Methods for diagnosing homologous recombination defects in human tumors

Abstract

The present invention relates to improved methods for diagnosing homologous recombination deficiency (HRD) in tumors. The method according to the invention comprises, inter alia, assessing the number of large scale genomic variations (LGAs) by obtaining a copy number variation (CNA) profile by low coverage whole genome sequencing (sWGS) in a tumor sample, and determining an LGA score corresponding to the number of LGAs adjusted according to the genomic complexity of the tumor and the presence of one or both markers, the marker is selected from the following groups of markers: (1) a cyclin dependent kinase 12 (CDK12) mutation related phenotype with multiple interstitial gains in a CNA spectrum, (2) cyclin E1 (CCNE1) amplification, (3) human epidermal growth factor receptor-2 (HER2) amplification and (4) a multifocal amplification phenotype.

Description

TECHNICAL FIELD

[0001] The present invention relates to a method for diagnosing homologous recombination deficiency in a tumor. BACKGROUND

[0002] BRCA1 / 2 Mutated tumor cells have a defective HR pathway function (homologous recombination deficiency or HRD) and thus rely on other DNA repair pathways to avoid cell death, such as non-homologous end joining (NHEJ), alternative end joining (AltEJ), single strand annealing (SSA) or base excision repair (BER), all of which involve the involvement of PARP1 and PARP2 enzymes (Groelly et al., Nat Rev Cancer , 2022). Thus, BRCA1 / 2 In defective tumor cells, inhibition of these PARP1 / 2 enzymes leads to cell death (Bryant et al., Nature 434:913-7, 2005; Farmer et al., Nature 434:917-21, 2005). PARP1 is the main breakthrough in BRCA1 / 2-deficient tumor treatment. Several clinical trials have shown that PARP inhibitors maintenance therapy improves the progression-free survival (PFS) of patients with advanced ovarian cancer (AOC) with HRD, regardless of whether it is combined with bevacizumab after platinum-based therapy. BRCA1 / 2 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). As in the PAOLA-1 trial, a significant PFS benefit was observed in patients with HRD-positive tumors, including those without BRCA mutations (BRCAmut), according to the Myriad MyChoice CDx Plus test (MG test). Thus, the olaparib (ola) + bevacizumab (bev) maintenance regimen has been approved in the USA / Europe / Japan for patients with BRCAmut or HRD-positive tumors. However, there is a large portion of invalid results in the MG test and it has only recently been centralized. BRCA1 / 2

[0003] Therefore, there is a need to develop new reliable and feasible decentralized HRD tests.

[0004] ​​The European HRD ENGOT project (EHEI) is a unique collaboration of European academic institutions aimed at providing a reliable HRD biomarker for AOC patients for the selection of patients who will benefit most from PARPi ± bevacizumab in first line treatment (Pujade-Lauraine et al., International Journal of Gynecologic Cancer 31 : A208-A208, 2021).

[0005] Recently, a new HRD test, called shallowHRD, based on shallow / low-coverage whole genome sequencing (sWGS), showed good performance on fresh-frozen samples (Eeckhoutte et al: Bioinformatics 36: 3888-3889, 2020). sWGS is a simple and cost-effective technology that can be applied to clinical sample analysis, including formalin-fixed paraffin-embedded (FFPE) samples. The bioinformatics analysis pipeline of shallowHRD is also quite simple and computationally inexpensive. However, this test method performs poorly on low-quality samples (FFPE, which is common in the clinic), and there are a large number of difficult-to-distinguish cases near the HRD cutoff, limiting its application in the clinic.

[0006] Therefore, although shallowHRD has great potential, it needs to be significantly improved to meet the specificity and selectivity required as an effective test for detecting HRD in samples such as FFPE samples. SUMMARY

[0007] The present invention is defined by the claims.

[0008] The inventors managed to significantly improve the performance of shallowHRD and make it suitable for routine clinical applications.

[0009] The inventors upgraded shallowHRD in particular in two fundamental aspects, (1) the detection of the number of LGAs, which is more reliable even in low-quality samples, and (2) the diagnostic decision rule is more detailed, thus providing a diagnosis for patients classified as having an intermediate number of LGAs, i.e. patients who cannot be directly classified as having HRD or homologous recombination proficiency (HRP).

[0010] The inventors added specific steps to the shallow HRD, as described above, which consist in assessing the number of large-scale genomic alterations (LGAs) from the copy number alteration (CNA) profile obtained by low-coverage whole genome sequencing (sWGS) to improve its accuracy. According to the shallow HRD, the CNA profile is directly used to calculate the number of LGAs, and the decision rule is based on the number of LGAs for which there is a large uncertainty zone around a threshold (cut-off). The main improvements brought by the inventors are a more accurate detection of LGAs (based on noise reduction and quality control), and a more reliable diagnosis of cases with a borderline number of LGAs (based on additional decision rules). These added steps specifically include noise reduction, quality classification of the CNA profile and selection of adaptive thresholds for LGA detection, determination of specific genomic markers (such as genomic complexity, amplifications, etc.), and a multi-step diagnostic process.

[0011] According to a first embodiment, the present application thus relates to a method for diagnosing homologous recombination deficiency in a tumor, said method comprising the steps of: - assessing the number of large-scale genomic alterations (LGAs) from the copy number alteration (CNA) profile obtained by low-coverage whole genome sequencing (sWGS) in a tumor sample, - determining a LGA score (also referred to as "HRD score") corresponding to the number of LGAs adjusted according to the genomic complexity of the tumor and the presence of markers selected from the group consisting of: (1) a cyclin-dependent kinase 12 (CDK12) mutation-related phenotype with multiple stromal gains in the CNA profile, (2) a cyclin E1 (CCNE1) amplification, (3) a human epidermal growth factor receptor-2 (HER2) amplification and (4) a multifocal amplification phenotype.

[0012] The method according to the present application is simple, easy to reproduce, and allows a correct diagnosis of HRD in poor quality samples, such as paraffin-embedded samples. The present application thus allows to identify patients who can benefit from a treatment comprising a PARP inhibitor and / or an alkylating agent.

[0013] According to another embodiment, the present application thus relates to a PARP inhibitor (PARPi) and / or an alkylating agent for use in a method of treating cancer in a patient, wherein said patient has been diagnosed as having a tumor presenting HRD according to the method of the present application.

[0014] The present application also relates to a method for predicting the efficacy of a treatment in a patient suffering from a cancer, wherein said treatment comprises a PARPi and / or an alkylating agent, and wherein said method comprises diagnosing HRD in a tumor sample as described herein. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1: Main steps of the shallow WGS approach and the shallowHRDv2 pipeline for efficient HRD diagnosis. Abbreviations: CNA stands for copy number alteration; FF stands for fresh frozen; FFPE stands for formaldehyde fixed and paraffin embedded; HRD stands for homologous recombination deficiency.

[0016] Figure 2 : Concept workflow of the shallowHRDv2 pipeline.

[0017] Figure 3 : Kaplan-Meier estimates: (A) PFS of PAOLA-1 patients according to homologous recombination status as determined by shallowHRDv2 or MyChoice and treatment arm. (B) OS of PAOLA-1 patients according to homologous recombination status as determined by shallowHRDv2 or MyChoice and treatment arm. (C) PFS of PAOLA-1 patients treatment arms according to homologous recombination status as determined by shallowHRDv2 and non-contributive results by MyChoice. HR stands for hazard ratio; PFS stands for progression free survival; OS stands for overall survival.

[0018] Figure 4 : Kaplan-Meier estimates: (A) PFS of PAOLA-1 patients according to homologous recombination status as determined by shallowHRDv2 or MyChoice and treatment arm. (B) OS of PAOLA-1 patients according to homologous recombination status as determined by shallowHRDv2 or MyChoice and treatment arm. (C) PFS of PAOLA-1 patients treatment arms according to homologous recombination status as determined by shallowHRDv2 and non-contributive results by MyChoice. HR stands for hazard ratio; PFS stands for progression free survival; OS stands for overall survival. BRCA1 / 2 (A) PFS; (B) OS of PAOLA-1 patients treatment arms of genotypically wild type tumors. HR stands for hazard ratio; PFS stands for progression free survival; OS stands for overall survival.

[0019] Figure 5: Two embodiments (A and B) of training sets and quality attributes used to adjust the decision rule. In embodiment A, the numbers represent the number of cases in the training set with the respective tumor content and noise characteristics.

[0020] Figure 6 : Embodiment (A) of the LGA score distribution in the training data set and adaptive model of score modification of the “cut-off” score (B).

[0021] Figure 7 : Another embodiment (A) of the LGA score distribution in the training data set, adaptive model of LGA score modification of the “cut-off” cases (B), and basic characteristics of HRD and non-HRD cases with cut-off LGA score. DETAILED DESCRIPTION

[0022] The present invention relates to an improved method for diagnosing Homologous Recombination Deficiency (HRD) in a tumor. The method according to the present invention aims at evaluating the number of large-scale genomic alterations by low-coverage whole genome sequencing (sWGS) in a tumor sample to adjust the obtained diagnosis. variant (LGA) to adjust the obtained diagnosis.

[0023] According to a first aspect, the present invention relates to a method for diagnosing Homologous Recombination Deficiency in a tumor, said method comprising the steps of: - evaluating the number of large-scale genomic alterations by low-coverage whole genome sequencing (sWGS) in a tumor sample to obtain a copy number alteration profile, - determining a LGA score corresponding to the number of LGAs adjusted according to the genomic complexity of the tumor and the presence of markers selected from the group of markers consisting of: (1) a Cyclin-Dependent Kinase 12 (CDK12) mutation associated phenotype, a multiple stromal gain in the CNA profile, (2) a Cyclin E1 (CCNE1) amplification, (3) a Human Epidermal Growth Factor Receptor-2 (HER2) amplification and (4) a multifocal amplification phenotype.

[0024] As used herein, the expression "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 caused by exogenous (such as ionizing radiation) or endogenous (such as cell replication byproducts). All organisms have evolved different strategies to cope with these damages. One of the most severe forms of DNA damage is a DSB. HR is the most accurate mechanism to repair DSBs as it utilizes an intact copy of DNA of a sister chromatid or homologous chromosome as a substrate to repair the break.

[0025] Cells (e.g., cancer cells) identified as having genomic DNA rearrangements (e.g., large-scale genomic alterations or LGAs) can be classified as having a higher likelihood of HR deficiency, i.e., one or more genes in the HR pathway are in a deficient state. As used herein, a "deficient state" of a gene refers to a deficiency in the sequence, structure, expression, and / or activity of the gene or its product as compared to a normal state. Examples include, but are not limited to, low or absent mRNA or protein expression, deleterious mutations, hypermethylation, reduced activity (e.g., enzymatic activity, and binding ability to another biomolecule), etc. As used herein, a deficient state of a pathway (e.g., the HR pathway) refers to at least one gene in the pathway (e.g., BRCA1) having a deficiency. Examples of highly deleterious mutations include frameshift mutations, stop codon mutations, and mutations that cause abnormal splicing of RNA. A deficient state of a gene in the HR pathway can result in a deficiency or reduction in HR activity in a cell (e.g., a cancer cell).

[0026] Examples of genes of 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 that are important for repairing double-stranded DNA breaks through the HR pathway. When the gene for any such protein, for example, is mutated or expressed at a lower amount, the change can lead to errors in DNA repair, which ultimately can cause cancer. Other proteins involved in the HR pathway include FANCA, FANCB, FANCC, FANCD2, FANCE, FANCG, FANCI, FANCL, FANCM, FAN1, SLX4 / FANCP, or ERCC1.

[0027] Accordingly, as used herein, the expression "HR pathway deficiency" or "HRD" refers to a condition in which one or more proteins of the HR pathway involved in DNA repair are defective or inactivated. In contrast, "HR pathway competent" or "HRP" refers to the absence of HRD.

[0028] Proteins involved in the HR pathway can encompass, 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.

[0029] As used herein, the term "inactivation", when referring to a gene, can refer to any type of defect in the gene. It includes, but is not limited to, germline mutations in the coding sequence, somatic mutations in the coding sequence, mutations in the promoter, and methylation of the promoter.

[0030] The method according to the present application can diagnose HRD in a tumor.

[0031] According to the application, the "tumor" can be any solid tumor or carcinoma. Preferably, the solid tumor or carcinoma is selected from the group consisting of breast cancer, ovarian cancer, colon cancer, lung cancer, prostate cancer, renal cancer, metastatic or invasive malignant melanoma, brain tumor, cancer of the fallopian tube, head and neck cancer, cancer of the peritoneum, liver cancer, cancer of the bladder, breast, colon, kidney, liver, lung, pancreas, stomach, esophagus, uterus, cervix, thyroid or skin, including squamous cell carcinoma. However, the application also relates to tumors of the hematopoietic system, such as leukemia, acute lymphoblastic leukemia, acute lymphoblastic leukemia, B-cell lymphoma, T-cell lymphoma, Hodgkin's lymphoma, non-Hodgkin's lymphoma, hairy cell lymphoma, Burkitt's lymphoma, acute and chronic myeloid leukemia and promyelocytic leukemia.

[0032] According to one particular embodiment, the tumor is selected from the group consisting of ovarian cancer, breast cancer, cancer of the fallopian tube, cancer of the peritoneum, lung cancer, pancreatic cancer, head and neck cancer, prostate cancer, gastric or esophageal cancer, uterine cancer, cervical cancer, renal cancer and bladder cancer.

[0033] According to one particular embodiment, the tumor is breast cancer or ovarian cancer, such as advanced ovarian cancer.

[0034] A "tumor" sample used in the context of the present application is typically obtained from a tumor biopsy. It can be, for example, a fresh or preserved sample, such as a frozen sample, or any tumor sample preserved by other means. The tumor sample can typically be in the form of a formalin-fixed paraffin-embedded (FFPE) sample. An FFPE sample refers to a tumor sample that is fixed with formaldehyde and then embedded in a paraffin block. FFPE samples are routinely prepared and used by the person skilled in the art.

[0035] As used herein, the term "patient" or "subject" denotes a mammal such as a rodent, a cat, a dog, a cow, a horse, a sheep, a pig or a primate. Preferably, the patient according to the present application is a human.

[0036] A "large genomic aberration" or "LGA" corresponds to a genomic rearrangement. An LGA refers to any somatic copy number transition (e.g., breakpoint) along the length of a chromosome, where 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 smaller 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), the copy number is stable. For example, if after filtering out regions shorter than 3 megabases, the copy number of the somatic cell is 1:1 for at least 10 megabases, then a breakpoint transitions to, for example, a region of at least 10 megabases with a copy number of 2:2, that is an LGA. Another way to define the same phenomenon is to define an LGA region as a region of the genome whose copy number is stable for 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) and which is bounded by a breakpoint (e.g., transition) where the copy number of another region also changes for 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 of 1:1, which is bounded on one side by a breakpoint transition to a region of at least 10 megabases with a copy number of 2:2, and which is bounded on the other side by a breakpoint transition to a region of at least 10 megabases with a copy number of 1:2, then it is two LGAs. Note that this is broader than allelic imbalance, as this copy number change is not considered an allelic imbalance (as the copy number ratio 1:1 and 2:2 are the same, e.g., there is no change in the copy number ratio). 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) describe LGAs and their use in determining HRD in detail.

[0037] According to the present invention, the number of LGAs is determined by low-coverage whole-genome sequencing. "Low-coverage whole-genome sequencing" or "sWGS" can detect the number of LGAs obtained by whole-genome sequencing (WGS) at low coverage, for example, a coverage of 0.3 or more, for example, at a ~1X coverage. Coverage refers to the average number of reads aligned to known reference bases. In the context of the present invention, sWGS is used to provide a "copy number variation profile" or "CNA" profile. Eeckhoutte et al.( Bioinformatics 36:3888-3889, 2020) fully discloses a method for determining the profile based on sWGS. This method (referred to as "ShallowHRD") consists of processing the sWGS data of tumor samples with the Control-FREEC software (Boeva et al. (2012), Bioinformatics ,28, 423–425). The input of this tool is "sample_name.bam_ratio.txt", which includes the CNA profile {x,g} 1, N , where x is the normalized read count in a sliding window, g is the genomic coordinate, and the spectral fragment is represented by S i , Z i representing the fragment median and size (in megabases, Mb).

[0038] ShallowHRD provides a CNA profile according to the following workflow: 1. Determine the CNAcut-off, and optimize the spectral fragments as follows: If Z i ≥ (Q1+Q3) / 2, then the fragment is defined as "large", where Q1, Q3 are the quartiles of the distribution of Z i (Z i >3Mb). Determine M as the first local minimum of the density of |(S i -S j ), where i, j are large fragments. CNAcut-off = min(max(0.025, M ), 0.45). If |(S i -S i+1 )|<CNAcut-off, then merge adjacent fragments; starting from the largest fragment.

[0039] 2. LGA is defined as a CNA break within a chromosome arm, and its adjacent fragments Z i , Z i +1 ≥ 10Mb, and are counted after removing fragments with a length <3Mb.

[0040] 3. The sample is labeled as "non-HRD" (LGA < 15), "borderline" (15 < LGA < 19) or "HRD" (LGA > 19).

[0041] 4. The sample quality is defined by M and cMAD cMAD = median(|(x-S x |)), where S x corresponds to the segment containing x before optimization: "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).

[0042] 5. If S c > 4 CNAcut-off, it is called "amplified", where CCNE1 is the segment containing the gene. c

[0043] In the context of the present invention, a diagnosis is provided based on the LGA score (also called "HRD score"), which corresponds to the number of LGA adjusted according to the genomic complexity of the tumor and the presence of (i.e. 1, 2, 3 or all 4) markers selected from the group of markers consisting of: (1) Cyclin-Dependent Kinase 12 (CDK12) mutation related phenotype, with multiple stromal gains in the CNA profile, (2) Cyclin E1 (CCNE1) amplification, (3) Human Epidermal Growth Factor Receptor-2 (HER2) amplification and (4) Multifocal amplification phenotype.

[0044] In the context of the present invention, the score is related to HRD: a high LGA score is related to HRD, a low LGA score is related to HRP. For example, according to the particular embodiment shown in the present example, a high score corresponds to a score of 20, or a score of 23 and above, and a low score corresponds to a score of 17 and below.

[0045] ​The estimation of the genomic complexity is performed by classifying the samples into two classes: "simple" and "complex", where "simple" genomes have two of the most common copy number levels, representing more than 70% of the genome. Other genomes are classified as "complex". According to another embodiment, the genomes can also be classified as "complex+", which is a subtype of "complex", containing more than three equally abundant copy number levels; all low-tumor content cases are labeled as "simple".

[0046] In the context of the present application, "simple" genomes are considered to be associated with HRD. Therefore, simple genomes will set a correction factor, which can be defined as "positive" or "bonus", which will adjust the diagnosis to be more prone to HRD.

[0047] On the contrary, "complex" genomes are considered to be associated with HRP. Therefore, "complex" genomes will set a correction factor, which can be defined as "negative" or "penalty", which will adjust the diagnosis to be more prone to HRP.

[0048] The method according to the present application comprises adjusting the number of LGAs based on the presence of 1, 2, 3 or 4 markers selected from the group consisting of: (1 ) Cyclin-Dependent Kinase 12 (CDK12) mutation-related phenotype with multiple stromal gains in CNA profile, (2) Cyclin E1 (CCNE1) amplification, (3) Human Epidermal Growth Factor Receptor-2 (HER2) amplification and (4) Multifocal amplification phenotype.

[0049] "Cyclin-Dependent Kinase 12" or "CDK12" is a protein kinase that is a key regulator of transcriptional elongation. It regulates the expression of genes involved in DNA repair and is essential for maintaining genomic stability. The amino acid sequence of human CDK12 is available at Uniprot database under reference number Q9NYV4. CDK12 is encoded by the gene having the nucleic acid sequence as indicated in Ensembl Genome Database under reference number ENSG00000167258. CDK12 The gene encoding CDK12 has the nucleic acid sequence as indicated in Ensembl Genome Database under reference number ENSG00000167258.

[0050] According to the present application, the number of LGAs is adjusted based on the presence of "CDK12 mutation-related phenotype with multiple stromal gains in CNA profile". This marker refers to the detection of a mutation-related (CDK12mut) tandem duplication phenotype (Popova et al., 2016) based on the number of stromal gains of 1 -10 Mb. CDK12 In the context of the present application, CDK12mut-related is detected based on the number of stromal gains of 1 -10 Mb, which are generally present compared to smaller stromal gains and / or stromal losses.

[0051] In the context of the present application, the presence of the phenotype associated with CDK12 mutations having multiple stromal gains in the CNA profile is considered as related to HRP. Therefore, the presence of this marker sets a correction coefficient which can be defined as a "negative value" or "penalty" ranging from HRP to HRD and thus adjusts the diagnosis towards HRP.

[0052] "Cyclin E1" or "CCNE1" is a regulatory protein which is essential for cell cycle regulation through its interaction with CDK2. The amino acid sequence of the CCNE1 protein is available in the Uniprot database under the reference number P24864, Homo sapiens CCNE1 The sequence of the gene is available in the Ensembl Genome database under the reference number ENSG00000105173.

[0053] According to the present application, the number of LGA is adjusted based on "CCNE1 amplification". This marker is defined as a high copy number gain (e.g. amplification) of the locus of the chromosome 19 carrying CCNE1 the gene CCNE1.

[0054] In the context of the present application, CCNE1 amplification is considered as related to the presence of HRP. Therefore, the presence of CCNE1 amplification sets a correction coefficient which can be defined as a "negative value" or "penalty" which will adjust the diagnosis towards HRP.

[0055] "Human epidermal growth factor receptor-2" or "HER2", also known as "receptor tyrosine-protein kinase erbB-2", "ERBB2", "cluster of differentiation 340" or "CD340", is a protein tyrosine kinase encoded by the by ERBB2 gene HER2. The amino acid sequence of the HER2 protein is available in the Uniprot database under the reference number P04626, Homo sapiens, and the sequence of the gene is shown under the reference number ENSG00000141736 in the Ensembl Genome database. ERBB2

[0056] According to the present application, the number of LGA is adjusted based on the detection of "HER2 amplification". This marker is defined as a high copy number gain (e.g. amplification) of the locus of the chromosome 17 carrying HER2 the gene HER2.

[0057] In the context of the present application, HER2 amplification is considered as related to the presence of HRP. Therefore, the presence of HER2 amplification sets a correction coefficient which can be defined as a "negative value" or "penalty" which will adjust the diagnosis towards HRP.

[0058] ​According to the application, the number of LGAs is adjusted based on the detection of a "multifocal amplification phenotype". This marker is defined as the presence of at least one high copy gain (amplification) on two or more chromosome arms in any given region.

[0059] In the context of the application, multifocal amplification is considered as being associated with the presence of HRP. Therefore, the presence of multifocal amplification sets a correction coefficient which can be defined as a "negative value" or "penalty" which will adjust the diagnosis to be more prone to HRP.

[0060] When the tumor sample is a FFPE sample, the method according to the application can advantageously comprise a step wherein the CNA profile from which the number of LGAs is calculated is corrected by eliminating (i.e. significantly reducing) false positive breakpoints associated with the noise profile (i.e. by determining a FFPE cumulative noise profile). The noise correction is advantageously associated with a segmentation optimization.

[0061] As mentioned above, this step is advantageously performed prior to determining the HRD score.

[0062] The FFPE cumulative noise profile is obtained from the segmented CNA profiles of ~100 almost normal genomes sequenced from FFPE tumor or normal samples. If each segment mean value is higher or lower than the global mean value, it is replaced by 1 or -1 respectively. Therefore, each genomic library is characterized by the sum of 1 / 0 / -1 over the ~100 profiles. When the sample is a FFPE tumor sample, the FFPE noise profile has clear peaks and holes at certain positions of the genome. A correlation between tumor CNA and FFPE cumulative profile >0.2 and a proportion of genomes falling into large segments (>20Mb) after segmentation <40% are characteristic of high FFPE noise.

[0063] Segmentation optimization is achieved by filtering small segments and merging segments with small differences in mean values (small differences meaning less than a threshold). FFPE denoising is performed under the same logic, adjacent segments are merged if their dynamics are locally correlated with the FFPE cumulative noise profile.

[0064] A threshold of inter-segment difference considered negligible is chosen according to the noise and tumor content estimation. After merging all adjacent segments with small differences in mean values, the profile is considered optimized. The CNA threshold, i.e. breakpoint calling, is fitted according to each sample quality class and is usually set to two times.

[0065] As mentioned above, the method according to the application can advantageously be used to solve borderline cases, i.e. cases for which the score does not correspond to HRP nor to HRD. For example, borderline cases correspond to LGA scores of 17 < LGA score < 20 or 23 when the score is calculated as illustrated in the following example. In these cases, the method according to the application further comprises solving by taking into account that the score is further adjusted by: genomic complexity of the tumor; the presence of a marker selected from the group of markers consisting of: (1) a cyclin-dependent kinase 12 (CDK12) mutation associated phenotype with multiple stromal gains in CNA profile, (2) cyclin E1 (CCNE1) amplification, (3) human epidermal growth factor receptor-2 (HER2) amplification, and (4) a multifocal amplification phenotype; and a boost cumulative LGA index (LGA_boost) or LGA_max, corresponding to the upper limit of LGA counts in the optimized segmented genomic profile, not limited by the threshold of LGA calling.

[0066] genomic complexity and the presence of a marker selected from the group of markers consisting of: (1) a cyclin-dependent kinase 12 (CDK12) mutation associated phenotype with multiple stromal gains in CNA profile, (2) cyclin E1 (CCNE1) amplification, (3) human epidermal growth factor receptor-2 (HER2) amplification, and (4) a multifocal amplification phenotype. Figure 6 or Figure 7 Methods of further adjusting critical case scores based on these elements are disclosed.

[0067] When the diagnosis is adjusted with a boost cumulative LGA index (LGA_boost), the index is calculated by the following formula: LGA_boost = LGA_chr_arm + LGA_at_telomere + LGA_20Mb + LGA_baseline + LGA_baseline_12 where LGA_chr_arm is the number of chromosome arms with LGA; LGA_at_telomere is the number of chromosome arms with LGA at telomere ends; LGA_20Mb is the number of LGAs with copy number (CN) breakpoints where the two genomic segments are greater than 19 Mb apart; LGA_baseline is the number of LGAs involving the most common CN layer; LGA_baseline_12 is the number of LGAs detected between the two most common CN layers.

[0068] Figure 6 The way LGA_boost is used to adjust the diagnosis is explained in the Background section.

[0069] When adjusting the diagnosis with the LGA_max index, the index is determined as the upper limit of the LGA count in the optimized segmented genome profile, not limited by the threshold of the LGA call. LGA_max = LGA + possibly missed LGAs, where the breakpoint amplitude of the possibly missed LGAs is less than the threshold of the LGA call.

[0070] The way of adjusting the diagnosis using LGA_max is explained in detail in the experimental section of the present application and Figure 7 in the following.

[0071] According to a preferred embodiment, the method according to the present application comprises the following steps ( Figure 2 ): 1) Obtain a determined copy number variation profile by assessing the number of LGAs of sWGS in the tumor sample (normalized by the sWGS read count profile in the tumor sample); 2) Denoising and segmentation optimization; 3) Genomic profile characterization: - Estimation of genomic complexity.

[0072] - Overall signal quality attribute (four categories: "good", "fair", "poor" and "very poor"), which is based on tumor content and noise, and defines the path of the final diagnosis, including the case of undetermined status (ND).

[0073] - CNA breakpoint analysis: based on the number of interstitial gains of 1-10 Mb, detect large-scale genomic alterations (LGAs), detect CDK12 mutant-related (CDK12mut) tandem duplication phenotypes (Popova et al, Cancer Res 2016; PMID 26787835); - Check CCNE1 amplification, ERBB2 amplification and multifocal amplification phenotypes.

[0074] 4) LGA score and HRD status attribution and final diagnosis (i.e. HRD or HRP): LGA_score = LGA + bonus - penalty The rules of the final diagnosis are based on the LGA score, the sample quality attribution and the CNA profile features.

[0075] Because it is possible to predict whether a particular patient has a cancer associated with HRD, it is also possible to select an appropriate treatment for said patient.

[0076] As described herein, a patient having cancer cells identified as having genomic DNA rearrangements (e.g., LGA) can be classified as likely to respond to a particular cancer treatment regimen. For example, a patient having cancer cells with a genome comprising genomic DNA rearrangements can be classified as likely to respond to a cancer treatment regimen comprising the use of a DNA damaging agent, a synthetic lethal agent (e.g., a PARP inhibitor), radiation, or a combination thereof.

[0077] Accordingly, another aspect of the application relates to a method of predicting the efficacy of a treatment in a patient having a cancer, wherein the treatment comprises a PARPi and / or an alkylating agent, and wherein the method comprises diagnosing HRD in a tumor sample as described above.

[0078] The application also relates to a PARPi and / or an alkylating agent for use in a method of treating a cancer in a patient, wherein the patient is diagnosed as having a tumor presenting a HRD according to the method of the application.

[0079] As used herein, the term "PARP inhibitor" or "PARPi" has its general meaning in the art. It refers to a compound capable of inhibiting the activity of the enzyme poly ADP ribose polymerase (PARP), a protein that is essential for the repair of single-strand breaks ("nicks") in DNA. If such nicks persist unrepaired until DNA replication (which must occur before cell division), replication itself will cause double-strand breaks to form. Drugs that inhibit PARP allow multiple double-strand breaks to form in this way, and in tumors carrying mutations in BRCA1, BRCA2, or PALB2, these double-strand breaks cannot be repaired efficiently, leading to cell death.

[0080] Typically, the PARP inhibitor according to the application can be selected from the group consisting of iniparib, olaparib, rucaparib, CEP972, MK4827, BMN-673 and 3-aminobenzamide.

[0081] As used herein, the term "alkylating agent" or "alkylating antineoplastic agent" has its general meaning in the art. It refers to a compound that links an alkyl group to DNA. Typically, the alkylating agent according to the application can be selected from platinum complexes (such as cisplatin, carboplatin and oxaliplatin), nitrogen mustards, chlorambucil, melphalan, cyclophosphamide, ifosfamide, estramustine, carmustine, lomustine, fotemustine, streptozocin, busulfan, pipobroman, procarbazine, dacarbazine, thiotepa and temozolomide.

[0082] The method according to the application can advantageously be implemented by a computer using a computer program tailored for reproducing the steps described above.

[0083] In another aspect, the present application provides a computer program product comprising a computer readable medium which, when executed on a computer, provides instructions for assessing the number of LGA in a tumor sample, correcting the sample noise spectrum, providing a HRD diagnosis according to the method of the present application.

[0084] Example Clinical confidence in the detection of tumor homologous recombination deficiency in a shallow whole genome sequencing approach Methods: Patients and tumor samples The first set consisted of FFPE-derived DNA from 449 AOC samples of the PAOLA-1 / ENGOT-ov25 trial in the EHEI project. All patients signed a written informed consent. The second set consisted of 109 consecutive FFPE AOC samples (8-20 5 pm sections of tumor area) sent to the Myriad Genetics central laboratory (Salt Lake City, UT, USA) as part of routine operations from March 2021 to January 2022. At the same time, the same FFPE samples were subjected to shallowHRDv2 in the genetics laboratory of the Institut Curie.

[0085] Statistical analysis Kaplan-Meier method was used to assess progression-free survival (PFS) and overall survival (OS), and stratified log-rank test was used to assess the difference between the ola+bev and bev groups. Hazard ratios (HR) and their 95% confidence intervals (95% CI) were calculated using stratified Cox proportional-hazards models. All statistical analyses were performed using Graphpad Prism software version 9.1.0.

[0086] Shallow WGS workflow 100 ng of FFPE DNA was used as starting sample. Mechanical DNA shearing was performed using a Covaris (model ME220 Focused-ultrasonicator) with 50 μΙ of DNA sample. We followed the supplier recommendations for the Agilent kit (SureSelect XT HS and XT Low Input Library Preparation, G9703A). Different steps included ligation, amplification and purification with AMpure XP magnetic beads (Beckman Coulter, code: A63882). Dose was performed with Thermo Fisher Scientific Qubit® dsDNA HS Assay Kit (code: Q32854) or Qubit® dsDNA BR Assay Kit (code: Q32853). After quality and quantity control with Agilent TapeStation and D1000 ScreenTape, we prepared 4 nM or 1.8 nM of library pool with NextSeq 550 S or NovaSeq 6000 sequencing system (Illumina Inc, San Diego, CA, USA), respectively.

[0087] ShallowHRDv2 bioinformatics pipeline After DNA extraction and whole genome sequencing at low coverage (~1X), we obtained normalized and corrected for GC content read count profiles (pool size ~50kb) by 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 ) ShallowHRDv2 bioinformatics pipeline includes copy number variation (CNA) profile analysis, providing HRD diagnosis, sample quality attribution and comprehensive quantitative and graphical outputs for manual control. Main steps of the pipeline, decision rules and summary of the diagnosis are described in Figure 1

[0088] Main steps of the CNA processing are the following Figure 2 (1) Three-way sample quality attribution: CNA profile classification according to tumor content (four categories), intrinsic sWGS noise (three categories) and FFPE noise (four categories), finally integrating signal vs. noise classification into four categories: "good", "fair", "poor" and "very poor" Figure 5B ​​).

[0089] (2) Noise reduction and breakpoint optimization in CNA profiles: filtering of small fragments and assembly (from ~100 normal profiles of FFPE samples, Figure 1 ) fragments with minor differences or local correlation with FFPE noise profiles, and thresholding of breakpoint calling for each mass category.

[0090] (3) Broad CNA profile features: - genomic complexity, where "simple" genomes have two most frequent copy number (CN) levels, accounting for more than 70% of the genome, otherwise, the genome is classified as "complex"; - a set of binary attributes, such as CCNE1 amplification, ERBB2 amplification, focal amplification phenotype (when more than two chromosome arms carry at least one amplification), CDK12 mutation-associated tandem duplication phenotype (when multiple interstitial 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, 1882-1891 (2016)); - a set of parameters characterizing breakpoints, including total number and number of large genomic alterations (LGAs), which contribute greatly to the HRD diagnosis; LGAs are defined as CN breaks between genomic fragments exceeding 9 Mb (fragment size rounded to the nearest integer).

[0091] (4) Multi-step HRD diagnosis is based on: - LGA score, which is basically the number of LGAs modified by PENALTY and BONUS, where PENALTY is defined by binary attributes and subtracted from the number of LGAs (PENALTY is set to 0, 5 or 8 if none, one or more binary attributes are true, respectively), and BONUS is defined by genomic complexity and added to the number of LGAs (BONUS is set to 5 for "simple" genomes, and 0 otherwise); - two thresholds for explicit HRD diagnosis, namely 17 and 20, where LGA score < 17 for non-HRD, and LGA score > 20 for HRD; - LGA score correction to resolve borderline cases (17 < LGA_score < 20). In short, if there is HRD evidence (genomic classification "simple" with penalty = 0 and LGA_max >= 14 or genomic classification "complex" and LGA_max >= 20), then the LGA score becomes 21 ( Figure 7 ).

[0092] decision rules The decision rule is multi-step, depends on sample quality attributes and includes the selection of thresholds for LGA calling. Two thresholds are used to call LGA at CNA breakpoints: strict (indicating simple genomes) and lenient (indicating complex genomes), which are applied in a conservative manner to high quality cases (LGA score based on LGA number, with lenient / strict threshold for non-HRD / HRD definite diagnosis) and fixed to strict / lenient strict thresholds for noisy / low tumor content samples, respectively. The simplified decision rule for poor quality samples consists in giving a diagnosis only for definite non-HRD cases with a small number of total breakpoints. Most poor quality cases are discarded. The robust decision rule for poor quality samples consists in giving a diagnosis only for definite cases, and in leaving an undetermined (ND) diagnosis for borderline cases. Additional rules for LGA_score correction procedure for good / average quality borderline cases help to resolve the diagnosis and reduce the number of uncertain cases.

[0093] comprehensive output (report) The final diagnosis is reported together with quality assessment and warning messages. Quantitative outputs provide complete information on the decisive genomic biomarker, LGA, LGA score and HRD diagnosis. The outputs include the profile with detected LGAs and the error profile to visually control the noise reduction quality and segmentation.

[0094] The cyclic binary segmentation of the CNA profile has a stochastic nature and can lead to differences in LGA number between runs, which can affect the diagnosis if close to the threshold. Therefore, the LGA number is reported as the mean value estimated from 21 segmentation / optimization runs with the standard error.

[0095] Detailed workflow of shallowHRDv2 Step 1) CNA profile segmentation with cyclic binary segmentation, classification of CNA profiles according to tumor content (four categories), intrinsic sWGS noise (three categories) and FFPE noise (four categories). Combining these attributes, an overall sample quality attribute ("good", "average", "poor" or "very poor") can be derived ( Figure 5B ), which is used to select the decisive procedure and reporting.

[0096] - The spectrum is characterized by the number of breakpoints; the variance of the CNA spectrum, the intra-segment variance, the inter-segment variance; the correlation with the FFPE noise; the percentage of the genome belonging to >20Mb segments after segmentation.

[0097] - ranking quality Characterized by the raw variance (intra-segment variance).

[0098] - tumor content Characterized by the median variance of large segments.

[0099] - FFPE noise Characterized by the number of breakpoints, the correlation with the FFPE cumulative spectrum, the variance of the error spectrum and the proportion of large segments after initial spectrum segmentation. The FFPE noise spectrum is obtained from the segmented CNA spectrum of ~100 almost normal genomes sequenced from FFPE samples. If the mean of each segment is higher or lower than the global mean, it is replaced by 1 or -1, respectively. Therefore, each genomic library is characterized by the sum of 1 / 0 / -1 of ~100 profiles. A correlation between tumor CNA and FFPE cumulative spectrum >0.2 and a proportion of the genome in large segments (>20Mb) after segmentation <40% are characteristic of high FFPE noise.

[0100] Step 2) Noise correction and segmentation optimization: filtering small segments and merging segments with small median difference or locally correlated with the FFPE noise spectrum. According to the noise and tumor content categories, a threshold is chosen whose inter-segment difference is considered negligible (no breakpoint). If the median difference is less than the threshold, adjacent segments are merged. If the breakpoint is followed in the FFPE covariate spectrum, even if the difference exceeds the threshold, the breakpoint is eliminated.

[0101] Step 3) Genomic profile characterization: - Estimation of genomic complexity (two categories: "simple" and "complex", where "simple" genomes have two most common copy number levels, representing more than 70% of the genome. Otherwise, the genome is classified as "complex". All low tumor content cases are labeled "simple".

[0102] - Overall spectrum quality attribution (four categories: "good", "fair", "poor" and "very poor"), based on tumor content and noise, characterizes the signal with respect to noise and defines the path to the final diagnosis, including the case of undetermined status (ND).

[0103] - CNA breakpoint analysis: if inter-fragment distance is < 3 Mb, filter out fragments < 3 Mb and merge adjacent large fragments before LGA calling; call LGA in an adaptive way, i.e. using two thresholds, strict (indicative of simple genomes) and relaxed (indicative of complex genomes); apply the strict threshold also in noisy samples, while apply the relaxed threshold in case of low tumor content. "LGA missed" counts LGA breakpoints with amplitude < relaxed threshold.

[0104] - Stromal gain number detection based on 1-10 Mb inter-stromal gain CDK12 Mutated (CDK12mut) tandem duplication phenotype {Popova et al., 2016}; - Check CCNE1 Amplification, ERBB2 HER2) amplification and amplified phenotype (when more than two chromosome arms carry at least one amplification).

[0105] Step 4) LGA score and HRD status attribution Figure 7 A): - LGA_score = LGA + bonus - penalty, where for "simple" genomes, bonus = 5; if one amplification or CDK12mut phenotype is detected, penalty = 5; if two or more of these features are detected, penalty = 8.

[0106] - The distribution of LGA scores in the training set is shown in Figure 7 B. LGA scores < 17 and LGA scores > 20 are considered clear-cut; LGA scores > 17 and < 20 are considered borderline.

[0107] - For borderline LGA scores, several correction rules are applied: if there is evidence of HRD (genome classified as "simple" with penalty = 0 and LGA_max > 14, or genome classified as "complex" and LGA_max > 20), the LGA score becomes 21 Figure 7 ).

[0108] - The auxiliary value LGA_max used to further clarify the HRD status of borderline cases corresponds to the maximum number of LGAs in the tumor segmentation profile (obtained if the threshold for LGA calling is ignored).

[0109] - If LGA score is > 20, HRD, if LGA score < 20, non-HRD. Borderline scores for POOR quality samples lead to a ND diagnosis.

[0110] - Final diagnosis is reported together with quality assessment and warning messages.

[0111] Random start of the segmentation algorithm and random spectrum optimization by a fixed threshold system can lead to variations in the number of LGAs, which ultimately affect the final diagnosis. Therefore, the complete workflow includes 11 runs to fix intermediate parameters and obtain a preliminary diagnosis and error estimate of the average LGA and LGA score, followed by 10 runs to narrow the confidence interval.

[0112] Results: Institut Curie joined the EHEI project and obtained DNA from 449 FFPE tumor samples extracted in the PAOLA-1 trial to validate the shallowHRDv2 test. The baseline characteristics of the 449 patients showed that they were globally representative. All samples had previously been tested by MyChoice within the PAOLA-1 trial framework, allowing us to establish a concordance table according to the binary classification results produced by GIS (HRD vs non-HRD) and shallowHRDv2. Among the 394 samples, the conclusive results of the two tests (394 / 449; 88%), we observed an overall concordance of 94% (369 / 394), a positive concordance of 95% (196 / 206) and a negative concordance of 92% (173 / 188). The percentage of tests with indeterminate results for GIS was 11% (51 / 449) while the failure rate for shallowHRDv2 was 3% (15 / 449). Cohen’s Kappa was 0.73 (p=1.95x10 -148 ), indicating that the two tests were substantially concordant. The correlation between MyChoice (GIS) scores and shallowHRDv2 (LGA) scores was good (R 2 =0.85), with discordant cases concentrated around the threshold of each test. Among the 15 GISHRP / shallowHRDv2 HRD cases, 6 tumors carried BRCA1 / 2 pathogenic variants. Among the 10 GISHRD / shallowHRDv2 HRP cases, 3 tumors carried BRCA1 / 2 pathogenic variants. BRCA1 / 2 The association of variants with loss of heterozygosity was not clear.

[0113] To assess the actual improvement of v2 compared to shallowHRD workflow version v1, we evaluated the performance of shallowHRDv1 on the same sample from the PAOLA-1 trial. Compared to GIS, shallowHRDv1 performed acceptablely, with an overall concordance of 93% (309 / 333), a positive concordance of 95% (169 / 177), a negative concordance of 90% (140 / 156), and 4% (17 / 449) of no-contribution results. Cohen's Kappa was 0.56 (p=8.25×10⁻⁶). -72 The results indicate that the two tests have moderate consistency, lower than those obtained using v2. Furthermore, shallowHRDv1 defines a “critical” state when the number of detected LGAs is 15–19 (“sensitive” and “specific” thresholds). According to shallowHRDv1, up to 15% (66 / 449) of PAOLA-1 samples are considered “critical” for which HRD status diagnosis is not possible.

[0114] Since shallowHRDv2 is highly correlated with MyChoice, we further evaluated whether HRD based on shallowHRDv2 could predict the clinical benefit of olaparib plus bevacizumab maintenance in a subgroup of the PAOLA-1 cohort (Ray-Coquard, I., et al. Olaparib plus Bevacizumab as First-Line Maintenance in Ovarian Cancer). N Engl J Med 381, 2416-2428 (2019); Ray-Coquard, IL, et al. Final overallsurvival (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, S808-S869 (2022)). The median follow-up time was 63 months (interquartile range [IQR]: 28.5–62.3). According to shallowHRDv2, the HRD group treated with olaxide plus bevacizumab (regardless of...) BRCA1 / 2The median progression-free survival (PFS) was 65.7 months in the HRD group (with or without disease status) and 20.3 months in the bevacizumab group (HR 0.36 [95% CI, 0.24–0.53]). According to MyChoice, the median PFS was 57.1 months in the HRD olaxaviva + bevacizumab group and 20.1 months in the HRD bevacizumab group (HR 0.40 [95% CI, 0.27–0.60]). Figure 3 A). Regarding OS results, ShallowHRDv2 also showed similar performance to MyChoice. According to shallowHRDv2 and MyChoice, the median OS was 75.2 months in the HRD group treated with olaxide plus bevacizumab and 66.4 months in the group treated with bevacizumab (shallowHRDv2 and MyChoice HR 0.49 [95% CI, 0.31-0.80] and 0.58 [95% CI, 0.36-0.91], respectively). Figure 3 B).

[0115] Importantly, shallowHRDv2 also predicted PARP1 benefit in patients for whom the MyChoice test did not contribute. According to shallowHRDv2, HRD patients treated with olaxostat plus bevacizumab did not reach the median PFS of 17.4 months compared with bevacizumab monotherapy (HR: 0.13 [95% CI, 0.04–0.47]). Figure 3 C).

[0116] The main advantage of detecting HRD based on genome mapping is that it identifies deficiencies in... BRCA1 / 2 In patients with pathogenic tumors, the tumor may lack homologous recombination. Therefore, we evaluated shallowHRDv2 in... BRCA1 / 2 Predictive value in wild-type (wt) AOC patient subgroups. According to shallowHRDv2, in BRCA1 / 2wt, HRD tumors, the median PFS with olarenaline plus bevacizumab was 40.8 months, and the median PFS with bevacizumab was 19.5 months (HR: 0.45 [95% CI, 0.26–0.76]). According to MyChoice, in BRCA1 / 2 In wt. HRD tumors, the median PFS with olaretinol plus bevacizumab was 40.8 months, and the median PFS with bevacizumab was 17.6 months (HR: 0.43 [95% CI, 0.24–0.77]). Figure 4 A). Notably, we observed that among BRCA1 / 2wt tumor patients receiving bevacizumab alone, HRD tumor patients with shallowHRDv2 tended to have a longer PFS than HRP patients, suggesting the prognostic value of HRD status, but the difference was not statistically significant (p=0.28).

[0117] In the HRD BRCA1 / 2 Patients treated with olaparib + bevacizumab did not reach median OS in wt tumors, while patients treated with bevacizumab had a median OS of 56.6 months and 55.0 months when HRD was defined by shallowHRDv2 or MyChoice, respectively (HR: 0.63 [95% CI, 0.33-1.19] and 0.60 [95% CI, 0.31-1.18] for the comparison between olaparib + bevacizumab and bevacizumab, respectively, by shallowHRDv2 or MyChoice; Figure 4 B). In contrast, according to shallowHRDv2, HRP BRCA1 / 2wt Patients with tumors tended to have a shorter median OS than patients receiving placebo + bevacizumab, although not significantly (38.2 vs. 42.1 months, respectively; log-rank test p=0.55 Figure 3 B).

[0118] ShallowHRDv2 showed good analytical performance and was equivalent to MyChoice in predicting PARPi benefit in the PAOLA-1 cohort. However, these conclusions were drawn based on a sample of patients in a clinical trial, which can be different in routine diagnostic practice. Therefore, we evaluated the performance of shallowHRDv2 in an independent cohort of 109 unselected consecutive FFPE AOC cases from our routine laboratory, which were also submitted to MyChoice. We confirmed that in this prospective cohort, the overall agreement between shallowHRDv2 and MyChoice was up to 91% (86 / 94), with a positive agreement of 92% (36 / 39) and a negative agreement of 91% (50 / 55), with a lower non-contributory result for shallowHRDv2 (5% vs 12%). Cohen’s Kappa was 0.69 (p=7.49 x 10 -29 ), confirming a moderate agreement between both tests.

[0119] Discussion For newly diagnosed OAC patients, HRD status must be assessed to balance the benefit and risk of PARPi maintenance. Therefore, there is a need for a local, reliable, and cost-effective HRD test. We here report the development of the shallowHRDv2 test and its clinical validation showing high concordance with MyChoice in predicting the benefit of olaparib + bevacizumab in the PAOLA-1 trial. 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, 3888-3889 (2020)), the main improvements of the v2 procedure (Eeckhoutte, A., et al. ShallowHRDv2: a robust and cost-effective HRD test for routine clinical practice. Submitted) include FFPE noise correction, critical diagnostic assessment based on tumor content and sWGS noise level, improved binary HRD classification, reduction of the number of unresolved cases, and auxiliary genomic features to refine the final conclusion. Thanks to these improvements, shallowHRDv2 reduces the number of non-conclusive results by about 60-75% compared to MyChoice (3% vs 11% in the PAOLA-1 cohort, 5% vs 12% in the routine cohort). More importantly, these patients with a shallowHRDv2 HRD status but no contribution from MyChoice significantly benefit from the combination of olaparib + bevacizumab, thus allowing more AOC patients to benefit from this combination. Only 3.3% of the analyses were non-contributory, and the robustness of shallowHRDv2 is similar to other genetic tests in routine clinical practice, such as BRCA1 / 2 tumor tests, 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, 917-923 (2021)).

[0120] We noticed that 6 cases carrying BRCA1 / 2 pathogenic variants were misclassified as HRP by MyChoice, while correctly classified as HRD with shallowHRDv2.

[0121] The clinical value of HRD testing can not be limited to olaparib use issues in AOC. In contrast to olaparib, which has not received a first-line treatment authorization (alone or in combination with bevacizumab) for HRP cases, niraparib has received approval for “all patients”.

[0122] Thus, using an inexpensive and robust HRD test comparable to MyChoice (such as shallowHRDv2) can also help estimate the benefit of prescribing niraparib in first-line treatment for BRCAwt AOC patients. Similarly and on a global scale, HRD tests can also be useful for other tumor types than ovarian cancer (Coussy, F. & Bidard, F. C. Expanding biomarkers for PARP inhibitors. Nat Cancer 3, 1141-1143 (2022); Gruber, J. J., 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, 1181-1191 (2022)) and our test is performed in pan-cancer samples.

[0123] Other teams are also involved in EHEI and new methods to test HRD have been validated on 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, 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 Cancer32, 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, 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, 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, A238-A239 (2022). All these reports claim that non-contributory results were reduced and that the clinical performance was overall satisfactory in predicting PFS benefit with olaparib + bevacizumab combination compared to MyChoice. However, most of these tests are based on NGS capture panel combined with sequencing of homologous recombination repair genes. Therefore, laboratories wishing to implement these tests will be forced to change their validated method for detection of homologous recombination repair gene variants, while this is not necessary with shallowHRDv2 based on pre-capture library sequencing. Detection of HRD status using single nucleotide polymorphism array is still independent of sequencing of homologous recombination repair genes, but the drawback is that it is more expensive and more time consuming than sWGS.

[0124] In conclusion, the shallowHRDv2 detection method is reliable, cost-effective, easy to implement, clinically validated and can be used as a reference method for detecting HRD together with the MyChoice test. Moreover, since a large dataset of multiple tumor types was used to develop the shallowHRD process, we believe that the shallowHRDv2 detection can also be applicable in the future to predict response to PARPi in clinical trials.

Claims

1. A method for diagnosing homologous recombination deficiency (HRD) in tumors, the method comprising the following steps: - Assess the quantity of large-scale genomic variants (LGAs) by obtaining copy number variant (CNA) profiles in low-coverage whole-genome sequencing (sWGS) of tumor samples. - Determine the LGA score, which corresponds to the number of LGAs adjusted for tumor genomic complexity and the presence of biomarkers selected from the group consisting of: (1) cyclin-dependent kinase 12 (CDK12) mutation-associated phenotype with multiple mesenchymal gains in the CNA profile; (2) cyclin E1 (CCNE1) amplification; (3) human epidermal growth factor receptor-2 (HER2) amplification; and (4) multifocal amplification phenotype.

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

4. The method according to any one of claims 1-3, wherein a high LGA score is associated with HRD and a low LGA score is associated with HR pathway proficiency (HRP), and wherein the critical case is addressed by considering the following factors: - The genomic complexity of the tumor; - LGA_max, the upper limit of the number of LGAs in the segmented copy number spectrum; - The presence of markers selected from the following group of markers: (1) cyclin-dependent kinase 12 (CDK12) mutation-associated phenotype with multiple interstitial gains in the CNA spectrum, (2) cyclin E1 (CCNE1) amplification, (3) human epidermal growth factor receptor-2 (HER2) amplification and (4) multifocal amplification phenotype.

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

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

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

8. PARPi and / or alkylating agents, in a method of treating a patient with cancer, wherein the patient has been diagnosed with a tumor presenting with HRD according to any one of claims 1-6.

Images