Background Error Rate Research Articles

Automated search methods for identifying wrong patient order entry-a scoping review.

To investigate: (1) what automated search methods are used to identify wrong-patient order entry (WPOE), (2) what data are being captured and how they are being used, (3) the causes of WPOE, and (4) how providers identify their own errors. A systematic scoping review of the empirical literature was performed using the databases CINAHL, Embase, and MEDLINE, covering the period from database inception until 2021. Search terms were related to the use of automated searches for WPOE when using an electronic prescribing system. Data were extracted and thematic analysis was performed to identify patterns or themes within the data. Fifteen papers were included in the review. Several automated search methods were identified, with the retract-and-reorder (RAR) method and the Void Alert Tool (VAT) the most prevalent. Included studies used automated search methods to identify background error rates in isolation, or in the context of an intervention. Risk factors for WPOE were identified, with technological factors and interruptions deemed the biggest risks. Minimal data on how providers identify their own errors were identified. RAR is the most widely used method to identify WPOE, with a good positive predictive value (PPV) of 76.2%. However, it will not currently identify other error types. The VAT is nonspecific for WPOE, with a mean PPV of 78%-93.1%, but the voiding reason accuracy varies considerably. Automated search methods are powerful tools to identify WPOE that would otherwise go unnoticed. Further research is required around self-identification of errors.

Abstract PR012: Ultra-sensitive detection of circulating tumour DNA enriches for patients with higher risk disease in clinically localised prostate cancer

Abstract Purpose Circulating tumour DNA (ctDNA) analysis has demonstrated utility for diagnostic and prognostic applications in many cancer types, including metastatic prostate cancer. However, localised prostate cancer yields relatively low levels of ctDNA and therefore it has been difficult to detect in this context using conventional methods. We aimed to assess the limits of detection of ctDNA in localised prostate cancer by leveraging thousands of patient-specific mutations per case with the high-sensitivity INVAR method, and to test the hypothesis that ctDNA detection is associated with high risk disease. Experimental Procedures A total of 128 individuals with clinically localised prostate cancer at the time of sample collection were selected from cohorts in Australia (n=48) and the UK (n=80). Additionally, 27 healthy individuals without prostate cancer were included as negative controls. Primary tumour tissue and matched bloods from all cases were whole genome sequenced (WGS) and somatic variants were called using pipelines from the Pan Prostate Cancer Group. Plasma cell-free DNA (cfDNA) samples from cases and controls were profiled using custom targeted sequencing panels, with saturating coverage of patient-specific mutations identified by WGS. We assessed ctDNA detection in cases using the highly sensitive INVAR pipeline, that leverages consensus read sequencing alignments, background error modelling and integration of signals across thousands of patient-specific variants. Biochemical recurrence and metastasis-free survival curves were used to assess the relationship between ctDNA detection and disease progression. Results We analysed pre-treatment blood plasma ctDNA in a large cohort of localised prostate cancer patients, using error-suppressed targeted sequencing of over 280k patient-specific mutations. To comprehensively assess ctDNA mutation analysis in this context, we combined signals across the maximum number of genome-wide patient specific mutations and leveraged an established analysis pipeline (INVAR) that corrects for background error rates and calculates a global integrated mutant allele fraction. ctDNA was detected in 9.3% of localised prostate cancer patients. In cases where ctDNA was detected we found significant associations with biochemical recurrence (p=0.01) and shorter metastasis-free survival (p < 0.0001). Conclusions Our study provides clear insights into the required analytical sensitivity and potential utility of ctDNA mutation analysis in localised prostate cancer. We found that mutation-based ctDNA detection rates were low in localised prostate cancer (<10% cases), but in ctDNA positive cases there was a significant association with relapse after surgical intervention alone. This raises the potential for including ctDNA detection as an additional tool for patient stratification in future neo/adjuvant treatment trials aiming to assess the impact of treatment escalation in men at high risk of relapse with current standard of care treatment alone. Citation Format: Bernard J. Pope, Gahee Park, Edmund Lau, Jelena Belic, Radoslaw Lach, Anne George, Patrick McCoy, Anne Nguyen, Corrina Grima, Chol-hee Jung, Emma-Jane Ditter, Hui Zhao, David Wedge, Rosalind A Eeles, Daniel Brewer, Andy G. Lynch, Harveer Dev, Christopher M Hovens, Vincent J. Gnanpragasam, Nitzan Rosenfeld, Niall M. Corcoran, Charles E. Massie. Ultra-sensitive detection of circulating tumour DNA enriches for patients with higher risk disease in clinically localised prostate cancer [abstract]. In: Proceedings of the AACR Special Conference: Advances in Prostate Cancer Research; 2023 Mar 15-18; Denver, Colorado. Philadelphia (PA): AACR; Cancer Res 2023;83(11 Suppl):Abstract nr PR012.

Improving the performance of reference-frame-independent quantum key distribution with advantage distillation technology.

The reference-frame-independent quantum key distribution (RFI-QKD) has the advantage of tolerating reference frames that slowly vary. It can generate secure keys between two remote users with slowly drifted and unknown reference frames. However, the drift of reference frames may inevitably compromise the performance of QKD systems. In the paper, we employ the advantage distillation technology (ADT) to the RFI-QKD and the RFI measurement-device-independent QKD (RFI MDI-QKD), and we then analyze the effect of ADT on the performance of decoy-state RFI-QKD and RFI MDI-QKD in both asymptotic and nonasymptotic cases. The simulation results show that ADT can significantly improve the maximum transmission distance and the maximum tolerable background error rate. Furthermore, the performance of RFI-QKD and RFI MDI-QKD in terms of the secret key rate and maximum transmission distance are still greatly improved when statistical fluctuations are taken into account. Our work combines the merits of the ADT and RFI-QKD protocols, which further enhances the robustness and practicability of QKD systems.

HashSeq: a Simple, Scalable, and Conservative De Novo Variant Caller for 16S rRNA Gene Data Sets.

ABSTRACT16S rRNA gene sequencing is a common and cost-effective technique for characterization of microbial communities. Recent bioinformatics methods enable high-resolution detection of sequence variants of only one nucleotide difference. In this study, we utilized a very fast HashMap-based approach to detect sequence variants in six publicly available 16S rRNA gene data sets. We then use the normal distribution combined with locally estimated scatterplot smoothing (LOESS) regression to estimate background error rates as a function of sequencing depth for individual clusters of sequences. This method is computationally efficient and produces inference that yields sets of variants that are conservative and well supported by reference databases. We argue that this approach to inference is fast, simple, and scalable to large data sets and provides a high-resolution set of sequence variants which are less likely to be the result of sequencing error.IMPORTANCE Recent bioinformatics development has enabled the detection of sequence variants with a high resolution of only one single-nucleotide difference in 16S rRNA gene sequence data. Despite this progress, there are several limitations that can be associated with variant calling pipelines, such as producing a large number of low-abundance sequence variants which need to be filtered out with arbitrary thresholds in downstream analyses or having a slow runtime. In this report, we introduce a fast and scalable algorithm which infers sequence variants based on the estimation of a normally distributed background error as a function of sequencing depth. Our pipeline has attractive performance characteristics, can be used independently or in parallel with other variant callers, and provides explicit P values for each variant evaluating the hypothesis that a variant is caused by sequencing error.

Development of a User-Friendly Pipeline for Mutational Analyses of HIV Using Ultra-Accurate Maximum-Depth Sequencing.

Human immunodeficiency virus type 2 (HIV-2) accumulates fewer mutations during replication than HIV type 1 (HIV-1). Advanced studies of HIV-2 mutagenesis, however, have historically been confounded by high background error rates in traditional next-generation sequencing techniques. In this study, we describe the adaptation of the previously described maximum-depth sequencing (MDS) technique to studies of both HIV-1 and HIV-2 for the ultra-accurate characterization of viral mutagenesis. We also present the development of a user-friendly Galaxy workflow for the bioinformatic analyses of sequencing data generated using the MDS technique, designed to improve replicability and accessibility to molecular virologists. This adapted MDS technique and analysis pipeline were validated by comparisons with previously published analyses of the frequency and spectra of mutations in HIV-1 and HIV-2 and is readily expandable to studies of viral mutation across the genomes of both viruses. Using this novel sequencing pipeline, we observed that the background error rate was reduced 100-fold over standard Illumina error rates, and 10-fold over traditional unique molecular identifier (UMI)-based sequencing. This technical advancement will allow for the exploration of novel and previously unrecognized sources of viral mutagenesis in both HIV-1 and HIV-2, which will expand our understanding of retroviral diversity and evolution.

Ultra-Deep Sequencing As a Predictive Tool for MDS Progression after Allogeneic Hematopoietic Cell Transplant

Background:Allogeneic hematopoietic cell transplantation (alloHCT) is the only curative treatment for patients with myelodysplastic syndromes (MDS). Up to 40% of MDS patients will die within the first year post-alloHCT due to progressive disease or treatment-related mortality. Predictors of post-alloHCT progression are needed. While recent data have shown that gene mutations in pre-transplant samples, including TP53, are associated with worse outcomes post-alloHCT, it is not known whether persistence of patient-specific mutations after alloHCT are predictive of progression. In this study we identified somatic mutations in MDS patients prior to alloHCT and then used ultra-deep sequencing of day 30 and day 100 post-alloHCT bone marrows to determine whether the extent of mutation clearance was associated with outcome.Methods:We selected 55 MDS patients who underwent alloHCT with a mean age of 55.6 years (range 23-71) and IPSS-R scores of very low (1), low (12), intermediate (6), high (23), and very high (13). Transplants included myeloablative and non-myeloablative conditioning regimens with related and unrelated donors. Of the 55 patients, 19 relapsed post-alloHCT and 36 did not with a mean of 242 days to relapse (range 27-1015) and a mean follow-up of 883 days (range 103-2106), respectively. Initial pre-alloHCT bone marrows and paired normal skin (as a source of control DNA) were enhanced exome sequenced (EES) to yield higher coverage in known recurrently mutated AML/MDS genes. To provide high sensitivity at post-alloHCT time points, an ultra-sensitive, error-corrected sequencing approach using unique molecular identifiers (UMIs) was combined with high-coverage depths. Probes were designed to target mutations detected by EES in the pre-alloHCT samples. Initial diagnostic marrow, skin, and day 30 post-alloHCT marrow samples were sequenced in all patients and day 100 post-alloHCT marrows were available in 37 patients. Variant allele fractions (VAFs) of somatic single nucleotide variant (SNV) mutations at day 30 and day 100 post-alloHCT were detected using a custom analysis pipeline employing read-level UMI-based error correction and base-level background error rate corrections. SNV calls were correlated with post-alloHCT progression-free survival and overall survival using Kaplan-Meier and Cox proportional hazards models, respectively.Results:EES was performed to a mean of 230x depth (1310x for recurrently mutated genes). A mean of 40.3 (range 17-151) somatic variants per case were identified with a mean VAF of 29.6%. Of these, a mean of 23 (range 1-70) SNVs per case were targeted and analyzed. Day 30 post-alloHCT marrows were sequenced to a mean depth of 23,062x corresponding to 2,753x unique coverage; day 100 samples were sequenced to 27,936x corresponding to 3,430x unique coverage. A SNV was included in the analysis if the error-corrected pre-transplant VAF was >10% and had >100x unique coverage at day 30. Nineteen of 55 (35%) and 18 of 37 (49%) cases had no mutations detected on post-alloHCT days 30 and 100, respectively. For patients with detectable mutations, the mean VAF per patient at day 30 was 2.35% (range of detected mutations 0.001-25.81%) and 3.44% (range 0.001-21.45%) at day 100. For patients with detectable mutations, the mean maximum VAF per patient at day 30 was 7.21% (range 0.020-88.19%) and 3.44% (range 0.023-38.98%) at day 100.The presence of at least 1 mutation with a VAF greater than 0.5% (equivalent to 1 heterozygous mutant cell in 100) at day 30 post-alloHCT was associated with a worse progression-free survival (median 3.8 vs. 21.9 months; p=0.031). Similar results were obtained using VAF cutoffs of 1% and 2.5% (p= 0.018, p=0.0002, respectively). In contrast, day 100 post-alloHCT marrows did not show a similar correlation. The effect of persistent mutations on overall survival was assessed with a Cox proportional hazards model. A higher mean VAF at day 30 post-alloHCT was associated with worse overall survival (p=0.028); for each mean VAF increase of 1% at day 30, the risk of death increased by about 10% [HR 1.10, 95% CI (1.01,1.20)].Conclusions:Detection of persistent mutations in the bone marrow by ultra-sensitive, error-corrected sequencing at day 30 post-alloHCT is associated with an increased risk of progression. These findings suggest that sequencing could be used as an individualized risk-assessment marker in MDS patients who have undergone alloHCT. DisclosuresUy:Novartis: Consultancy, Other: Travel Suppport; Boehringer Ingelheim: Consultancy; GlycoMimetics: Consultancy.

Leveraging phased variants for personalized minimal residual disease detection in localized non-small cell lung cancer.

8518 Background: Detection of circulating tumor DNA (ctDNA) has prognostic value in lung cancer and could facilitate minimal residual disease (MRD) driven approaches. However, the sensitivity of ctDNA detection is suboptimal due to the background error rates of existing assays. We developed a novel method leveraging multiple mutations on a single cell-free DNA molecule (“phased variants” or PVs) resulting in an ultra-low error profile. Here we develop and apply this approach to improve MRD in localized NSCLC. Methods: To identify the prevalence of PVs, we reanalyzed whole genome sequencing (WGS) from 2,538 tumors and 24 cancer types from the pan-cancer analysis of whole genomes (PCAWG). We applied Phased Variant Enrichment and Detection Sequencing (PhasED-Seq) to track personalized PVs in localized NSCLC. We compared PhasED-Seq to a single nucleotide variant (SNV)-based ctDNA method. Results: In the PCAWG dataset, we found that PVs were common in both lung squamous cell carcinomas (LUSC, median 1,268/tumor; rank 2nd) and adenocarcinomas (LUAD, median 655.5/tumor; rank 3rd). However, PVs did not occur in stereotyped genomic regions. Thus, to leverage PhasED-Seq, we performed tumor/normal WGS to identify PVs, followed by design of personalized panels targeting PVs to allow deep cfDNA sequencing. We performed personalized PhasED-Seq for 5 patients with localized NSCLC. PVs were identified from WGS of tumor FFPE and validated by targeted resequencing in all cases (median 248/case). The background rate of PVs was lower than that of SNVs, even when considering duplex molecules (background: SNVs, 3.8e-5; duplex SNVs, 1.0e-5; PVs, 1.2e-6; P < 0.0001). We next assessed PhasED-Seq for MRD detection in 14 patient plasma samples. Both SNVs and PhasED-Seq had high specificity in healthy control cfDNA (95% and 97% respectively). Using SNVs, ctDNA was detected in 5/14 samples; PhasED-Seq detected all of these with nearly identical tumor fractions (Spearman rho = 0.97). However, PhasED-Seq also detected MRD in an additional 5 samples containing tumor fractions as low as 0.000094% (median 0.0004%). We analyzed serial samples from a patient with stage III LUAD treated with chemoradiotherapy (CRT) and durvalumab. SNV-based ctDNA and PhasED-Seq detected similar MRD levels (0.8%) prior to therapy. However, 3 samples collected during CRT, as well as before and during immunotherapy, were undetectable by SNVs. SNV-based ctDNA then re-emerged at disease recurrence. PhasED-Seq detected MRD in all 3 samples not detected by SNVs with tumor fractions as low as 0.00016%, including prior to immunotherapy (8 months prior to progression). Similar improvements were seen in samples not detected by SNVs from 2 additional patients. Conclusions: Personalized ctDNA monitoring via PVs is feasible and improves MRD detection in localized NSCLC. PhasED-Seq allows clinical studies testing personalized treatment based on MRD.

Cfcloud: A Cloud-Based Workflow for Cell-Free DNA Data Analysis

Introductions:Cell-free DNA (cfDNA) are extracellular nucleic acids circulating in blood and other biological fluids. They are non-randomly fragmented, mainly short (< 200 bp) DNA molecules released as a result of apoptosis, necrosis or active secretion from cells. They are believed to be protected from degradation by nucleases due to nucleosome structure. cfDNA is a promising biomarker for prenatal testing, hemoglobinopathies, inflammatory progress, and detection of tumor-derived DNA in plasma. The amount of donor derived cell-free DNA circulating in the recipient's plasma can be used as a signature of dying cells from the transplanted organ. Donor derived cell-free DNA percent (%dd-cfDNA) can be quantified using next generation sequencing (NGS)-based method that utilizes genotype information of recipient and donor by measuring single nucleotide polymorphism (SNP) differences. This method was used in Genome Transplant Dynamics (GTD) cohort study. This study showed that elevated dd-cfDNA levels correlate with rejection of the transplanted organ. GTD method was shown to be a rigorous and highly reproducible across different genotyping or sequencing platforms. Here, we present thecfCloud, an open source, cloud-based implement of the GTD method in a Snakemake pipeline which automates the quantification of dd-cfDNA amount. Methods:As illustrated in Figure 1, the analysis workflow contains several modules and rules; 1) Align reads rule - aligns sequencing reads against the reference genome (hg19), and marks duplicated reads. 2) Prepare genotype rule - given the input VCF file, for every recipient/donor pair, generates a SNP table by classifying SNPs as i) Group A, SNPs where both recipient (R) and donor (D) are homozygous with the same allele, like R=AA, D=AA, and ii) Group B, SNPs where both recipient and donor are homozygous with different alleles, like R=AA, D=BB. 3) Make CallsTable rule - for every SNP position, if the observed base of an overlapping read matches the recipient allele, it classifies that read as recipient-derived, if the observed base matches the donor allele then it classifies as donor-derived, otherwise as background-error. CallsTable lists chromosome, position, base quality and mapping quality scores of all the sites as well as the assigned category. 4) Call % dd-cfDNA rule - using the CallsTable, filters out the reads based on the input parameters. Then, calculates % dd-cfDNA by using SNPs in Group A and estimates background-error rate by using SNPs in Group B. 5) Prepare FinalReport rule - generates a final report showing total counts, donor counts, dd-cfDNA percentages and estimated background-error rates for all the samples. Results:Using the workflow management system, Snakemake, we have developed a user friendly, efficient, and comprehensive pipeline that is fully automated, reproducible, and scalable. While the pipeline is fully customizable, it can be run using a simple command line. Upon execution, Snakemake infers the combination of rules necessary to achieve a specific output and or a report. With a basic understanding of the underlying framework of Snakemake, cfCloud is highly scalable and can be easily customized. In addition to local implementation, cfCloud also runs as an Amazon Machine Instance (AMI) in the Amazon Elastic Compute Cloud (EC2). The operating system, software tools and all settings, are encapsulated into a single cloud image of the computing system that is ready to deploy and can be easily saved and restored for later use. A snapshot is also an executable machine image that can be shared with other users of the cloud, allowing collaborators to share their configurations and analysis results as a single image. Conclusions:Here we present a fully automated pipeline to systematize the quantification of donor derived cell-free DNA amount. The Amazon Web Machine Image is available from the AWS Market Place. The cfCloud provides users with improved usability, efficiency, and scalability in local and cloud environment. The cfCloud software, installation instructions, tutorials and example data can be freely obtained from https://github.com/NHLBI-BCB/cfCloud under the MIT license. Figure Disclosures No relevant conflicts of interest to declare.

Detection of genome-wide low-frequency mutations with Paired-End and Complementary Consensus Sequencing (PECC-Seq) revealed end-repair-derived artifacts as residual errors.

To improve the accuracy and the cost-efficiency of next-generation sequencing in ultralow-frequency mutation detection, we developed the Paired-End and Complementary Consensus Sequencing (PECC-Seq), a PCR-free duplex consensus sequencing approach. PECC-Seq employed shear points as endogenous barcodes to identify consensus sequences from the overlap in the shortened, complementary DNA strand-derived paired-end reads for sequencing error correction. With the high accuracy of PECC-Seq, we identified the characteristic base substitution errors introduced by the end-repair process of mechanical fragmentation-based library preparations, which were prominent at the terminal 7bp of the library fragments in the 5'-NpCpA-3' and 5'-NpCpT-3' trinucleotide context. As demonstrated at the human genome scale (TK6 cells), after removing these potential end-repair artifacts from the terminal 7bp, PECC-Seq could reduce the sequencing error frequency to mid-10-7 with a relatively low sequencing depth. For TA base pairs, the background error rate could be suppressed to mid-10-8. In mutagen-treated (6μg/mL methyl methanesulfonate or 12μg/mL N-nitroso-N-ethylurea) TK6, increases in mutagen treatment-related mutant frequencies could be detected, indicating the potential of PECC-Seq in detecting genome-wide ultra-rare mutations. In addition, our finding on the patterns of end-repair artifacts may provide new insights into further reducing technical errors not only for PECC-Seq, but also for other next-generation sequencing techniques.

Abstract PR08: Validation and clinical implementation of MSK-ACCESS, an ultra-deep sequencing assay for noninvasive somatic mutation profiling

Abstract Introduction: Circulating cell-free DNA (cfDNA) is a source of tumor-derived DNA to interrogate somatic alterations when tissue is not available or of insufficient quantity for analysis. At MSKCC, we have developed and validated MSK-ACCESS (Analysis of Circulating cfDNA to Evaluate Somatic Status), a targeted next-generation sequencing assay that can detect ultra-low frequency somatic variants in select exons and introns of 129 genes. MSK-ACCESS can identify mutations, copy number alterations, gene fusions, and MSI status in plasma and was recently approved by the NYS-DOH for clinical testing. Here, we present the results of the validation study and our clinical experience with MSK-ACCESS since June 2019. Methods: Target regions from 129 genes were selected to maximize coverage of actionable, oncogenic, and hotspot mutations based on the first 25,000 tumors sequenced using MSK-IMPACT, our institutional clinical sequencing assay. Plasma cfDNA and buffy-coat DNA were extracted from whole blood collected in cell-stabilizing tubes (STRECK BCT cell-free DNA tube). Unique molecular indexes were introduced during DNA library construction, allowing for error suppression from consensus reads collapsed by Marianas, an in-house-developed algorithm. These consensus reads enable variant calling at low allelic frequency (AF) based on a 10−6 background error rate. Results: Analytical validation of MSK-ACCESS demonstrated 93% accuracy (n=100 variants), 99% precision (n=153 variants), and 100% sensitivity based on an assay limit of detection of 0.5% AF (n=19 variants). Variants were detected down to 0.1% AF. To date, 240 clinical cfDNA and matched normal DNA pairs have been sequenced, analyzed for somatic alterations, and clinically reported to guide patient management. Most clinical cases were from lung (55%) or prostate (13%) cancers and submitted for diagnostic purposes (71%). Median raw coverage was 18,367X, and median consensus coverage was 1411X. Mutations were detected in 180 (75%) samples with a median variant AF of 1.8% (0.02% - 95%). Comparison of concurrent commercial plasma testing results to MSK-ACCESS revealed multiple variants that were of clonal hematopoiesis or germline origin incorrectly reported as somatic variants. In the lung cohort, 48 patients had tissue testing with MSK-IMPACT; among 32 patients with a driver alteration detected by MSK-ACCESS, 91% had the identical driver alteration reported by MSK-IMPACT. Additionally, MSK-ACCESS identified a MET exon 14 alteration in one lung cancer patient that led to protocol enrollment and partial response. Conclusions: Liquid biopsy testing using MSK-ACCESS reliably detected clinically actionable mutations, reducing the need for multiple biopsies. These results also illustrate the importance of analyzing a matched normal sample when interpreting cfDNA results and highlight the potential of using cfDNA analysis to guide treatment selection, assess for treatment response, and identify mechanisms of treatment resistance. This abstract is also being presented as Poster A20. Citation Format: A. Rose Brannon, Gowtham Jayakumaran, Monica Diosdado, Yu Hu, Anna Razumova, Fanli Meng, Emily Lebow, Juber Patel, Ian Johnson, Preethi Srinivasan, Maysun Hasan, Jenna-marie Dix, Aijazuddin Syed, Brian Houck-Loomis, Bob T. Li, Charles Rudin, David Solit, Marc Ladanyi, Maria Arcila, Dana Tsui, Ahmet Zehir, Michael Berger, Ryma Benayed. Validation and clinical implementation of MSK-ACCESS, an ultra-deep sequencing assay for noninvasive somatic mutation profiling [abstract]. In: Proceedings of the AACR Special Conference on Advances in Liquid Biopsies; Jan 13-16, 2020; Miami, FL. Philadelphia (PA): AACR; Clin Cancer Res 2020;26(11_Suppl):Abstract nr PR08.

Abstract PR01: ctDNA shedding dynamics dictate early lung cancer detection potential

Abstract Early cancer detection aims to find tumors before they progress to an incurable stage. Prospective studies with tens of thousands of healthy participants are ongoing to determine whether asymptomatic cancers can be accurately detected by analyzing circulating tumor DNA (ctDNA) from blood samples. We developed a stochastic mathematical model of tumor evolution and ctDNA shedding to investigate the potential and the limitations of ctDNA-based cancer early detection tests. We inferred ctDNA shedding rates of early-stage lung cancers and calculated that a 15-mL blood sample contains on average only 1.5 genome equivalents (GE) of ctDNA for lung tumors with 1 billion cells (concentration of 0.19 GE per plasma mL; tumor fraction of 0.02%). This low level of ctDNA can be explained by the relatively low turnover rate of lung cancer cells. Similarly, fast-growing tumors led to lower levels of ctDNA because a lower number of cell death events decreases the amount of released ctDNA. We designed a detection test of virtual tumors that accounts for the varying plasma DNA concentrations in cancer and cancer-free patients and that incorporates various sources for technical and biologic errors. Two important considerations for disease detection are the expected number of mutations per tumor covered by the sequencing panel and the underlying error rate of the sequencing assay. To determine the potential performance of mutation-based ctDNA detection tests, we computed receiver operating characteristic (ROC) curves and calculated area under the curve (AUC) values. In the case of early relapse detection, we assumed that the sequencing panel covers 20 known clonal mutations per primary tumor with a background error-rate of 1.5 × 10−5 per base pair. For monthly relapse testing with a specificity of 99.5%, we found a median detection size of 0.19 cm3. Moreover, we observed a lead time of 180 days compared to monthly imaging-based relapse detection. While tracking fewer than 20 mutations drastically decreased the expected lead time, tracking more than 20 mutations only minimally increased the expected lead time. Similarly, the increases in lead time gained slowed down for more frequent testing than one month for relapsing tumors with a growth rate of 1%. In the case of cancer screening, we assumed that the sequencing panel has the same error rate as above and covers 2,000 base pairs, resulting in approximately one detectable mutation per lung cancer, which is similar to the CancerSEEK panel. For an annual screening test with a specificity of 99.5%, we found a median lung cancer detection size of 4.4 cm3. This detection size constitutes a more than 80% reduction from the current median detection size of 22.5 cm3 for lung cancers and suggests a potential lead time of 410 days. This mathematical framework provides a mechanistic interpretation of ctDNA-based test results and informs the design of future screening and relapse detection initiatives across cancer types and subpopulations with varying risks. This abstract is also being presented as Poster A01. Citation Format: Stefano Avanzini, David M. Kurtz, Jacob J. Chabon, Sharon S. Hori, Ash A. Alizadeh, Maximilian Diehn, Johannes G. Reiter. ctDNA shedding dynamics dictate early lung cancer detection potential [abstract]. In: Proceedings of the AACR Special Conference on Advances in Liquid Biopsies; Jan 13-16, 2020; Miami, FL. Philadelphia (PA): AACR; Clin Cancer Res 2020;26(11_Suppl):Abstract nr PR01.

UMI-Gen: A UMI-based read simulator for variant calling evaluation in paired-end sequencing NGS libraries

MotivationWith Next Generation Sequencing becoming more affordable every year, NGS technologies asserted themselves as the fastest and most reliable way to detect Single Nucleotide Variants (SNV) and Copy Number Variations (CNV) in cancer patients. These technologies can be used to sequence DNA at very high depths thus allowing to detect abnormalities in tumor cells with very low frequencies. Multiple variant callers are publicly available and are usually efficient at calling out variants. However, when frequencies begin to drop under 1%, the specificity of these tools suffers greatly as true variants at very low frequencies can be easily confused with sequencing or PCR artifacts. The recent use of Unique Molecular Identifiers (UMI) in NGS experiments has offered a way to accurately separate true variants from artifacts. UMI-based variant callers are slowly replacing raw-read based variant callers as the standard method for an accurate detection of variants at very low frequencies. However, benchmarking done in the tools publication are usually realized on real biological data in which real variants are not known, making it difficult to assess their accuracy. ResultsWe present UMI-Gen, a UMI-based read simulator for targeted sequencing paired-end data. UMI-Gen generates reference reads covering the targeted regions at a user customizable depth. After that, using a number of control files, it estimates the background error rate at each position and then modifies the generated reads to mimic real biological data. Finally, it will insert real variants in the reads from a list provided by the user. AvailabilityThe entire pipeline is available at https://gitlab.com/vincent-sater/umigen under MIT license.

Phased Variant Enrichment for Enhanced Minimal Residual Disease Detection from Cell-Free DNA

Background: Circulating tumor DNA (ctDNA) is an emerging biomarker in non-Hodgkin lymphomas (NHLs). Current methods for ctDNA minimal residual disease (MRD) are limited by two factors - low input DNA amounts and high background error rates. VDJ sequencing (i.e., IgHTS) has low background but is limited by low cell-free DNA (cfDNA). Tracking multiple mutations via CAPP-Seq improves sensitivity, but detection is limited by background errors. Clustered mutations have been described in multiple cancers including NHLs and potentially have lower error rates. We explored clustered mutations from whole-genome sequencing (WGS) to identify 'phased variants' (PVs), defined as multiple mutations on a single DNA molecule (Fig 1A). We designed a method to capture PVs for improved ctDNA detection and explored its utility for MRD in DLBCL. Methods: We reanalyzed WGS from 1455 tumors across 11 cancer types. We identified genomic regions recurrently containing PVs and designed an assay for deep cfDNA sequencing. We applied this assay to 171 patients with large B-cell lymphomas. We compared the performance of PVs for disease detection to current ctDNA techniques, including CAPP-Seq and duplex sequencing. Results: To utilize PVs, mutations must occur within a typical cfDNA strand (~170bp). We measured the frequency of putative PVs in WGS, focusing on pairs of mutations occurring within <170bp. PVs were more frequent in NHLs than any other histology (median: DLBCL, 642; FL, 307; Burkitt, 89.5; CLL, 34; breast, 46; lung, colorectal, melanoma, bladder, cervical, head & neck < 10 per case; P < 0.001 for NHLs vs others). PVs in NHLs were enriched in single base substitution mutational signatures associated with activation-induced cytidine deaminase (AID) (SBS84 & 85). PVs in NHLs occurred in stereotyped regions, including canonical AID targets such as IGH, IGK, and IGL, as well as 44 other AID targets (Schmitz, NEJM 2018) (Fig 1B). We additionally identified novel regions not previously implicated as targets of AID, including LPP, XBP1, BZRAP1, and HLA-DQ. We designed an approach for enriching PVs from ~115kb (Phased variant Enrichment Sequencing, PhasE-Seq) and other regions recurrently mutated in B-NHLs. We compared PhasE-Seq and CAPP-Seq using tumor and plasma samples from 16 patients. Compared to CAPP-Seq, PhasE-Seq yielded more SNVs and PVs per case (median SNVs: 331 vs 114, P<0.001; PVs: 729 vs 222.5, P <0.001). We next applied PhasE-Seq to 171 patients with untreated lymphomas (DLBCL, 148; primary mediastinal B-cell lymphoma, PMBCL, 23) profiling 58 tumor and 171 plasma samples with matched germline. We observed significant differences in the distribution of PVs between subtypes - for example, GCB-DLBCL had more PVs in BCL2, MYC, and SGK1, while ABC-DLBCL had more in PIM1 and IGHV4-34 (Fig 1C). Similarly, we noted enrichment in PVs in PMBCL in CIITA, SOCS1, CD83, and ITPKB. We then compared PhasE-Seq to alternative methods for MRD detection. We used limiting dilutions of patient ctDNA down to 1:1,000,000 to establish the detection limit (LOD, Fig 1D). PhasE-Seq outperformed CAPP-Seq and duplex sequencing for recovery of expected tumor content, with a high degree of linearity down to ~1:1,000,000. We applied standard CAPP-Seq and PhasE-Seq to patient cfDNA samples after two cycles of front-line therapy (n=92). We previously reported a 2.5-log reduction in ctDNA as prognostic at this time-point (Kurtz, JCO 2018). Using CAPP-Seq, 58% (53/92) of samples were undetectable. Using PhasE-Seq, 30% (16/53) of samples not detected by CAPP-Seq had evidence of MRD, with levels as low as 2:1,000,000. In patients with ctDNA undetected by CAPP-Seq, detection by PhasE-Seq significantly stratified outcomes (Fig 1E). Conclusions: PVs are frequent in NHLs, likely due to AID, and correlate with disease biology. PhasE-Seq allows for superior detection of ctDNA, including MRD detection in the majority of patients after 2 cycles. Targeted sequencing of ctDNA should consider PVs to maximize detection and guide precision approaches. Figure 1: A) Structure of phased variants B) Distribution of putative PVs from WGS data C) Genomic enrichment in PVs in lymphoma subtypes D) Dilution series comparing PhasE-Seq, CAPP-Seq, and duplex sequencing E) Waterfall plot showing ctDNA level vs outcome; undetectable ctDNA by CAPP-Seq is highlighted F) EFS of patients with undetectable ctDNA by CAPP-Seq after 2 cycles, stratified by PhasE-Seq Disclosures Kurtz: Roche: Consultancy. Diehn:Roche: Consultancy; AstraZeneca: Consultancy; Novartis: Consultancy; BioNTech: Consultancy; Quanticell: Consultancy. Alizadeh:Pharmacyclics: Consultancy; Janssen: Consultancy; Genentech: Consultancy; Roche: Consultancy; Gilead: Consultancy; Celgene: Consultancy; Chugai: Consultancy; Pfizer: Research Funding.

Impact of Polymerase Fidelity on Background Error Rates in Next-Generation Sequencing with Unique Molecular Identifiers/Barcodes

Liquid biopsy and detection of tumor-associated mutations in cell-free circulating DNA often requires the ability to identify single nucleotide variants at allele frequencies below 0.1%. Standard sequencing protocols cannot achieve this level of sensitivity due to background noise from DNA damage and polymerase induced errors. Addition of unique molecular identifiers allows identification and removal of errors responsible for this background noise. Theoretically, high fidelity enzymes will also reduce error rates in barcoded NGS but this has not been thoroughly explored. We evaluated the impact of polymerase fidelity on the magnitude of error reduction at different steps of barcoded NGS library construction. We find that barcoding itself displays largest impact on error reduction, even with low fidelity polymerases. Use of high fidelity polymerases in the barcoding step of library construction further suppresses error in barcoded NGS, and allows detection of variant alleles below 0.1% allele frequency. However, the improvement in error correction is modest and is not directly proportional to polymerase fidelity. Depending on the specific application, other polymerase characteristics such as multiplexing capacity, PCR efficiency, buffer requirements and ability to amplify targets with high GC content may outweigh the relatively small additional decrease in error afforded by ultra-high fidelity polymerases.

SmCounter2: an accurate low-frequency variant caller for targeted sequencing data with unique molecular identifiers.

MotivationLow-frequency DNA mutations are often confounded with technical artifacts from sample preparation and sequencing. With unique molecular identifiers (UMIs), most of the sequencing errors can be corrected. However, errors before UMI tagging, such as DNA polymerase errors during end repair and the first PCR cycle, cannot be corrected with single-strand UMIs and impose fundamental limits to UMI-based variant calling.ResultsWe developed smCounter2, a UMI-based variant caller for targeted sequencing data and an upgrade from the current version of smCounter. Compared to smCounter, smCounter2 features lower detection limit that decreases from 1 to 0.5%, better overall accuracy (particularly in non-coding regions), a consistent threshold that can be applied to both deep and shallow sequencing runs, and easier use via a Docker image and code for read pre-processing. We benchmarked smCounter2 against several state-of-the-art UMI-based variant calling methods using multiple datasets and demonstrated smCounter2’s superior performance in detecting somatic variants. At the core of smCounter2 is a statistical test to determine whether the allele frequency of the putative variant is significantly above the background error rate, which was carefully modeled using an independent dataset. The improved accuracy in non-coding regions was mainly achieved using novel repetitive region filters that were specifically designed for UMI data.Availability and implementationThe entire pipeline is available at https://github.com/qiaseq/qiaseq-dna under MIT license.Supplementary information Supplementary data are available at Bioinformatics online.

Abstract 1286: Analytical validation of an integrated next-generation sequencing pan-cancer liquid biopsy approach for detection of microsatellite instability

Abstract Failure of the DNA mismatch repair (MMR) pathway during DNA replication in cancer leads to the increased accumulation of somatic mutations. The molecular hallmark of MMR deficiency is microsatellite instability (MSI), which presents as polymorphism of tandem nucleotide repeat lengths ubiquitously distributed throughout the genome. Furthermore, the presence of MMR-deficiency or MSI across solid tumors has been recently demonstrated to be a robust biomarker for immunotherapy response with checkpoint inhibition. We have previously developed technology to detect MSI in tumor tissue through hybrid-capture based targeted next-generation sequencing analyses of a 125 gene panel, with probes capturing mononucleotide repeat markers. Using a set of 32 MSI and 27 microsatellite stable (MSS) formalin-fixed paraffin-embedded tumor tissue specimens, we obtained 100% sensitivity and 100% specificity with a limit of detection of 20% tumor purity. The clinical importance of MSI status now requires a fast, objective, highly sensitive screening method, particularly in late-stage patients where tumor material may not be readily obtained. However, to extend this approach to a liquid biopsy panel requires technological advances to overcome the inherent challenges associated with low circulating tumor DNA (ctDNA) levels that are compounded by polymerase slippage in mononucleotide repeat regions during PCR amplification. To overcome these limitations, we applied a multifactorial error correction approach together with a novel peak finding algorithm to more accurately identify the specific mononucleotide sequences in cell-free DNA (cfDNA) analyses. By eliminating a significant majority of sequencing errors and polymerase slippage artifacts, we were able to reduce background error rates by > 90%. Combined with implementation of a distribution modeling and a peak finding algorithm, we were able to accurately sequence the mononucleotide tracts to minimize false discovery rates for cfDNA analyses. To evaluate the performance of this approach in clinical samples, for a cohort of plasma samples obtained from 13 late-stage clinical cases with matched tumor tissue data available (9 MSI and 4 MSS) together with 66 healthy donor plasma samples, we obtained 89% sensitivity and 100% specificity, with a lowest reported mutant allele fraction of 0.4%. Taken together, this integrated next-generation sequencing liquid biopsy approach is valuable for the determination of MSI status to inform immunotherapy treatment. Citation Format: Andrew Georgiadis, Derrick Wood, Derek Murphy, Sonya Parpart-Li, David Riley, Naomi Sengamalay, Marian Novak, Sian Jones, Samuel A. Angiuoli, Mark Sausen. Analytical validation of an integrated next-generation sequencing pan-cancer liquid biopsy approach for detection of microsatellite instability [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 1286.

Prospective Molecular MRD Detection By NGS: A Powerful Independent Predictor for Relapse and Survival in Adults with Newly Diagnosed AML

Introduction:Although the majority of Acute Myeloid Leukemia (AML) patients achieve a complete morphological remission (CR) after induction therapy, relapse rates remain high. Molecular Minimal Residual Disease (MRD) detection by PCR-based technologies has been shown to improve relapse prediction but has been restricted to specific genetically-defined subsets of AML only. Next-Generation Sequencing (NGS) has the advantage that it allows for the assessment of a broad range of disease-related gene mutations in a single assay. Residual leukemia-specific mutations in bone marrow in morphological CR after induction therapy are supposed to represent the source of relapse. However, persistent mutations may also represent clonal hematopoiesis, analogous to age-related clonal hematopoiesis of indeterminate potential (CHIP) present in healthy individuals. It is currently unknown which and to what extent the persisting somatic mutations after induction therapy contribute to AML relapse. Here, we present a comprehensive study, detailing the value of molecular MRD detection by NGS, in a large prospective cohort of newly diagnosed AML.Methods:482 AML patients (<65 years) were treated with 2 cycles of standard induction chemotherapy followed by consolidation in HOVON-SAKK clinical trials (www.hovon.nl). NGS was performed to detect mutations in a panel of 54 genes frequently mutated in myeloid malignancies (Illumina) at diagnosis and in bone marrow in morphological CR after completion of induction therapy. Thompson-Tau outlier testing was performed to reliably detect persisting mutations above background error rates. The primary and secondary endpoints of the study were relapse and overall survival, respectively. To establish and subsequently test our definition of NGS MRD, the cohort was split into a representative training (n=283) and validation cohort (n=147). The Cumulative Incidence of Relapse (CIR) was estimated with competing-risks regression analyses according to the method of Fine & Gray. The Cox proportional hazard model was used to calculate overall survival estimates. P-values <0.05 were considered significant.Results:In 430 out of 482 (89.2%) AML patients somatic driver mutations were present at diagnosis. In 51.4% of subjects persisting mutations were detected in bone marrow in morphological CR at highly variable variant allele frequencies (VAF 0.0002-0.47), predominantly persisting in DNMT3A (78.7%), TET2 (54.2%) and ASXL1 (51.6%). These persistent mutations in DNMT3A, TET2 and ASXL1 (DTA) in the training cohort did not associate with the incidence of relapse at any VAF cut-off, indicating a stage of clonal hematopoiesis rather than a condition of impending relapse. In contrast, in the subset of AML patients with persisting DTA mutations, a significant correlation with relapse was observed when any other persistent non-DTA mutation was considered (training cohort: 5-years CIR 76.4% vs. 39.4%; p=0.002). In the training cohort NGS MRD, as defined by persistent non-DTA mutations, was found to be highly associated with the risk of relapse (SHR:1.85 [95%CI 1.27-2.70]; p=0.001), which was confirmed in the validation set (SHR:2.81 [95%CI 1.64-4.79]; p<0.001) (Fig. 1). In fact, NGS MRD was significantly associated with CIR when the training and validation series were combined (5-years CIR 58.3% versus 33.9% (p<0.001)) (Fig. 2). In addition, NGS MRD predicted for reduced survival in both cohorts (training: HR:1.64 [95%CI 1.12-2.42]; p=0.012 and validation: HR:3.08 [95%CI 1.87-5.08]; p<0.001). Finally, multivariable analysis including the data of all 430 AML patients, with adjustment for age, WBC, ELN2017 risk and number of induction cycles needed to achieve CR, revealed that NGS MRD expresses profound independent prognostic significance for relapse (SHR:1.89 [95%CI:1.34-2.65]; p<0.001) and overall survival (HR:1.64 [95%CI:1.18-2.27]; p=0.003). In sensitivity analysis with time-dependent correction for allogeneic stem cell transplantation NGS MRD remained highly prognostic for relapse and survival.Conclusions:In an unprecedentedly large prospective study including training and validation cohorts, targeted NGS MRD detection is established as a powerful and independent predictor for relapse and survival. NGS MRD is applicable in virtually all newly diagnosed adults with AML while persistent CHIP-related mutations lack prognostic value. [Display omitted] DisclosuresOssenkoppele:J&J: Consultancy, Honoraria; Celgene: Honoraria, Research Funding; Karyopharm: Consultancy, Research Funding; Roche: Honoraria; Novartis: Research Funding.

Abstract 2759: Technical limitations of capture-based, targeted deep sequencing for the detection of somatic variations in cell-free DNA

Abstract Despite widespread applications of capture-based targeted deep sequencing, its limitations on detecting circulating tumor DNA (ctDNA) have not been systematically elucidated. Here, we analyze and characterize parameters critically determining its detection limits; depth of coverage and background errors. The depth increase by using more cell-free DNA (cfDNA) was saturated at around ~3000x with ~50 ng of cfDNA primarily due to a deduplication artifact, which limited the detection sensitivity of the method. Next, characterizing background errors, we uncovered that a majority of errors were not derived from biological origins but technical artifacts, especially during sample preparation and library construction rather than sequencing. Thus, it is important to use an appropriate protocol for those steps minimizing background errors. Considering the depth of coverage and the background error rate, we estimated the detection limit of the method approximately to be 0.4% variant allele frequency at 95% confidence level. Applying the method to pancreatic cancer patients, we found that the levels of ctDNAs measured by cfDNA sequencing well correlated with clinical responses to therapy and/or disease progression, but biopsy-free profiling of somatic variants was limited by the moderate sensitivity of 72% and PPV of 74% at the variant level. Our results clarify technical limitations of targeted deep sequencing and provide a guideline for applying the technique to analyze cfDNA. Citation Format: Gahee Park. Technical limitations of capture-based, targeted deep sequencing for the detection of somatic variations in cell-free DNA [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 2759. doi:10.1158/1538-7445.AM2017-2759

Abstract 2740: High-throughput and sensitive quantification of circulating tumor DNA

Abstract Circulating tumor DNA (ctDNA) has a great potential to serve as non-invasive biomarkers for monitoring disease progression and treatment response. However, its broad clinical applications have been hindered by sensitivity, throughput and complexity of existing ctDNA detection methods. Here we report an adaptation of a multiplex PCR pre-amplification (PreAmp) into the microfluidic multiplex PCR (MMP-Seq) technology for high-throughput and sensitive quantitation of ctDNA. We have demonstrated that PreAmp MMP-Seq enabled sensitive (down to 0.6% allele frequency), reproducible (R2=0.97), accurate and quantitative mutation detection from as low as 1.6 ng DNA input. An Empirical Bayesian Model has been established to estimate background error rates for every base of amplicons in the ctDNA panel using data from 43 normal plasmas. 0.1-0.3% of median error rates were observed across all positions. 149 longitudinal plasma samples from 12 ovarian and 10 pancreatic cancer patients of PhI DMUC5754A anti-MUC16 ADC trials were then profiled using PreAmp MMP-Seq. 79% (11/14) driver mutations found in tissue samples were also identified in matching baseline plasma samples at frequencies well above estimated error rates. Moreover, we found that changes of corresponding ctDNA concentration in post-treatment plasma samples were correlated with changes in tumor burdens and circulating protein marker levels. In conclusion, the MMP-seq PreAmp workflow provides an inexpensive, reproducible and flexible solution for mutation detection in plasma DNA samples. Citation Format: Yinghui Guan, Oleg Mayba, Shan Lu, Thomas Sandmann, Younjeong Choi, Walter Darbonne, Vincent Leveque, Lisa Ryner, Eric Humke, Nga W. Tam, Sundari Sujathasarma, Anna Cheung, Richard Bourgon, Yulei Wang. High-throughput and sensitive quantification of circulating tumor DNA [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 2740. doi:10.1158/1538-7445.AM2017-2740

Abstract 2071: Sub-clonal heterogeneity and risk of treatment failure in Philadelphia-positive leukemias

Abstract Molecularly targeted therapies have been exceptionally successful in the treatment of some malignancies, most notably chronic myeloid leukemia (CML). However in other cancers, e.g. Philadelphia-positive acute lymphoblastic leukemia (Ph+ ALL), drug resistance is a frequent occurrence. Resistance mutations may pre-exist as a consequence of intra-tumor heterogeneity; thus the extent of heterogeneity may determine the likelihood of treatment failure. Intra-tumor heterogeneity is challenging to quantitate, as heterogeneous mutations are likely to occur below the background error rate of conventional approaches for mutation detection. To overcome this limitation and explore the relationship between intra-tumor heterogeneity and drug resistance in Ph+ leukemias, we utilized single-molecule Duplex Sequencing (DS) of the ABL1 gene. DS eliminates sequencing errors by independently tagging the two strands of individual DNA molecules; true mutations are present at the same position in both strands, while PCR and sequencing errors are only present in one strand. DS has an error rate below 0.00001%, and thus enables quantitation of heterogeneous mutations with unprecedented resolution. We find that CP-CML, a disease that is often well controlled with targeted therapy, has an extremely low burden of intra-tumor heterogeneity at the time of diagnosis, with a sub-clonal mutation burden similar to that of normal individuals. Simulations which take into account the low cancer stem cell fraction of CP-CML and the sub-clonal mutation burden we measured reveal that only a minority of CP-CML patients will harbor pre-existing drug resistance, which may explain the unusual success of targeted therapy in this setting. Next, we studied patients with advanced Ph+ leukemias (refractory BP-CML and Ph+ ALL), which have a high rate of treatment failure and poor outcomes. We find that patients with BP-CML and Ph+ALL have a significant elevation in sub-clonal heterogeneity relative to those with CP-CML. We demonstrate that the higher extent of intra-tumor heterogeneity correlates with a significantly higher rate of drug resistance mutations and compound resistance (i.e. two concurrent resistance mutations) in patients. Unlike CP-CML, pre-existing mutations conferring resistance to targeted therapy are nearly certain in refractory BP-CML and Ph+ ALL as a consequence of the load of intra-tumor heterogeneity and high cancer stem cell fraction. We support this concept by three distinct approaches: (i) detection of pre-existing sub-clones that drive treatment failure; (ii) modeling the growth kinetics of resistant clones; and (iii) extrapolation of the mutation burden measured by DS. Our results suggest that intra-tumor heterogeneity influences the likelihood of resistance to molecularly targeted therapy. Quantitation of sub-clonal mutation load may be broadly applicable in other malignancies for predicting the likelihood of response to targeted therapy and monitoring disease progression. Citation Format: Michael W. Schmitt, Justin R. Pritchard, Lan Beppu, J. Graeme Hodgson, Victor M. Rivera, Lawrence A. Loeb, Jerald P. Radich. Sub-clonal heterogeneity and risk of treatment failure in Philadelphia-positive leukemias [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 2071. doi:10.1158/1538-7445.AM2017-2071