAMP (dLAMP) is the first dINAA developed [31,32]. Later, other dINAAs have been reported, such as digital multiple displacement amplification (dMDA), digital isothermal multiple-self-matchinginitiated amplification (dIMSA), digital recombinant polymerase amplification (dRPA), and so on [33][34][35][36][37][38]. However, the development of dINAA devices is still in the research stage, as the commercial products have not been launched yet.
As of now, more and more researchers are enthusiastic about the potential of digital NA detection (dNAD) based on microfluidic Lab-on-a-Chip (LOC) devices, since an increasingly significant role has been played in single-cell analysis, early diagnosis of cancer, prenatal diagnosis, and so on. In this chapter, we will concentrate on the principle, classification and advances, analysis and evaluation, application, and future prospects of dNADs that are accomplished either through commercialized LOC devices or the devices our laboratory or other laboratories have established.

Principle of dNAD
According to the strategies of amplifying NA, dNAD or single-molecule NA detection can be divided into dPCR and dINAA. However, both of them share the same principle.
Generally speaking, the principle of dNAD is composed of three core steps [39]. First, the original sample should be partitioned into thousands or hundreds of thousands of individual microreactions, endeavoring to make each contain nearly one target molecule. Second, the number of "positive" microreactors indicated either in a real-time reaction or in an endpoint reaction is counted and analyzed. Third, the concentration of nonpartitioned sample is calculated using certain statistical methods. In theory, if the number of microreaction is more enough or the number of target molecules is less enough, one reaction unit with positive signal represents one target molecule. However, in fact, a positive partition may contain more than one molecule. Therefore, in calculating the target's true concentration, a Poisson distribution is adopted in hope to correct the results. Therefore, dNAD can be considered as a binary output (present or absent like "1" or "0" in computer science) measurement, giving a direct and high-confidence NA molecule's measurement method [10,40]. Compared to conventional tubebased NA detection, digital analysis is superior in realizing the absolute quantification with high sensitivity, high precision, and low ambiguity, avoiding the requirement of establishing a standard curve.

Classification and advances of dNAD
In the early stage of digital detection, the used materials are 96-and 384-microwell plates. Then, due to the rapid development of microfluidic chip techniques, an increasing number of digital detection devices emerge. Also, a variety of materials have been used individually or jointly, such as silicon wafer, quartz, glass, polydimethylsiloxane (PDMS), polymethyl methacrylate, and so on. According to the approaches to partition reaction mixture, the currently launched dNAD methods can be roughly grouped into three categories: plate-based dNAD (pdNAD), droplet-based dNAD (ddNAD), and chip-based dNAD (cdNAD). On structural design, each classified dNAD has the advantages and disadvantages, and the corresponding commercial devices are also developed. In this subchapter, we are going to narrate their features and recent advances in either commercial or research aspects.

pdNAD
At present, most of the pdNADs are established as dPCR devices, but they are not hard to be developed as dINAA platforms. As the first generation of dNAD, plate-based dPCR (pdPCR) was first conducted using plenty of commercially available 96-and 384-microwell plates [23,41]. The biggest benefit for this kind of digital platform is saving to create the plates that have been widely used in conventional PCR. Each microwell undertakes each microreaction; therefore, the high sensitivity and accuracy of detection lean entirely on the enough number of microwells. However, actually, the number is hard to be reached just using microwell plates.
Another problem causing the embarrassment is the volume of reagents required [42]. For each microwell, more than 5 μL are needed, and the cost of reagents inevitably daunts most researchers, let alone the application for POCT. To break the barriers, some researchers made modification. As shown in Figure 2, Morrison et al. deceased the volume of microreaction into 33 nL using a stainless steel plate (25 mm in width and 75 mm in length) in which up to 3072 microholes (320 μm in diameter) were created [43]. In contrast, the required volume was reduced to 1/64, and the throughput was increased by 24-fold, although it had the comparable sensitivity to the past. At present, this technique has been applied to commercial devices in 2009, the OpenArray RealTime PCR System from Life Technologies. However, as the number of reaction units increases, the problem turns into how to efficiently load the reagents. Consequently, it has to use some ancillary equipment-like microarray spotter or mechanical arms, which in turn raises the cost and is cumbersome.
Considering the embarrassing situation, in the second half of 2013, Life Technologies launched the next-generation digital detection device, the QuantStudio 3D dPCR system [42,44]. It is a simple and affordable platform to provide the reliable and robust dPCR. The device used a special plate (10 mm in width and 10 mm in length) where a total of 20,000 hexagonal microwells are fabricated. The volume of each microwell is 0.8 nL, and each reaction well is isolated absolutely from its neighbors. At present, the system has been applied to the absolute quantification of viral load, low-level pathogen detection, sensitive genetically modified organism (GMO) detection, differential gene expression, copy number variation (CNV), NGS library quantification, and rare mutation analysis [45][46][47][48][49][50]. Although the cost of reagents is reduced, the system still calls for supporting instruments to load the reagents, amplify the sample, and read the results.
For high-throughput sample analysis, 96-and 384-microplate formats are still of use. Formulatrix introduced a new commercial high-throughput pdPCR device termed as constellation dPCR. The device brings the digital analysis to a 96-sample microplate format, and the socalled high-throughput results from the preformation of dPCR on 96 samples at once and up to 384 samples per hour. As required, the number of partitions for each microwell in the plate can be easily increased, and it reaches 496 for the 96-microplate format.

ddNAD
ddNAD can go back to emulsion PCR (ePCR) [51][52][53][54]. ePCR is widely used for NGS ( Figure  3) [55]. After generating a DNA library, the fragments of genomic DNA are attached to the beads, because their surface is modified with oligonucleotide probes whose sequences are complementary to the sequences of the fragments. When the beads are compartmentalized into water (the PCR reagent)-oil emulsion droplets, plenty of microreactors are produced. Since each bead captures single-stranded DNA fragment, in theory, ePCR can amplify it down to one DNA molecule. However, it is not easy to partition the fragments and beads into one droplet simultaneously, and then the performance of ePCR suffers from variety. Benefitting from the rapid development of microfluidic LOC techniques, ddNAD also enjoys a huge boom in recent years. dPCR is still the main part in ddNAD, but droplet-based dINAAs including dLAMP, dRPA, digital rolling circle amplification (RCA), and digital hyperbranched RCA (HRCA) are showing up more and more [34][35][36][37]56]. Beer et al. successfully created picoliter-scale water-in-oil droplets by using a shearing Tjunction in a fused-silica device in 2008 [57]. The NA used for the device was RNA; therefore, an off-chip valving system was integrated to stop the droplet motion, because a different thermal cycling was required for reverse transcription and subsequent PCR amplification. Each droplet contained the PCR mixture of single-copy template, primers, and reaction buffer, which was really termed as digital detection. One year later, Mazutis et al. developed a method for high-throughput dINAA platform in a 2 pL droplet-based microfluidic system [35]. The isothermal HRCA was used to perform the DNA amplification in droplets. This platform was demonstrated to allow fast and accurate digital quantification of the template. In 2011, Zhong et al. reported another picoliter-scale droplet-based multiplexing dPCR platform, breaking the one target per color barrier of qPCR [58]. The number of droplets generated reached more than 10 6 , which was enough for enhancing the likelihood that only one DNA molecule was amplified in each droplet. Given its great potential in application, RainDance Technologies launched the commercial digital detection system with the highest droplet throughout, the RainDrop dPCR system. Unfortunately, the system may consume up to 50 μL reagents per sample. Considering this point, the system may be not proper for rare sample detection. At the same year, Hindson et al. also established a high-throughput droplet-based dPCR (ddPCR) platform [59]. A total of 2 million droplets were generated, and the droplets were then transferred into a 96-well plate for TaqMan probe-based PCR. Finally, to read out the results, a flow cytometry-like double-channel fluorescence detection device was used in a microfluidic chip, in which droplets went through one by one. The platform was confirmed to realize the accurate measurement of germ-line CNV, discriminate the mutant molecules from the wild molecules with 10 5 -fold excess, and absolutely quantify circulating fetal and maternal DNA from cell-free plasma. Based on the platform, the first commercial ddPCR system was launched by QuantaLife in 2011, but in the end of that year Bio-Rad Laboratories purchased the company and launched the QX100 ddPCR. Recently, the new version, QX200 ddPCR system, is also available.
Apart from the process of droplet generation and subsequent NA amplification, other approaches to generate droplet are also reported. As shown in Figure 4, Shen et al. described a SlipChip to create droplet array [26]. The SlipChip was composed of two glass plates, in which elongated wells were designed to overlap and form the fluidic path for reagent loading. After sample loading, the simple slipping of the two plates broke the path, removing the overlap among wells and generating 1280 droplet array (2.6 nL for each). The device had a reservoir preloaded with oil, so each microreactor was absolutely isolated from each other during thermal cycling. Finally, the results were read out using endpoint fluorescence intensity. The biggest advantage of SlipChip is the capability of realizing multistep manipulation of plenty of microvolumes to form droplet array in parallel. Attributed to the remarkable feature, until now, the SlipChip has been applied to perform immunoassays, protein crystallization, multiplex PCR, dPCR, dLAMP, dRPA, and so on [26,34,[60][61][62][63][64].
To make the droplet more stable and to easily collect the amplified products, Leng et al. invented an agarose droplet-based single-molecule ePCR device [51]. The agarose performed the unique thermoresponsive sol-gel switching property, and a microfluidic chip was designed to produce uniform agarose solution droplets. Schuler et al. applied centrifugal step emulsification to the fast and easy generation of monodisperse droplets [37]. Only by adjusting the nozzle geometry (depth, width, and step size) and interfacial tensions droplets with desirable diameters could be produced. Using this droplet device, dRPA was successfully established for the absolute quantification of Listeria monocytogenes DNA concentration standards within 30 min.
In ddNAD, the microreactors are generated by carefully titrating emulsions of water, oil, and chemical stabilizer; therefore, there is no requirement of the walls of microwells to separate the microreactors. Compared to pdNAD, ddNAD can easily achieve higher throughput via a microdroplet generator to produce hundreds of thousands of droplet reactions per sample. However, the workflow of ddPCR is complicated, referring to generating droplet, transferring droplet, sealing microplate, conventional PCR, and reading out the signal by other devices.

cdNAD
The development of cdNAD is greatly attributed to the rapid progress of microfluidic techniques, which can realize the low cost, low volume, and high-throughout paralleled NA detections. In the last several decades, microreactors in cdNAD are mainly formed either by the mechanical compartmentalization of PDMS or by the succeeding isolation via immiscible liquid phase. In particular, for PDMS-based chips, the establishment of multilayer soft lithography (MSL) techniques developed by Unger et al. in 2000 also gives a huge boost, making the high-density microwells, micropumps, and microvalves easily fabricated [65]. Based on different power sources to partition reagents, cdNAD can be divided into three categories: integrated fluidic circuit (IFC) cdNAD, self-priming compartmentalization (SPC) cdNAD, and localized temporary negative pressure (LTNP)-assisted cdNAD as well as other cdNADs.

IFC cdNAD
The outstanding feature of IFC chip is the special design of separated and interlaced liquid and gas channels, as shown in Figure 5. Taking advantage of the high elasticity of PDMS, hundreds or thousands of microreaction units are formed rapidly when gas channels are added with pressure.
In 2006, Ottesen et al. used the IFC chip to achieve dPCR analysis [66]. A total of 1176 microreaction units were produced by controlling accurately the integrated microvalves and removing the conventional microarray spotter and plates. Now, the IFC chip-based dPCR (IFC cdPCR) platform is successfully established for commercial purpose by Fluidigm. As the first vendor to commercialize dPCR device, Fluidigm provides two IFC-based systems, the BioMark HD and EP1 systems. In the two systems, the PCR reagents are mixed and partitioned automatically, the thermocycling is integrated, and the results can be read out after reaction. The BioMark HD system can offer real-time detection for each tiny reaction and eliminate false positives according to the data, so the system is also available to qPCR. Compared to BioMark HD, EP1 is just an endpoint detection machine, giving the binary output-like data whether or not the microreaction occurs. Recently, Fluidigm also combines with Olink to detect human protein biomarkers based on proximity extension assay (PEA) technology. Until now, the dPCR device from Fluidigm has been applied to single-cell analysis, early diagnosis of cancer, and prenatal diagnosis.
In 2011, Heyries et al. developed a megapixel dPCR platform in which 10 6 microunits were fabricated, and the microreaction's volume reached down to 10 pL (Figure 6) [67]. The density of the microreactors reached up to 440,000/mm 2 , which was the highest density for IFC platform. On detection performance, this device was able to discriminate one mutant molecule from 10 5 wild molecules and achieve the discrimination of a 1% difference in chromosome copy number. After the platform, in 2012, Men et al. published anther dPCR platform possessing the lowest volume (36 fL) of microreactors until now. Its density of microreactors was more than 20,000/mm 2 [24]. After loading the reagents into all microreactors simultaneously, the deformation of a PDMS membrane was used to completely seal the filled microreactors. Due to the femtoliter-level microreactors fabricated, the device can greatly reduce the consumption of reagent and sample. For dINAA, IFC chip is also combined with isothermal MDA to develop dMDA for enumeration of total NA contamination [33]. On detection of microbial genomic DNA fragments, dMDA performs higher sensitivity with orders of magnitude than qPCR.
By making the microchamber smaller or increasing its number, IFC cdNAD has a potential to be developed into a digital detection platform with higher throughout, higher density, and higher discrimination ability, but this platform still relies on the control system of integrated microvalves and micropumps to load and partition the reagents, which is hard to be applied towards POCT. Furthermore, narrowing the size of microchamber endlessly may have an impact on the efficiency of NA amplification.

SPC cdNAD
Targeting practical POCT devices, currently proposed plate-based, droplet-based, and IFC cdNADs are confronted with the huge challenge of demand for peripheral control instrument, for instance, external syringe pumps, droplet generation devices, and plenty of integrated microvalves and micropumps. Upon this challenge, the built-in power-driving, self-partitioning, easy-to-use, and low-priced SPC cdNADs were developed by our laboratory. The builtin power results from the gas solubility and permeability of PDMS, because PDMS remains absorbing gas and letting gas go through them, although PDMS is in a solid state in chips [68].
The chip possesses the prominent feature of SPC, resulting from the used material of silicone elastomer PDMS, a relatively cheap material, which possesses high gas solubility and permeability. When the fabricated chips are evacuated, a negative pressure environment is formed due to the gas solubility of PDMS, which can service as a self-priming power to let the sample solutions be sucked into each reaction chamber and sequentially the biocompatible oil to seal and separate each filled chamber. Thus, in realizing dNAD, thousands of independent microwells can be created automatically, avoiding the external control system, which is superior to IFC cdNAD. Currently, SPC cdPCR and dINAAs [SPC chip-based dLAMP (cdLAMP) and SPC chip-based dIMSA (cdIMSA)] have been developed by our laboratory [30,31,69,70].

SPC cdPCR
The chip shown in Figure 7 is composed of three PDMS layers, two glass coverslips, and a waterproof layer [30]. The three PDMS layers include an inlet and outlet layer, a microwell array layer, and a blank layer. In the microwell array layer, a total of 5120 reaction microwells (150 μm in width, 150 μm in length, and 250 μm in height) are equally distributed in four separate panels. Each microwell contains down to 5 nL solution. The inlet and outlet layers have four 0.5-mm holes and four 2.5-mm holes in diameter punched as injection ports and suction chambers when aligning to the outlet of the microwell array layer, respectively. For mechanical stability, the blank layer coats the microwell array layer with the waterproof layer embedded. The waterproof layer is made of low permeability fluorosilane polymer, which beneficially prevents the evaporation during the step of denaturing template DNA at 95°C in PCR. One of the glass coverslips with plasma pretreated are used to seal the microwell array, and the other one is pressed on the upper surface of the SPC chip for mechanical stability at the end of microchip operation.
MSL techniques are used to fabricate the SPC chip. The chip patterns are designed by a software of CorelDRAW X4 and printed on transparency films using a high-resolution printer to create the masks of channels and microwells. The photoresist material used are SU8 serials, which are a high-contrast, epoxy-based negative photoresist. Several 4-inch silicon wafers are adopted as the mold substrate. The PDMS to replicate the SPC chip is the silicone elastomer PDMS, which is composed of PDMS base (A) and catalyst (B) at certain ratios. Because dPCR is an endpoint detection, the reaction components including PCR buffer, primers, labeled probes, and templates have to be mixed before loading into the chip. Each diluted template in the mixture is individually injected into the three panels of the chip, allowing the samples to be compartmentalized completely and each chamber contains down to 5 nL solution. A Maestro Ex In-Vivo Imaging System (CRI Maestro, USA) is used to capture the fluorescent image of the microchip after dPCR. As a new generation of microfluidic chips, SPC cdPCR has been successfully applied to the absolute quantification of β-actin DNAs and the lung cancerrelated genes.

SPC chip-based dINAAs (cdINAAs)
Although PCR is widely adopted and used as a standard analytical technique in molecular diagnosis, it is remarkably confined when applied to field and POCT due to the facts that it requires nonportable thermocycling facilities, its process of obtaining results is cumbersome, and the whole amplification takes 2 h or more. Also, SPC cdPCR confronts the same defects. Accordingly, SPC cdLAMP and SPC cdIMSA are established by our laboratory. As simple and easy world-to-chip fluidic devices, SPC cdINAAs have the great potential in POCT for the developing countries.
Similar to SPC cdPCR, SPC cdLAMP is also the completely valve-free and SPC device ( Figure  8) [31]. It is also made mainly of PDMS and fabricated by MSL techniques. In size, the SPC chip used for dLAMP is the same as the dPCR-used chip; however, in composition, it does not contain a waterproof layer because of the absence of the DNA denaturing step in LAMP. For the microwell array PDMS layer of dLAMP-used SPC chip, a total of 4800 microwells (150 μm in width, 150 μm in length, and 300 μm in height) are fabricated and they are also equally distributed into four panels (each contains 1200 chambers), and the interval for two closed chambers is 150 μm. The big difference from the dPCR-used SPC is that the rectangular chambers are located vertically on the main channels and the branch channels link to chamber without orthogonal turning points. On performance, the SPC cdLAMP can precisely calculate the absolute DNA concentration. To conduct the data acquisition and analysis of SPC cdLAMP, the Maestro Ex In-Vivo Imaging System is employed. However, the imaging system is too cumbersome and expensive to allow the e POCT, especially in the less developed regions.
Herein, an easy-to-use and cost-efficient smartphone-based dLAMP POCT device platform is also established by our laboratory.
SPC cdIMSA is an updated version of SPC cdLAMP, in which the LAMP is replaced by IMSA and a mixed dye is used to establish a label-free and sensitive dual-fluorescence detection for on-chip IMSA [70,71]. The used SPC chip for dIMSA is the same as that for dPCR without any Similarly, the mixed dye-based dual-fluorescence detection has a potential in the POCT application of SPC cdLAMP and other SPC chip-based dINAAs.

LTNP cdNAD
As another POCT-oriented LOC device, LTNP chip resembling SPC chip also uses the gas solubility of PDMS to load solution into chambers (Figure 9) [29]. However, the measure of evacuating the LTNP chip is different from the SPC chip. For the former, the gas solubility and permeability of the chip itself and another PDMS layer coating on the chip services simultaneously as the real-time power source of evacuating with a syringe filter. For the latter, preevacuating by a vacuum pump is indispensable, which calls for robust and efficient approaches of sealing and packaging chips [30].
The LTNP cdPCR has been already exploited [29]. The chip consists of three parts, which are a lamina-chip layer (LCL), a vaporproof layer (VPL), and a syringe filter-like microfluidic device (μfilter) with helical channels. The μfilter has two parts. One is designed for generating the LTNP through pulling the plug of the syringe connected at one end of helical channels. The helical channel in the μfilter is 200 μm in width and 40 μm in depth. The other is used for  Currently, LTNP cdPCR has been used for the detection of keratin 19 in A549 lung carcinoma cells [29]. Additionally, the integrated LTNP chip with NA enrichment, isolation, and digital detection functions has been developed and successfully applied to detect bovine DNA in ovine meat for food adulteration detection [72,73]. Figure 10 [32]. In this device, the reagents were partitioned into the microchambers based on an inherent fluidic phenomenon that the interplay between fluidic forces and interfacial tension could cause the self-dispersion of an income aqueous fluid into an array of chambers prefilled with an immiscible fluid. Therefore, the sample loading could be realized with manual or automated syringe pumps or external air pressure, removing the hydraulic valves or mechanical action. Less than 2 μL sample was used for the accurate quantification of relative and absolute DNA concentrations. To improve the efficiency of partitioning samples, a new generation of SD chip was invented with close to 100% efficiency in 2013 [74]. In 2014, based on the SD chip, digital RT-PCR was developed to absolutely quantify mRNA from single cells [75]. Due to the simplicity and robustness of the SD chip, the SD cdNAD is an inexpensive and easy-to-operate digital detection device. In 2010, Sundberg et al. designed a spinning disk platform to achieve dPCR [25]. The disk was an inexpensive and disposable plastic disk-like chip. Differing from other approaches, 1000 nL microwells were generated by passive compartmentalization through centrifugation, and the volume of each well was 33 nL in average. The whole process, including disk loading, thermocycling, and fluorescent imaging, only costs less than 35 min. However, this kind of cdNAD performed some defects, such as the tedious plastic disk manufacturing and the probable NA adsorption to the disk.

Analysis and evaluation of cdNAD
This part focuses on the analysis and evaluation of cdNAD. The contents involve four aspects, which are the dehydration of microreaction, the dynamic range, the response of cdNAD, and the precision of cdNAD.

Dehydration of microreaction
In PCR, dehydration is attributable to the repeated denaturing steps, whereas, for isothermal NA amplification, reacting at a consistent temperature (e.g., 63°C) for at least 1 h results in dehydration. Also, the influence of dehydration is different according to the different materialbased chips for cdNAD.
As we know, the microfluidic PDMS chips suffer from dehydration heavily. Actually, the dehydration degree is determined by the thickness of the PDMS layer between the top of the chamber and the waterproof layer. If v f is the total volume fraction of reaction reagent, v f is defined by the function of where A and h refer to the designed area and height, respectively. Using the formula of C sat_25°C ×P vap_70°C /P vap_25°C , a saturated concentration of water vapor in PDMS at 70°C (C sat_70°C ) can be calculated as 400 mol/m 3 . Then, the maximum fractional loss of water (fl max ) from the reaction chambers is defined by where ρ water is the density of water and M water is the molar mass of water.

Dynamic range
When high concentration targets are loaded, the chip panel can be completely saturated, whereas, for low targets, the chip panel still possesses the capability of realizing digital detection. Therefore, the theoretical dynamic range is determined by the high concentrations that make the chip completely saturated.
In this situation, the occurrence of an empty chamber is a small probability event, and due to statistical independence between the chamber number and the total empty chamber number, the event can be modeled as a random Poisson process. Then, the occurrence probability of t empty chamber number, P (n=t, λ), is defined by P(n = t, λ) = λ t e −λ t ! . In the function, λ equals to the mean number of empty chambers in panels. For each empty chamber in each panel, namely, a chamber containing 0 molecule, the probability P(n=0,λ)=e -λ =e -m/N , where λ is the ratio of the loaded molecule number (m) to the chamber numbers (N). Therefore, λ is equal to the product of the probability P(n=0,λ) and the N chamber numbers, namely, λ=Ne -m/N . Finally, If λ is more than 10, the probability of the chip panel is completely saturated, P(n=0,λ)=e-10, which is less than 10-4 and can be as the acceptable failure rate. Then, λ=N e -m/N >10, and the maximum loaded molecular number (mmax) can be calculated.

Response of cdNAD
According to the distribution of molecules across the chip, the Poisson distribution is adapted to calculate the original concentration of target stock solution.
For each chip panel, the probability (P) of having the number (n) of NA molecules per chamber is P(n,λ)=(λ n e -λ )/n!, where λ is the ratio of loaded molecule number to chamber number of each panel. For λ, it is equal to the average molecule number per chamber, which can be expressed by the equation of λ=C 0 X dil V, where C 0 is the target's original concentration, V is the volume of each chamber, and X dil is the dilution factor of diluted targets used in the panel.
If the stock solution is diluted in k fold, the X dil =1/k, and particularly, the X dil of stock solution is 1. When a chamber contains 0 molecule, the probability P(n=0,λ)=e -λ . As for a chamber capturing one or more molecules, the probability P(n≧1,λ)=1-P(n=0,λ)=1-e -λ . After dPCR, a chamber with observed positive signal suggests that at least one target molecule is captured. Therefore, the ratio (f) of observed positive chamber number to chamber number of each panel, namely, the observed fraction of positive chambers, equals to the probability P(n≧1,λ), which is f=P(n≧1,λ)=1-e -λ . Then, λ=ln(1-f)=-C 0 X dil V, and a linear variation relationship is exhibited in terms of the regression curve equation between ln(1-f) and X dil , because C 0 and V are constant values. Then, the target's original concentration C 0 can be calculated based on the slope of the curve (-C 0 V).
In addition, because the Poisson distribution is a particular case of a binomial distribution, the distribution of the positive chamber number (x) in each panel is classified as the binomial distribution. The probability distribution P(x) is therefore influenced by the probability p=P(n≧1,λ)=1-P(n=0,λ)=1-e -λ . Based on the nature of binomial distribution, x is equal to the mean, namely, x=Np=N (1-e -λ

Precision of cdNAD
When the chamber number in each panel is very large, the Poisson distribution is approximate to normal distribution, which is also called Gaussian distribution. Use the parameters in the discussion of response and dynamic range above, and let y be a random variable representing the number of positive chambers that capture at least one molecule. Also, we can know that its mathematical expectation or mean μ is N (1-e -λ ), and its variance σ 2 is Ne -λ (1-e -λ ). Then, the probability density function associated with y is as follows: If the precision of cdNAD is defined as the minimum difference in concentration (∆λ/λ) that is reliably detected with more than 99% true positive and more than 99% true negative rate, this situation corresponds to 4.6σ separation in the mean (μ) for two Gaussian distributions. That is to say, in the case with small ∆λ, ∆μ/∆λ=4.6σ. Then, ∂[N(1-e -λ )]∆λ/∂λ=Ne -λ ∆λ=4.6[Neλ (1-e -λ )] 1/2 , and ∆λ/λ=4.6(e λ -1) 1/2 /(λN 1/2 ). For different total chamber numbers (N) in chip panel, the precision (∆λ/λ) to expected molecules per chamber (λ) could be plotted.

Application of dNAD
Compared to conventional NA detection methods, dNAD provides better sensitivity, better precision, higher tolerance, and the NA's absolute quantification. Until now, dNAD has been applied to a variety of research fields, including pathogen detection, food safety, clinical diagnosis (genetic instability estimation and early cancer), prenatal diagnosis, quantitative analysis of gene expression, and NGS library quantification. Although dPCR is still the most widespread type of dNAD, in resource-limited regions, it is confined due to the requirement of thermal cycling and robust temperature control. To overcome this awkwardness, dINAAs without thermal cycling are of great interest, which also enlarge the application of dNAD, especially for POCT. For this subchapter, the recent advances on application of dNAD will be narrated.

Pathogen detection
Honestly speaking, on pathogen detection, the advantages of most conventional NADs on both analytical sensitivity and specificity are inherited by the corresponding digital formats. In addition, dNAD has the capability of realizing NA's absolute quantification. Therefore, dNAD can be used as the ultrahigh sensitivity and more accurate method for viral load determination or bacterial quantification. Kelley et al. established a duplex ddPCR assay for high-precision methicillin-resistant Staphylococcus aureus (MRSA) analysis [76]. By assaying 397 clinical samples, a good agreement with the reference assay for both qPCR and ddPCR assays was indicated. Strain et al. demonstrated and estimated the application of ddPCR for a highly precise measurement of HIV DNA [77]. Total HIV DNA and episomal 2-LTR (long terminal repeat) circles in cells that were all isolated from infected patients were targeted. Compared to qPCR, ddPCR performed a significantly increased precision (5-fold for total HIV DNA and >20-fold for 2-LTR circles), making it an alternative for the measurement of HIV DNA from clinical specimens.
dNAD also enables the direct detection without NA extraction due to the ability of partitioning target NA and nontarget components (e.g., inhibitors) into different microwells. Pavšič et al. employed two dPCR platforms (QX100 ddPCR system from Bio-Rad Laboratories and the BioMark HD IFC cdPCR system from Fluidigm) for the direct quantification of two wholevirus materials of human cytomegalovirus (HCMV) without DNA extraction [78]. It was demonstrated that direct quantification by both dPCRs could provide repeatable measure-ments of viral DNA copy numbers, giving a closer agreement with the actual viral load than that with either dPCR or qPCR of extracted DNA.
Although dNAD certainly possesses the superiority of absolute quantitation, some reports fail to demonstrate the advantage, which calls for the requirement of optimization when establishing dNAD. For instance, Boizeau et al. used ddPCR to provide an absolute quantitation of HBV genome molecules [79]. However, the results indicated that qPCR assays remained more sensitive than ddPCR when used for low HBV DNA levels, suggesting that optimization of ddPCR was still necessary, especially on accurately differentiating the positive from negative in samples with very low levels of target DNA molecules.

Food safety
Food safety mainly refers to two parts: the minoring of foodborne pathogens and the detection of GMO. For example, human noroviruses (NoV) and hepatitis A virus (HAV) are two foodborne enteric viruses that have caused the vast majority of nonbacterial gastroenteritis or some fatal infectious hepatitis. Coudray-Meunier et al. conducted a comparative study of IFC RT-cdPCR and conventional RT-qPCR when quantifying the NoV and HAV from lettuce and water samples, proving that the IFC RT-cdPCR assay was more tolerant to inhibitory substances from lettuce samples [80]. Also, the IFC RT-cdPCR may be useful for standardizing the quantification of enteric viruses in bottled water and lettuce samples. Fu et al. used the BioMark HD system equipped with a 48.770 digital array to develop cdPCR for GMO detection without pretreatment [81]. The CaMV35s promoter and the NOS terminator were selected as the targets, and nine events of GMOs (MON810, MON863, TC1507, MIR604, MIR162, GA21, T25, NK603, and Bt176) were collected to determine the specificity. The results showed that the cdPCR could achieve a discrimination of down to 0.1%, lower than the labeling threshold level of the EU, allowing highly sensitive, specific, and stable GMO screening detection. Dalmira et al. developed a duplex ddPCR assay to characterize the certified reference materials (CRMs) in terms of T-nos/hmg copy number ratio in maize [82]. After optimization using a central composite design, the duplex ddPCR method realized the absolute limit of detection (LOD) and limit of quantification (LOQ) of 11 and 23 copy number T-nos, namely, relative LOD of 0.034% and relative LOQ of 0.08%, respectively. The dynamic range of T-nos/hmg ratio ranged from 0.08% to 100%. The results indicated that the duplex ddPCR assay was useful for characterizing CRM candidates on T-nos/hmg ratio.
The bias caused by reliance on quantitative standards may have an impact on the results of qPCR, which is not beneficial for water monitoring and microbial source identification. Therefore, Cao et al. employed a duplex ddPCR to simultaneously quantify Enterococcus spp. and the human fecal-associated HF183 marker for evaluating the water quality [83]. The results demonstrated that ddPCR performed greater tolerance of inhibition than qPCR, with one to two orders of magnitude higher at inhibitor concentrations. Also, ddPCR brought about remarkably improved precision, although a lower upper LOQ than qPCR was indicated.
For food safety, it is indispensable to identify and quantify the meat products for unveiling species fraud and product mislabeling during food processing. Tian et al. used LTNP cdPCR to detect bovine meat in ovine meat [72]. Floren et al. successfully applied two-step ddPCR for precise quantification of cattle, horse, and pig in processed meat product, with the nuclear F2 gene targeted [84].

Clinical diagnosis
At present, clinical diagnosis including genetic instability estimation and early cancer is also the main application field for dNAD. It has been clearly confirmed that the genetic instability of human cells is one of the causes of cancer, including somatic mutation, allelic imbalance, loss of heterozygosity, CNVs, and single nucleotide variations (SNVs). Accordingly, how to discriminate the rare mutant gene from abundant normal NAs attracts great concerns of researchers. Interestingly, it is also the question that initiates the concept formation of dNAD.
As de novo CNV may be caused by reprogramming somatic cells into induced pluripotent stem cells, Abyzov et al. used qPCR and ddPCR to detect and estimate this phenomenon. The results showed that, in parental fibroblasts, at least half of the CNVs are indicated as low-frequency somatic genomic variants [85]. Boettger et al. successfully applied a ddPCR approach to analyze father-mother-offspring trios from HapMap at specific sites within region 1 on the investigation of inference of complex CNV and single nucleotide polymorphism (SNP) haplotypes at the human 17q21.31 locus [86].
Detecting biomarkers associated with tumor formation, development, and drug evaluations on gene expression is the useful approach for early cancer diagnosis. Up to now, various biomarkers have been identified, and dNAD reflects the unique advantage of absolute quantitation of their gene expression [87]. Zhu et al. applied SPC cdPCR to the detection of three lung cancer-related genes (PLAU, ENO2, and PLAT). cdPCR yielded comparable results to qPCR, illustrating that the established platform had the ability of realizing absolute quantitation for gene expression [30]. Floren et al. employed ddPCR to detect the BRAF-V600E and V600K mutations in melanoma circulating tumor with high sensitivity [84]. The study demonstrated that ddPCR performed 200-fold increased sensitivity than competitive allelespecific PCR (castPCR), giving an LOD of 0.0005% when combined with whole-genome amplification (WGA). Through noninvasive analysis of circulating free plasma DNA, Gevensleben et al. determined the presence of oncogenic amplification by developing a plasma DNA ddPCR assay targeting HER2 [88]. In the independent validation cohort, ddPCR could reach a positive and negative predictive value of 70% and 92%, respectively. The results suggested that ddPCR had the potential to the analysis of any locus amplified in cancer, not only in metastatic breast cancer. Beaver et al. also employed a ddPCR assay for the detection of circulating plasma tumor DNA (ptDNA) in patients with early-stage cancer [89]. A total of 30 tumors were first analyzed by Sanger sequencing for common PIK3CA mutations, and ddPCR was then used to analyze their extracted DNA for the same mutations. This ddPCR-based accurate mutation detection platform was demonstrated to be of great use for early-stage breast cancer.
As a new generation of cancer biomarkers, abnormally DNA methylation in the gene's regulatory regions can affect identical residues that may cause the cancer. The analysis of methylated genes therefore becomes more and more important in cancer research. For example, Li et al. developed a sensitive bead-based ddPCR for the quantification of DNA methylation, which could detect down to one methylated DNA molecule in approximately 5000 unmethylated DNA molecules from plasma or fecal samples [20]. Weisenberger et al. developed an improved DNA methylation detection with single-molecule high-resolution based on IFC cdPCR and successfully identified the breast cancer-specific hypermethylation phenomenon in the CpG islands of RUNX3, CLDN5, and FOXE1 [90].

Prenatal diagnosis
Prenatal diagnosis is usually carried out by invasive or noninvasive approaches [91]. The invasive method (e.g., amniocentesis) refers to inserting needles into the uterus, which is timeconsuming (several weeks required) and risky to the fetus [92]. In contrast, noninvasive method seems to be more rapid and safer. Currently, noninvasive prenatal diagnosis mainly involves ultrasonography of the womb or detecting cell-free fetal DNA (cffDNA) in maternal serum and plasma. The latter is also termed as noninvasive prenatal testing (NIPT). Appling dNAD to NIPT is the emerging noninvasive approach with high sensitivity and high precision [93][94][95][96][97]. Gu et al. employed the QX100 ddPCR system from Bio-Rad Laboratories to detect cffDNA for the risk of methylmalonic acidemia, confirming that ddPCR is a cost-effective and noninvasive prenatal method when diagnosing known mutations associated with Mendelian disorders [98]. Pornprasert and Prasing implemented a ddPCR for the deletion of α(0)thalassemia Southeast Asian (SEA)-type deletion [99]. The study showed that ddPCR might be an alternative technology to routine clinical diagnosis. Barrett et al. adopted 12×765 digital array chips to establish an IFC cdPCR to analyze the cffDNA for NIPT of sickle cell anemia [94].
The results suggested that the built IFC cdPCR is a useful method to determine the genotype of fetuses at risk for sickle cell anemia. Meantime, the report also illustrated that it was essential to optimize the fractional fetal DNA concentration.
Because most cffDNA fragments were approximately 200 bp in size and, in the early gestation, the cffDNA occupies low percentage (mostly <10%) in maternal plasma, efficient methods for the extraction of cffDNA are in great need. Holmberg et al. thereby estimated two commercial platforms to extract cffDNA: the Akonni Biosystems TruTip technology and the Circulating Nucleic Acids Kit from Qiagen [100]. Determined by QX100 ddPCR system and qPCR, the extracted products from two platforms performed similar results.

NGS library quantification
As known to all, the establishment of NGS libraries still leans mainly on manual bench-top procedures, which is slow and inefficient [101]. To solve it, Kim et al. invented an automated digital microfluidic LOC-based sample preparation for the NGS. Compared to the conventional methods, digital microfluidic LOC platform is cost-efficient and has high throughout [102]. Similarly, Thaitrong et al. integrated a droplet-based microfluidic LOC system with a unit of capillary-based reagent delivery and the quantitative CE module to develop an automated quality-control platform for NGS [103]. Besides, White et al. confirmed that dPCR is able to provide sensitive and absolute calibration for NGS, enabling direct sequencing without titration runs and with sufficient precision [11]. Fu et al. developed a picoliter-scaled droplet digital WGA (ddWGA) platform for realizing uniform and accurate single-cell sequencing, bringing about significantly improved amplification evenness and accuracy for the simultaneous detection of CNVs and SNVs in single-cell level [104]. Weisenberger et al. applied the Fluidigm BioMark Digital Array to establish digital bisulfite genomic DNA sequencing with high resolution and high sensitivity [105]. The results showed that IFC cdPCR was a fast and reliable method for the single-molecule-scaled detection of DNA methylation information.

Conclusion and future prospects
Undoubtedly, there is growing interest in dNAD, because it allows the more precise NA quantification, the higher discrimination of rare NA mutants, and the more reproducible and less susceptible to inhibitors than the traditional NA methods. Consequently, dNAD has full potential to influence the development of biology research, clinical diagnosis, the safety of food and environment, and other research fields.
To enhance the impact of this promising technique and push it towards clinical application, the MIQE for dNAD (dMIQE) was also published [106]. Based on dMIQE, the experimental protocols are standardized, the efficient utilization of resources is maximized, and the data are adequately assisted. However, the promising technique of dNAD still confronts some shortcomings. On one hand, although the development of microfluidic LOC offers a lot of dNAD device platforms, these devices perform low functional integration, and the supporting detection approaches lean primarily on real-time fluorescence scanning or the endpoint analysis of CCD camera-captured images, which, to some extent, adds the real cost and also has impacts on the true detection accuracy. Therefore, in the future, it will become a general trend that dNAD devices are highly integrated with multiple functions including cell or singlecell capture, cell lysis, and NA enrichment and purification, employing more advanced supporting detection technology. Particularly, for ddNAD, the strategy of droplet generation is one of the developing directions. For instance, Tanaka et al. currently created a hands-off autonomous preparation method of monodisperse emulsion droplets using a degassed PDMS chip [107]. Jeong et al. used a specially designed three-dimensional monolithic elastomer device to create a kiloscale droplet generation [108]. According to the snap-off mechanism, Barkley et al. also invented a novel technique to generate monodisperse droplets [109].
On the other hand, dNAD is actually the digital version of NAD, thereby possessing the same disadvantages (e.g., bias or nonspecific amplification) as most sequence-based NA amplification methods. Based on this point, how to improve or guarantee the NA amplification fidelity in microreactors is one of the future prospects of dNAD. It should be noted that the optimized reaction conditions for NA amplification in microreactors might be different from those in bulk state, meaning that the optimization of reaction system is indispensable. Furthermore, the precision of dNAD is greatly influenced by the number and size of microchambers (for cdNAD) or droplet (for ddNAD), which in turn challenges the fabrication of microfluidic chip and the uniformity of partitioning. Accordingly, we also anticipate that, in the future, there will be some novel strategies developed to realize digital detection not only based on partitioning reagents.
Certainly, the application of dNAD will be enlarged by combining with other molecular assays, especially for single-cell analysis and single-cell genomic sequencing. For example, dNAD can combine with proximity ligation or PEAs to achieve single-molecule protein biomarker detection. Additionally, in the future, ongoing comparison tests of dNAD and qPCR will roundly prove the detection superiority of dNAD in many research fields. Also, dNAD will become the promising POCT-oriented research area for the ambitious plan of precision medicine.
Conclusively, dNAD based on microfluidic LOC devices will continue to provide further opportunities for determining NA molecules, protein molecules, and other biomolecules towards deep analysis with high sensitivity and precision.