Methods for diagnosing homologous recombination deficiencies in human tumors

Abstract

The present invention relates to an improved method for diagnosing homologous recombination deficiencies (HRDs) in tumors. The method according to the present invention comprises the steps of: evaluating the number of large genomic alterations (LGAs) in a tumor sample by obtaining a copy number alteration (CNA) profile by shallow coverage whole-genome sequencing (sWGS); and determining an LGA score corresponding to the number of LGAs adjusted for the complexity of the tumor genome and the presence of one or two markers selected from a group of markers consisting of (1) phenotypes associated with mutations in cyclin-dependent kinase 12 (CDK12) with multiple intermediate gains in the CNA profile; (2) amplification of cyclin E1 (CCNE1); (3) amplification of human epidermal growth factor receptor-2 (HER2); and (4) phenotypes of amplification at multiple sites.

Description

[Technical Field]

[0001] The present invention relates to a method for diagnosing homologous recombination deficiencies in tumors. [Background technology]

[0002] BRCA1 / 2 mutant tumor cells have a defect in the HR pathway (homologous recombination deficiency or HRD), and therefore rely on alternative DNA repair pathways such as non-homologous end joining (NHEJ), alternative end joining (AltEJ), single-strand annealing (SSA), or base excision repair (BER) to avoid cell death, all of which involve the PARP1 and PARP2 enzymes (Groelly et al., Nat Rev Cancer, 2022). As a result, inhibiting these PARP1 / 2 enzymes in BRCA1 / 2-deficient tumor cells leads to cell death (Bryant et al., Nature 434: pp. 913-917, 2005; Farmer et al., Nature 434: pp. 917-912, 2005). PARPi represents a significant advance in the treatment of BRCA1 / 2-deficient tumors. Several clinical trials have demonstrated that maintenance with or without bevacizumab-containing PARP inhibitors, followed by platinum-based therapy, improves progression-free survival in patients with advanced ovarian cancer (AOC) with BRCA1 / 2 deficiency or HRD (Moore et al., N Engl J Med, 2018; Ray-Coquard et al., N Engl J Med 381:2416-2428, 2019; Gonzalez-Martin et al., N Engl J Med 381:2391-2402, 2019; Coleman et al., Lancet 390:1949-1961, 2017). Following the Myriad MyChoice CDx Plus test (MG test), as in the PAOLA-1 trial, showed a significant PFS benefit in patients with HRD-positive tumors, including those without BRCA1 / 2 mutations (BRCAmut). The olaparib (ova) + bevacizumab (bev) maintenance regimen was thus approved in the USA / Europe / Japan for patients with BRCAmut or HRD-positive tumors. However, MG testing was largely inconsistent and remained inconsistent until very recently.

[0003] Therefore, there is a need to develop novel, highly reliable, and feasible non-aggregated HRD tests.

[0004] The European HRD ENGOT Initiative (EHEI) is a unique European academic research collaboration aimed at providing reliable HRD biomarkers for selecting AOC patients who are most likely to benefit from PARPi ± bevacizumab as first-line treatment (Pujade-Lauraine et al., International Journal of Gynecologic Cancer 31:A208~A208, 2021).

[0005] Recently, a novel HRD test called Shallow HRD, based on shallow / low-coverage whole-genome sequencing (sWGS), has shown good performance in fresh, frozen samples (Eeckhoutte et al., Bioinformatics 36: pp. 3888-3889, 2020). sWGS is an easy and inexpensive technique and can be applied to clinical samples, including formalin-fixed paraffin-embedded (FFPE) samples. The bioinformatics pipeline for Shallow HRD is very simple and computationally easy. However, poor performance in noisy samples (which is common in clinical FFPE) and a high number of uninterpreted cases around the HRD cutoff have hindered the clinical application of this test. [Prior art documents] [Non-patent literature]

[0006] [Non-Patent Document 1] Groelly et al., Nat Rev Cancer, 2022 [Non-Patent Document 2] Bryant et al., Nature 434: pp. 913-917, 2005. [Non-Patent Document 3] Farmer et al., Nature 434: pp. 917-921, 2005. [Non-Patent Document 4] Moore et al., N Engl J Med, 2018 [Non-Patent Document 5] Ray-Coquard, I., Olaparib plus Bevacizumab as First-Line Maintenance in Ovarian Cancer. N Engl J Med 381, 2416~2428 pages (2019) [Non-licensed Document 6] Gonzalez-Martinら, N Engl J Med 381:2391~2402 pages, 2019 [Non-licensed Document 7] Coleman, Lancet 390:1949~1961 pages, 2017 [Non-licensed Document 8] Pujade-Lauraine, International Journal of Gynecologic Cancer 31:A208~A208 pages, 2021 [Non-licensed Document 9] Eeckhoutte, Bioinformatics 36:3888~3889 pages, 2020 [Non-licensed Document 10] Popova, Ploidy and large-scale genomic instability consistently identify basal-like breast carcinomas with BRCA1 / 2 inactivation, CANCER RES. (2012) 72:5454~5462 pages [Non-licensed Document 11] Boeva(2012), Bioinformatics, 28, pages 423~425 [Non-licensed Document 12] Popovaら、Cancer Res 2016、PMID 26787835 [Non-licensed Document 13] Boeva, V., Control-FREEC: a tool for assessing copy number and allelic content using next-generation sequencing data. Bioinformatics 28, pp. 423~425 (2012) [Non-Patent Document 14] Popova, T. et al., Ovarian Cancers Harboring Inactivating Mutations in CDK12 Display a Distinct Genomic Instability Pattern Characterized by Large Tandem Duplications. Cancer Res 76, pp. 1882-1891 (2016) [Non-Patent Document 15] Ray-Coquard, IL et al., Final overall survival (OS) results from the phase III PAOLA-1 / ENGOT-ov25 trial evaluating maintenance olaparib (ola) plus bevacizumab (bev) in patients (pts) with newly diagnosed advanced ovarian cancer (AOC). Ann Oncol 33, pp. S808-S869 (2022) [Non-Patent Document 16] Callens, C. et al., Concordance Between Tumor and Germline BRCA Status in High-Grade Ovarian Carcinoma Patients in the Phase III PAOLA-1 / ENGOT-ov25 Trial. J Natl Cancer Inst 113, pp. 917-923 (2021) [Non-Patent Document 17] Coussy, F. and Bidard, FC. Expanding biomarkers for PARP inhibitors. Nat Cancer 3, pp. 1141-1143 (2022). [Non-Patent Document 18] Gruber, JJ et al., A phase II study of talazoparib monotherapy in patients with wild-type BRCA1 and BRCA2 with a mutation in other homologous recombination genes. Nat Cancer 3, pp. 1181-1191 (2022) [Non-Patent Document 19] Loverix, L. et al., Predictive value of the Leuven HRD test compared with Myriad myChoice PLUS on 468 ovarian cancer samples from the PAOLA-1 / ENGOT-ov25 trial (LBA 6). Gynecologic Oncology 166, pp. S51-S52 (2022) [Non-Patent Document 20] Willing, E.-M. et al., 2022-RA-873-ESGO Validation study of the'NOGGO-GIS ASSAY' based on ovarian cancer samples from the first-line PAOLA-1 / ENGOT-ov25 phase-III trial. International Journal of Gynecologic Cancer 32, pp. A370-A370 (2022) [Non-Patent Document 21] Buisson, A. et al., 2022-RA-913-ESGO Clinical performance evaluation of a novel deep learning solution for homologous recombination deficiency detection. International Journal of Gynecologic Cancer 32, pp. A277-A278 (2022) [Non-Patent Document 22] Leman, R. et al., 2022-RA-935-ESGO Development of an academic genomic instability score for ovarian cancers. International Journal of Gynecologic Cancer 32, pp. A280 - A280 (2022)

Non-Patent Document 23

Summary of the Invention

Problems to be Solved by the Invention

[0007] Therefore, although shallow HRD is very promising, significant improvements are needed to meet the specificity and selectivity required to qualify as an efficient test for detecting HRD in samples such as FFPE samples.

Means for Solving the Problems

[0008] The present invention is defined by the claims.

[0009] The inventors have achieved a remarkable improvement in the performance of shallow HRD and made shallow HRD suitable for normal clinical applications.

[0010] The inventors have upgraded shallow HRD in two key respects: (1) detection of LGA count, which has become more reliable even in low-quality samples; and (2) determination rules for diagnosis, which have become more detailed and therefore provide a diagnosis for patients classified as having an intermediate LGA count, i.e., patients who could not be directly classified as exhibiting HRD or homologous recombination function (HRP).

[0011] The inventors have improved the accuracy of shallow HRD by adding a specific step to it, which consists of evaluating the number of large genomic alterations (LGAs) from copy number alteration (CNA) profiles obtained by shallow coverage whole-genome sequencing (sWGS), as described above. According to shallow HRD, the number of LGAs is calculated directly from the CNA profile, and the determination rule is based on the number of LGAs that have a large uncertain area around the cutoff (borderline). The main improvements brought about by the inventors consist of more precise LGA detection (due to noise reduction and quality control) and more reliable diagnosis in cases with borderline LGA counts (due to additional determination rules). These added steps include, in particular, noise reduction, classification of CNA profile quality and selection of adaptive thresholds for LGA calling, determination of specific genomic markers, such as genomic complexity and amplification, and multi-step diagnostic procedures.

[0012] Therefore, according to the first embodiment, the present invention is a method for diagnosing homologous recombination deficiency in a tumor, - A step in which the number of large-scale genomic alterations is evaluated by obtaining a copy number variation profile in a tumor sample using shallow coverage whole-genome sequencing (sWGS). - A process to determine the LGA score (also referred to as the "HRD score"), which corresponds to the LGA number adjusted for the complexity of the tumor genome and the presence of markers selected from a group of markers consisting of (1) phenotypes associated with mutations in cyclin-dependent kinase 12 (CDK12) with multiple intermediate gains in the CNA profile, (2) amplification of cyclin E1 (CCNE1), (3) amplification of human epidermal growth factor receptor-2 (HER2), and (4) phenotypes of amplification at multiple sites. This includes methods.

[0013] The method according to the present invention is simple, easily reproducible, and enables accurate diagnosis of HRD in degraded samples such as paraffin-embedded specimens. The present invention therefore enables the identification of patients who would benefit from treatment including PARP inhibitors and / or alkylating agents.

[0014] Accordingly, according to another embodiment, the present invention relates to a PARP inhibitor (PARPi) and / or alkylating agent for use in a method for treating cancer in a patient diagnosed according to the method of the present invention as having a tumor exhibiting HRD.

[0015] The present invention also relates to a method for predicting the effectiveness of a treatment in a patient with cancer, wherein the treatment comprises a PARPi and / or an alkylating agent, and comprises a step of diagnosing HRD in a tumor sample as described herein. [Brief explanation of the drawing]

[0016] [Figure 1] This figure shows the main steps of the shallow WGS approach and shallow HRDv2 pipeline for effective HRD diagnosis. Abbreviations: CNA = copy number variation, FF = fresh frozen sample, FFPE = formaldehyde-fixed and paraffin-embedded sample, HRD = homologous recombination defect. [Figure 2] This diagram illustrates the workflow of the shallow HRDv2 pipeline concept. [Figure 3](A) This figure shows the Kaplan-Meier estimates of PFS for PAOLA-1 patients according to homologous recombination status and treatment arm determined by Shallow HRDv2 or MyChoice. (B) This figure shows the Kaplan-Meier estimates of OS for PAOLA-1 patients according to homologous recombination status and treatment arm determined by Shallow HRDv2 or MyChoice. (C) This figure shows the Kaplan-Meier estimates of PFS according to homologous recombination status and treatment arm determined by Shallow HRDv2 for PAOLA-1 patients in whom MyChoice did not contribute results. HR is the hazard ratio, PFS is progression-free survival, and OS is overall survival. [Figure 4] (A) Kaplan-Meier estimates of PFS, and (B) Kaplan-Meier estimates of OS according to the homologous recombination status and treatment arm determined by Shallow HRDv2 or MyChoice in PAOLA-1 patients with tumors in the wild type of the BRCA1 / 2 gene. HR is the hazard ratio, PFS is progression-free survival, and OS is overall survival. [Figure 5A] This figure shows two examples of training sets (A and B) and the quality attributes used to refine the decision rules. In A, the numbers refer to the number of cases in the training set with corresponding tumor volume and noise features. [Figure 5B] This figure shows two examples of training sets (A and B) and the quality attributes used to adjust the decision rules. [Figure 6] This figure shows an example of the LGA score distribution in the training dataset (A), and an adaptive model for score correction at the "borderline" score (B). [Figure 7] This figure shows another example of the LGA score distribution in the training dataset (A), an adaptive model for LGA score correction in the "borderline" case (B), and the basic characteristics of HRD and non-HRD cases with borderline LGA scores. [Modes for carrying out the invention]

[0017] The present invention relates to an improved method for diagnosing homologous recombination deficiencies (HRDs) in tumors. The method according to the present invention aims to refine the diagnosis obtained by evaluating the number of large genomic alterations (LGAs) by shallow coverage whole-genome sequencing (sWGS).

[0018] According to a first aspect, the present invention is a method for diagnosing homologous recombination deficiency in a tumor, - A step in which the number of large-scale genomic alterations is evaluated by obtaining a copy number variation profile in a tumor sample using shallow coverage whole-genome sequencing (sWGS). - A process to determine an LGA score that corresponds to the complexity of the tumor genome and the presence of a marker selected from a group of markers consisting of (1) a phenotype associated with mutations in cyclin-dependent kinase 12 (CDK12) with multiple intermediate gains in the CNA profile, (2) amplification of cyclin E1 (CCNE1), (3) amplification of human epidermal growth factor receptor-2 (HER2), and (4) a phenotype of amplification at multiple sites. This includes methods.

[0019] As used herein, the term “homologous recombination (HR) pathway” has its general meaning in the art. It refers to a cellular pathway that repairs double-strand DNA breaks (DSBs) through a mechanism called homologous recombination. Within mammalian cells, DNA is continuously exposed to damage from exogenous sources such as ionizing radiation or endogenous sources such as byproducts of cell replication. All organisms have evolved various strategies to cope with these damages. One of the most detrimental forms of DNA damage is DSBs. HR is the most precise mechanism for repairing DSBs because it uses intact DNA copies derived from sister chromatids or homologous chromosomes as substrates for repairing the breaks.

[0020] Cells identified as having genomic DNA rearrangements (e.g., large genomic alterations or LGAs) (e.g., cancer cells) can be classified as having an increased likelihood of HR deficiency, i.e., one or more genes in the HR pathway being in a deficiency state. As used herein, “deficiency state” of a gene means that the sequence, structure, expression, and / or activity of the gene or its product is deficient compared to normal. Examples include, but are not limited to, low or absent mRNA or protein expression, harmful mutations, hypermethylation, and weakened activity (e.g., enzyme activity, ability to bind to another biomolecule). As used herein, a deficiency state of a pathway (e.g., the HR pathway) means that at least one gene in that pathway (e.g., BRCA1) is deficient. Examples of highly harmful mutations include frameshift mutations, stop codon mutations, and mutations that alter RNA splicing. A deficiency state of a gene in the HR pathway can result in a deficiency or reduction of HR activity within the cell (e.g., cancer cells).

[0021] Examples of genes involved in the HR pathway include, but are not limited to, BRCA1, BRCA2, PALB2 / FANCN, BRIP1 / FANCJ, BARD1, RAD51, and RAD51 paralogs (RAD51B, RAD51C, RAD51D, XRCC2, XRCC3). These genes encode proteins crucial for repairing double-strand DNA breaks via the HR pathway. If any of these protein genes are, for example, mutated or underexpressed, the changes can lead to errors in DNA repair, which can ultimately result in cancer. Other genes involved in the HR pathway include FANCA, FANCB, FANCC, FANCD2, FANCE, FANCG, FANCI, FANCL, FANCM, FAN1, SLX4 / FANCP, or ERCC1.

[0022] Therefore, the terms "HR pathway deficiency" or "HRD," as used herein, refer to a state in which one or more proteins involved in the HR pathway for DNA repair are missing or inactivated. Conversely, "HR pathway functionality" or "HRP" refers to the absence of HRD.

[0023] Proteins involved in the HR pathway may include, but are not limited to, inactivation of at least one of the following genes: BRCA1, BRCA2, PALP2 / FANCN, BRIP1 / FANCJ, BARD1, RAD51, RAD51 paralogs (RAD51B, RAD51C, RAD51D, XRCC2, XRCC3), FANCA, FANCB, FANCC, FANCD2, FANCE, FANCG, FANCI, FANCL, FANCM, FAN1, SLX4 / FANCP, and ERCC1.

[0024] As used herein, the term “inactivation” can mean any type of deletion of a gene, including, but not limited to, germline mutations in the coding sequence, somatic mutations in the coding sequence, mutations in the promoter, and methylation of the promoter.

[0025] The method according to the present invention enables the diagnosis of HRD in tumors.

[0026] According to the present invention, “tumor” can be any solid tumor or cancer. Preferably, the solid tumor or cancer is selected from cancers of the bladder, breast, colon, kidney, liver, lung, pancreas, stomach, esophagus, uterus, cervix, thyroid, or skin, including breast cancer, ovarian cancer, colon cancer, lung cancer, prostate cancer, kidney cancer, metastatic or invasive malignant melanoma, brain tumor, bladder cancer, fallopian tube cancer, head and neck cancer, peritoneal cancer, liver cancer, and squamous cell carcinoma. However, the present invention also considers hematopoietic malignancies such as leukemia, acute lymphoblastic leukemia, acute lymphoblastic leukemia, B-cell lymphoma, T-cell lymphoma, Hodgkin lymphoma, non-Hodgkin lymphoma, hairy cell lymphoma, Burkitt lymphoma, acute and chronic myeloid leukemia, and promyelocytic leukemia.

[0027] In certain embodiments, the tumor is selected from ovarian cancer, breast cancer, fallopian tube cancer, peritoneal cancer, lung cancer, pancreatic cancer, head and neck cancer, prostate cancer, gastric or esophageal cancer, uterine cancer, cervical cancer, kidney cancer, and bladder cancer.

[0028] In certain embodiments, the tumor is breast cancer or ovarian cancer, for example, advanced ovarian cancer.

[0029] As used in the context of this invention, a tumor "sample" is typically obtained from a tumor biopsy. Such a sample may be, for example, a fresh sample or a preserved sample, such as a frozen sample or any tumor sample preserved by other means. A tumor sample may typically be in the form of a formalin-fixed paraffin-embedded (FFPE) sample. An FFPE sample consists of a tumor sample fixed in formaldehyde and then embedded in a paraffin wax block. FFPE samples are prepared and used as is customary by those skilled in the art.

[0030] As used herein, the terms “patient” or “subject” refer to mammals, such as rodents, cats, dogs, cattle, horses, sheep, pigs, or primates. Preferably, the patient according to the present invention is a human.

[0031] A "Large Genome Alteration" or "LGA" corresponds to a genome rearrangement. An LGA refers to any somatic cell copy number transition (e.g., a breakpoint) along the length of a chromosome between two regions of at least a certain minimum length (e.g., at least 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, or 20, or more, megabases) after filtering out regions shorter than a certain maximum length (e.g., 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1, 1.5, 2, 2.5, 3, 3.5, 4, or more megabases). For example, if, after filtering out regions shorter than 3 megabases, somatic cells have a 1:1 copy number over, for example, at least 10 megabases, followed by a breakpoint transition to, for example, a region of at least 10 megabases with a 2:2 copy number, this is one LGA. An alternative way to define the same phenomenon is the LGA region, which is a genomic region with a stable copy number over at least a certain minimum length (e.g., at least 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, or 20 megabases) with a breakpoint (e.g., transition) as its boundary, where the copy number changes at this breakpoint to form another region of at least this minimum length. For example, if, after filtering out regions shorter than 3 megabases, a somatic cell has a region of at least 10 megabases with a copy number 1:1, on one side of which a breakpoint transition is boundary to a region of at least 10 megabases with a copy number 2:2, and on the other side a breakpoint transition is boundary to a region of at least 10 megabases with a copy number 1:2, then this is two LGAs. It should be noted that this is broader than allele imbalance, because such copy number variations are not considered allele imbalance (for example, because copy ratios 1:1 and 2:2 are identical, meaning there is no change in copy ratio).The use of LGA in determining HRD is described in detail by Popova et al. (Ploidy and large-scale genomic instability consistently identify basal-like breast carcinomas with BRCA1 / 2 inactivation, CANCER RES. (2012) 72:5454~5462).

[0032] According to the present invention, the number of LGAs is determined by shallow coverage whole-genome sequencing. "Shallow coverage whole-genome sequencing" or "sWGS" can detect the number of LGAs obtained from whole-genome sequencing (WGS) at low coverage, e.g., 0.3 or higher, e.g., approximately 1x coverage. Coverage corresponds to the average number of reads aligned to a known reference base. In the context of the present invention, sWGS is used to provide a "copy number variation profile" or "CNA" profile. The method enabling the determination of the said profile based on sWGS is fully disclosed in Eeckhoutte et al. (Bioinformatics 36: pp. 3888-3889, 2020). This method, called "shallow HRD," consists of processing sWGS of tumor samples with Control-FREEC software (Boeva ​​et al. (2012), Bioinformatics, 28, pp. 423-425). This tool takes CNA profile {x,g} as input. 1,N Using "sample_name.bam_ratio.txt", which contains the following, in the formula, x is the normalized read count within the sliding window, g is the genome coordinate, and profile segmentation is the median and size (megabases, Mb units) of the segment. i , Z i It will be held at [location].

[0033] Shallow HRD provides CNA profiles based on the following workflow: 1. Detect CNA cutoff and optimize profile segmentation as follows: Z i ≧(Q1 + Q3) / 2 (where Q1 and Q3 are the quartiles of the i (Z i > 3Mb) distribution), then the segment is defined as "large". Detect M as the first minimum of |(S i - S j )| density (where i and j are large segments). CNA cutoff = min(max(0.025, M), 0.45). If |(S i - S i+1 )| < CNA cutoff, merge adjacent segments starting from the largest segment. 2. Define LGA as a CNA break within a chromosomal arm where adjacent segments Z i , Z i+1 ≧ 10Mb, and count after removing segments < 3Mb. 3. Annotate the sample as "non - HRD" (LGA < 15), "borderline" (15 ≦ LGA ≦ 19), or "HRD" (LGA > 19). 4. Define the quality of the sample by M and cMAD (cMAD = median(|(x - S x )|), where S x corresponds to the segment containing x), and then optimize: "poor" (cMAD > 0.5 | cMAD > 0.14 and M > 0.45), "average" (cMAD > 0.14 and M < 0.45 | cMAD < 0.14 and M > 0.45), or "normal or highly contaminated" (M < 0.025). 5. If S c ≧ 4 · CNA cutoff (where c is the segment containing the gene), call CCNE1 amplification.

[0034] In the context of the present invention, the diagnosis is provided based on an LGA score, also referred to as the "HRD score," which corresponds to the complexity of the tumor genome and the LGA number, adjusted by the presence of one (i.e., one, two, three, or all four) markers selected from a group of markers consisting of (1) phenotypes associated with mutations in cyclin-dependent kinase 12 (CDK12) with multiple intermediate gains in the CNA profile, (2) amplification of cyclin E1 (CCNE1), (3) amplification of human epidermal growth factor receptor-2 (HER2), and (4) phenotypes of amplification at multiple sites.

[0035] In the context of the present invention, the score correlates with HRD, a high LGA score is associated with HRD, and a low LGA score is associated with HRP. For example, according to the specific embodiments shown in the examples of the present invention, a high score corresponds to a score of 20 or 23 or higher, and a low score corresponds to a score of 17 or lower.

[0036] The complexity of a genome is estimated by assigning samples to two categories: "simple" and "complex," where a "simple" genome has two most frequent copy number levels that account for more than 70% of the genome. Otherwise, the genome is classified as "complex." According to further embodiments, a genome may also be classified as "complex+", which is a subtype of "complex" and means that it has four or more copy number levels of equal frequency. All cases of low tumor burden are annotated as "simple."

[0037] In the context of this invention, a “simple” genome is considered to correlate with HRD. Therefore, a simple genome is assigned a modification constant that may qualify as a “positive” or “bonus” that adjusts the diagnosis toward HRD.

[0038] Conversely, "complex" genomes are considered to correlate with HRP. Therefore, "complex" genomes are assigned a modification constant that may qualify as a "negative" or "penalty" that adjusts the diagnosis toward HRP.

[0039] A method according to the present invention includes a step of adjusting the LGA number based on the presence of one, two, three, or four markers selected from a group of markers consisting of (1) a phenotype associated with a mutation in cyclin-dependent kinase 12 (CDK12) with multiple intermediate gains in the CNA profile, (2) amplification of cyclin E1 (CCNE1), (3) amplification of human epidermal growth factor receptor-2 (HER2), and (4) a phenotype of amplification at multiple sites.

[0040] Cyclin-dependent kinase 12, or CDK12, is a protein kinase that functions as an important regulator of transcriptional elongation. It regulates the expression of genes involved in DNA repair and is necessary for maintaining genomic stability. The amino acid sequence of human CDK12 is available in the Uniprot database under reference Q9NYV4. CDK12 is encoded by the CDK12 gene, which has the nucleic acid sequence shown in the Ensembl genome database under reference ENSG00000167258.

[0041] According to the present invention, the LGA number is adjusted based on the presence of a phenotype associated with a CDK12 mutation accompanied by multiple intermediate gains in the CNA profile. This marker indicates the detection of a tandem overlap phenotype associated with a CDK12 mutation (CDK12mut) based on the number of intermediate gains between 1 and 10 Mb (Popova et al., 2016). In the context of the present invention, association with CDK12mut is detected based on the number of intermediate gains between 1 and 10 Mb being greater than the number of smaller intermediate gains and / or intermediate losses.

[0042] In the context of this invention, the presence of a phenotype associated with CDK12 mutations accompanied by multiple intermediate gains in the CNA profile is considered to correlate with HRP. Therefore, the presence of this marker may qualify as a “negative” or “penalty” on the scale from HRP to HRD, and thus a modification constant is assigned to adjust the diagnosis toward HRP.

[0043] Cyclin E1, or CCNE1, is a regulatory protein essential for regulating the cell cycle through its interaction with CDK2. The amino acid sequence of the CCNE1 protein is available in the Uniprot database under reference P24864, and the sequence of the human CCNE1 gene is shown in the Ensembl genome database under reference ENSG00000105173.

[0044] According to the present invention, the LGA number is adjusted based on "CCNE1 amplification." This marker is defined as the fact that chromosome 19 has a high copy number gain (e.g., amplification) at the CCNE1 locus.

[0045] In the context of this invention, the amplification of CCNE1 is considered to correlate with the presence of HRP. Therefore, the presence of CCNE1 amplification is assigned a modification constant that may qualify as a “negative” or “penalty” that adjusts the diagnosis toward HRP.

[0046] Human epidermal growth factor receptor-2 or HER2, also known as "receptor tyrosine-protein kinase erbB-2," "ERBB2," "differentiation antigen group 340," or "CD340," is a protein tyrosine kinase encoded by the ERBB2 gene. The amino acid sequence of the HER2 protein is available in the Uniprot database under reference P04626, and the sequence of the human ERBB2 gene is shown in the Ensembl genome database under reference ENSG00000141736.

[0047] According to the present invention, the LGA number is adjusted based on the detection of "HER2 amplification." This marker is defined as the fact that chromosome 17 has a high copy number gain (e.g., amplification) at the HER2 locus.

[0048] In the context of this invention, HER2 amplification is considered to correlate with the presence of HRP. Therefore, the presence of HER2 amplification is assigned a modification constant that may qualify as a “negative” or “penalty” that adjusts the diagnosis toward HRP.

[0049] According to the present invention, the LGA number is adjusted based on the detection of a "phenotype of amplification at multiple locations." This marker is defined as the fact that three or more chromosomal arms have a high copy gain (amplification) in at least one given region.

[0050] In the context of this invention, amplification at multiple locations is considered to correlate with the presence of HRP. Therefore, the presence of amplification at multiple locations is assigned a modification constant that may qualify as a “negative” or “penalty” that adjusts the diagnosis toward HRP.

[0051] If the tumor sample is an FFPE sample, the method according to the present invention may, advantageously, include a step of correcting the CNA profile from which the LGA count is calculated by eliminating, i.e., significantly reducing, false positive breakpoints that correlate with the noise profile, i.e., by determining the cumulative noise profile of the FFPE. The noise correction may, advantageously, involve segmentation optimization.

[0052] This process is advantageously performed before determining the HRD score, as explained above.

[0053] The cumulative noise profile of FFPE is obtained from segmented CNA profiles of approximately 100 nearly normal genomes sequenced from FFPE tumor or FFPE normal samples. The mean of each segment is replaced with 1 or -1 if it is greater than or less than the normal mean, respectively. Each genome bin is thus characterized by the sum of 1s, 0s, and -1s obtained from approximately 100 profiles. When the sample is an FFPE tumor sample, the noise profile of FFPE has distinct peaks and valleys at certain locations in the genome. A correlation of >0.2 between the tumor CNA and the cumulative profile of FFPE, and a percentage of genomes that are classified as large segments (>20Mb) after segmentation (<40%) are characteristic of high FFPE noise.

[0054] Segmentation optimization is performed by filtering out small segments and merging segments with small differences in their mean values ​​(small difference means below a threshold). FFPE noise reduction is performed under the same logic: if the dynamics of adjacent segments locally correlate with the cumulative noise profile of FFPE, those segments are merged together.

[0055] The threshold for considering differences between segments negligible is selected based on noise and estimated tumor volume. After merging all adjacent segments with small mean differences, the profile is considered optimized. The threshold for CNA, i.e., the breakpoint call, is tailored to the quality category of each sample and is typically set to twice the size of the previous threshold.

[0056] As described above, the method according to the invention can advantageously be used to determine border cases, i.e. cases where the score does not correspond to either HRD or HRP. For example, when the score is calculated as shown in the following examples, the borderline cases correspond to an LGA score of 17 < LGA score < 20 or an LGA score of 23. In such cases, the method according to the invention further comprises the step of determining, and more precisely adjusting, the score by taking into account the following: the complexity of the tumor genome, (1) the phenotype associated with a mutation in cyclin-dependent kinase 12 (CDK12) with multiple intermediate gains in the CNA profile, (2) the amplification of cyclin E1 (CCNE1), (3) the amplification of human epidermal growth factor receptor-2 (HER2), and (4) the presence of a marker selected from a marker group consisting of the phenotype of amplifications at multiple locations, and an auxiliary cumulative LGA indicator (LGA_boost), or an LGA_max corresponding to the upper limit of the LGA count seen in an optimized segmented genome profile not restricted by the threshold of the LGA call.

[0057] The complexity of the genome, and the presence of a marker selected from the marker group consisting of (1) the phenotype associated with a mutation in cyclin-dependent kinase 12 (CDK12) with multiple intermediate gains in the CNA profile, (2) the amplification of cyclin E1 (CCNE1), (3) the amplification of human epidermal growth factor receptor-2 (HER2), and (4) the phenotype of amplifications at multiple locations are determined as described above. A method for further adjusting the score for borderline cases based on these elements is disclosed in FIGS. 6 or 7.

[0058] When adjusting the diagnosis with the auxiliary cumulative LGA indicator (LGA_boost), said indicator is calculated by the following formula: LGA_boost = LGA_chr_arm + LGA_at_telomere + LGA_20Mb + LGA_baseline + LGA_baseline_12 where, in the formula, LGA_chr_arm is the number of chromosome arms that have LGA. LGA_at_telomere is the number of chromosomal arms with LGA at the telomere end. LGA_20Mb is the number of LGAs where both genomic segments exceed 19Mb at the copy number (CN) break. LGA_baseline is the number of LGAs with the most frequent CN layer. LGA_baseline_12 is the number of LGAs detected between the two most frequent CN layers.

[0059] The method LGA_boost, used to adjust the diagnosis, is explained in Figure 6.

[0060] When adjusting the diagnosis using the LGA_max index, the index is determined to be the upper limit of LGA counts seen in an optimized segmented genomic profile that is not limited by the threshold for LGA calling. LGA_max = LGA + potentially missed LGA, where potentially missed LGA has a breakpoint size smaller than the threshold for LGA calling.

[0061] The method LGA_max used for adjusting the diagnosis is described in detail in the Experiments section and Figure 7 of this application.

[0062] According to a preferred embodiment, the method according to the present invention includes the following steps (Figure 2): 1) A step to obtain a copy number change profile determined by evaluating the LGA number by sWGS in tumor samples (by normalizing the sWGS read count profile in tumor samples), 2) Noise reduction and segmentation optimization, 3) Characterization of genome profiles: - Estimation of genome complexity. - Overall signal quality attributes (four categories: "Good," "Moderate," "Poor," and "Bad"): These are based on tumor volume and noise and define the path to final diagnosis, including the criteria for an undetermined state (ND). - Analysis of CNA breakpoints: Detection of tandem duplication phenotypes associated with CDK12 mutations (CDK12mut) based on the call of large genomic alterations (LGAs) and the number of intermediate gains of 1-10 Mb (Popova et al., Cancer Res 2016, PMID 26787835). - Verification of the phenotypes of amplification at CCNE1, amplification at ERBB2, and amplification at multiple points. 4) Attributes of LGA score and HRD status and final diagnosis (i.e., HRD or HRP): LGA Score = LGA + Bonus - Penalty The final diagnostic criteria are based on the LGA score, sample quality attributes, and CNA profile characteristics.

[0063] Since it is possible to predict whether a given patient has cancer associated with HRD, it is also possible to select appropriate treatment for the patient.

[0064] Where applicable, patients with cancer cells identified as having genomic DNA rearrangements (e.g., LGA) may be classified as more likely to respond to certain cancer treatment regimens. For example, patients with cancer cells having a genome containing genomic DNA rearrangements may be classified as more likely to respond to cancer treatment regimens involving the use of DNA damaging agents, synthetic lethal agents (e.g., PARP inhibitors), radiation therapy, or a combination thereof.

[0065] Accordingly, another aspect of the present invention relates to a method for predicting the effectiveness of a treatment in a patient with cancer, wherein the treatment comprises PARPi and / or an alkylating agent, and comprises the step of diagnosing HRD in a tumor sample as described above.

[0066] The present invention also relates to PARPi and / or alkylating agents for use in a method for treating cancer in a patient diagnosed according to the method of the present invention as having a tumor exhibiting HRD.

[0067] As used herein, the terms “PARP inhibitor” or “PARPi” have their usual meanings in the art. This refers to compounds that can inhibit the activity of poly-ADP-ribose polymerase (PARP), an enzyme that is a protein crucial for repairing single-strand breaks (“nicks” in DNA). If such nicks remain unrepaired until DNA replication occurs (which must occur before cell division), the replication itself forms double-strand breaks. Drugs that inhibit PARP thus cause the formation of multiple double-strand breaks, and in tumors with BRCA1, BRCA2, or PALB2 mutations, these double-strand breaks cannot be efficiently repaired, leading to cell death.

[0068] Typically, PARP inhibitors according to the present invention may be selected from the group consisting of iniparib, olaparib, rucaparib, CEP 9722, MK 4827, BMN-673, and 3-aminobenzamide.

[0069] As used herein, the terms “alkylating agent” or “alkylating anti-cancer agent” have their common meanings in the art. This refers to a compound that binds an alkyl group to DNA. Typically, alkylating agents according to the present invention may be selected from platinum complexes such as cisplatin, carboplatin, and oxaliplatin, chlormethine, chlorambucil, melphalan, cyclophosphamide, ifosfamide, estramustine, carmustine, lomustine, fotemustine, streptozosin, busulfan, pipobromane, procarbazine, dacarbazine, thiotepa, and temozolomide.

[0070] The method according to the present invention can be advantageously carried out by a computer by using a computer program adapted to reproduce the steps disclosed above.

[0071] In another aspect, the present invention provides a computer program product mounted on a computer-readable medium that, when executed by a computer, provides instructions for evaluating the number of LGAs in a tumor sample, correcting the sample noise profile, and providing an HRD diagnosis according to the method of the present invention. [Examples]

[0072] Achieving clinical reliability in shallow whole-genome sequencing approaches for detecting homologous recombination defects in tumors. method: Patient and tumor samples The first cohort consisted of FFPE-derived DNA obtained from 449 AOC samples in the PAOLA-1 / ENGOT-ov25 trial within the EHEI framework. All patients submitted written informed consent. The second cohort consisted of 109 consecutive FFPE AOC samples (8 to 20 5 μm slides depending on the tumor segment), which were sent to Myriad Genetics' central laboratory (Salt Lake City, UT, USA) from March 2021 to January 2022 as part of routine procedures. In parallel, shallow HRDv2 was performed on the same FFPE samples in the genetics laboratory at the Curie Institute.

[0073] statistical analysis Progression-free survival (PFS) and overall survival (OS) were estimated using the Kaplan-Meier method, and the difference between the olaparib (ola) + bevacizumab (bev) group and the bev group was evaluated using a stratified log-rank test. Hazard ratios (HR) and their corresponding 95% confidence intervals (95% CI) were calculated using a stratified Cox proportional hazards model. All statistical analyses were performed using GraphPad Prism software version 9.1.0.

[0074] Shallow WGS Workflow 100 ng of FFPE DNA was used as input. Mechanical DNA fragmentation was performed using a 50 μl DNA sample with a Covalis (Model ME220 single-point focused ultrasonic device). The inventors followed the supplier's recommendations for Agilent kits (SureSelect XT HS and XT Low Input Library Preparation, G9703A). Different steps consisted of ligation, amplification, and purification with AMpure XP beads (Beckman Coulter, reference number A63882). Dosage was performed using Thermo Fisher Scientific Qubit® dsDNA HS assay kit (reference number: Q32854) or Qubit® dsDNA BR assay kit (reference number: Q32853). After adjusting the quality and quantity using Agilent TapeStation and D1000 ScreenTape, the inventors prepared 4nM and 1.8nM library pools in NextSeq 550 S or NovaSeq 6000 sequencing systems, respectively (Illumina Inc., San Diego, CA, USA).

[0075] Shallow HRDv2 Bioinformatics Pipeline Following DNA extraction and whole-genome sequencing at low coverage (approximately 1x), normalized and corrected read count profiles (approximately 50kb bin size) for GC levels were obtained using ControlFreec (Boeva, V. et al., Control-FREEC: a tool for assessing copy number and allelic content using next-generation sequencing data. Bioinformatics 28, 423-425 (2012)) (Figure 1). The shallow HRDv2 bioinformatics pipeline consists of analysis of copy number variation (CNA) profiles, sample quality attributes, and comprehensive quantitative and graphic outputs for manual control, providing HRD diagnostics. The main steps of the pipeline, an overview of the decision rules, and diagnostics are described below and in Figure 1.

[0076] The main steps of the CNA process are as follows (Figure 2): (1) Three types of sample quality attributes: classification of CNA profiles according to tumor volume (four categories), intrinsic sWGS noise (three categories), and FFPE noise (four categories), and a final integrated classification of the signal to four categories of noise: "good," "moderate," "insufficient," and "poor" (Figure 5B).

[0077] (2) Noise reduction and breakpoint optimization in CNA profiles: Small segments are filtered out, and thresholds for breakpoint calls appropriate to each quality category are used to assemble segments with small differences or segments that have a local correlation with the FFPE noise profile (obtained from approximately 100 normal profiles obtained from FFPE samples, Figure 1).

[0078] (3) Characterization of broad CNA profiles by the following: - Genome complexity: Here, a "simple" genome has two most frequent copy number (CN) levels that account for more than 70% of the genome; otherwise, the genome is classified as "complex." - A set of binary attributes, e.g., CCNE1 amplification, ERBB2 amplification, local width phenotype (called when three or more chromosomal arms have at least one amplification), and tandem duplication phenotype associated with CDK12 mutations (called when multiple intermediate gains of 1-10 Mb are detected) (Popova, T. et al., Ovarian Cancers Harboring Inactivating Mutations in CDK12 Display a Distinct Genomic Instability Pattern Characterized by Large Tandem Duplications. Cancer Res 76, pp. 1882-1891 (2016)). - A set of parameters that characterize breakpoints, including the total number and the number of large genomic alterations (LGAs), which significantly contribute to HRD diagnosis. LGAs are defined as CN breaks between genomic segments larger than 9 Mb (segment sizes are rounded to integers).

[0079] (4) Multi-stage HRD diagnosis based on the following: - LGA Score: This is essentially the LGA number modified by penalties and bonuses, where penalties are determined by binary attributes and subtracted from the LGA number (the penalty is set to 0, 5, or 8 if there are 0, 1, or 2 or more binary attributes), and bonuses are determined by genome complexity and added to the LGA number (the bonus is set to 5 for a "simple" genome and 0 otherwise). - Two thresholds for a clear HRD diagnosis: 17 and 20: LGA score < 17 for non-HRD, LGA score ≥ 20 for HRD. - Modification of the LGA score applied to determine the borderline cases (17 < LGA score < 20). Briefly stated, the LGA score is shifted to 21 if HRD is proven (the genome is classified as "simple" with penalty = 0 and LGA_max ≥ 14, or the genome is classified as "complex" with LGA_max ≥ 20) (Figure 7).

[0080] Decision rules. The decision rules are multi - step, determined by the quality attributes of the sample, and include the selection of thresholds for LGA calls. To call LGA at CNA breakpoints, two thresholds are used: stringent (suggesting a simple genome) and soft (suggesting a complex genome). In good - quality cases, these are applied in a conservative manner (the LGA score is based on the number of LGAs at the soft / stringent thresholds for a clear non - HRD / HRD diagnosis), and in noisy samples / low - tumor - volume samples, they are fixed at the stringent / soft thresholds. The simplified decision rule for poor - quality samples consists of providing a diagnosis only for clear non - HRD cases with a small total number of breakpoints. Most poor - quality cases are excluded. The robust decision rule for insufficient - quality samples consists of providing a diagnosis only for clear cases and leaving borderline cases with an undetermined (ND) diagnosis. Additional rules for the LGA score modification procedure in good / moderate - quality borderline cases assist in the determination of the diagnosis and reduce undetermined cases.

[0081] Comprehensive output (report). The final diagnosis is reported together with the quality assessment and warning messages. The quantitative output provides complete information about definitive genomic biomarkers, LGA, LGA score, and HRD diagnosis. The output includes a segmented profile with detected LGAs and an error profile for visually controlling the quality of noise reduction and segmentation. <communicated - by - reference - number -

[0082] <communicated - by - reference - number - The circular binary segmentation of the CNA profile has a probabilistic component, which can lead to variations in the LGA count between runs, potentially affecting the diagnosis when close to the threshold. Therefore, the LGA count is reported as the mean estimated from 21 segmentation / optimization runs, along with the associated standard error.

[0083] Detailed workflow for Shallow HRDv2. Step 1) Segmentation of CNA profiles using a circular binary segmentation method, and classification of CNA profiles according to tumor volume (4 categories), intrinsic sWGS noise (3 categories), and FFPE noise (4 categories). Combinations of these attributes provide an overall sample quality attribute ("good," "medium," "insufficient," or "poor"), which is used for definitive pipeline selection and reporting (Figure 5B). - The profile is characterized by the number of breakpoints; variance of the CNA profile, variance within segments, variance between segments; correlation with FFPE noise; and the percentage of genomes belonging to segments >20Mb after segmentation. - The quality of sequencing is characterized by raw variance (variance within segments). - Tumor volume is characterized by the variance of the median of the large segments. - FFPE noise is characterized by the number of breakpoints, the correlation with the cumulative FFPE profile, the variance of the error profile, and the proportion of large segments after the initial profile segmentation. The FFPE noise profile is obtained from segmented CNA profiles of approximately 100 nearly normal genomes sequenced from the FFPE sample. The mean of each segment is replaced with 1 or -1 if it is greater than or less than the normal mean, respectively. Each genome bin is thus characterized by the sum of 1 / 0 / -1 obtained from approximately 100 profiles. A correlation of >0.2 between tumor CNA and the cumulative FFPE profile, and a proportion of genomes in large segments (>20Mb) after segmentation (<40%) are characteristic of high FFPE noise.

[0084] Step 2) Noise Correction and Segmentation Optimization: Small segments are filtered out, and segments with small median differences or local correlations to the FFPE noise profile are merged. A threshold for considering segment differences negligible (no breakpoint calls) is selected depending on the noise and tumor volume categories. Adjacent segments were merged if the median difference was below the threshold. Additionally, if the breakpoints followed the breakpoints in the FFPE covariate profile, these breakpoints were excluded even if the difference exceeded the threshold.

[0085] Step 3) Characterization of the genome profile: - Estimation of genome complexity (two categories: "simple" and "complex": a "simple" genome has two most frequent copy number levels that account for more than 70% of the genome. If not, the genome is classified as "complex." All cases with low tumor burden are annotated as "simple"). - Overall profile quality attributes (four categories: "Good," "Moderate," "Poor," and "Bad"): These are based on tumor volume and noise, characterizing the signal compared to noise and defining the path to final diagnosis, including conditions for an undetermined state (ND). - CNA breakpoint analysis: LGAs are called after filtering out segments smaller than 3Mb and merging adjacent large segments if the distance between segments is less than 3Mb. LGAs are called in adaptive mode, using two thresholds: stringent (suggesting a simple genome) and soft (suggesting a complex genome). The stringent threshold is applied even to noisy samples, while the soft threshold is applied in cases of low tumor volume. "Potentially missed LGAs" counts LGA breakpoints smaller than the soft threshold. - Detection of phenotypes of CDK12 mutation-related (CDK12mut) tandem duplication based on the number of intermediate gains between 1 and 10 Mb {Popova et al., 2016}. - Confirmation of CCNE1 amplification, ERBB2 (HER2) amplification, and amplification phenotype (called when three or more chromosomal arms have at least one amplification).

[0086] Step 4) Attributes of LGA score and HRD status (Figure 7A): - LGA score = LGA + bonus - penalty, Here, a "simple" genome has a bonus of 5, the detection of one amplification or CDK12mut phenotype results in a penalty of 5, and the detection of two or more of these features results in a penalty of 8. - Figure 7B shows the distribution of LGA scores in the training set. LGA scores below 17 and LGA scores of 20 or higher are considered definite (clear). LGA scores greater than 17 and below 20 are considered borderline. - For borderline LGA scores, several modification rules apply: the LGA score is shifted to 21 if HRD is proven (either the genome is classified as "simple" with a penalty of 0 and LGA_max ≥ 14, or the genome is classified as "complex" with LGA_max ≥ 20) (Figure 7). - An auxiliary value, LGA_max, used to further clarify the HRD status in borderline cases, corresponds to the maximum number of LGAs in the segmented profile of the tumor (obtained when the threshold for LGA calls is ignored). - If the LGA score is ≥20, HRD is called; if the LGA score is <20, no HRD is called. Borderline scores in samples of insufficient quality result in an ND diagnosis. - The final diagnosis will be reported along with a quality assessment and warning messages.

[0087] The random start and constant threshold of the segmentation algorithm, combined with the probabilistic profile optimization of the system, can lead to variations in the number of LGAs, which ultimately affects the final diagnosis. The complete workflow therefore includes 11 runs to fix intermediate parameters and obtain a preliminary diagnosis and error estimates of the mean LGA and LGA score, followed by 10 runs to narrow the confidence interval.

[0088] result: The Curie Institute, participating in the EHEI initiative, accessed 449 DNA samples extracted from FFPE tumor specimens for the PAOLA-1 test to validate the shallow HRDv2 test. The baseline characteristics of these 449 patients demonstrated that they were representative of the global cohort. All samples had been previously tested by MyChoice within the framework of the PAOLA-1 test, which allowed the inventors to create a correspondence table with the binary classifications obtained from GIS (HRD and non-HRD) and shallow HRDv2. Of the 394 samples for which conclusive results for both tests were obtained (394 / 449, 88%), the inventors observed 94% (369 / 394) overall agreement, 95% (196 / 206) positive agreement, and 92% (173 / 188) negative agreement. The percentage of non-conclusive tests was 11% (51 / 449) for GIS, while the failure rate for Shallow HRDv2 was 3% (15 / 449). Cohen's kappa (p=1.95×10) -148 ) showed substantial agreement between the two tests. The correlation between MyChoice(GIS) scores and Shallow HRDv2(LGA) scores was good (R 2 (=0.85), and cases that did not match were concentrated around the threshold for each test. Of the 15 GIS HRP / shallow HRDv2 HRD cases, 6 tumors had BRCA1 / 2 pathogenic variants. Of the 10 GIS HRD / shallow HRDv2 HRP cases, 3 tumors had BRCA1 / 2 pathogenic variants. Information regarding the relationship with loss of heterozygosity could not be obtained for the RCA1 / 2 variant.

[0089] To evaluate the actual improvement of v2 compared to version v1 of the Shallow HRD pipeline, the inventors evaluated the performance of Shallow HRD v1 in the same samples for the PAOLA-1 test. The performance of Shallow HRD v1 compared to GIS was acceptable, with 93% (309 / 333) overall agreement, 95% (169 / 177) positive agreement, 90% (140 / 156) negative agreement, and 4% (17 / 449) non-contributory results. Cohen's kappa at 0.56 (p=8.25×10⁶) -72 The results showed moderate agreement between the two tests, lower than that obtained with v2. Furthermore, Shallow HRDv1 calls the condition "borderline" when the number of LGAs detected is between 15 and 19 ("highly sensitive" and "specific" cutoffs). According to Shallow HRDv1, as many as 15% (66 / 449) of PAOLA-1 samples would be called "borderline," and a diagnosis of the HRD status for these samples cannot be obtained.

[0090] Because shallow HRDv2 correlated strongly with MyChoice, the inventors further evaluated whether HRD according to shallow HRDv2 predicted the clinical benefit of ola+bev maintenance in this PAOLA-1 cohort subset (Ray-Coquard, I. et al., Olaparib plus Bevacizumab as First-Line Maintenance in Ovarian Cancer. N Engl J Med 381, pp. 2416-2428 (2019); Ray-Coquard, IL et al., Final overall survival (OS) results from the phase III PAOLA-1 / ENGOT-ov25 trial evaluating maintenance olaparib (ola) plus bevacizumab (bev) in patients (pts) with newly diagnosed advanced ovarian cancer (AOC). Ann Oncol 33, pp. 808-869 (2022)). The median follow-up period was 63 months (interquartile range [IQR]: 28.5–62.3). According to Shallow HRDv2, the median PFS was 65.7 months in the HRD group treated with ola+bev (regardless of BRCA1 / 2 status) and 20.3 months in the bev-only group (HR 0.36 [95% CI, 0.24–0.53]), while according to MyChoice, the median PFS was 57.1 months in the HRD ola+bev group and 20.1 months in the HRD bev group (HR 0.40 [95% CI, 0.27–0.60], Figure 3A). Shallow HRDv2 also showed similar performance to MyChoice in terms of OS results. According to Shallow HRDv2 and MyChoice, the median OS was 75.2 months in the HRD group treated with ola+bev and 66.4 months in the bev-only group (HR 0.49 [95% CI, 0.31~0.80] and HR 0.58 [95% CI, 0.36~0.91] for Shallow HRDv2 and MyChoice, respectively, Figure 3B).

[0091] Importantly, shallow HRDv2 was also able to predict the benefits of PARPi in patients not supported by the MyChoice trial. Treatment with ola + bev did not achieve the median PFS in HRD patients following shallow HRDv2, whereas the median PFS was 17.4 months when treated with bev alone (HR: 0.13 [95% CI, 0.04~0.47], Figure 3C).

[0092] The primary advantage of detecting HRD based on genomic profiling is the identification of patients with tumors that may have homologous recombination deficiencies in the absence of BRCA1 / 2 pathogenic variants. Therefore, we evaluated the predictive value of shallow HRDv2 in this patient subgroup with BRCA1 / 2 wild-type (wt) AOC. According to shallow HRDv2, the median PFS in tumors with BRCA1 / 2wt, HRD was 40.8 months in the ola+bev group and 19.5 months in the bev group (HR: 0.45 [95% CI, 0.26~0.76]). According to MyChoice, the median PFS in tumors with BRCA1 / 2wt, HRD was 40.8 months in the ola+bev group and 17.6 months in the bev group (HR: 0.43 [95% CI, 0.24~0.77]) (Figure 4A). Notably, the inventors observed that among patients with BRCA1 / 2wt tumors treated with bev alone, those with HRD tumors following shallow HRDv2 tended to have a longer PFS than those with HRP. This suggests a prognostic value for HRD status, although the difference was not statistically significant (p=0.28).

[0093] In HRD BRCA1 / 2wt tumors, the median OS was not achieved in patients treated with ola+bev, regardless of the test used. In contrast, the median OS in patients treated with bev was 56.6 and 55.0 months, respectively, when HRD was defined by shallow HRDv2 or MyChoice (HRs for the comparison between ola+bev and bev were 0.63 [95% CI, 0.33~1.19] and 0.60 [95% CI, 0.31~1.18], respectively, when defined by shallow HRDv2 or MyChoice, Figure 4B). Conversely, patients with HRP BRCA1 / 2wt tumors treated with ola+bev tended to have a shorter median OS than patients treated with placebo+bev, but this was not statistically significant (38.2 vs. 42.1 months, respectively, log-rank p=0.55, Figure 3B).

[0094] Shallow HRDv2 demonstrated good analytical performance and was comparable to MyChoice in predicting the benefits of PARPi in the PAOLA-1 cohort. However, these conclusions were obtained from samples derived from patients included in the clinical trial and may differ in normal diagnostic procedures. Therefore, we evaluated the performance of shallow HRDv2 in an independent cohort consisting of 109 randomly selected, sequential FFPE AOC cases derived from our normal laboratory, and also applied these to MyChoice. In this prospective cohort, we confirmed high overall agreement of 91% (86 / 94), positive agreement of 92% (36 / 39), and negative agreement of 91% (50 / 55) between shallow HRDv2 and MyChoice, with fewer non-contributing results in shallow HRDv2 (5% vs. 12%). Cohen's kappa (p=7.49×10) -29 This supported moderate agreement between the two tests.

[0095] Consideration Assessment of HRD status is essential to balance the benefits and risks of PARPi maintenance in newly diagnosed OAC patients. Therefore, a localized, reliable, and inexpensive HRD test is needed. We hereby report the development of the ShallowHRDv2 test, whose clinical validation has shown its high agreement with MyChoice in the PAOLA-1 trial to predict the benefits of ola+bev in AOC patients. Key improvements in the v2 pipeline compared to the first version of the ShallowHRD test (Eeckhoutte, A. et al., ShallowHRD: detection of homologous recombination deficiency from shallow whole genome sequencing. Bioinformatics 36, pp. 3888-3889 (2020)) include correction of FFPE noise, a more precise diagnostic assessment based on tumor volume and sWGS noise levels, an improved binary HRD classification that reduces the number of undetermined cases, and auxiliary genomic features to refine the final conclusions. Due to these improvements, shallow HRDv2 reduces the number of non-conclusive outcomes by approximately 60 to 75% compared to MyChoice (3% vs. 11% in the PAOLA-1 cohort, and 5% vs. 12% in the regular cohort). More importantly, in shallow HRDv2, these patients who had HRD status but did not contribute to MyChoice greatly benefit from the ola+bev combination treatment, thus enabling more AOC patients to benefit from ola+bev. Since only 3.3% of analyses were not contributing, the robustness of shallow HRDv2 is similar to that of other genetic tests in routine clinical practice, such as BRCA1 / 2 tumor testing, which showed a 4.4% failure rate in the PAOLA-1 clinical trial (Callens, C. et al., Concordance Between Tumor and Germline BRCA Status in High-Grade Ovarian Carcinoma Patients in the Phase III PAOLA-1 / ENGOT-ov25 Trial. J Natl Cancer Inst 113, pp. 917-923 (2021)).

[0096] The inventors noted that six cases with BRCA1 / 2 pathogenic variants were incorrectly classified as HRP by MyChoice, while they were accurately classified as HRD by Shallow HRDv2.

[0097] The clinical values ​​of HRD testing are not limited to the issue of olaparib in AOC. In contrast to olaparib, which is not approved as a first-line treatment for HRP cases (either alone or in combination with bevacizumab), niraparib has received "all-comer" approval.

[0098] The use of inexpensive and robust HRD tests comparable to MyChoice, such as Shallow HRDv2, can therefore also help estimate the benefits of prescribing niraparib as a first-line treatment in BRCAwt AOC patients. Similarly, and moreover, HRD testing may be useful for other tumor types besides ovarian cancer (Coussy, F. and Bidard, FC, Expanding biomarkers for PARP inhibitors. Nat Cancer 3, pp. 1141-1143 (2022); Gruber, JJ et al., A phase II study of talazoparib monotherapy in patients with wild-type BRCA1 and BRCA2 with a mutation in other homologous recombination genes. Nat Cancer 3, pp. 1181-1191 (2022)), and our test was developed for all cancer samples.

[0099] Other teams also participated in EHEI, and new tests for detecting HRD were validated in tumor samples from the PAOLA-1 trial (Loverix, L. et al., Predictive value of the Leuven HRD test compared with Myriad myChoice PLUS on 468 ovarian cancer samples from the PAOLA-1 / ENGOT-ov25 trial (LBA 6). Gynecologic Oncology 166, pp. S51-S52 (2022); Willing, E.-M. et al., 2022-RA-873-ESGO Validation study of the 'NOGGO-GIS ASSAY' based on ovarian cancer samples from the first-line PAOLA-1 / ENGOT-ov25 phase-III trial. International Journal of Gynecologic Cancer 32, pp. A370-A370 (2022); Buisson, A. et al., 2022-RA-913-ESGO Clinical performance evaluation of a novel deep learning solution for homologous recombination deficiency detection. International Journal of Gynecologic Cancer 32, pp. A277~A278 (2022); Leman, R. et al., 2022-RA-935-ESGO Development of an academic genomic instability score for ovarian cancers. International Journal of Gynecologic Cancer 32, pp. A280~A280 (2022); Christinat, Y. et al., 2022-RA-567-ESGO The Geneva HRD test: clinical validation on 469 samples from the PAOLA-1 trial.International Journal of Gynecologic Cancer 32, pp. A238-A239 (2022). All of these studies report a reduction in non-contributing outcomes compared to MyChoice, and overall satisfactory clinical performance in predicting the PFS benefit of the ola+bev combination. However, most of these tests are based on NGS capture panels combined with homologous recombination repair gene sequencing. Therefore, laboratories wishing to perform these tests must modify their already validated methods for detecting homologous recombination repair gene changes, whereas this modification is unnecessary when using shallow HRDv2 based on pre-capture library sequencing. Detecting HRD status using mononucleotide polymorphism arrays still does not require homologous recombination repair gene sequencing, but it has the disadvantages of being more expensive and time-consuming than sWGS.

[0100] In conclusion, the shallow HRDv2 test is robust, cost-effective, easy to implement, clinically validated, and can be considered a reference test for detecting HRD, in conjunction with the MyChoice trial. Furthermore, having used a large dataset consisting of different tumor types to develop the shallow HRD pipeline, the inventors are confident that the shallow HRDv2 test will be applicable to predicting responses to PARPi in future clinical trials.

Claims

1. A method for diagnosing homologous recombination deficiency (HRD) in tumors, - A process to evaluate the number of large genomic alterations (LGA) in tumor samples by obtaining copy number variation (CNA) profiles using shallow coverage whole-genome sequencing (sWGS). - A process for determining the LGA score, which corresponds to the LGA number adjusted for the complexity of the tumor genome and the presence of markers selected from a group of markers consisting of (1) phenotypes associated with mutations in cyclin-dependent kinase 12 (CDK12) with multiple intermediate gains in the CNA profile, (2) amplification of cyclin E1 (CCNE1), (3) amplification of human epidermal growth factor receptor-2 (HER2), and (4) phenotypes of amplification at multiple sites. Methods that include...

2. The method according to claim 1, wherein the sample is selected from fresh tumor samples and preserved tumor samples, such as frozen tumor samples and formalin-fixed paraffin-embedded (FFPE) tumor samples.

3. The method according to claim 2, wherein the tumor sample is a formalin-fixed paraffin-embedded (FFPE) sample of the tumor, and the sWGS CNA profile is corrected by eliminating the FFPE noise profile.

4. A high LGA score is associated with HRD, a low LGA score is associated with HR pathway functionality (HRP), and there are borderline cases. - Complexity of the tumor genome, - LGA_max, which is the upper limit of the number of LGAs seen in the segmented copy number profile. - The presence of a marker selected from a group of markers consisting of (1) a phenotype associated with mutations in cyclin-dependent kinase 12 (CDK12) with multiple intermediate gains in the CNA profile, (2) amplification of cyclin E1 (CCNE1), (3) amplification of human epidermal growth factor receptor-2 (HER2), and (4) a phenotype of amplification at multiple sites. The method according to any one of claims 1 to 3, which is determined by taking into consideration the following.

5. The method according to any one of claims 1 to 4, wherein the tumor is a breast tumor.

6. The method according to any one of claims 1 to 4, wherein the tumor is an ovarian tumor.

7. A method for predicting the effectiveness of a treatment in a patient with cancer, wherein the treatment comprises PARPi and / or an alkylating agent, and the method comprises diagnosing HRD in a tumor sample according to the method of any one of claims 1 to 6.

8. PARPi and / or alkylating agents for use in a method for treating cancer in a patient diagnosed with having a tumor exhibiting HRD according to the method of any one of claims 1 to 6.

Images