key: cord-0295962-dlxctyxc
authors: Marcozzi, Alessio; Jager, Myrthe; Elferink, Martin; Straver, Roy; van Ginkel, Joost H.; Peltenburg, Boris; Chen, Li-Ting; Renkens, Ivo; van Kuik, Joyce; Terhaard, Chris; de Bree, Remco; Devriese, Lot A.; Willems, Stefan M.; Kloosterman, Wigard P.; de Ridder, Jeroen
title: Accurate detection of circulating tumor DNA using nanopore consensus sequencing
date: 2020-07-15
journal: bioRxiv
DOI: 10.1101/2020.07.14.202010
sha: bb65464b902f4b80dcccd76ebd70c37ee9f5a594
doc_id: 295962
cord_uid: dlxctyxc

Levels of circulating tumor DNA (ctDNA) in liquid biopsies may serve as a sensitive biomarker for real-time, minimally-invasive tumor diagnostics and monitoring. However, detecting ctDNA is challenging, as much fewer than 5% of the cell-free DNA in the blood typically originates from the tumor. To detect lowly abundant ctDNA molecules based on somatic variants, extremely sensitive sequencing methods are required. Here, we describe a new technique, CyclomicsSeq, which is based on Oxford Nanopore sequencing of concatenated copies of a single DNA molecule. Consensus calling of the DNA copies increased the base-calling accuracy ∼60x, enabling accurate detection of TP53 mutations at frequencies down to 0.02%. We demonstrate that a TP53-specific CyclomicsSeq assay can be successfully used to monitor tumor burden during treatment for head-and-neck cancer patients. CyclomicsSeq can be applied to any genomic locus and offers an accurate diagnostic liquid biopsy approach that can be implemented in point-of-care clinical workflows.

Solid tumors constantly shed small DNA molecules into the bloodstream, which are cleared within a few hours 1, 2 . Determining the circulating tumor DNA (ctDNA) content in the blood of cancer patients offers a unique opportunity for real-time detection and monitoring of solid tumors 3, 4 , as levels of these ctDNA molecules are associated with tumor presence, tumor type, tumor size, tumor stage, prognosis, response to therapy, and recurrent disease 2,5-9 . Furthermore, obtaining blood from a patient is minimally-invasive and therefore, in contrast to biopsies of solid tumors, more suited to generate serial measurements of the tumor within the same patient. Moreover, tumor locations (primary tumors or metastases) are not always easily accessible for taking biopsies and complications can occur. In this context, it has been shown that ctDNA detection in blood and other fluids ("liquid biopsies") is complementary to solid biopsies for detection of targets for precision medicine 10 .

The presence of somatic mutations in cell-free DNA (cfDNA) molecules is commonly used to approximate ctDNA content 8, 11 . However, detection of ctDNA is challenging, since noncancerous cells also shed cfDNA into the blood. The fraction of tumor-derived molecules in the blood is typically much lower than 5% and fractions as low as 0.1% have been observed 5, 12, 13 .

Therefore, a diagnostic ctDNA assay must be fast and cheap as well as highly sensitive. ctDNA can be detected with good sensitivity by digital droplet PCR (ddPCR), but this technique requires quite some time since it can typically only interrogate a single locus per assay and variants must be known a priori 2, 14, 15 . Alternatively, next-generation sequencing (NGS) approaches are used, but these suffer from a lower sensitivity and require highly optimized lab workflows to become cost-effective 9, 16 .

Oxford Nanopore Technology (ONT) recently emerged as a powerful sequencing platform that offers advantages in terms of speed (real-time sequencing), cost-efficiency (low capital investment), and flexibility (distributed sequencing instead of centralized sequencing) 17 .

ONT sequencing could, therefore, be very relevant for rapid and point-of-care clinical liquid biopsy testing. There are, however, two important limitations for ONT sequencing that hamper its use in a clinical setting. Firstly, current protocols are optimized for long DNA molecules. The shortest fragment sequenced on this platform to date is~425 bp, which is much longer than the average 145bp ctDNA 18, 19 . Secondly, the basal error rate is~5-10%, which is too high to reliably detect ctDNA 20, 21 . Several studies have shown that reading the same molecule multiple times can reduce the sequencing error rate [22] [23] [24] [25] . However, some of these methods can only detect ctDNA fractions of >10% 25 , while others rely on self-circularization which is not possible for short ctDNA molecules 26 .

Here, we present a new technique, called CyclomicsSeq, that utilizes circularization and concatemerization of short DNA molecules and an optimized DNA backbone sequence in combination with ONT sequencing. As proof of concept, we developed a TP53 -specific CyclomicsSeq protocol and a dedicated software pipeline to determine the mutation burden of a series of cfDNA samples obtained from liquid biopsies from patients with Human Papilloma Virus (HPV) negative head-and-neck squamous cell carcinoma (HNSCC). TP53 is the most commonly mutated tumor suppressor gene in human cancer and therefore serves as a widely applicable target for cancer monitoring based on liquid biopsies 27, 28 . There are relatively few hotspot mutations 29 , making this gene especially suitable for NGS-based approaches. The application to HPV-negative HNSCC is motivated by the fact that five-year survival rates are relatively low and substantial treatment benefits may be obtained by early diagnosis of recurrent disease and/or treatment response [30] [31] [32] . Moreover, differentiation between residual or recurrent tumor and radiation effects is often difficult during response evaluation or in case of suspicion of recurrency, even using modern imaging techniques. Approximately 90% of the HPV-negative HNSCC patients have a somatic mutation in TP53 33 . These TP53 mutations occur early in the tumorigenesis of HNSCC and as such are present in (virtually) all tumor cells including subclones that metastasize 31, 34 . For this reason, the detection of mutated TP53 ctDNA molecules in liquid biopsies is suggested to be an ideal biomarker for HNSCC 14, 35 .

We demonstrate that CyclomicsSeq leads to highly accurate consensus sequences, suitable for mutation detection at single-molecule resolution. Longitudinal liquid biopsy testing using CyclomicsSeq correctly identifies the presence and absence of ctDNA content, which could be informative for the management of HNSCC patients. CyclomicsSeq can be applied to a single or multiple genomic regions of choice, in principle, thereby representing a new liquid biopsy test that is relevant for diagnostic monitoring of any solid tumor for which ctDNA is a suitable biomarker.

CyclomicsSeq is a protocol designed to produce and sequence long (>1Kb) DNA concatemers with a linear repetition of a sequence of interest called "insert", and a DNA adaptor, referred to as "backbone". The molecular protocol of CyclomicsSeq is divided into four main steps: 1) circularization of insert and backbone, 2) rolling circle amplification (RCA), 3) long-read sequencing, and 4) data processing ( Fig. 1a-b ; Supplementary Fig. 1 ). In step 4, the long reads are split based on the backbone and insert sequences and individual copies are extracted. Based on these individual copies, a consensus sequence is constructed for the backbone and insert separately. The backbones are optimized for e.g. flexibility while retaining a short length of around~250 bp (Methods; Supplementary Fig. 2 , Supplementary Table S1). They serve as a molecular adaptor to mediate the circularization of the insert and are used to split and filter the reads during the data processing step. Backbones also includes barcodes and restriction sites utilized for quality control of the concatemers before sequencing (Methods; Supplementary Fig.  1 ). The inserts can be, in principle, any double-stranded DNA fragment. We have tested the method with inserts ranging from 90 to 700 bp. In this study, short (<200bp) PCR amplicons from the TP53 gene amplified from (cf)DNA were used.

As a proof-of-principle, we performed a CyclomicsSeq test with a TP53 insert and backbone BB24 (Methods; Sample CY_SM_PC_HN_0002_001_000; Supplementary Table   S2 -S3), and sequenced the resulting concatemeric DNA molecules on a Nanopore MinION instrument. This MinION run (Fig. 1c -e) yielded 7.2Gb of data, with reads containing concatemers of up to 250 repeats and an average number of concatemers of 24 repeats (Fig. 1c) .

The majority of the data (70%) consisted of concatemers with alternating backbone and insert sequences (Fig. 1d) . The main byproducts of the RCA reaction were backbone-only concatemers (30%) that are filtered out during data processing (Fig. 1e ).

In addition to the single amplicon used in the above pilot test, we tested whether CyclomicsSeq can be paired with amplicon panels covering multiple genomic loci and entire coding regions of genes. As an example, a multiplex PCR method was used to amplify all the TP53 exons from cfDNA (Sample CY_SM_PC_HC_0004_003; Supplementary Data,   Supplementary Table S3 ). Consensus reads spanned across all TP53 exons, with a relatively even distribution of the coverage across all exons (Fig. 2a) . Using a single MinION workflow, we obtained a coverage >1,000X for the vast majority (74%) of the exonic bases of TP53 (Fig,   2b ).

To evaluate person-, time-and sequencing-dependent variability in CyclomicsSeq results, CyclomicsSeq was performed three times by two different operators (only one of which had experience with the protocol) on two different days using the same insert and backbone, and subsequently each CyclomicsSeq product was sequenced on two separate MinION flow cells ( Supplementary Fig. 3 , Supplementary Table S2-S3). The insert used for these experiments was a mixture of four versions of a 151bp synthetic insert with 0 -4 mutations across the insert. In total, between 12,242 and 125,446 reads were obtained. The ratio of PASS and FAIL reads and the read length distribution were highly similar between runs, although there is some inter-individual difference ( Supplementary Fig. 3) . Nevertheless, the observed ratios of the four inserts were highly similar as well, indicating that CyclomicSeq provides reproducible results ( Supplementary Fig. 3 ). is circularized with an optimized DNA backbone. Rolling circle amplification generates a long DNA molecule with alternating insert and backbone sequences, which is sequenced using ONT sequencing. Consensus calling of the DNA sequence allows discrimination between mutations and sequencing artifacts. b Schematic overview of the bioinformatic pipeline. c Distribution of insert copies versus the number of reads for a representative CyclomicsSeq run ( #CY_SM_PC_HN_0002_001_000 ). d Ratio of insert versus backbone for CyclomicsSeq reads for a representative CyclomicsSeq run ( #CY_SM_PC_HN_0002_001_000 ). Each read is represented by a data point (dot). Colours, noted in the legend, represent the different categories a read can belong to. Optimal CyclomicsSeq reads result from a one to one ratio of insert and backbone copies and contain at least 10 repeats (Blue). The other categories include: reads with fewer repeats (Green), reads without a backbone (Orange), reads without the target insert (Gray). Reads with BB:I ratios between 0.35 and 3 are defined as "Good" (Cyan), while the others are classified as "Off-ratio" (Purple). e Ratio of sequencing data grouped by read type for a representative CyclomicsSeq run ( #CY_SM_PC_HN_0002_001_000 ). In this case, more than 60% of the data was used to generate consensus reads. The remnant data was discarded because it contained backbone-only sequences. 

To evaluate the effect of CyclomicsSeq consensus calling on Nanopore sequencing accuracy, we performed 19 CyclomicsSeq experiments with three different backbone sequences, one backbone for each experiment (Supplementary Data, Supplementary Table S2-S3). In total, between 0.86 and 11.9 million (mean 3.09 million) sequencing reads were obtained (Supplementary Table S3 ).

For each experiment, we determined the false positive rate of single-nucleotide errors (snFP rate) in the consensus backbone sequences as a function of the number of copies of the backbone in a read (Fig. 3a) . For reads with a single copy of the backbone, the mean snFP rate was 0.0184 (minimum and maximum values were 0.0166 -0.0210) (Fig. 3a) . Consensus calling reduced the snFP rate to 0.0038 (0.0028 -0.0057) for 5 and 0.0016 (0.001 -0.0024) for 10 repeats (Fig. 3a) .

The snFP rate did not decrease substantially after~10 repeats. Similar to false positive single-nucleotide errors, the number of short deletions decreased with an increased number of repeats. This reduction plateaus after~10 repeats ( Supplementary Fig. 4 ). This indicates that applying a threshold of at least 10 repeats for consensus calling will result in accurate mutation calls without unnecessary loss of data. Using this threshold, the mean false positive rate for single-nucleotide errors was 5.10 -4 (2.10 -4 -6.10 -4 ) in the backbone sequences.

Although 91.9% of the positions in the backbone sequences had an snFP rate below 0.001, some positions had an snFP rate exceeding 0.004 ( Supplementary Fig. 5 ). This suggests that there were non-random sequencing errors in the sequencing data that cannot be resolved by standard consensus calling. Non-random sequencing errors can depend on the sequence context and, therefore, considering only reads with a forward or a reverse orientation for some positions might reduce these non-random errors. Indeed, the snFP rate could be further improved by at least 0.1% at 11 of the 243 positions in BB24 by considering only forward or reverse reads for those positions (Fig. 3b ). This especially reduced the number of false positives at positions with a high snFP rate. The improvement was consistent between sequencing runs, confirming the non-randomness of errors at these positions ( Supplementary Fig. 5 ). After correction for forward or reverse orientation, 92.7% of the positions had a mean snFP rate < 0.001 in consensus called reads with at least 10 repeats of the insert, and 0% of the positions had an snFP rate >0.01 ( Fig.   3c -d, Supplementary Fig. 6 ). Furthermore, only 2.1% of the positions had a combined snFP and deletion rate >0.01 ( Supplementary Fig. 4 ). Using both the threshold for consensus calling (at least 10 repeats) and the forward/reverse orientation correction, the snFP rate was 3.10 -4 (minimum and maximum values were 2.10 -4 -5.10 -4 ) in the backbone sequences. CyclomicsSeq thus lowers the sequence error rate of ONT sequencing by~60x, which is a rate compatible with mutation frequencies in circulating DNA of cancer patients 5, 12, 13 .

Recently, ONT released the Flongle flow cell with R9-like pores and reusable parts 36 c Mean snFP rate across BB24 in reads with at least 10 repeats. Reference sequence is depicted below the x-axis. 

To evaluate the use of CyclomicsSeq for detection of cancer mutations in liquid biopsies from cancer patients, we focused on sequencing the TP53 gene in cfDNA. In a first experiment, we estimated the false positive rate for detection of known TP53 mutations as catalogued in the COSMIC database 38 in four sequencing runs based on a TP53 amplicon covering one, multiple or all TP53 exons, amplified from control cfDNA samples from individuals without cancer (Fig.   4 ). For all four runs, the median snFP rate was less than 6.10 -4 across the TP53 exon(s). For ~90% of the COSMIC mutations the snFP rate was lower than 1.10 -3 and between 20% and 30% of all COSMIC bases have a snFP rate lower than 1.10 -4 .

Next, we aimed to test CyclomicsSeq in a situation which mimics low ctDNA amounts in the blood. To this end, we generated a 141 bp (17:7577010-7577150 in GRCh37, covering a Fig. 9 ). In total, between 2.5 and 3.9 million sequencing reads were obtained for WT, MUT and the mixed WT/MUT sample (Supplementary Data, Supplementary   Table S3 ). Because pJET is~10x longer than the backbone, a threshold of at least 5 repeats was applied during consensus calling. Even so, only 7.8 -10.9% of these reads contained enough copies of the insert and were useful for data analysis. We found that for molecules with only one insert (i.e. without consensus calling) the snFP rate was~1.08x lower compared to inserts amplified with PCR ( Supplementary Fig. 9 ). This indicates that the PCR used to amplify insert prior to CyclomicsSeq introduces errors. In the consensus called reads (i.e. in molecules with at least 5 repeats of the insert), the snFP rate was~1.26x lower compared to inserts that underwent a PCR step (Fig. 5a) . A PCR-free approach can thus improve the results obtained by CyclomicsSeq even further, at the cost of sequencing depth, simplicity of the protocol and sample processing time.

In the MUT sample, 0.018%, 0.17%, and 0.013% of the reads contained a false positive WT call at the three assessed positions, respectively (Fig. 5b) . Furthermore, 99.9%, 99.5% and 99.9% of the reads contained true positive mutation calls in the MUT sample, at the three assessed positions (Fig. 5b) . The three synthetic mutations were observed in less than 0.004% of the reads in the WT sample (Fig. 5b) . In the mixed WT/MUT, the observed ctDNA fraction was notably higher than in the WT sample for all three positions (Fig. 5c) . These experiments confirm that CyclomicsSeq can be used to accurately detect low amounts of mutated ctDNA in the blood. Blue is CY_PJET_12WT_0001_000, yellow is CY_SS_PC_HC_0001_001_000, green is CY_SM_PC_HC_0002_001_000, and red is CY_SM_PC_HC_0004_001_000. 

To confirm whether CyclomicsSeq can be used to detect mutated ctDNA in the blood of patients, we focused on HPV-negative HNSCC patients, because 90% of these tumors contain TP53 mutations 33 . We isolated cfDNA from the blood of three advanced stage HPV-negative HNSCC patients (denoted as patient A, B and C) before, during and after treatment (2 -6 time points per patient) and performed CyclomicsSeq on each sample (Fig. 6a-c) . Each patient's HNSCC tumor contained a known TP53 mutation, as determined by sequencing of tumor tissue (Supplementary Table S2 ). All three patients received daily radiotherapy treatment for five to seven weeks. In detected 0.5% ctDNA before treatment (Fig. 6a,d) . After an initial increase, the amount of mutated ctDNA dropped but never reached 0% in the CyclomicsSeq measurements. CyclomicsSeq and ddPCR, both assays detect 0% ctDNA three weeks after treatment initiation, in line with the observed recurrence-free survival (Fig. 6c,f) . Although the snFP rate is compatible with the detection of ctDNA in the majority of cancer patients, some genomic positions still suffer from a relatively higher snFP rate. In addition, 0.25% of the bases have a high deletion rate (>10%, Supplementary Fig. 4 ) that will decrease sensitivity to detect single nucleotide variants at those positions. We aim to lower the error rate further by e.g. removing the PCR step from the protocol. Furthermore, implementation of forward/reverse correction for deletions will likely reduce these mutation rates as well.

Finally, the implementation of the 4-nucleotide interspersed barcode sequence in the PCR can aid in deduplication of PCR-amplified molecules and removal of chimeric reads 39 .

We demonstrate CyclomicsSeq using the ONT sequencing platform in the current study, Informed consent was provided by all participants.

Plasma was isolated within a few hours after blood collection. First, the blood was centrifuged at 800g for 10 minutes. The upper layer was subsequently centrifuged at >14,000g (non-barcode) bases guarantees that their signal is well isolated and easily discerned.

The backbones were generated by annealing and fill-in of two semi-complementary synthetic phosphorylated oligos purchased from Integrated DNA Technologies (https://www.idtdna.com).

A polymerase with error-correction activity was used for the fill-in reaction in order to obtain blunt-end products, with phosphorylated ends. The fill-in reaction consisted of a 25 µl Phusion

High-Fidelity Master Mix 2X (New England Biolabs), 23 µl of water and 1ul of each oligo at a concentration of 10 µM. The reaction was subjected to 5 cycles of DNA melting (1 minute at 98°C), annealing (30 seconds at 65°C) and elongation (15 seconds at 72°C). All the backbones were gel-purified.

Per sample, 2 to 10 ng of cfDNA was used for PCR-based enrichment of TP53 sequences. Table) of 30 seconds at 98°C and 15 seconds at 59°C, and finally 2 minutes incubation at 72°C. PCR products were gel-purified using the Wizard SV Gel and PCR Clean-Up System (Promega) according to the manufacturer's protocol.

PCR products were kept at -20°C. Sample CY_SM_PC_HC_0004_003 and CY_SM_PC_HC_0004_004 were amplified using the CleanPlex TP53 Panel of Paragon Genomics according to manufacturer's protocol.

The reaction mix for circularization of the backbone and insert (3:1 

Circular DNA obtained by the circularization reaction was combined with 12 μl 5X Annealing buffer (50 mM Tris @ pH 7.5-8.0, 250 mM NaCl, 5 mM EDTA) and 1 μl Exo-resistant random primers (Thermofisher), heated for 5 minutes at 98°C and then cooled down at room temperature. Subsequently, the RCA mix (previous reaction mixture, 10 μl 10X Phi29 Buffer (Thermofisher), 2 μl BSA (New England Biolabs), 10 μl dNTPs (Thermofisher), 4 μl pyrophosphatase (Thermofisher), 2 μl Phi29 Polymerase (Thermofisher), and MQ (to a volume of 100 μl)) was prepared. RCA was performed overnight at 30°C. The RCA-reaction was inactivated by 10 minute incubation at 70°C.

To test whether CyclomicsSeq worked, 4 μl of RCA mixture was incubated with a restriction enzyme that specifically cuts backbone-backbone interactions, but not backbone-insert interactions. Briefly, 4 μl of RCA mixture was combined with 4 μl Restriction enzyme buffer (New England Biolabs), 13 μl MilliQ, and 1 μl BglII (New England Biolabs). The reaction mixture was incubated for 1 hour at 37°C and then ran on a 1.5% Agarose gel.

Synthetic sense and antisense oligos were purchased from Integrated DNA Technologies in order to produce the following two dsDNA strands, encoding for a single exon of the TP53 gene. ONT sequencing RCA products are purified using AMPure beads. Subsequently, branched DNA (which can be a consequence of the RCA) was resolved by a 1 hour incubation at 37°C with 4 μl T7 endonuclease (New England Biolabs) and re-purified using AMPure beads. ONT libraries were prepared according to manufacturer's protocol version SQK-LSK109 using 1500ng as input DNA, extending the DNA repair step to 50 minutes and the adapter ligation to 30 min.

Sequencing data was processed using the cyclomics consensus pipeline available in our GitHub 

The file structure.txt, generated by the pipeline, is parsed to determine the read length distribution and the ratio between backbone and insert for each read of a run. These features are then used to group the reads into the categories found in Fig. 1c -e and in the SupplementaryData.

The code used is available in the jupyter notebook Stats_from_structure.ipynb , available in the GitHub repository ( https://github.com/UMCUGenetics/CyclomicsManuscript ).

The coverage of consensus reads on TP53 is computed using samtools depth, without coverage limit (option -d=0), on the bam files generated by the pipeline, suffixed with "full_consensus.sorted.bam" . The resulting table was used to generate the plot of Figure 2 using the jupyter notebook TP53_panel_coverage.ipynb , available in the shared GitHub repository.

To determine the false positive rate specifically for COSMIC mutations (Figure 4 ), the number of the consensus bases called at each COSMIC position was counted. The false-positive rate, for each position, was calculated as the percentage of COSMIC mutation over the total coverage.

For the position for which there exists a bias in the sequencing accuracy between the forward and the reverse strand, the consensus was computed separately, and the base counts and coverage from the most accurate strand were used. The code used is available in the jupyter notebook

COSMIC_analysis.ipynb , available in the shared GitHub repository. For each base position in the 10+ consensus called files, we next determined whether taking only forward or reverse reads would reduce the mean snFP rate by more than 1/1000. If so, only forward or reverse measurements were considered for these positions specifically. Mean snFP and deletion rates and standard deviations were calculated per base position across the samples. Furthermore, the mean number of bases per sample with error rates <0.1%, 0.1-1% and >1% were computed.

For this analysis all samples with BB24 that were sequenced with the R9 flow cell were included. Four samples were chosen at random as 'training' samples and the remaining four samples were the 'testing' samples. First, the bases that need forward/reverse correction were defined using the 'training' samples only, similar as described previously. Subsequently, the 'testing' samples were forward/reverse corrected. We then plotted both the uncorrected and the forward/reverse corrected snFP rate.

ddPCR was performed as described previously 14 . Briefly, a ddPCR reaction was prepared (13 μl mastermix and 9 μl cfDNA) and subsequently ran on a QX200 ddPCR system according to protocol (Bio-Rad Laboratories). Data analysis was performed using QuantaSoft v1.7.4.0917

(Bio-Rad Laboratories). Each experiment was carried out in duplicate, and mean number of positive droplets were used as a proxy for ctDNA concentrations.

The sequencing datasets generated during the current study are available at EGA 

P2018-004) and is part of the Oncode Institute, which is partly financed by the Dutch Cancer

were involved in the conceptual design of the CyclomicsSeq protocol. A.M. developed the wet-lab protocol

wrote the manuscript. All authors provided textual comments and have approved the manuscript

Rapid clearance of fetal DNA from maternal plasma

Liquid biopsies come of age: towards implementation of circulating tumour DNA

Circulating tumor DNA as a non-invasive substitute to metastasis biopsy for tumor genotyping and personalized medicine in a prospective trial across all tumor types

Analysis of circulating tumor DNA to monitor metastatic breast cancer

Detection of Circulating Tumor DNA in Early-and Late-Stage Human Malignancies

Liquid biopsy: monitoring cancer-genetics in the blood

Monitoring response to therapy in melanoma by quantifying circulating tumour DNA with droplet digital PCR for BRAF and NRAS mutations

Circulating mutant DNA to assess tumor dynamics

An ultrasensitive method for quantitating circulating tumor DNA with broad patient coverage

Clinical utility of circulating tumor DNA as a response and follow-up marker in cancer therapy

Detection of chromosomal alterations in the circulation of cancer patients with whole-genome sequencing

Direct detection of early-stage cancers using circulating tumor DNA

Genome-wide cell-free DNA fragmentation in patients with cancer

Droplet digital PCR for detection and quantification of circulating tumor DNA in plasma of head and neck cancer patients

High-Throughput Droplet Digital PCR System for Absolute Quantitation of DNA Copy Number

Integrated digital error suppression for improved detection of circulating tumor DNA

Three decades of nanopore sequencing

Rapid Multiplex Small DNA Sequencing on the MinION Nanopore Sequencing Platform

Enhanced detection of circulating tumor DNA by fragment size analysis

Evaluation of Oxford Nanopore's MinION Sequencing Device for Microbial Whole

Playing a long game

INC-Seq: accurate single molecule reads using nanopore sequencing

Improving nanopore read accuracy with the R2C2 method enables the sequencing of highly multiplexed full-length single-cell cDNA

Mutational and fitness landscapes of an RNA virus revealed through population sequencing

High-Fidelity Nanopore Sequencing of Ultra-Short DNA Targets

DNA flexibility studied by covalent closure of short fragments into circles

TP53 mutations in human cancers: origins, consequences, and clinical use

Role of p53 in Cell Death and Human Cancers

Why are there hotspot mutations in the TP53 gene in human cancers?

Cancer survival in Europe 1999-2007 by country and age: results of EUROCARE--5-a population-based study

The molecular biology of head and neck cancer

The molecular landscape of head and neck cancer

Comprehensive genomic characterization of head and neck squamous cell carcinomas

Targeted sequencing reveals TP53 as a potential diagnostic biomarker in the post-treatment surveillance of head and neck cancer

Circulating tumor DNA detection in head and neck cancer: evaluation of two different detection approaches

New 'R10' nanopore released into early access

COSMIC: the Catalogue Of Somatic Mutations In Cancer

Enabling high-accuracy long-read amplicon sequences using unique molecular identifiers with Nanopore or PacBio sequencing

Adaptive seeds tame genomic sequence comparison

Split-alignment of genomes finds orthologies more

The authors would like to thank Manon Huibers for input on the study, Floris Reinders for input on the manuscript and the Utrecht Sequencing Facility for sequencing. The Utrecht Sequencing Facility is subsidized by the University Medical Center Utrecht, Hubrecht Institute, and UtrechtUniversity. This study was financially supported by the Oncode Institute (project number

The authors declare the following financial competing interest: A.M., R.S., W.P.K., and J.d.R filed patents and A.M., W.P.K., and J.d.R founded a company (Cyclomics) based on CyclomicsSeq.

Correspondence to Wigard P. Kloosterman ( Wigard@cyclomics.com ) and Jeroen de Ridder ( J.deridder-4@umcutrecht.nl ).