Main

Precise control over the binding affinities between biomolecules is one of the critical means by which nature allows thousands of types of biomolecule to operate in tandem with molecular-level control in heavily crowded environments. Moreover, molecules with well-defined binding affinities for a specific target—thus called affinity reagents—are essential for many applications, including biosensing, diagnostics, imaging and therapeutics. For example, the binding affinity between a target analyte and its specific biorecognition element is the determinative factor that defines the dynamic range of a biosensor1,2. High-affinity recognition molecules with low dissociation constants (Kd) are sought for point-of-care applications to enable rapid binding events and low limits of detection. On the other hand, low-affinity reagents with fast association and dissociation rates have been used in real-time monitoring applications3,4,5, because they can resolve the dynamic changes in their target analyte level. The advancement of detection modalities with extremely high resolution and low background levels has also made it possible for low-affinity binding reagents to achieve high detection sensitivity2,6,7. Hence, to meet the diverse requirements of different sensing scenarios, it is highly desirable to be able to generate recognition molecules with desired binding affinities against a specific target.

Nucleic-acid aptamers have garnered substantial interest as versatile affinity reagents and as promising alternatives to antibodies over the past three decades8,9,10. In recent years, aptamers have demonstrated great potential in real-time monitoring and wearable sensing applications11,12,13,14,15. Compared to antibodies, aptamers offer several unique features, including high stability, compact size, cost-effective mass synthesis, minimal batch-to-batch variation, non-immunogenic properties and amenability to site-specific chemical modification8,16,17. Despite these attractive features, the use of aptamers in biosensing is relatively limited. One reason for this is a lack of high-quality aptamers for many biomolecular targets16. Conventional procedures for aptamer selection through SELEX are labour-intensive, time-consuming and incapable of producing aptamers with programmable binding affinities. A typical SELEX experiment requires 10 to 15 rounds of selection to allow the target-binding sequences to dominate the nucleic acid pool16,18,19. This inefficiency is due to the inability of traditional SELEX to separate aptamer candidates with diverse binding affinities in each round, as well as the introduction of undesired biases—such as polymerase chain reaction (PCR) bias and amplification of parasitic sequences—through repeated rounds of selection. These factors also make the binding affinities of the selected aptamers unpredictable, requiring a number of sequences to be tested to identify an aptamer with desired properties.

Recent advancements in SELEX technologies with improved sequence partitioning efficiency, such as CE-SELEX20, Non-SELEX21, M-SELEX22 and particle display selection technology23, have provided solutions for the aptamer-selection challenge. However, none of these methods are able to quantitatively generate aptamers with defined binding affinities. Here we describe a new approach, referred to as Pro-SELEX, for quantitative isolation of aptamers with programmable binding affinities. Our workflow consists of particle display, magnetic particle sorting, high-dimensional and parallelized selection and high-content bioinformatics. Using this workflow, we are able to profile the binding performance of individual aptamer candidates at different target concentrations in a single round of selection. As a proof of concept, we selected aptamers specific to human myeloperoxidase (MPO), an emerging diagnostic and prognostic biomarker of coronary artery disease24,25. We successfully isolated a number of anti-MPO aptamers with pre-defined binding affinities (Kd) and developed algorithms relating selection data to this important biological parameter.

Results

The overall Pro-SELEX approach is summarized in Fig. 1. Aptamer libraries displayed on particles are incubated with the binding target, where they are expected to bind to differing numbers of analyte molecules based on affinity. Magnetic nanoparticles are used to label the aptamer particles (APs) according to number of analyte molecules bound, and the aptamer particles are then sorted into different compartments of a microfluidic chip based on magnetic content. Pools of aptamers are generated based on the material collected in each compartment and sequenced. Comparative bioinformatics are employed to analyse the representation of individual aptamers and confirm the validity of sequences as true-positive hits as well as infer relative affinity ranges.

Fig. 1: Quantitative isolation of aptamers with different binding affinities via a high-dimensional microfluidic approach.
figure 1

a, Overall workflow. Shaded bands represent the target concentration (darker colour indicates higher concentration). b, Isolation of aptamer particles with different levels of target binding using Pro-SELEX chips. c, Separation rationale: simulated linear flow velocity profile. The formation of capture pockets was observed in the valley of X-shaped microstructures. d, Magnitude of linear flow velocity. e, Determination of the median magnetic load of particles captured by Pro-SELEX at different flow rates (8, 16 and 32 ml h−1). Data are presented as mean values, and the error bars represent the standard deviation of three replicates. W represents particles collected from the outlet (waste).

Source data

Development of Pro-SELEX selection platform

To measure the relative affinity of each aptamer sequence in a nucleic acid library, particle display techniques23 can be used to transform individual aptamer candidates from a solution-based aptamer library into monoclonal APs via emulsion PCR, where each particle presents many copies of a single nucleic-acid sequence on its surface (Fig. 1a). The AP libraries can then be incubated with a specific target over a range of concentrations. Based on the Langmuir isotherm, the target binding level of an AP is dependent on the target concentration and the binding affinity of the aptamer coated on its surface, where half of the aptamer molecules will be bound to the target when the target concentration is equal to the Kd of the aptamer. APs with different binding levels can then be separated and collected using a high-resolution microfluidic sorting approach followed by sequencing analysis to identify aptamers with different binding affinities based on their binding performance at different target concentrations.

To accurately partition APs with different levels of target binding, we developed a magnetic labelling and microfluidic sorting approach (Fig. 1b). Polystyrene particles (7 µm in diameter) were used to prepare APs. After incubation of the APs with a biotinylated target of interest, the particles were labelled with anti-biotin magnetic nanoparticles (MNPs, 50 nm in diameter) to convert the target binding into a magnetic content. The labelled mixture was then sorted and isolated using a microfluidic device (Pro-SELEX chip). This device features four capture zones with differing linear velocities to allow differential sorting of APs with varying levels of magnetic content. Owing to the low magnetic susceptibility of MNPs, each capture zone contains microfabricated structures to create localized regions of low flow velocity and enhanced capture dynamics (Supplementary Fig. 1). The first zone exhibits the highest linear velocity and thus retains APs with high magnetic content, as the retaining magnetic force overcomes the drag force exerted by the high flow velocity. The ensuing three zones exhibit gradually reduced linear velocities (Fig. 1d; for simulation information see Fig. 1c). This design allows APs to be separated and captured in different zones according to their target binding levels. We demonstrated that the Pro-SELEX chip is able to sort the particles based on their magnetic content (Fig. 1e and Supplementary Fig. 2). It is worth noting that the stringency and resolution of magnetic separation of the Pro-SELEX chip are programmable by adjusting the chip flow rate (Fig. 1e). Similar device design has been used to perform profiling of cellular proteins and nucleic acids as well as rare cell capture, demonstrating superior throughput and recovery rate compared to fluorescence-activated cell sorting (FACS) and magnetic-activated cell sorting (MACS)26,27,28,29. We now demonstrate the utility of this type of device for aptamer selection.

Validation of the sorting performance of the Pro-SELEX chip

In the first suite of experiments, we used thrombin aptamers as a test case to validate the approach (Fig. 2a). We first tested whether the Pro-SELEX chip could sort APs with different levels of target binding. Particles coated with thrombin-03 aptamer23 (03-particles) with varying target binding levels were prepared by incubating the particles with a range of thrombin concentrations, from 1 pM to 1 nM, followed by labelling with MNPs (Fig. 2b). These 03-particles were then sorted using Pro-SELEX chips at flow rates of 8 ml h−1 and 16 ml h−1, and distinctive sorting profiles were generated (Fig. 2c,d). The particles with high magnetic content (1 nM and 100 pM) were primarily captured in zone 1 of the chip at a flow rate of 8 ml h−1 and partially shifted to zone 2 upon increasing the flow rate to 16 ml h−1, whereas particles with lower magnetic content were mainly collected in later zones and almost all of the non-target bound particles were detected in the waste. This observation indicates that the Pro-SELEX chip is able to sort and capture APs based on their target binding levels, employing several flow rates to stratify the sorting profile.

Fig. 2: Validation of the sorting performance of the Pro-SELEX chip using anti-thrombin aptamers with different binding affinities.
figure 2

a, Workflow of the validation experiments. Z1–Z4, zones 1–4; W, waste. Shaded bands represent the target concentration (darker colour indicates higher concentration). b, Evaluation of the binding performance of thrombin-03-coated APs at thrombin concentrations of 1 nM, 100 pM, 10 pM, 1 pM and 0 pM. The particles are labelled with MNPs and anti-MNPs labelling check reagent-APC and analysed by flow cytometry. c,d, Determination of the capture profile of particle-displayed thrombin-03 aptamer prepared using different thrombin concentrations at flow rates of 8 ml h−1 (c) and 16 ml h−1 (d). Data are presented as mean values, and the error bars represent the standard deviation of three replicates. eg, Analysis of particle-displayed thrombin aptamer 03, TBA and a primer sequence after incubation with 1 µM (e), 1 nM (f) and 1 pM (g) thrombin molecules. The particles are labelled with MNPs and anti-MNPs labelling check reagent-APC and analysed by flow cytometry. hj, Sorting of the particle-displayed thrombin aptamers 03, TBA and a primer sequence using Pro-SELEX chip at a flow rate of 8 ml h−1 with 1 µM (h), 1 nM (i) and 1 pM (j) thrombin molecules. Data are presented as mean values, and the error bars represent the standard deviation of three replicates.

Source data

Next, we explored whether the Pro-SELEX approach could separate APs with different binding affinities. Two types of AP were prepared carrying comparable copy numbers of either thrombin-03 (Kd = 7.04 pM) or thrombin-TBA (Kd = 2.6 nM)23,30 (Supplementary Fig. 3). Both APs, together with primer-coated particles, were tested with three thrombin concentrations and analysed using flow cytometry (Fig. 2e–g) and the Pro-SELEX chip (Fig. 2h–j). When using a thrombin concentration of 1 µM, which is substantially higher than the Kd of both aptamers, both APs became saturated by the target and were captured in zone 1 of the chip. When the target concentration dropped to 1 nM, a substantial separation was observed such that the high-affinity 03-particles remained in zone 1 and the low-affinity TBA-particles were binned in zones 2 and 3. When the target concentration further decreased to 1 pM, most TBA-particles were collected in the waste, whereas a substantial fraction of the 03-particles were still captured in the chip. Notably, less than 0.5% of the primer-coated particles were captured by the chip at a flow rate of 8 ml h−1, regardless of the thrombin concentrations, showcasing the high capture specificity of the Pro-SELEX chip. We further investigated the effect of flow rate on the nonspecific capture of aptamers to ensure that nonbinders can efficiently be removed during the selection process (Supplementary Fig. 4). Together, these results demonstrated that the Pro-SELEX chip is capable of separating APs with distinctive binding affinities, where target concentrations play a prominent role in determining the separation efficiency.

Selection of MPO aptamers using Pro-SELEX

We next explored whether we could use the Pro-SELEX platform to efficiently generate aptamers. We selected MPO as the model target for testing the capability of Pro-SELEX for aptamer selection. An initial random DNA library containing ~1 × 1014 random DNA molecules was first subjected to a pre-enrichment step consisting of three rounds of both counter selections against undesired targets and conventional SELEX towards MPO immobilized beads to remove nonspecific binders and reduce the sequence space of the library (Supplementary Table 1 and Supplementary Fig. 5). The enriched library was then used as a template to prepare AP libraries through emulsion PCR (Supplementary Fig. 6). The AP libraries were tested at five MPO concentrations (10 pM, 100 pM, 1 nM, 10 nM and 100 nM), in an attempt to isolate aptamers with Kd within this range. It was critical to ensure that the MPO molecules were available in excess and not depleted during the target incubation, especially when testing low target concentrations, to confirm that the binding between APs and MPO follows the Langmuir isotherm (Supplementary Table 2). The AP libraries were then sorted by Pro-SELEX chips at two flow rates, 16 ml h−1 and 32 ml h−1 (Fig. 3a). It required less than 20 min for the Pro-SELEX chip to sort 5 × 107 APs, which is faster than the FACS-based sorting method23. Particles captured by each zone within the ten chips were then collected (Fig. 3b,c). As expected, we clearly observed that the fractions of APs captured by Pro-SELEX chips decreased upon lowering the target concentration, and the sorting profiles of the AP libraries were shifted when using a faster flow rate, highlighting the high resolution of our approach.

Fig. 3: Isolation of anti-MPO aptamers with different binding affinities via a high-dimensional microfluidic-based approach.
figure 3

a, Workflow of the selection experiments. Five MPO concentrations (100 nM, 10 nM, 1 nM, 100 pM and 10 pM) were tested at flow rates of 16 and 32 ml h−1 for the screening experiments. Particles were collected from individual capture zones (Z1–Z3) and analysed by high-throughput sequencing and the AptaZ tool for aptamer identification. b,c, APs collected from zones 1–4 of the Pro-SELEX chips at flow rates of 16 ml h−1 (b) and 32 ml h−1 (c). d, Distribution of calculated Z score over the pool of sequences presented in MPO selection. e, Heatmap of fold enrichment in all 30 conditions of sequences with different Z scores. The AptaZ algorithm comprehensively considers the fold enrichment from different conditions to generate Z scores to quantitatively rank the aptamers. FC, fold change. f, Eight candidates with different Z scores were expressed on particles and incubated with Cy5-labelled MPO before analysis via flow cytometry. Each line represents a fitted Langmuir isotherm. Data are presented as mean values, and error bars represent the standard deviation of three replicates. g, The relationship between calculated Z score and validated Kd derived from the eight candidates. The Z score correlates with the Kd value in a linear fashion according to curve fitting.

Source data

The particles collected from zones 1–3 of each chip, together with the unsorted aptamer population, for a total of 31 DNA pools, were subjected to high-throughput sequencing (HTS). We also created a scoring tool, AptaZ, which was employed to comprehensively analyse the fold of enrichment of each aptamer candidate from 30 DNA pools and generate a Z score (details are provided in Supplementary Discussion and Fig. 3d). Aptamer candidates that can survive multiple target concentrations, especially low concentrations at two flow rates, and be repeatedly captured by the chips would generate high Z scores (Fig. 3e). Eight MPO-aptamer candidates, MPO-01 to MPO-08, with diverse Z scores, were randomly chosen for characterization (Supplementary Table 3). All eight aptamers demonstrated an affinity for MPO with Kd values ranging from 227 pM to 27.8 nM (Fig. 3f and Supplementary Fig. 7). A linear relationship was observed between the Kd values of the eight candidates and their calculated Z scores based on curve fitting (Fig. 3g), indicating the potential of using Z scores to search for the aptamer candidates with desired binding affinities.

Quantitative selection of MPO aptamers with desired affinities

We next explored whether the Z score could be used to quantitatively screen aptamers with desired binding affinities. A search band was generated based on the correlation between the Z score and the Kd values of the validated MPO aptamers (Fig. 4a). We set the desired Kd values as 1,000 pM, 3,000 pM, 9,000 pM and 27,000 pM and calculated their theoretical Z score ranges based on the search band. Aptamer candidates within this range were then chosen and examined for their binding affinities, and the differences between their experimental Kd and the desired Kd were calculated. We defined a difference smaller than 25% as a ‘great fit’.

Fig. 4: Quantitative isolation of anti-MPO aptamers with desired affinities based on their Z score.
figure 4

a, Strategy for validation. A Kd search band was generated based on the KdZ score curve fit. The Z-score ranges of four desired values of Kd (1,000 pM, 3,000 pM, 9,000 pM and 27,000 pM) were calculated. Several aptamer candidates within the Z-score ranges could be tested and the difference between their actual Kd and the desired Kd then calculated to evaluate the accuracy of Z-score-based aptamer identification. b, Evaluation of the accuracy of using Z scores to identify MPO aptamers with Kd values of 1,000 pM, 3,000 pM, 9,000 pM and 27,000 pM. Red dots represent the aptamers with an actual Kd with <25% difference from the desired Kd. The data points for MPO-13 (Kd = 9,580 pM) and APO-27 (Kd = 27,700 pM) are outside the axis limits and are not shown. c, The binding curve of the best-fit aptamer candidates. Polystyrene particles coated with aptamers MPO-14, MPO-18, MPO-25 or MPO-34 were incubated with Cy-5-labelled MPO before analysis via flow cytometry. The red line represents the results of fitting a Langmuir binding isotherm to the MPO binding data. Error bars represent the standard deviation of three replicates, and the centres of the error bars represent mean values.

Source data

By evaluating the binding affinity of multiple MPO-aptamer candidates in different Z-score ranges (Supplementary Fig. 8 and Supplementary Table 4), we found that the success rate in identifying a ‘great fit’ was between 20% and 50%, suggesting that by testing no more than five aptamer candidates within the Z-score range, we can identify one aptamer with the desired binding affinity (Fig. 4b). The ‘best fits’ for the requested affinities are MPO-14 with a Kd of 897 pM, MPO-18 with a Kd of 3,192 pM, MPO-25 with a Kd of 8,778 pM and MPO-34 with a Kd of 24,910 pM (Fig. 4c; their secondary structures are provided in Supplementary Fig. 9). The Kd values of the tested aptamers spanned more than 100-fold. The selected aptamers also demonstrated excellent specificities for MPO (Supplementary Fig. 10) as well as the potential to be used to analyse biological samples (Supplementary Fig. 11). MPO-16 revealed the highest binding affinity (Kd = 166 pM); its sequence, secondary structure, binding specificity and binding performance in biological samples are highlighted in Supplementary Fig. 12. It should be noted that the normal MPO level from a healthy middle-aged population has been reported to be <640 pM, with higher MPO levels being associated with increased risk of cardiovascular diseases31. We believe that the anti-MPO aptamers with different binding affinities reported here will become valuable tools in future MPO monitoring applications.

Discussion

We developed an aptamer-generation method, Pro-SELEX, which enables efficient and quantitative generation of aptamers with desired binding affinities. Unlike conventional SELEX methods, which rely on iterative rounds of selection to screen for aptamer candidates (Fig. 5a), our approach monitors the binding performance of individual aptamer candidates at different target concentrations in a single round of selection (Fig. 5b), taking advantage of the high level of parallelization enabled by high-dimensional microfluidics. Current iterations of SELEX, such as CE-SELEX and particle display, despite their substantially improved efficiency in generating aptamers, still select for aptamers with the highest affinities (Fig. 5c). Unfortunately, high-affinity aptamers might not be suitable for all applications. For example, affinity receptors with suitable Kd should be selected based on the target concentration range to achieve quantitative biomolecular recognition, as the dynamic range of single-site binding typically spans 81-fold changes in target concentrations, which is dependent on the Kd of the receptors2. If the Kd of a receptor is substantially lower than the relevant concentration range, the resulting saturated signal will not change over the detection range2. A strategy to manipulate the dynamic range of sensing is to combine receptors with different target affinities1,32. This naturally raises the need to generate aptamers with different Kd. In addition, lower-affinity receptors with fast binding kinetics (high on rate and off rate) are ideal for continuous monitoring as they allow for rapid equilibration with the surrounding target4,33. The development in recent years of sophisticated detection modalities with considerably lower levels of background and digital outputs makes it possible to quantify molecules in log dynamic ranges at concentrations much lower than the Kd (refs. 34,35,36). By exploiting these types of sensing mechanism, low-affinity receptors can also achieve high detection sensitivity. Collectively, it will be advantageous to have the capability of generating receptors with customizable binding affinities to meet the requirements of diverse applications. The Pro-SELEX platform was developed to fulfil this need.

Fig. 5: Comparison of Pro-SELEX with current aptamer-selection technologies.
figure 5

a, Conventional SELEX strategy. b, Pro-SELEX strategy. c, A table summarizing the performance criteria of Pro-SELEX versus current aptamer-selection technologies.

The Pro-SELEX approach is also highly programmable and can be easily customized to suit various other applications. For example, the particle size used in this platform is large enough (diameter of 7 µm) to capture a wide range of targets, from small molecules to viruses or bacteria. Given the high efficiency of Pro-SELEX and the broad applications of aptamers during the SARS-CoV-2 pandemic37,38,39, our platform can be applied for the rapid generation of antivirus aptamers during the early stage of a pandemic. The time required for Pro-SELEX and SELEX to complete is summarized in Supplementary Table 5. In addition, recent advances in incorporating non-natural nucleotides with diverse structures and functional groups into aptamers would lead to aptamers with enhanced binding performance and expanded target range40,41,42,43. We believe our approach can be adapted to accommodate these unnatural nucleotides to efficiently produce modified aptamers.

Methods

Materials

Carboxyl polystyrene particles (7 µm, #PC06004) were purchased from Bangs Laboratories. Dulbecco’s phosphate-buffered saline (DPBS; #14190250), streptavidin (Cy5, #SA1011), Dynal MyOne C1 streptavidin Dynabeads (65001), biotinylated peroxidase (#432040), a PureLink Quick gel extraction kit (K210012), an EZ-Link Micro NHS-PEG4-biotinylation kit (#21955) and dNTP Mix (#R0192) were ordered from Thermo Fisher Scientific. Light mineral oil (#M5310), 2-butanol (#294810), N-(3-dimethylaminopropyl)-N′-ethylcarbodiimide hydrochloride (EDC, #161462), N-hydroxysuccinimide (NHS, #130672), human serum (H4522-100ML) and imidazole hydrochloride (#13386) were purchased from Sigma Aldrich. Anti-biotin microbeads ultrapure (#130-105-637) and labelling check reagents-APC (#130-122-219) were purchased from Miltenyi Biotech. Recombinant human MPO (3174-MP-250) was ordered from R&D Systems. Lambda exonuclease (#M0262S) was ordered from New England Biolabs (NEB). GoTaq G2 Hot Start Taq Polymerase (#M7405) was purchased from Promega. ABIL® EM 90 was obtained from Evonik. Recombinant human BNP protein (#ab87200) and recombinant human RANTES protein (#ab269212) were ordered from Abcam. Biotinylated thrombin (#69723) was obtained from Fisher Scientific. DT-20-M Gamma tubes were ordered from IKA Works. Water was purified with a Millipore Milli-Q Biocel water purification system.

All DNA oligonucleotides (Supplementary Table 1) were obtained from Integrated DNA Technologies and purified by 10% denaturing (8 M urea) polyacrylamide gel electrophoresis, and their concentrations were determined using a NanoDrop spectrophotometer. The library was 90-nt long, consisting of a 50-nt variable region flanked by two 20-nt primer sites (5′-AGCAGCACAGAGGTCAGATG-N50-AAGTGTAGTGTCTCCGTGGC-3′). The forward primer sequence was 5′-AGCAGCACAGAGGTCAGATG-3′ and the reverse primer was 5′-GCCACGGAGACACTACACTT-3′. The sequencing adaptors were 5′-ACACTCTTTCCCTACACGACGCTCTTCCGATCT-NNNN-AGCAGCACAGAGGTCAGATG-3′ for the forward-primer adaptor and 5′-GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTGCCACGGAGACACTACACTT-3′ for the reverse-primer adaptor.

Chip fabrication and operation

The microfluidic device was fabricated using a protocol combining three-dimensional (3D) printing and soft lithography techniques, as described in ref. 44. In brief, the master mould was manufactured by a stereolithographic 3D printer (Microfluidics Edition 3D Printer, Creative CADworks) using the ‘CCW master mould for PDMS’ resin (Resinworks 3D) with a layer thickness of 50 µm. The actual chips were fabricated by casting polydimethylsiloxane (Sylgard 184, Dow Chemical) on the printed moulds, followed by 120-min incubation at 70 °C for curing. Cured polydimethylsiloxane replicas were peeled off, punched and plasma-bonded to no. 1 glass thickness coverslips (260462, Ted Pella) to finish the chip. Before use, the chips were conditioned with a binding buffer (DPBS with 2.5 mM MgCl2, 1 mM CaCl2 and 0.01% TWEEN-20) to remove air bubbles. During operation, each chip was attached to arrayed N52 NdFeB magnets (D14-N52, K&J Magnetics) and fluidically connected to a digital syringe pump (Fusion 100, Chemyx) under the withdrawn mode.

Finite-element simulation

Numerical simulation of the flow velocity pattern within the microfluidic device was carried out using COMSOL Multiphysics (version 5.6, COMSOL) using the 3D creeping flow module. The key parameters for the simulation were as follows: boundary condition—pressure of 0 Pa with suppression of backflow; wall condition—no slip; mesh size—physics-controlled, fine.

Scanning electron microscopy

The fabricated chips were coated with 15-nm Au using a desktop sputter coater (Denton Desk II, Leica). The coated chips were imaged under a field-emission scanning electron microscope (SU5000, Hitachi) using 5-kV accelerating voltage and high-vacuum mode.

Preparation of biotin-labelled protein targets

MPO protein, as well as the proteins used for counter selection, including human serum proteins, BNP and RANTES, were biotinylated using an EZ-Link Micro NHS-PEG4-biotinylation kit (Thermo Fisher Scientific). The free biotin was removed by a Zeba Desalt spin column with 7-K molecular-weight cutoff (10 ml, Thermo Fisher Scientific). We measured the biotinylated MPO protein concentration based on absorbance at 280 nm using a NanoDrop spectrophotometer.

Pre-enrichment by SELEX

The DNA library (6 × 1014 random DNA molecules) was diluted with 1 ml of binding buffer and heated at 90 °C for 1 min, followed by annealing at room temperature (r.t.) for 10 min. Magnetic beads (Dynabeads MyOne streptavidin C1, Thermo Fisher Scientific) conjugated with human serum proteins, thrombin, BNP and RANTES were washed twice with binding buffer, and incubated with the DNA library at r.t. for 2 h. The DNA in the supernatant was collected, mixed with MPO (R&D Systems)-conjugated magnetic beads, and incubated at r.t. for 1 h. The beads were washed five times with 500 μl of binding buffer. The MPO-bound DNA molecules were then eluted by adding 100 μl of water and heating at 95 °C for 15 min. The eluted DNA was PCR-amplified and used for the following round of enrichment and AP synthesis. Three rounds of pre-enrichment were performed to remove both nonspecific DNA molecules that bind to other protein targets and inactive DNA molecules that do not bind to MPO protein.

Preparation of forward primer particles

A 1-ml volume of 7-μm carboxyl polystyrene particles (Bangs Laboratories) was washed once with 1 ml of 0.01 N NaOH and three times with 1 ml of nuclease-free water, then resuspended in a 500-µl reaction mixture containing 200 mM NaCl, 0.2 mM 5′-amino-modified FP (or amino-modified thrombin aptamers), 1 mM imidazole chloride, 50% v/v dimethyl sulfoxide and 250 mM EDC (Sigma Aldrich). The free carboxyls on the particles were then converted into amino-reactive NHS ester in the presence of 250 mM EDC and 100 mM NHS in 2-(N-morpholino)ethanesulfonic acid (MES) buffer (100 mM, pH 4.7; Thermo Scientific) for 30 min at r.t. The particles were then conjugated with 20 mM amino-PEG12 (Thermo Scientific) in MES buffer for 1 h. The particles were washed three times with 1 ml of TE (10 mM Tris, pH 8.0, 0.1 mM EDTA), suspended in 1 ml of TE buffer and stored at 4 °C.

Preparation of aptamer particles

Emulsion PCR

The oil phase (prepared freshly each time) was composed of 3% ABIL® EM90 and 0.05% Triton X-100 in mineral oil. The aqueous phase consisted of 1× GoTaq PCR buffer (Promega), 25 mM MgCl2, 3.5 mM of each dNTP (Thermo Fisher Scientific), 3 μM 5′ phosphorylated reverse primer (Phos-RP), 0.5 U μl−1 of GoTaq Hot Start Polymerase (Promega), 2 pM template DNA and 5 × 107 forward-primer-coated particles in a total volume of 1 ml. Water-in-oil emulsions were prepared by adding 1 ml of the aqueous phase to 7 ml of the oil phase in a DT-20 tube (IKA) locked into the Ultra-Turrax device (IKA). This addition was performed dropwise over 30 s while the mixture was stirred at 650 r.p.m. in the Ultra-Turrax. After adding the aqueous phase, we continued stirring the mixture for 5 min. The emulsions were distributed in 100-μl aliquots into ~80 wells of a 96-well PCR plate. We performed PCR under the following cycling conditions: 94 °C for 3 min, followed by 50 cycles of 94 °C for 15 s, 58 °C for 30 s and 72 °C for 60 s.

Emulsion PCR clean-up

After PCR, each PCR well was mixed with 50 μl of 2-butanol to break the emulsion. The broken emulsions were then transferred to a 50-ml tube. Next, 150 μl of 2-butanol were added to each PCR well to collect the leftover emulsions, then transferred to the 50-ml collection tube. After vortexing for 30 s, the particles were pelleted by centrifugation at 2,500g for 5 min. After carefully removing the oil phase, we resuspended the particles in 1 ml of emulsion breaking buffer (100 mM NaCl, 1% Triton X-100, 10 mM Tris-HCl, pH 7.5 and 1 mM EDTA) and transferred them to a new 1.5-ml tube. After vortexing for 30 s and centrifugation for 90 s at 15,000g, we removed the supernatant. Particles were then washed three times with TE buffer by centrifugation followed by resuspension in 500 μl of TE.

Single-strand generation

To generate single-stranded DNA, we first removed the supernatant by centrifugation. We then treated the particles with 100 U of Lambda exonuclease (NEB) in 500 μl of reaction buffer (67 mM glycine-KOH (pH 9.4), 2.5 mM MgCl2, 0.01% (vol/vol) Triton X-100) at 37 °C for 1 h, followed by inactivation of the reaction by heating at 80 °C for 15 min. The particles were then washed three times with 1 ml of TE buffer, suspended in 1 ml of TE buffer and stored at 4 °C.

AP quality control

Forward primer conjugation

To test the conjugation efficiency of the forward primers, we incubated 1 µM Cy5-labelled FP complementary sequence (Cy5-FP′) with 1 µl of forward-primer particles in 100 µl of binding buffer at 60 °C for 10 min, then cooled them at r.t. for 10 min. The particles were washed twice with 100 µl of binding buffer and analysed with a Cytoflex flow cytometer (Beckman Coulter Life Science). Acquired data were analysed using FlowJo software (version 10.8.1, FlowJo).

AP monoclonality

Based on the Poisson distribution, most particles would be monoclonal when <35% of the particles contain PCR products23. To confirm this, we annealed the APs with Cy5-labelled reverse primer in binding buffer at 60 °C for 10 min then cooled them at r.t. for 10 min. The particles were then washed twice with 100 µl binding buffer and analysed by flow cytometry.

Determining the aptamer copy number on each AP

Quantitative PCR was performed with a CFX384 Real-Time PCR detection system (Bio-Rad) to estimate the aptamer copy number for each AP. Calibration samples were prepared by adding 6 × 10, 6 × 102, 6 × 103, 6 × 104, 6 × 105, 6 × 106, 6 × 107, 6 × 108 or 6 × 109 templates into a 10-μl reaction containing 250 nM each of forward and reverse primer, ~8,000 forward-primer-coated particles, 5 μl of GoTaq PCR Master Mix (Promega) and 0.5× SYBR green (Life Technologies). Test samples were prepared identically, but with ~10,000 APs. From the threshold cycle, we quantified ~6 × 108 sequences on 10,000 APs. Because only ~20% of APs displayed template sequences, the average copy number of sequences on each template-bearing AP was ~3 × 105.

Pro-SELEX screening

During Pro-SELEX screening, we incubated 5 × 107 aptamer particles with biotin-labelled MPO protein at different concentrations (see Supplementary Table 2 for detailed experimental conditions) in selection buffer (DPBS with 2.5 mM MgCl2, 1 mM CaCl2, 0.01% TWEEN-20, 0.2% BSA). To ensure the MPO molecules are not depleted during selection, the reaction volumes were increased when using lower concentrations of the target. After 1 h of incubation, the beads were washed three times with selection buffer. We then magnetically labelled the aptamer particle-captured MPO proteins by incubating the particles with 20 μl of anti-biotin microbeads (Miltenyi Biotech) in 1 ml of selection buffer for 15 min at r.t. The particles were then washed with binding buffer twice and sorted by Pro-SELEX microfluidic devices. The APs captured at each capture zone of the Pro-SELEX microfluidic devices were collected and stored at 4 °C.

High-throughput sequencing

For each collected aptamer pool and the unsorted DNA pool, two rounds of PCR were performed to add adaptor sequences and sequencing indexes. The amplified sequences were then purified using agarose gel electrophoresis and a PureLink Quick gel extraction kit, quantified using a NanoDrop spectrophotometer, and sent to McMaster Genomics Facility for sequencing on an Illumina MiSeq.

Bioinformatics

Raw sequencing data were processed using the Illumina Basespace online platform to sort tagged sequence pools and output sequence data in FASTQ format. Paired-end reads were merged using USEARCH (v10)45 and the primers were then trimmed using Cutadapt (V3.5)46. Sequences containing less than perfect complementarity were discarded to minimize sequencing errors in the dataset. FASTA format trimmed sequences were dereplicated and tagged with the copy number using USEARCH. Read counts were then generated in the format of tab-delimited text files. The files were processed by the AptaZ script written in MATLAB (version 2021a, MathWorks) to calculate the corresponding Z score for each sequence. Calculated Z scores were exported as CSV files for visualization and validation. The source code of AptaZ can be accessed, free of charge, at https://github.com/dwangnu/AptaZ.

Affinity measurement

After sequencing analysis, we performed particle PCR to synthesize APs displaying each unique aptamer candidate sequence. Each particle PCR reaction (100 µl) consisted of 1× GoTaq PCR Master Mix (Promega), 25 mM MgCl2, 2 µM phos-RP, 10 nM aptamer template and 5 × 106 FP-coated particles. PCR was carried out under the following cycling conditions: 94 °C for 3 min, followed by 30 cycles of 94 °C for 15 s, 58 °C for 30 s and 72 °C for 30 s. To avoid particle aggregation and increase particle PCR efficiency, we vortexed the reaction every four cycles after the 94 °C denaturing step. We then removed the reverse strands using Lambda exonuclease as described above. We then incubated 1 µl of the APs with Cy5-RP and measured the AP fluorescence intensity using flow cytometry to test for particle PCR efficiency. For affinity measurement, different concentrations of biotinylated MPO (10 pM to 200 nM) were incubated with a fixed amount of the APs (2 × 104 particles) in selection buffer for 1 h at room temperature. The unbound MPO was washed away with selection buffer, after which streptavidin-labelled Cy5 was introduced and incubated for 15 min to label the bound MPO. The particles were washed with binding buffer, and median fluorescence intensities were quantified via flow cytometry and FlowJo software (version 10.8.1, FlowJo).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.