Digital and Absolute Assays for Low Abundance Molecular Biomarkers

CONSPECTUS:

There has been a recent surge of advances in biomolecular assays based on the measurement of discrete molecular targets as opposed to signals averaged across molecular ensembles. Many of these “digital” assay designs derive from now-mature technologies involving single-molecule imaging and microfluidics and provide an assortment of new modalities to quantify nucleic acids and proteins in biospecimens such as blood and tissue homogenates. A primary new benefit is the robust detection of trace analytes at attomolar to femtomolar concentrations for which many ensemble assays cannot distinguish signals above noise levels. In addition, multiple biomolecules can be differentiated within a mixture using optical barcodes, with much faster and simpler readouts compared with sequencing methods. In ideal digital assays, signals should, in theory, further represent absolute molecular counts, rather than relative levels, eliminating the need for calibration standards that are the mainstay of typical assays. Several digital assay platforms have now been commercialized but challenges hinder the adoption and diversification of these new formats, as there are broad needs to balance sensitivity and dynamic range of detection, increase analyte multiplexing, improve sample throughput, and reduce cost. Our lab and others have developed technologies to address these challenges by redesigning molecular probes and labels, improving molecular transport within detection focal volumes, and applying solution-based readout methods in flow.

This Account describes the principles, formats, and design constraints of digital biomolecular assays that apply optical labels toward the goal of simple and routine target counting that may ultimately approach absolute readout standards. The primary challenges can be understood from fundamental concepts in thermodynamics and kinetics of association reactions, mass transport, and discrete statistics. Major advances include (1) new inorganic nanocrystal probes for more robust counting compared with dyes, (2) diverse molecular amplification tools that endow attachment of numerous labels to single targets, (3) specialized surfaces with patterned features for electromagnetic coupling to labels for signal amplification, (4) surface capture enhancement methods to concentrate targets through disruption of diffusion depletion zones, and (5) flow counting in which analytes are rapidly counted in solution without pull-down to a surface. Further progress and integration of these tools for biomolecular counting could improve the precision of laboratory measurements in life sciences research and benefit clinical diagnostic assays for low abundance biomarkers in limiting biospecimen volumes that are out of reach of traditional ensemble-level bioassays.

Graphical Abstract

INTRODUCTION

Quantification of molecules in biological fluids is a core part of clinical diagnostic testing and analytical studies in the life sciences. In the most common assay designs, target biomolecules like proteins or nucleic acids react with dye-based sensors in solution or are captured on a surface for labeling with dye-based probes (Figure 1a). An analog optical intensity readout is collected statically or over time, and the concentration of the target is calculated by comparison of measurements with those of calibration standards. Many such assay formats have remained largely unchanged for decades, including enzyme-linked immunoassays,⁵ nucleic acid and protein microarrays,^6,7 and quantitative polymerase chain reaction (qPCR).⁸ Despite their proven value and robust design, applications are often hindered by lower limits of detection (LOD) in the picomolar concentration range (~10⁷ copies μL⁻¹), the necessity for a fairly large sample volume (~100 μL) for one technical replicate, and/or a limited number of different targets that can be simultaneously determined (multiplexing).^9-11 In recent decades, a variety of low-abundance biomolecules present at attomolar to femtomolar concentrations (1–10³ copies μL⁻¹) have been identified and have drawn broad interest, especially blood-based biomarkers associated with a variety of clinical conditions and diseases. A clinical goal is to now detect and quantify these circulating tumor DNA fragments, tumor-associated microRNA, and trace proteins such as cytokines, mutated p53, phosphorylated tau, and neurofilament light chain which may predict disease onset, patient outcome, and drug response.^12,13 Furthermore, there is an ongoing drive to advance the practical use of smaller biospecimen volumes collected using minimally invasive methods for point-of-care testing using portable instruments with rapid readout.^14,15 These collective initiatives are driven in large part by the goals of precision medical care to harness trace analytes using accessible instruments to stratify patients for treatments based on their individualized and dynamically changing molecular expression profiles.¹²

Figure 1.

Examples of analog molecular assays and commercialized digital assay platforms. (a) Surface-capture analog assay showing protein target capture and labeling with antibodies in wells with typical sizes near 2.5 mm. Calibration standards are used to convert ensemble well intensities to target concentrations. (b) Nanostring nCounter. Surface-captured target nucleic acids are labeled with a reporter probe which further binds to numerous fluorescent probes functioning as a digital fluorescent barcode that is linearized by an electric field. High-resolution multicolor imaging allows counting and identification of multiple distinct targets¹⁶. (c) Quanterix single-molecule array (Simoa). Antigens are captured by antibodies on a microbead, labeled using enzyme-conjugated antibodies, and then isolated in arrayed femtoliter microwells. Colorimetric readouts of the enzymes allow digital counting of single protein targets. Distinct colors of fluorescent beads bound to different antibodies can be used to identify different proteins.¹⁰ (d) Droplet digital PCR. Individual nucleic acid targets and PCR reagents are compartmentalized in single aqueous nanoliter droplets. PCR-based amplification results in either positive or negative fluorescence intensity in each droplet. Readout of individual droplets is performed in a flow stream or through imaging.¹⁷

Breakthroughs in single-molecule detection and optical microscopy in the 1990s set the stage to push the lower LOD of diverse biomolecular assays to single-molecule levels.^9,18 When the number of target analytes in analog assays is small, LODs are defined by faint background signals emanating from samples and instrument components that mask ensemble signals of small numbers of dyes that serve as probes and sensors. This was surpassed by microscopically imaging individual molecules with spatial resolution near the diffraction limit of light (<1 μm²), a region much smaller than that of a typical analog assay chamber (~5 mm²) or microarray spot (~500 μm²). Life sciences communities applied these new single-particle imaging methods to study DNA and protein interactions, to count nucleic acid targets in cells, and to study membrane receptor and molecular motor transport,^19,20 primarily using fluorescence-based methods that benefitted from advances in high-gain detectors, high-numerical aperture objectives integrated into confocal and wide-field microscopes, and new fluorophores with improved brightness and photo-stability. Simultaneously, advances in microfabrication led to the accessible production of microfluidic systems and arrayed solid substrates, allowing use of smaller sample volumes.^10,15,17

Integration of these tools led to “second generation” molecular assays that digitally read out single molecules in a handful of commercialized analytical instruments. These digital platforms can be categorized broadly as either surface-capture assays or solution-partitioning assays. A notable surface-capture assay platform is the Nanostring nCounter (Figure 1b) which quantifies nucleic acids adsorbed to a glass surface by single-molecule imaging and counting of targets labeled with fluorescent dyes, achieving a lower LOD of 100–500 aM.¹⁶ Other examples include second- and third-generation DNA sequencing technologies from Illumina, Inc., and Pacific Biosciences which use digitized molecular readouts and multicolor detection in microfabricated reaction chambers on chips, but the more complex dynamic readouts are beyond the cost and speed requirements for routine measurements of biomarker abundance.²¹ Digital readouts can also be achieved by fractionating biospecimen volumes to capture individual molecules within isolated microwells or droplets for digitized enzymatic amplification. The Quanterix Simoa single-molecule array performs enzyme-linked immunosorbent assays on bead surfaces (Figure 1c) to achieve 50 aM LOD for proteins.^10,22,23 Droplet digital PCR (ddPCR, Figure 1d) similarly detects single nucleic acids isolated in aqueous droplets using PCR-driven exponential amplification and fluorescent sensors, with resulting counts considered to be measures of “absolute” abundance in biospecimen samples.^17,24 Each of these new digital assay classes has been compared with analog assays^10,16,17 and is being evaluated in clinical trials for measuring low-abundance blood biomarkers (e.g., ClinicalTrials.gov Identifiers NCT02279004, NCT05542511, NCT03727334). One ddPCR test (Bio-Rad QXDx BCR-ABL%IS Kit) is now FDA-approved to monitor minimal residual disease in chronic myeloid leukemia patients based on BCR::ABL1 oncogene transcripts in RNA extracts from blood.²⁵

ADVANTAGES AND CHALLENGES OF DIGITAL AND ABSOLUTE ASSAYS

Digital assays involve the measurement and quantification of discrete molecules as counts of recorded events as opposed to continuous intensity values. These assays have intrinsic advantages over their analog counterparts, including improved lower LOD, measurement robustness at low analyte concentrations,²⁶ and the potential for higher degrees of molecular multiplexing within a limited sample volume.^27,28 Absolute assays are digital assays in which counts are equal to the number of countable targets in the sample volume. Such assays provide the potential benefit of requiring no calibration of counts to calculate target concentrations. The limitations by which digital assays fail to specifically and sensitively count all targets in the sample volume, and therefore reach an absolute readout, are a primary focus of the discussion below. However, it is important to note that for any assay, measured counts may underestimate targets in the original biospecimen due to preanalytical processing steps that cause losses from extraction, target degradation, and surface adsorption. Even absolute assays require internal references or controls to account for the deviation between the test sample and the original biospecimen.

Figure 2a shows an example of measurements from a typical digital assay in comparison with its analog counterpart in a surface-capture format with readout by fluorescence-based imaging. Background fluorescent signals set the lower LOD for the analog assay in the picomolar range. This global background can be effectively subtracted by collecting high-resolution images of the faint discrete molecules captured on the surface so that their distinct features can be identified as diffraction-limited point spread functions (PSFs) commonly approximated as 2-dimensional Gaussian functions. Using counts of spots in the field of view instead of intensities, LODs can be extended to the femtomolar level and below.¹ However, both assay types are limited in their dynamic range of detection. At the upper limit of a digital assay, PSFs spatially overlap and cannot be discretely identified. This is problematic because many molecules span 10⁶-fold variations in abundance, requiring higher concentration samples to be diluted and tested again to fit within the detection window. The upper limit can be expanded by calibrating the analog intensity to the individual PSF intensity (Figure 2b), allowing nearly 5 orders of magnitude of dynamic range, down to the femtomolar level, in a single sample. It is still a challenge to balance the lower LOD and dynamic range in practical surface-capture setups, as extending the detection range to low attomolar levels is usually limited by the presence of off-target positive counts and the large area of capture surface that must be interrogated. This requires the highest fidelity in identifying individual labeled targets relative to interfering background signals and off-target labeling, as well as important statistical considerations,²⁹ described below.

Figure 2.

Comparison between digital and analog fluorescence-based surface-capture assays and their integration.¹ (a) Digital counts and analog intensity measurements are shown for microRNA-375 targets labeled with dye-based probes after rolling circle amplification (RCA). Analog intensities are calibrated to digital counts using the method shown in panel b. (b) Fluorescently labeled targets on a surface are imaged at high resolution. Low-intensity images are analyzed digitally using a spot-detection algorithm that identifies PSFs to measure the average spot intensity (I_S). For high-intensity images, pixel intensities are calibrated to spot counts using I_S, the number of pixels (N_p), and the average pixel intensity with target (I₊) and without target (I_{_}). Adapted with permission from ref 1. Copyright 2018 American Chemical Society.

A second advantage of digital readouts is that many different molecules can be labeled with distinguishable optical tags and discretely counted in the same sample, which is generally constrained to just a few targets in ensemble analog readouts. Surface-capture platforms are particularly suitable for this multiplexing technique, as multicolor single-molecule fluorescence microscopy is well-developed and widely accessible, and diverse classes of optical molecular barcodes have been developed.^30,31 As an example, the Nanostring nCounter can distinguish and quantify >500 genes in the same sample using bar-coded fluorescent tags.¹⁶ Small degrees of multiplexing have also been achieved in digital assays based on solution partition including ddPCR and Simoa. ^27,32 These readouts use spectrally distinct sensors or fluorescent beads for each target, which is intrinsically limited compared with more information-dense codes of molecular probes that include both ratiometric intensities and spatial ordering.

Despite limitations in multiplexing relative to surface-capture assays, assays that fractionate samples into compartments or droplets have significant advantages. In particular, measurements can be collected from the entire sample volume, whereas surface-capture assays often cannot comprehensively or reliably capture all targets.^17,24,26,29 Enzymatic amplification within spatially organized compartments can also enhance single-target signals by a magnitude and specificity that is difficult to match in a homogeneous mixture. Count measures are therefore more precise for attomolar samples and are more readily converted into an absolute readout of target copies with linearity with respect to target concentration. This enhanced precision is the primary value of ddPCR in practice compared with analog qPCR, which can already detect targets reliably at femtomolar levels across a wide dynamic range.^17,33 However, microwell and droplet components require complex instrumentation and processing steps that increase cost, limit throughput, and hinder device miniaturization.³⁴ Further, ddPCR assays require dilute targets to ensure discrete isolation in single droplets for amplification and digital measurements, which necessitates screening large numbers of droplets to achieve accurate readings. In contrast, surface-capture assays in which labeled targets are randomly adsorbed on a surface for imaging and counting through high-magnification microscopy can be achieved using devices with small form-factors and with minimal manual processing steps.^4,9,35 As a result, surface-based digital assay classes are flourishing and rapidly advancing in research laboratories and are a primary focus of the following sections.

OVERCOMING CHALLENGES TOWARD ABSOLUTE MOLECULAR ASSAYS

Challenges of developing robust digital assays can be greater than those of analog assays due to the higher complexity of digital readouts and data interpretation as well as a greater number of technological components. Often there is a trade-off between sensitivity, specificity, and dynamic range that arises from the limiting kinetics of pull-down to a surface and the finite differences in affinity between labeling reagents and the target molecules relative to off-target molecules that are present in much higher numbers. Here we describe how our group and others are addressing key challenges toward the development of absolute biomolecular counting assays for routine and fast quantification of low-abundance biomarkers. We focus primarily on assays for which known proteins or nucleic acids are labeled with molecular probes such as dye conjugates of antibodies or single-stranded DNA (ssDNA), respectively.

Labels

A probe is a conjugate between a label for optical measurement (e.g., a fluorescent dye) and a molecular tag that binds specifically to a target (e.g., antibody or ssDNA). Fluorescent labels have garnered the most use in digital assays due to compatibility with advanced single-molecule imaging methods and include organic dyes, semiconductor nanocrystal quantum dots (QDs), fluorescent proteins, and polymeric fluorescent beads.^1,36 Other commonly applied label classes include fluorescence-based photon-upconversion nanoparticles, light-scattering colloidal metals, and surface-enhanced Raman scattering nanoparticles.³⁶ Smith et al. compared the performance of 4 fluorescent probe classes for single-molecule counting assays, including organic molecular dyes, QDs, a bright fluorescent protein (phycoerythrin, PE), and fluorescent beads (FluoSphere, bead)¹ which, in order, exhibit increasing intensity (Figure 3a) but also total size. As shown in Figure 3b,c, the signal-to-noise ratio of PSF detection for digital counting increased with label size, progressively improving the fidelity of single-target detection. Background signals from cleaned glass coverslips set the PSF detection threshold such that absolute counts depended on an arbitrarily chosen threshold for discriminating positive signals from background. The brightness and uniformity of signals from PE and beads resulted in consistently zero background counts at low thresholds, whereas smaller QDs and dyes exhibited lower uniformity of single-molecule intensity. This is an important fundamental trade-off in labels because dyes and QDs provide higher degrees of spectral tunability needed for multiplexing, as well as reduced steric hindrance needed for absolute readouts of labeled targets. Readouts of dyes and QDs can be improved by shifting the emission band to spectral ranges where the background signals are lower.³⁷ Le et al. reported that QDs provide major advantages as digital labels compared with dyes in noisy background samples (Figure 4).² In the case of autofluorescent cells, single-molecule imaging for growth factor detection was more accurate for QD labels compared with dyes, especially when shifting the QD emission band to longer wavelengths where background counts from the sample are reduced. However, current QD variants exhibit heterogeneous intensities at the single-molecule level due to intermittent “blinking”, which can make comprehensive target counting difficult, although newer variants have substantially enhanced homogeneity of emission³⁸ and plasmonic surfaces can suppress blinking.⁴ Additional approaches to reduce background signals include time-resolved measurements of labels with long excited state lifetimes (e.g., QDs) or using fluorescence upconversion labels with anti-Stokes emission.³⁶

Figure 3.

Photophysical properties of avidin-conjugated fluorophores for single-molecule surface-capture counting assays.¹ (a) Extinction coefficient (ε) spectra and fluorescence emission spectra (FL) are shown for a dye (red), QD (orange), PE protein (green), and bead (blue). (b) Widefield single-molecule fluorescence images of labels sparsely bound to coverglass. (c) Spot detection counts are shown for different PSF detection thresholds for bare coverglass (gray) and coverglass with several hundred fluorophores per field of view (black). The red line indicates the difference between the two. Error bars indicate standard deviation. Adapted with permission from ref 1. Copyright 2018 American Chemical Society.

Figure 4.

Controlling emission wavelength to enhance signal-to-noise ratio for digital target counting.² (a) Images of cells labeled with epidermal growth factor (EGF) bound to either a dye or QD with 3 emission wavelengths (565, 605, or 744 nm). Brightfield (B.F.) micrographs are overlaid with nuclear stain (blue). Blackboxes indicate zoomed-in areas in fluorescence images. Yellow arrows indicate background fluorescence; red arrows indicate single dye or QD. (b) Fluorescence spectra of mean background, dye, and QDs. (c) Intensities of background, single dyes, and single QDs in each spectral band. Gray corresponds to background and color corresponds to specific signals of labeled foci. (d) Receiver operating characteristic (ROC) curves show higher detection accuracy of single QD744 (dark red) compared with dye (blue), QD565 (green), and QD605 (orange) amid background signals from cells. Numbers show area under the ROC curve. Adapted with permission from ref 2. Published 2019 by Springer Nature under a Creative Commons Attribution 4.0 International License.

Weak signals from single-dye and single-QD readouts can be overcome using amplification processes that result in attachment of multiple labels to each target. For specific target classes such as mRNA and DNA, fluorescence in situ hybridization (FISH) methods can append ~50 ssDNA–dye labels to a single target gene localized in a diffraction-limited spot, resulting in amplification of the signal well above background levels, with readouts considered to be absolute in fixed cells.³⁹ As shown in Figure 5a–c, Liu et al. showed that this labeling scheme enhances the difference between signal and noise to a greater extent when using ssDNA–QD labels compared with ssDNA–dye labels.⁴⁰ However, counts were diminished when using larger size QD labels, and specially designed compact QDs were necessary to match absolute counts of standards. This labeling strategy is only possible for target classes bearing multiple tandem regions for attachment of multiple probes. Therefore, diverse ways have been devised to append numerous ssDNA repeats to short nucleic acids and proteins. Figure 5d–f shows an example of rolling circle amplification (RCA) for enzymatic extension of short ~22-mer microRNA sequences to generate long sequences with numerous probe hybridization sites, greatly enhancing digital signal-to-noise ratio.¹ A wide variety of such amplification methods have recently been advanced and reviewed.⁴¹

Figure 5.

Examples of amplification methods to enhance digital molecular readouts. (a) Fluorescence in situ hybridization using ssDNA–dye (top row) or ssDNA–QD (bottom row) probes for glyceraldehyde-3-phosphate dehydrogenase (GAPDH) mRNA in fixed HeLa cells.⁴⁰ Fluorescence micrographs show mRNA foci (red) and nuclei (blue). (b) Intensity histograms are shown for FISH spots (black) and single fluorophores (gray). (c) Counts per cell are shown with different PSF detection thresholds for FISH (blue) or cells without labels (red), with flatter curves for QDs indicating higher count fidelity. Shading indicates standard deviation. (d) Fluorescence images and spot intensities of single fluorescent molecules (PE) in comparison with (e) RCA-amplified nucleic acids labeled with PE.¹ The schematic depicts numerous PE labels per target molecule. (f) Intensity histograms are shown for targets labeled with PE (cyan) or RCA-PE (red) compared with background (gray). Adapted with permission from refs 1 and 40. Copyright 2018 American Chemical Society and published 2018 by Springer Nature under a Creative Commons Attribution 4.0 International License.

Bioaffinity Tags

For analog assays measuring abundant targets, specific capture of targets among an excess of off-target molecules is effective when using antibodies or complementary oligonucleotides as bioaffinity tags. However, for trace analytes in plasma extracts, targets can be outnumbered by a billion-fold or more, so example of these effects for DNA capture using complementary ssDNA tags, with the goal of detecting a single nucleotide polymorphism. The affinity of the capture tag for the target can be designed to span equilibrium dissociation constants (K_D) from less than 10⁻¹⁵ M to 10⁻⁹ M by adjusting the sequence length between 7 and 15 nucleotides (nt). As shown in Figure 6b, targets at 1 fM will be fully bound to the capture tag when the tag is also present at femtomolar concentrations, but only for tag lengths of 13 nt and longer. Higher tag concentrations are needed to capture all targets when using shorter tags due to lower affinity. As shown in Figure 6c, off-target binding of the wild-type sequence is lower than that for the target, but the number of molecules bound depends on tag concentration because the difference in K_D is small (~100-fold). Therefore, the capture tag concentration must be precisely tuned to minimize capture of off-target molecules which may be present in large excess relative to the target. Lower tag concentrations are needed for high target discrimination (Figure 6d), so there is a fundamental trade-off between target capture (sensitivity) and off-target discrimination (specificity). As shown in Figure 6e, off-target counts elicit both a shift in total measured counts and a reduction in dynamic range by a magnitude that depends on the abundance of wild-type sequences. For many trace RNA and DNA classes, these trade-offs are more easily balanced than for single nucleotide polymorphisms. For example, for targets with longer distinguishable sequences, tag affinity differences between targets and off-targets can be designed to be much larger, allowing high target selectivity even at high tag concentrations. Such longer targets can be detected near 1 aM in competition with a billion-fold excess of off-target molecules,¹⁷ a scenario common in undiluted plasma extracts of total RNA (~2.5 ng mL⁻¹) and DNA (~30 ng mL⁻¹).

Figure 6.

Impact of off-target molecules and tag affinity, showing equilibrium calculations. (a) Schematic depiction of a probe bearing a 13 nucleotide (nt) tag that binds to a target sequence or a single-base mismatch. (b) Counts per microliter for the indicated probe concentrations with target present at 1000 copies μL⁻¹, calculated for tag lengths between 7 and 15 nt. (c) Counts per microliter for target or single-base mismatch at 1000 copies μL⁻¹ with 13 nt tag. (d) Target-to-mismatch count ratio with both target and single-base mismatch present at 1000 copies μL⁻¹. (e) Counts at the indicated target concentrations with different off-target sequences present at the indicated ratios from 1:0 to 1:50 with 10,000 probes μL⁻¹. Values in panels b–e were calculated from the simple thermodynamic models depicted in panel a.

Creative techniques have been developed to bypass limits of static binding assays for higher specificity nucleic acid detection in digital platforms. The Walter group developed a single-molecule kinetic fingerprinting approach to image the dynamic association and dissociation of individual dye-labeled ssDNA probes to target nucleic acids captured and isolated on a surface.⁴² Targets with similar equilibrium binding constants are more distinguishable by their rates of association and dissociation, which provided a signature of specificity orthogonal to counts. This allowed rejection of off-target events to boost the count specificity, resulting in the detection of a single base mutation in DNA amid a million-fold excess of wild-type DNA.⁴³ A second approach is to use locked nucleic acids which exhibit greater sequence specificity for shorter binding regions. This approach was used by Cohen et al. in digital readouts of short-strand miRNA in a Simoa assay.⁴⁴

The challenge is greater for low-abundance protein targets. Undiluted plasma contains ~1 mM total protein (70 mg mL⁻¹), for which a 1 aM target is outnumbered by a quadrillion-fold (1:10¹⁵). Antibodies typically have K_D values across a broad range of 0.01–10 nM, so high tag concentrations (nanomolar) are needed for complete target capture.²⁶ Antibody off-target binding occurs with K_D that is near 1 mM at best,⁴⁵ so specific capture of attomolar targets is effectively impossible in plasma using a single antibody-based tag. This can be partially addressed using a secondary antibody in a sandwich immunoassay for which two independent antigen binding events compound to reduce off-target event counts. Specificity is further enhanced by isolation steps that wash away competing off-target molecules, as in the case of Simoa, but at the cost of greater workflow complexity and desorptive losses of targets to degrees that depend on the specific antibody affinity.²⁶ This is further complicated for important targets like cytokines and growth factors that equilibrate between monomeric and oligomerized forms, with antibody selectivity typically favoring oligomers.⁴⁶ For such targets, counts can be expected to nonlinearly depend on concentration. Analytical challenges are greater for proteins than for nucleic acids in part due to the sheer diversity of the proteome compared with the genome, with protein isoforms, posttranslational modifications, dynamically changing conformations, and natively interacting molecules like heterophilic antibodies all potentially impacting capture efficiency by bioaffinity tags. Moreover, proteins have a propensity to denature, adsorb to surfaces, and aggregate during processing so readouts can be less reliable and particularly prone to matrix interferences.

An alternative approach is to use in situ reactions with additive selectivity that eliminate washing steps. The groups of Nilsson and Landegren advanced the application of enzymatic ligations between the ends of capture tags such as padlock probes and proximity probes catalyzed by target binding in digital molecular assays.^47,48 In the case of padlock probes, this step enhances the specificity of a second RCA step facilitated by the ligated tag–target hybrid, resulting in attomolar digital detection of targets.⁴⁷ These in situ methods may be more adaptable to streamlined workflows with less complex liquid handling than those involving purification and washing. Newer methods for enhancing selectivity include the use of Cas13 and other enzymes with high sequence discrimination,⁴⁹ although these have yet to be adopted for protein quantification in digital readouts.

Target Capture from Attomolar Solutions

There are distinct challenges in the capture and detection of attomolar targets compared to more abundant targets. As shown in Figure 7a, for samples smaller than ~100 μL, the exact volume dictates measurement precision because the average solution volume occupied per target is on the same scale as the total volume. For 10 μL samples of 1 aM biospecimens, copies will randomly span 3 to 13 (the 90% prediction interval); this range is 0 to 3 discrete copies for 1 μL samples. This sampling uncertainty compounds with measurement uncertainties that derive from imperfections in sensitivity and specificity and the potentially incomplete capture of targets. For surface-capture assays, a primary goal is therefore to maximize the number of targets collected on a substrate for detection. Fundamental limits in pull-down kinetics due to diffusion from solution to a surface are shown in Figure 7b, for which complete capture is faster when the solution is compressed to a smaller thickness above the capture surface. For sub-millimeter thicknesses, complete capture can be achieved in minutes, but the solution must be spread across a large area. As a result, large areas must be visualized and analyzed in order to achieve high certainty of the number of targets present in the sample. The Nanostring nCounter applies 1 μL samples in 30 μm height channels for rapid capture of labeled nucleic acids. An area of 10 mm² (600 fields of view) is then imaged at high resolution over 4 h, with an estimated ~1% of targets counted, resulting in a 100–500 aM detection limit.¹⁶ Low attomolar detection is unrealistic with these parameters. It is notable that target capture can be much slower and less complete than in the calculated example in Figure 7b showing the smallest, fastest-moving targets (22-mer ssDNA) and assuming instantaneous and irreversible capture after collision. Larger nanoparticle labels or products of RCA will diffuse much more slowly, while association rates can be limiting, especially for nucleic acids.⁵⁰

Figure 7.

Sampling and surface-capture considerations for attomolar digital assays. (a) Dependence of target counts on sample volume from 0.1 nL to 100 μL, assuming samples collected randomly from biospecimens with concentrations from 1 aM to 100 fM. Shaded regions are 90% prediction intervals of the Poisson distribution. (b) Surface capture dependence on time and solution thickness. Random walks of 60 individual molecules were simulated in a 10 μL (1 aM) volume with thickness from 0.5 mm to 10 mm above a flat capture surface. Diffusion coefficient corresponds to that of a 22-mer RNA (138 μm² s⁻¹).⁵¹ Counts correspond in time for target collisions with the surface, assuming instantaneous, irreversible binding. (c) Enhancement of mass transport of trace targets to surfaces using an aqueous two-phase system of saline (salt phase) and poly(ethylene glycol) (PEG phase) that condenses the targets to the surface.⁵² (d) Nanoporous substrates eliminate nonslip layers on surfaces for more efficient surface-capture. Fluid near the surface can flow through surface pores (red lines) to increase the rate of capture. (e) Three-dimensional single molecule deconvolution microscopy image analysis to digitally count molecules in 3D substrates such as single cells across a z-height of 20 μm or more.² (f) Photonic crystal capture surface propagates evanescent waves from line-scanned excitation beams to enhance the emission of proximal QDs for digital counting of surface-captured targets in an excess of unbound QD probes in solution.⁴ Adapted with permission from refs 2, 4, and 52. Published 2019 and 2022 by Springer Nature under a Creative Commons Attribution 4.0 International License and Copyright 2021 American Chemical Society.

The trade-off between measurement precision and time for processing and analysis has been addressed using a number of innovative techniques. A well-developed approach is to use dispersible microbeads to capture targets in solution for concentration into smaller volumes by sedimentation,^23,26 although this can add considerable complexity to processing workflows. Li et al. showed that an immiscible fluid can be used to concentrate attomolar nucleic acid solutions above a capture surface for digital counting, an approach that may be more amenable to simple fluidic systems (Figure 7c).⁵² Gao and co-workers also observed that probe capture could be enhanced by cycling solutions above a capture surface with a gas phase through draining, which replenishes the “depletion zone” of targets in the solution proximal to the capture surface.⁵³ Xiong et al. showed that microscopic stir bars that enhance convective transport can substantially increase capture efficiency in surface capture assays.⁵⁴ Zhang et al. showed that microfluidics together with 3-dimensional (3D) porous capture regions disrupt immobile and depleted surface layers for enhanced capture (Figure 7d).⁵⁵ Imaging and analysis of such 3D capture structures is more challenging for digital readouts but has been made feasible through methods developed for imaging cells through 3D PSF interrogation (Figure 7e)² or simply by projecting a 3D image onto a single 2D image.³⁹

A final challenge associated with surface-capture assays is the finite dissociation rate between the target and capture tag. For a steady-state counting assay, target or label desorption will reduce absolute counts, increase the lower LOD, and cause sublinearity with analyte concentration. It is therefore beneficial to refrain from washing away unbound labels and targets in order to reach an equilibrium of captured targets and labels. To do so amid background signals from unbound labels requires the use of reduced focal volumes near the capture substrate. Localized excitation can be achieved within a few hundred nanometers of the surface using specialized optical setups (e.g., total internal reflectance fluorescence, TIRF). An alternative approach is to apply capture substrates that selectively propagate evanescent electromagnetic waves, such as photonic crystals (Figure 7f) which were shown by Xiong et al.⁴ to enhance fluorescence excitation and emission of proximal fluorophores and can be produced at scale to avoid the high cost and large size of TIRF setups.

In-Solution Readouts

Many aforementioned challenges with surface-capture assays can be avoided by counting labeled targets in solution. Single analytes can be digitally detected in solution using fluorescence confocal microscopy to calculate concentrations using time-series correlation methods.¹⁸ However, diffusion into and out of a focal volume is too slow to practically interrogate 10 μL volumes needed for attomolar targets. This can be partially overcome by moving the focal volume throughout the solution, a method incorporated into the MilliporeSigma SMCxPRO digital assay.⁵⁶ Smith et al. reported a single-molecule flow assay in which target nucleic acids are amplified by RCA, labeled with ssDNA–dye probes, and counted as they are rapidly passing through a focal volume in a flow stream (Figure 8a), allowing the collection of hundreds to thousands of counts in seconds from femtomolar samples.³ This was performed in a commercial benchtop flow cytometer conventionally used for cell analysis. Because RCA products are long tandem repeats encoded by a designer template sequence, they can be labeled with a large number of probes which yield events with intensity much greater than that of excess unbound labels remaining in solution, requiring no purification. High detection fidelity was verified by correlation across multiple optical channels of scattering and fluorescence (Figure 8b,c), with 4 orders of magnitude of dynamic range (Figure 8d). The sequence repeats also serve as barcodes to evaluate multiple targets using orthogonal excitation and emission channels (Figure 8e,f).

Figure 8.

Digital counting of nucleic acids by single-molecule flow analysis.³ (a) Schematic illustration depicting labeling of RCA products with fluorescent probes and counting using a flow cytometer. (b, c) Scatter plots show detected events for microRNA RCA amplicons labeled with both ssDNA–dye probes (Cy3 and Cy5) and intercalating dye SYBR Green, with an additional measure of side scattering. Black lines indicate gates of specific events. (d) Counts measured for indicated microRNA concentrations. Red line indicates linear dynamic range. (e) Fluorophore absorbance (A, red) and fluorescence (FL, blue) spectra for dyes to create optical barcodes. Gray highlight indicates emission bandpass filters. (f) Events in 2-color ratiometric channels using Cy3 and Cy5 dye combinations. (g) Intensity distribution of RCA products calibrated to the number of bound ssDNA–dye is shown in comparison with (h) the calculated number of free dye–ssDNA probes remaining in the focal volume for different target concentrations (3,000 fL focal volume). (i) Dependence of RCA product detection efficiency on target concentration in different focal volumes. Adapted with permission from ref 3. Copyright 2020 American Chemical Society.

Flow counting was applied for protein quantification by Wu et al., who captured targets on dispersed microbeads before in-solution RCA amplification. With this combination, they achieved detection limits in the low attomolar range at higher throughput than for bead-based detection in microwells and were able to distinguish 8 different proteins based on the bead optical codes.⁵⁷ A flow-based protein quantification platform from Singulex Inc. and MilliporeSigma (Erenna) was also commercialized that digitally counted fluorophores dissociated from microwell immunoassays plates. However, this platform was replaced with the static solution SMCxPRO system apparently to avoid complications and costs of the fluidics.⁵⁶

Challenges still remain with flow-based detection to achieve absolute counting at attomolar concentrations, primarily due to the imbalance between bound and unbound probes. Figure 8g shows normalized intensities of RCA products with a median of 271 dye labels with log-normal distribution. The lower LOD of the assay depends on the labeling density relative to the concentration of unbound dyes. When using ~1 nM probes for complete labeling in less than an hour, >1,000 free dyes remain in nanoliter focal volumes when targets are present at femtomolar and attomolar levels. As targets increase in concentration, unbound labels become depleted to shift the background level as shown in Figure 8h. Only <1% of targets can be counted individually at sub-picomolar concentrations using these conditions. To overcome this, Figure 8i shows that a reduced focal volume that effectively partitions the flow stream into smaller increments creates background levels containing fewer free dyes to reveal the densely labeled targets. A <100 fL focal volume is needed to detect all targets at attomolar concentrations, requiring flow streams narrower than ~1 μm, which have been created using custom microfluidic setups to interrogate individual molecules in solution.⁵⁸

OUTLOOK TOWARD ABSOLUTE COUNTING ASSAYS

Digital molecular analysis platforms are expanding in diversity and many early reports have shown outstanding promise toward future performance characteristics that far exceed those of analog assays (see representative metrics in Table 1). Nevertheless, at present, current commercial platforms remain more expensive, lower in throughput, and more challenging to use than their analog counterparts. The biophysical, thermodynamic, and kinetic limitations of conventional assays are magnified in digital platforms but can be anticipated to be overcome by integrating advanced probe, labeling reagents, fluid processing methods, and readout modalities. Orthogonal internal standards are likely to be needed to provide an assessment of the specificity and sensitivity of counts to real targets and will make a large impact when using real samples with variable matrices toward the goal of developing calibration-free absolute assays. Surface capture platforms will likely play a key role in the next decade in “second generation” analytical assays that use digitized counts as readouts due to the advanced instrumentation and established analytical framework that is widely available and amenable to short readout times and minimal sample processing. Electrochemical and label-free readouts have also seen substantial advances that could further simplify surface-capture assay designs.^59,60 However, throughput will continue to be limiting at the levels of surface capture and analysis for the lowest abundance targets. Absolute readouts will also be difficult to achieve in most cases without compartmentalized reactions, for which there are significant limitations in multiplexing, cost, and throughput. Flow-based analysis we believe is the “third generation” assay format that could achieve fast rates of comprehensive counting with multidimensional and absolute quantification and could mature in tandem with nanopore-based sequencing.^61,62 In fact, flow-based single molecule counting was called the “holy grail” of microscale molecular analysis in 2005.⁶³ However, new instrument designs are needed that can continuously tune flow rates and focal volumes and further enable parallelization and miniaturization.

Table 1.

Summary of Representative Analog and Digital Molecular Biomarker Assays

assay type	readout	biomarker class	t (h)a	dynamic range (log10)	LOD (copies)	multiplexing capacity	ref
sandwich immunoassay	analog	protein	~6	2	1 × 106	1	10
qPCR	analog	nucleic acid	1.5–5	7–8	~5	>3c	64
ddPCR	absolute	nucleic acid	~6	5	1	>3c	24, 32
Simoa	digital	protein	~4	>3	3 × 103b	6	23, 65
		nucleic acid	~5	4	2 × 105b	3	44
nCounter	digital	nucleic acid	~23	~3	5 × 103	>500	16
SMCxPRO	digital	protein	~4	>4	5 × 104	1	56
surface capture and counting	digital	nucleic acid	~3	5	600	1	35
		protein	~2	6	1 × 106	1	66
kinetic fingerprinting	digital	nucleic acid	~3	>4	60	1	52
		protein	3–5	>3	1 × 105	1	67
flow counting	digital	nucleic acid	4–9	4	6 × 105	12	3, 68
		protein	~3	~3	2 × 103b	8	57

^aEstimate of hands-on and automated run time from sample to results.

^bMedian across multiple reported targets or sequences.

^cSingle target analysis is most common.

Biographies

Chia-Wei Kuo is a Ph.D. student in Bioengineering in the lab of Prof. Andrew Smith at the University of Illinois Urbana–Champaign. He received a B.S. in Chemical Engineering from National Taiwan University. His current research focuses on developing new flow-based platforms for quantifying biomarkers and nanocrystal-conjugated DNA probes for analytical biological studies (e.g., DNA-PAINT and DNA-origami).

Andrew M. Smith is a Professor of Bioengineering, Materials Science & Engineering, and Technology Entrepreneurship at the University of Illinois Urbana–Champaign, and of Medicine at the Carle Illinois College of Medicine. He received a B.S. in Chemistry and a Ph.D. in Bioengineering, both from Georgia Institute of Technology. His research focuses on fluorescent molecular probes, single-molecule analytical assays, and therapeutic nanomaterials.