Abstract
Background
Tracing patient-specific tumor mutations in cell-free DNA (cfDNA) for minimal residual disease (MRD) detection is promising but challenging. Assaying more mutations and cfDNA stands to improve MRD detection but requires highly accurate, efficient sequencing methods and proper calibration to prevent false detection with bespoke tests.
Methods
MAESTRO (Minor Allele Enriched Sequencing Through Recognition Oligonucleotides) uses mutation-specific oligonucleotide probes to enrich cfDNA libraries for tumor mutations and enable their accurate detection with minimal sequencing. A new approach, MAESTRO-Pool, which entails pooling MAESTRO probes for all patients and applying these to all samples from all patients, was used to screen for 22 333 tumor mutations from 9 melanoma patients in 98 plasma samples. This enabled quantification of MRD detection in patient-matched samples and false detection in unmatched samples from other patients. To detect MRD, a new dynamic MRD caller was used that computes a probability for MRD detection based on the number of mutations and cfDNA molecules sequenced, thereby calibrating for variations in each bespoke test.
Results
MAESTRO-Pool enabled sensitive detection of MRD down to 0.78 parts per million (ppm), reflecting a 10- to 100-fold improvement over existing tests. Of the 8 MRD positive samples with ultra-low tumor fractions <10 ppm, 7 were either in upward-trend preceding recurrence or downward-trend aligning with response. Of 784 patient-unmatched tests, only one was found as MRD positive (tumor fraction = 2.7 ppm), suggesting high specificity.
Conclusions
MAESTRO-Pool enables massively parallel, tumor-informed MRD testing with concurrent benchmarking of bespoke MRD tests. Meanwhile, our new MRD caller enables more mutations and cfDNA molecules to be tested without compromising specificity. These improve the ability for detecting traces of MRD from blood.
Introduction
Using circulating tumor DNA (ctDNA) to detect minimal residual disease (MRD) from blood stands to inform better care for cancer patients. It could enable early intensification or switching of therapy long before clinical relapse, or de-escalation of therapy when no longer needed, thereby sparing the side effects of therapy (1–3). Multiple studies have shown that ctDNA detection is highly predictive for tumor recurrence (2, 4, 5). However, the sensitivity of detecting MRD at key clinical time points such as within a few months following surgery is low, often below 25%, especially in cancer types considered to be “low ctDNA shedders” (6–11). Given this, there is a critical unmet need to improve sensitivity to the maximum possible levels. Being able to detect MRD in more patients and with longer lead time to recurrence could enable ctDNA testing to inform therapy escalation and de-escalation for cancer patients.
Higher sensitivity for detecting MRD is attainable by tracking more tumor mutations per patient or sampling more cell-free DNA (cfDNA) molecules from blood. Given the genetic diversity of most patients’ tumors, tracking more tumor mutations requires “tumor-informed” assays (12–14): sequencing a patient’s tumor, identifying a unique tumor mutation fingerprint, and assaying for this fingerprint in each patient’s own plasma DNA. Yet, this becomes challenging due to the massive excess of normal cfDNA in blood and the extreme amounts of sequencing required to accurately resolve low-abundance mutations. Further as bespoke assays are developed to detect increasingly lower tumor fractions in cfDNA, it becomes crucial to mitigate the potential for false detection. This has been largely underexplored as bespoke assays are generally not benchmarked in many samples on an individualized basis. We reasoned that plasma samples from other patients could be used as controls for one another, coupled with improved methods for MRD detection, to characterize and safeguard against false detection.
Here we introduce MAESTRO-Pool, a modification of the recently developed MAESTRO (Minor Allele Enriched Sequencing Through Recognition Oligonucleotides) method (13) for mutation enrichment sequencing of rare mutations, which has been used to improve MRD detection from liquid biopsies (15). With MAESTRO-Pool, MAESTRO probes from a cohort of patients are synthesized and pooled and subjected to all samples from all patients, thereby enabling the detection of MRD in patient-matched samples and the quantification of false detection in patient-unmatched samples. We also introduce a new “dynamic” MRD caller that provides a probabilistic score for MRD detection, allowing more mutations and cfDNA molecules to be sequenced while limiting false detection.
Materials and Methods
Patients and samples
All patients provided written, informed consent to allow the collection of blood and tumor tissue and analysis of clinical and genetic data for research purposes. Patients with high-risk melanoma were prospectively identified for enrollment into an Institutional Review Board-approved tissue analysis and banking cohort (Dana Farber/Harvard Cancer Center [DF/HCC] protocol identifier 11–181) of which 9 were selected for analysis. DNA was extracted and made into sequencing libraries as previously described (16, 17). Further details are in the online Supplemental Methods.
Designing tumor fingerprint panels
Patient-specific tumor fingerprint panels were designed following the MAESTRO (13) or MRD Tracker (16) methods previously described. However, additional selection criteria were imposed for MAESTRO-Pool which is further described in the online Supplemental Methods.
Hybrid capture and sequencing
Hybrid capture for MAESTRO-Pool and MRD Tracker followed the previously published methods (13, 16, 17). Samples were pooled for sequencing on an Illumina NovaSeq S4 with a target raw depth of 50 million read pairs per sample. Further details are in the online Supplemental Methods.
Data processing and filtering
Consensus calling followed the same protocol previously described (13, 16). The resulting consensus data was then processed using Miredas, our suite of custom MRD calling scripts (16), which applied fragment-level and site-level filtering before quantifying the number of consensus molecules per site. Additional filtering strategies were imposed for MAESTRO-Pool samples which are further described in the online Supplemental Methods.
Dynamic MRD Calling
The dynamic MRD calling algorithm was designed to weigh sample-specific attributes, such as the observed tumor fraction, validated fingerprint size, and cfDNA mass, and quantify the probability of the observed data being the result of true tumor signal rather than false, spontaneous errors. For the detailed method, please refer to the online Supplemental Methods.
Tumor fraction and limit of detection estimations
Tumor fraction and limit of detection estimations relied on methods previously described (13, 15, 16). However, the limit of detection estimation was adapted for the dynamic MRD caller. Further details about these methods are in the online Supplemental Methods.
MAESTRO Enrichment and sequencing requirements
For calculating enrichment and sequencing requirements, we followed a similar strategy to our previous work (13). Both estimations were restricted to sites that overlapped the MAESTRO and MRD Tracker fingerprints, allowing us to make direct comparisons per site. Further details are in the online Supplemental Methods.
Results
MAESTRO-Pool and an improved MRD Caller
MAESTRO-Pool enables massively parallel, tumor-informed MRD testing for patient cohorts. It involves performing whole-genome sequencing (WGS) of each patient’s tumor, designing MAESTRO probes that target patient-specific tumor mutations, and combining these into a single assay that is applied to all samples from all patients (Fig. 1). Mutations are verified in each patient’s tumor DNA and excluded if found in their own germline DNA as previously described (13, 15, 16). MRD detection is performed for each patient’s own plasma samples (i.e., “matched samples”) as well as plasma samples from other patients (i.e., “unmatched samples”) to assess specificity of each bespoke assay. Mutations shared in tumor biopsies from multiple patients are excluded when computing the specificity of each MRD test in unmatched samples.
Fig. 1.
MAESTRO-Pool schematic. One MAESTRO-Pool assay, comprising each patient’s fingerprint, is applied to all samples from all patients. Each patient’s own (matched) plasma samples were tested for MRD while other patients’ (unmatched) plasma samples were used as controls to assess the specificity of each bespoke assay. MRD was detected using our new dynamic MRD caller which calculates the probability of MRD detection based on sample-specific attributes. Abbreviations: MAESTRO, Minor Allele Enriched Sequencing Through Recognition Oligonucleotides; MRD, minimal residual disease.
To this point, we have used a “fixed” threshold for MRD detection with MAESTRO of ≥2 mutated duplexes when assaying approximately 1000 genome-wide mutations from standard blood volumes (e.g., 1 to 3 × 10 cc tubes). This has shown high specificity in benchmarking experiments and cancer patients (13, 16), but we reasoned that it may be insufficient as we seek to track even more mutations and cfDNA molecules per patient (Fig. 2). Therefore, we developed a new “dynamic” MRD caller that computes a probability for MRD detection based on sample-specific attributes such as observed tumor fraction, fingerprint size, and cfDNA molecules, against an assumed error rate of 1/10 M (see Materials and Methods). We set a threshold of P ≥ 0.95—a tunable parameter—for a sample to be considered MRD positive. At P ≥ 0.95, our model suggests this will limit false detection while maintaining high sensitivity down to low parts per million (ppm) levels of tumor-derived cfDNA (Fig. 2). We first confirmed that the dynamic caller had negligible impact on prior results (n = 133/134 concordant MRD test results) when up to 1000 mutations per patient were assayed in up to 148 ng cfDNA with MAESTRO (online Supplemental Fig. 1). Notably, the one sample with a discordant MRD call was borderline negative with dynamic MRD calling with a probability score of 0.94.
Fig. 2.
Predicted performance of the fixed vs dynamic MRD caller when assaying 50 ng (A), 100 ng (B), 250 ng (C), 500 ng (D), and 1000 ng (E) of cfDNA with increasing numbers of mutations. The left and center plots portray the probability of false detection with fixed, mutation-based thresholding, or dynamic, probabilistic-based thresholding, respectively, assuming a fixed error rate of 1 × 10−7. The right plot compares the limit of detection of both MRD calling approaches. All plots assumed 75 duplexes per site per 1 ng cfDNA. Abbreviations: cfDNA, cell-free DNA; ppm, parts per million.
MAESTRO-Pool testing of melanoma patients
Next, we sought to push the bounds on the number of mutations assayed and used MAESTRO-Pool to experimentally validate our new MRD caller in matched vs unmatched samples. We identified 9 patients treated with curative intent for Stage III melanoma with a median of 10 (range 5 to 22) plasma samples each (98 total) collected over a median of 2.5 years (range 0.5 to 4.0) of clinical follow-up. Eight of the patients experienced recurrence, 7 of which had subsequent samples collected, while one patient remained disease-free but passed away due to unrelated circumstances. We performed WGS of tumor and normal DNA from each patient and identified a median of 40 348 mutations (range 5932 to 160 598) per patient. We created one MAESTRO-Pool assay for all patients (median 1856 mutations per patient, range 447 to 5571) and applied it to each patient’s tumor and normal DNA, as well as the 98 plasma DNA samples from all patients. For comparison, we also created 9 personalized MRD Tracker assays, optimized from our previously validated Parsons et al. method (16), comprising a median 1000 (range 334 to 1000) mutations per patient. These were capped at 1000 mutations per patient and only applied to each patient’s own samples given the substantially greater sequencing requirements without MAESTRO enrichment.
By applying MRD assays to each patient’s tumor and normal DNA, we first validated each patient’s tumor mutation fingerprint. With vs without MAESTRO enrichment, we found a median of 78% (range 3% to 89%) and 97% (range 3% to 99%) of mutations were confirmed, respectively (online Supplemental Fig. 2A). Expectedly, the specimens with the lowest validation rates had the lowest median variant allele frequency (VAF) in WGS which is likely indicative of low tumor purity. Only the mutations that validated as present in the tumor biopsy and absent from the normal DNA were used for MRD detection from plasma (see Materials and Methods). Further, we computed the estimated detection limits at 90% power for each sample based on the duplex depths and validated fingerprint sizes (see Materials and Methods). We found most MAESTRO tests to be powered to detect low ppm, except for the 3 patients with validated fingerprint sizes well below 1000 mutations (online Supplemental Fig. 2B).
Applying the dynamic MRD caller with P ≥ 0.95, we found 47 of 98 matched samples to be MRD positive (median tumor fraction 1.1 × 10−4, range 7.8 × 10−7 to 0.13) and only 1 of 784 unmatched samples to be MRD positive (tumor fraction 2.7 × 10−6, Fig. 3A and B). This resulted in median experimental specificities of 100% (range 100% to 100%) and 100% (range 98% to 100%) for detecting ≥10 ppm and <10 ppm ctDNA, respectively, in a median of 88 (range 76 to 93) unmatched patient samples. Of note, had a fixed threshold of ≥2 mutations been used, an additional 3 matched and 20 unmatched samples would have been found MRD positive with most false detection occurring in the largest panels of 4337, 3816, and 3479 mutations, as expected (Fig. 3A and B). While a stricter threshold of ≥3 or ≥4 mutations achieved specificities comparable to dynamic MRD calling, it also reduced the number of MRD positive calls in matched samples (online Supplemental Fig. 3A). These results highlight the importance of dynamic MRD calling as more tumor mutations and cfDNA molecules are sequenced per patient.
Fig. 3.
Application of MAESTRO-Pool and the dynamic MRD caller to 9 melanoma patients. (A), The MRD results from applying MAESTRO-Pool across all samples from all patients and using either fixed MRD calling with a threshold of ≥2 mutations (left) or dynamic MRD calling with a threshold of P ≥ 0.95 (right). Patients are sorted by descending validated fingerprint size while samples are sorted by descending cfDNA mass within each patient; (B), Experimental specificity was quantified for each patient’s bespoke MRD test (“patient fingerprint”) using unmatched samples and the tumor fraction range is depicted for samples called MRD positive. Again, the results are split between fixed MRD calling (left) and dynamic MRD calling (right). Abbreviations: MAESTRO, Minor Allele Enriched Sequencing Through Recognition Oligonucleotides; MRD, minimal residual disease; cfDNA, cell-free DNA; ppm, parts per million.
Interestingly, of the 3 samples that were MRD positive with fixed MRD calling but negative with dynamic calling, all had 2 or 3 mutations detected and tumor fractions below 1 ppm (range 5.7 × 10−7 to 9.9 × 10−7) suggesting that these were borderline MRD calls initially. This is further reflected in their dynamic probability scores ranging between 0.69 and 0.92. Although dynamic MRD calling successfully reduced the false detection rates, we noticed that the P ≥ 0.95 threshold was quite conservative (e.g., median specificity of 1). We thus explored tuning this parameter. We observed that the predicted specificity underestimated the measured specificity and, therefore, lower probability thresholds, such as P ≥ 0.80, still yielded higher specificity and comparable recall to fixed MRD calling (online Supplemental Fig. 3B). Additionally, the dynamic probability score enabled the comparison of MRD signal between samples while correcting for sample-specific attributes. This can be leveraged to quantify a confidence score by comparing the probability score of matched samples to the probability score distribution in unmatched samples. For example, matched samples with borderline-negative MRD signals (i.e., dynamic scores of 0.50 ≤ P < 0.95) can be assigned a confidence score by quantifying the fraction of unmatched samples with a probability score greater than or equal to its own (online Supplemental Fig. 4). This could further inform whether there may be traces of MRD in borderline-negative samples.
As further validation, we compared MAESTRO-Pool results using the dynamic caller for matched patient samples against the MRD results with MRD Tracker. For MRD Tracker, we used our standard cutoff of ≥2 mutations detected (16). We observed high concordance in MRD detection—70 of 77 samples with concordant MRD calls—and corresponding tumor fractions (r2 = 0.97) in cfDNA between MAESTRO-Pool and MRD Tracker (online Supplemental Fig. 5A). Of the 7 samples with discordant MRD calls, 5/7 of the MRD positive samples had observed tumor fractions below the detection limit of the MRD negative samples (online Supplemental Fig. 5B) and 4/7 of the samples were from patient 1406 which only had 22 and 28 tumor-validated mutations with MAESTRO-Pool and MRD Tracker, respectively. This suggested that the discordant samples were caused by one of the samples being underpowered. Reassuringly, of the 30 samples that were negative for MRD by MAESTRO, 29 were also negative for MRD with MRD Tracker.
Lastly, we explored the effect of MAESTRO enrichment on sequencing efficiency. For mutations tracked with MAESTRO-Pool and MRD Tracker, we found a median 115-fold enrichment (range 0.44 to 12 056) in VAF when using MAESTRO (Fig. 4A). Expectedly, the fold-enrichment was most substantial among samples with low tumor fractions, e.g., median 777 (range 1.3 to 8631) for samples with <100 ppm tumor DNA (Fig. 4B). Next, we examined the reads required to uncover mutated DNA duplexes with MAESTRO-Pool vs MRD Tracker on a sample-by-sample basis (Fig. 4C and D). We found a median 33-fold reduction in reads required (range 1.5 to 439) to uncover 90% of the mutated DNA duplexes in each sample using MAESTRO (Fig. 4D). Our results using MAESTRO-Pool are consistent with our prior observations that MAESTRO enables highly sensitive MRD detection using substantially less sequencing.
Fig. 4.
Efficiency of MAESTRO mutation enrichment sequencing. (A), Comparison of duplex VAF between MAESTRO and MRD Tracker for sites with ≥1 mutant duplex in both fingerprints; (B), Association between MAESTRO’s median VAF enrichment per sample and tumor fraction which establishes an upper bound for VAF enrichment. The size of the marker indicates the number of sites used to estimate the median VAF fold enrichment (i.e., number of sites with ≥1 mutant duplex in both MAESTRO and MRD Tracker fingerprints); (C), An example of downsampling which was used to quantify the read pairs required for mutant duplex saturation. For this, we restricted to sites present in both MAESTRO and MRD Tracker fingerprints; (D), Reduction in read pairs required to reach mutant duplex saturation due to MAESTRO enrichment (see Materials and Methods). Patients are sorted by descending validated MAESTRO fingerprint size. Abbreviations: MAESTRO, Minor Allele Enriched Sequencing Through Recognition Oligonucleotides; VAF, variant allele frequency; ppm, parts per million.
Case studies of individual melanoma patients
Finally, we examined how the MRD testing results related to the treatments and outcomes of individual patients. As this was a cohort study and not a clinical trial, there was wide variation in treatment and follow-up for individual patients, lending each to be examined individually (Fig. 5 and online Supplemental Figs. 6–9). In one exemplary patient, MRD was detected following surgery at 2.6 ppm immediately prior to adjuvant dabrafenib and trametinib therapy (Fig. 5A). MRD then became undetectable at 2 subsequent time points before rising to 2.1 ppm 138 days before recurrence was detected in the brain. Following definitive craniotomy and radiation treatment, MRD was undetectable at 2 time points. Then, immediately before pembrolizumab, MRD was detected at 3.1 ppm, and at 2 time points thereafter it remained at similar levels. The patient continued therapy for another 355 days until MRD levels rose to 6612 ppm and a gastric metastasis was detected 442 days after initial detection. Then, the patient was given ipilimumab and nivolumab and MRD became undetectable while imaging scans were inconclusive.
Fig. 5.
Aligning the MRD test results with treatment and response data for 2 exemplary patients. The graph displays MAESTRO’s MRD results, detection limit of leading commercially available MRD tests (100 ppm), and significant clinical developments along the top, for Pt_966 (A) and Pt_1361 (B). Time points with dynamic scores 0.5 ≤ P < 0.95 were marked as “Borderline” and explicitly annotated. Detailed clinical annotations, including status at last clinical follow-up, are shown below. Abbreviations: MRD, minimal residual disease; MAESTRO, Minor Allele Enriched Sequencing Through Recognition Oligonucleotides; ppm, parts per million; LOD, limit of detection.
In a second exemplary patient, there were no samples available shortly after surgery, but MRD was detected 48 and 84 days preceding both local and distant recurrence, respectively (Fig. 5B). MRD levels subsequently declined down to 0.8 ppm before rising again. Of note, an increase in MRD from 0.8 to 24 ppm was detected when imaging scans indicated stable disease and 35 days before scans showed progression. The patient was then given ipilimumab and nivolumab and showed a stark decline in MRD levels at least down to 0.9 ppm preceding a long and durable response which was ongoing at last follow-up. These cases exemplify the deep levels of MRD that can be detected with MAESTRO and suggest the potential for using these measurements to guide care.
Interestingly, of the 8 MRD positive samples with ultra-low tumor fractions (<10 ppm) detected among all patients in the cohort, 7 were found either preceding recurrence (in an upward trend of increasing tumor fraction) or after a decline in tumor fraction leading to a period of therapy response. Notably, 4 of these samples occurred at time points for which patients had no evidence of disease via imaging scans. These observations, together with the experimental specificities, suggest that deep detection of MRD could fill meaningful voids in our monitoring of cancer treatment response.
Discussion
Liquid biopsies hold tremendous promise for detecting MRD and informing more precise cancer care but require higher sensitivity. Assaying more mutations in more cfDNA could improve sensitivity using bespoke tests, but this creates opportunities for false detection if not properly accounted. Here, we introduced MAESTRO-Pool for massively parallel, tumor-informed MRD testing in cohort studies. By pooling multiple patients’ MRD tests together, MAESTRO-Pool assays thousands of patient-specific tumor mutations while concurrently benchmarking each patient’s bespoke MRD test using unmatched samples as controls. This has streamlined MRD testing and enabled the detection of low ppm levels of ctDNA in patients with melanoma, including when scans showed no evidence of disease. Additionally, our new dynamic MRD caller accounted for differences in the numbers of mutations and cfDNA molecules sequenced by computing a probability score for MRD in each sample. Using the dynamic MRD caller with MAESTRO-Pool, we observed median experimental specificities of 100% (range 100% to 100%) and 100% (range 98% to 100%) for detecting ≥10 ppm and <10 ppm ctDNA, respectively, when each patient’s test was benchmarked in a median of 88 unmatched samples (range 76 to 93) from other patients. We also showed the ability to identify borderline-negative patients who may benefit from follow-up testing using the dynamic caller and its probability score distribution in pooled testing.
The massive excess of normal cfDNA in blood and the large number of sequencing reads that are required to overcome sequencing errors often prohibits the routine benchmarking of individualized MRD assays in control samples. The challenge is further compounded when more mutations and more cfDNA are assayed to enhance MRD detection at key clinical time points. MAESTRO-Pool addresses these challenges by leveraging mutation enrichment to reduce the amount of sequencing required per sample and allow multiple patients’ MRD tests to be pooled and applied to many samples at once. A similar approach would be impractical using standard duplex sequencing approaches, as it would be significantly slower and financially inviable, both of which would ultimately burden healthcare systems and patients. Meanwhile, the dynamic caller further accounts for varied numbers of mutations and cfDNA molecules to limit false detection. Of note, the dynamic caller works with both MAESTRO and MAESTRO-Pool MRD testing.
Despite these technical improvements, there are limitations to this study. The patient cohort that we analyzed was relatively small, had nonuniform blood collections, and had nonuniform treatment courses which led to translational analysis on a case-by-case basis. We are in the process of applying MAESTRO-Pool and dynamic MRD calling to several additional clinical cohorts to understand their translational impact on a larger scale. Furthermore, due to the lack of commercially available tests with ppm-level sensitivity, it is yet to be determined how MRD detection in the ppm range should affect clinical decision-making. Prospective trials will be needed to address this. Additionally, tumor-informed MRD testing involves upfront costs such as the sequencing of tumor and germline DNA and the establishment of individualized assays. For MAESTRO-Pool, this entailed ≥30 × WGS of tumor DNA, ≥15 × WGS of germline DNA, and the synthesis of thousands of approximately 30nt biotinylated MAESTRO probes. Yet, the upfront costs were effectively amortized over multiple tested time points per patient. Although MAESTRO reduced the sequencing required per sample by a median of 33-fold, the number of MAESTRO assays and probes that were pooled for MAESTRO-Pool testing affected the final cost per assay.
In summary, MAESTRO-Pool and dynamic MRD calling enabled ctDNA detection down to 1 ppm while simultaneously benchmarking each patient’s bespoke MRD test using unmatched patient samples as controls. These features streamlined MRD testing and enabled higher sensitivity and specificity as more mutations and cfDNA were analyzed. In turn, these approaches for enhanced MRD detection stand to enable more precise care for cancer patients.
Supplemental Material
Supplemental material is available at Clinical Chemistry online.
Supplementary Material
Acknowledgments
We acknowledge the patients and their families for their contributions to this study. We thank the Gerstner Family Foundation for its generous support of this work.
Contributor Information
Timothy Blewett, Gerstner Center for Cancer Diagnostics, Broad Institute of MIT and Harvard, Cambridge, MA, United States.
Justin Rhoades, Gerstner Center for Cancer Diagnostics, Broad Institute of MIT and Harvard, Cambridge, MA, United States.
Ruolin Liu, Gerstner Center for Cancer Diagnostics, Broad Institute of MIT and Harvard, Cambridge, MA, United States.
Kan Xiong, Gerstner Center for Cancer Diagnostics, Broad Institute of MIT and Harvard, Cambridge, MA, United States.
Sainetra Sridhar, Gerstner Center for Cancer Diagnostics, Broad Institute of MIT and Harvard, Cambridge, MA, United States.
Andjela Crnjac, Gerstner Center for Cancer Diagnostics, Broad Institute of MIT and Harvard, Cambridge, MA, United States.
Ju Cheng, Gerstner Center for Cancer Diagnostics, Broad Institute of MIT and Harvard, Cambridge, MA, United States.
Aleigha R Lawless, Massachusetts General Hospital Cancer Center, Department of Medicine, Massachusetts General Hospital, Boston, MA, United States.
Dennie T Frederick, Gerstner Center for Cancer Diagnostics, Broad Institute of MIT and Harvard, Cambridge, MA, United States.
Keith T Flaherty, Massachusetts General Hospital Cancer Center, Department of Medicine, Massachusetts General Hospital, Boston, MA, United States.
Gerassimos Mike Makrigiorgos, Department of Radiation Oncology, Dana-Farber Cancer Institute and Brigham and Women’s Hospital, Boston, MA, United States.
Viktor A Adalsteinsson, Gerstner Center for Cancer Diagnostics, Broad Institute of MIT and Harvard, Cambridge, MA, United States.
Nonstandard Abbreviations
cfDNA, cell-free DNA; MRD, minimal residual disease; MAESTRO, Minor Allele Enriched Sequencing Through Recognition Oligonucleotides; ppm, parts per million; ctDNA, circulating-tumor DNA; WGS, whole-genome sequencing.
Author Contributions
The corresponding author takes full responsibility that all authors on this publication have met the following required criteria of eligibility for authorship: (a) significant contributions to the conception and design, acquisition of data, or analysis and interpretation of data; (b) drafting or revising the article for intellectual content; (c) final approval of the published article; and (d) agreement to be accountable for all aspects of the article thus ensuring that questions related to the accuracy or integrity of any part of the article are appropriately investigated and resolved. Nobody who qualifies for authorship has been omitted from the list.
Timothy Blewett (Data curation-Lead, Formal analysis-Lead, Investigation-Lead, Methodology-Lead, Project administration-Supporting, Software-Lead, Visualization-Lead, Writing—original draft-Lead, Writing—review & editing-Lead), Justin Rhoades (Data curation-Supporting, Formal analysis-Supporting, Methodology-Supporting, Software-Supporting, Writing—review & editing-Supporting), Ruolin Liu (Formal analysis-Supporting, Methodology-Lead, Software-Supporting, Writing—review & editing-Supporting), Kan Xiong (Methodology-Lead, Project administration-Supporting, Writing—original draft-Supporting, Writing—review & editing-Supporting), Sainetra Sridhar (Methodology-Supporting), Andjela Crnjac (Methodology-Supporting), Ju Cheng (Methodology-Supporting), Aleigha Lawless (Data curation-Supporting, Resources-Supporting, Writing—original draft-Supporting), Dennie Frederick (Data curation-Supporting, Resources-Supporting), Keith Flaherty (Conceptualization-Supporting, Investigation-Supporting, Resources-Supporting, Writing—original draft-Supporting), G. Makrigiorgos (Conceptualization-Lead, Funding acquisition-Lead, Investigation-Supporting, Resources-Supporting, Visualization-Supporting, Writing—original draft-Lead, Writing—review & editing-Lead), and Viktor A Adalsteinsson (Conceptualization-Lead, Formal analysis-Lead, Funding acquisition-Lead, Investigation-Lead, Methodology-Lead, Project administration-Lead, Resources-Lead, Supervision-Lead, Visualization-Lead, Writing—original draft-Lead, Writing—review & editing-Lead).
Authors’ Disclosures or Potential Conflicts of Interest
Upon manuscript submission, all authors completed the author disclosure form.
Research Funding
National Institutes of Health grant 2R01 CA221874 (G.M. Makrigiorgos and V.A. Adalsteinsson). Funding from Gerstner Family Foundation (T. Blewett, J. Rhoades, R. Liu, K. Xiong, S. Sridhar, A. Crnjac, J. Cheng, and V.A. Adalsteinsson).
Disclosures
A patent application has been filed on the methods described in this paper (T. Blewett, R. Liu, and V.A. Adalsteinsson). V.A. Adalsteinsson and G.M. Makrigiorgos are co-inventors on a patent application covering MAESTRO (US 2023/0203568, pending) which has been licensed to Exact Sciences. V.A. Adalsteinsson receives research funding from Exact Sciences. V.A. Adalsteinsson is a cofounder and advisor to Amplifyer Bio which was not involved in this study. K.T. Flaherty served on the Board of Directors of Clovis Oncology, Strata Oncology, Checkmate Pharmaceuticals, Kinnate Pharmaceuticals, and Scorpion Therapeutics; served on the Scientific Advisory Boards of PIC Therapeutics, Apricity, Fog Pharma, Tvardi, xCures, Monopteros, Vibliome, ALX Oncology, OMRx, Soley Therapeutics, Quanta Therapeutics, Alterome, and IntrECate; served as consultant to Lilly, Novartis, Genentech, Takeda, and Transcode Therapeutics; and received research funding from Novartis and Sanofi.
Role of Sponsor
The funding organizations played no role in the design of study, choice of enrolled patients, review and interpretation of data, preparation of manuscript, or final approval of manuscript.
Data Availability Statement
The data underlying this article are available through dbGaP with the accession number: phs003447.v1.p1.
References
- 1. Etienne G, Guilhot J, Rea D, Rigal-Huguet F, Nicolini F, Charbonnier A, et al. Long-term follow-up of the French stop imatinib (STIM1) study in patients with chronic myeloid leukemia. J Clin Oncol 2017;35:298–305. [DOI] [PubMed] [Google Scholar]
- 2. Magbanua MJM, Swigart LB, Wu H-T, Hirst GL, Yau C, Wolf DM, et al. Circulating tumor DNA in neoadjuvant-treated breast cancer reflects response and survival. Ann Oncol 2021;32:229–39. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Radovich M, Jiang G, Hancock BA, Chitambar C, Nanda R, Falkson C, et al. Association of circulating tumor DNA and circulating tumor cells after neoadjuvant chemotherapy with disease recurrence in patients with triple-negative breast cancer: preplanned secondary analysis of the BRE12-158 randomized clinical trial. JAMA Oncol 2020;6:1410–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Garcia-Murillas I, Chopra N, Comino-Méndez I, Beaney M, Tovey H, Cutts RJ, et al. Assessment of molecular relapse detection in early-stage breast cancer. JAMA Oncol 2019;5:1473–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Lipsyc-Sharf M, de Bruin EC, Santos K, McEwen R, Stetson D, Patel A, et al. Circulating tumor DNA and late recurrence in high-risk hormone receptor-positive, human epidermal growth factor receptor 2-negative breast cancer. J Clin Oncol 2022;40:2408–19. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Azad TD, Chaudhuri AA, Fang P, Qiao Y, Esfahani MS, Chabon JJ, et al. Circulating tumor DNA analysis for detection of minimal residual disease after chemoradiotherapy for localized esophageal cancer. Gastroenterology 2020;158:494–505.e6.. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Bettegowda C, Sausen M, Leary RJ, Kinde I, Wang Y, Agrawal N, et al. Detection of circulating tumor DNA in early- and late-stage human malignancies. Sci Transl Med 2014;6:224ra24. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Cohen JD, Li L, Wang Y, Thoburn C, Afsari B, Danilova L, et al. Detection and localization of surgically resectable cancers with a multi-analyte blood test. Science 2018;359:926–30. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Corrò C, Hejhal T, Poyet C, Sulser T, Hermanns T, Winder T, et al. Detecting circulating tumor DNA in renal cancer: an open challenge. Exp Mol Pathol 2017;102:255–61. [DOI] [PubMed] [Google Scholar]
- 10. Kim Y-W, Kim Y-H, Song Y, Kim H-S, Sim HW, Poojan S, et al. Monitoring circulating tumor DNA by analyzing personalized cancer-specific rearrangements to detect recurrence in gastric cancer. Exp Mol Med 2019;51:1–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Yamamoto Y, Uemura M, Fujita M, Maejima K, Koh Y, Matsushita M, et al. Clinical significance of the mutational landscape and fragmentation of circulating tumor DNA in renal cell carcinoma. Cancer Sci 2019;110:617–28. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Coombes RC, Page K, Salari R, Hastings RK, Armstrong A, Ahmed S, et al. Personalized detection of circulating tumor DNA antedates breast cancer metastatic recurrence. Clin Cancer Res 2019;25:4255–63. [DOI] [PubMed] [Google Scholar]
- 13. Gydush G, Nguyen E, Bae JH, Blewett T, Rhoades J, Reed SC, et al. Massively parallel enrichment of low-frequency alleles enables duplex sequencing at low depth. Nat Biomed Eng 2022;6:257–66. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Wan JCM, Heider K, Gale D, Murphy S, Fisher E, Mouliere F, et al. ctDNA monitoring using patient-specific sequencing and integration of variant reads. Sci Transl Med [Internet] 2020;12:eaaz8084. [DOI] [PubMed] [Google Scholar]
- 15. Parsons HA, Blewett T, Chu X, Sridhar S, Santos K, Xiong K, et al. Circulating tumor DNA association with residual cancer burden after neoadjuvant chemotherapy in triple-negative breast cancer in TBCRC 030. Ann Oncol 2023;34:899–906. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Parsons HA, Rhoades J, Reed SC, Gydush G, Ram P, Exman P, et al. Sensitive detection of minimal residual disease in patients treated for early-stage breast cancer. Clin Cancer Res 2020;26:2556–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Xiong K, Shea D, Rhoades J, Blewett T, Liu R, Bae JH, et al. Duplex-repair enables highly accurate sequencing, despite DNA damage. Nucleic Acids Res 2022;50:e1. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data underlying this article are available through dbGaP with the accession number: phs003447.v1.p1.
References
- 1. Etienne G, Guilhot J, Rea D, Rigal-Huguet F, Nicolini F, Charbonnier A, et al. Long-term follow-up of the French stop imatinib (STIM1) study in patients with chronic myeloid leukemia. J Clin Oncol 2017;35:298–305. [DOI] [PubMed] [Google Scholar]
- 2. Magbanua MJM, Swigart LB, Wu H-T, Hirst GL, Yau C, Wolf DM, et al. Circulating tumor DNA in neoadjuvant-treated breast cancer reflects response and survival. Ann Oncol 2021;32:229–39. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Radovich M, Jiang G, Hancock BA, Chitambar C, Nanda R, Falkson C, et al. Association of circulating tumor DNA and circulating tumor cells after neoadjuvant chemotherapy with disease recurrence in patients with triple-negative breast cancer: preplanned secondary analysis of the BRE12-158 randomized clinical trial. JAMA Oncol 2020;6:1410–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Garcia-Murillas I, Chopra N, Comino-Méndez I, Beaney M, Tovey H, Cutts RJ, et al. Assessment of molecular relapse detection in early-stage breast cancer. JAMA Oncol 2019;5:1473–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Lipsyc-Sharf M, de Bruin EC, Santos K, McEwen R, Stetson D, Patel A, et al. Circulating tumor DNA and late recurrence in high-risk hormone receptor-positive, human epidermal growth factor receptor 2-negative breast cancer. J Clin Oncol 2022;40:2408–19. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Azad TD, Chaudhuri AA, Fang P, Qiao Y, Esfahani MS, Chabon JJ, et al. Circulating tumor DNA analysis for detection of minimal residual disease after chemoradiotherapy for localized esophageal cancer. Gastroenterology 2020;158:494–505.e6.. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Bettegowda C, Sausen M, Leary RJ, Kinde I, Wang Y, Agrawal N, et al. Detection of circulating tumor DNA in early- and late-stage human malignancies. Sci Transl Med 2014;6:224ra24. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Cohen JD, Li L, Wang Y, Thoburn C, Afsari B, Danilova L, et al. Detection and localization of surgically resectable cancers with a multi-analyte blood test. Science 2018;359:926–30. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Corrò C, Hejhal T, Poyet C, Sulser T, Hermanns T, Winder T, et al. Detecting circulating tumor DNA in renal cancer: an open challenge. Exp Mol Pathol 2017;102:255–61. [DOI] [PubMed] [Google Scholar]
- 10. Kim Y-W, Kim Y-H, Song Y, Kim H-S, Sim HW, Poojan S, et al. Monitoring circulating tumor DNA by analyzing personalized cancer-specific rearrangements to detect recurrence in gastric cancer. Exp Mol Med 2019;51:1–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Yamamoto Y, Uemura M, Fujita M, Maejima K, Koh Y, Matsushita M, et al. Clinical significance of the mutational landscape and fragmentation of circulating tumor DNA in renal cell carcinoma. Cancer Sci 2019;110:617–28. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Coombes RC, Page K, Salari R, Hastings RK, Armstrong A, Ahmed S, et al. Personalized detection of circulating tumor DNA antedates breast cancer metastatic recurrence. Clin Cancer Res 2019;25:4255–63. [DOI] [PubMed] [Google Scholar]
- 13. Gydush G, Nguyen E, Bae JH, Blewett T, Rhoades J, Reed SC, et al. Massively parallel enrichment of low-frequency alleles enables duplex sequencing at low depth. Nat Biomed Eng 2022;6:257–66. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Wan JCM, Heider K, Gale D, Murphy S, Fisher E, Mouliere F, et al. ctDNA monitoring using patient-specific sequencing and integration of variant reads. Sci Transl Med [Internet] 2020;12:eaaz8084. [DOI] [PubMed] [Google Scholar]
- 15. Parsons HA, Blewett T, Chu X, Sridhar S, Santos K, Xiong K, et al. Circulating tumor DNA association with residual cancer burden after neoadjuvant chemotherapy in triple-negative breast cancer in TBCRC 030. Ann Oncol 2023;34:899–906. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Parsons HA, Rhoades J, Reed SC, Gydush G, Ram P, Exman P, et al. Sensitive detection of minimal residual disease in patients treated for early-stage breast cancer. Clin Cancer Res 2020;26:2556–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Xiong K, Shea D, Rhoades J, Blewett T, Liu R, Bae JH, et al. Duplex-repair enables highly accurate sequencing, despite DNA damage. Nucleic Acids Res 2022;50:e1. [DOI] [PMC free article] [PubMed] [Google Scholar]





