Droplet Digital Enzyme-Linked Oligonucleotide Hybridization Assay for Absolute RNA Quantification

We present a continuous-flow droplet-based digital Enzyme-Linked Oligonucleotide Hybridization Assay (droplet digital ELOHA) for sensitive detection and absolute quantification of RNA molecules. Droplet digital ELOHA incorporates direct hybridization and single enzyme reaction via the formation of single probe-RNA-probe (enzyme) complex on magnetic beads. It enables RNA detection without reverse transcription and PCR amplification processes. The magnetic beads are subsequently encapsulated into a large number of picoliter-sized droplets with enzyme substrates in a continuous-flow device. This device is capable of generating droplets at high-throughput. It also integrates in-line enzymatic incubation and detection of fluorescent products. Our droplet digital ELOHA is able to accurately quantify (differentiate 40% difference) as few as ~600 RNA molecules in a 1 mL sample (equivalent to 1 aM or lower) without molecular replication. The absolute quantification ability of droplet digital ELOHA is demonstrated with the analysis of clinical Neisseria gonorrhoeae 16S rRNA to show its potential value in real complex samples.

sensitivity simultaneously. For example, Walt and Duffy et al. have successfully implemented single enzyme based approaches for absolute quantification of bacterial genomic DNAs 17 . Pioneering works in single enzyme-based digital assays [18][19][20] have led us to explore the possibility of using linear hybridization for digital RNA detection and quantification.
In this study, we report a continuous-flow droplet digital enzyme-linked oligonucleotide hybridization assay (droplet digital ELOHA) capable of absolute quantification of RNA molecules with high sensitivity. Droplet digital ELOHA incorporates direct hybridization and single enzyme reactions [18][19][20] , enabling a digital reverse-transcription-free RNA detection. Specifically, RNA targets are first allowed to hybridize to DNA capture probes that have been conjugated to magnetic beads. This is followed by hybridization to enzyme-labeled single-strand DNA detection probes. The magnetic beads are subsequently encapsulated into a large number of droplets, mixed with the enzyme substrate, incubated, and directly detected in a single integrated microfluidic chip. This enables facile digital assays without the need for material transfer between instruments. Furthermore, the continuous-flow operation removes the restriction of the total number of reactions imposed by the footprint of the device that is prevalent in micro-fabricated chamber-based digital assays.

Design and Fabrication of Microfluidic Devices.
A casting mold was fabricated using SU8-3050 photoresist and single layer soft lithography on 4 inch silicon wafer. The detailed layout of microfluidic design is presented in Supplementary Information (Figure S1). In our microfluidic devices, we used length-variable Tygon tubing to perform the incubation process in a continuous-flow fashion, which offers both manufacturing ease and incubation time flexibility as compared with designs that use several SU8 thicknesses for droplet generation and incubation 21,22 . The size of the nozzle for droplet generation is 10 μ m and the height of the microfluidic chip is 20 μ m. The microfluidic chips were made of polydimethylsiloxane (PDMS) by casting onto SU8 mold. The PDMS replica was permanently bonded with cover glass (130 μ m thickness, Ted Pella) through oxygen plasma treatment. The microfluidic chips were further treated with Aquapel to render microfluidic channel surfaces hydrophobic before each use.
Operation of Microfluidic Devices. A 50 μ L solution of prepared magnetic beads and 50 μ L RGP substrate solution were first introduced into Tygon tubes of diameter ~500 μ m. This was then pushed by FC-40 oil using syringe pumps (Harvard Apparatus). We used a 60 μ L/h pumping volumetric flow rate for the oil phase and 10 μ L/h for both aqueous phases. The oil phase used in our experiments consists of FC-40 oil and 5% poly(ethylene glycol) di-(krytox-FSH amide) surfactant by weight. A 50 cm long Tygon tube (~102 μ L in volume) was used for continuous incubation of droplets for about 77 min at room temperatures. The microfluidic chip was placed on a custom designed optical stage. The excitation laser beams were focused into the detection region on the chip using a 40× objective (Thorlabs RMS40X-PF, NA 0.75, focal depth~0.6 μ m). The detection region is a 10 μ m sized restriction which examines the incubated droplets one by one. Fluorescence data was continuously acquired from the detection region while upstream droplets were still being generated (Supplementary Movie). Running the assay at room temperature helped avoid droplet evaporation issues that may arise from thermal cycling on chip 8 . The droplets were imaged with a 4× objective before and after passing through the incubation tubing to verify the droplet stability.
Optical Setup and Data Acquisition. The custom built optical setup for measuring droplet fluorescence is sketched in Fig. 1(b). The optical setup is capable of dual laser excitation (488 nm and 552 nm, OBIS lasers from Coherent, Inc.), as well as simultaneous dual color detection, including the 'green' (centered at 520 nm) and the 'red' (centered at 628 nm) wavelength bands. Laser power of 1 mW was used for excitation. The optical setup also has a trans-illumination LED source for imaging the droplets on the chip. A custom-built LabVIEW program was used to control fluorescent data acquisition using single photon counting module (SPCM-AQRH13) with 0. Due to a 'digital' nature of the assay, the detection is insensitive to the droplet size variation. A total amount of N (~1 million) droplets were analyzed for a specific experiment. The droplet intensity of these N droplets usually follows a bimodal distribution. The droplet populations with lower fluorescence intensity correspond to 'negative' reactions, where no single enzyme is present within these droplets.
, where I bg is the droplet background intensity that corresponds to the highest probability in the negative droplet population. This method greatly removes false positives from random changes in fluorescence during data acquisition.
Single Enzyme Activity in Droplets. As-received Sβ G was mixed with 500 μ L of 2 mM sodium azide to yield 2 mg/mL Sβ G stock solution, which was stored at 4 °C. Various Sβ G concentrations were prepared before each experiment by diluting the stock solution with the following buffer: 100 mM Tris-HCl (pH 8.0), 2 mM KCl, 1 mM MgCl 2 and 0.2 mg/mL BSA. RGP substrate at the concentration of 500 μ M was prepared in 100 mM Tris-HCl (pH 8.0), 2 mM KCl, and 1 mM MgCl 2 .

Magnetic Beads Captured Single Enzyme Activity in Droplets.
A total of 20 million streptavidin coated magnetic beads were resuspended into 20 μ L of 2X binding and washing (BW) buffer (10 mM Tris-HCl (pH 8.0), 1 mM EDTA, 2 M NaCl, and 0.1% Tween-20), incubated with 3 pmol biotinylated capture oligos (5-/5BiotinTEG/GT TGT CAA GAT GCT ACC GTT CAG AG-3) for 30 min at room temperature. The beads were then washed 3 times with 1X BW buffer and resuspended into 20 μ L of 1X BW buffer, which were further incubated with 7 pmol biotinylated complementary oligos (5-/5Biosg/CT CTG AAC GGT AGC ATC TTG ACA AC-3) for 3 hours at room temperature. Beads were then washed 3 times and made into 20 aliquots of 10 μ L solutions (each had about 1 million magnetic beads). Each aliquot was incubated with Sβ G for 4 hours and washed 3 times. Mock-up ssDNA Quantification. Streptavidin coated magnetic beads were functionalized with biotinylated capture oligos (5-/5BiotinTEG/GT TGT CAA GAT GCT ACC GTT CAG AG-3). With excessive capture oligos removed, bead solution was used to capture mock-up ssDNA (5-TTG ACG  GCG AAG ACC TGG ATG TAT TGC TCC TCT GAA CGG TAG CAT CTT GAC AAC-3) at various concentrations. After that, 1 pmol biotinylated detection oligos (5-TA CAT CCA GGT CTT CGC CGT CAA/3Bio/-3) were used to form a sandwich structure, from which Sβ G was tagged to the biotin site.
Synthetic Neisseria gonorrhoeae 16S rRNA in Human Total RNA Background. In order to test the assay's specificity towards RNA quantification in human total RNA background, we spiked clean 16S rRNA synthetic samples into 150 ng of human total RNA (equivalent to around 15,000 human cells).
Clinical Neisseria gonorrhoeae 16S rRNA from Isolated Cell Culture. A commercially available RNA extraction kit (miRNeasy Mini Kit, Qiagen) was utilized to extract RNA from bacterial cells, which were cultured from an isolation collected from a patient (Division of Infectious Diseases-Johns Hopkins Medicine). The concentration of eluted 80 μ L of RNA sample was around 192.9 ng/μ L (evaluated by Nanodrop). The protocols used to capture and label the Gonorrhoeae 16S rRNA were similar as those used in the synthetic targets.

Results
Two Steps of Dependent Poisson Process. Our sample preparation steps (forming enzyme labeled complexes on magnetic beads) share a similar procedure as that of digital ELISA 9,18 . As shown in inset i of Figure (a), RNA molecules were firstly captured and concentrated on magnetic beads through specific capture oligonucleotides, and then further hybridized with detection oligonucleotides labeled with enzyme reporters (Sβ G). The probability of finding x enzymes on a single magnetic bead is given by the Poisson statistics, where λ eb is the average enzyme per magnetic bead (equivalently, average RNA molecules per magnetic bead). When RNA molecules are much less than magnetic beads (e.g., λ eb ~ 0.1), most magnetic beads (~99.5% for λ eb ~ 0.1) will accommodate only 1 or 0 enzyme labeled complex. After sample preparation step, these single enzyme labeled beads, together with fluorogenic substrate, were loaded into our microfluidic chip to generate digitalized droplets using a flow focusing method (inset ii of Fig. 1(a)). The probability of finding y beads in a single droplet also follows the Poisson statistics, where λ bd is the average number of magnetic beads in each droplet. If λ bd ~ 1, Poisson statistics predicts more than 8% droplets contain more than 2 magnetic beads, which will greatly introduce the non-linearity for 'digital' counting of single molecules. In our experiment, 1 million magnetic beads in 50 μ L of loading buffer were merged with another 50 μ L of fluorogenic substrate to form droplets of size ~10 pL. As a result, λ bd is around 0.1, which indicates that 90.48% droplets will contain no magnetic Scientific RepoRts | 5:13795 | DOi: 10.1038/srep13795 beads, 9.05% droplets will contain just one magnetic bead, and only 0.47% will contain more than 2 magnetic beads. In this regime, digital counting of single molecules is feasible. The next section details the quantification method for the aforementioned Poisson processes (Eqs (1) and (2)).

Absolute Quantification Based on Two Dependent Poisson Processes. After incubation of
magnetic beads with fluorogenic substrate, each droplet develops either a high or low concentration of florescent product, depending on whether the droplet contains enzyme reporter(s) (Fig. 1(a)). The droplet florescence intensity is measured by a custom designed optical system ( Fig. 1(b)). We define those droplets without any enzyme reporters as negative droplets. There are several scenarios that yield negative droplets. For example, the droplet contains 0 beads, the probability of which is P(Y = 0), or the droplet contains 1 bead and this particular bead carries no enzyme, the probability of which is P(Y = 1)P (X = 0), or droplet contains 2 beads and both beads carries zero enzyme, the probability of which is P(Y = 2) P(X = 0) P(X = 0), and so forth. Therefore, the probability of all negative droplets is given by, As a result, the probability of positive droplets is p = 1 − p neg . Substituting Eqs (1 and 2) into Eq. (3), the probability of positive droplets can be expressed as, Note that for arbitrary λ ed and λ bd , positive droplets can have many sophisticated combinations of beads and enzymes, including those with multiple beads and enzymes per droplet (e.g., two beads, one bead has 2 RNA linked enzymes and the other one has 1 RNA linked enzymes). For a linear 'digital' counting of RNA molecules in our continuous flow chip, we need both λ ed and λ bd to be small (~0.1) such that each single droplet will only contain 0 or 1 magnetic bead, and each single magnetic bead carries 0 or 1 enzyme.
To relate the number of captured RNA molecules with the measured percentage of positive droplets, we can derive the average RNA number per bead (equivalent toλ ed ) from Eq. (4), By knowing the total number of beads (N b , usually 10 6 beads in our experiments), total testing sample volume loaded into the microfluidic chip (v s , 100 μ L) and the droplet size (v d , 10 pL), the absolute number of captured RNA molecules is related to the percentage of positive droplets as, Since percentage of positive droplets (p) can be determined by examining the droplet florescence intensity one-by-one (inset iii of Fig. 1 (a)), absolute RNA quantification is possible through Eq. (6), with known experimental parameters of N b , v s , and v d . It is not required to know exactly which droplet contains a single magnetic bead. We believe this novel absolute quantification method can be broadly extended to other cases involving dependent Poisson processes, like double emulsions 24 .
System Evaluation with ssDNA. Since droplet digital ELOHA depends on the single enzyme activity in pL-sized droplets, we first verified that single enzyme activity is indeed functional and detectable in our droplet platform (Supplementary Text and Figures S2-S5). Before running absolute RNA quantification, we performed proof-of-concept experiments with synthetic single stranded DNA. As shown in Fig. 2(a-c), with increasing ssDNA concentration, more droplets have higher fluorescence intensity, indicating an increased population of droplets containing enzyme reporters. Figure 2(d) shows the change in the percentage of positive droplets as a function of ssDNA concentrations ranging from 100 fM to 1 aM. The percentage of positive droplets is saturated at around 10% due to Poisson distribution of magnetic beads into droplets. If the average number of beads per droplet (λ bd ) is ~0.1, ~90% of the droplets will contain no magnetic beads (Supplementary Figure S6). As a result, the maximal percentage of positive droplet is ~10%, no matter how many enzyme labeled complexes are associated with each magnetic bead during hybridization process. On the other hand, the false positive background, determined by experiments with no ssDNA target (mean+3σ ), is found to be ~0.01% (dashed red line in Fig. 2). The saturation level of ~10% and the background level of ~0.01% sets a ~3 log linear range of our continuous flow droplet digital ELOHA based on two Poisson process.
The false positive background of ~0.01% is likely due to the non-specific binding of Sβ G enzyme onto the magnetic bead surface, which determines the limit of detection (LOD). The LOD of ssDNA quantification in 100 μ L testing sample is around 10 aM, or 602 copies of ssDNA. Among different non-amplification based nucleic acid detection methods, the LOD of droplet digital ELOHA is comparable to SiMoA assay 17 , and better than methods that use gold nanoparticle scattering 25 , NanoString nCounter 26 or imaging 27 (see Supplementary Table 1

for comparison).
We also found the false positive background level remains almost the same when a 10 times larger volume of ssDNA is evaluated. Figure 3 shows the results for ssDNA in 1 mL volume, with LOD around 1 aM for 1 mL of ssDNA sample, or ~600 copies of ssDNA. This absolute detectable number (~600) is very close to that of the 100 μ L sample (Fig. 2(d)), indicating that the background is mainly due to the enzyme non-specific binding to the magnetic beads. In addition, the ssDNA capture efficiency in 1 mL is very close to that in 100 μ L sample when operating in the linear region (69 ± 7% for 100 μ L sample and 63 ± 6% for 1 mL sample, see Supplementary Tables 2 and 3). The pre-concentrating effect of magnetic beads makes our droplet digital ELOHA an ultrasensitive tool for clinically relevant mL-sized sample volume 2,28 .
Synthetic RNA Quantification. After system validation using ssDNA, we set out to explore our continuous flow droplet digital ELOHA for absolute RNA quantification without reverse transcription. Neisseria gonorrhoeae16S rRNA was used as a model RNA target. Similar protocols were used to form the enzyme labeled DNA-RNA-DNA complex on the magnetic beads, as in the case of ssDNA. Device operation was the same as before. Figure 4 shows the results for clean synthetic 16S rRNA samples (triangles). The measured percentage of positive droplets agrees very well with the values predicted from the two-step Poisson statistics, indicating that the capturing and formation of the enzyme labeled complex is indeed functional with RNAs. No reference or standard is needed for quantification using Eq. (6). The false positive background is ~0.01%, attributed to the non-specific binding of enzyme to the magnetic beads.
To further verify the assay's specificity towards the target RNA sequence, we spiked 16S rRNA sample into 150 ng of total human RNA background, which directly mimics the clinical test samples using an RNA extraction kit. The empty squares in Fig. 4 show the results from the spiked sample. It is found that the quantification of synthetic 16S rRNA is not affected by the background of 150 ng of human total RNAs (~15,000 copies of background molecules). The RNA capture efficiency by magnetic beads is found to be ~56 ± 1% for clean RNA sample and ~53 ± 5% for samples with total RNA background (see Supplementary Tables 4 and 5). The RNA capture efficiency is found to be less than the ssDNA case. The detection limit for both clean and spiked RNA sample is determined to be around ~600 molecules copies, an intrinsic background limit in our system mainly set by the sample preparation step (non-specific binding of enzymes).
Clinical RNA Sample Quantification. Lastly, we applied our continuous flow chip for clinical RNA samples to evaluate its performance in real world situations. RNAs extracted from an isolation collected from patients were used as testing targets, which were specifically captured and hybridized on magnetic beads to form the enzyme labeled complex. Before we ran RNA quantification using our droplet digital ELOHA, the copy number of Gonorrhoeae 16S rRNA was evaluated by real-time quantitative reverse transcription PCR (RT-qPCR). By using a reference standards (See Methods for details), the unknown starting 16S rRNA in 1 μ L volume is found to be ~3 × 10 6 copies ( Fig. 5(a)). Gel-electrophoresis was adapted to verify the major amplification product is indeed the designed target (54 bp length, Fig. 5(b)). Figure 5(c) shows the results from droplet digital ELOHA, where the RNA copy numbers were calculated through the measured value of positive percentage of droplets. It is found that the quantified number of RNA falls short of the expected value determined from the aforementioned RT-qPCR method. This is due to the fact that the RNA capture efficiency in ELOHA is not 100%. In fact, the RNA numbers determined by ELOHA are about half of the input copy number, which can be justified by the ~50% RNA capture efficiency (Supplementary Tables 4 and 5). Though sample-preparation induced quantification bias exists in almost any analytical assay, it is noteworthy that the RNA capture efficiency in ELOHA is fairly stable (~50%) across the linear dynamic range of the assay, therefore no calibration of the capture efficiency is required for each measurement. Currently we do not have a definitive conclusion to the ~50% capturing efficiency. We speculate this may due to the hybridization kinetics in low concentrations (digital assay region). In fact, a similar digital assay also reported a capture efficiency around 50-60% 9 . This is an interesting biophysical phenomenon worthy of further exploration, though it's beyond the scope of this work. In addition, droplet digital ELOHA shows a consistently smaller coefficient of variation (<10%, Supplementary Figure S8) for triplicate and is able to reliably differentiate 40% difference in RNA quantity (Fig. 5(c)). This high accuracy of RNA quantification makes ELOHA very suitable for various applications such as absolute quantification of viral load 29 and mRNA gene expression 30 .  Discussion By using magnetic beads to capture individual RNA molecules and digitally counting these molecules in a statistically significant quantity of droplets, digital ELOHA enables sensitive, specific and accurate quantification of RNA molecules without the need to generate cDNA using reverse transcription (as in digital RT-PCR), an enzymatic process that may introduce quantification bias. This makes digital ELOHA suitable for applications such as gene expression evaluation. Droplet digital ELOHA provides a novel method for a magnetic bead-based digital assay with a two-step Poisson process, which is particularly suitable for end-point detection based continuous-flow platforms, since no imaging is required. Our continuous-flow platform permits a small device footprint and a high throughput of analysis.
Though a very promising technology, some foreseeable limitations exist in the current platform and warrant further optimization. First of all, as with any hybridization based approach, the specificity may be compromised for long RNA strands which may result in false positive signals. On the other hand, for short RNA molecules (e.g., miRNA, ~22 nucleotides), the approach may suffer from low hybridization efficiency which will results in negative signals. These issues can be mitigated by using more specific probes 31 . Second, non-specific binding of enzyme reporters to magnetic beads should be suppressed for an improved dynamic range and detection limit. The current sample preparation results in a background of ~0.01%, which we believe can be further decreased by optimizing the surface block beyond the conventional BSA method. Third, the singleplex nature of the current platform is insufficient for multiplexed quantitative microRNA (miRNA) profiling 32 . Nevertheless, by incorporating multiple colors into our 'cytometry' system, it is a reasonable goal to achieve multiplexed detection in droplet digital ELOHA.