Assessing and treating cancer

Abstract

This document provides methods and materials for assessing and / or treating mammals (e.g., humans) suspected of having cancer. For example, methods and materials for identifying a mammal as having cancer are provided. In some cases, the representation of particular Alu elements (e.g., a reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction) in cell-free DNA (cfDNA) in a fluid sample obtained from a mammal (e.g., a mammal suspected of having cancer) can be used to identify the mammal as having cancer.

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. 63 / 538.719, filed on September 15, 2023. The disclosure of the prior application is considered part of, and is incorporated by reference in, the disclosure of this application.

[0004] STATEMENT REGARDING FEDERAL FUNDING

[0005] This invention was made with government support under grant CA006973 awarded by the National Institutes of Health. The government has certain rights in the invention.

[0006] SEQUENCE LISTING

[0007] This application contains a Sequence Listing that has been submitted electronically as an XML file named “44807-0467W01_SL.xml.” The XML file, created on August 8, 2024. is 2,149 bytes in size. The material in the XML file is hereby incorporated by reference in its entirety.

[0008] TECHNICAL FIELD

[0009] 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 identifying a mammal as having cancer. In some cases, the representation of particular Alu elements (e.g., a reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction) in cell-free DNA (cfDNA) in a fluid sample obtained from a mammal (e.g., a mammal suspected of having cancer) can be used to identify the mammal as having cancer.

[0010] BACKGROUND

[0011] Alu elements are short interspersed nuclear elements (SINEs) of approximately 300 base pairs (bp), with more than 1 million copies spread throughout the genome (Deininger, Genome Biol. 12: 1-12 (2011)). In cancer cells, SINEs can participate in structural changes (Deininger et al., Mol. Genet. Metab. 67: 183-193 (1999)). Moreover, Alu elements can be hypomethylated early during tumor progression (Feinberg et al., Nat. Rev. Cancer 4: 143-153 (2004); Rodriguez el al., Nucleic Acids Res. 36:770-784 (2008); Daskalos et al., Ini. J. Cancer 124:81-87 (2009); Cho et al., J. Pathol. 211 :269-277 (2007); Choi et al, Mod. Pathol. 20:802-810 (2007); Richards et al., PLOS ONE .e (2009); and Hunt et al., Sci. Rep. 12:5776 (2022)), and this feature has been incorporated into methods for the earlier detection of cancer through plasma cfDNA analysis (Zhou et al., Proc. Natl. Acad. Sci. 119:e2209852119 (2022)). Alu elements can also reflect the altered fragmentation patterns found in cfDNA in cancer patients - one of the first plasma multi-cancer biomarkers used qPCR to calculate the ratio of short and long Alu segments (Agostini et al., Ann. Surg. Oncol. 18:2461-2468 (2011); Mead et al.. Br. J. Cancer 105:239-245 (2011); and Iqbal et al., SpringerPlus 4:265 (2015)).

[0012] SUMMARY

[0013] This document provides methods and materials for assessing and / or treating mammals (e.g.. humans) having cancer. In some cases, this document provides methods and materials for identifying a mammal as having cancer. For example, the representation of particular Alu elements (e.g., a reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction) in cfDNA in a fluid sample obtained from a mammal (e.g., a mammal suspected of having cancer) can be used to identify the mammal as having cancer. In some cases, this document provides methods and materials for treating a mammal identified as having cancer. For example, a mammal identified as having cancer as described herein (e.g., based, at least in part, on a reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction in cfDNA in a fluid sample obtained from the mammal) can be administered one or more cancer treatments. As demonstrated herein, fluid samples from mammals (e.g.. humans) having cancer include a reduced AluS fraction in cfDNA as compared to the AluS fraction in cfDNA in fluid samples from mammals that do not have cancer (e.g., healthy mammals) (Figure 5). As demonstrated herein, fluid samples from mammals (e.g., humans) having cancer include an increased AluJ fraction in cfDNA as compared to the AluJ fraction in cfDNA in fluid samples from mammals that do not have cancer (e.g., healthy mammals).

[0014] Having the ability to determine 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 reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction in cfDNA in a fluid sample 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 specificity7and in a non-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 reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction in cfDNA in a fluid sample obtained from the mammal) can be performed on a fluid sample obtained by a liquid biopsy.

[0015] In general, one aspect of this document features methods for assessing a mammal where the methods can include, or consist essentially of, (a) determining a representation of Alu elements in cell-free DNA (cfDNA) in a fluid sample obtained from the mammal; and (b) identifying the mammal as having cancer if the representation of Alu elements includes a presence of at least one of (1) a reduced AluS fraction. (2) an increased AluJ fraction, and (3) an increased AluY fraction. The mammal can be a human. The representation of Alu elements can be determined using whole genome sequencing, and the reduced AluS fraction can be less than about 0.66, the increased AluJ fraction can be greater than about 0.235. and the increased AluY fraction can be greater than about 0. 12. The representation of Alu elements can be determined using RealSeqS, and the reduced AluS fraction can be less than about 0.75, the increased AluJ fraction can be greater than about 0.065, and the increased AluY fraction can be greater than about 0.205. The cancer can include one or more solid tumors. The fluid sample can be a plasma sample.

[0016] In another aspect, this document features methods for assessing a mammal where the methods can include, or consist essentially of, (a) determining a representation of Alu elements in cell-free DNA (cfDNA) in a fluid sample obtained from the mammal; and (b) identifying the mammal as not having cancer if the representation of Alu elements includes an absence of each of (1) a reduced AluS fraction, (2) an increased AluJ fraction, and (3) an increased AluY fraction. The mammal can be a human. The representation of Alu elements can be determined using whole genome sequencing, and the reduced AluS fraction can be less than about 0.66, the increased AluJ fraction can be greater than about 0.235, and the increased AluY fraction can be greater than about 0. 12. The representation of Alu elements can be determined using RealSeqS, and the reduced AluS fraction can be less than about 0.75, the increased AluJ fraction can be greater than about 0.065, and the increased AluY fraction can be greater than about 0.205. The cancer can include one or more solid tumors. The fluid sample can be a plasma sample.

[0017] In another aspect, this document features methods for treating for a mammal having cancer where the methods can include, or consist essentially of, (a) determining that a representation of Alu elements in cell-free DNA (cfDNA) in a fluid sample obtained from the mammal includes the presence of at least one of (1) a reduced AluS fraction, (2) an increased AluJ fraction, and (3) an increased AluY fraction; and (b) administering a cancer treatment to the mammal. The mammal can be a human. The representation of Alu elements can be determined using whole genome sequencing, and the reduced AluS fraction is less than about 0.66, the increased AluJ fraction is greater than about 0.235, and the increased AluY fraction is greater than about 0.12. The representation of Alu elements can be determined using RealSeqS, and the reduced AluS fraction is less than about 0.75, the increased AluJ fraction is greater than about 0.065, and the increased AluY fraction is greater than about 0.205. The cancer can include one or more solid tumors. The cancer treatment can include performing surgery. The cancer treatment can include radiation therapy. The cancer treatment can include administering, to the mammal, an anti-cancer agent such as a chemotherapy or an angiogenesis inhibitor. The fluid sample can be a plasma sample.

[0018] In another aspect, this document features methods for treating cancer where the methods can include, or consist essentially of, administering a cancer treatment to a mammal identified as having a fluid sample having cell-free DNA (cfDNA) including a representation of Alu elements having a presence of at least one of (1) a reduced AluS fraction, (2) an increased AluJ fraction, and (3) an increased AluY fraction. The mammal can be a human. The representation of Alu elements can be determined using whole genome sequencing, and the reduced AluS fraction is less than about 0.66, the increased AluJ fraction is greater than about 0.235, and the increased AluY fraction is greater than about 0.12. The representation of Alu elements can be determined using RealSeqS, and the reduced AluS fraction is less than about 0.75, the increased AluJ fraction is greater than about 0.065, and the increased AluY fraction is greater than about 0.205. The cancer can include one or more solid tumors. The cancer treatment can include performing surgery. The cancer treatment can include radiation therapy. The cancer treatment can include administering, to the mammal, an anti-cancer agent such as a chemotherapy or an angiogenesis inhibitor. The fluid sample can be a plasma sample.

[0019] 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 entirety7. 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.

[0020] 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.

[0021] BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1. Schematic overview of the studies presented in Example.

[0023] Figures 2A-2C. Cohort 2 performance metrics. Figure 2A) Evaluation of the Alu Profile Learning Using Sequencing (A-PLUS), global aneuploidy scores (GAS), and protein score (PROT) individual performances as well as that of the multi-analyte classifier (A-PLUS+GAS+PROT). Sensitivities were calculated at 99% specificity in each case. Error bars represent 95% CI. Figure 2B) Receiver operating characteristic (ROC) curve for the samples in Cohort 2. Figure 2C) Venn Diagrams of the overlap in positive biomarker predictions for Cohort 2 samples. Samples outside all three circles are negative for all three biomarkers.

[0024] Figures 3A-3D. Cohort 3 performance metrics. Figure 3A) Comparison of Cohort 2 and Cohort 3 Performance for A-PLUS predictions. Only cancer types common to both Cohorts are depicted. Figure 3B) Comparison of Cohort 2 and Cohort 3 Performance for the Multi-Analyte Classifier (A-PLUS+GAS+PROT). Only cancer types common to both Cohorts are depicted. Figure 3C) Full Performance Metrics for the individual assays and Multi-Analyte Classifier in Cohort 3. Figures 3D) Venn Diagrams of the overlap in positive biomarker predictions for Cohort 3 samples. Samples outside all three circles are negative for all three biomarkers.

[0025] Figures 4A-4B. Reproducibility. Technical replicates were assessed for concordance of A-PLUS (Figure 4A) and GAS (Figure 4B).

[0026] Figures 5A-5C. Cancer samples exhibit a global reduction in sequencing depth of AluS when compared to the total depth of the Alu elements (AluS fraction). Figure 5A) AluS fraction of RealSeqS data from samples used in the current study (Cohorts 1, 2, 3, and 4). Figure 5B) AluS fraction for whole genome sequencing (WGS) samples Figure 5C) AluS fraction on a publicly available WGS dataset. There are 312,138 AluJ loci, 686, 962 AluS loci, and 143,178 AluY loci distributed throughout the genome. RealSeqS, however, only amplifies a subset of the total Alu loci (AluJ - 125,797; AluS - 485,614; AluY - 86252) hence the substantial difference between the AluS fraction between RealSeqS and WGS samples.

[0027] Figures 6A-6D. Distribution of Non-Cancer Cohort 2 scores and the selected thresholds for positivity for: A-PLUS (Figure 6A); GAS (Figure 6B); PROT (Figure 6C); and A-PLUS+GAS+PROT (Figure 6D).

[0028] Figures 7A-7F. Cancer samples exhibit a global increase in sequencing depth of AluJ and AluY Fraction when compared to the total depth of the Alu elements. AluJ (Figure 7A) and AluY (Figure 7B) Fraction of RealSeqS data from samples used in the current study (Cohorts 1. 2, 3, and 4). AluJ (Figure 7C) and AluY (Figure 7D) Fraction for WGS samples; AluJ (Figure 7E) and AluY (Figure 7F) Fraction on a publicly available WGS dataset.

[0029] DETAILED DESCRIPTION

[0030] This document provides methods and materials for assessing and / or treating mammals (e.g., humans) having cancer. In some cases, this document provides methods and materials for identifying a mammal as having cancer. For example, the representation of particular Alu elements (e.g., a reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction) in cfDNA in a fluid sample obtained from a mammal (e.g.. a mammal suspected of having cancer) can be used to identify the mammal as having cancer. In some cases, this document provides methods and materials for treating a mammal identified as having cancer. For example, a mammal identified as having cancer as described herein (e.g., based, at least in part, on a reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction in cfDNA in a fluid sample obtained from the mammal) can be administered one or more cancer treatments.

[0031] An Alu element can include about 300 base pairs and particular Alu elements can be distinguished from each other by their sequences. For example, a canonical Alu element sequence can be as shown in Example 2, and particular Alu elements can be distinguished based on having one or more deviations (e.g., one or more nucleotide insertions, one or more nucleotide deletions, one or more nucleotide indels, and / or one or more nucleotide substitutions) from the canonical Alu element sequence. Exemplary Alu element sequences can be as described at, for example, the Institute for Systems Biology (ISB) summary of the human ALU subfamily available at repeatmasker.org / AluSubfamilies / humanAluSubfamilies.html.

[0032] The representation of a particular Alu element can be referred to as a fraction (e.g., an AluS fraction and an AluJ fraction). As used herein, an “AluS fraction” refers to the number of AluS elements present within the total population of Alu elements. An AluS fraction can include AluS elements of any AluS sub-family (e.g., AluSz, AluSz6, AluSx, AluSxl, AluSq, AluSq2, AluSq4, AluSp, AluSg, AluSqlO, AluSx3, AluSx4, AluSg4, AluSg7, AluSc, AluSc8, and AluSc5 elements). When an AluS fraction is a reduced AluS fraction, AluS elements of any AluS sub-family can be present in an AluS fraction in any representation, provided that the total AluS fraction is reduced.

[0033] As used herein, an “AluJ fraction” refers to the number of AluJ elements present within the total population of Alu elements. An AluJ fraction can include AluJ elements of any AluJ sub-family (e.g., AluJr, AluJr4, AluJo, and AluJb elements). When an AluJ fraction is an increased AluJ fraction, AluJ elements of any AluJ sub-family can be present in an AluJ fraction in any representation, provided that the total AluJ fraction is increased.

[0034] As used herein, an “AluY fraction” refers to the number of AluY elements present within the total population of Alu elements. An AluY fraction can include AluY elements of any AluY sub-family (e.g., Yb8, Yb9, Yh9, Yc3, Yc3, Yd8, Ye5, Ye6, Yg6, Yk4, Ykl 1, Ykl2. and Ya5 elements). When an AluY fraction is an increased AluY fraction, AluY elements of any AluY sub-family can be present in an AluY fraction in any representation, provided that the total AluY fraction is increased.

[0035] Any appropriate method can be used to determine the representation of a particular Alu element (e.g., an AluS fraction, an AluJ fraction, and / or an AluY fraction). Examples of methods that can be used to determine the representation of a particular Alu element include, without limitation sequencing techniques such as WGS, RealSeqS, hybrid capture panels, single base extension, and polymerase chain reaction (PCR) based assays such as quantitative PCR (qPCR), digital PCR (dPCR), TaqMan® Real-Time PCR. For example, when the representation of a particular Alu element (e.g., an AluS fraction, an AluJ fraction, and / or an AluY fraction) is determined using a sequencing technique, the fraction can be expressed as the number of the particular Alu reads (e.g., AluS reads, AluJ reads, or AluY reads) divided by the number of all Alu reads. In some cases, the representation of a particular Alu element can be measured as described in Example 1. In some cases, the representation of a particular Alu element can be measured as described elsewhere (see, e.g., Douville et al., Proc. Natl. Acad. Set. 117:4858-4863 (2020); and International Patent Application Publication No. WO 2020 / 236625).

[0036] In cases where an AluS fraction is determined using WGS, a reduced AluS fraction can be less than about 0.66. For example, a reduced AluS fraction (e.g., as determined using WGS), can be from about 0.63 to about 0.66.

[0037] In cases where an AluS fraction is determined using RealSeqS, a reduced AluS fraction can be less than about 0.75. For example, a reduced AluS fraction (e.g., as determined using RealSeqS), can be from about 0.72 to about 0.75.

[0038] In some cases, a reduced AluS fraction can be determined by comparing the AluS fraction of a fluid sample obtained from a mammal (e.g., a mammal suspected of having cancer) to a reference AluS fraction and a reduced AluS fraction can refer to any AluS fraction that is lower than the reference AluS fraction. In some cases, a reduced AluS fraction can be an AluS fraction that is at least 1 (e g., at least 1.5, at least 2, at least 2.5, at least 3, or at least 3.5) fold lower than a reference fraction AluS fraction. In some cases, a reduced AluS fraction can be an AluS fraction that is at least 5% (e.g., at least 8%, at least 10%, at least 12%, at least 15%. or at least 20%) lower than a reference AluS fraction. In cases where an AluJ fraction is determined using WGS, an increased AluJ fraction can be greater than about 0.235. For example, an increased AluJ fraction (e.g., as determined using WGS), can be from about 0.235 to about 0.245.

[0039] In cases where an AluJ fraction is determined using RealSeqS, an increased AluJ fraction can be greater than about 0.065. For example, an increased AluJ fraction (e.g.. as determined using RealSeqS), can be from about 0.065 to about 0.08.

[0040] In cases where an AluY fraction is determined using WGS, an increased AluY fraction can be greater than about 0. 12. For example, an increased AluY fraction (e.g., as determined using WGS). can be from about 0. 12 to about 0.14.

[0041] In cases where an AluY fraction is determined using RealSeqS, an increased AluY fraction can be greater than about 0.205. For example, an increased AluY fraction (e.g., as determined using RealSeqS), can be from about 0.205 to about 0.22.

[0042] In some cases, an increased Alu fraction (e.g., an increased AluJ fraction and / or an increased AluY fraction) can be determined by comparing the particular Alu fraction (e.g., an AluJ fraction or an AluY fraction) of a fluid sample obtained from a mammal (e g., a mammal suspected of having cancer) to a reference fraction of that particular Alu element and an increased Alu fraction can refer to any Alu fraction that is greater than the reference Alu fraction. In some cases, an increased Alu fraction can be an Alu fraction that is at least 1 (e.g., at least 1.5, at least 2, at least 2.5, at least 3, or at least 3.5) fold higher than a reference fraction Alu fraction. In some cases, an increased Alu fraction can be an Alu fraction that is at least 5% (e.g., at least 8%, at least 10%, at least 12%, at least 15%, or at least 20%) higher than a reference Alu fraction].

[0043] The term reference as used herein with respect to any particular Alu fraction refers to the fraction of that Alu element 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 cases where an AluS fraction is determined using WGS, a reference AluS fraction can be about 0.66 or more. In cases where an AluS fraction is determined using RealSeqS, a reference AluS fraction can be about 0.75 or more. In cases where an AluJ fraction is determined using WGS, a reference AluJ fraction can be about 0.235 or less. In cases where an AluJ fraction is determined using RealSeqS, a reference AluJ fraction can be about 0.065 or less. In cases where an AluY fraction is determined using WGS, a reference AluY fraction can be about 0. 12 or less. In cases where an AluY fraction is determined using RealSeqS, a reference AluY fraction can be about 0.205 or less. It will be appreciated that Alu fractions from comparable samples are typically used when determining whether or not a particular Alu fraction is a reduced Alu fraction or an increased Alu fraction. For example, when assessing a human for the fraction of a particular Alu element in cfDNA of a fluid sample, the reference Alu fraction is a reference Alu fraction of human control samples.

[0044] Any appropriate mammal can be assessed and / or treated as described herein. Examples of mammals that can be assessed and / or treated as described herein include, without limitation, humans, non-human primates (e.g., monkeys), dogs, cats, horses, cows, pigs, sheep, mice, and rats. In some cases, a human (e.g., a human suspected of having cancer) can be assessed and / or treated as described herein.

[0045] When assessing a mammal (e.g., a human such as 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. In some cases, 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. A cancer assessed and / or treated as described herein can be any stage of cancer (e.g., Stage I, Stage, II, Stage III, or Stage IV). Examples of cancers that can be assessed and / or treated as described herein include, without limitation, breast cancers, colorectal cancers, esophageal cancers, lung cancers, liver cancers, pancreatic cancers, ovarian cancers, stomach cancers, and melanomas.

[0046] Any appropriate fluid sample from a mammal (e.g., a human such as a human suspected of having cancer) can be assessed as described herein (e.g., for a reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction in cfDNA in 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). Examples of fluid samples that can be assessed as described herein include, without limitation, urine samples, blood samples (e.g., whole blood samples plasma samples, and serum samples), saliva samples, stool samples, and cerebrospinal fluid (CSF) 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 can be isolated from a fluid sample and can be assessed as described herein.

[0047] Any appropriate method can be used to obtain a fluid sample from a mammal (e.g., a human such as a human suspected of having cancer). When a fluid sample is a blood sample, the blood sample can be obtained using phlebotomy techniques (e.g., can be a liquid biopsy).

[0048] In some cases, identifying a mammal (e.g., a human such as a human suspected of having cancer) also can include assessing aneuploidy of cfDNA in a fluid sample obtained from the mammal. Any appropriate method can be used to assess aneuploidy. In some cases, a global aneuploidy score (GAS) can be used to assess aneuploidy. In some cases, aneuploidy (e.g., a GAS) can be determined as described in Example 1. In some cases, aneuploidy (e.g., a GAS) can be determined as described elsewhere (see, e.g., Douville et al., Proc. Natl. Acad. Sci. 117:4858-4863 (2020)). For example, when cfDNA in a fluid sample obtained from a mammal (e.g.. a human such as a human suspected of having cancer) is determined to have a GAS of less than about 0.64, the mammal can be identified as having cancer. For example, when cfDNA in a fluid sample obtained from a mammal (e.g., a human such as a human suspected of having cancer) is determined to have a GAS of greater than 0.64, the mammal can be identified as not having cancer.

[0049] In some cases, identifying a mammal (e.g., a human such as a human suspected of having cancer) also can include assessing circulating polypeptides in a fluid sample obtained from the mammal. Any appropriate method can be used to assess circulating polypeptides in a fluid sample. In some cases, a protein (PROT) score can be used to assess circulating polypeptides in a fluid sample. A PROT score can be based on the presence or absence of any appropriate polypeptides. Examples of polypeptides that can be used to determine a PROT score include, without limitation, OPN polypeptides, HGF polypeptides. AFP polypeptides, CA125 polypeptides, CA15-3 polypeptides, CEA polypeptides. CAI 9-9 polypeptides, TIMP-1 polypeptides, and PSA polypeptides. In some cases, circulating polypeptides can be assessed (e.g., a PROT score can be determined) as described in Example 1. For example, when a fluid sample obtained from a mammal (e.g., a human such as a human suspected of having cancer) is determined to have a PROT of less than about 0.73, the mammal can be identified as having cancer. For example, when a fluid sample obtained from a mammal (e.g., a human such as a human suspected of having cancer) is determined to have a PROT of greater than 0.73, the mammal can be identified as not having cancer.

[0050] As described herein, the representation of a particular Alu element (e.g., a reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction) in cfDNA of a fluid sample obtained from a mammal (e.g., a human) can be used to identify that mammal as having cancer. In some cases, a reduced AluS fraction in cfDNA of a fluid sample obtained from a mammal (e.g., as compared to a reference sample) can be used to identify that mammal as having cancer. For example, when a fluid sample obtained from a mammal (e.g., a human suspected of having cancer) is determined to have a reduced AluS fraction in cfDNA in the fluid sample, the mammal can be identified as having cancer. For example, when a fluid sample obtained from a mammal (e.g., a human suspected of having cancer) is determined to have an AluS fraction of less than about 0.66 as determined using WGS or an AluS fraction of less than about 0.75 as determined using RealSeqS, the mammal can be identified as having cancer. In some cases, the absence of a reduced AluS fraction in cfDNA of a fluid sample 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 obtained from a mammal (e.g., a human suspected of having cancer) is determined to lack a reduced AluS fraction in cfDNA in the fluid sample, the mammal can be identified as not having cancer. For example, when a fluid sample obtained from a mammal (e.g., a human suspected of having cancer) is determined to have an AluS fraction of about 0.66 (or more) as determined using WGS or an AluS fraction of about 0.75 (or more) as determined using RealSeqS. the mammal can be identified as having cancer.

[0051] In some cases, an increased AluJ fraction in cfDNA of a fluid sample obtained from a mammal (e.g., as compared to a reference sample) can be used to identify7that mammal as having cancer. For example, when a fluid sample obtained from a mammal (e.g., a human suspected of having cancer) is determined to have an increased AluJ fraction in cfDNA in the fluid sample, the mammal can be identified as having cancer. For example, when a fluid sample obtained from a mammal (e.g., a human suspected of having cancer) is determined to have an AluJ fraction of greater than about 0.235 as determined using WGS or an AluJ fraction of greater than about 0.065 as determined using RealSeqS, the mammal can be identified as having cancer. In some cases, the absence of an increased AluJ fraction in cfDNA of a fluid sample 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 obtained from a mammal (e.g., a human suspected of having cancer) is determined to lack an increased AluJ fraction in cfDNA in the fluid sample, the mammal can be identified as not having cancer. For example, when a fluid sample obtained from a mammal (e g., a human suspected of having cancer) is determined to have an AluJ fraction of about 0.235 (or less) as determined using WGS or an AluJ fraction of about 0.065 (or less) as determined using RealSeqS, the mammal can be identified as having cancer.

[0052] In some cases, the representation of a particular Alu element (e.g., a reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction) in cfDNA of a fluid sample obtained from a mammal (e.g., a human) can be used to select one or more actions for that mammal. For example, a mammal (e.g., a human) identified as having cancer as described herein (e.g., based, at least in part, on a reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction in cfDNA of a fluid sample obtained from the mammal) can be selected for one or more cancer treatments.

[0053] In some cases, a mammal (e.g., a human) identified as having cancer based, at least in part, on having an AluS fraction of less than about 0.66 as determined using WGS can be selected for one or more cancer treatments. In some cases, a mammal (e.g., a human) identified as having cancer based, at least in part, on having an AluS fraction of less than about 0.75 as determined using RealSeqS, can be selected for one or more cancer treatments. For example, a mammal (e.g., a human) identified as not having cancer as described herein (e.g.. based, at least in part, on the absence of a reduced AluS fraction in cfDNA of a fluid sample obtained from the mammal) can be selected to receive more frequent diagnostic testing. In some cases, a mammal (e.g., a human) identified as not having cancer based, at least in part, on having an AluS fraction of about 0.66 (or more) as determined using WGS can be selected to receive more frequent diagnostic testing. In some cases, a mammal (e g., a human) identified as not having cancer based, at least in part, on having an AluS fraction of about 0.75 (or more) as determined using RealSeqS, can be selected to receive more frequent diagnostic testing.

[0054] In some cases, a mammal (e.g., a human) identified as having cancer based, at least in part, on having an AluJ fraction of greater than about 0.235 as determined using WGS can be selected for one or more cancer treatments. In some cases, a mammal (e.g.. a human) identified as having cancer based, at least in part, on having an AluJ fraction of greater than about 0.065 as determined using RealSeqS, can be selected for one or more cancer treatments. For example, a mammal (e.g., a human) identified as not having cancer as described herein (e.g.. based, at least in part, on the absence of an increased AluJ fraction in cfDNA of a fluid sample obtained from the mammal) can be selected to receive more frequent diagnostic testing. In some cases, a mammal (e.g., a human) identified as not having cancer based, at least in part, on having an AluJ fraction of about 0.235 (or less) as determined using WGS can be selected to receive more frequent diagnostic testing. In some cases, a mammal (e.g., a human) identified as not having cancer based, at least in part, on having an AluJ fraction of about 0.065 (or less) as determined using RealSeqS, can be selected to receive more frequent diagnostic testing.

[0055] 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 reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction in cfDNA of a fluid sample obtained from the mammal) can be administered to instructed to self-administer one or more cancer treatments. For example, a mammal (e.g., a human) identified as having cancer based, at least in part, on having an AluS fraction of less than about 0.66 as determined using WGS can be administered to instructed to self-administer one or more cancer treatments. For example, a mammal (e.g., a human) identified as having cancer based, at least in part, on having an AluS fraction of less than about 0.75 as determined using RealSeqS, can be administered to instructed to self-administer one or more cancer treatments. For example, a mammal (e.g., a human) identified as having cancer based, at least in part, on having an AluJ fraction of greater than about 0.235 as determined using WGS can be administered to instructed to self-administer one or more cancer treatments. For example, a mammal (e.g., a human) identified as having cancer based, at least in part, on having an AluJ fraction of greater than about 0.065 as determined using RealSeqS, can be administered to instructed to self-administer one or more cancer treatments.

[0056] When a mammal (e.g., a human) identified as having cancer as described herein (e.g., based, at least in part, on a reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction in cfDNA of a fluid sample 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 anti-cancer 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 reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction in cfDNA of a fluid sample 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 reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction in cfDNA of a fluid sample obtained from the mammal) include, without limitation, radiation therapies, and / or surgeries (e.g., to remove at least a portion of one or more solid tumors).

[0057] The invention will be further described in the following examples, which do not limit the scope of the invention described in the claims.

[0058] EXAMPLES

[0059] Example 1: Short interspersed nuclear elements (SINEs) in cancer

[0060] As noted above, there is much precedent for Alu sequence elements being especially prone to epigenetic changes in various cancers.

[0061] This Examples describes the discovery' that representation of specific Alu elements in the cell-free DNA (cfDNA) of cancer patients differs from representation of those specific Alu elements in cfDNA of normal controls. Because there are so many Alu elements in the genome, a machine-learning based approach, referred to herein as Alu Profile Learning Using Sequencing (A-PLUS), was used to distinguish individuals with cancer from those without cancer on the basis of the representation of particular Alu elements in their cfDNA.

[0062] Results

[0063] Rationale and background of the assay

[0064] During the development and implementation of a sensitive PCR-based assay called Repetitive Element AneupLoidy Sequencing System (RealSeqS), substantial differences in read depth at specific loci were observed (see, e.g.. Douville et al.. Proc. Natl. Acad. Sci. 117:4858-4863 (2020)). These loci did not appear to correlate with cancer-specific copy number variations in the non-cancer controls. Thus, an unbiased supervised machine-learning method was utilized to select cancer-specific Alu element representations from RealSeqS data as a metric for cancer patient classification.

[0065] The detection of cancer in asymptomatic patients, which is the primary goal of multi -cancer earlier detection tests, requires very high specificity. Designing a highly specific machine-learning algorithm to predict cancer status from the~350,000 features assessed in RealSeqS sequencing data thereby posed technical challenges. To address such challenges, several principles were incorporated into the development of A-PLUS. First, confounding loci associated with technical noise, ethnicity, sex, and batch differences were identified and eliminated. Second, the number of features was reduced from 350,000 using Principal Component Analysis (PCA). Third, an order of magnitude more samples (thousands rather than hundreds) were used than are typically used in initial studies on new tests of cfDNA performance. Fourth, samples were divided into four prespecified and non-overlapping cohorts to minimize over-fitting: Cohort 1 was used to choose features and train the ML model; Cohort 2 was used to establish thresholds for scoring samples as positive or negative; Cohort 3 was used to independently test (validate) the ML model based on Cohort 1 and the thresholds based on Cohort 2; and Cohort 4 was used to evaluate reproducibility of the scoring system (Figure 1).

[0066] Technical nuances underlying the development of A-PLUS are detailed in the Materials and Methods section and the code is available at doi.org / 10.5281 / zenodo.7853623. Cohort 1: A-PLUS Feature selection and model training

[0067] Cohort 1 consisted of 566 individuals (715 total samples) from 354 individuals without cancer (400 total samples) and 202 (315 total samples) from solid cancer patients. Individuals with multiple samples were always from the same time point and considered technical replicates. Cohort 1 has 459 samples previously analyzed for aneuploidy using RealSeqS (Douville et al., Proc. Natl. Acad. Sci. 117:4858-4863 (2020)) and an additional 250 samples from controls of Vietnamese, Han Chinese, South Asian, or Native American / Inuit ethnicities. These additional samples were included because inherited polymorphisms within Alu elements could alter alignments and the subsequent read depth representation of Alu loci. Of the total samples from patients included those with cancers of the breast, colorectal, esophagus, lung, liver, pancreas, ovary, and stomach. Slightly fewer cancer samples than control samples were used because specificity was valued over sensitivity; cancer samples erroneously classified as controls were deemed less harmful to performance of the final classifier than the reverse.

[0068] Important elements of the training included normalization of read depths and the removal of amplicons with insufficient coverage, removal of amplicons that were unstable based on T-tests (Methods). After these steps, there were 121,197 loci of the original 350,000 that remained. PCA was then used to reduce dimensionality. Finally, a Support Vector Machine (SVM) was used to identify the 60 first PCA components. This feature number (60) was -10% of the total number of unique patients in the training set which was considered a reasonable compromise to cope with the p»n conundrum (Verleysen in Computational Intelligence and Bioinspired Systems. Lecture Notes in Computer Science. Cabestany et al.. Eds. (Springer. Berlin. Heidelberg, 2005). pp. 758- 770). Performance was not assessed in Cohort 1. The non-cancer samples were used to generate a euploid reference panel for aneuploidy calls. Both the cancer samples and noncancer samples were used to generate and optimize model building.

[0069] To reduce the chance of over-fitting and clearly establish quality metrics prior to evaluation of samples, previously published metrics and thresholds for inclusion of patients were used rather than excluding any participants because of any metric related to A-PLUS performance. Quality Control metrics were designed to identify technical outliers (Douville et al., Proc. Natl. Acad. Sci. 117:4858-4863 (2020)). The first metric identified samples likely impacted from unusual GC bias. This metric assessed whether the proportional representation at various low GC chromosomes fell outside our pre- established range. Next, two different size metrics were used to identify the presence of large molecular weight DNA. Large molecular weight DNA is frequently attributed to cellular genomic DNA, not cfDNA. Contamination from genomic DNA can limit the sensitivity of the assay. These previously defined criteria excluded 3.1% of Cohort 1 samples. A small fraction of patients were identically excluded in the other cohorts.

[0070] Cohort 2: Analyte integration and determination of thresholds

[0071] Cohort 2 included samples from 704 solid cancer patients (total of 852 samples) and 958 controls without cancer (total of 1402 samples). Cancers included those from colon, esophagus, stomach, breast, colorectum, lung, ovary, and pancreas and none had metastases at the time of the blood draw. The A-PLUS score corresponding to 99% specificity among the control samples was 0.28 (Figure 6A). At this threshold, samples from patients with cancers of the esophagus and stomach had the highest sensitivities (86 - CI: 68% to 96% and 87% - CI: 73% to 94%, respectively) and samples from breast cancer patients had the lowest sensitivities (34%; Figure 2A).

[0072] It was also investigated whether the number of principal components as well as machine-learning algorithm would impact performance. An A-PLUS model using the first 60 principal components outperformed models with 15, 30, 90, and 240 PCs. Across Cohort 2, 51.5% of cancers were detected using a model built with 60 PCs, while 36.5% with 15, 46.0% with 30, 49.9% with 90, and 49.2% with 240 based on a threshold that produces > 99% specificity. The SVM model outperformed the Random Forrest model (51.5% vs 39.8%).

[0073] Global aneuploidy scores (GAS) were then generated for Cohort 2 samples with the same RealSeqS data used to generate A-PLUS scores. The GAS uses a different machine learning technique to generate a single score that reflects gains or losses of 39 chromosome arms, focusing on those arms that are typically altered in cancers (Douville et al., Proc. Natl. Acad. Sci. 115:1871-1876 (2018)). A GAS threshold of >0.64 yielded a 99% specificity in the Cohort 2 control samples (Figure 6B). The highest sensitivities at this 99% specificity were achieved for cancers of the esophagus and liver (43% - CL 26% to 62% and 37% - CI: 17% to 61%, respectively) and the lowest sensitivities in breast (6% - CI 4% to 10%; Figure 2A). In the 687 cancers scoring negatively (i.e., below the 99%-specificity threshold) in the GAS assay, 318 (46%) scored positively (i.e., above the 99% specificity threshold) in the A-PLUS assay. (Figure 2C). Conversely, 81% of the cancer samples that scored positively in GAS also scored positively in A-PLUS (Figure 2C).

[0074] Next, a panel of eight protein markers previously shown to be useful for cancer detection when employed at high thresholds (e.g.. OPN. HGF, AFP, CA125, CA15-3, CEA, CAI 9-9, and TIMP-1) was evaluated. To integrate these eight protein values into a single score, Logistic Regression with binary classification was used to generate a protein score. Performance was assessed using 10-fold cross validation and a protein score ("PROT") of >0.73 generated 99% specificity in the samples from patients without cancer (Figure 6C). The highest sensitivities at this threshold were achieved for cancers of the liver and stomach (84% - CI: 60% to 96% and 69% - CI: 54% to 81%, respectively) and the lowest sensitivities in breast (18% - CI: 14% to 23%; Figure 2A). In the 535 cancers scoring negatively with the protein assay, 44% scored positively with A-PLUS (Figure 2C)

[0075] Logistic Regression was then used to integrate A-PLUS and GAS with the proteins into a multi-analyte classifier. Performance was assessed using 10-fold cross validation and a threshold of >0.87 generated 99% specificity (Figure 6D). The highest sensitivities were achieved for patients with cancers of the esophagus and liver (90% - CI: 72% to 97% and 90% - CI: 66% to 98%, respectively), and the lowest sensitivities in patients with breast cancers (32%; Figure 2A). The sensitivity of the multi-analyte classifier was equivalent or higher than any individual analyte in every cancer type, and high levels of specificity were maintained.

[0076] Cohort 3: Independent validation

[0077] Cohort 3 samples were from 2960 individuals, including 1167 patients with solid cancers of eleven types: breast, colorectum, esophagus, head and neck, kidney, lung, ovary, pancreas, prostate, stomach, and uterus (Table 1).

[0078] Attorney Docket No. 44807-0467WO1 / P17983-01

[0079] Table 1. Description of samples evaluated.

[0080] The 99% thresholds defined by Cohort 2 controls were used to assess performance in Cohort 3 for each of the assays described above. For A-PLUS, the specificity observed in Cohort 3 (98.5%) slightly decreased different from the 99% expected from Cohort 2 (p=0.03, one sided Two sample Z-test for equality of proportions). In the seven cancer types that were evaluated in both Cohorts 2 and 3, the cancer-type sensitivities were similar with the exception of lung, where the sensitivities were 54% and 27%, respectively (Figure 3 A). This decrease in Cohort 3 sensitivity for lung cancers could not be attributed to differences in cancer subtype between the two cohorts. The sensitivities and specificities of aneuploidy alone as well as proteins alone were similar in Cohorts 2 and 3. The multi-analyte test incorporating A-PLUS, aneuploidy, and proteins was able to detect more than 2 / 3 of cancers of each of the following organs: esophagus, pancreas, ovary, stomach, colorectum at an observed specificity of 98.9% (Figure 3A). The sensitivities of the two other cancer types evaluated in both Cohorts 2 and 3 were lower (37% - CI: 29% to 46% and 26% - CI: 16% to 38% in lung and breast, respectively). Four cancer types (head and neck, kidney, prostate, and uterus) were represented in Cohort 3 but not Cohorts 1 or 2. Even though no algorithmic training was performed on these cancer types, the sensitivity for their detection ranged from 17% to 36%. Cancers of the liver were not available in Cohort 3, though the sensitivity for their detection in Cohort 2 was high (90%).

[0081] The overlap in analyte detections (A-PLUS, GAS, and PROT) at the pre-defined thresholds is depicted as a Venn diagram in Figure 3D. A-PLUS made a greater contribution to positive calls than aneuploidy or proteins, and A-PLUS detected 41% of the samples that were not detected by either aneuploidy or proteins.

[0082] The vast majority of cancers represented in Cohort 3, and all the samples in Cohorts 1 and 2, were relatively early in the sense that few had any distant metastatic lesions evident at diagnosis. However, there were 15 cases (0.6% of 2349), all ovarian and all in Cohort 3, that were later re-staged to IV. When these were removed, the sensitivity for ovarian cancer detection decreased from 75% to 73%. The sensitivity was 72% for ovarian in Cohort 2, which did not contain any Stage IV cancers. When categorized by stage I, II, or III, the average sensitivities for all cancers using each individual analyte or the multi-analyte classifier increased with stage in both Cohorts 2 and 3. With the multi-analyte classifier, for example, sensitivity averaged 57% in Stage I, 60% in Stage II, and 65% in Stage III.

[0083] Cohort 4: Reproducibility

[0084] The technical reproducibility of the A-PLUS and GAS assays (both based on RealSeqS sequencing data) was evaluated in 1686 individuals (1539 without cancer and 147 with cancer) from Cohort 2 or Cohort 3. Repeated samples are all from the same timepoint. When an individual had three or more technical replicates available, only the first two replicates were considered. Separate aliquots of purified DNA (i.e., separate template molecules) from the same plasma sample were independently amplified using the single RealSeqS primer pair and the PCR products sequenced. The amplification and sequencing of each DNA sample from the same patient were performed on different days. Of the 1686 pairs, 54 pairs had at least one of samples with a QC fail.

[0085] Using the thresholds defined by Cohort 2, 95.8% of the 1632 pairs scored concordantly (either positively or negatively) with a Cohen’s Kappa of 0.56 (95% CI 0.47- 0.65) for A-PLUS (Figure 4A). Of the 70 discordant samples, 23 scored just below the threshold in one of the two DNA aliquots (i.e., between the scores required for 98% and 99% specificity) and above the score required for 99% specificity in the other aliquot. With GAS, 99.3% pairs were concordant (Figure 4B) with a Cohen’s Kappa of 0.66 (CI 0.48-0.85).

[0086] Alu subsets

[0087] It was next determined whether there was a subset of the Alu loci included in the A- PLUS heuristic that were particularly important for its success in distinguishing samples from individuals with or without cancer. The most striking observation was a global reduction in read depth across all AluS loci. The AluS sub-family is the largest sub-family and older than AluJ but younger than the AluY (Jurka et al., Proc. Natl. Acad. Sci. U. S. A., 85:4775-4778 (1988); and Willard et al., J. Mol. Evol. 26: 180-186 (1987)). Using only a single feature - the average normalized read depth of AluS elements (herein dubbed AluS Fraction) from RealSeqS data - without any machine learning algorithms at all - samples from cancer patients could be distinguished from those of controls (p<2e-176 one sided t- test; AUC = 0.70; Figure 5A). A-PLUS scores and the proportional representation of AluS elements were inversely correlated (-0. 19; p<2.2e-16 via Pearson’s). It was noted that PC3 from the A-PLUS model was associated with the global change in read depth across AluS with a Pearson correlation coefficient of 0.72. Given that AluJ fraction, AluS fraction, and AluY fraction add to 1, a change in AluS naturally produces a change in the other fractions. The decrease in AluS fraction for cancer samples was associated with an increase in AluJ fraction (p<4e-38 one sided t-test; AUC=0.46; Figure 7A) and AluY fraction (p<5e-100 one sided t-test; AUC=0.66; Figure 7B) fraction albeit less statistically significant.

[0088] It was evaluated whether the reduced representation of AluS elements in the cfDNA from cancer patients was the result of some unknown bias in amplification efficiency or sequencing generated from the RealSeqS approach employing a single primer pair. To address this question, WGS data from a recently published study that included 65 cfDNA samples from individuals without cancer and 58 samples from cancer patients (55 colorectum; 2 Lung; and 1 pancreas) was evaluated (Curtis et al., medRxiv, DOI: 10.1101 / 2022.08.02.22278319 (2022)). Samples from cancer patients indeed had a global reduction of AluS representation compared to controls (p<2.6e-5, one sided t-test, Figure 5B). Moreover, this single feature (AluS Fraction, defined as # of AluS reads divided by # of all Alu reads) could distinguish samples from cancer patients and controls with an AUC of 0.73 in ROC analysis. The AluS Fraction value corresponds to >99% specificity among the control samples was 0.6546. At this threshold, 36% samples from patients with cancers were positive. The increased representation of AluJ (p>0.05 one sided t-test) and AluY (p<le-5) were also observed in cancer patients in WGS (Figures 7C Fig 7D). Even though AluJ fraction was not statistically significant for this cohort, at high specificities (>99%) 16% of cancers could still be classified with this metric. AluY fraction while statically very significant could only identify 10% of cancers at high specificities (>99%).

[0089] Next, a publicly available dataset that included 266 cfDNA samples from cancer patients (25 bile duct; 54 breast; 22 colorectum; 27 stomach; 76 lung; 26 ovary; 35 pancreatic) and 260 samples from individuals without cancer was evaluated (Zheng et al, Bioinformatics, 37:2502-2503 (2021)). Again, samples from cancer patients had a global reduction of AluS (p<4e-43, one sided t-test; Figure 5C). Using the pre-selected threshold for AluS Fraction, 21% of cancers were detected at a specificity of 98.8% (Table 2). The increased representation in AluJ (p<3e-16 one sided t-test; Figure 7E) and AluY (p<0.02 one sided t-test; Figure 7F) was observed in the publicly available WGS dataset.

[0090] Table 2. Evaluation of AluS Fraction and Aneuploidy in WGS data.

[0091] Finally, it was evaluated whether AluS Fraction in WGS data could be combined with the evaluation of aneuploidy in the same dataset. WiseCondorX were used to count the number of aberrant chromosomal regions throughout the genome in the FinaleDB dataset. Aneuploidy was defined by the WiseCondorX default settings. The scores for aneuploidy and AluS Fraction were combined with a Boolean OR to avoid the possible pitfalls of machine learning algorithms mentioned above. The addition of AluS analysis enhanced the sensitivity of detection of cancers from 42% with aneuploidy alone and 21% with AluS alone to 52% in combination, with an aggregate specificity of 98.5% (Table 2). Similar to the A-PLUS results, patients with cancers of the stomach (67%) and colorectum (65%) had the highest sensitivities.

[0092] Taken together, these results demonstrate that the presence of a reduced AluS fraction, an increased AluJ fraction, and / or an increased AluY fraction in cfDNA of a fluid sample from a mammal (e.g., a human) can be used to identify that mammal as having cancer. Materials and Methods

[0093] Patient Samples

[0094] Plasma was purified from 5108 healthy individuals and 2037 patients with solid cancer using a BioChain Cell-free DNA Extraction Kit (Cat X K5011625). All patients were de-identified and patients are not known to anyone outside the research group.

[0095] RealSeqS Experimental Protocol

[0096] RealSeqS was performed as described elsewhere (Douville et al., Proc. Natl. Acad. Sci. 115: 1871-1876 (2018)). Briefly, PCR was performed in 25 pL reactions containing 7.25 pL of water, 0.125 pL of each primer, 12.5 pL of NEBNext Ultra II Q5 Master Mix (New England Biolabs cat # M0544S), and 5 pL of DNA. Eight independent reactions were performed in ~0.1 ng to 0.25 ng of DNA. A second round of PCR was then performed to add dual indexes to each PCR product prior to sequencing. The second round of PCR was performed in 25 pL reactions containing 7.25 pL of water, 0.125 pL of each primer, 12.5 pL of NEBNext Ultra II Q5 Master Mix (New England Biolabs cat # M0544S), and 5 pL of DNA containing 5% of the PCR product from the first round. Amplification products from the second round were purified with AMPure XP beads (Beckman cat # a63880), as per the manufacturer's instructions, prior to sequencing. As noted above, each sample was amplified in eight independent PCRs in the first round. Each of the eight independent PCRs was then re-amplified using index primers in the second PCR round. The sequencing reads from the 8 replicates were summed for the bioinformatic analysis but could also be assessed individually for quality control purposes.

[0097] Sequence Analysis

[0098] Massively parallel sequencing was performed using a Hiseq4000. During the first round of PCR, degenerate bases at the 5’ end of one of the primers were used as molecular barcodes (unique identifiers, UIDs) to uniquely label each DNA template molecule. This ensured that each DNA template molecule was counted only once, as described elsewhere (Kinde et al., Proc. Natl. Acad. Sci., 108:9530-9535 (2011)). The term “reads” refers to uniquely identified reads (UIDs). Depending on the experiment, each read was sequenced on average 1.1 times. An average of 10.4 million reads per sample (IQR 7.9 M to 12.7 M) was assessed. If multiple reads had the same UID, at least 50% of the reads were required to map to the same genomic location. Reads with the same UID, but with discordant genomic locations were discarded from analysis. Only samples with sufficient read depth (>2.5M) were included.

[0099] Alu Profile Learning Using Sequencing Model Building

[0100] Alu Profile Learning Using Sequencing (A-PLUS) is a supervised machine learning approach to identify differences in normalized read depth for RealSeqS loci between noncancer and cancer cell samples. The following steps were employed to build A-PLUS:

[0101] 1. Assemble a diverse and balanced training set of non-cancer and non-metastatic cancer samples. Alu SINEs are known to have ethnic specific single nucleotide polymorphisms that could impact alignment and potentially alter the normalized read depth representation of various RealSeqS loci. The potential for possible confounders to impact predictions was limited. The training set consisted of 134 previously published non-cancer and 198 solid cancer samples (breast, colorectum, esophagus, lung, liver, pancreas, ovary, and stomach) from Stages I, II, and III. For the noncancers with self-reported ancestry information, the cohort consisted of 41 African / Black, 24 Hispanic, 33 White / Caucasian, and 27 multiple ancestries. For the cancers, the self-reported ancestry consisted of 65 Asian, 3 African / black, 120 Caucasian / White. 264 non-cancer samples and 13 cancer samples were added to diversify the ancestry representation of samples. The additional samples with ancestry information included 20 Vietnamese, 2 Pacific Islander, 59 Black / African, 335 Asian, 3 Native American / Inuit samples, and 136 Caucasian / White. Only samples with sufficient read depths were considered.

[0102] 2. The 99 patients with cancer were selected for this training set because each had a second technical replicate.

[0103] 3. All samples were normalized for total read depth. To do this, a sample’s autosomal loci was divided by its total autosomal coverage and multiplied by 10,000,000. This normalizing step allows all samples to be compared with each other regardless of differences in total coverage. Performed Amplicon Feature Selection. a. Loci with insufficient coverage were removed. Only loci with an average of 1 normalized read in our training set were considered and 445,711 loci (AluJ loci=93,030; AluS loci=242,659; AluY loci=33,803; other loci that are not Alu SINEs=76,219) were removed. After applying this filter, only 299,473 loci remained. Note, RealSeqS is theoretically able to amplify up to 745,184 repetitive elements but typically only 350,000 Alu SINEs were observed. b. The paired T-test was performed for 99 cancer patients with a technical replicate. Loci (AluJ loci=16,377; AluS loci=109,728; AluY loci=27,389; other loci that are not Alu SINEs=5) that were statistically significant (p<0.01) were discarded. After applying this filter, only 145,974 loci of the original 745,184 loci remained. We acknowledge that a statistically significant difference in paired samples could still be informative if the effect size is greater than the technical variability. This filtering step is a conservative choice and likely removed informative loci. Given the large number of loci, we prioritized stability over power and decided to remove loci. c. The Kolmogorov-Smirnov test was performed on samples from different ancestries in the samples from individuals without cancer. Loci that were statistically significant (p<0.01) were discarded (AluJ loci=l,559; AluS loci=l 8,286; AluY loci=4,932; other loci that are not Alu SINEs=0). After applying this filter, 121,197 loci remained. The remaining loci consisted of 9,373 AluJ loci; 95,116 AluS loci; 16,702 AluY loci; and 6 loci that were not Alu SINEs. PCA was perform on the 121,197 loci. PC A was performed using the prcomp function in R version 3.4. A SVM was generated using the first 60 principal components from Step 5 as predictive features. The number of components (n=60) was based on -10% of the total number of samples in the Training set (-600). We used the el071 (vl .7-9) library with a radial basis kernel, cost =1, and nu=0.5.

[0104] Detection of Aneuploidy

[0105] An algorithm to detect the presence of aneuploidy from amplicon sequencing data is described elsewhere (see, e.g., Douville et al., Proc. Natl. Acad. Sci. 117:4858-4863 (2020); and Douville et al., Proc. Natl. Acad. Sci. 115:1871-1876 (2018)), and uses the normalized read counts of 500-kb intervals across the genome and performs a “within- sample” comparison. Outlier intervals were bioinformatically identified and filtered. The remaining intervals were aggregated across the chromosome arm and a circular binary algorithm (CBS) was performed. Any aberrant segment <5 MB was assumed to be a germline copy number variation and bioinformatically removed. The statistical significance was then calculated for the chromosome arm without the aberrant <5 MB segments. The 39 non-acrocentric chromosome arm statistical significances were then used as predictive features in a supervised machine learning model. A SVM was trained on 363 presumably euploid samples from individuals without cancer, from 128 samples of cancer patients, and from 648 in silico- generated aneuploid samples derived from the samples of individuals without cancer. The model was built using R v3.4 with the el 071 library and generates a Global Aneuploidy Score (GAS) which ranges from 0 to 1. Several studies have evaluated copy number alterations in clonal hematopoiesis of indeterminate potential (CHIP) in large population databases (Terao et al., Nature, 584: 130-135 (2020); Saiki et al., Nat. Med., 27: 1239-1249 (2021); and Loh et al., Nature, 559;350-355 (2018)). the possibility of CHIP was not explicitly accounted for, but many of the candidate regions were <5MB and would already be excluded using the previously described bioinformatic filters. Based on the CHIP studies, one additional bioinformatic filter was instituted to limit the possibility that a positive GAS was derived from CHIP. CHIP samples most frequently exhibit only one aneuploid arm and are limited to a small number of candidate arms (Zheng et al., Bioinformatics, 37;2502-2503 (2021)). Any aneuploid sample (positive GAS) with only one aneuploid arm from chr 5q, 13q, 14q, 20q was therefore bioinformatically labeled as CHIP and the GAS set to 0. Evaluation of Plasma Proteins

[0106] The Bioplex 200 platform (Biorad, Hercules CA) was used to determine the concentration of multiple target proteins in the plasma samples. Luminex bead-based immunoassays (Millipore, Bilerica NY) were performed following the manufacturers protocols and concentrations were determined using 5 parameter log curve fits (using Bioplex Manager 6.0) with vendor-provided standards and quality controls. The HCCBP1MAG-58K panel was used to detect OPN, HGF, AFP, CA125, CA15-3, CEA, and CAI 9-9. The HTMP1MAG-54K panel was used to detect TIMP-1 .

[0107] Multi-analyte Classifier

[0108] Using 1402 non-cancer samples and 852 cancer samples (Cohort 2), a multi-analyte classifier was generated with A-PLUS scores, GAS, and the eight proteins described above as predictive features. The protein values were also supplemented with 126 negative controls from the Luminex bead-based immunoassays. Some samples were not assessed for proteins. To account for missing feature values in these samples, non-cancer samples were randomly sampled and assigned values. Feature values <98thpercentile in the non-cancer Cohort 2 samples were set to 0 to reduce possible overfitting to features with low scores. Cross validation was performed by randomly partitioning the dataset into 10 equally sized folds. The same seed was used to preserve the random partitions. Then a model was generated on 9 folds and tested on the 1 fold that was withheld. This protocol was repeated the remaining 9 folds. At the conclusion of cross validation, each sample was withheld once and the scores on the withheld fold were recorded.

[0109] External Validation using WGS data

[0110] Samples from Cristiano et al., (Nature, 570:385-389 (2019)) were downloaded from FinaleDB. These included 266 samples from patients with cancers of the Bile Duct, Breast, Colorectum, Stomach, Lung, Ovary, and Pancreas and 260 samples from individuals without cancer. Molecules that mapped to autosomal Alu elements were retrieved using overlapSelect from the kent src tools package using a Thresholds (the entire read had to fall within the Alu element). The AluS Fraction was calculated as the fraction of AluS molecules divided by the total molecules in Alu elements. WisecondorX was applied using default parameters and only considered autosomal chromosomes for the evaluation of aneuploidy in these samples. Default thresholds (z>5 or z<-5) were used and >3 autosomal copy number alterations were used as a threshold for positivity.

[0111] Statistical analyses

[0112] The two sample Z-test for equality of proportions was used to determine whether the sensitivity and specificity of Cohort 2 v . Cohort 3 were different. Cohen’s Kappa Statistic was used to assess whether concordance between technical replicates was different from random chance. Error bars represent confidence intervals as calculated by the Wilson score interval. Pearson’s correlation coefficient was used to compare A-PLUS scores to the AluS Fraction. Cohort samples size was not selected for statistical power but on the basis of sample availability and no samples previously reported were used in the validation set (Cohort 3).

[0113] Example 2: Exemplary Alu Element

[0114] Canonical nucleic acid sequence of a human Alu element (SEQ ID NO:1):

[0115] GGCCGGGCGCGGTGGCTCACGCCTGTAATCCCAGCACTTTGGGAGGCCGAGGCGGGAGGATC GCTTGAGCCCAGGAGTTCGAGACCAGCCTGGGCAACATAGCGAGACCCCGTCTCTACAAAAA ATACAAAAATTAGCCGGGCGTGGTGGCGCGCGCCTGTAGTCCCAGCTACTCGGGAGGCTGAG GCAGGAGGATCGCTTGAGCCCAGGAGTTCGAGGCTGCAGTGAGCTATGATCGCGCCACTGCA CTCCAGCCTGGGCGACAGAGCGAGACCCTGTCTC

[0116] Example 3: Assessing A Human Suspected of Having Cancer

[0117] A fluid sample (e.g., a plasma sample) is obtained from a human. The obtained fluid sample is examined for the AluS fraction in cfDNA.

[0118] If the cfDNA in the fluid sample has a reduced AluS fraction, then the human is identified as having cancer.

[0119] If the cfDNA in the fluid sample lacks a reduced AluS fraction, then the human is identified as not having cancer. Example 4: Treating cancer

[0120] A fluid sample (e.g., a plasma sample) is obtained from a human. The obtained fluid sample is examined for the AluS fraction in cfDNA using WGS.

[0121] If the cfDNA in the fluid sample has an AluS fraction of less than 0.66, then the human is administered or instructed to self-administer one or more (e.g., one, two, three, four, five, or more) cancer treatments.

[0122] Example 5: Treating cancer

[0123] A fluid sample (e g., a plasma sample) is obtained from a human. The obtained fluid sample is examined for the AluS fraction in cfDNA using RealSeqS. If the cfDNA in the fluid sample has an AluS fraction of less than 0.75, then the human is administered or instructed to self-administer one or more (e.g., one, two, three, four, five, or more) cancer treatments.

[0124] OTHER EMBODIMENTS

[0125] 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 assessing a mammal, wherein said method comprises:(a) determining a representation of Alu elements in cell-free DNA (cfDNA) in a fluid sample obtained from said mammal; and(b) identifying the mammal as having cancer if the representation of Alu elements comprises a presence of at least one of (1) a reduced AluS fraction, (2) an increased AluJ fraction, and (3) an increased AluY fraction.

2. A method for assessing a mammal, wherein said method comprises:(a) determining a representation of Alu elements in cell-free DNA (cfDNA) in a fluid sample obtained from said mammal; and(b) identifying the mammal as not having cancer if the representation of Alu elements comprises an absence of each of (1) a reduced AluS fraction, (2) an increased AluJ fraction, and (3) an increased AluY fraction.

3. The method of any one of claims 1-2, wherein said mammal is a human.

4. The method of any one of claims 1 -3, wherein said representation of Alu elements is determined using whole genome sequencing.

5. The method of claim 4, wherein said reduced AluS fraction is less than about 0.66, said increased AluJ fraction is greater than about 0.235, and said increased AluY fraction is greater than about 0.12.

6. The method of any one of claims 1-3, wherein said representation of Alu elements is determined using RealSeqS.

7. The method of claim 6, wherein said reduced AluS fraction is less than about 0.75, said increased AluJ fraction is greater than about 0.065, and said increased AluY fraction is greater than about 0.205.

8. The method of any one of claims 1-7, wherein said cancer comprises one or more solid tumors.

9. The method of any one of claims 1-8, wherein said fluid sample is a plasma sample.

10. A method for treating for a mammal having cancer, wherein said method comprises:(a) determining that a representation of Alu elements in cell-free DNA (cfDNA) in a fluid sample obtained from said mammal comprises the presence of at least one of (1) a reduced AluS fraction, (2) an increased AluJ fraction, and (3) an increased AluY fraction; and(b) administering a cancer treatment to said mammal.

11. A method for treating cancer, wherein said method comprises administering a cancer treatment to a mammal identified as comprising a fluid sample having cell-free DNA (cfDNA) comprising a representation of Alu elements having a presence of at least one of (1) a reduced AluS fraction, (2) an increased AluJ fraction, and (3) an increased AluY fraction.

12. The method of any one of claims 10-11, wherein said mammal is a human.

13. The method of any one of claims 10-12. wherein said representation of Alu elements is determined using whole genome sequencing.

14. The method of claim 13, wherein said reduced AluS fraction is less than about 0.66, said increased AluJ fraction is greater than about 0.235, and said increased AluY fraction is greater than about 0. 12.

15. The method of any one of claims 10-12, wherein said representation of Alu elements is determined using RealSeqS.

16. The method of claim 15. wherein said reduced AluS fraction is less than about 0.75, said increased AluJ fraction is greater than about 0.065, and said increased AluY fraction is greater than about 0.205.

17. The method of any one of claims 10-16, wherein said cancer comprises one or more solid tumors.

18. The method of any one of claims 10-17, wherein said cancer treatment comprises performing surgery.

19. The method of any one of claims 10-17, wherein said cancer treatment comprises radiation therapy.

20. The method of any one of claims 10-17, 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.

21. The method of any one of claims 10-20, wherein said fluid sample is a plasma sample.

Images