Nanopore sequencing of DNA-barcoded probes for highly multiplexed detection of microRNA, proteins and small biomarkers

There is an unmet need to develop low-cost, rapid and highly multiplexed diagnostic technology platforms for quantitatively detecting blood biomarkers to advance clinical diagnostics beyond the single biomarker model. Here we perform nanopore sequencing of DNA-barcoded molecular probes engineered to recognize a panel of analytes. This allows for highly multiplexed and simultaneous quantitative detection of at least 40 targets, such as microRNAs, proteins and neurotransmitters, on the basis of the translocation dynamics of each probe as it passes through a nanopore. Our workflow is built around a commercially available MinION sequencing device, offering a one-hour turnaround time from sample preparation to results. We also demonstrate that the strategy can directly detect cardiovascular disease-associated microRNA from human serum without extraction or amplification. Due to the modularity of barcoded probes, the number and type of targets detected can be significantly expanded.

There is an unmet need to develop low-cost, rapid and highly multiplexed diagnostic technology platforms for quantitatively detecting blood biomarkers to advance clinical diagnostics beyond the single biomarker model.Here we perform nanopore sequencing of DNA-barcoded molecular probes engineered to recognize a panel of analytes.This allows for highly multiplexed and simultaneous quantitative detection of at least 40 targets, such as microRNAs, proteins and neurotransmitters, on the basis of the translocation dynamics of each probe as it passes through a nanopore.Our workflow is built around a commercially available MinION sequencing device, offering a one-hour turnaround time from sample preparation to results.We also demonstrate that the strategy can directly detect cardiovascular disease-associated microRNA from human serum without extraction or amplification.Due to the modularity of barcoded probes, the number and type of targets detected can be significantly expanded.
The detection of blood serum biomarkers is one of the most common methods for diagnosis, prognosis, predicting future disease and monitoring response to treatment.Biochemicals have been used for over 60 years to aid diagnosis of a wide range of conditions, including cancer 1 , pregnancy 2 and cardiac disease 3 .Biomarker tests traditionally rely on the detection of proteins to indicate a condition.However, these tests often lack the specificity to provide clinically useful detail on the pathology 4,5 .To increase the specificity in detecting diseases, multiplexed biomarker assays have been developed.Such strategies have been attempted in several areas, including Alzheimer's disease 6,7 , amyotrophic lateral sclerosis 8 , cardiovascular disease [9][10][11] , chronic obstructive pulmonary disease 12 , infection 13 and cancers [14][15][16][17] .Many of the methodologies employed rely upon antibody recognition of an epitope and an associated optical readout for each analyte, for example, ELISA, which can be sensitive to pM levels of analyte.Commercially available systems can detect up to 80 proteins simultaneously 18,19 .Other methods use modified DNA-based aptamers, which allow the detection of multiple proteins in a single sample with a sensitivity down to 125 fM (refs.20,21).These techniques offer promising tools for high-throughput protein biomarker detection but are unable to observe multiple molecular species simultaneously.
Clinical diagnostics are now moving beyond protein biomarkers and genetic testing.One example is microRNAs (miRNAs), short non-coding RNAs that regulate gene expression 22,23 .Alterations in miRNA expression have been identified in a wide range of clinical areas, such as cardiology 24 , hepatology 25 , nephrology 26 , neurology 27 , oncology 28 and vascular disease 29 .These molecules are detected in the blood primarily through reverse transcription-quantitative PCR (RT-qPCR) 30 , which is sensitive down to fM.However, this method requires multiple steps and relies on signal amplification, which may introduce bias in the measurement.Furthermore, the rapid degradation of miRNAs provides a particular problem when considering their https://doi.org/10.1038/s41565-023-01479-zBiological nanopores are advantageous over solid-state devices as they are highly reproducible and can be engineered for specific functionality 44 .In particular, biological nanopores are useful for DNA/RNA sequencing, as shown by Oxford Nanopore Technologies (ONT).ONT has commercialized sequencing devices that use biological nanopore arrays to allow for high-throughput, simultaneous DNA/RNA reads 45 of fragments ≥20 base pairs 46 .
This study showcases a multiplexed analyte detection strategy, combining nanopore sequencing with DNA-barcoded molecular probes.The platform allows accurate demultiplexing of events, with simultaneous quantitative detection of at least 40 molecules that can consist of miRNAs, proteins and small molecules, such as neurotransmitters.The presence of each analyte is determined by the translocation dynamics of each probe as it passes through a nanopore.In this study, we selected 40 miRNAs and proteins implicated in cardiac disease.The method established is easily adaptable and scalable, meaning the number of detected biomarkers can be extended to cover multiple diseases.The assay requires a sample volume of less than 30 μl, does not require sample labelling or amplification and costs less than US$100.Furthermore, the technology bears the potential for pooled use in a clinical setting.A rapid miRNA profiling platform would offer the potential to capture short-lived events and perform frequent longitudinal testing.
There is, therefore, a great need to develop technologies that can perform highly multiplexed detection of various analyte classes, including nucleic acids, proteins and small molecules.Single-molecule nanopore sensing offers the ideal platform for performing this task.Nanopores have previously been shown to enable efficient detection of DNA, RNA, proteins and other molecules [31][32][33][34][35] , albeit not in a highly multiplexed configuration.Analyte detection by nanopores depends on measuring current fluctuations as charged molecules are electrophoretically driven through a nanoscale aperture.The translocation of a molecule through a nanopore causes a change in the ionic current, which is dependent on the molecule's charge, size and conformation 36 .However, the method generally lacks selectivity.Strategies to address these limitations include chemical modifications of the pore [37][38][39][40] , the use of molecular carriers 41 and the use of electro-optical methods 42,43 .These methods are excellent for the detection of molecules at very low concentrations without amplification; however, their throughput and multiplexing ability are limited.

5' --3' 2C3
Motor protein Motor protein 100 ms 0.8 nA Fig. 1 | Demultiplexing of 40 miRNAs, proteins and small molecules using barcoded probes and nanopore sequencing.a, Workflow for detection of miRNAs, proteins and small molecules.Barcoded probes (i) were incubated with synthetic targets or serum samples from healthy participants (ii).Nanopore sequencing was performed to classify the barcode, and the presence of target analytes (iii) was determined.b, Barcoded probe design.The probe consists of a Y-adapter containing the motor protein; the barcode region, containing 35 bases; and the target binding region at the 5′ end of the probe, which includes either a complementary sequence to a miRNA or an aptamer designed to specifically bind a protein or small molecule.c, Characteristic current traces.Example events are highlighted for barcoded probe only and barcoded probe events with bound target analytes.d, Base calling and analysis of events from barcoded probes result in current traces without delay (i).Base calling and analysis of events from barcoded probes with analyte bound results in current traces with a delay (ii).patient analysis, which could further reduce the cost per test.Due to the simplicity of the experimental protocol, the portability of the platform and the rapid turnaround time for experiments, we are confident that this approach could have an extensive impact on current diagnostics.

Strategy for the multiplexed detection of analytes
A highly multiplexed detection strategy was achieved by combining nanopore sequencing with barcoded molecular probes that selectively bind to target analytes (Fig. 1a and Supplementary Table 1).Barcoded probes were incubated with target analytes (miRNAs, proteins, small molecules), sequenced with the MinION device (ONT) and, subsequently, the presence of target analytes was determined.The barcoded probes consist of three key regions: (1) adapter; (2) barcode; and (3) target binding region (Fig. 1b).The adapter is identical for all probes.The barcode sequence acts as a unique identifier and can have many permutations, with theoretically up to 1.18 × 10 21 distinctive arrangements of the nucleotides for the barcode length used in these experiments.The target binding region can either be a complementary sequence (to bind miRNA or DNA) or an aptamer (to bind proteins and small molecules).The translocation of hybridized probes (target is bound) is slowed since the nanopore geometry does not allow the passage of a double-strand or protein/small molecule to feed through the nanopore (Fig. 1c).The slowed translocation can be observed in the event current signal as a 'delay' period.Events can then be subclassified with or without delay (Fig. 1d), revealing at the single-molecule level the presence of an analyte (Supplementary Fig. 1), whilst the sequenced barcode classifies the analyte being targeted.
To establish the multiplexed platform, we designed 40 unique barcoded probes to detect miRNAs implicated in cardiac disease.The accuracy of barcode base calling and classification was determined in experiments without target analytes (Fig. 2).Events were sequenced and aligned against a barcode library with known barcode sequences (Supplementary Table 1).For example, the alignment score of a unique barcode, 'barcode 38', is shown in Fig. 2a.The normalized alignment score of events for the true sequence was significantly higher compared to all other barcode sequences in the library (Fig. 2a, Supplementary Fig. 2a and Supplementary Table 2).However, alignment scoring was not suitable in multiplexed experiments due to the occurrence of false positives.To address this, a series of thresholds were tested, including: (1) the number of mismatches; (2) x mismatches in the first y bases; (3) number of aligned bases; and (4) sequence beginning with the bases 'GGG' (Fig. 2b).It was found that reducing the total number of mismatched bases worked poorly on its own (area under the curve (AUC) = 0.414).We observed that mismatches at the start of the sequenced event were particularly indicative of poorly resolved events.Consequently, a threshold based on allowing x mismatches in the first y bases was found to be moderately effective in removing unwanted events (AUC = 0.735).Increasing the total number of aligned bases was the second-best applied threshold (AUC = 0.828).Requiring the sequenced events to start with G residues (as this is common among all barcoded probes) proved to be the most efficient at separating true and false events (AUC = 0.846).Finally, we combined each criterion to determine the optimum threshold: ≤5 mismatched bases total; 1 mismatch in the first 10 bases; ≥15 bases aligned in total; sequence starts with 'GGG' (AUC = 0.805).This configuration did not improve sensitivity but markedly reduced the false-positive rate (FPR).The optimized thresholds resulted in an accuracy of >95% in the alignment of base-called events for single barcoded probe experiments (Fig. 2c).
To determine the sensitivity of the platform, we tested a barcode against two further probes with one and two mismatched bases.In a multiplexed experiment with all three of the barcoded probes present, the alignment score for the correct barcode was significantly higher than the alignment scores of the other sequences (Fig. 2d).

Hybridization of barcoded probes with miRNA
Many miRNAs share similar sequences 47 .To demonstrate sequence-specific miRNA detection, a mixture of 40 barcoded probes (30 nM) was incubated with each miRNA (10 nM) individually (Fig. 3a).It was found that the percentage delayed events of each barcoded probe increased significantly (P < 0.01) when in the presence of its corresponding target miRNA.We assessed the increase in percentage delayed events by comparing the miRNA sequence homologies (Supplementary Fig. 3a).Using a threshold of ≥90% miRNA sequence similarity (Supplementary Fig. 4a) and a significant increase in percentage delay (P ≤ 0.01), we found that 2.65% of all classifications were true positive, 0.95% false positive, 0% false negative and 96.40% true negative (Fig. 3b), resulting in a platform accuracy of 99.05%, specificity of 99.02% and sensitivity of 100%.Importantly, we observed that the barcode sequence does not influence the percentage delayed events detected for our probes (Supplementary Fig. 4b).In single barcoded probe experiments (for example, barcoded probe 38 and miR-221-5p), the total event time increased when 50 nM of miRNA was present compared to the control (Fig. 3c and Supplementary Fig. 2b).However, it is known that the DNA-controlling motor enzyme does not move the probe through the pore at a consistent rate; therefore, the speed at which the barcoded probes are translocated varies 48 .Hence, rather than relying on the translocation time, the moving standard deviation of the electrical signal was used to identify signal delays.Using this method, it was possible to determine a concentration-percentage delay curve for each barcoded probe and miRNA combination (R 2 ≥ 0.989) (Fig. 3d and Supplementary Fig. 5).

Multiplexed detection of miRNA, protein and small molecules
Barcoded probes were incubated with synthetic miRNAs to identify whether event delays could be observed in multiplexed conditions   https://doi.org/10.1038/s41565-023-01479-z(Fig. 4a).It was found that the mean percentage delayed events increased when miRNAs were present (0 nM versus 50 nM, 12.32 ± 2.07% versus 46.99 ± 12.95%, two-tailed t-test, ****P = 3.577 × 10 −4 , n = 5) (Fig. 4b(i)).Without demultiplexing, it was found that there was a linear increase in percentage delayed events between 0.1 nM and 10 nM (R 2 ≥ 0.988), followed by a plateau (Fig. 4b(ii)).The signal was demultiplexed using our alignment protocol, identifying individual populations of each barcoded probe.We then constructed bespoke concentration-percentage delay curves for each probe (Fig. 4b(iii)), and data points for each curve were normalized with background subtraction.We next tested our ability to quantify multiple miRNAs using a single-blinded test.A total of 40 barcoded probes were incubated with a mix of 40 miRNAs at various concentrations, ranging from 0.25 nM to 20 nM (Fig. 4c).The event files were sent to a blinded researcher who determined the percentage delay of each barcoded probe and interpolated the value with the standard curves to determine the concentration of the miR-NAs.After estimating each miRNA concentration, the experiment was unblinded and compared to the actual concentration (Fig. 4c(i) and Supplementary Fig. 6).Comparison of each miRNA prediction to its true value can be made by comparing the colour changes vertically.Similar colours mean a close prediction of miRNA concentration to the true value.The ratio of predicted concentration:actual concentration was plotted, allowing comparison across all 40 barcoded probes; values >1 were overpredictions and <1 were underpredicted (Fig. 4c(ii)).The accuracy of all predictions was within one order of magnitude of the true concentration.
This platform, due to its adaptability, has the potential to detect proteins, miRNAs and small molecules in a single experiment.Barcoded probes were designed with aptamers specific for each protein and small molecule (Supplementary Table 3; aptamer sequences were taken from published literature).In single barcoded probe experiments, a significant increase in translocation time could be observed when in the presence of its corresponding protein in pathophysiological concentrations: thrombin (Fig. 5a, * This corresponded with a significant increase in the percentage delayed events (Fig. 5a-d), suggesting that the detection of proteins is possible with this method.Moreover, a significant increase in translocation time and percentage delayed events was observed when serotonin (150 nM) was incubated with its corresponding barcoded probe (Fig. 5e, two-tailed t-test, **P = 0.005), despite our delay algorithm identifying a high number of delayed events in the control condition.Multiplexed experiments were conducted by observing analytes of different molecular species.First, a duplex experiment of a small molecule and a miRNA was conducted.Incubation of barcoded probes with 50 nM small molecule and ≥3 nM miRNA significantly increased the percentage delayed events for each probe (Fig. 5f).We also performed a triplex experiment observing a miRNA, small molecule and protein simultaneously.The percentage delayed events for each barcoded probe increased significantly from the controls when miRNA (≥10 nM), small molecule (≥150 nM) and protein (≥50 nM) were present (Fig. 5g).

Detection of miRNAs in human serum
Blood serum from eight healthy participants was tested for the presence of 40 miRNAs and compared to the negative control (no serum, only buffer).To reduce pore blockage (Supplementary Fig. 7), serum was centrifuged through a 10 kDa molecular weight cut-off spin filter to remove large proteins.The list of miRNAs selected for detection in this assay have all been previously associated with cardiovascular disease (Supplementary Table 1).The mean percentage delay of all miRNAs (not demultiplexed) was increased in each of the eight participant samples in three independent experiments; however, no significance was determined (Fig. 6a).Interestingly, when the signal was demultiplexed, significant changes in percentage delay of miRNAs could be observed (Fig. 6b).Of the 40 miRNAs tested, we saw a significant increase in 24 miRNAs across all eight participants (Fig. 6b,c and Supplementary Fig. 8).The most frequently observed miRNAs across all samples were: miR-1233-5p, miR146a-5p, miR211-5p, miR30c-5p, miR18a-5p, miR-126-5p and miR193b-3p.To further verify our results, we compared percentage delayed events with RT-qPCR of miR-29a in four samples.We found that our nanopore-based platform agreed with the RT-qPCR assay (Supplementary Fig. 9).

Conclusions
There is a genuine clinical need to develop platforms that can rapidly detect multiple biomarkers in patient samples.By expanding the panel of molecules observed, it is possible that disease subpopulations could be identified and treatment regimens optimized.Detection of biomarkers of different molecular species within the same sample could also massively reduce the analysis pipeline.
Our data show that barcoded probes, in combination with the ONT sequencing platform, offer a highly accurate and sensitive analyte sensing platform with single-molecule resolution.We detected 40 different barcoded probes in the same sample and constructed independent concentration-delay curves.We then used these standard curves to predict miRNA concentrations in a single-blinded experiment.The multiplexed method was applied to multiple molecular species, allowing https://doi.org/10.1038/s41565-023-01479-zus to observe an increase in percentage delayed events due to proteins and small molecules (as well as miRNAs).Finally, we showed that the platform is compatible with human serum, indicating its potential as a biomarker detection platform.
The assay time can be less than 60 minutes in the platform's current configuration.This is a clear improvement compared to other technologies, which take several hours or even days to complete 49 .ONT sequencing software can now spot specific barcode sequences, allowing for live demultiplexing, which will further reduce assay time and provide a critical advantage in the detection of analytes with short half-life.Moreover, since the MinION device is highly portable, assays could be performed away from the clinic.The speed and portability of the platform offer further advantages over other biomarker detection strategies.
The platform can correctly distinguish between barcodes with a single nucleotide change in the sequence.Interaction of probes was predicted (Supplementary Fig. 3); however, there was no clear relationship between dimer formation and event capture rate.This indicates that any future applications of this technology will not be limited by barcode sequences, but perhaps by the selectivity of the binding regions.Our data suggest that miRNA sequences with a similarity ≥90% have the potential to hybridize with the incorrect probe (Supplementary Fig. 10); strategies to distinguish between such miR-NAs must therefore be further developed.There is also potential for the large number of biomarkers assessed to result in 'data blindness', to address this a graphical interface or risk-scoring method must be developed to assist the end user.
Currently, the detection limit of the platform for miRNA is approximately 50 pM.There is potential to push the technology further by reducing the false-negative rate of event alignment.At the moment, there are a large number of events that are rejected due to https://doi.org/10.1038/s41565-023-01479-z the thresholds applied to ensure high accuracy.One way to reduce this event loss and increase events available for delay analysis would be to switch to a different nanopore model.A particular issue of assaying blood serum is that nanopore lumen are small and are easily blocked by proteins.Pore blockage reduces the barcoded probe capture rate, meaning assay times must be increased or repeated.To reduce the blockage of nanopores by proteins in this study, we added a filtration step, which removed the largest serum proteins (>10 kDa) but prevented simultaneous miRNA and protein detection in serum.In future applications, it would be preferable to develop our methodology further to reduce the sample processing required before detection whilst also reducing the unwanted pore blocking.Limiting proteolysis whilst preventing miRNA degradation is a key hurdle to overcome in the development of an assay that can detect both proteins and miRNAs in serum simultaneously.
The platform shows great potential for use in clinical environments, for example, to offer expansive, longitudinal disease tracking or early disease detection.The assay could be adapted for use with a variety of complex fluids, for example, saliva, urine and cerebrospinal fluid.With further optimization, this strategy could significantly reduce testing time, assay cost and sample volume, whilst increasing the data available to the clinician.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41565-023-01479-z.
Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material.If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.To view a copy of this license, visit http://creativecommons. org/licenses/by/4.0/.© The Author(s) 2023 https://doi.org/10.1038/s41565-023-01479-z The experiments performed to generate the data in Fig. 4c were single-blinded (the analyst was blinded to sample contents).The experiments performed to generate the data in Fig. 6 and Supplementary Figs. 8 and 9 were fully blinded to all researchers.

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

Fig. 3 |
Fig. 3 | Sensitivity and selectivity of single miRNA detection.a, Heat map showing significance for selective miRNA binding.The 40 barcoded probes (30 nM) were incubated with each miRNA (10 nM), respectively.A significant increase (two-tailed t-test, **P ≤ 0.01) in percentage delayed events was observed for all barcoded probes when in the presence of their miRNA target (n = 3, n total events = 712,400).b, Confusion matrix showing true positive (TP), false positive (FP), false negative (FN) and true negative (TN) occurrence.The accuracy was 99.05%, specificity was 99.02% and sensitivity was 100%.c, Translocation time of single barcoded probe and single miRNA experiments (n = 5, n total events = 54,197).d, Concentration-percentage delay relationship of single miRNA and single barcoded probe experiment (barcoded probe 38, miR-221-5p).Data were fitted using the Hill equation with n H = 1 (dissociation constant (K d )= 1.77 nM and V max = 61.03%,R 2 = 96.25,n = 5, n total events = 54,197).Data presented as mean ± s.d.All experiments were performed in sequencing buffer.

Fig. 6 |
Fig. 6 | Multiplexed detection of miRNAs in human serum.a, Mean percentage delay of barcoded probes (not demultiplexed).Serum, after centrifugation with 10 kDa molecular weight cut-off spin filter, was incubated with 40 barcoded probes (30 nM) and compared to zero serum control (one-tailed t-test P > 0.05, n = 3, n total events = 194,115).b, Heat map of P values comparing percentage delayed events ±serum (one-tailed t-test, n = 3, n total events = 194,115).c, Significantly upregulated miRNAs per patient (one-tailed t-test *P ≤ 0.05, **P ≤ 0.01, n = 3, n total events = 194,115).Exact P values are available from the source data file for this figure.Data are presented as mean ± s.d.All experiments were performed in sequencing buffer.