Duplex Sequencing for Ultra-Low Frequency Measurable Residual Disease Detection in Adult Acute Myeloid Leukemia

Background: The presence of measurable residual disease (MRD) is strongly associated with treatment outcomes in acute myeloid leukemia (AML). However, > 20% of MRD negative cases (as assessed by flow cytometry) subsequently relapse. We sought to determine if ultrasensitive Duplex Sequencing (DS), which relies on double-stranded consensus-making to achieve an error rate below one-in-ten-million, yields better prognostic performance via molecular MRD detection. Methods: Retrospective targeted DNA sequencing of 29 genes recurrently mutated in adult AML was performed on paired diagnostic and remission bone marrow samples from patients enrolled on the SWOG trial S0106 (randomized 7+3 versus 7+3 + gemtuzumab ozogamicin (GO)). Patients were selected if they had remission samples with flow cytometry results (n=67). Non-error corrected sequencing was performed on diagnostic samples (average depth 279x) and DS was performed on remission samples (average duplex molecular depth 27,002x). For each patient, potential germline variants were identified and excluded from the analysis if the variant allele fraction (VAF) was ≥ 35% at both diagnosis and remission, or ≥ 40% at either time point and a gnomAD allele frequency ≥ 0.05. Somatic variants present at diagnosis were classified as potentially deleterious if computationally predicted as such and with a VAF ≥ 5% (≥ 1% for FLT3-ITD/NPM1 insertions). For analysis of residual disease in remission, we evaluated the following outcomes (events): overall survival (OS; death), relapse-free survival (RFS) and time to relapse (TTR; relapse with death a competing event). All outcomes were measured from date of morphologic remission to date of event, with patients without event censored at date of last contact. Associations between residual disease and outcomes were assessed using Cox regression models (cause-specific model for TTR). Results: The median age was 48 years (range 8-60). 32 patients were randomized to 7+3 and 30 to 7+3+GO. A total of 172 potentially deleterious variants were identified in the diagnostic samples. Variants had an average VAF of 31% (range 1.4-91.5%) at diagnosis and were detected in 23 of the 29 genes, with FLT3 being the most frequently mutated. Of the 67 patients analyzed, 93% (n=62) had at least one variant detected at diagnosis (median 2, range 0-9) and 68% (n=42) had at least one residual diagnostic variant also found in the remission sample. We defined the presence of DS MRD as non-DTA (DNMT3A, TET2, ASXL1) time-of-diagnosis mutations identified at the remission time point with a VAF > 0.1% and/or an NPM1 VAF > 0.01% (PMID:31860405). DS MRD was strongly associated with all outcomes, with hazard ratios (and 95% CI) for TTR: 7.1 (2.7-18.9); RFS: 4.9 (2.2-10.9) and OS: 5.1 (2.1-12.3). As a comparator, we correlated treatment outcomes with the results of a flow cytometry MRD assay previously carried out on the same samples during the S0106 trial. The prognostic association of flow MRD with TTR, RFS and OS was less strong (i.e., a smaller hazard ratio) than DS MRD, with TTR: 2.5 (0.9-6.7); RFS: 2.2 (0.9-5.4) and OS: 2.4 (1.0-6.1). RFS and TTR for DS MRD and flow cytometry are plotted below for the 62 patients with a variant detected at diagnosis. Comparing DS MRD with flow cytometry, discordance was found in 20 cases: 15 cases where DS was positive and flow negative, and 5 cases in the opposite direction. Among the 15 discordant cases with DS positive and flow negative, 9 relapsed and 2 died without relapse and 4 were alive without relapse at last contact. Among the 5 discordant cases with DS negative and flow positive, 1 relapsed, 1 died without relapse, and 3 were alive without relapse at last contact. Conclusions: Among the 67 patients evaluated in this prospectively collected study, the presence of MRD defined by DS was strongly associated with adverse disease outcomes. The vast majority of patients had at least one time-of-diagnosis mutation that could be tracked as a measure of MRD. When compared with flow cytometry, DS exhibited superior negative and positive predictive values for foretelling future relapse, although this could potentially reflect the historical version of the flow cytometry assay used. These findings suggest that DS is a powerful tool that could be used in patient management and for early treatment assessment in clinical trials. Figure 1 Figure 1. Disclosures Hourigan: Sellas: Research Funding. Radich: Amgen: Membership on an entity's Board of Directors or advisory committees; Novartis: Membership on an entity's Board of Directors or advisory committees; Genentech: Membership on an entity's Board of Directors or advisory committees; BMS: Membership on an entity's Board of Directors or advisory committees.

Duplex-Repair enables highly accurate sequencing, despite DNA damage

Accurate DNA sequencing is crucial in biomedicine. Underlying the most accurate methods is the assumption that a mutation is true if altered bases are present on both strands of the DNA duplex. We now show that this assumption can be wrong. We establish that current methods to prepare DNA for sequencing, via ‘End Repair/dA-Tailing,’ may substantially resynthesize strands, leading amplifiable lesions or alterations on one strand to become indiscernible from true mutations on both strands. Indeed, we discovered that 7–17% and 32–57% of interior ‘duplex base pairs’ from cell-free DNA and formalin-fixed tumor biopsies, respectively, could be resynthesized in vitro and potentially introduce false mutations. To address this, we present Duplex-Repair, and show that it limits interior duplex base pair resynthesis by 8- to 464-fold, rescues the impact of induced DNA damage, and affords up to 8.9-fold more accurate duplex sequencing. Our study uncovers a major Achilles’ heel in sequencing and offers a solution to restore high accuracy.

High prevalence of somatic PIK3CA and TP53 pathogenic variants in the normal mammary gland tissue of sporadic breast cancer patients revealed by duplex sequencing.

The mammary gland undergoes hormonally stimulated cycles of proliferation, lactation and involution. We hypothesized that these factors increase the mutational burden in glandular tissue and may explain high cancer incidence rate in the general population and recurrent disease. Hence, we investigated the DNA sequence variants in the normal mammary gland, tumor and peripheral blood from 52 reportedly sporadic breast cancer patients, including breast-conserving surgery cases. Targeted resequencing of 542 cancer associated genes revealed mosaic somatic pathogenic variants of: PIK3CA, TP53, AKT1, MAP3K1, CDH1, RB1, NCOR1, MED12, CBFB, TBX3 and TSHR in the normal mammary gland, at considerable allelic frequencies (9x10-2 to 5.2x10-1) indicating clonal expansion. Further evaluation of the frequently damaged PIK3CA and TP53 genes by ultra-sensitive duplex sequencing demonstrated a diversified picture of multiple low level-mosaic (in 10-2 to 10-4 alleles) hotspot pathogenic variants. Our results raise a question about the oncogenic potential in non-tumor mammary gland tissue of breast-conserving surgery patients.

CODEC enables 'single duplex' sequencing

Detecting mutations as rare as a single molecule is crucial in many fields such as cancer diagnostics and aging research but remains challenging. Third generation sequencers can read a double-stranded DNA molecule (a 'single duplex') in whole to identify true mutations on both strands apart from false mutations on either strand but with limited accuracy and throughput. Although next generation sequencing (NGS) can track dissociated strands with Duplex Sequencing, the need to sequence each strand independently severely diminishes its throughput. Here, we developed a hybrid method called Concatenating Original Duplex for Error Correction (CODEC) that combines the massively parallel nature of NGS with the single-molecule capability of third generation sequencing. CODEC physically links both strands to enable NGS to sequence a single duplex with a single read pair. By comparing CODEC and Duplex Sequencing, we showed that CODEC achieved a similar error rate (10-6) with 100 times fewer reads and conferred 'single duplex' resolution to most major NGS workflows.

Duplex-Repair enables highly accurate sequencing, despite DNA damage

Accurate DNA sequencing is crucial in biomedicine. Underlying the most accurate methods is the assumption that a mutation is true if altered bases are present on both strands of the DNA duplex. We now show that this assumption can be wrong. We establish that current methods to prepare DNA for sequencing, via End Repair/dA-Tailing, may substantially resynthesize strands, leading amplifiable lesions or alterations on one strand to become indiscernible from true mutations on both strands. Indeed, we discovered that 7-17% and 32-57% of interior duplex base pairs from cell-free DNA and formalin-fixed tumor biopsies, respectively, could be resynthesized in vitro and potentially introduce false mutations. To address this, we present Duplex-Repair, and show that it limits interior duplex base pair resynthesis by 8- to 464-fold, rescues the impact of induced DNA damage, and affords up to 8.9-fold more accurate duplex sequencing. Our study uncovers a major Achilles heel in sequencing and offers a solution to restore high accuracy.

Phased variants improve DLBCL minimal residual disease detection at the end of therapy.

7565 Background: Detection of circulating tumor DNA (ctDNA) has prognostic value in diverse tumors, including DLBCL. Despite uses for assessing molecular response to therapy, current methods using immunoglobulin or hybrid-capture sequencing have suboptimal sensitivity, particularly when disease-burden is low. This contributes to a high false negative rate at key milestones such as at the end of therapy (EOT; Kumar A, ASH 2020). We explored the utility of detecting multiple mutations (phased variants, PVs) on individual cell-free DNA (cfDNA) strands to improve MRD in DLBCL. Methods: We applied Phased Variant Enrichment and Detection Sequencing to track PVs from 485 specimens from 117 DLBCL patients undergoing first-line therapy. We sequenced cfDNA prior to, during, and after therapy to assess the prognostic value of MRD. We compared the performance of PhasED-Seq to current techniques, including SNV-based CAPP-Seq and duplex sequencing. Results: To establish its detection limit for ctDNA, we compared the background error-profile of of PVs and SNVs in cfDNA sequencing from healthy subjects. PV-detection by PhasED-Seq demonstrated a lower background profile than SNVs, even when considering duplex molecules (n = 12; 8.0e-7 vs 3.3e-5 and 1.2e-5; P < 0.0001). We also assessed analytical sensitivity within a ctDNA limiting dilution series from 3 patients, simulating tumor fractions from 0.1% to 0.00005% (1:2,000,000). PhasED-Seq outperformed SNV-based methods and duplex sequencing for recovery of expected tumor content below 0.01% (P < 0.0001 and P = 0.005 respectively by paired t-test). We then explored disease detection in clinical samples. We identified SNVs and PVs from pretreatment tumor or plasma and followed these variants in serial cfDNA. Using SNV-based methods, 40% and 59% of patients had undetectable ctDNA after 1 or 2 cycles (n = 82 and 88). However, 24% and 25% of these cases had detectable ctDNA by PhasED-Seq. Importantly, MRD detection by PhasED-Seq was prognostic for event-free survival even in patients with undetectable ctDNA by SNVs. We next explored the utility of PhasED-Seq at the EOT in 19 subjects, 5 of whom experienced eventual disease progression. While only 2/5 cases with progression had detectable disease at EOT using SNVs, PhasED-Seq detected all 5/5 cases. PhasED-Seq also correctly identified all patients (14/14) without clinical relapse as having no residual disease, including one patient who discontinued therapy after 1 cycle due to toxicity, but remains in remission > 5 years after this single treatment. This resulted in superior classification of patients for EFS using PVs compared with SNVs (C-statistic: 0.98 vs 0.60, P = 0.02). Conclusions: Tracking PVs results in significantly lower background rates than SNV-based approaches, enabling detection to parts per million range. PhasED-Seq improves on disease detection in DLBCL at the EOT, allowing possible MRD-driven consolidative approaches.

Somatic mutation landscapes at single-molecule resolution

Somatic mutations drive cancer development and may contribute to ageing and other diseases. Yet, the difficulty of detecting mutations present only in single cells or small clones has limited our knowledge of somatic mutagenesis to a minority of tissues. To overcome these limitations, we introduce nanorate sequencing (NanoSeq), a new duplex sequencing protocol with error rates <5 errors per billion base pairs in single DNA molecules from cell populations. The version of the protocol described here uses clean genome fragmentation with a restriction enzyme to prevent end-repair-associated errors and ddBTPs/dATPs during A-tailing to prevent nick extension. Both changes reduce the error rate of standard duplex sequencing protocols by preventing the fixation of DNA damage into both strands of DNA molecules during library preparation. We also use qPCR quantification of the library prior to amplification to optimise the complexity of the sequencing library given the desired sequencing coverage, maximising duplex coverage. The sample preparation protocol takes between 1 and 2 days, depending on the number of samples processed. The bioinformatic protocol is described in:https://github.com/cancerit/NanoSeqhttps://github.com/fa8sanger/NanoSeq_Paper_Code