Assessing and treating cancer

Abstract

This document relates to methods and materials for assessing and / or treating mammals (e.g., humans) suspected of having cancer. For example, this document provides methods and materials for detecting circulating tumor DNA (ctDNA) in a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a human) suspected of having cancer. In some cases, this document provides methods and materials for assessing and, optionally, treating a mammal having cancer. For example, a mammal (e.g., a human) identified as having the presence of ctDNA in a fluid sample (e.g., a fluid sample containing cfDNA) obtained from the mammal can be administered one or more cancer treatments.

Description

[0001] ASSESSING AND TREATING CANCER

[0002] CROSS-REFERENCE TO RELATED APPLICATIONS

[0003] This application claims the benefit of U.S. Patent Application Serial No.

[0004] 63 / 545.047, filed on October 20. 2023. The disclosure of the prior application is considered part of, and is incorporated by reference in, the disclosure of this application.

[0005] STATEMENT REGARDING FEDERAL FUNDING

[0006] This invention was made with government support under AG066101, CA212495, CA229240, and CA230995 awarded by the National Institutes of Health. The government has certain rights in the invention.

[0007] TECHNICAL FIELD

[0008] This document relates to methods and materials for assessing and / or treating mammals (e.g.. humans) suspected of having cancer. For example, this document provides methods and materials for detecting circulating tumor DNA (ctDNA) in a fluid sample (e.g., a fluid sample containing cfDNA). In some cases, this document provides methods and materials for assessing and, optionally, treating a mammal having cancer. For example, a mammal (e.g., a human) identified as having cancer based, at least in part, on the presence of ctDNA in a fluid sample (e.g., a fluid sample containing cfDNA) obtained from the mammal can be administered one or more cancer treatments.

[0009] BACKGROUND

[0010] Cell-free DNA (cfDNA) refers to small fragments of DNA that are released into the blood through processes such as cell death, apoptosis, or necrosis, fragmented by nucleases, and the circulate freely in the bloodstream (Diehl et al., Nature Medicine, 14(9): 985-990 (2008)). The cfDNA in blood-plasma is double stranded, varies in length up to -160 bases and in healthy tissues is constantly being removed from the bloodstream (Song et al., Nature Biomedical Engineering, 6(3):232-245 (2022)). cfDNA can originate from various cells in the body, with approximately 0.01% of cfDNA being circulating tumor DNA (ctDNA); i.e.. cfDNAthat originated from tumor cells (Sant et al., Cancers. 14(2):310 (2022)). As cfDNA carries genetic information that reflects the physiological and pathological states of the cells from which it originated, it has emerged as one of the leading analytical biomolecules for determining underlying pathological states, including cancer (van der Pol et al., Cancer Cell, 36(4):350-368 (2019)), pregnancy-related complications (Lo et al., Science Translational Medicine, 2(61), p. 61ra91 (2010)), autoimmune disorders (Duvvuri et al., Frontiers in Immunology, 10:502 (2019)), and organ transplant rejection (Knight et al., Transplantation, 103(2):273-283 (2019)).

[0011] SUMMARY

[0012] This document provides methods and materials for assessing and / or treating mammals (e.g., humans) suspected of having cancer. In some cases, this document provides methods and materials for detecting circulating tumor DNA (ctDNA) in a fluid sample (e.g., a fluid sample containing cfDNA). For example, this document provides methods and materials for determining an epigenetic instability index (Eli) of cfDNA, also referred to as a cell free epigenetic instability index (cfEII), where the cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) can be used to identity7the presence or absence of ctDNA within cfDNA. In some cases, a cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a human) suspected of having cancer can be used to identify that mammal as having cancer. For example, a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a human) suspected of having cancer can be assessed to determine if the mammal has cancer based, at least in part, the cfEII of the fluid sample.

[0013] Traditional methylation-based biomarker approaches employ genomic sites with significant tumor-specific methylation that are identified by analyzing methylation patterns across tumor and normal samples (Teschendorff et al., A (2012) Bioinformatics (Oxford, England). 28(11): 1487-1494; and Ritchie et al., (2015) Nucleic Acids Research, 43(7): e47). However, individual tumor sample methylation patterns show wide variation in methylation of CpGs at defined genomic sites, such as CpG-islands (CGIs). Adding to this patient-specific variation is the methylation heterogeneity amongst the tumor cells within an individual patient’s tumor.

[0014] This document describes the development of metrics to address variation in methylation, referred to herein as an epigenetic instability index (Eli). The Eli gives us a score of the degree of epigenetic perturbation of the genome for each patient sample based on DNA methylation. The metric distinguishes tumors from normal samples in a way that performs better than traditional measures of the degree of methylation at a given genome site (CpG dinucleotide) or regions (CpG-islands). As described herein, the genomic CGIs shown in Table 1 have increased Eli across various cancers and can be used to identify the presence of a cancer, regardless of the tumor location and regardless of the tumor origin. Also as described herein, the genomic CGIs shown in Table 1 can be used to determine the cfEII of a fluid sample (e.g., a fluid sample containing cfDNA). For example, the cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) can be calculated based on the variance of methylation of a plurality of sequencing reads of CpG sites in CpG-islands (ReadVarCGI) present in cfDNA in the fluid sample, thereby determining the presence or absence of ctDNA within the fluid sample. For example, a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a human) suspected of having cancer that includes a cfEII that is greater than 0.5, can be used to identify’ that mammal as having the presence of ctDNA in the fluid sample, thereby identifying the mammal as having cancer. For example, a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a human) suspected of having cancer that includes a cfEII that is greater than a cfEII of a reference sample, can be used to identify that mammal as having the presence of ctDNA in the fluid sample, thereby identifying the mammal as having cancer. Once a mammal (e.g., a human) is identified as having cancer, the mammal can be administered one or more cancer treatments.

[0015] Having the ability to determine whether a mammal (e.g., a human) has cancer as described herein (e.g., based, at least in part, on a cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from the mammal) provides a unique and unrealized opportunity for early cancer detection in a sensitive, non-invasive assay and can provide opportunities for early cancer detection (e.g., in asymptomatic patients) with high specificity and in anon-invasive manner. For example, methods for determining whether a mammal (e.g., a mammal suspected of having cancer) has cancer as described herein (e.g., based, at least in part, on a cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from the mammal) can be performed on a fluid sample obtained by a liquid biopsy.

[0016] In one aspect, methods for determining a cell free epigenetic instability index (cfEII) of a fluid sample comprising cell-free DNA (cfDNA) are provided. Such methods typically include determining a variance of methylation of a plurality of sequencing reads of CpG sites in CpG-islands (ReadVarCGI) of the cfDNA, wherein sequencing reads having fewer than 4 CpG sites are excluded, wherein the plurality of sequencing reads comprise sequencing reads of at least 15% of the CpG-islands set forth in Table 1; and wherein the variance represents the cfEII of the fluid sample.

[0017] In some embodiments, the determining the ReadVarCGI comprises calculating a variance across a cfDNA read that maps to a CGI of interest and has more than 4 CpG sites within the read. In some embodiments, each of the sequencing reads is less than 160 base pairs in length. In some embodiments, the plurality of sequencing reads comprise sequencing reads of from about 15% to about 30% of the CpG-islands set forth in Table 1.

[0018] In some embodiments, the fluid sample is selected from the group consisting of a whole blood sample, a plasma sample, a serum sample, a urine sample, a cerebrospinal fluid sample, and a saliva sample.

[0019] In some embodiments, the sample is obtained from a mammal. In some embodiments, the mammal is a human.

[0020] In some embodiments, such methods further include plotting a cumulative distribution function of the ReadVarCGI.

[0021] In some embodiments, the method comprises using a machine learning (ML) system. In some embodiments, the machine learning system is a Random Forest based ML algorithm.

[0022] In another aspect, methods for detecting circulating tumor DNA (ctDNA) in a fluid sample comprising cfDNA obtained from a mammal are provided. Such methods typically include (a) determining a cfEII of the cfDNA; and (b) identifying the cfDNA as comprising ctDNA when the cfEII is greater than a cfEII of a reference sample.

[0023] In some embodiments, the fluid sample is selected from the group consisting of a whole blood sample, a plasma sample, a serum sample, a urine sample, a cerebrospinal fluid sample, and a saliva sample.

[0024] In some embodiments, the mammal is a human.

[0025] In still another aspect, methods for assessing a mammal are provided. Such methods typically include (a) determining a cfEII of a fluid sample obtained from the mammal; and (b) identifying the mammal as having cancer when the cfEII is greater than a cfEII of a reference sample.

[0026] In yet another aspect, methods for assessing a mammal are provided. Such methods typically include (a) determining a cfEII of a fluid sample obtained from the mammal; and (b) identifying the mammal as not having cancer when the cfEII is not greater than a cfEII of a reference sample. In some embodiments, the fluid sample is selected from the group consisting of a whole blood sample, a plasma sample, a serum sample, a urine sample, and a cerebrospinal fluid sample.

[0027] In some embodiments, the mammal is a human.

[0028] In some embodiments, the cancer comprises one or more solid tumors. In some embodiments, the cancer is selected from the group consisting of a breast cancer, a lung cancer, a liver cancer, a colorectal cancer, a prostate cancer, a brain cancer, a skin cancer, a head and neck cancer, a pancreatic cancer, a genitourinary cancer, a gynecologic cancer, a gastrointestinal cancer, sarcomas, and a hepatobiliary cancer.

[0029] In another aspect, methods for treating for a mammal having cancer are provided. Such methods typically include (a) determining that a cfEII of a fluid sample obtained from the mammal is greater than a cfEII of a reference sample; and (b) administering a cancer treatment to the mammal.

[0030] In still another aspect, methods for treating cancer are provided. Such methods typically include administering a cancer treatment to a mammal identified as comprising a fluid sample having a cfEII having a variance greater than a cfEII of a reference sample.

[0031] In some embodiments, the fluid sample is selected from the group consisting of a whole blood sample, a plasma sample, a serum sample, a urine sample, and a cerebrospinal fluid sample.

[0032] In some embodiments, the mammal is a human.

[0033] In some embodiments, the cancer comprises one or more solid tumors. In some embodiments, the cancer is selected from the group consisting of a breast cancer, a lung cancer, a liver cancer, a colorectal cancer, a prostate cancer, a brain cancer, a skin cancer, a head and neck cancer, a pancreatic cancer, a genitourinary cancer, a gynecologic cancer, a gastrointestinal cancer, sarcomas, and a hepatobiliary cancer.

[0034] In some embodiments, the cancer treatment comprises performing surgery; In some embodiments, the cancer treatment comprises radiation therapy. In some embodiments, the cancer treatment comprises administering, to the mammal, an anticancer agent selected from the group consisting of a chemotherapy, and an angiogenesis inhibitor.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. Although methods and materials similar or equivalent to those described herein can be used to practice the invention, suitable methods and materials are described below. All publications, patent applications, patents, and other references mentioned herein are incorporated by reference in their entirety. In case of conflict, the present specification, including definitions, will control. In addition, the materials, methods, and examples are illustrative only and not intended to be limiting.

[0036] The details of one or more embodiments of the invention are set forth in the accompanying drawings and the description below. Other features, objects, and advantages of the invention will be apparent from the description and drawings, and from the claims.

[0037] BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figures 1 A-1H. The Eli enabled the distinction of healthy vs cancer samples in a pan-cancer investigation. Figure 1A) Schematic of Eli measure 1, Methylation Heterogeneity In CpG sites (MHIC). Figure IB) Pan-cancer TCGA screen of MHIC measurements between healthy and cancerous samples across different genomic elements (CpG Island TSS, Bivalently marked CpG Island Promoters, Non-bivalently marked CGI- TSS and Imprinting genes). Each arrow indicates the increased MHIC for bivalent promoter CGIs. Figure 1C) Standard Deviation of methylation across bivalently marked CpG Island Promoters in 5 different TCGA cancer studies. The vertical line indicates the minimal standard deviation in the healthy samples. Figure ID) Schematic of Eli measure 2, Variance in methylation across regulatory regions like CpG Islands (VarCGI). Figure IE) VarCGI across normal and TCGA LUAD samples. Figure IF) MHIC and VarCGI across Boruta269 CGI in Healthy and TCGA LUAD samples. Figure 1G) Verification of Boruta269 CGI in WGBS lung and liver cancers via MHIC. Figure 1H) Venfication of Boruta269 CGI in WGBS lung and liver cancers via VarCGI.

[0039] Figures 2A-2G. The Eli detected differences between healthy and cancerous samples via cfDNA targeted and whole genome bisulfite sequencing. Figure 2A) Schematic of ReadV arCGI, the adaptation of the second Eli metric VarCGI for use in cfDNA targeted sequencing. Figure 2B) Cumulative distribution function (CDF) of the ReadV arCGI metric within Boruta269 CGI across different stages of lung cancer from cfDNA-based targeted sequencing and healthy controls. There were 143 Stage IA samples, 29 Stage IB samples, 36 Stage II samples, 27 Stage III samples, and 240 healthy samples. Figure 2C) CDF plot of solely methylation across those Boruta269 CGI in the same lung cancer dataset used in Figure 2B. Figure 2D) CDF plot of the ReadV arCGI metric across regions identified in the original study of cfDNA-based targeted methylation sequencing in different stages of lung cancer. Figure 2E) CDF plot of those regions based purely on methylation signatures alone. Figure 2F) CDF plot of cfDNA- based WGBS of a breast cancer study using the ReadV arCGI metric and the Boruta269 CGI. Data included 15 early / non-metastatic samples and 15 advanced / fully metastatic samples in addition to healthy controls (60 samples). Figure 2G) CDF plot of those Boruta269 CGI plotted in the same breast cancer dataset but calculated solely on the basis of the methylation of those sites.

[0040] Figures 3A-3H. The Epigenetic Instability Index (Eli) enables the distinction of healthy vs cancer samples in a pan-cancer investigation. Figure 3A) Density estimates of beta values across all CGI which reside along TSS within healthy and COAD DNA methylation from TCGA, each red line is a cancer sample, and each blue line is a healthy- normal sample. Figure 3B) Density estimates of beta values across all poly comb marked CGI which reside along TSS within healthy and COAD DNA methylation from TCGA, each red line is a cancer sample, and each blue line is a healthy normal sample. Figure 3C) Boxplots of niean-B methylation per sample across poly comb marked CGI which reside at TSS in a pan cancer TCGA screen. Figure 3D) Schematic representation of MHIC (Methylation Heterogeneity in CpG Sites). Figure 3E) Schematic representation of V arCGI (Variance in methylation in regulatory regions, such as CGI. Figure 3F) Schematic representation of the difference between MHIC and VarCGI scores and alternative metrics of methylation assessment. Figure 3G) MHIC scores across different TCGA cancers and their associated normal samples. Each point represents the MHIC score for one patient sample. Figure 3H) VarCGI scores across different TCGA cancers and their associated normal samples. Each point represents the VarCGI score for one patient sample.

[0041] Figures 4A-4H. Using the Ell to elucidate regions of the epigenome most susceptible to epigenetic instability. Figure 4A) MHIC scores of samples across different TCGA cancers (BRCA, COAD, GBM, LUAD, PAAD) and associated normal samples within different genomic elements (CpG Islands that reside along TSS, Poly comb marked promoter CGI which reside along TSS, Non-polycomb marked promoter CGI which reside along TSS). Figure 4B) MHIC scores of samples across different TCGA cancers and associated normal samples within 553 Boruta feature selected CGI. Figure 4C) VarCGI scores of samples across different TCGA cancers and associated normal samples within 553 Boruta feature selected CGI. Figure 4D) Density distribution of standard deviation across poly comb marked CGI in normal samples of different TCGA cancers. Figure 4E) Density distribution of standard deviation across poly comb marked CGI in cancerous samples of different TCGA cancers. Figure 4F) Schematic representation of the feature selection process followed to obtain regions of the epigenome likely to be the most susceptible to epigenetic instability . Figure 4G) MHIC scores of normal and cancerous samples across different TCGA cancers within the 269 CGI highlighted from Boruta feature selection. Figure 4H) VarCGI scores of normal and cancerous samples across different TCGA cancers within the 269 CGI highlighted from Boruta feature selection.

[0042] Figures 5A-5D. Verification of the Eli in higher resolution technologies. Figure 5 A) MHIC scores within Boruta selected 269 CGI between Lung and Liver WGBS and associated normal samples. Figure 5B) VarCGI scores within Boruta selected 269 CGI between Lung and Liver WGBS and associated normal samples. Figure 5C) Mean methylation measurements per sample within the Boruta selected 269 CGI between Lung and Liver WGBS and associated normal samples. Figure 5D) Mean methylation measurements per sample within sites from the TCGA screen that were deemed to be differentially methylated between Liver and Lung WGBS and associated normal samples. Differential methylation criteria defined as a delta beta > 0.2 | < -0.2 and FDR < 0.05.

[0043] Figures 6A-6E Eli can detect differences in DNA methylation between normal and breast cancer samples within cfDNA. Figure 6A) Schematic of the adaptation of the second Eli metric VarCGI for use in cfDNA. Figure 6B) VarCGI scores of the Boruta selected 269 CGI per sample between; normal, early breast cancer (non-metastatic) and advanced breast cancer samples (fully metastatic). Figure 6C) Mean methylation measurements of the Boruta selected 269 CGI per sample between; normal, early breast cancer and advanced breast cancer samples. Figure 6D) Cumulative Distribution Function (CDF) of the VarRead metric within Boruta selected 269 CGI across different stages of Breast cancer. Figure 6E) CDF of mean methylation within Boruta selected 269 CGI across different stages of Breast cancer.

[0044] Figures 7A-7C. Eli can detect differences in DNA methylation between normal and NSCLC samples within cfDNA. Figure 7A) VarCGI scores of the Boruta selected 269 CGI per sample between; normal, and Stage 1 A - Stage III NSCLC samples. Figure 7B) CDF of VarRead measurements of the Boruta selected 269 CGI per sample between; normal, and Stage 1A - Stage III NSCLC samples. Figure 7C) CDF of mean methylation within Boruta selected 269 CGI across different stages of Breast cancer.

[0045] Figures 8-8G. Random Forest-based Eli classifiers outperform alternative assessed methodologies of cancer detection. Figure 8A) Schematic of classifier design. Figure 8B) ROC of Eli based Random Forest classifiers tasked with differentiating normal and breast cancer samples. Figure 8C) ROC of Eli based Random Forest classifiers tasked with differentiating normal and NSCLC samples. Figure 8D) ROC of Eli based Random Forest classifiers tasked with differentiating normal and Stage I NSCLC samples. Figure 8E) ROC of Eli based Random Forest classifiers tasked with differentiating normal and Stage II & III NSCLC samples. Figure 8F) ROC of Eli based Random Forest classifiers tasked with differentiating normal and early (non-metastatic) breast cancer samples. Figure 8G) ROC of Eli based Random Forest classifiers tasked with differentiating normal and advanced (metastatic) breast cancer samples. AUC listed with 95% Confidence Intervals.

[0046] Figures 9A-9E. Relationship between Eli measures (MHIC and VarCGI) with mean-B methylation in poly comb marked CGI. Figure 9A) MHIC score per cancer sample calculated within poly comb marked CGI as a function of mean-B methylation per sample calculated across the same genomic intervals. Figure 9B) VarCGI score per cancer sample calculated within poly comb marked CGI as a function of mean-B methylation per sample across the same genomic intervals. Figure 9C) MHIC score per cancer sample calculated across poly comb marked CGI as a function of MHIC score per cancer sample calculated across the same genomic intervals. Black numbers at the top LHS of each facet represent the correlation coefficient of each comparison. Figure 9D) Top 100 cancer- derived poly comb marked CGI (which reside along TSS) with the highest SD across the CGI as a function of the mean-B of those CGI within different TCGA cancers. Figure 9E) Top 100 cancer-derived poly comb marked CGI (which reside along TSS) with the highest mean-B methylation as a function of the standard deviation of those CGI within different TCGA cancers.

[0047] Figures 10A-10D. Alternative MHIC criteria, mean-B assessments of the Boruta selected 269 CGI and mean-B assessments of differentially methylated sites within TCGA cancers included. Figure 10A) MHIC scores per sample within poly comb marked CGI which reside along TSS across different TCGA cancers and associated normal samples, using a criterion of 0.3-0.7 beta value as the definition of the intermediate methylation zone as opposed to 0.2-0.8. Figure 10B) Mean-B methylation per sample of the Boruta selected 269 CGI across different TCGA cancers and their associated normal samples. Figure IOC) Mean-B methylation per sample of sites marked as differentially methylated across different TCGA cancers and associated normal samples. Differential methylation analysis was carried out per cancer and differential methylation was defined as delta beta > 0.2 or < -0.2 and FDR < 0.05. Figure 10D) Table detailing the number of differentially methylated sites across TCGA cancers studied split by hypermethylated / hypomethylated.

[0048] Figures 11A-11D. Additional comparisons of VarRead and Mean-B metrics differentiating normal and different stages of breast cancer. Figure HA) Mean-B methylation of differentially methylation sites identified in a TCGA screen across normal, early and advanced breast cancer. Figure 1 IB) CDF of VarRead metric and mean-B methylation across B-269 CGI within normal, early and advanced breast cancer. Figure 11C) CDF of VarRead metric and mean-B methylation across differentially methylated regions within normal, early and advanced breast cancer. Figure 1 ID) Methylation density between B-269 CGI and differentially methylated sites across normal, early breast cancer samples and advanced breast cancer samples.

[0049] Figures 12A-12F. Supplemental data to support the capacity ofVarCGI to differentiate normal vs cancerous samples from the DNA methylation of cfDNA in comparison to mean-B methylation. Figure 12A) CDF of the VarRead metric (left) and mean-P methylation (right) within the Boruta selected 269 CGI across different stages of NSCLC. Figure 12B) CDF of the VarRead metric (left) and mean-B methylation (right) within differentially methylated sites across different stages of NSCLC. Figure 12C) CDF of the VarRead metric (left) and mean-B methylation (right) within the Liang, 2021 CGI across different stages of NSCLC. Figure 12D) VarCGI metric across different stages of lung cancer within B-269 CGI (left). DM Sites (middle) and Liang 2021 CGI (right). Figure 12E) Mean-B methylation metric across different stages of lung cancer within B- 269 CGI (left), DM Sites (middle) and Liang 2021 CGI (right). Figure 12F) Methylation density of B-269 CGI and DM Sites across normal and different stages of NSCLC.

[0050] Figures 13A-13D. Supplemental data to support the claim that the differentiating ability of VarRead to separate normal and cancerous samples is not due to mathematical artifact. Figure 13A) VarRead metric (left) and mean methylation metric (right) within all CGI genome-wide across cfDNA methylation. Figure 13B) VarRead metric (left) and mean methylation metric (right) within 269 random CGI throughout the genome. Figure 13C) VarRead metric (left) and mean methylation metric (right) of all poly comb marked CGI which reside along TSS. Figure 13D) VarRead metric (left) and mean methylation metrics (right) across 269 random poly comb marked CGI which reside along TSS. Each figure referenced above is comprised of 15 early breast cancer samples, 15 advanced breast cancer samples and 30 associated normal samples (cfDNA WGBS) or 143 Stage IA, 29 Stage IB, 36 Stage II, 27 Stage III NSCLC samples and 240 associated normal samples (cfDNA deep targeted sequencing).

[0051] Figures 14A-14B. CpG density across CGI throughout the genome in different publicly available cfDNA WGBS or targeted sequencing preparations. Figure 14A) Density of CpG residues per CGI within reads covering the Boruta selected 269 CGI (upper left), All CGI which reside along TSS (upper right), all polycomb marked CGI which reside along TSS (lower left) and all sites deemed to be differentially methylated from a TCGA screen (lower right). Each figure referenced above is comprised of 15 early breast cancer samples, 15 advanced breast cancer samples and 30 associated normal samples (cfDNA WGBS). Figure 14B) Density of CpG residues per CGI within reads covering the Boruta selected 269 CGI (upper left), All CGI which reside along TSS (upper right), all poly comb marked CGI which reside along TSS (middle left), all sites deemed to be differentially methylated from a TCGA screen (lower right) and sites highlighted by Liang et al., 2021 (lower left). With reference to Figure 14B, each figure is comprised of 143 Stage IA, 29 Stage IB. 36 Stage II, 27 Stage III NSCLC samples and 240 associated normal samples (cfDNA deep targeted sequencing).

[0052] DETAILED DESCRIPTION

[0053] This document provides methods and materials for assessing and / or treating mammals (e.g., humans) suspected of having cancer. In some cases, this document provides methods and materials for detecting circulating tumor DNA (ctDNA) in a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g.. a human) suspected of having cancer. For example, this document provides methods and materials for determining an epigenetic instability index (Eli) of cfDNA, also referred to as a cell free epigenetic instability index (cfEII), where the cfEII can be used to identity the presence or absence of ctDNA within cfDNA. In some cases, a cfEII of cfDNA within a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a human) suspected of having cancer can be determined. For example, a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a human) suspected of having cancer can be assessed to determine if the mammal has cancer based, at least in part, the cfEII of cfDNA within in the fluid sample.

[0054] A fluid sample (e.g., a fluid sample containing cfDNA) can be assessed for the presence or absence of ctDNA by detecting the cfEII of the fluid sample. For example, a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a human) suspected of having cancer can be assessed to determine if the mammal has cancer based, at least in part, on the cfEII of the fluid sample. For example, a cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal suspected of having cancer can be calculated based on at least 15% (e.g., at least 20%, at least 25%, at least 30%, at least 35%, at least 40%, at least 50%, at least 60%, at least 70%, at least 80%, at least 90%, or at least 95%) of the genomic CGIs listed in Table 1. In some cases, a cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal suspected of having cancer can be calculated based on from about 15% to about 100% (e.g., from about 15% to about 90%, from about 15% to about 80%, from about 15% to about 70%. from about 15% to about 50%, from about 15% to about 40%, from about 15% to about 30%, from about 20% to about 100%, from about 30% to about 100%, from about 40% to about 100%, from about 50% to about 100%, from about 60% to about 100%, from about 70% to about 100%, from about 80% to about 100%, from about 90% to about 100%, from about 20% to about 90%, from about 30% to about 80%, from about 40% to about 70%. from about 50% to about 60%, from about 20% to about 40%, from about 30% to about 50%, from about 40% to about 60%, from about 50% to about 70%, from about 60% to about 80%, or from about 70% to about 90%) of the genomic CGIs listed in Table 1. For example, a cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal suspected of having cancer can be calculated based on at least 40 (e.g.. at least 50, at least 60, at least 70, at least 80, at least 90, at least 100, at least 125, at least 150, at least 175, at least 200, at least 225, or at least 250) of the genomic CGIs listed in Table 1. In some cases, a cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal suspected of having cancer can be calculated based on from about 40 to about 269 (e.g., from about 40 to about 250. from about 40 to about 225, from about 40 to about 200, from about 40 to about 175, from about 40 to about 150, from about 40 to about 125, from about 40 to about 100, from about 40 to about 80, from about 40 to about 60, from about 50 to about 269, from about 75 to about 269, from about 100 to about 269, from about 125 to about 269, from about 150 to about 269, from about 175 to about 269, from about 200 to about 269, from about 225 to about 269, from about 250 to about 269, from about 50 to about 250, from about 75 to about 225, from about 100 to about 200, from about 125 to about 175, from about 50 to about 150, from about 75 to about 125, from about 100 to about 200, from about 125 to about 175, from about 150 to about 250, or from about 40 to about 269) of the genomic CGIs listed in Table 1. In addition to CGIs, the same metrics described herein can be applied to other features within the mammalian genome (e.g., enhancers, promoters, coding sequences).

[0055] Regulatory regions (e.g., one or more CGIs) in nucleic acid (e.g., cfDNA) are subject to high epigenetic heterogeneity methylation patterns, and the epigenetic instability- (Eli) of such regulatory regions (e.g., one or more CGIs) can be determined. In some cases, Eli can be calculated based on a methylation heterogeneity in CpG sites (MHIC). A MHIC calculates the proportion of CG sites across the genome in a sample which register a beta value of between 0.2 - 0.8 (e.g., the proportion of the genome having intermediate methylation). Any beta value range greater than 0.2 to any value less than 1 (e.g., 0.3 to 0.7) can be used. MHIC is represented in Figure 1A. In some cases, Eli can be calculated based on a variance in methylation across regulatory regions like CpG Islands (VarCGI). If all sites in the CGI are unmethylated the variance in that CGI’s methylation will be 0. However, if CpG sites within the CGI start to gain methylation, the variance of that CGI will increase. CGI vary in length throughout the human genome. In some cases, a VarCGI calculation includes only variance along individual sequencing reads containing 4 or more CG sites within them. VarCGI is represented in Figure ID.

[0056] When determining the Eli of one or more regulatory regions (e.g., one or more CGIs), any appropriate method can be used to detect methylation of a CpG site. Examples of methods that can be used to detect methylation of a CpG site include, without limitation, bisulfite conversions (e.g., sodium bisulfite conversion and enzy matic conversion).

[0057] When determining the Eli of one or more regulatory regions (e.g., one or more CGIs), any appropriate method can be used to obtain a sequencing read of a CpG site. Examples of methods that can be used to obtain a sequencing read of a CpG site include, without limitation, sequencing reads derived from long read sequencing (e.g., nanopore sequence and PacBio HiFi sequencing).

[0058] In some cases, the Eli of one or more regulatory regions (e.g., one or more CGIs) can include reduced representation bisulfite sequencing (RRBS).

[0059] In some cases, the Eli of one or more regulatory regions (e.g.. one or more CGIs) can include whole genome bisulfite sequencing (WGBS).

[0060] A cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) is calculated based on the ReadVarCGI, where the plurality of sequencing reads cover a plurality' of CGIs set forth in Table 1. The cfEII of a fluid sample is calculated by determining the ReadVarCGI. Any appropriate method can be used to determine the ReadVarCGI. In some cases, a RLM package can be used to determine the ReadVarCGI. Any sequencing read having fewer than 4 CpG sites generally is excluded from the calculation. The variance of sequencing reads that map to CGIs is determined and this variance represents the cfEII of the fluid sample. VarCGI is represented in Figure 2A. In some cases, a Boruta package is used to identify genomic CGIs having increased Eli across various cancers (see, e.g., Table 1).

[0061] In some cases, a cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) can be calculated as described in Example 1 or Example 2.

[0062] In some cases, a cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) can be used to determine whether or not the fluid sample contains ctDNA. For example, when a fluid sample (e g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a human) suspected of having cancer includes a cfEII that is greater (e.g.. statistically significantly greater) than a cfEII of a reference sample, that mammal can be identified as having the presence of ctDNA in the fluid sample, thereby identifying the mammal as having cancer. For example, when a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a human) suspected of having cancer includes a cfEII that is greater than 0.5, that mammal can be identified as having the presence of ctDNA in the fluid sample, thereby identifying the mammal as having cancer. In some embodiments, the cfEII is determined by a machine learning (ML) algorithm (e.g., a Random Forest based ML algorithm).

[0063] The term “greater than’?as used herein with respect to a cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a human) suspected of having cancer refers to a cfEII value that higher than a cfEII value of a reference sample. The term reference as used herein with respect to a cfEII value of a reference sample refers to the cfEII typically observed in a sample (e.g., a control sample) from one or more mammals (e.g., humans) without cancer (e.g., healthy mammals such as healthy humans). In some cases, a cfEII value of a reference sample is determined using a ML algorithm. It will be appreciated that cfEII from comparable samples are typically used when determining whether or not a particular cfEII is greater than a cfEII of a reference sample. For example, when assessing a human for the cfEII of a fluid sample, the cfEII of a reference sample is a cfEII of human control samples.

[0064] In some cases, a cfEII that is greater than a cfEII of a reference sample can be at least 5% (e.g., about 5%, about 10%, about 20%, about 30%, about 40%, about 50%, about 60%, about 70%, about 80%, about 90%, about 95%, or more) greater than the cfEII of the reference sample. In some cases, a cfEII that is greater than a cfEII of a reference sample can be at least 2 (e g., at least 5, at least 10, at least 15, at least 20, at least 25, at least 35, or at least 50) fold greater than the cfEII of the reference sample.

[0065] Any appropriate fluid sample (e.g., a fluid sample containing cfDNA) can be assessed as described herein (e.g., for the presence or absence of ctDNA based, at least in part, on the cfEII of the fluid sample). In some cases, a fluid sample can be a biological sample. In some cases, a fluid sample can contain one or more biological molecules (e.g., nucleic acids such as DNA and RNA, polypeptides, carbohydrates, lipids, hormones, and / or metabolites). In some cases, a fluid sample can contain cfDNA. In some cases, a fluid sample does not contain genomic DNA. Examples of fluid samples that can be assessed as described herein include, without limitation, blood samples (e.g., whole blood samples, plasma samples, and serum samples), urine samples, cerebrospinal fluid (CSF) samples, and saliva samples. A sample can be a fresh fluid sample or a fixed fluid sample. In some cases, one or more biological molecules can be isolated from a fluid sample. For example, nucleic acid (cfDNA) can be isolated from a fluid sample and can be assessed as described herein.

[0066] When a fluid sample (e.g., a fluid sample containing cfDNA) is obtained from a mammal (e.g., a human) such as a mammal suspected of having cancer, any appropriate method can be used to obtain the fluid sample. When a fluid sample is a blood sample, the blood sample can be obtained using phlebotomy techniques (e.g., can be a liquid biopsy). Any appropriate mammal suspected of having cancer can be assessed and / or treated as described herein. Examples of mammals that can have cancer and can be assessed and / or treated as described herein include, without limitation, humans, nonhuman primates (e.g., monkeys), dogs, cats, horses, cows, pigs, sheep, mice, and rats. In some cases, a human suspected of having cancer can be assessed and / or treated as described herein.

[0067] When assessing a mammal (e.g., a human) suspected of having cancer as described herein and / or treating a mammal (e.g., a human) identified as having cancer as described herein, the cancer can be any type of cancer. For example, a cancer assessed and / or treated as described herein can include one or more solid tumors. In some cases, a cancer assessed and / or treated as described herein can be a blood cancer. In some cases, a cancer assessed and / or treated as described herein can be a primary cancer. In some cases, a cancer assessed and / or treated as described herein can be a metastatic cancer. In some cases, a cancer assessed and / or treated as descnbed herein can be a refractory cancer. In some cases, a cancer assessed and / or treated as described herein can be a relapsed cancer. Examples of cancers that can be assessed and / or treated as described herein include, without limitation, breast cancers, lung cancers, liver cancers, colorectal cancers, prostate cancers, brain cancers, skin cancers, head and neck cancers, pancreatic cancers, genitourinary' cancers, gynecologic cancers, gastrointestinal cancers, sarcomas, and hepatobiliary cancers.

[0068] As described herein, the cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a human) can be used to identify that mammal as having cancer. For example, the cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a human) suspected of having cancer can be used to determine the presence or absence of cfDNA, such that the presence of ctDNA can be used to the identify the mammal as having cancer and the absence of ctDNA can be used to identify the mammal as not having cancer.

[0069] In some cases, a cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a human) can be used to identify that mammal as having cancer. For example, when a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a mammal such as a human suspected of having cancer) is determined to have a fluid sample including a cfEII that is greater than, e.g., 0.5 and / or greater than a cfEII of a reference sample, the mammal can be identified as having cancer. For example, when a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a mammal such as a human suspected of having cancer) is determined to have the presence of ctDNA based, at least in part, on the fluid sample having a cfEII that is greater than, e.g., 0.5 and / or greater than a cfEII of a reference sample, the mammal can be identified as having cancer.

[0070] In some cases, a cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a human) can be used to identify that mammal as not having cancer. For example, when a fluid sample (e g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a mammal such as a human suspected of having cancer) is determined to have a fluid sample including a cfEII that is not greater than 0.5 and / or not greater than a cfEII of a reference sample, the mammal can be identified as not having cancer. For example, when a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a mammal such as a human suspected of having cancer) is determined to have the absence of (e.g., to lack) ctDNA based, at least in part, on the fluid sample having a cfEII that is not greater than 0.5 and / or not greater than a cfEII of a reference sample, the mammal can be identified as not having cancer.

[0071] In some cases, the cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from a mammal (e.g., a mammal such as a human suspected of having cancer) can be used to select one or more actions for that mammal. For example, a mammal (e.g., a human) identified as having cancer based, at least in part, on having a fluid sample (e.g., a fluid sample containing cfDNA) including a cfEII that is greater than 0.5 and / or greater than a cfEII of a reference sample can be selected for one or more cancer treatments. For example, a mammal (e.g., a human) identified as having cancer based, at least in part, on having a fluid sample (e.g., a fluid sample containing cfDNA) including a cfEII that is greater than 0.5 and / or greater than a cfEII of a reference sample can be selected for to receive additional and / or more frequent diagnostic testing.

[0072] In some cases, a mammal (e.g., a human) identified as having cancer as described herein (e.g., based, at least in part, on a cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from the mammal) can be administered to instructed to selfadminister one or more cancer treatments. For example, a mammal (e.g., a human) identified as having cancer based, at least in part, on a cfEII of a fluid sample (e.g.. a fluid sample containing cfDNA) obtained from the mammal can be administered to instructed to self- administer one or more cancer treatments. When a mammal (e.g., a human) identified as having cancer as described herein (e.g., based, at least in part, on a cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from the mammal) is selected for and / or administered one or more cancer treatments, the cancer treatment(s) can be any appropriate cancer treatment(s). In some cases, a cancer treatment can include administering to the mammal one or more (e.g., one, two, three, or more) anti-cancer agents. For example, an anti-cancer agent can be a chemotherapeutic agent. For example, an anti-cancer agent can be a cytotoxic agent. For example, an anti-cancer agent can be an angiogenesis inhibitor. Examples of anticancer agents that can be administered to a mammal (e.g.. a human) identified as having cancer as described herein (e.g., based, at least in part, on a cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from the mammal) include, without limitation, busulfan, cisplatin, carboplatin, paclitaxel, docetaxel, nab-paclitaxel, altretamine, capecitabine, cyclophosphamide, etoposide (vp-16). gemcitabine, ifosfamide, irinotecan (cpt-11), liposomal doxorubicin, melphalan, pemetrexed, topotecan, vinorelbine, goserelin, leuprolide, tamoxifen, letrozole, anastrozole, exemestane, bevacizumab, olaparib, rucaparib, niraparib, and any combinations thereof. In some cases, a cancer treatment can include subjecting the mammal to one or more (e g., one, two, three, or more) cancer therapies. Examples of cancer therapies that can be used to treat a mammal (e.g., a human) identified as having cancer as described herein (e.g., based, at least in part, on a cfEII of a fluid sample (e.g., a fluid sample containing cfDNA) obtained from the mammal) include, without limitation, immunotherapy, radiation therapies, and / or surgeries (e.g., to remove at least a portion of one or more solid tumors).

[0073] The invention will be further described in the following examples, which do not limit the scope of the invention described in the claims.

[0074] EXAMPLES

[0075] Example 1: Detection ofctDNA using epigenetic instability-based metrics

[0076] This Example describes the development of methods for using the epigenetic instability’ index (Eli) of cell-free DNA (cfDNA) to detect circulating tumor DNA (ctDNA) in body fluids, such as blood. For example, this Example describe that discovery that CGIs subject to high epigenetic heterogeneity and development of measures that capture the heterogeneity in methylation patterns can enable extraction of majority of cancer-specific methylation-based alterations and thus better identify tumor derived DNA.

[0077] An epigenetic instability7(Eli) metric was created to capture this variability and to provide a sample-based score of epigenetic perturbation with clinical translation. The Eli is made up of two measures, each uncorrelated to the other and thus providing different information with regards to the epigenetic state of the patient. The first measure is termed “MHIC”, methylation heterogeneity7in CpG sites (Figure 1 A). This measure captures the proportion of Illumina methylation array probes within a sample that falls within 0.2 to 0.8 beta value (approximates 20-80% methylation), referred to as the “intermediate methylation zone”. Cancer samples had a higher proportion of intermediate zone methylation than normal samples (Figure IB). Secondly , the alternations in methylation within regulatory regions, such as CGI, which are mostly unmethylated in normal samples and thus exhibit a high correlation were measured. To do this, measure termed “VarCGI”, variance in methylation across CpGs in the CGI regions was created (Figure 1 D). This measure captures the variance across all CGI within a sample and then computes the root mean variance of those CGI to give a sample specific measure of methylation alterations. Again, higher root mean variance was seen in tumor samples in comparison to healthy samples (Figure IE).

[0078] To examine which genomic elements exhibited the greatest epigenetic instability, and thus biomarker potential, MHIC was calculated across different cancers in all CGI associated with transcription start sites (TSS), CGI which reside at TSS and marked with both H3K27me3 and H3K4me3 in stem cells (referred to as “bivalent”) and CGI which reside at TSS but are not bivalently marked (Figure IB). The highest increase in sample MHIC was found in bivalent CGI which reside within TSS, highlighting these sites as the sites most likely to exhibit the highest epigenetic instability. To investigate this epigenetic perturbation, the standard deviation of each bivalently marked CGI within the human genome was calculated in order to identify those regions that contributed to an increase in sample MHIC. (Figure 1C). The majority of variation within normal samples did not exceed a SD of 0.2, but this was not the case in the cancer samples, which exhibited vast heterogeneity past this threshold. CGI with SD greater than 0.2 in any of the studied cancers were then deemed likely to be regions which exhibit great epigenetic instability. Although, there were approximately -22,000 of these regions at the CpG level and it is unlikely that each CpG contributed equally to the epigenetic instability of the sample, therefore, feature selection via Boruta was used to select the most informative CGI. This resulted in 269 CGI (referred to as “Boruta HighVarCGI”) covered by approximately -3500 probes which improved, if not maintained, the median distinction between healthy and cancer samples (Figure IF). These Boruta HighVarCGI sites were verified in publicly available tissue-based WGBS (Figure 1G. H).

[0079] While it is useful to be able to distinguish healthy and cancer samples within tissue samples, it is of greater clinical benefit to be able to distinguish these sample groups within cfDNA. An algorithm, VarMethEstimator, was thus developed to adapt the measure of VarCGI so that variance could be calculated per cfDNA fragment read instead of per CGI - termed “ReadV arCGI” (Figure 2A). To this end, publicly available cfDNA- based WGBS from different stages of breast cancer patients (30 early / non-metastatic samples and 30 advanced / fully metastatic samples) in addition to healthy controls (60 samples) were obtained. This meant that every individual read available across the genome from the WGBS data could be utilized instead of a traditional method of estimating the cumulative percentage measurement to represent the methylation of all reads at one particular CpG. Using ReadVarCGI to compute variance across each CGI for the 269 Boruta HighVarCGI and plotting as a cumulative distribution function (CDF, Figure 2F). a discrimination was seen between healthy, early breast cancer samples and advanced breast cancer samples, especially in comparison to a CDF computed solely using the traditional methylation values of those Boruta HighVarCGI regions (Figure 2G). A similar distinction between healthy and different stages of lung cancer was found using the same Boruta HighVarCGI regions and the same style of variance computations with ReadVarCGI (Figure 2B) in comparison to a CDF computed solely using the traditional methylation values of the Boruta HighVarCGI regions (Figure 2C). Throughout the range of readV arCGI or mean methylation values computed for the CGI regions, the former shows a clear separation of cancer cases from normal indicating the number of cfDNA fragments that can be positively identified to be abnormal, i.e., ctDNA, was better captured using the Eli metric.

[0080] The WGBS data from publicly available cfDNA samples used above was of very’ low coverage, which allowed analyses of only -30% of the 269 CGI in any given sample. To combat against bias from small cfDNA fragments the ReadVarCGI was conducted only on reads with 4 or greater CpG among them, which highly limits the number of fragments available for analyses. Even with these limitations the Ell-based ReadVarCGI performed beter in discriminating cancer from healthy, even early-stage cancer, than traditional methylation-based approach. As the proportion of ctDNA in cfDNA ranges from 0.01%, increasing coverage of reads for the target CGIs is expected to improve analytical sensitivity, which is achievable given current advances in methylation profiling and target enrichment protocols.

[0081] Example 2: Protocol for developing cfEII metrics of ctDNA

[0082] This Example describes an exemplary method for determining a cfEII of ctDNA. The Examples described herein are based on (a) the MHIC of a set of 269 CGI genomic regions, and (b) the VarCGI of those genomic regions, however, the same protocols can be used when assessing regions other than the 269 CGIs.

[0083] From Illumina 450k or EPIC array data

[0084] 1) Data were processed to obtain a beta value matrix (rows equal CG sites, columns equal samples). Any currently available program can be used to do this, popular programs include minfi, RnBeads, ChAMP.

[0085] 2) When the beta matrix contained multiple cancer ty pes and their subsequent healthy samples, the data were parsed by cancer type and by disease state of the sample (normal, tumor).

[0086] 3) Beta value matrix table was further transformed to remove non-cancerous cell ty pe methylation paterns using various deconvolution approaches (e.g., surrogate variable analysis (SV A)).

[0087] 4) The measures of epigenetic instability were calculated: a. MHIC calculates the proportion of CG sites across the genome in a sample which register a beta value of between 0.2 - 0.8. Since cancer samples are characterized as a gain of methylation across time, normal samples should have a lower proportion of CG sites with a beta value of 0.2 - 0.8 than tumor samples. Any beta value range greater than 0.2 to any value less than 1 can be used; such as 0.3 to 0.7. MHIC was calculated within the sample and produced a sample specific score of epigenetic instability7.

[0088] MHIC is represented in Figure 1 A, where the filled pins indicate sites with a methylation of between 0.2 and 0.8 and the open pins indicate a methylation value outside of 0.2-0.8. MHIC was calculated within the sample itself as opposed to calculating changes in methylation between CpG sites across samples. b. VarCGI calculates the variance in methylation at regulatory regions, such as CGIs and is another within sample metric of epigenetic instability’. In the human genome, the majority of CpG Islands are unmethylated, it is generally only in disease states such as cancer where gains in methylation within the CGI itself are seen. If all sites in the CGI are unmethylated the variance in that CGI’s methylation was 0. However, if CG sites within the CGI start to gain methylation, the variance of that CGI increased. Minor alterations to CGI methylation are acceptable and happen during aging, so to correct for this, VarCGI measured the variance in methylation across each CGI in the human genome having data in an acceptable range. CGIs vary in length throughout the human genome so in order to characterize the CGIs effectively, a criterion of 10 or more CG sites reporting methylation was imposed for inclusion of that CGI into the VarCGI calculation. Once the variance for each acceptable CGI was computed per sample, this was formed into a sample specific score by taking the root mean variance of those individual variance scores. VarCGI is characterized in Figure ID, where the filled pins indicate methylated CGIs and the open pins indicate unmethylated CGIs. Cancerous samples had a higher degree of VarCGI than normal samples.

[0089] From reduced representation bisulfite sequencing (RRBS) or from whole genome bisulfite sequencing (WGBS) on tissues

[0090] MHIC and VarCGI worked very similarly to how they work in array technologies. There was a higher genome-wide coverage of CG sites in RRBS or WGBS and there was no beta matrix that hailed from RRBS or WGBS analysis. The representation of methylation in this case was a BED file (chromosome, start site, end site, methylation as a beta value). All BED files were read into R (the current version is R 4.x.x, but any recent version is acceptable) and a long format of a beta matrix was created.

[0091] From RRBS or WGBS on cfDNA

[0092] MHIC and VarCGI calculated on cfDNA yielded differences between different stages of cancers and normal samples. These metrics were based on the small amount of sequencing reads derived from circulating tumor DNA (ctDNA) from the cfDNA. Therefore, an alternative metric of Eli from cfDNA was developed to extract as much information as possible from the cfDNA sequences. This technique is referred to as cfEII or cell free Eli (cfEII).

[0093] 1) To calculate cfEII the methylation of individual sequencing reads within CGI of interest was extract using with the RLM package.

[0094] 2) Exclude any sequencing read that has less than 4 CG sites on it.

[0095] 3) Obtain reads that map to CGIs of interest and calculate the variance across the reads that map to those CGI. These variance measures per sample represent that sample’s cfEII. Figure 3 represents the process of calculating cfEII.

[0096] 4) Plot these values as a Cumulative Distribution Function with comparison to the normal samples.

[0097] When compared to the normal / healthy samples, even early-stage cancer samples displayed higher values for cfEII and reached statistical significance.

[0098] Genomic regions that capture majority of cancer-specific epigenetic instability (269 CGI regions):

[0099] The approach of using the cfEII metric to discriminate cancers (including early- stage breast and lung cancer) can be used agnostically for any set of CGIs from even a low coverage cfDNA data. However, the cfEII metric was estimated from a smaller set of defined set of genomic regions to get a more robust and reproducible implementation. The smallest set of genomic regions that sufficiently captured the epigenetic instability across various cancer types were identified using the below method:

[0100] 1) a matrix was created with the standard deviation (SD) of beta values across all bivalently marked CGI (rows = patients, columns = CGI)

[0101] 2) a column was added detailing whether the sample was normal or cancerous

[0102] 3) this data was input into feature selection via the Boruta package

[0103] 4) the function TentativeRoughFix was used to classify features with undecided classifications

[0104] 5) print.Boruta was used to get the CGI classified as informative. 269 CGIs for use in determining a Eli such as a cfEII were identified. These 269

[0105] GCIs are shown in Table 1. Table 1. Exemplary genomic CGIs.

[0106] Example 3: Protocol for cfDNA extraction

[0107] This Example describes an exemplary method for extracting cfDNA from a blood sample. cfDNA Extraction Protocol

[0108] Whole blood was collected in BD Vacutainer EDTA tubes or PAXgene blood ccfDNA tubes and was processed within two hours of collection. Plasma and cellular components were separated via centrifugation at 800g for 10 minutes at 4 degrees Celsius. Plasma was spun again at 17,000g for 10 minutes at room temperature. Plasma was processed immediately or aliquoted into 1 mL aliquots and stored at -80 degrees Celsius. DNA was isolated from plasma using the Qiagen QIAamp Circulating Nucleic Acids Kit and eluted in LoBind tubes. Concentration and quality of cfDNA were determined using the Bioanalyzer 2100.

[0109] QIAamp Circulating Nucleic Acids Kit

[0110] In brief, l-5mL of plasma was combined with appropriate quantities of Proteinase K and prepared ACL Buffer with carrier RNA, vortexed for thirty seconds, and incubated at 60 degrees Celsius for 30 minutes. ACB buffer was added and the lysate was vortexed and placed on ice for five minutes. The lysate was applied to the QIAamp Mini column and column extender. Vacuum pressure of -800 to -900 mbar w as applied to the column until the lysate was completely drawn through. Subsequently vacuum treatments were performed after each of 600 LIL of ACW1 buffer, 750 pL of ACW2 buffer, and 750 pL of 100% Ethanol was applied to the column. The QIAamp Mini column was placed in a wash tube and centrifuged at 17,000g for 3 minutes. The column was placed in anew wash tube and dried at 65 degrees Celsius for 10 minutes. Lastly, the column was placed in a LoBind tube and 54 pL of AVE buffer was applied, incubated for 3 minutes at room temperature, and centrifuged at 17,000g for 3 minutes.

[0111] Sample Preparation for Bisulfite Sequencing

[0112] Next-generation bisulfite sequencing libraries were prepared using the NEBNext Low Input Enzymatic Methyl-seq kit (Development) without modification. In brief, cfDNA w as mixed with NEBNext Ultra II End Prep Reaction Buffer and NEBNext Ultra II End Prep Enzyme Mix and incubated for 15 minutes at 20 degrees Celsius then 15 minutes at 65 degrees Celsius. Next. NEBNext EM-seq adaptor, NEBNext Ligation Enhancer, and NEBNext Ultra II Ligation Master Mix were added to the mixture and were incubated at 15 minutes at 20 degrees Celsius. The reaction w as purified using NEBNext Sample purification beads in a 1 : 1 ratio to reaction volume. Oxidation of cfDNA was performed by mixing adaptor-ligated cfDNA with TET2 Reaction Buffer, Oxidation Supplement. DTT, TET2 enzyme, Diluted Oxidation Enhancer, and 500 mM Fe(II) with subsequent incubation for one hour at 37 degrees Celsius. NEB Stop reagent was then added and the mixture was incubated for 30 minutes at 37 degrees Celsius. The oxidation reaction was purified using NEBNext Sample purification beads in a 1 : 1 ratio to reaction volume. Denaturation of DNA was performed with Formamide for 10 minutes at 85 degrees Celsius. Denatured DNA was mixed with Deamination Reaction Buffer, Recombinant Albumin, and APOBEC for three hours at 37 degrees Celsius. Immediately following deamination, PCR amplification was performed using NEBNext Q5U Polymerase with EM-seq Index Primers. The reaction was purified using NEBNext Sample purification beads in a 1 : 1 ratio to reaction volume. Concentration and quality7of cfDNA libraries were assessed using the Bioanalyzer 2100.

[0113] Example 4: Epigenetic Instability Based Metrics for Early Cancer Detection

[0114] The results in this Example re-present and expand on at least some of the results provided in other Examples.

[0115] This Example describes the development of methods for using the degree of epigenetic perturbation, termed the epigenetic instability7index (Eli), for multiple cancer screening through cfDNA. These studies demonstrate that the Eli performs better in identifying cancer cases from cfDNA than standard approaches based on absolute methylation values.

[0116] Results

[0117] Eli analyses improve the distinction between normal and tumor samples

[0118] Hypermethylation of promoter CGIs is a well described cancer-specific epigenetic alteration. Inspection of the methylation gains in individual cancer samples shows that the majority of methylation gains are heterogenous at individual CpG sites. For example, plotting the per sample methylation distribution of CpG sites within CGIs for colon cancer (TCGA COAD) revealed this heterogeneity phenomenon, with the majority7of the changes falling in an intermediate range of 20 to 80% (Figure 3A). Further, these heterogenous gains were more prominent at poly comb marked CGIs that are prone to undergo cancer-specific hypermethylation (Easwaran et al.. Genome Res., 22(5):837-49 (2012)). In contrast to the above dynamics, most normal samples have little methylation at the above sites, but each COAD sample exhibited heterogeneity7within these regions, the extent of which is not captured by the locally weighted smoothing estimate (Figure 3B). In concordance with the heterogenous methylation in each sample, the mean methylation heterogeneity across CGIs within individual samples showed wide variation in the colon adenocarcinoma (COAD) samples, as well as in other cancer types, such as breast (BRCA), glioblastoma (GBM), lung adenocarcinoma (LUAD) and pancreatic adenocarcinoma (PAAD) compared to the corresponding normal samples (Figure 3C). Tumors have a heterogenous mix of cells, such as normal cells and infiltration of immune cells noted in the TCGA datasets (Chakravarthy et al., Nat Commun., 9(1 ):3220 (2018)). Surrogate variable analyses did not reveal hidden surrogate variables for any of the datasets analyzed, indicating that cell ty pe heterogeneity', while it exists, is not a significant component of the methylation heterogeneity. Thus, the heterogenous methylation changes mainly represent the methylation heterogeneity associated with tumor state.

[0119] Metrics to capture the epigenetic perturbation in a given tumor sample based on the high DNAme variability at the individual CGIs, and without need for normal reference samples, were defined. Two measures were employed: Methylation Heterogeneity In CpG sites (MHIC) and Variation in CpG sites across the CGI (VarCGI), which showed clear separation of normal and tumor samples (Figures 3G and 3H). MHIC measures methylation heterogeneity in CpG sites as the proportion of methylation between 0.2 - 0.8 b-values (schematic shown in Figure 3D), where b represents the approximate percent methylation at individual CpG sites measured by the methylation array probes, b-values between 0.3 - 0.7, representing less extreme definition of intermediate methylation, also captured a similar distinction between healthy and cancerous samples (Figure 10A). The MHIC measure was mainly used to explore the shift in methylation from the unmethylated or methylated state of the CpGs. The second measure, VarCGI (schematic shown in Figure 3E), captured the variation in the methylation of individual CpG residues within a CGI, which was expected to be high in cancerous samples (Figure 9B) due to the heterogeneity in methylation alterations at individual CpG sites in the CGI region, but not in healthy samples. When assessing the relationship between MHIC and VarCGI, the measures were less correlated to each other than their correlations with the mean methylation (mean-b) of the corresponding CGI, indicating that each measure may separately provide information relevant to the epigenetic instability in tumor samples from a subject (Figures 9A, 9B, and 9C). The MHIC (Figure 3G) and VarCGI (Figure 3H) scores computed for the same poly comb marked CGI regions showed differences between normal and cancer samples, similar to the mean-b measurements for all the cancer types assessed (Figure 3C). Further, the CGI regions with highest epigenetic instability in cancers, identified as the top 100 CGIs with the highest SD in the cancer samples, represented regions with a broad range of methylation changes with normal samples generally having low methylation (Figure 9D); while CGIs representing highest methylation in the cancer samples, identified as the top 100 CGIs with highest methylation, had lower SD values with some of these regions also exhibiting high methylation in normal samples (Figure 9E). These data suggest that CGIs that exhibited high methylation variability were distinct CGI regions not highlighted in traditional approaches that consider mean-B methylation alone, thus providing novel regions for identifying cancer related epigenetic changes.

[0120] Examination of MHIC and VarCGI in different genomic elements to elucidate regions with the most informative epigenetic variation using TCGA data

[0121] To test the utility of epigenetic instability in distinguishing cancers from normal samples, regions with most unstable methylation patterns were identified in the TCGA dataset. As stated earlier, poly comb-marked CGI regions represent a subset of the CGI exhibiting heterogenous methylation gains within individual samples (Figure 3B). This was also observed in the increased MHIC score for the poly comb-marked CGI-TSS (Figure 4A). To capture the most informative epigenetically unstable regions, the SD of each CGI in each of the 179 normal and 1905 tumor samples from TCGA were input into Boruta feature selection to discriminate the normal and cancer samples which resulted in 553 informative CGI corresponding to ~8K probes from the methylation array. The MHIC and VarCGI for these probes resulted in a distinction between normal and tumor samples in the TCGA dataset, which was the same if not better in some cases than using all poly comb marked CGI probes (Figures 4B and 4C). As mentioned previously, the poly comb marked CGI were among most likely to undergo methylation alterations during carcinogenesis. Therefore, to get a smaller set of regions with highest discriminatory power, CGIs w ere further filtered with standard deviation (SD) of methylation across the probes in that region greater than 0.2 (Figures 4D and 4E) and Boruta feature selection was used for those CGI to obtain regions with a higher threshold of variation compared to normal samples, resulting in 269 informative CGI (-3.5K probes, referred to as the B- 269 CGI”) which sufficiently captured the epigenetic instability for a given sample and consequently separated cancer and normal samples (schematic of the feature selection process shown in Figure 4F, Figure 4G - MHIC, and Figure 4H - VarCGI).

[0122] B-269 CGIs discriminate tumors from normal samples in WGBS data as well as in arrays

[0123] Methylation results w ithin the TCGA data assessed were based on Illumina HumanMethylation450 BeadChip array technology, which assessed genome wide methylation at -450,000 sites, with a mean density of 7 probes in the CGIs, with these sites primarily focused on promoter / CGI areas. While the array has been shown to be representative of genome wide methylation in normal and cancer samples (Muller et al., Genome Biol., 20(l):55 (2019)), it is important to ensure that the measures of epigenetic instability and epigenetically unstable regions were also representative in independent non-array based higher resolution technologies. To this end, MHIC and VarCGI for the B-269 CGI were computed in WGBS data from healthy / lung cancer (n=7) and healthy / liver cancer samples (n=l l) (Jenkinson et al., Nat Genet., 49(5):719-29 (2017)). These samples were chosen due to availability of high-quality (majority of sequences with a MAPQ >= 30 and normal samples without histories of cancer) WGBS data from normal and tumor tissue samples. VarCGI and MHIC measures for the B-269 CGI clearly distinguished cancer and normal samples (Figures 5A and 5B), verifying the epigenetic instability- concept in higher resolution technologies from independent datasets. It was found that comparing mean methylation values at the B-269 CGI was able to maintain if not improve the distinction between tumor and normal samples than the CGI containing the most significant differentially methylated CpG sites from our TCGA investigation (Figures 5C and 5D).

[0124] Epigenetic Instability metric performs better than mean methylation in identifying cancer samples from cfDNA

[0125] While MHIC and VarCGI showed capabilities in distinguishing healthy and cancerous tissue samples, this capability would be of even greater clinical benefit if the epigenetic instability could be detected using cfDNA. To investigate this, WGBS of cfDNA isolated from plasma of 30 patients without cancer, 15 patients with early non- metastatic breast cancer, and 15 patients with advanced metastatic breast cancer (Gao et al., Clin TranslMed., 12(8):el014 (2022)) was analyzed. Given the low genomic coverage of cfDNA in typical WGBS experiments, and to enable the maximum amount of information to be obtained from the WGBS reads, VarCGI was adapted to allow the calculation of CGI methylation variance from individual sequencing reads followed by summarizing the variance metric as the root mean of the individual read variances as in the original definition of VarCGI (Figure 6A). VarCGI was then computed within the cfDNA of normal, early, and advanced breast cancer samples across the most variable B- 269 CGI discussed previously, with coverage parameters of at least 5 CpGs in the sequencing read for it to be considered in the variance calculation (Figure 6B). In comparison to mean-b measures across those same B-269 CGI (Figure 6C) or for the CGIs with the most differentially methylated CpG sites (Figure 1 1 A). VarCGI better differentiated healthy, early breast cancer, and advanced breast cancer samples (Figure 6B).

[0126] The VarCGI metric for cfDNA analysis was enhanced by directly computing only the variance in methylation at CpG sites across individual sequencing reads containing at least 5 CpGs for each of the B-269 CGI. This w as done for all the B-269 CGI for which sequenced reads were available in the dataset, with the hypothesis that the root mean variance step of the VarCGI calculation was minimizing the effect of epigenetic alterations in the small pool of cfDNA reads available. To avoid confusion, this metric is referred to as ‘VarRead’ (Figure 6A, bottom arrow for schematic of calculation). Following the calculation of VarRead, the results were plotted as a cumulative distribution function (CDF) between the sample types of healthy, early breast cancer, and advanced breast cancer. The breast cancer samples exhibited higher probability of highly variable reads (Figure 6D, combined cancer stages found in Figure 1 IB), in comparison to the healthy samples. The VarRead metric was better able to separate breast cancers from healthy cfDNA (Figure 6D) compared to similar assessment of mean methylation across the B-269 CGI (Figure 6E). and also for the CGI with the top differentially methylated (DM) CpG sites derived from the TCGA datasets (Figure 11C). These differences between epigenetic instability based-VarRead and traditional methylation metrics was most striking for the early breast cancers in comparison to healthy cfDNA samples.

[0127] To validate the performance of VarRead in another cancer type, similar analyses were performed on publicly available targeted deep coverage BS-seq data of cfDNA from different stages of lung cancer ( Liang et al., Nat Biomed Eng., 5(6):586-99 (2021)). Similar to the findings in breast cancer data, the different stages of lung cancer (stage IA, IB. Il, 111) were better distinguished using the VarCGI (Figure 7A for B-269. Figure 12D for DM sites identified from our analyses of the TCGA data, as well as the DM sites discovered in the original study in which the lung cfDNA dataset originated) and VarRead (Figure 7B) approach compared to the mean methylation of the B-269 CGI regions (Figure 7C, combined stages in Figure 12A), the most differentially methylated sites based on TCGA-LUAD data (Figures 12B and 12E) or the most differentially sites discovered in the Liang et al. study from which the data was obtained (Liang et al., Nat Biomed Eng., 5(6):586-99 (2021)) (Figures 12C and 12E).

[0128] It was further tested if the B-269 CGIs represent the best targets for cfDNA based discrimination of cancer from healthy samples. To this end alternative regions of the genome beyond the B-269 regions used above were assessed using VarCGI metric (Figure 6A, schematic of the calculation), mean methylation across those regions, and VarRead (Figure 4D, schematic of the calculation). The B269 CGIs performed the best in separating cancers from healthy (Figures 6D, 7A, and 7B) compared to regions representing all the CGIs (Figure 13A), randomly picked 269 CGI (Figure 13B), PcG- marked CGIs (Figure 13C) or randomly picked 269 PcG-marked CGIs (Figure 13D). Especially, in comparison to traditional methylation values for all the different CGI categories analyzed (Figures 13A-13D, right-two panels), the VarCGI metric (Figures 13A-13D, left-two panels) was better at discriminating not only late but also early-stage breast and lung cancers from healthy samples. Thus, the epigenetic instability metric of VarCGI applied to the B-269 regions identified from TCGA array data proficiently identified cancer samples from cfDNA methylation profiles.

[0129] It is interesting to note that VarRead and VarCGI values better discriminated cancers compared to DNA methylation values, even when the most significantly cancerspecific differentially methylated CGIs were considered (Figures 12A-12E). To probe this further, the methylation densities of the B-269 CGI and the cancer-specific differentially methylated CGIs were assessed and showed differing densities of methylation between both the 2 sets of regions and between the normal and stages of cancer (Figure 1 ID - breast cfDNA, Figure 12F - lung cfDNA). Read length density plots indicated no coverage bias for the Boruta 269 or the differentially methylated regions (Figures 14A- 14B). These analyses showed that the regions identified to have high Eli capture CGIs that tend to have more DNA methylation heterogeneity in patient cfDNA samples, which the Ell metric was able to capture to discriminate cancer samples.

[0130] Classifier models using the Eli metrics perform better than DNA methylation-based metrics for cfDNA evaluation

[0131] To test if VarRead measurements could accurately classify Healthy and Lung cancer samples within the B-269 CGI, a random forest classifier was built (80:20 train to test ratio, 5-fold cross validation performed to ensure similar AUC results obtained throughout, further details in Figure 8 A) which achieved an AUC of 0.96 and approximately 80% sensitivity at 95% specificity (Figure 8C). AUC and specificity / sensitivity measurements were lower for random forest classifiers built using mean methylation measurements across the CGI identified to be differentially methylated in Liang et al. (Liang et al., Nat Biomed Eng., 5(6):586-99 (2021)) or the differentially methylated CGIs identified from TCGA data (Figure 8C). Similar classifiers were also built for the breast cancer dataset with the VarRead measurements across the CGI achieving a higher AUC (0.87) and sensitivity at 95% specificity than differentially methylated sites identified from TCGA data (AUC = 0.77, Figure 8B). By stage, random forest classifiers were also built for the breast and lung cancer cfDNA data and achieved a similar result (Figure 8D - Healthy vs Stage I NSCLC; AUC = 0.92, Figure 8E - Healthy vs Stage II & III NSCLC; AUC 0.99, Figure 8F -Healthy vs Early Breast Cancer; AUC = 0.74, Figure 8G - Healthy vs Advanced Breast Cancer; AUC=0.76). These results indicate that epigenetic instability metrics, such as VarRead, provide significant advancements for the examination of DNA methylation dynamics during carcinogenesis and detection of cancers, even early stages, in comparison to metrics based on capturing absolute methylation values.

[0132] Methods

[0133] Genomic Annotations

[0134] Genomic intervals of bivalent CGI were obtained from Easwaran et al. (Easwaran et al.. Genome Res., 22(5):837-49 (2012)). CGI annotations and TSS were obtained from UCSC genome browser for Human GRCh37 / hgl9. Promoters were defined as 1500bp upstream and downstream of a TSS.

[0135] Analysis of TCGA Data

[0136] Level 1 TCGA Illumina HumanMethylation450 BeadChip Array data was downloaded from TCGA for BRCA, COAD, GBM, LU AD and PAAD cancers and normal. Data was loaded into R 4.1.3, preprocessed to remove all probes on the X and Y chromosomes due to potential cross hybridization risks, probes which reside on SNPs were removed and those with detection p-values > 0.05 were removed. Probes on the Illumina probe blacklist for the HumanMethylation450 BeadChip Array were also removed (Pidsley et al., Genome Bio , 17(l):208 (2016)). Data was normalized using Functional Normalization and background corrected using Noob (Minfi vl.42.0). Differential methylation analysis was carried out using Limma (version 3.52.2), differential sites were considered to be those with a q value of < 0.05 and delta beta greater than 0.2 or less than -0.2.

[0137] For surrogate variable analysis, the full model used health status (NT. TP) as the variable of interest and the null model used array position and slide ID. Number of latent factors was determined by the SVA package. For this, both models and a beta matrix were input into the SVA function recursively for each TCGA cancer analyzed. No surrogate variables were found.

[0138] Analysis of WGBS

[0139] Raw fastq files of tissue-based WGBS of Lung / Liver cancer samples and associated normals were downloaded from GSE86340 and processed through Bismark (version 0.23.1). Duplicate sequences were removed and only alignments with a MAPQ > 30 were retained. Raw cfDNA-based bisulfite sequencing files were obtained from PRJNA494975 (WGBS) and PRJNA534206 (targeted sequencing), aligned using Bismark (version 0.23.1) with duplicates removed and only alignments with MAPQ > 30 retained. Output BAM files were then input into RLM (version 1.10) for the extraction of read-level methylation.

[0140] Statistical Considerations

[0141] All statistical analyses were performed using R 4. 1.3. All statistical tests are two- tailed. In instances where dimensionality of the data would make statistical testing difficult, data was bootstrapped around the mean and the resulting distribution for each test case was used to generate p values. R packages caret (version 6.0-94) and randomforest (version 4.7- 1.1) were used to obtain the classification of normal vs cancerous samples. Areas Under the Curve (AUC) and Receiver Operating Characteristics (ROC) were obtained using pROC (version 1.18.0).

[0142] Definitions of Eli metrics

[0143] MHIC: The MHIC metric calculates the percentage of probe sites measured as intermediate methylation (0.2 to 0.8 methylation) in selected regions. At represents the probe sites, and x is the average methylation of all CpGs in that region. 0.8

[0144] VarCGI: VarCGI estimates the overall variance of methylation sites across selected CpG islands for each sample. Here X means the methylation level at CpG i in

[0145] CpG island j, and u is the average methylation of all CpGs in the CpG island.

[0146] VarCGI (cf): VarCGI (cf) is derived from the VarCGI, but it calculates the variance at the read level rather than CpG level. Here, the methylation status of each read is extracted, and it should follow the Bernoulli distribution. The parameter ry means the i th read in CpG island

[0147] VarRead is also derived from VarCGI, but the distinction is that there is no root mean variance step included. Variance is reported at the read level and results in a vector of read level variance scores of reads aligning to CGI regions. VarRead also follows the Bernoulli distribution and parameter means the i th read in CpG island / .

[0148] Data Availability

[0149] All the datasets used in this study are publicly available from the NCBI Sequence Read Archive database (ncbi.nlm.nih.gov / sra) or from The Cancer Genome Atlas (portal.gdc.cancer.gov). The Lung / Liver cancer dataset used in this study is available under accession number GSE86340. The cfDNA Breast cancer dataset is available under accession number PRJNA494975. The cfDNA Lung cancer dataset is available under accession PRJNA534206. Code Availability

[0150] Any code or scripts utilized as part of this project can be found at github . com / sj thursby / Eli .

[0151] OTHER EMBODIMENTS

[0152] It is to be understood that while the invention has been described in conjunction with the detailed description thereof the foregoing description is intended to illustrate and not limit the scope of the invention, which is defined by the scope of the appended claims. Other aspects, advantages, and modifications are within the scope of the following claims.

Claims

WHAT IS CLAIMED IS:

1. A method for determining a cell free epigenetic instability index (cfEII) of a fluid sample comprising cell-free DNA (cfDNA), said method comprising: determining a variance of methylation of a plurality of sequencing reads of CpG sites in CpG-islands (ReadVarCGI) of said cfDNA, wherein sequencing reads having fewer than 4 CpG sites are excluded, wherein said plurality of sequencing reads comprise sequencing reads of at least 15% of the CpG-islands set forth in Table 1; and wherein said variance represents the cfEII of said fluid sample.

2. The method of claim 1, wherein said determining said ReadVarCGI comprises calculating a variance across a cfDNA read that maps to a CGI of interest and has more than 4 CpG sites within the read.

3. The method of any one of claims 1-2, wherein each of said sequencing reads is less than 160 base pairs in length.

4. The method of any one of claims 1-2, wherein said plurality of sequencing reads comprise sequencing reads of from about 15% to about 30% of the CpG-islands set forth in Table 1.

5. The method of any one of claims 1-4, wherein said fluid sample is selected from the group consisting of a whole blood sample, a plasma sample, a serum sample, a urine sample, a cerebrospinal fluid sample, and a saliva sample.

6. The method of any one of claims 1-5, wherein said sample is obtained from a mammal.

7. The method of claim 6, wherein said mammal is a human.

8. The method of any one of claims 1-7, further comprising plotting a cumulative distribution function of said ReadVarCGI.

9. The method of any one of claims 1-8, wherein the method comprises using a machine learning (ML) system.

10. The method of claim 9, wherein the machine learning system is a Random Forest based ML algorithm.

11. A method for detecting circulating tumor DNA (ctDNA) in a fluid sample comprising cfDNA obtained from a mammal, said method comprising:(a) determining a cfEII of said cfDNA;(b) identifying said cfDNA as comprising ctDNA when said cfEII is greater than a cfEII of a reference sample.

12. The method of claim 11, wherein said fluid sample is selected from the group consisting of a whole blood sample, a plasma sample, a serum sample, a urine sample, a cerebrospinal fluid sample, and a saliva sample.

13. The method of any one of claims 11-12, wherein said mammal is a human.

14. A method for assessing a mammal, wherein said method comprises:(a) determining a cfEII of a fluid sample obtained from said mammal;(b) identifying the mammal as having cancer when said cfEII is greater than a cfEII of a reference sample.

15. A method for assessing a mammal, wherein said method comprises:(a) determining a cfEII of a fluid sample obtained from said mammal;(b) identifying the mammal as not having cancer when said cfEII is not greater than a cfEII of a reference sample.

16. The method of any one of claims 14-15, wherein said fluid sample is selected from the group consisting of a whole blood sample, a plasma sample, a serum sample, a urine sample, and a cerebrospinal fluid sample.

17. The method of any one of claims 14-16, wherein said mammal is a human.

18. The method of any one of claims 14-17, wherein said cancer comprises one or more solid tumors.

19. The method of any one of claims 14-18, wherein said cancer is selected from the group consisting of a breast cancer, a lung cancer, a liver cancer, a colorectal cancer, a prostate cancer, a brain cancer, a skin cancer, a head and neck cancer, a pancreatic cancer, a genitourinary cancer, a gynecologic cancer, a gastrointestinal cancer, sarcomas, and a hepatobiliary cancer.

20. A method for treating for a mammal having cancer, wherein said method comprises:(a) determining that a cfEII of a fluid sample obtained from said mammal is greater than a cfEII of a reference sample; and(b) administering a cancer treatment to said mammal.

21. A method for treating cancer, wherein said method comprises administering a cancer treatment to a mammal identified as comprising a fluid sample having a cfEII having a variance greater than a cfEII of a reference sample.

22. The method of any one of claims 20-21, wherein said fluid sample is selected from the group consisting of a whole blood sample, a plasma sample, a serum sample, a urine sample, and a cerebrospinal fluid sample.

23. The method of any one of claims 20-22, wherein said mammal is a human.

24. The method of any one of claims 20-23, wherein said cancer comprises one or more solid tumors.

25. The method of any one of claims 20-24, wherein said cancer is selected from the group consisting of a breast cancer, a lung cancer, a liver cancer, a colorectal cancer, a prostate cancer, a brain cancer, a skin cancer, a head and neck cancer, a pancreatic cancer, a genitourinary cancer, a gynecologic cancer, a gastrointestinal cancer, sarcomas, and a hepatobiliary cancer.

26. The method of any one of claims 20-25, wherein said cancer treatment comprises performing surgery.

27. The method of any one of claims 20-26, wherein said cancer treatment comprises radiation therapy.

28. The method of any one of claims 20-27, wherein said cancer treatment comprises administering, to said mammal, an anti-cancer agent selected from the group consisting of a chemotherapy, and an angiogenesis inhibitor.

Images