Methods and systems for adjuvant therapy

Abstract

Provided herein are methods for identifying a subject for an adjuvant therapy by detecting MRD. In some embodiments, detecting MRD comprises identifying circulating tumor DNA in cell-free DNA from a biological sample of the subject. In some embodiments, it includes methods for identifying a subject as a candidate for adjuvant therapy, the method comprising: a) isolating cell-free DNA (cfDNA) from a biological sample obtained from the subject; detecting minimal residual disease (MRD) of the subject by assessing presence of circulating tumor DNA (ctDNA) molecules in the cfDNA; and identifying the subject as a candidate for adjuvant therapy if MRD is detected, wherein the subject has previously undergone curative-intent resection of a cancer.

Description

METHODS AND SYSTEMS FOR ADJUVANT THERAPYCROSS-REFERENCE

[0001] This application claims priority to U.S. Provisional Patent Application No.63 / 693,985, filed September 12, 2024, U.S. Provisional Patent Application No. 63 / 691,926, filed September 6, 2024, and U.S. Provisional Patent Application No. 63 / 650,776, filed May 22, 2024, which are each incorporated herein by reference in their entireties.BACKGROUND

[0002] The detection of circulating tumor DNA (ctDNA) minimal residual disease (MRD) after curative-intent therapy for early-stage non-small cell lung cancer (NSCLC) portends an exceptionally high risk of recurrence, ranging from -80% to 100%. Although the positive predictive value of tumor-informed first generation ctDNA MRD assays is high, the clinical sensitivity of these assays is suboptimal, and the majority of patients destined to recur are ctDNA MRD negative after completion of standard-of-care treatment. This high false negative rate has hampered clinical implementation of ctDNA MRD testing in early stage NSCLC, since patients with negative results cannot be assumed to be cured. This shortcoming has also made it challenging to perform clinical trials in which ctDNA MRD is an integral biomarker used to select patients for therapeutic escalation. Improved methods and systems are needed to detect MRD after curative-intent therapy or resection.SUMMARY

[0003] In some aspects, provided herein are methods for identifying a subject as a candidate for adjuvant therapy, the method comprising: a) isolating cell-free DNA (cfDNA) from a biological sample obtained from the subject; detecting minimal residual disease (MRD) of the subject by assessing presence of circulating tumor DNA (ctDNA) molecules in the cfDNA; and identifying the subject as a candidate for adjuvant therapy if MRD is detected, wherein the subject has previously undergone curative-intent resection of a cancer, and wherein the cancer is lung cancer, bladder cancer, or breast cancer. In some embodiments, assessing the presence of ctDNA comprises assaying the cfDNA for single nucleotide variants (SNVs). In some embodiments, the subject has a history of tobacco exposure. In some embodiments, the SNVs are enriched for SBS4 variants. In some embodiments, the subject does not have a history of tobacco exposure. In some embodiments, the SNVs are enriched for SB SI variants. In some embodiments, assessing the presence of ctDNA comprises assaying the cfDNA for phased variants (PVs). In some embodiments, the subject has received an earlier adjuvant therapy.

[0004] In some aspects, provided herein are methods for identifying a subject as a candidate for adjuvant therapy, the method comprising: a) isolating cell-free DNA (cfDNA) from a biological sample obtained from the subject; detecting minimal residual disease (MRD) of the subject by assessing presence of circulating tumor DNA (ctDNA) molecules in the cfDNA by assaying the cfDNA for phased variants (PVs); and identifying the subject as a candidate for adjuvant therapy if MRD is detected, wherein the subject has previously undergone curativeintent resection of a cancer. In some embodiments, the cancer is lung cancer, bladder cancer, or breast cancer.

[0005] In some embodiments, the cancer is lung cancer. In some embodiments, the lung cancer is non-small cell lung cancer.

[0006] In some embodiments, the subject has a history of tobacco exposure. In some embodiments, PVs or SNVs are enriched for SBS4 variants.

[0007] In some embodiments, the subject does not have a history of tobacco exposure. In some embodiments, the PVs are enriched for SBS2 variants.

[0008] In some embodiments, the PVs have a background error rate less than about 50 ppm.

[0009] In some embodiments, the method has a median LOD95 below 10 ppm.

[0010] In some embodiments, the biological sample is collected between about 2 and about 8 weeks post-resection.

[0011] In some embodiments, the subject received neo-adjuvant systemic therapy prior to resection.

[0012] In some embodiment, the MRD is at a concentration of about 0.19 to about 22,000 ppm. In some embodiments, the MRD is at a concentration of about 0.19 to about 5 ppm.

[0013] In some embodiments, the adjuvant therapy is consolidation radiotherapy, adjuvant chemotherapy, adjuvant immunotherapy, adjuvant hormonal therapy, or adjuvant targeted therapy.

[0014] In some embodiments, prior to detecting MRD, the method comprises a longitudinal testing step of the subject. In some embodiments, the longitudinal testing occurs every 3-6 months. In some embodiments, the subject is identified as a candidate for adjuvant therapy if MRD becomes detectable.

[0015] In some embodiments, the biological sample is a blood sample.

[0016] In some embodiments, assessing presence of ctDNA molecules in the cfDNA comprises whole genome sequencing. In some embodiments, the method further comprises targeted enrichment of patient-specific sets of genomic regions that were determined to harbor PVs in the resected cancer.

[0017] In some embodiments, the subject has received an earlier adjuvant therapy.

[0018] In aspects, provided herein is a method of treating cancer with an adjuvant therapy, the method comprising: identifying a subject as a candidate for adjuvant therapy according to any one of the methods described herein; and treating the subject with an adjuvant therapy selected from consolidation radiotherapy, adjuvant chemotherapy, adjuvant immunotherapy, adjuvant hormonal therapy, or adjuvant targeted therapy.INCORPORATION BY REFERENCE

[0019] All publications, patents, and patent applications mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent, or patent application was specifically and individually indicated to be incorporated by reference.BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Various features of the disclosure are set forth with particularity in the appended claims. A better understanding of the features and advantages of the present disclosure will be obtained by reference to the following detailed description that sets forth illustrative embodiments, in which the principles of the disclosure are utilized, and the accompanying drawings of which:

[0021] FIG. 1 depicts modeling of LOD95 requirements for MRD detection in early stage NSCLC. FIG. 1A and IB show representative exponential (FIG. 1A) and log-linear model (FIG. IB) for change in ctDNA variant allele fraction (VAF) over time for patient CRUK0762 from the TRACERx study. FIG. 1C shows a histogram of the distribution of ctDNA doubling times for the 28 patients included in the modeling analyses. Doubling times were calculated for each patient from the slopes of the log-linear models. FIG. ID shows ctDNA doubling times stratified by stage. FIG. IE shows ctDNA doubling times stratified by histologic type. FIG. IF shows projected clinical sensitivity for detection of MRD at the post-surgery landmark in patients destined to develop recurrence using the patient-specific log-linear models for hypothetical assays with LOD95 ranging from 10'4to 10'6.

[0022] FIG. 2A is a study schematic. FIG. 2B shows patient characteristics and tumor mutations. Each column represents an individual patient. Color bars depict key patient characteristics. Recurrently mutated genes of interest are shown.

[0023] FIG. 3 depicts association of PVs and SNVs with different mutagenic processes in smokers versus never-smokers. FIG. 3A shows mean VAFs of PVs in pre-neoadjuvant therapy biopsy specimens and post-neoadjuvant surgical resection specimens for 5 patients. Two technical replicates were performed using independent sections from FFPE blocks for the biopsy specimens. FIG. 3B shows a volcano plot depicting enrichment of mutational signatures in PVs versus SNVs across the entire cohort. FIG. 3C shows a heatmap depicting case-level enrichmentof mutational signatures in PVs and SNVs in current or former smokers and never-smokers.FIG. 3D shows column scatter plots depicting mutational signatures differentially enriched in PVs and SNVs from current or former smokers and never-smokers.

[0024] FIG. 4 depicts an analysis of pre-treatment ctDNA detection. FIG. 4A shows background error rates for SNVs and PVs. FIG. 4B shows LOD95 for all plasma samples using START. FIG. 4C shows LOD95 for all plasma samples using the SNV-based MRD assay. FIG. 4D shows pre-treatment sensitivity for MRD detection by stage and histologic type for START and the SNV-based MRD assay applied to the same plasma samples. FIG. 4E shows mean VAF in plasma samples with detectable ctDNA using START. FIG. 4F shows pre-treatment mean VAF detected using START stratified by stage and tumor histologic type. FIG. 4G shows a scatter plot depicting the relationship between mean VAF and radiologically-assessed tumor volumes for plasma samples analyzed in 2 previous studies (Abbosh, C. et al., Nature 616, 553- 562 (2023); Chabon, J. J. et al., Nature 580, 245-251 (2020)) and in the current study. Median tumor volumes for each cohort are listed in the legend. FIG. 4H shows a cross-cohort comparison of pre-treatment sensitivity for MRD detection by stage and histologic type for the current study and several previously published studies focused on early stage NSCLC employing first generation MRD assays (Abbosh, C. et al., Nature 545, 446-451 (2017); Gale, D. et al., Ann Oncol 33, 500-510 (2022); Abbosh, C. et al., Nature 616, 553-562 (2023); Chabon, J. J. et al., Nature 580, 245-251 (2020))Wilcoxon rank-sum test; ns, P > 0.05; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

[0025] FIG. 5 depicts an analysis of ctDNA MRD detection at the post-treatment landmark. FIG. 5A shows the percent of patients with detectable or undetectable ctDNA MRD at the posttreatment landmark who did or did not develop recurrence. FIG. 5B shows sensitivity (Sn), specificity (Sp), positive predictive value (PPV), and negative predictive value (NPV) for detecting MRD in patients destined to develop recurrence using START or the SNV-based MRD assay. FIG. 5C shows a Kaplan Meier analysis of freedom from recurrence on the basis of the ctDNA MRD status at the post-treatment landmark using the SNV-based MRD assay. FIG. 5D shows a Kaplan Meier analysis of freedom from recurrence on the basis of the ctDNA MRD status at the post-treatment landmark using START. FIG. 5E shows mean VAF at the posttreatment landmark using START in patients who did or did not develop recurrence. Samples with detectable ctDNA MRD using the SNV-based assay are colored blue. FIG. 5F shows the percentage of post-operative plasma samples that were ctDNA MRD positive using the START assay in patients who were destined to recur, based on the time between plasma collection and recurrence. FIG. 5G shows a Kaplan-Meier analysis of patients with undetectable ctDNA MRDpostoperatively by START, stratified by whether they subsequently received adjuvant therapy (systemic therapy and / or radiotherapy). FIG. 5H as in FIG. 5G, but for MRD-positive patients.

[0026] FIG. 6 depicts representative patient vignettes. FIG. 6A shows a vignette for patient in whom ctDNA MRD that was detected postoperatively by START, but not by the SNV-based MRD assay, was cleared after radiotherapy. FIG. 6B shows a vignette for patient in whom ctDNA MRD detected postoperatively by START but not the SNV-based MRD assay was cleared after radiotherapy. FIG. 6C shows a vignette for patient in whom ctDNA MRD that was detected postoperatively by START, but not by the SNV-based MRD assay, was cleared after adjuvant chemotherapy. FIG. 6D shows a vignette for patient treated with neoadjuvant immunotherapy in whom ctDNA MRD was detected postoperatively by START, but not by the SNV-based MRD assay, and who developed a biopsy-proven right adrenal recurrence. Bx, biopsy; Carbo, carboplatin; Cis, cisplatin; doc, docetaxel; Ipi, ipilimumab; nivo, nivolumab; pac, paclitaxel; PORT, postoperative radiotherapy.

[0027] FIG. 7 depicts postoperative ctDNA levels over time for patients with NSCLC in the TRACERx study. Log-linear model for change in ctDNA variant allele fraction (VAF) over time for all patients included in the mathematical modeling analysis of ctDNA MRD dynamics.

[0028] FIG. 8 depicts the correlation between ctDNA concentrations determined using the START assay and the SNV-based approach. Scatter plot showing ctDNA VAF for all plasma samples using the two assays.

[0029] FIG. 9 depicts disease-free and overall survival based on detection of ctDNA MRD at the post-treatment landmark. FIG. 9A shows Kaplan Meier analysis of disease-free survival based on ctDNA MRD status at the post-treatment landmark using the SNV-based MRD assay. FIG. 9B, as in FIG. 9A, but using START. FIG. 9C shows Kaplan Meier analysis of overall survival based on ctDNA MRD status at the post-treatment landmark using the SNV-based MRD assay. FIG. 9D, as in FIG. 9C but using START, also known as PhasED-Seq.

[0030] FIG. 10 shows an overview of the patient cohort.

[0031] FIG. 11 is a table that provides characteristics of an early-stage breast cancer cohort.

[0032] FIG. 12 shows disease-free survival stratified by MRD status post-surgery.

[0033] FIGS. 13A and 13B show disease-free survival stratified by MRD clearance after adjuvant therapy. Specifically, FIG. 13A shows Kaplan-Meier analysis of disease-free survival based on ctDNA MRD status for patients with detectable ctDNA post-surgery. FIG. 13B shows a swimmer plot showing the 8 patients who had detectable ctDNA post-surgery and the details of the therapy and timing of therapy received.

[0034] FIG. 14 shows disease-free survival stratified by pre-treatment ctDNA detection. No recurrence was observed in patients who had undetectable ctDNA at baseline.

[0035] FIG. 15 shows an overview of the MRD testing process for the breast cancer cohort using START.

[0036] FIG. 16 shows a graph demonstrating the limit of detection of the MRD assay. The data points depict the observed fraction as a function of the expected fraction of three limiting dilution series of cfDNA from patients with limited-stage non-small cell lung cancer diluted into healthy donor background cfDNA. Replicates were performed in triplicate from the expected tumor fraction of 1 part in 107to 1 part in 2xlO7Additional negative samples were assessed for specificity and the LOD95 was determined by Probit modeling.

[0037] FIGS. 17A and 17B show ctDNA levels at landmark time points in ctDNA + early- stage breast cancer patients. The dashed line in both charts indicates the approximate LOD for first generation assays. FIG. 17A shows treatment landmarks related to NACT, with patients grouped by pathological response. Patients who eventually relapse are indicated by red circles. FIG. 17B shows treatment landmarks related to surgery, with patients grouped by eventual relapse.

[0038] FIGS. 18A, 18B, 18C, 18D, and 18E show the distribution of ctDNA levels for the early-stage breast cancer patient cohort before treatment based on different characteristics. FIG. 18A shows the distribution of ctDNA levels by subtype. FIG. 18B shows the distribution of ctDNA levels by stage. FIG. 18C shows the distribution of ctDNA levels by T status. FIG. 18D shows the ctDNA distribution by nodal status. FIG. 18E shows the ctDNA distribution by Grade.

[0039] FIGS. 19A, 19B, 19C, 19D, and 19E show individual patient vignettes for the early- stage breast cancer patient cohort. FIG. 19A shows the ctDNA levels over time for a patient that did not clear ctDNA after treatment. FIG. 19B shows the ctDNA levels over time for a patient with undetectable ctDNA after neo-adjuvant chemotherapy. FIG. 19C shows the ctDNA levels over time for a patient with undetectable ctDNA levels after surgery. FIG. 19D shows the ctDNA levels over time for a patient with undetectable ctDNA levels after adjuvant chemotherapy. FIG. 19E shows the ctDNA levels over time for a patient with undetectable ctDNA levels after endocrine therapy.

[0040] FIG. 20 shows swimmer plots for patients from the early-stage breast cancer cohort organized by subtype. Each line represents an individual patient with details regarding treatment landmarks and ctDNA status given by the shape and color of markers placed on each line relative to time since surgery. The vertical gray line across all patient traces indicates surgery. The placement of each blue or red circle indicates the timing of extraction of blood samples used to analyze ctDNA status relative to other landmark events. Blue circles indicate undetected ctDNA using START whereas red circles indicated detected ctDNA using START. Therapies areindicated by rectangles. The color of the rectangle indicates the type of therapy received and the length of the rectangle represents the relative duration. Gray circles indicate the absence of a distant relapse whereas purple triangles indicate the presence of a distant relapse.DETAILED DESCRIPTION

[0041] In some aspects, provided herein are methods for identifying a subject as a candidate for adjuvant therapy. In some embodiments, the method comprises isolating cell-free DNA (cfDNA) from a biological sample obtained from the subject; detecting minimal residual disease (MRD) of the subject by assessing presence of circulating tumor DNA (ctDNA) molecules in the cfDNA; and identifying the subject as a candidate for adjuvant therapy if MRD is detected. In some embodiments, the subject has previously undergone curative-intent resection of a cancer. In some embodiments, the cancer is lung cancer, bladder cancer, or breast cancer. In some embodiments, assessing the presence of ctDNA comprises assaying the cfDNA for single nucleotide variants (SNVs). In some embodiments, assessing the presence of ctDNA comprises assaying the cfDNA for phased variants (PVs).

[0042] In some aspects, provided herein are methods for identifying a subject as a candidate for adjuvant therapy. In some embodiments, the method comprises isolating cell-free DNA (cfDNA) from a biological sample obtained from the subject; detecting minimal residual disease (MRD) of the subject by assessing presence of circulating tumor DNA (ctDNA) molecules in the cfDNA by assaying the cfDNA for phased variants (PVs); and identifying the subject as a candidate for adjuvant therapy if MRD is detected. In some embodiments, the subject has previously undergone curative-intent resection of a cancer.

[0043] In some embodiments, the cancer is lung cancer, bladder cancer, or breast cancer. In some embodiments, the lung cancer is non-small cell lung cancer (NSCLC).

[0044] In some embodiments, the PVs or SNVs are enriched for a single-base substation (SBS) mutational signature. In some embodiments, wherein the subject has a history of tobacco exposure, the SBS mutational signature is an SBS4 signature. In some embodiments, PVs are enriched for SBS4 variants in subjects with a history of tobacco exposure. In some embodiments, SNVs are enriched for SBS4 variants in subjects with a history of tobacco exposure. In some embodiments, wherein the subject does not have a history of tobacco exposure, the SBS mutational signature is an SBS2 or SBS1 signature. In some embodiments, PVs are enriched for SBS2 variants in subjects that do not have a history of tobacco exposure. In some embodiments, SNVs are enriched for SBS1 variants in subjects that do not have a history of tobacco exposure.

[0045] The term “background error rate” refers to the rate of detecting false positive molecules. In some embodiments, wherein detecting minimal residual disease (MRD) of thesubject comprises assessing presence of circulating tumor DNA (ctDNA) molecules in the cfDNA by assaying the cfDNA for phased variants (PVs), the PVs have a background error rate less than about 50 ppm, e.g., less than about 50 ppm, less than about 45 ppm, less than about 40 ppm, less than about 35 ppm, less than about 30 ppm, less than about 25 ppm, less than about 20 ppm, less than about 15 ppm, less than about 10 ppm, less than about 5 ppm, less than about 1 ppm, less than about 0.5 ppm. In some embodiments, the background error rate is about 0.014 ppm.

[0046] In some of any of the embodiments herein, 95% limit of detection (LOD95) is less than about 1 out of 100,000, less than about 1 out of 500,000, less than about 1 out of 1,000,000, less than about 1 out of 1,500,000, or less than about 1 out of 2,000,000 observations from the sequencing data. In some embodiments, the LOD95 is less than about 10 ppm, 5 ppm, 1 ppm, or less than about 0.5 ppm. In some embodiments, the LOD95 is about 1.4 ppm.

[0047] In some embodiments, MRD is detected at a concentration of about 0.19 ppm to about 22,000 ppm. In some embodiments, MRD is detected at a concentration of about 0.19 to about 0.5 ppm, about 0.5 to about 1 ppm, about 1 ppm to about 5 ppm, about 5 ppm to about 10 ppm, about 10 ppm to about 20 ppm, about 20 ppm to about 30 ppm, about 30 ppm to about 40 ppm, about 40 ppm to about 50 ppm, about 50 ppm to about 100 ppm, about 100 ppm to about 200 ppm, about 200 ppm to about 300 ppm, about 300 ppm to about 400 ppm, about 400 ppm to about 500 ppm, about 500 ppm to about 1000 ppm, about 1000 ppm to about 2000 ppm, about 2000 ppm to about 5000 ppm, about 5000 ppm to about 10000 ppm, or about 10000 ppm to about 22000 ppm. In some embodiments, MRD is detected at a concentration of about 0.19 ppm to about 5 ppm.

[0048] In some embodiments, the biological sample is collected between about 2 and about 8 weeks after curative-intent resection of a cancer. In some embodiments, the biological sample is collected about 2 to about 3 weeks after resection of the cancer. In some embodiments, the biological sample is collected about 3 to about 4 weeks after resection of the cancer. In some embodiments, the biological sample is collected about 4 to about 5 weeks after resection of the cancer. In some embodiments, the biological sample is collected about 5 to about 6 weeks after resection of the cancer. In some embodiments, the biological sample is collected about 6 to about 7 weeks after resection of the cancer. In some embodiments, the biological sample is collected about 7 to about 8 weeks after resection of the cancer. In some embodiments, the biological sample is collected about 2 weeks after resection of the cancer. In some embodiments, the biological sample is collected about 3 weeks after resection of the cancer. In some embodiments, the biological sample is collected about 4 weeks after resection of the cancer. In some embodiments, the biological sample is collected about 5 weeks after resection of the cancer. Insome embodiments, the biological sample is collected about 6 weeks after resection of the cancer. In some embodiments, the biological sample is collected about 7 weeks after resection of the cancer. In some embodiments, the biological sample is collected about 8 weeks after resection of the cancer.

[0049] The term “biological sample” or “bodily sample,” as used interchangeably herein, generally refers to a tissue or fluid sample derived from a subject. A biological sample can be directly obtained from the subject. Alternatively, a biological sample can be derived from the subject (e.g., by processing an initial biological sample obtained from the subject). The biological sample can be or can include one or more nucleic acid molecules, such as DNA or ribonucleic acid (RNA) molecules. The biological sample can be derived from any organ, tissue or biological fluid. A biological sample can comprise, for example, a bodily fluid or a solid tissue sample. An example of a solid tissue sample is a tumor sample, e.g., from a solid tumor biopsy. Non-limiting examples of bodily fluids include blood, serum, plasma, tumor cells, saliva, urine, cerebrospinal fluid, lymphatic fluid, prostatic fluid, seminal fluid, milk, sputum, stool, tears, and derivatives of these. In some cases, one or more cell-free nucleic acid molecules as disclosed herein can be derived from a biological sample, such as from serum or plasma sample.

[0050] In some embodiments, after identifying the subject as a candidate for adjuvant therapy, the subject is administered an adjuvant therapy. In some embodiments, the adjuvant therapy comprises adjuvant chemotherapy, adjuvant radiotherapy, adjuvant chemoradiotherapy, adjuvant immunotherapy, adoptive cell therapy, adjuvant hormone therapy, and / or adjuvant targeted drug therapy. In some embodiments, the adjuvant therapy is consolidation radiotherapy.

[0051] In some embodiments, prior to curative-intent resection of the cancer, the subject received neo-adjuvant systemic therapy. In some embodiments, the neo-adjuvant systemic therapy comprises chemotherapy, radiotherapy, chemoradiotherapy, immunotherapy, adoptive cell therapy, hormone therapy, and / or targeted drug therapy. In some embodiments, the subject has received one or more cycles of the neo-adjuvant therapy.

[0052] In some embodiments, prior to detecting MRD, the method comprises a longitudinal testing step. In some embodiments, longitudinal testing comprises isolating cfDNA from a biological sample obtained from the subject and assessing for the presence of ctDNA molecules in the cfDNA. In some embodiments, longitudinal testing occurs every 3 to 6 months. In some embodiments, longitudinal testing occurs every 3 to 6 months until ctDNA is detected in the cfDNA. In some embodiments, the method comprises the method comprises isolating cfDNA from a biological sample obtained from the subject; assessing the cfDNA for the presence of ctDNA; and repeating the isolating and assessing steps every 3-6 months. In some embodiments, the method comprises isolating cfDNA from a biological sample obtained from the subject;assessing the cfDNA for the presence of ctDNA; repeating the assessing step every 3-6 months until MRD is detected; and upon detecting MRD, identifying the subject as a candidate for adjuvant therapy. In some embodiments, the subject has previously undergone curative-intent resection of a cancer. In some embodiments, the cancer is lung cancer, bladder cancer, or breast cancer. In some embodiments, assessing the presence of ctDNA comprises assaying the cfDNA for single nucleotide variants (SNVs). In some embodiments, assessing the presence of ctDNA comprises assaying the cfDNA for phased variants (PVs).

[0053] In some embodiments, assessing the presence of ctDNA molecules in the cfDNA comprises sequencing the cfDNA. Sequencing methods include first-generation sequencing methods (e.g., Maxam-Gilbert sequencing, Sanger sequencing). The sequencing method can be a high-throughput sequencing method, such as next-generation sequencing (NGS) (e.g., sequencing by synthesis). A high-throughput sequencing method can sequence simultaneously (or substantially simultaneously) at least about 10,000, at least about 100,000, at least about 1 million, at least about 10 million, at least about 100 million, at least about 1 billion, or more polynucleotide molecules (e.g., cell-free nucleic acid molecules or derivatives thereof). NGS can be any generation number of sequencing technologies (e.g., second-generation sequencing technologies, third-generation sequencing technologies, fourth-generation sequencing technologies, etc.). Non-limiting examples of high-throughput sequencing methods include massively parallel signature sequencing, polony sequencing, pyrosequencing, sequencing-by- synthesis, combinatorial probe anchor synthesis (cPAS), sequencing-by-ligation (e.g., sequencing by oligonucleotide ligation and detection (SOLiD) sequencing), semiconductor sequencing (e.g., Ion Torrent semiconductor sequencing), DNA nanoball sequencing, singlemolecule sequencing, and sequencing-by-hybridization.

[0054] In some embodiments of any one of the methods disclosed herein, the sequences of the cfDNA can be obtained based on any of the disclosed sequencing methods that utilizes nucleic acid amplification (e.g., polymerase chain reaction (PCR)). Non-limiting examples of such sequencing methods can include 454 pyrosequencing, polony sequencing, and SoLiD sequencing. In some cases, amplicons (e.g., derivatives of the plurality of cell-free nucleic acid molecules that is obtained or derived from the subject, as disclosed herein) that correspond to a genomic region of interest (e.g., a genomic region associated with a disease) can be generated by PCR, optionally pooled, and subsequently sequenced to generating sequencing data. In some examples, because the regions of interest are amplified into amplicons by PCR before being sequenced, the nucleic acid sample is already enriched for the region of interest, and thus any additional pooling (e.g., hybridization) may not and need not be needed prior to sequencing (e.g., non-hybridization based NGS). Alternatively, pooling via hybridization can further be performedfor additional enrichment prior to sequencing. Alternatively, the sequencing data can be obtained without generating PCR copies, e.g., via cPAS sequencing.

[0055] In some embodiments, provided herein is a method of treating cancer with an adjuvant therapy, the method comprising: identifying a subject as a candidate for adjuvant therapy according to any one of the methods described herein; and treating the subject with an adjuvant therapy selected from consolidation radiotherapy, adjuvant chemotherapy, adjuvant immunotherapy, adjuvant hormonal therapy, or adjuvant targeted therapy.

[0056] The term “about” means a range of up to 10% of a given value.

[0057] The term “phased variants,” “variants in phase,” “PV,” or “somatic variants in phase,” as used interchangeably herein, generally refers to two or more mutations (e.g., SNVs or indels) that occur in cis (i.e., on the same strand of a nucleic acid molecule) within a single cell- free nucleic acid molecule. In some cases, a cell-free nucleic acid molecule can be a cell-free deoxyribonucleic acid (cfDNA) molecule. In some cases, a cfDNA molecule can be derived from a diseased tissue, such as a tumor (e.g., a circulating tumor DNA (ctDNA) molecule). Examples of changes in nucleic acid sequence relative to a reference genomic sequence (e.g., a sequence derived from one of more healthy cells or a consensus sequence from a cohort) can include: a somatic single nucleotide variant (SNV), a somatic indel, a somatic translocation breakpoint, a somatic amplification or deletion breakpoint (e.g., the boundary of a large genomic copy number alteration, such as a large-scale deletion or a large-scale amplification), a germline SNV, a germline indel, a germline translocation breakpoint, a germline amplification or deletion breakpoint, or a region of localized hypermutation (kataegis). In some cases, phased variants may occur in cis (i.e., on the same strand of a nucleic acid molecule) within a single molecule, such as a single cell-free nucleic acid molecule. In some cases, a cell-free nucleic acid molecule can be a cell-free deoxyribonucleic acid (cfDNA) molecule. In some cases, a cfDNA molecule can be derived from a diseased tissue, such as a tumor (e.g., a circulating tumor DNA (ctDNA) molecule). In some cases, the cell-free nucleic acid molecule can be a cell-free ribonucleic acid molecule. The term “phased variant” can refer to one of the plurality of variants that are required to occur in proximity to one another to constitute phased variants, while the term “set of phased variants,” as used in the claims, can refer to the plurality of variants that together form phased variants (i.e., the variants that are within 170 bp of each other with respect to the reference genome, occurring on the same strand of DNA).

[0058] The term “biological sample” or “bodily sample,” as used interchangeably herein, generally refers to a tissue or fluid sample derived from a subject. A biological sample can be directly obtained from the subject. Alternatively, a biological sample can be derived from the subject (e.g., by processing an initial biological sample obtained from the subject). Thebiological sample can be or can include one or more nucleic acid molecules, such as DNA or ribonucleic acid (RNA) molecules. The biological sample can be derived from any organ, tissue or biological fluid. A biological sample can comprise, for example, a bodily fluid or a solid tissue sample. An example of a solid tissue sample is a tumor sample, e.g., from a solid tumor biopsy. Non-limiting examples of bodily fluids include blood, serum, plasma, tumor cells, saliva, urine, cerebrospinal fluid, lymphatic fluid, prostatic fluid, seminal fluid, milk, sputum, stool, tears, and derivatives of these. In some cases, one or more cell-free nucleic acid molecules as disclosed herein can be derived from a biological sample, such as from serum or plasma.

[0059] The term “subject,” as used herein, generally refers to any animal, mammal, or human. A subject can have, potentially have, or be suspected of having one or more conditions, such as a cancer. In some examples, the subject is a human.

[0060] The term “cell-free DNA” or “cfDNA,” as used interchangeably herein, generally refers to DNA fragments circulating freely in a blood stream of a subject. Cell-free DNA fragments can have dinucleosomal protection (e.g., a fragment size of at least 240 base pairs (“bp”)). These cfDNA fragments with dinucleosomal protection were likely not cut between the nucleosome, resulting in a longer fragment length (e.g., with a typical size distribution centered around 334 bp). Cell-free DNA fragments can have mononucleosomal protection (e.g., a fragment size of less than 240 base pairs (“bp”)). These cfDNA fragments with mononucleosomal protection were likely cut between the nucleosome, resulting in a shorter fragment length (e.g., with a typical size distribution centered around 167 bp).EXAMPLES

[0061] The following examples are provided to further illustrate some embodiments of the present disclosure, but are not intended to limit the scope of the disclosure; it will be understood by their exemplary nature that other procedures, methodologies, or techniques known to those skilled in the art may alternatively be used.Example 1: Modeling LOD95 Requirements for MRD Detection in Early Stage NSCLC

[0062] To investigate the expected clinical impact of improvements in LOD95 on MRD assay performance, we built mathematical models of ctDNA dynamics after curative-intent treatment of early-stage NSCLC using ctDNA surveillance data from the TRACERx study (Abbosh, C. et al., Nature 616, 553-562 (2023)). This TRACERx study employed a tumor- informed first generation ctDNA MRD assay that tracked SNVs with an LOD95 of 80 ppm (0.008%). Among 70 patients with disease recurrence, we identified 38 patients in whom no intervening therapy was given between consecutive postoperatively detectable ctDNA MRD measurements, including a subset of 23 (33%) patients with 3 or more consecutive positivesamples, and another subset of 15 (21%) patients with two consecutive positive samples. Analysis of ctDNA concentrations in patients with 3 or more plasma samples who did not receive intervening therapy suggested that ctDNA MRD concentrations increase exponentially over time (FIG. 1 A). After generation of log-transformed linear models for each patient (FIG. IB), we found these models to be strongly correlated with the observed ctDNA MRD measurements, having a median Pearson correlation of 0.89, and in 18 of the 23 patients (78%), having correlations greater than 0.5. Cases with correlations below this threshold tended to have ctDNA MRD measurements near the LOD95 of the assay, suggesting their ctDNA concentrations may not have been reliably quantified (Armbruster, D. A. & Pry, T., Clin Biochem Rev 29 Suppl 1, S49-52 (2008)). Thus, in the absence of adjuvant therapy, ctDNA MRD levels in most patients with NSCLC demonstrate exponential growth kinetics after primary surgical resection (FIG. 7).

[0063] Having established that exponential growth is an appropriate model for ctDNA MRD, we used the slopes of the log-linear models for the 28 (74%) patients whose data fit well with an exponential growth model to calculate patient-specific ctDNA doubling times (FIG. 1C). The distribution of doubling times was bimodal (overall median, 51 days; interquartile range, 40-79 days), with most patients demonstrating doubling times of ~40 days, and a minority having longer doubling times centered at -150 days. Doubling times were significantly longer for patients with stage I disease than for patients with stage II or III disease (P = 0.041), suggesting slower growing disease may have contributed to establishing a diagnosis at an earlier stage in these patients (FIG. ID). Notably, no significant difference in doubling times was observed between adenocarcinoma and non-adenocarcinoma histologic types (P = 0.65) (Fig. le). This suggests that, unlike steady-state levels of ctDNA shedding in established tumors (Abbosh, C. et al., Nature 545, 446-451 (2017); Abbosh, C. et al., Nature 616, 553-562 (2023) ; Chabon, J. J. et al., Nature 580, 245-251 (2020)), histology does not seem to significantly impact ctDNA dynamics in the MRD setting.

[0064] Using the individual patient-specific mathematical models, we then projected back in time to predict ctDNA concentrations immediately after surgery for each patient. This allowed us to estimate the clinical sensitivity at the post-operative landmark for ctDNA MRD assays with varying analytical sensitivities (i.e., LOD95s). Our models indicated that assays with an LOD95 similar to that of first generation ctDNA MRD assays (i.e., 100 ppm, or 0.01%) would be expected to achieve 39% clinical sensitivity at this landmark (FIG. IF). This result is consistent with the -30% to 40% clinical sensitivity observed at this post-operative landmark in studies using first generation MRD assays (Abbosh, C. et al., Nature 545, 446-451 (2017); Zhou, C. et al., Ann Oncol 32, S 1373-S 1391 (2021); Gale, D. et al., Ann Oncol 33, 500-510 (2022); Zhang,J. T. et al., Cancer discovery 12, 1690-1701 (2022); Abbosh, C. et al., Nature 616, 553-562 (2023); Chen, K. et al., Cancer Cell 41, 1749-1762 el746 (2023)). Furthermore, our modeling predicted that improving assay sensitivity to an LOD95 of 1 ppm would lead to a more than twofold increase of clinical sensitivity to 85%. Thus, a meaningful increase in clinical sensitivity for MRD detection in early stage NSCLC will require assays that can achieve analytical sensitivities near 1 ppm.Example 2: Identification of Patient-Specific Phased Variants

[0065] To test these model -based projections, we applied both the ultrasensitive START ctDNA MRD assay, also known as PhasED-Seq (see, e.g., WO2021 / 092476 and WO2022 / 236221, each of which is incorporated by reference herein in its entirety), which targets PVs, and a first generation MRD method that tracks SNVs to 269 tumor and blood samples from 46 patients with previously untreated early stage NSCLC (FIG. 2A). All patients underwent curative-intent resection, of whom 30% (n=14) received pre-operative (neoadjuvant) systemic therapy and 48% (n=22) received post-operative (adjuvant) therapy (FIG. 10). Median follow-up of the full cohort was 4.6 years. The cohort had a high prevalence of clinical characteristics associated with low ctDNA shedding, including lung adenocarcinoma (LU AD) (76% [n=35]), stage I and II disease (61% [n=28]), and low tumor volumes (<14 mL [volume of 3 cm diameter sphere] in 85% of patients [n=39]). Formalin-fixed paraffin-embedded (FFPE) tumor resection specimens and matched leukocytes were subjected to whole genome sequencing (WGS) for identification of SNVs and PVs. Recurrently mutated genes included common lung cancer drivers such as TP53 (48% [n=22]), KRAS (33% [n=16]), and EGFR (11% [n=5]) (FIG. 2B). A median of 2,020 PVs were identified and targeted via personalized capture panels for each case.

[0066] TABLE 1 : Patient-level clinical characteristics and outcomes.

[0067] For 5 patients treated with neoadjuvant therapy, pre-treatment core-needle biopsy FFPE specimens were available in addition to the corresponding post-systemic therapy surgical resection specimen. In these cases, we isolated tumor DNA from 2 separate sets of tissue sections from each of the pre-treatment biopsy samples and sequenced these 10 technical replicates using each patient’s personalized PV panel. As expected, the median PV variant allele fractions (VAFs) were higher in the pre-neoadjuvant therapy biopsy specimens in all cases (median 1.7-fold, range 1.4 to 2.4 fold), reflecting the impact of neoadjuvant therapy. These results also show that PVs can be successfully detected in pre-treatment FFPE core needle biopsy samples from patients with early stage NSCLC (FIG. 3 A). Furthermore, given the high concordance of recovered PVs before and after neo-adjuvant therapy, these results suggest that resection specimens harboring residual tumor cells after neoadjuvant therapy can serve to faithfully identify PVs for MRD detection.Example 3: Mutational Processes Associated With Phased Variants

[0068] To explore the mutagenic processes that give rise to PVs and SNVs in NSCLC, we compared the single-base substitution (SBS) mutational signatures associated with each mutation type (Alexandrov, L. B. et al., Nature 500, 415-421 (2013)). Across the entire cohort, PVs were significantly enriched for SBS4, which is associated with a history of tobacco exposure, whereas SNVs were enriched for SBS1, which is associated with aging (Alexandrov, L. B. et al., Nat Genet 47, 1402-1407 (2015)) (FIG. 3B). However, when the enrichment of mutational signatures was examined at an individual case level, PVs in never smokers were not enriched for SBS4 (FIG. 3C). We therefore compared SBS mutational signatures of PVs and SNVs in smokers versus never smokers. Both PVs and SNVs in smokers were significantly enriched for SBS4, compared with never smokers (FIG. 3D). Interestingly, whereas in never-smokers, SNVs were more strongly enriched for SBS1, PVs were enriched for SBS2, which is attributed to activity of the APOBEC family of cytidine deaminases. Thus, different mutagenic processes are responsible for PV accumulation in never-smokers and smokers, as well as for SNVs and PVs in never- smokers.Example 4: LOD95 and Pre-Treatment Detection of ctDNA

[0069] To confirm the ultra-low error profile of PVs, we assessed the background error rate of PVs and SNVs using the 46 patient-specific panels and plasma from 47 healthy controls (1,200 total comparisons). PVs demonstrated a background error rate of 0.014 ppm, which is more than 3,500-fold lower than that of SNVs in the same samples (FIG. 4A). To enable comparison between technical approaches, we next performed both START and a tumor- informed SNV-based MRD assay tracking the 16 SNVs with the highest VAFs in tumor tissue (CAPP-Seq; Methods) on all plasma samples. Across the 167 plasma samples analyzed in the study, START achieved a median LOD95 of 1.4 ppm (FIG. 4B), 60-fold lower than that of the SNV-based MRD assay (median LOD95 84 ppm) (FIG. 4C). Both assays had a specificity of 97% in control samples, with only 37 (3%) and 31 (3%) of the 1200 case-control tests displaying tumor signal for START and SNV-based monitoring, respectively. In the examination of pretreatment samples obtained before surgery or any preoperative treatment, START detected ctDNA in more than twice as many stage I and II LUADs than the SNV-based MRD assay (13 / 21 versus 6 / 21; P=0.03) (FIG. 4D). Rates of pre-treatment detection for the two MRD approaches were similar for patients with stage III LU AD and patients with stage I to III non- LUAD, which is consistent with the expected higher levels of ctDNA shedding in these cases.

[0070] When START was used, ctDNA was detected in 46% of all plasma samples (81 / 177), with a median concentration of 120 ppm (range, 0.19 - 21,860 ppm) (FIG. 4E). Although ctDNA concentrations determined by the two methods were highly correlated (R=0.9), in 29 plasmasamples, ctDNA was only detected using the PV-based approach while in one sample it was exclusively detected using the SNV-based method. As expected on the basis of known associations between NSCLC histologic types or stage and ctDNA shedding Abbosh, C. et al., Nature 545, 446-451 (2017); Abbosh, C. et al., Nature 616, 553-562 (2023); Chabon, J. J. et al., Nature 580, 245-251 (2020)), patients with LU AD had lower pre-treatment ctDNA concentrations than non-LUAD patients, and patients with stage I or II LU AD had the lowest levels (FIG. 4F). Even after controlling for stage and histologic type, the ctDNA concentrations in our cohort appeared to be significantly lower than those in 2 key prior studies (Abbosh, C. et al., Nature 616, 553-562 (2023); Chabon, J. J. et al., Nature 580, 245-251 (2020)), suggesting that our patients may have had a higher frequency of smaller tumors. Indeed, the median tumor volume of patients with detectable pre-treatment ctDNA in our study was 8.6 mL compared with 18.5 and 24.1 mL in the two previous studies. Notably, the relationship between tumor volume and pre-treatment ctDNA concentration remained remarkably consistent across the three studies (FIG. 4G). Even considering the significant differences in tumor volumes, the sensitivity of pretreatment ctDNA detection we observed using START was higher than that in previous studies using first-generation SNV-based MRD methods, most significantly for stage I LUADs (P=0.0013, one-sided Fisher exact test) (FIG. 4H).Example 5: Detection of ctDNA MRD at the Post-Treatment Landmark

[0071] Post-surgery plasma samples were available for 45 patients (98%), of whom 18 (40%) developed recurrence. To perform outcomes analyses, we defined the post-treatment landmark as the time of the first plasma sample obtained after completion of all treatment; and, when not available, as the time of the last post-surgical sample collected during adjuvant therapy. When START was used, ctDNA MRD was detected at the post-treatment landmark in 12 patients, compared with 6 patients when the SNV-based MRD assay was used. Of the 12 patients with positive START results for ctDNA MRD, all 12 (100%) developed recurrence; of the 6 patients with positive MRD results by the SNV-based approach, 5 (83%) developed recurrence (Fig. 5a). Conversely, of the 33 patients with negative START ctDNA-MRD results at the posttreatment landmark, 6 (18%) developed recurrence; of the 39 patients with negative results by the SNV-based approach, 13 (33%) developed recurrence. START achieved significantly superior clinical sensitivity (67% vs 28%, P=0.022, one-sided Fisher exact test) as well as higher specificity, positive predictive value, and negative predictive value, compared with the MRD- based approach (FIG. 5B).

[0072] Kaplan-Meier analyses revealed that patients who were positive for ctDNA MRD had significantly worse freedom from recurrence regardless of the MRD method used. However, stratification of outcomes based on MRD detection was greatest using START, with a largerfraction of recurrences present in the MRD-positive subgroup (FIG. 5C-D). Nearly identical results were observed for disease-free survival (FIG. 9A and 9B). Importantly, START, but not the SNV-based MRD assay, was also able to appropriately stratify overall survival (FIG. 9C and 9D). Additionally, START detected significantly more patients whose tumors recurred six months or more after surgery (9 / 14 vs. 3 / 14, P=0.027, one-tailed Fisher exact test). This suggests that increasing the analytical sensitivity of MRD detection significantly improves the ability to detect tumors that will ultimately recur at later times after treatment.

[0073] At the post-treatment landmark, the median concentration of ctDNA MRD in patients who developed recurrence was 3 ppm (range, 0-21,860 ppm). Samples that were falsely negative by the SNV-based approach but detected by START had ctDNA concentrations below the detection limit of the SNV-based assay (FIG. 5E). Post-treatment longitudinal sampling was sparse due to the collection period overlapping with the COVD-19 pandemic and associated logistical challenges, which prevented an accurate estimate of median lead time. As an alternative, we examined the sensitivity of MRD detection in relation to the time between posttreatment blood collection and recurrence. Sensitivity was stable, at approximately 65%, when the time between blood collection and recurrence was 0 to 24 months, and it decreased to 14% when samples were collected more than 24 months prior to recurrence (FIG. 5F).

[0074] We also examined outcomes of patients who received adjuvant treatment (including adjuvant systemic therapy, postoperative radiation therapy, or both) after tumor resection. Patients who were ctDNA MRD negative by START after surgery had similar outcomes regardless of receipt of adjuvant therapy (FIG. 5G); however, patients who were ctDNA MRD positive and received adjuvant therapy had statistically significantly better outcomes than those who did not (P=0.00035) (FIG. 5H). A similar benefit for adjuvant therapy was not observed in patients who were MRD-positive by the SNV-based assay, as the majority of patients who developed recurrence were found to be MRD-negative with this assay.Example 6: Patient Vignettes Highlighting the Utility of Ultrasensitive MRD Detection

[0075] The additional clinical sensitivity that we achieved using the START assay has potential utility in a number of clinical scenarios. For example, data from 2 of our patients suggest the potential use of ultrasensitive MRD analysis for selecting patients for consolidation radiotherapy. Patient MSK-FDX-026 received 4 cycles of neoadjuvant carboplatin and paclitaxel for a clinical stage IIIB (cT3N2) right upper lobe adenocarcinoma with biopsy -proven N2 nodal disease. The patient then underwent right upper lobectomy with en bloc right lower lobe wedge resection and mediastinal lymph node dissection, demonstrating nodal downstaging but a large (>7 cm) residual primary tumor (ypT4N0). ctDNA was detectable before, during, and after neoadjuvant chemotherapy using both assays (FIG. 6A). However, post-resection ctDNA wasdetected by START at a VAF of 0.31 ppm but undetectable using the SNV-based MRD assay. The patient then received postoperative radiation to the right pleural apex in an area of parietal pleural invasion noted on final pathology. After completion of radiotherapy, the patient’s ctDNA became undetectable, and he remains disease free at last follow-up more than 4 years after surgery. Similarly, patient MSK-FDX-034 had a clinical stage IIIA (cT2aN2) undifferentiated NSCLC with biopsy-proven N2 nodal involvement for which he completed four cycles of neoadjuvant cisplatin and docetaxel. Using START, ctDNA was detectable before initiation of therapy at 3800 ppm; dropped precipitously after one cycle of chemotherapy to 5.9 ppm, indicating an initial response, but then increased again to 171 ppm after completion of chemotherapy, just prior to surgery (FIG. 6B). The patient subsequently underwent a left upper lobectomy with mediastinal lymph node dissection, demonstrating a complete response in the primary tumor bed but residual disease in a single N2 node (ypT0N2al). ctDNA was detectable in the initial postoperative blood sample at 4.9 ppm, indicating residual disease. The patient then underwent postoperative radiotherapy to the mediastinum, after which ctDNA became undetectable and the patient remained disease free at last follow-up. In contrast, using the SNV- based MRD approach ctDNA was not detectable during neoadjuvant chemotherapy or postoperatively prior to radiotherapy. The results for these 2 patients suggest that ultrasensitive MRD analysis enables detection of microscopic residual local disease, which could help to inform personalization of consolidation radiotherapy.

[0076] Ultrasensitive MRD detection may also allow personalization of adjuvant systemic therapy. For example, patient MSK-FDX-014 underwent a right upper lobectomy with mediastinal lymph node dissection for a pathological stage IIB (TlcNl) LUAD with an EGFR exon 19 deletion and a de novo T790M mutation. Using the SNV-based MRD assay, ctDNA was undetected in all samples, including before surgery (FIG. 6C). By contrast, using START, the patient had detectable ctDNA before treatment at 4.1 ppm, and in the first postoperative blood sample at 3.5 ppm. After completing 4 cycles of adjuvant cisplatin and pemetrexed chemotherapy and starting EGFR tyrosine kinase inhibitor therapy with osimertinib, the patient’s ctDNA levels became undetectable. The patient remained recurrence free until his death from other causes.

[0077] Similarly, ultrasensitive MRD detection could improve the accuracy of identifying patients with residual disease after neoadjuvant immunotherapy followed by surgery. Patient MSK-FDX-039 received 3 cycles of neoadjuvant immunotherapy for a clinical T3N1 right middle lobe LUAD before undergoing right middle and lower bilobectomy with mediastinal lymph node dissection. Surgical pathology demonstrated significant residual tumor, consistent with ypT2bN2 LUAD. Using both methods, ctDNA was detectable before and duringneoadjuvant therapy, at ~50 ppm (FIG. 6D). Although ctDNA was never detected again after surgery using the SNV-based approach, ctDNA remained detectable using START, including in the initial postoperative blood sample at a concentration of 0.19 ppm. Consistent with the MRD result, 5 months after apparent complete resection the patient developed a right adrenal metastasis, which was not present on pretreatment, preoperative, or initial postoperative CT and PET / CT imaging studies. As the adrenal metastasis was the only site of disease, the patient underwent right adrenalectomy. ctDNA remained detectable after adrenalectomy and the patient subsequently received four cycles of carboplatin, pemetrexed and pembrolizumab. Additional follow-up blood draws were missed due to the COVID-19 pandemic. This vignette highlights the value of the START assay’s additional analytic sensitivity for correctly assessing response to neoadjuvant treatment and detecting MRD after resection.

[0078] Methods

[0079] Dataset

[0080] To model changes in ctDNA concentration over time, we used an existing data set of longitudinal ctDNA surveillance samples from patients who underwent curative-intent therapy for early-stage NSCLC in the TRACERx study. In this study, patients with early stage NSCLC underwent resection and were subsequently managed according to the National Institute of Clinical Excellence approved care pathways. Plasma samples for TRACERx ctDNA analysis were collected prospectively at 3-month intervals for the first 2 years after surgical resection and at 6-month intervals for the subsequent 3 years. These plasma samples were previously analyzed using tumor-informed personalized anchored-multiplex polymerase chain reaction with an LOD95 of 80 ppm (0.008%), and overall tumor VAF was calculated by adding the total number of reads containing tumor-specific variants divided by the total depth at those positions.

[0081] To enable modeling of ctDNA MRD growth dynamics, we assessed the 70 patients in the TRACERx cohort who experienced disease recurrence and identified the subset of patients who had three or more consecutive postoperative ctDNA MRD-positive samples, and underwent no intervening adjuvant therapy, for inclusion in our cohort. We also identified patients who had two consecutive ctDNA MRD-positive samples, and underwent no intervening therapy.

[0082] Generation of mathematical model for ctDNA growth dynamics

[0083] To generate a mathematical model for ctDNA growth dynamics, we used the 23 patients who had a recurrence and at least three consecutive ctDNA-positive samples without intervening therapy. We started from the assumption that ctDNA concentrations expand exponentially over time in patients with active disease. We then generated log-transformed linear models for each of the patients, using a correlation threshold > 0.5, with the actual ctDNA MRD data to ensure goodness of fit to exponential growth kinetics. After confirming that anexponential growth model was appropriate for the majority of these patients, we also generated log-linear models for patients with 2 consecutive ctDNA-positive samples.

[0084] Calculation of ctDNA doubling times and association with clinical variables

[0085] Using the patient-specific log-linear model slopes, which measure change in VAF over days post-surgery, we calculated ctDNA VAF doubling times for each patient using the following formula:

[0087] We included 18 patients with 3 or more consecutive ctDNA positive samples who had a log-linear model correlation > 0.5 and 10 patients with 2 consecutive ctDNA positive samples with a doubling time between 0 and 365 days in our modeling cohort.

[0088] We then assessed the distribution of ctDNA VAF doubling times and explored the associations between doubling time and stage (comparing stage I with II and III) or histologic type (comparing LU AD with non-LUAD) using the Wilcoxon rank sum test.

[0089] Simulating ctDNA VAF distributions to project clinical sensitivity

[0090] We used the patient-specific log-linear models to simulate a distribution of ctDNA VAFs at each day post-surgery. For each of the 28 patients at each day post-surgery, we generated a distribution of 100 VAFs using the log-linear model projections and standard errors. These simulated ctDNA VAF distributions were combined to create a projected distribution of 2800 ctDNA VAFs at each day post-surgery. We then calculated the projected clinical sensitivity at day 1 after surgical resection for assays with LOD95s of 100 ppm (0.01%), 10 ppm (0.001%), and 1 ppm.

[0091] Patient cohort

[0092] A total of 46 patients with stage I to III NSCLC (American Joint Committee on Cancer 8th edition) treated with curative intent were prospectively enrolled at Memorial Sloan Kettering Cancer Center (FIG. 10). Forty -four patients underwent complete (R0) resection. Thirty patients received perioperative systemic therapy, including 9 patients who received neoadjuvant therapy, 16 patients who received adjuvant therapy and 5 patients who received both. For patients who received neoadjuvant therapy, blood samples for cfDNA analysis were collected before treatment, during treatment, and after treatment / before surgery. For patients who received adjuvant therapy or no systemic therapy, blood samples were collected pre-operatively. Forty-five patients had at least 1 postoperative blood sample collected between 2 and 8 weeks postoperatively, with 32 (71%) having at least one additional follow-up blood collection. . Regular interval follow-up blood sample collection was hampered by the COVID-19 pandemic due to a combination of factors including patients choosing to obtain their surveillance imaging locally and following-up with their Memorial Sloan Kettering Cancer Center (MSKCC)physicians via telehealth visits. Additionally, the core lab that processed all blood collections for this study temporarily suspended operations during the pandemic. For patients treated with neoadjuvant therapy, clinical staging was used, whereas for the remaining patients, pathologic staging was used. Patients were selected such that a sufficient number with recurrence were included.

[0093] Blood collection

[0094] Blood was collected in two Cell-Free DNA BCT (Streck) 10 mL blood collection tubes and processed within 24 hours. Whole blood was centrifuged at 800 g for 10 minutes followed by a second spin at 18,000 g for 10 minute, both at room temperature. The plasma and buffy coat were each separated from the red blood cells. Plasma was stored at -80 °C until cfDNA isolation. Buffy coat was stored at -80 °C for later DNA isolation from leukocytes.

[0095] Sample processing and DNA extraction

[0096] FFPE specimens, including pre-treatment biopsy and / or resected tumors, were sectioned to prepare both hematoxylin-and-eosin-stained and unstained slides at 10 micron thickness. All subsequent experiments were performed in Foresight Diagnostic’s CLIA laboratory (Aurora, CO). Following pathology review to indicate regions of higher tumor purity, marked hematoxylin-and-eosin slides were used to guide macrodissection of unstained tissue for processing using the DNAstorm FFPE DNA Extraction Kit (CellDataSci) in accordance with the manufacturer’s instructions. Purified DNA samples were sonicated using a Covaris M220 ultrasonicator followed by solid-phase reversible immobilization (SPRI) bead cleanup. DNA sample quality was then assessed by qPCR using the KAPA Human gDNA Quant and QC Kit (Roche) in accordance with the manufacturer’s instructions. Buffy coat samples were extracted using the QIAamp DNA Mini Kit (QIAGEN) in accordance with the manufacturer’s instructions, followed by sonication and cleanup using the QIAquick PCR Purification Kit (QIAGEN). Plasma samples were manually extracted using either the QIAamp Circulating Nucleic Acid Kit or the QIAsymphony DSP Circulating Nucleic Acid Kit (QIAGEN) in accordance with the manufacturer’s instructions.

[0097] DNA sequencing

[0098] Sequencing libraries for WGS and targeted capture sequencing were prepared using the KAPA HyperPrep Kit (Roche) as previously described in detail elsewhere (Kurtz, D. M. et al., Nat Biotechnol (2021)). In brief, median DNA input was 237 ng (range, 13-500 ng) for FFPE tumor samples, 60 ng (range, 60-80 ng) for germline (peripheral blood mononuclear cell (PBMC)) samples, and 40 ng (range, 4-80 ng) for cfDNA (plasma) samples. For cfDNA samples, target enrichment with patient-specific panels was performed using the xGen Hybridization and Wash Kit (IDT) in accordance with the manufacturer’s instructions.Sequencing was performed using 2x 150 paired end reads using the Illumina NovaSeq 6000 system. The median depth for WGS samples was 70x.

[0099] Mutational signature analysis

[0100] The contribution of previously defined mutational signatures (Alexandrov, L. B. et al., Nature 500, 415-421 (2013)) to tumor PVs or SNVs was assessed using deconstructSigs vl.9.0 (available on the world wide web at https: / / github.com / raerose01 / deconstructSigs). PVs or SNVs identified from tumor WGS data were evaluated for each sample using the whichSignatures function from the deconstructSigs package using the COSMIC SBS signature set (signatures. exome.cosmic.v3.may2019). SNVs contained within PVs were excluded from the SNV signature analysis. Signature contributions to PVs or SNVs were compared across samples with a Wilcoxon rank-sum test and P values adjusted for multiple hypothesis testing using the Benjamini -Hochberg method.

[0101] Foresight START MRD assay

[0102] The START assay is based on the previously described PhasED-Seq method, with minor modifications (Kurtz, D. M. et al., Nat Biotechnol (2021)). In brief, sample fastq files were pre-processed using fastp and aligned to the GRCh37 reference genome using bwa-mem2. WGS bam files were then deduplicated using either the samtools or the picard deduplication function, whereas bam files for cfDNA samples were deduplicated using an in-house deduplication algorithm.

[0103] PVs were genotyped as previously described (Kurtz, D. M. et al., Nat Biotechnol (2021)), ignoring centromeres and microsatellites. We only considered properly paired reads with high mapping quality and hypermutated outlier fragments were removed. PVs present in the matched germline were not considered. Variants were then filtered against an in house database of normal and tumor WGS samples to remove frequently observed variants.

[0104] To select PVs for inclusion in the targeted capture panel, variants were filtered on the basis of a series of heuristics, as previously described (Kurtz, D. M. et al., Nat Biotechnol (2021)). In brief, these included PV spanning distance, support in matching PBMC, VAF, and MAPQ. Adjacent PVs were combined into the same region using bedtools merge and a median of 2,020 PVs (range, 524-2,172 PVs) were tracked per patient.

[0105] Capture panels were used to sequence the matching patient’s tumor and germline DNA and only variants with no background in the matched normal sample and a minimum VAF of 0.03 in the tumor sample were considered for subsequent tracking in plasma samples. Each capture panel was also used to sequence a set of 24 healthy control cfDNA samples, and variants with background signal were eliminated. The panels were then applied to the matching patient’s plasma samples and presence of ctDNA was determined using a previously described MonteCarlo-based approach (Kurtz, D. M. et al., Nat Biotechnol (2021)). The median depth for plasma samples was 4,685X. The ctDNA concentration was defined as the number of reads containing a targeted variant over the total number of reads covering variant positions. Specificity was tested by applying each patient-specific capture panel to cfDNA from 47 independent healthy controls.

[0106] SNV-based MRD assay

[0107] The SNV-based CAncer Personalized Profiling by Deep Sequencing (CAPP-Seq) assay was performed in parallel on the same cfDNA libraries, as previously described (Chabon, J. J. et al., Nature 580, 245-251; Newman, A. M. et al., Nat Med 20, 548-554 (2014)). In brief, the 16 SNVs with the highest VAFs in tumor were targeted to achieve a median LOD95 of 84 ppm (0.008%) matching those of clinically-available first generation SNV-based ctDNA MRD assays (Abbosh c / a / .(2023); Kandasamy, R. et al., J Clin Oncol 40, el3582-el3582 (2022); Russell, H. et al., Cancer Res 83, 3384 (2023); Newman, A. M. et al., Nat Biotechnol 34, 547- 555 (2016)). The presence of ctDNA was determined using a previously described Monte Carlobased approach and ctDNA concentration was corrected on the basis of the copy number at each variant locus in the tumor sample (Chabon, J. J. et al., Nature 580, 245-251; Newman, A. M. et al., Nat Med 20, 548-554 (2014)). Specificity was tested by applying each patient-specific capture panel to cfDNA from 47 independent healthy controls.

[0108] Limit of detection analysis

[0109] The LOD95 for both the SNV-based and START MRD assays was modeled as a binomial sampling process. First, the error-rate of tracking variants was assessed by monitoring patient-specific variants in cfDNA samples from a cohort of 47 healthy controls. This yielded an error-rate of ~5* 10'5for SNVs, and -2* 1 O'8for PVs, consistent with previous results (Kurtz, D. M. et al., Nat Biotechnol (2021); Newman, A. M. et al., Nat Biotechnol 34, 547-555 (2016)).

[0110] Using this background error-rate, for each sample, we then assessed what the 95th percentile of the expected number of detected reads would be in a negative sample from the binomial distribution, using the MATLAB function ‘binoinv’ and the number of unique sequencing reads that were assessable for tumor variants (i.e., informative reads). This yielded the threshold number of detected reads above which tumor signal can be detected with >95% specificity.

[0111] We then determined the minimum tumor fraction above which we would expect to see this number of detected reads, or more, given the number of informative reads sequenced. We performed this using the right-hand portion of the binomial cumulative distribution function ‘binocdf in MATLAB, with minimization using the ‘fmincon’ package. This procedure was performed for each sample’s specific number of sequencing reads and variants assessed by each method.

[0112] Tumor volume calculations

[0113] Gross tumor volume for each patient was estimated on computed tomography (CT) by calculating the ellipsoid volume of the tumor (transverse x anteroposterior x craniocaudal x 0.523) (Pathak, R. S. et al., J Med Imaging Radiat Oncol 60, 661-667, (2016)).Example 7: Analysis of disease-free survival using MRD clearance after Adjuvant Therapy in early stage breast cancer cohort.

[0114] Patient cohort

[0115] Samples from a total of 51 patents with early-stage breast cancer were analyzed (FIG. 11). Three different subtypes of early stage cancer were represented in the patient cohort, including HR+ / HER2-, TNBC, and HER2+. Each patient of the cohort was treated with surgery, and a subset of patients were treated with adjuvant therapy after surgery. 390 samples from 50 patients were analyzed.

[0116] ctDNA analysis

[0117] Blood samples were collected from patients at multiple landmarks throughout the treatment process and analyzed for the presence of ctDNA. Detecting the presence of ctDNA involved first isolating cell-free DNA from plasma in the patient blood samples and next analyzing the isolated cell-free DNA to detect ctDNA using START as shown in FIG. 15. The START workflow included isolating tumor DNA from the breast tumor tissue and germline DNA from a corresponding non-cancerous sample for each patient. Whole genome sequencing was performed on each of these samples and patient-specific variants were identified by comparing the sequencing data between the tumor sample and germline sample to identify phased variant (PV)- containing signatures. A custom set of oligonucleotide probes was designed to target specific regions within the patient tumor DNA sample that harbor the identified patientspecific PVs and in some cases other low-error rate tumor-specific alterations. These probes were then used to isolate sequences within the patient ctDNA samples collected throughout the treatment process, and the enriched sequences were then sequenced. Detection of tumor fraction within each sample was based on sequencing data associated with the patient-specific variants, and ctDNA detection status was determined based on the detection limit of START. The limit of detection (LOD95) of the MRD assay was 0.3 parts per million. Background signal rate was 1 in 35 million. Specificity was 100% in negative samples.

[0118] Samples were extracted from each patient during multiple landmarks of treatment. Multiple samples were collected before and after surgery as well as during follow-up visits. In some cases, patients received adjuvant therapy after surgery. The adjuvant therapy included chemotherapy, HER2 antibody therapy, Endocrine-based therapy, immune checkpoint inhibitortherapy, or a combination thereof. For patients who received adjuvant therapy, multiple samples were collected throughout the treatment process.

[0119] A majority of the patients (72%) had detectable ctDNA pre-treatment. 36% of samples with detectable ctDNA had levels <0.01% (<100 parts per million). Levels of ctDNA were assessed prior to treatment (FIG. 18A-E). Pre-treatment ctDNA detection was associated with tumor stage, T status, nodal status and grade. Tumor subtype was not statistically associated with ctDNA detection. Levels of ctDNA were assessed at landmarks related to neo-adjuvant chemotherapy (NACT) for patients who received NACT (FIG. 17A-B). Many samples at pre- and post-operative time points, as well as during follow-up, had ctDNA detectable with tumor fractions below 10'4. Patients who did not experience disease relapse had lower levels of ctDNA at post-operative and follow-up time points than those who experiences disease relapse. All patients who experienced long-term disease-free status persistently cleared their ctDNA-MRD, both post-operatively and post-adjuvant therapy. Levels of ctDNA changes during therapy, with clearance of ctDNA-MRD observed at various landmark time points throughout therapy, including after neo-adjuvant chemotherapy, after surgery, after adjuvant chemotherapy, and after endocrine therapy (FIG. 19A-E). FIG. 20 shows the clinical history and ctDNA detection for all samples for all patients considered. All patients with disease progression had ctDNA detected at or prior to the time of relapse (n=7 / 7, 100% sensitivity). In contrast, all patients with durable remission had persistently undetectable ctDNA during follow-up, generally after clearance of prior detectable ctDNA from NACT, surgery, or adjuvant therapy.

[0120] Analysis of ctDNA detection using START was used to determine disease status of each patient post-surgery and pre-adjuvant therapy for 49 patients. Disease free survival for the patient cohort stratified by MRD status post-surgery is shown in FIG. 12. As shown, of the 49 patients analyzed, 41 had undetected ctDNA levels post-surgery based on the START analysis, whereas 8 patients showed detectable levels of ctDNA at this landmark. As shown, the survival probability of the patients with detectable disease post-surgery and pre-adjuvant therapy was lower than for patients with undetected ctDNA at this landmark. The 8 patients with detectable ctDNA levels post-surgery were further stratified by ctDNA levels after adjuvant therapy (FIG. 13 A). As shown, the four patients with undetectable ctDNA post adjuvant therapy had an overall survival of 100% during the time horizon analyzed whereas the four patients with detectable ctDNA post adjuvant therapy had low survival probability within less than 20 months postsurgery. The therapy and START analysis details for each of the 8 patients depicted in FIG. 13 A are shown in 13B.Example 8: Detecting nucleic acids with a low limit of detection

[0121] In this example, cfDNA was analyzed using a dilution series to determine a limit of detection with a 95% confidence interval (LOD95).

[0122] Blood samples from limited-stage non-small cell lung cancer patients were processed to extract cfDNA from each sample. Patient-specific variants were identified for each patient such that cfDNA from non-small cell lung cancer patients could be distinguished from healthy donor extracted cfDNA based on the workflow outlined in FIG. 15. Briefly, for each patient, tumor DNA was isolated from tissue and germline DNA was isolated from a non-cancerous sample (e.g., blood, plasma, buccal, or a combination thereof). Whole genome sequencing was performed. Based on the comparison between the tumor DNA and germline DNA, a patientspecific assay targeting up to 5,000 targets per patient was designed. The targets included phased variants and other low-error-rate tumor specific alterations (e.g., single nucleotide variants, insertions, and deletions). Using each of the patient-specific set of targets, cfDNA was isolated from the plasma of each patient and analyzed using a custom panel sequencing based on the patient-specific targets. For cancer detection, the results of this assay can enable minimal residual disease detection. In the context of this experiment, this analysis was used for analyzing the LOD95 of the assay.

[0123] Three dilution series were performed with the limited-stage non-small cell lung cancer patient cfDNA diluted into a healthy donor cfDNA background. Replicates were performed in triplicate from expected tumor fraction of 1 part in 107to 1 part in 2xl07and observed fraction relative to expected fraction for each dilution is shown in FIG. 16. Negative samples without any limited-stage non-small cell lung cancer patient cfDNA were also analyzed to evaluate specificity. The LOD95 was determined to be 0.3 parts per million using Probit modeling. The background signal rate was 1 in 35 million and specificity was 100% in negative samples.

[0124] While preferred embodiments of the present disclosure have been shown and described herein, it will be obvious to those skilled in the art that such embodiments are provided by way of example only. Numerous variations, changes, and substitutions will now occur to those skilled in the art without departing from the disclosure. It should be understood that various alternatives to the embodiments of the present disclosure may be employed in practicing the present disclosure. It is intended that the following claims define the scope of the present disclosure and that methods and structures within the scope of these claims and their equivalents be covered thereby.

Claims

CLAIMSWhat is claimed is:

1. A method for identifying a subject as a candidate for adjuvant therapy, the method comprising:(a) isolating cell-free DNA (cfDNA) from a biological sample obtained from the subj ect;(b) detecting minimal residual disease (MRD) of the subject by assessing presence of circulating tumor DNA (ctDNA) molecules in the cfDNA; and(c) identifying the subject as a candidate for adjuvant therapy if MRD is detected, wherein the subject has previously undergone curative-intent resection of a cancer, and wherein the cancer is lung cancer, bladder cancer, or breast cancer.

2. The method of claim 1, wherein assessing the presence of ctDNA comprises assaying the cfDNA for single nucleotide variants (SNVs).

3. The method of claim 1 or 2, wherein the subject has a history of tobacco exposure.

4. The method of claim 3, wherein the SNVs are enriched for SBS4 variants.

5. The method of claim 1 or 2, wherein the subject does not have a history of tobacco exposure.

6. The method of claim 5, wherein the SNVs are enriched for SBS1 variants.

7. The method of any one of the preceding claims, wherein assessing the presence of ctDNA comprises assaying the cfDNA for phased variants (PVs).

8. The method of any one of claims 1 to 7, wherein the subject has, prior to (b), received an earlier adjuvant therapy.

9. A method for identifying a subject as a candidate for adjuvant therapy, the method comprising:(a) isolating cell-free DNA (cfDNA) from a biological sample obtained from the subj ect;(b) detecting minimal residual disease (MRD) of the subject by assessing presence of circulating tumor DNA (ctDNA) molecules in the cfDNA by assaying the cfDNA for phased variants (PVs); and(c) identifying the subject as a candidate for adjuvant therapy if MRD is detected, wherein the subject has previously undergone curative-intent resection of a cancer.

10. The method of claim 9, wherein the cancer is lung cancer, bladder cancer, or breast cancer.

11. The method of any one of claims 9 or 10, wherein the cancer is lung cancer.

12. The method of claim 11, wherein the lung cancer is non-small cell lung cancer.

13. The method of any one of claims 9-12, wherein the subject has a history of tobacco exposure.

14. The method of claim 13, wherein PVs or SNVs are enriched for SBS4 variants.

15. The method of any one of claims 9-14, wherein the subject does not have a history of tobacco exposure.

16. The method of claim 15, wherein the PVs are enriched for SBS2 variants.

17. The method of any one of claims 9-16, wherein the PVs have a background error rate less than about 50 ppm.

18. The method of any one of claims 9-17, wherein the method has a median LOD95 below 10 ppm.

19. The method of any one of claims 9-18, wherein the biological sample is collected between about 2 and about 8 weeks post-resection.

20. The method of any one of claims 9-19, wherein the subject received neo-adjuvant systemic therapy prior to resection.

21. The method of any one of claims 9-19, wherein the MRD is at a concentration of about 0.19 to about 22,000 ppm.

22. The method of claim 21, wherein the MRD is at a concentration of about 0.19 to about 5 ppm.

23. The method of any one of the preceding claims, wherein the adjuvant therapy is consolidation radiotherapy, adjuvant chemotherapy, adjuvant immunotherapy, adjuvant hormonal therapy, or adjuvant targeted therapy.

24. The method of any one of the preceding claims, wherein prior to detecting MRD, the method comprises a longitudinal testing step of the subject.

25. The method of claim 24, wherein the longitudinal testing occurs every 3-6 months.

26. The method of claim 23 or 24, wherein the subject is identified as a candidate for adjuvant therapy if MRD becomes detectable.

27. The method of any one of the preceding claims, wherein the biological sample is a blood sample.

28. The method of any one of the preceding claims, wherein assessing presence of ctDNA molecules in the cfDNA comprises whole genome sequencing.

29. The method of claim 28, further comprising targeted enrichment of patient-specific sets of genomic regions that were determined to harbor PVs in the resected cancer.

30. The method of any one of claims 9-29, wherein the subject has, prior to (b), received an earlier adjuvant therapy.

31. A method of treating cancer with an adjuvant therapy, the method comprising:(a) identifying a subject as a candidate for adjuvant therapy according to any one of the preceding claims; and(b) treating the subject with an adjuvant therapy selected from consolidation radiotherapy, adjuvant chemotherapy, adjuvant immunotherapy, adjuvant hormonal therapy, or adjuvant targeted therapy.

Images