High-Throughput, High-Multiplex Digital Protein Detection with Attomolar Sensitivity

Abstract

A major challenge in many clinical diagnostic applications is the measurement of low-abundance proteins and other biomolecules in biological fluids. Digital technologies such as the digital enzyme-linked immunosorbent assay (ELISA) have enabled 1000-fold increases in sensitivity over conventional protein detection methods. However, current digital ELISA technologies still possess insufficient sensitivities for many rare protein biomarkers and require specialized instrumentation or time-consuming workflows that have limited their widespread implementation. To address these challenges, we have developed a more sensitive and streamlined digital ELISA platform, Molecular On-bead Signal Amplification for Individual Counting (MOSAIC), which attains low attomolar limits of detection, with an order of magnitude enhancement in sensitivity over these other methods. MOSAIC uses a rapid, automatable flow cytometric readout that vastly increases throughput and is easily integrated into existing laboratory infrastructure. As MOSAIC provides high sampling efficiencies for rare target molecules, assay bead number can readily be tuned to enhance signal-to-background with high measurement precision. Furthermore, the solution-based signal readout of MOSAIC expands the number of analytes that can simultaneously be measured for higher-order multiplexing with femtomolar sensitivities or below, compared with microwell- or droplet-based digital methods. As a proof of principle, we apply MOSAIC toward improving the detectability of low-abundance cytokines in saliva and ultrasensitive multiplexed measurements of eight protein analytes in plasma and saliva. The attomolar sensitivity, high throughput, and broad multiplexing abilities of MOSAIC provide highly accessible and versatile ultrasensitive capabilities that can potentially accelerate protein biomarker discovery and diagnostic testing for diverse disease applications.

Graphical Abstract

Many proteins and other biomolecules of potential diagnostic utility exist at very low concentrations in biological fluids. Achieving sufficient analytical sensitivity remains a major obstacle toward accurate measurements of these low-abundance biomarker candidates, particularly for proteins, which comprise a rich yet still relatively unmapped reservoir of biological information. To address this challenge, various ultrasensitive enzyme-linked immunosorbent assay (ELISA) tools have been developed, with digital detection methods such as digital ELISA providing up to 1000-fold improvements in sensitivity over conventional bulk analytical techniques.^1–5 In many existing digital measurement techniques, individual molecules are isolated in ultrasmall volume containers for generation of a signal at a locally high concentration, thus enabling single molecule counting. Single Molecule Arrays (Simoa) technology, the current state-of-the-art for ultrasensitive protein detection, utilizes such a signal compartmentalization approach with microwell arrays to achieve subfemtomolar limits of detection.¹ In digital ELISA methods, single protein molecules are captured on antibody-coated beads, which are used at a high excess to ensure a Poisson distribution in which each bead carries either one or zero target molecules. Upon formation of single immunocomplex sandwiches and labeling with an enzyme or other signal amplification moiety, single beads are isolated in individual compartments such as microwells or droplets, along with fluorogenic enzyme substrate or other reagent necessary for signal generation. As a result, single target molecules are isolated in individual compartments, and the number of “on” and “off” compartments is subsequently counted.

Despite vast improvements in sensitivity, many potential protein biomarkers remain undetectable in accessible biological fluids by digital ELISA techniques, thus necessitating sensitivity beyond subfemtomolar limits of detection. A key obstacle toward further improving sensitivity, however, is inefficient sampling of rare target molecules. For instance, in Simoa, only about 5% of the assay beads are analyzed, resulting in measurement imprecision at very low target concentrations due to Poisson noise, thus limiting sensitivity. At low attomolar concentrations where only a few hundred target molecules are present in a typical 100 μL sample volume, the Poisson noise, √N, where N is the number of events, contributes significantly to measurement uncertainty, in addition to experimental errors that further decrease measurement precision.⁶ Digital methods improve upon traditional ELISA by allowing individual analyte molecules to be counted, but in turn are still limited by the >90% of molecules that they are unable to count. The widespread use of digital ELISA methods has also been constrained because they typically require specialized instrumentation or have limited throughput due to extensive workflows involving, e.g., highly controlled droplet generation or microwell loading.^3,4

We previously developed digital ELISA platforms, including dropcast single molecule assays (dSimoa) and droplet digital ELISA, that improved sampling efficiencies and analytical sensitivity over Simoa by approximately 10-fold.^4,7,8 In particular, by generating a localized, nondiffusible fluorescent signal on each bead carrying a target molecule, dSimoa enabled a simple bead dropcasting method for counting single molecules captured on the beads. However, these methods require long imaging times of 20−30 minutes per sample to capture all the beads across multiple fields of view, thus limiting throughput. More recently, a magnetic-meniscus sweeping strategy was introduced to increase bead loading into microwell arrays and integrated with improved optics and image analysis to increase the overall percentage of beads analyzed.⁹ This enhanced sampling efficiency improved sensitivity over standard digital ELISA by up to two orders of magnitude. The requirement for highly specialized instrumentation and incubation times of up to 24 h, however, present practical limitations on the broad utility of this technology.

Here, we introduce a streamlined digital ELISA technology, Molecular On-bead Signal Amplification for Individual Counting (MOSAIC), that achieves attomolar sensitivities and requires only common laboratory infrastructure, thus vastly increasing the accessibility of ultrasensitive protein detection. By integrating our previously developed method for localized signal amplification of single molecules with the rapid, high-throughput detection capabilities of flow cytometry, we attain low attomolar limits of detection, with an order of magnitude improvement over the gold standard digital ELISA method, Simoa. Flow cytometry is a tool often used in the biological sciences for single cell analysis. It can achieve high sampling efficiencies because it avoids loading into microwells, but unlike dSimoa and droplet digital ELISA, its optics enable many thousands of events to be analyzed per second for high throughput.¹⁰ Particles are focused into a single-file line using sheath flow and detected using light sources and detectors that measure size and structure (with scattering) and fluorescence (with lasers and photomultipliers). While many bead-based flow cytometric immunoassays such as Luminex have been developed, these platforms utilize a bulk signal readout, i.e., signal detection from multiple target molecules per bead, with picomolar limits of detection.^11,12 In recent years, a few digital measurement strategies with flow cytometric readouts have been developed but have been limited to femtomolar limits of detection or above, with attomolar sensitivities yet to be demonstrated for protein detection.^13–15 Using a powerful on-bead signal amplification strategy, MOSAIC enables the detection of localized high-intensity signals from single target molecules with flow cytometry. We further exploit the high sampling efficiencies of MOSAIC to systematically reduce assay bead numbers for enhancing sensitivity to detect low attomolar protein concentrations. As a proof of concept, we demonstrate the abilities of MOSAIC to measure previously undetectable levels of cytokines in saliva. Furthermore, MOSAIC expands the multiplexing capabilities of digital ELISA, as the number of bead types that can be analyzed within a single sample is no longer limited by the total number of microwells or other compartments. To demonstrate the feasibility of increased multiplexing in MOSAIC, we simultaneously measure eight protein targets with attomolar to low femtomolar sensitivities using low volumes of plasma and saliva. MOSAIC simultaneously achieves enhanced sensitivity, high accessibility, a high-throughput readout, and expanded multiplexing for digital ELISA, thus providing a considerable advance over both existing digital ELISA and bead-based flow cytometry methods. These multifaceted capabilities can make ultrasensitive protein detection widely accessible and enable the discovery of previously undetectable biomarkers for diverse applications.

RESULTS AND DISCUSSION

Development of the MOSAIC Platform.

To establish an ultrasensitive digital ELISA platform with a rapid flow cytometric signal readout, we first utilized our recently developed method for generating a localized fluorescent signal from single target molecules captured on an excess number of antibody-coated beads (Figure 1).⁷ Upon formation of single immunocomplex sandwiches on the beads and labeling with streptavidin conjugated to a DNA primer-template pair, rolling circle amplification (RCA) is carried out to form a DNA concatemer attached to each immunocomplex sandwich. The incorporation of a fluorescently labeled DNA probe in the RCA reaction enables in situ hybridization to the concatemer, thus producing a strong fluorescent signal on each bead carrying a target molecule. As the amplified signal is attached to each immunocomplex sandwich, individual beads are subsequently analyzed by flow cytometry for counting of fluorescent “on” and “off” beads.

Figure 1.

Molecular On-bead Signal Amplification for Individual Counting (MOSAIC). Single target molecules are first captured with an excess number of antibody-coated paramagnetic beads, such that each bead carries either zero or one target molecule according to a Poisson distribution. Upon formation of single immunocomplex sandwiches with a biotinylated detector antibody and labeling with a streptavidin-DNA conjugate, rolling circle amplification (RCA) is carried out to generate a long DNA concatemer attached to each immunocomplex. Fluorescently labeled DNA probes are hybridized to the RCA product, allowing “on” and “off” beads to be counted via flow cytometry.

To evaluate the feasibility of single molecule counting using this readout method, we first applied MOSAIC toward the detection of interleukin-10 (IL-10), an anti-inflammatory cytokine with diverse roles across cancers, autoimmune diseases, and infectious diseases.^16,17 The “on” beads spanned a wide range of fluorescence intensities due to the heterogeneous size distribution of RCA concatemers (Figure S1). The cutoff value for distinguishing “on” from “off” beads was determined by fitting a two-Gaussian mixture model to the fluorescence intensity values using expectation maximization.¹⁸ At lower concentrations, almost all beads fell into the lower (“off”) Gaussian, and beads were counted as “on” if they were more than five standard deviations above the mean of this Gaussian. At higher concentrations, labels were predicted for the beads based on the likelihood that they were drawn from the “off” or “on” Gaussian. Using this algorithm, we were able to consistently quantify the average molecules per bead (AMB), a measure of the fraction of “on” beads. Importantly, the flow cytometric counting of “on” and “off” beads enables a rapid, high-throughput, and automatable digital readout, requiring less than a minute per sample. This readout time is similar to that of the current Simoa technology while utilizing common laboratory instrumentation and much faster than our recently developed dSimoa and droplet digital ELISA platforms.

High Sampling Efficiencies in MOSAIC Enable Improved Sensitivities.

Importantly, because nearly the entire bead solution can be analyzed by flow cytometry, MOSAIC attains high sampling efficiencies. Approximately 50−60% of the total initial assay beads are analyzed per sample, representing an over 10-fold increase in sampling efficiency over Simoa. We thus reasoned that the analytical sensitivity of MOSAIC can be maximized by reducing the number of total assay beads, increasing the ratio of target molecules to beads and thereby signal-to-background ratio, while maintaining sufficient measurement precision. We define the signal-to-background as the ratio of the AMB of a given sample concentration to the AMB of the blank sample, or background. The ability to enhance sensitivity using lower bead numbers in digital ELISA has been demonstrated in our previous work as well as in a recent microwell-based platform.^4,7,9 To exploit this possibility in MOSAIC for maximal sensitivity, we systematically varied assay bead number from 100 000 to 2000 for IL-10. Consistent with our hypothesis, the signal-to-background improved with decreasing bead number (Figure 2A). To quantify the improvement in signal-to-background, we normalized the signal-to-background of a given calibrator in each calibration curve to that of the 100 000-bead calibration curve, defining this normalized value as the relative signal-to-background. This relative signal-to-background increased as the number of beads was reduced, and a corresponding improvement in the limit of detection (LOD) was observed, with approximately an order of magnitude enhancement in sensitivity compared with 100 000 beads (Figure 2B). As bead number was further reduced to 2000 beads, however, the analytical sensitivity did not improve further, due to increasing effects of Poisson noise on measurement precision at very low bead numbers. Maximal sensitivity was attained at 20 000 beads, with an LOD of 15.9 aM, representing an over 12-fold enhancement in sensitivity over the corresponding Simoa assay (Figure 2C and Table 1). Standard Simoa assays typically use 500 000 total beads, of which approximately 5%, or 25 000 beads, are loaded and imaged in the microwell array. As shown by the present work, the signal-to-background is often maximized at lower assay bead numbers. To reduce assay bead number with Simoa while maintaining consistent bead loading, some of the beads may be “helper beads”—unconjugated dye-encoded beads that are ignored in the analysis but help ensure efficient loading of the assay beads.¹⁹ For a closer comparison to MOSAIC, we used 100 000 assay beads with 400 000 helper beads for improved signal-to-background, despite fewer total beads analyzed.

Figure 2.

Analytical sensitivities of MOSAIC assays across different analytes and assay bead numbers. (A) Calibration curves for IL-10 MOSAIC assays using different assay bead numbers. The signal readout is denoted by average molecules per bead (AMB). (B) Effect of MOSAIC assay bead number on the limit of detection (LOD) and signal-to-background for IL-10 detection. LOD values were calculated as three standard deviations above the background AMB. Relative signal-to-background for each assay bead number was determined as the signal-to-background of a specific calibrator normalized to the signal-to-background of the same calibrator using 100 000 assay beads. (C) Calibration curve for the corresponding IL-10 Simoa assay, using 100 000 assay beads and 400 000 helper beads. (D) Effects of MOSAIC assay bead number on LOD and relative signal-to-background across additional analytes. All error bars represent the standard deviations of three replicates, with six replicates performed for the blank.

Table 1.

Limit of Detection (LOD) and Lower Limit of Quantification (LLOQ) Values of Optimized MOSAIC Assays and the Corresponding Simoa Assays for Various Analytes^a

		limit of detection (aM)		lower limit of quantification (aM)
analyte	MOSAIC assay bead number	MOSAIC (3×)	Simoa (3×)	MOSAIC (10×)	Simoa (10×)
IL-10	20 000	15.9	200.3	45.9	753.3
IFN-γ	10 000	18.9	125.1	47.0	441.8
IL-6	20 000	227.9	661.2	699.0	1965.5
IL-1β	50 000	124.0	690.2	401.5	2032.8
IL-8	10 000	37.9	179.9	120.2	639.1
IL-12p70	100 000	5.8	69.6	19.9	217.6

^aLOD and LLOQ values are calculated as three and ten standard deviations, respectively, above the background.

To assess whether reducing bead number can similarly enhance the sensitivity of MOSAIC for other analytes, we expanded the MOSAIC platform to additional cytokines and performed similar variations of bead number for each analyte (Figure 2D, Figure S1, and Table S1). Improved signal-to-background upon decreasing bead number from 100 000 to 10 000 was also observed for interferon-gamma (IFN-γ), interleukin-8 (IL-8), and interleukin-12 p70 (IL-12p70), with optimal sensitivity achieved at 10 000 beads for IFN-γ and IL-8. For IL-12p70, despite the increased signal-to-background, sensitivity remained similar across all bead numbers, with the lowest LOD of 5.8 aM attained using 100 000 assay beads. For IL-6 and IL-1β, there was little increase in the signal-to-background as bead number was reduced from 100 000. Consistent with this observation, the LOD did not improve much with decreasing assay bead number. We observed only 3-to 5-fold improvements in sensitivity over the corresponding Simoa assays for these analytes, as expected from the lack of signal-to-background enhancement; these improvements can be attributed to increased sampling efficiencies. Our results demonstrate that MOSAIC achieves 3−12 times higher sensitivity over Simoa, the current state-of-the-art for ultrasensitive protein detection (Table 1 and Figures S2–S3). These improvements were also reflected in the lower limits of quantification (LLOQs), which represent the lowest analyte concentrations that can be quantified with acceptable precision and accuracy. Because of the much higher sampling efficiencies of MOSAIC compared to standard digital ELISA, much lower bead numbers down to 10 000 beads can be used to enhance the signal-to-background for several analytes, while maintaining sufficient measurement precision. While the extent of sensitivity improvement is dependent upon the particular antibody pair, assay bead number can be readily optimized for each analyte. The high-throughput readout of MOSAIC further facilitates rapid optimization of assay bead number, which is less practical with our previously developed dropcast film- or droplet-based digital ELISA platforms. We hypothesize that differences in capture antibody affinity play an important role in determining the degree of improvement in signal-to-background for the different assays and lead to sensitivity differences when reducing assay bead numbers. Although the total capture antibody concentration decreases as assay bead number is reduced, high-affinity capture antibodies can maintain high target capture efficiencies and thereby increase the overall fraction of “on” beads with reduced assay bead number. However, for lower-affinity capture antibodies, kinetic limitations may decrease the number of captured target molecules as assay bead number is reduced, yielding little change in the fraction of “on” beads.

As decreasing the number of assay beads also reduces the number of capture antibody molecules present, capture kinetics may be slower. We thus investigated whether longer target capture times could further enhance the signal-to-background ratio and analytical sensitivity. For both IL-10 and IFN-γ, however, increasing target capture times from 1 h to 4 h resulted in similar or slightly worse LODs (Figure S4). We therefore used 1 h target capture times in subsequent experiments for a more efficient workflow.

To determine the extent of sensitivity improvement arising from improved sampling efficiencies in MOSAIC, we performed random sampling of subsets of beads among the calibration curves obtained for IL-10 and IFN-γ and examined the effects of number of beads analyzed on sensitivity and precision. Consistent with the high Poisson noise that occurs at low event numbers, the LODs obtained when only a few hundred beads were analyzed were very high, across all starting assay bead numbers (Figure 3A,B). For each subset size, random sampling was repeated 100 times, and high variation in calculated LODs among replicate subsets was observed when 1000 or fewer beads were analyzed, in line with the poor precision and reproducibility expected at these low sampling efficiencies. As increasing numbers of assay beads are analyzed, analytical sensitivities improve considerably, with smaller gains in sensitivity as the number of analyzed beads increases past several thousand. Furthermore, the improvement in sensitivity with increasing sampling efficiency corresponded to a decrease in the measurement coefficients of variation of the background signal (Figure 3C,D). Our results thus empirically support the important role of improved sampling efficiencies in the enhanced sensitivity of MOSAIC.

Figure 3.

Increasing sampling efficiency improves analytical sensitivity and precision. (A,B) Limits of detection for calibration curves generated from subsets of varying bead numbers randomly sampled from MOSAIC calibration curves for IL-10 (A) and IFN-γ (B). Each color denotes the starting total assay bead number. Each point represents the median of 100 randomly sampled subsets, with error bars representing the interquartile range. Open circles denote values for which the upper quartile had a positive infinity value due to very high measurement coefficients of variation (CVs) at very low bead subset sizes. (C,D) Measurement CVs of the background signal for randomly sampled bead subsets of varying bead numbers for IL-10 (C) and IFN-γ (D).

MOSAIC Improves Detectability of Low-Abundance Analytes in Biological Fluids.

We next explored the performance of MOSAIC in biological fluids and whether its enhanced sensitivity can improve the detectability of low abundance biomarkers. As a representative cytokine, we measured IFN-γ using both MOSAIC and the corresponding Simoa assay in a cohort of human plasma samples (Figure 4A,B). The IFN-γ concentrations measured by MOSAIC, using 10 000 assay beads, showed good correlation with those measured by Simoa, supporting the accuracy of the MOSAIC assay. Furthermore, although IFN-γ was generally detectable in plasma using both methods, one out of the 17 plasma samples remained undetectable by Simoa, while the more sensitive MOSAIC assay achieved 100% detectability.

Figure 4.

Measurements of IFN-γ concentrations in human plasma and saliva using MOSAIC and Simoa. (A−D) Measured IFN-γ concentrations in the 17 plasma (A,B) and 26 saliva (C,D) samples using MOSAIC and Simoa. Concentrations shown represent the measured concentration values in the 4-fold diluted plasma samples. The Pearson correlation coefficients were 0.80 and 0.31 for the plasma and saliva (among detectable values) samples, respectively. The low correlation coefficient in saliva may be attributed to the small fraction of detectable samples using Simoa as well as some samples with IFN-γ levels near the LOD of Simoa or MOSAIC. Red dashed lines denote assay LODs, which were calculated as three standard deviations above the AMB of the background (buffer only). Samples with measurements below the assay LOD were assigned a value equal to the LOD. Samples for which measured values were below the LOD of either or both assays are denoted by an open circle. Error bars represent the standard deviations of two replicates.

To further assess whether the superior sensitivity of MOSAIC can improve the detectability of low abundance analytes in biological fluids, we applied MOSAIC to saliva. As saliva contains a minimal serum component filtered from blood through the salivary glands, many potential biomarkers exist at much lower levels in saliva than in blood, necessitating ultrasensitive techniques. As a proof of concept, we used MOSAIC to measure the levels of IFN-γ in a cohort of saliva samples. While IFN-γ was detectable in only 42% (11/26) of the saliva samples using Simoa, the enhanced sensitivity of the low-bead MOSAIC assay improved the detectability of IFN-γ to over 65% (17/26) of the saliva samples (Figure 4C,D). All saliva samples with detectable IFN-γ levels using Simoa were detectable by MOSAIC, with positively correlated measurements. The wider variability between MOSAIC and Simoa measurements in saliva compared with plasma may be attributed to differing matrix interference effects with the different assays, which may particularly affect the detectability of low target concentrations near the assay LODs. However, we observed recoveries within an acceptable range of 70−130% when performing spike and recovery experiments in saliva (Table S2). The ability of MOSAIC to detect IFN-γ levels in several saliva samples that were undetectable by Simoa despite being above its LOD may be attributed not only to its enhanced signal-to-background and sensitivity at low bead numbers, but also to higher sampling efficiencies that increase measurement precision at low concentrations. In addition, the more extensive washing performed in MOSAIC compared to Simoa, in order to minimize RCA products amplified from excess streptavidin-DNA, may contribute to improved removal of interfering components. The improved detectability of endogenous proteins at concentrations above the LOD of Simoa is consistent with our observation of a similar phenomenon in our previous evaluations of droplet digital ELISA and dSimoa assays.^4,7

MOSAIC’s ability to detect attomolar concentrations of endogenous proteins, which are undetectable by Simoa, highlights its potential diagnostic utility for rare biomarkers. Using MOSAIC, we also achieved improved detectability of an even lower-abundance analyte in saliva, IL-12p70, from 0% by Simoa to approximately 12%, or 3/26 saliva samples (Figure S5). While the sensitivity of MOSAIC remained insufficient to detect IL-12p70 in the majority of the tested saliva samples, our results demonstrate that its enhanced sensitivity can begin to uncover the “tip of the iceberg” of such rare analytes in saliva but also underscore the need for even more sensitive methods.

Multiplexing Capabilities of MOSAIC.

In addition to the enhanced sensitivity of MOSAIC, its on-bead signal generation strategy and flow cytometric readout can expand the multiplexing capabilities of digital ELISA. The ability to simultaneously measure multiple analytes in a single sample can accelerate sample throughput and biomarker signature discovery and is especially critical in applications with limited sample volumes, such as neonatal saliva or fingerprick blood. Multiplexing in digital ELISA as well as in a recently developed ultrasensitive planar ELISA platform has shown great utility in various diagnostic applications, but the number of bead types that can be analyzed simultaneously in bead confinement methods is limited by the total number of compartments and low bead analysis efficiencies.^20–23 In contrast, the solution-based readout of MOSAIC enables, in principle, an unlimited number of bead types to be analyzed in one sample with readily tunable sensitivities and dynamic ranges via assay bead number for each analyte (Figure 5A). Furthermore, the versatility of flow cytometry enables a wide range of beads to be distinguished by fluorescence wavelengths and intensities, as well as bead sizes. To verify the multiplexing capabilities of MOSAIC, we first developed a multiplexed assay for IL-6, IL-1β, IL-10, and IFN-γ, using beads encoded with different fluorescent dyes. This assay demonstrated mid- to high-attomolar sensitivities and acceptable recoveries in human plasma, with little cross-reactivity in the concentration ranges of the cytokines present in plasma (Table S3 and Figure S6). As further validation, its measurements in plasma correlated well with those of the corresponding four-plex Simoa assay, particularly for analyte concentrations well above the LODs of both methods (Figure S7).

Figure 5.

Multiplexing with MOSAIC technology. (A) Schematic of multiplexing with MOSAIC. Beads coated with antibodies to different target analytes are encoded by using different fluorescent dyes with different wavelengths, intensities, and/or using multiple bead sizes. Upon capture of the single analyte molecules on each bead type, formation of single immunocomplex sandwiches, and labeling with streptavidin-DNA, rolling circle amplification is carried out and the mixture of beads is analyzed by flow cytometry. Beads are differentiated by a series of gates in different fluorescence channels, and the average molecules per bead for each bead type is then determined from the intensities in the fluorescence channel corresponding to the probe color. (B,C) Measured concentrations of eight protein analytes in human plasma (top) and saliva (bottom) using an eight-plex MOSAIC assay (B) and two four-plex Simoa assays (C). Concentrations shown are the measured concentration values in the 16-fold and 8-fold diluted plasma and saliva samples, respectively. Measurements below the assay LOD are assigned a value equal to the LOD and denoted by open symbols.

Having established the ability of MOSAIC to measure multiple analytes simultaneously with high accuracy, we then explored the increased multiplexing versatility of MOSAIC by incorporating additional beads with different fluorescent dyes and intensities. As a proof of principle, we integrated MOSAIC assays for the cytokines IL-6, IL-1β, IL-10, IFN-γ, IL-12p70, IL-5, and IL-18, and vascular endothelial growth factor (VEGF) into an eight-plex assay. High analytical sensitivities were maintained, with LOD values ranging from mid-attomolar to low femtomolar concentrations (Figure S8 and Tables S4–S5). The simultaneous digital detection of eight proteins with attomolar to low femtomolar sensitivities surpasses the multiplexing capabilities of any previously reported digital ELISA method. We applied the assay to both human plasma and saliva, using 16-fold and 8-fold dilution factors, respectively, to ensure acceptable recoveries across all analytes in spike and recovery experiments and consistent dilution linearity (Figures S9–S10 and Tables S6–S8). Importantly, there was little cross-reactivity across the measured concentration ranges (Figure S8). The exceptional sensitivity of MOSAIC further enables the use of higher dilution factors to minimize matrix interference effects and potential false signals due to cross-reactivity, in addition to allowing more clinical information to be collected from limited sample volumes. Using less than 15 and 30 μL of plasma and saliva, respectively, across two replicates, we measured these eight proteins in a cohort of plasma and saliva samples and compared the measurements to those of two four-plex Simoa assays (Figure 5B,C and Figure S12). The eight-plex MOSAIC assay successfully detected these endogenous proteins at attomolar to femtomolar concentrations in both plasma and saliva. The measured concentrations were generally correlated between the two methods for analytes with high detectability, further supporting the accuracy of the high-multiplex MOSAIC assay. Thus, MOSAIC enables ultrasensitive multiplexed measurements of a broad biomarker panel using very small sample volumes. In addition, although the majority of the proteins in our eight-plex MOSAIC assay were detected at attomolar to low femtomolar concentrations in the plasma and saliva samples, the multiplex assay was also able to measure several proteins at higher concentrations in the same sample volume. Thus, the versatility of MOSAIC allows simultaneous measurements of proteins across a wide range of concentrations. Although we achieved sensitivities ranging from mid-attomolar to low femtomolar concentrations across the eight-plex assay, further improvements in sensitivity to low attomolar concentrations across all analytes in a biomarker panel may be necessary for specific clinical applications. As the number of bead types that can be analyzed by MOSAIC in a single sample is not constrained by compartments, we expect that such sensitivity improvements can be achieved by selection of higher-affinity antibodies and optimization of surface chemistries across all beads to minimize nonspecific binding.

CONCLUSIONS

We have developed a single-molecule protein detection platform, MOSAIC, that simultaneously achieves attomolar sensitivities, high accessibility, a high-throughput readout, and expanded multiplexing capabilities with digital ELISA. While digital measurement technologies have enabled 1000-fold increases in sensitivity, proteins at attomolar or lower concentrations remain inaccessible, limiting efforts in biomarker discovery and early detection of diseases in which biomarker levels may be very low during initial stages. One notable example is saliva, which is collected noninvasively and contains diverse proteins of potential diagnostic value but at much lower levels than in plasma, representing an exceptional challenge to existing digital ELISA technologies. Our proof-of-principle measurements of IFN-γ and IL-12p70 in saliva highlight the potential utility of MOSAIC for unveiling uncharted territories of rare biomarker candidates. Because of the high sampling efficiencies of MOSAIC, we were able to reduce the number of assay beads to enhance the signal-to-background while maintaining sufficient measurement precision. Importantly, MOSAIC circumvents the requirement for specialized instrumentation, which has restricted widespread implementation of digital ELISA technologies. By eliminating the need to isolate beads into individual compartments, MOSAIC enables single molecule counting using flow cytometry, thus translating digital ELISA to existing, widely available laboratory instrumentation. In MOSAIC, each sample is analyzed in less than a minute, overcoming one major weakness of newer, more-sensitive digital ELISA techniques (including dSimoa and droplet digital ELISA) compared to commercial, automated Simoa instrumentation.^4,7,23 Automation via the 96-well plate sampling modes that are already built into many benchtop flow cytometers further provides a streamlined workflow.

In addition to enhanced sensitivity and an accessible workflow, MOSAIC introduces increased multiplexing capabilities to digital ELISA by eliminating the physical constraint of compartments on the total number of bead types that can simultaneously be analyzed. As sampling efficiencies in MOSAIC are unaffected by the total number of beads, the desired dynamic range for each analyte in a multiplex MOSAIC assay is readily tunable by the assay bead number, with no upper limit on the number of beads or bead types in each sample. Although our recently developed dSimoa technology also removes the constraint of compartments on the number of beads that can be analyzed, practical implementation of high-multiplex dSimoa assays is limited by the need to increase dropcast bead film sizes with increasing bead numbers, requiring extended imaging times. In addition to multiplexing by color and fluorescence intensity, the versatile flow cytometric readout provides the ability to distinguish beads by size, thus expanding the palette of multiplexable bead types. While this work demonstrates an eight-plex MOSAIC assay as a proof of concept, the potential to expand digital multiplexing up to a hundred or more proteins by encoding beads with ratios of fluorescent dyes provides a highly promising direction for future exploration. The multiplexing capabilities of MOSAIC can be further expanded by using combinatorial pairs of encoded capture beads and detector antibodies barcoded with different DNA primer sequences, with corresponding fluorophore-labeled DNA probes. While cross-reactivity remains a potential limitation in higher order multiplex assays, the attomolar sensitivity of MOSAIC can enable higher dilution factors to be used while maintaining good target detectability, thus keeping measurements in concentration ranges where cross-reactivity is minimal. In addition, smaller multiplex panels may be used with subsequent combination of all beads into a single sample for signal readout, as well as sequential target capture.²⁴

Despite the low attomolar sensitivities of MOSAIC, further work is required to achieve even higher sensitivity and detectability for rare analytes. With the development and screening of higher-affinity reagents, we expect that sub-attomolar LODs can be attained. While most of the MOSAIC assays developed in this study maintained about three orders of magnitude in dynamic range, the current MOSAIC platform and analysis method are limited to the digital measurement regime because of the heterogeneity in the fluorescence intensities of the “on” beads. As certain analytes may exist over a broader range of concentrations in biological fluids, higher assay bead numbers can be used for a wider dynamic range, thus requiring specific optimization of bead number for each application. Furthermore, while washing steps in the current MOSAIC workflow are automated using a microplate washer, future work will explore integration of the entire assay into an automated liquid handling platform for improved measurement precision. We will also develop MOSAIC assays with shorter incubation and signal amplification periods to reduce overall assay times from the current three hours for specific clinical applications that require shorter turnaround times. In summary, by leveraging the power of on-bead signal amplification and the multifaceted capabilities of flow cytometry, MOSAIC provides an ultrasensitive protein detection method with attomolar sensitivity, higher-order multiplexing, and a rapid high-throughput readout, accessible to any lab with a flow cytometer.

METHODS

Materials.

All affinity reagents, recombinant proteins, and DNA oligos used in this work are listed in the Supporting Information (Tables S9–S10). Buffers and paramagnetic beads were purchased from Quanterix Corporation and Bangs Laboratories. Custom DNA oligos were purchased from Integrated DNA Technologies and MilliporeSigma.

Preparation of Capture and Labeling Reagents.

Capture antibodies were buffer exchanged with Bead Conjugation Buffer (Quanterix) using a 50K Amicon Ultra-0.5 mL centrifugal filter (MilliporeSigma). After Bead Conjugation Buffer was added to the antibody solution in the filter up to 500 μL, buffer exchange was carried out by centrifuging three times at 14 000g for five minutes, with addition of 450 μL Bead Conjugation Buffer between centrifugation cycles. The buffer-exchanged antibody was recovered by inverting the filter into a new tube, centrifuging at 1000g for two minutes, rinsing the filter with 50 μL Bead Conjugation Buffer, and centrifuging one more time at 1000g for two minutes. The concentration of the buffer-exchanged antibody was then measured using a NanoDrop spectrophotometer. For each bead type, beads were washed three times with 300 μL of Bead Wash Buffer (Quanterix) and two times with 300 μL Bead Conjugation Buffer (Quanterix) before resuspending in cold Bead Conjugation Buffer. Bead number and conjugation conditions for each analyte are shown in Table S11. A 1 mg vial of 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide hydrochloride (EDC) (Thermo Fisher Scientific) was dissolved in 100 μL of cold Bead Conjugation Buffer, and the desired volume was added to the beads. The beads were shaken for 30 min at either room temperature or 4 °C. After EDC activation of the carboxyl groups on the beads, the beads were washed once with 300 μL cold Bead Conjugation Buffer before resuspension in the buffer-exchanged antibody solution. Antibody conjugation was carried out by shaking the beads for two hours at either room temperature or 4 °C, followed by washing twice with 300 μL Bead Wash Buffer. The antibody-coupled beads were then blocked for 30 min at room temperature with shaking in 300 μL Bead Blocking Buffer (Quanterix). After washing once each with 300 μL Bead Wash Buffer and Bead Diluent (Quanterix), the beads were resuspended in Bead Diluent, counted with a Beckman Coulter Z1 Particle Counter, and stored at 4 °C.

Detector antibodies were obtained in biotinylated form except for the IFN-γ detector antibody, which was obtained in unmodified form. The IFN-γ detector antibody was biotinylated by reconstituting to 1 mg/mL in Biotinylation Reaction Buffer (Quanterix) and adding a 40-fold molar excess of NHS-PEG4-Biotin (Thermo Fisher Scientific) freshly dissolved in water. The biotinylation reaction was carried out for 30 min at room temperature, followed by purification of the biotinylated antibody with a 50K Amicon Ultra-0.5 mL centrifugal filter. Five centrifugation cycles of 14 000g for five minutes with addition of 450 μL of Biotinylation Reaction Buffer between cycles were performed, with subsequent recovery of the purified antibody via inversion of the filter into a new tube and centrifuging at 1000g for two minutes. The filter was then rinsed with 50 μL Biotinylation Reaction Buffer before centrifuging one more time at 1000g for two minutes and quantifying the antibody concentration with a NanoDrop spectrophotometer.

Preparation of Streptavidin-DNA Conjugate.

A 5′ azide-modified primer was annealed to the DNA template for RCA by heating a solution of 33.8 μM primer and 40.6 μM template in NEBNext Quick Ligation Buffer (New England Biolabs) at 95 °C for two minutes and allowing to cool to room temperature over 90 min. Ligation was then performed with addition of T4 DNA ligase and incubation at room temperature for three hours. The ligation reaction was then heated at 65 °C for 10 min to inactivate the ligase, with subsequent cooling to room temperature. The ligation reaction was buffer exchanged into phosphate buffered saline (PBS) with 1 mM EDTA using a 7K MWCO Zeba spin desalting column (Thermo Fisher Scientific). For conjugation, streptavidin (Biolegend 280302) was buffer exchanged into PBS using a 10K Amicon Ultra-0.5 mL centrifugal filter, incubated with a 20-fold molar excess of dibenzocyclooctyne-PEG4-N-hydroxysuccinimidyl ester (DBCO-PEG4-NHS, MilliporeSigma) for 30 min at room temperature, and purified with a 10K Amicon Ultra-0.5 mL centrifugal filter in PBS with 1 mM EDTA. A 2-fold molar excess of the ligated primer-template was then added to the DBCO-modified streptavidin and incubated overnight at 4 °C. The conjugate was stored in aliquots at −80 °C in PBS with 5 mM EDTA, 0.1% BSA, and 0.02% sodium azide.

MOSAIC Assays.

MOSAIC assays were performed in a 96-well plate (Greiner Bio-One, 655096), with antibody-coated beads and detector antibodies diluted to the desired concentrations in Homebrew Sample Diluent (Quanterix). Assay conditions for each analyte are listed in Table S12. Sample volumes of 100 μL were used, with 10 μL of antibody-coated beads. For two-step assays, 10 μL of detector antibody was added to each sample. The plate was sealed and shaken for one hour for target capture, followed by washing with System Wash Buffer 1 (Quanterix) using a BioTek 405 TS Microplate Washer. For three-step assays, 100 μL of detector antibody was added to the beads after target capture and washing steps, followed by a 10 min incubation and additional washing. The immunocomplex sandwiches were then labeled with streptavidin-DNA by adding 100 μL of the conjugate diluted in Sample Diluent with 5 mM EDTA to the beads and shaking for the desired time. The samples were then washed with System Wash Buffer 1 for eight cycles, transferred to a new 96-well plate, and washed an additional time with 200 μL of System Wash Buffer 1 before being resuspended in 60 μL of the RCA reaction mixture. The RCA mixture consisted of 0.5 mM deoxynucleotide mix (New England Biolabs), 0.33 U/uL phi29 DNA polymerase (Lucigen), 0.2 mg/mL bovine serum albumin (BSA, New England Biolabs), 1 nM fluorescently labeled DNA probe (Integrated DNA Technologies), and 0.1% Tween-20 in 50 mM Tris-HCl (pH 7.5), 10 mM (NH₄)₂SO₄, and 10 mM MgCl₂. ATTO-647N and ATTO-565 labeled DNA probes were used for single target and multiplex MOSAIC assays, respectively. Upon addition of the RCA mixture to each sample, the plate was shaken for 1 h at 37 °C, followed by addition of 150 μL PBS with 0.1% Tween-20 and 5 mM EDTA to stop the reaction. Samples were washed one time with 200 μL of the same PBS-Tween-EDTA buffer and resuspended in 100 μL of the buffer with added 0.1% BSA. Samples were measured using a CytoFlex LX flow cytometer (Beckman Coulter) equipped with six lasers, in either tube or plate sampling mode. Bleach and buffer wells were included between different samples to minimize potential sample carryover. Multiplex MOSAIC assays were carried out following the same protocol as for the single-plex MOSAIC assays, with different fluorescent dye-encoded beads combined in the same sample.

Plasma and saliva samples were diluted in Sample Diluent or StartingBlock Blocking Buffer (Thermo Fisher Scientific), respectively, with protease inhibitor (Halt Protease Inhibitor Cocktail, Thermo Fisher Scientific). Plasma samples were obtained from BioIVT and the Mass General Brigham Biobank, and saliva samples were obtained from BioIVT. All human samples were deidentified, and experiments were performed under Institutional Review Board approval by Mass General Brigham. All plasma and saliva samples were centrifuged at 2000g for 10 min or 21 000g for 20 min, respectively, at 4 °C before diluting for measurements.

Simoa Assays.

All Simoa assays were performed on an HD-X Analyzer (Quanterix), with automated sample processing, image analysis, and calculations of average enzymes per bead (AEB). Assay conditions for each analyte are listed in Table S12. The same antibody-coated beads used in the single-plex MOSAIC assays were used for all Simoa assays, with the same detector antibody concentrations as in the MOSAIC assays. For each single-plex Simoa assay, 100 000 antibody-coated beads and 400 000 helper (nonconjugated) beads were used; for each four-plex Simoa assay, 125 000 antibody-coated beads per analyte were used. Streptavidin-β-galactosidase (SβG) Concentrate (Quanterix) was diluted in SβG Diluent (Quanterix) to the desired concentration for each assay. All assay reagents and consumables were loaded to the HD-X Analyzer according to the manufacturer’s instructions.

Data Analysis.

Flow cytometry data were first analyzed with FlowJo Software (Becton, Dickinson, and Company); beads were identified using gates on forward scatter and side scatter. Single beads were additionally gated using forward scatter (Figure S1). Beads were distinguished from other particulate matter, and multiplex beads of different colors were separated, based on encoded fluorescent dyes (Figure S13). These populations were typically easily separable, so consistent gates could be used for each batch, and often across batches. Next, the probe fluorescence intensities for each bead population were exported for automated analysis in Python.

RCA probe fluorescence intensities were analyzed in Python with the packages jax, numpy, pandas, scikit-learn, and waltlabtools. The analysis proceeded using the following steps: (1) A two-Gaussian mixture model was fit to the log-transformed fluorescence intensities for each well using expectation maximization.²⁵ This modeled each bead’s fluorescence intensity as being drawn from one of two normal distributions: the “off” distribution (with a lower mean) or the “on” distribution (with a higher mean). (2) Two methods of counting “on” beads were calculated. In the first method, beads were assigned as “on” or “off” based on their predicted membership in one or the other Gaussian according to the expectation maximization algorithm. In the second method, beads were counted as “on” if they were at least five standard deviations above the mean of the lower (“off”) Gaussian peak. The first method yielded better results as long as two peaks were identifiable, but at very low numbers of “on” beads, the higher Gaussian was harder to fit, so the second method provided more reliable estimates. Accordingly, the metric used was a weighted average of the two, with a sinusoidal weighting function: the weight of the first method is given by sin⁴(πf_on/2), where f_on is the fraction of “on” beads according to the first method (Gaussian mixture assignment). At very low concentrations, the few “on” beads were identified based on a fluorescence intensity much higher than the overall population; at higher concentrations, they were identified on the basis of their membership in the “on” group. (3) The fraction of “on” beads was then converted to an AMB using Poisson statistics. (4) These AMBs were mapped to concentrations using a four-parameter logistic calibration curve. The LOD and LLOQ were calculated as three and ten standard deviations above the background, respectively, where a correction factor c₄(n) was applied for unbiased estimation of the standard deviation.²⁶ Pearson correlation coefficients were calculated using GraphPad Prism.