WO2023085932A1 - Prediction of response following folfirinox treatment in cancer patients - Google Patents

Abstract

The invention relates to a method of typing a tissue sample of an individual suffering from gastrointestinal cancer, comprising (i) providing a non-cancerous tissue sample from the individual, wherein said tissue sample comprises genomic nucleic acids; (ii) determining a copy number of at least three genomic marker regions in said tissue sample; and(iii) comparing the copy number of the at least three genomic marker regions to a baseline or reference to thereby type the sample for a response following FOLFIRINOX therapy. The invention further relates to methods of treating an individual with gastrointestinal cancer using this method of typing.

Description

Title: Prediction of response following FOLFIRINOX treatment in cancer patients

FIELD: The invention relates to methods for typing of cancer, especially pancreatic cancer. The invention is directed to the copy number variation of genomic regions to predict response following cancer therapy.

1 INTRODUCTION

Pancreatic ductal adenocarcinoma (PDAC) is the twelfth most common cancer worldwide with a median survival of less than 18 months, it is one of the four most lethal forms of cancer worldwide(Ferlay et al., 2018. Eur J Cancer 103: 356-387). Experts estimated that PDAC will become the second leading cause of cancer death by the year 2030 (Rahib et al., 2014. Cancer Res 74: 2913-2921). A major factor contributing to its lethality is that PDAC is rarely diagnosed during the early stage due to a lack of effective screening methods (Lau & Cheung, 2017. World J Gastrointest Oncol 9: 281-292). While surgical resection is considered as the only potentially curative treatment, this is only possible in about 20% of the patients. Therefore, in the past, the main focus of treatment for patients with PDAC has been palliative care.

The standard of care treatment for PDAC patients with advanced disease is chemotherapy (Gresham et al., 2014. BMC Cancer 14: 471). While several different chemotherapeutics are available, two systemic combination treatments have shown the most positive results in metastatic disease: FOLFIRINOX and gemcitabine/nab-paclitaxel (Ducreux et al., 2019. Semin Oncol 46: 28-38). FOLFIRINOX is a combination of 5-fluorouracil, leucovorin, irinotecan and oxaliplatin, which was designed initially as a treatment for metastatic pancreatic cancer in 2010 (Conroy et al., 2010. J Clin Oncol 28: 4010). FOLFIRINOX has been shown to significantly improve overall survival, with a median survival of 14 months as compared to 9 months for gemcitabine monotherapy (Walma et al., 2020. Eur J Surg Oncol 47: 699-707). However, it has also been associated with more adverse side-effects, such as neutropenia, thrombocytopenia, diarrhoea, fatigue and neuropathy (Gresham et al., 2014. BMC Cancer 14: 471). Additionally, the percentage of patients responding to FOLFIRINOX treatment is quite low, namely around 16-32% in metastatic disease (Thibodeau and Voutsadakis, 2018. J Clin Med 7:7) and around 13% in locally advanced disease (Walma et al., 2021. Eur J Surg Oncol 47: 699-707). Therefore, discriminating responding from non- responding patients beforehand could profoundly aid in selecting suitable treatment and prevent unnecessary toxicity and discomfort for the patient. Ideally, biomarkers predictive of response to FOLFIRINOX would be identified that are affordable and easy to measure.

There is thus a need for a method to predict a response following FOLFIRINOX therapy in gastrointestinal cancer patients that is sensitive, accurate, affordable and easy to perform.

2 BRIEF DESCRIPTION OF THE INVENTION

It was found that changes in DNA copy number (i.e. copy number variations, or CNVs) of specific genomic regions covering genes, or in the vicinity of genes, such as NF1, MSH2, and FANCM, are associated with FOLFIRINOX therapy responsiveness.

Accordingly, the invention relates to a method of typing a tissue sample of an individual suffering from gastrointestinal cancer, comprising: (i) providing a non- cancerous tissue sample from the individual, wherein said tissue sample comprises genomic nucleic acids; (ii) determining a copy number of at least three genomic marker regions in said tissue sample, wherein a genomic marker region is covering at least part of a neurofibromatosis 1 gene (NF1 ), a genomic marker region is covering at least part of FA Complementation Group M (FANCM) gene and a genomic marker region is covering at least part of MutS Homolog 2 (MSH2) gene; (iii) comparing the copy number of the at least three genomic marker regions to a baseline or reference, thereby typing the sample for a response following FOLFIRINOX therapy, whereby an increase in copy number in at least one of the three marker regions is an indicator of FOLFIRINOX therapy non-responsiveness and whereby a no increase in copy number is an indicator of FOLFIRINOX therapy responsiveness.

In a preferred method of the invention, a genomic marker region is the genomic marker region corresponding to the NF1 gene listed in Table 1, a genomic marker region is the genomic marker region corresponding to the FANCM gene listed in Table 1 and a genomic marker region is at least one of the genomic marker regions corresponding to the MSH2 gene listed in Table 1.

A preferred method of the invention comprises determining a copy number of at least one further genomic marker region selected from Table 1.

A preferred tissue sample comprises mucous membrane, provided by a swab, preferably a cheek swab, or wherein the tissue sample is a liquid tissue sample, preferably a blood sample.

Preferably, said individual who is typed as being responsive for FOLFIRINOX therapy is subsequently treated with FOLFIRINOX therapy.

Said gastrointestinal cancer preferably is a pancreatic cancer, more preferably a pancreatic ductal adenocarcinoma (PDAC).

Furthermore, the invention provides a method of treating an individual with gastrointestinal cancer, comprising typing a non-cancerous tissue sample from said individual using the method according to the invention, treating an individual that is typed as responsive to FOLFIRINOX therapy with FOLFIRINOX therapy, and treating the individual that is typed as non-responsive to FOLFIRINOX therapy with gemcitabine.

In methods of the invention, said individual that is typed as being responsive to FOLFIRINOX therapy is treated with FOLFIRINOX in combination with gemcitabine and/or radiotherapy.

In methods of the invention, said individual that is typed as being non- responsive to FOLFIRINOX therapy is treated with gemcitabine in combination with paclitaxel or capecitabine, and/or radiotherapy.

Furthermore, the invention provides a use of a copy number of at least three genomic marker regions comprising at least part of a neurofibromatosis 1 gene (NF1), at least part of FA Complementation Group M (FANCM) gene and at least part of MutS Homolog 2 (MSH2) gene, which copy number is determined in a non- cancerous tissue sample from an individual suffering from gastrointestinal cancer, to predict a response of the individual following FOLFIRINOX therapy.

3 BRIEF DESCRIPTION OF THE FIGURES Figure 1: The twenty most occurring genes within all unique enriched pathways. Pathway analyses were performed using two different methods, namely a gene-set enrichment method implemented within Plink and a CNV Collapsing Random Effects Test (CCRET) method. Only genes affected by any of the detected CNVs were included. Pathways overlapping between Plink and CCRET were only included once.

Figure 2: Receiver Operating Characteristics for modelling of the candidate biomarker CNV regions. The y-axis shows the True Positive Rate and the x-axis shows the False Positive Rate. Each line represents performance of the respective colours model over different cut offs.

4 DETAILED DESCRIPTION OF THE INVENTION

For the purpose of clarity and a concise description, features are described herein as part of the same or separate embodiments, however, it will be appreciated that the scope of the invention may include embodiments having combinations of all or some of the features described.

4.1 Definitions

As are used herein, the singular forms "a", "an" and "the", are intended to include the plural forms as well.

As is used herein, the term "or" includes any and all combinations of one or more of the associated listed items, unless the context clearly indicates otherwise (e.g. if an “either .or” construction is used).

As are used herein, the terms "comprise" and "comprising", and conjugations thereof, are open language and specify the presence of stated features but do not preclude the presence or addition of one or more other features.

It will be understood that when a particular step of a method is referred to as subsequent to another step, it can directly follow said other step or one or more intermediate steps may be carried out before carrying out the particular step, unless specified otherwise.

As is used herein, the term “cancer” or “tumour”, refers to a disease or disorder resulting from the proliferation of oncogenically transformed cells. As used herein, the term “gastrointestinal cancer”, is any cancer originating from cells of the gastrointestinal tract and accessory organs of digestion, including oesophagus cancer, stomach cancer, biliary system cancer, small intestine cancer, large intestine (i.e. colon) cancer, colorectal cancer, pancreatic cancer, rectum cancer and anus cancer.

As is used herein, the term “pancreatic cancer”, is any cancer originating from cells of the pancreas.

As used herein, the term "PDAC" refers to pancreatic ductal adenocarcinoma, which is pancreatic cancer that originates in the ducts of the pancreas.

As is used herein, the term “individual”, refers to a human.

As is used herein, the term “tissue sample”, refers to any tissue that can be completely or partly obtained from an individual by various means including, for example, a swab, a biopsy such as needle biopsy and surgery. Suitable tissue samples according to the invention are tissue samples from which genomic nucleic acids can be isolated, a cheek swab, blood sample or other biological fluid containing genomic nucleic acids. The act of obtaining a sample by surgery is not part of this invention.

As is used herein, the term “non-cancerous tissue sample”, refers to a tissue sample comprising healthy tissue cells and not comprising cancerous tissue cells.

As is used herein, the term “somatic genomic nucleic acid”, refers to a nucleic acid including chromosomal DNA that originates from a healthy cell. The term “somatic genomic nucleic acid” is used herein to distinguish a nucleic acid used in a method according to the invention, from a “tumour nucleic acid”, which originates from a tumour cell. The term “somatic genomic nucleic acid” may be equal to the term “germ line genomic nucleic acid”.

As is used herein, the term “typing of a sample”, refers to the classification of a sample based on characterized features. In this invention typing includes assisting in the prediction of a response following FOLFIRINOX therapy. Said assisting preferably is by characterisation of copy numbers of genomic regions in a sample, and predicting a response to a subsequent treatment with FOLFIRINOX, based on the determined copy numbers of the genomic regions. As is used herein, the term “FOLFIRINOX therapy”, refers to a therapy comprising oxaliplatin, irinotecan, fluorouracil (5-FU) and leucovorin (i.e. folinic acid).

As is used herein, the term “therapy responsiveness”, refers to an individual's therapeutic response when treated with a specific therapy, here FOLFIRINOX therapy. A therapeutic response includes, fore example, an increase in survival time, an inhibition of tumour growth, and/or a reduction in tumour volume. For example a response to FOLFIRINOX therapy can be measured using the RECIST criteria (Eisenhauer et al., 2009. Eur J Cancer 45: 228-247). The term responsiveness is used interchangeably with ‘sensitivity’.

As is used herein, the term “genomic marker region”, refers to a region in a human genome of which copy number, alone or in combination with other genomic regions, is correlated with a prediction of an effect, in this application FOLFIRINOX therapy responsiveness.

As is used herein, the term “copy number” refers to the number of copies of a nucleic acid sequence that is present in a genomic nucleic acid of a tissue sample.

As is used herein, the term “copy number variation (CNV)” refers to variation in the number of copies of a nucleic acid sequence that is present in a genomic tissue sample in comparison with the copy number of the nucleic acid sequence present in a reference sample. Copy number variations include deletions, duplications and multiplications. CNV encompasses complete chromosomal aneuploidies and partial aneuploidies.

As is used herein, the term “a baseline”, refers to a certain value which may be used to reference the copy number of an individuals’ nucleic acid sequence to. A baseline can be an integer e.g. 2. A baseline can be based on a measurement e.g. based on the measured intensity in a (micro)array or based on a Ct (threshold cycle) value measured using qPCR. For example, in a SNP array, a genomic region may be covered by several individual SNPs. An increased or decreased signal for an individual SNP, or an group of SNPs, when compared to the SNPs upstream or downstream from that individual SNP, or group of SNPs, is indicative for the presence of a CNV.

In addition, for a specific SNP the array platform may include two types of hybridization probes, specific to two types of known alleles, usually coded as A and B. A SNP genotype can be determined by measuring the ratios of the hybridization intensities for A and B probes. An increased signal for allele A and a decreased signal for allele B, or an increased signal for allele B and a decreased signal for allele A, is indicative for a loss of heterozygosity (LOH), which can also be used for assisting in the prediction of a response following FOLFIRINOX therapy.

CNVs such as duplications and deletions result in an increase or decrease of the total measured intensity of a specific SNP, compared to a reference. Intensity ratios for CNVs spanning multiple SNPs will have patterns distinct from normal (disomic) genomic regions. Computational methods such as PennCNV (Wang et al., 2007. Genom Res 17: 1665-1674) can be used to detect CNVs using hybridization intensities and allele frequencies from SNP markers.

As is used herein, the term “a reference”, refers to any particular known, sequenced or characterized nucleic acid sequence or sample, which may be used to reference an individuals’ nucleic acid sequence or sample to. A reference may comprise nucleic acid sequences or samples from multiple individuals. Preferably, a reference comprises a nucleic acid sequence or sample derived from a non- responsive individual or healthy individual e.g. to UCSC Genome Browser hg19 human genome assembly.

As used herein, the term “combination”, refers to the administration of effective amounts different therapies, e.g. FOLFIRINOX therapy and gemcitabine, to an individual in need thereof. Said different therapies may be provided in one pharmaceutical preparation, or as two or more distinct pharmaceutical preparations. When administered as distinct pharmaceutical preparations, they may be administered on the same day or on different days to a patient in need thereof, and using a similar or dissimilar administration protocol, e.g. daily, twice daily, biweekly, orally and/or by infusion. Said combination is preferably administered repeatedly according to a protocol that depends on the patient to be treated (age, weight, treatment history, etc.), which can be determined by a skilled physician.

4.2 Sample collection and pre-processing

A tissue sample according to the invention may be obtained from an individual with a cancer such as a gastrointestinal cancer. Said individual with gastrointestinal cancer can be an individual diagnosed with gastrointestinal cancer or likely to be diagnosed with gastrointestinal cancer. Said individual with gastrointestinal cancer is an individual suffering from gastrointestinal cancer or likely to suffer from gastrointestinal cancer. Said gastrointestinal cancer preferably is a pancreatic cancer, more preferably a pancreatic ductal adenocarcinoma (PDAC).

The tissue sample may comprise any tissue sample comprising genomic nucleic acids from said individual such as blood, serum, plasma, mucosa, saliva, urine, stool, lymph fluid, and/or cerebrospinal fluid. Said tissue sample is a non- cancerous tissue sample, i.e. not comprising cancerous tissue cells. Said tissue sample does not, or is not known to, comprise gastrointestinal cancer cells. It is explicitly stated that the act of obtaining a sample from an individual is not part of this invention.

A tissue sample may be collected in any clinically acceptable manner. A tissue sample preferably comprises mucous membranous tissue, provided by a swab such as a cheek swab. A swab is to be understood to encompass standard medical swabs, i.e. swabs designed for taking biological samples such as mucous membranes. Swabs may be processed using any means known in the art. A further preferred tissue sample is a liquid tissue sample, preferably a blood sample. Said liquid sample may be obtained by phlebotomy, including venipuncture. The main advantage of using a non-cancerous tissue sample such as a swab or blood sample, in contrast to a cancerous tissue sample, is that the tissue collection method is less invasive e.g. no tumour biopsy is needed.

Genomic nucleic acids can be obtained from a tissue sample by methods known in the art. There are several DNA extraction techniques known to date, including organic extraction such as phenol-chloroform extraction, non-organic extraction using e.g. proteinase K and salting out, Chelex extraction (Chelex, Bio- Rad), silica-based column techniques (e.g. QIamp, Qiagen) and magnetic beads- based techniques (e.g. AMPure beads, Beckman Coulter or ReliaPrep, Promega). A preferred method of obtaining genomic nucleic acids from a tissue sample uses a lysis buffer for breaking open the cells, followed by nucleic acid extraction by binding of the nucleic acids to the particles having magnetic properties e.g. magnetizable cellulose (as described before US6855499B1). The nucleic acids bound to the particles are then washed and the bound nucleic acids are eluted. In this way the nucleic acids in the tissue sample are separated from all non-nucleic acid material. The ReliaPrep series from Promega provides commercial kits for this preferred method of obtaining nucleic acids from a tissue sample.

4.3 Genomic regions associated with FOLFIRINOX therapy response

The invention provides a set of genomic marker regions whose copy number is correlated with a response, here FOLFIRINOX therapy responsiveness in gastrointestinal cancer patients. Said set of genomic marker regions comprises at least three genomic marker regions, wherein a genomic marker region is covering at least part of neurofibromatosis 1 (NF1 ) gene, a genomic marker region is covering at least part of FA Complementation Group M (FANCM) gene, and a genomic marker region is covering at least part of MutS Homolog 2 (MSH2) gene. The NF1 gene is located on the human chromosome 17 at location 29421945 to 29704695 according to UCSC Genome Browser hgl9 human genome assembly. The FANCM gene is located on the human chromosome 14 at location 45605136 to 45670093 according to UCSC Genome Browser hgl9 human genome assembly. The MSH2 gene is located on the human chromosome 2 at location 47630206 to 47710367 according to UCSC Genome Browser hgl9 human genome assembly. Preferably, said set of genomic marker regions comprises at least three genomic marker regions, wherein a genomic marker region is the genomic marker region corresponding to the NF1 gene listed in Table 1, a genomic marker region is the genomic marker region corresponding to the FANCM gene listed in Table 1 and a genomic marker region is at least one of the genomic marker regions corresponding to the MSH2 gene listed in Table 1.

Said set of genomic marker regions preferably comprises a region on chromosome 14, from nucleotide 45628281 to nucleotide 45628393, a region on chromosome 17, from nucleotide 29663346 to nucleotide 29665822, and a region on chromosome 2, from nucleotide 47630494 to nucleotide 47630496; a region on chromosome 14, from nucleotide 45628281 to nucleotide 45628393, a region on chromosome 17, from nucleotide 29663346 to nucleotide 29665822, and a region on chromosome 2, from nucleotide 47630496 to nucleotide 47630511; a region on chromosome 14, from nucleotide 45628281 to nucleotide 45628393, a region on chromosome 17, from nucleotide 29663346 to nucleotide 29665822, and a region on chromosome 2, from nucleotide 47630511 to nucleotide 47630513; a region on chromosome 14, from nucleotide 45628281 to nucleotide 45628393, a region on chromosome 17, from nucleotide 29663346 to nucleotide 29665822, a region on chromosome 2, from nucleotide 47630494 to nucleotide 47630496, and a region on chromosome 2, from nucleotide 47630496 to nucleotide 47630511; a region on chromosome 14, from nucleotide 45628281 to nucleotide 45628393, a region on chromosome 17, from nucleotide 29663346 to nucleotide 29665822, a region on chromosome 2, from nucleotide 47630494 to nucleotide 47630496, and a region on chromosome 2, from nucleotide 47630511 to nucleotide 47630513; a region on chromosome 14, from nucleotide 45628281 to nucleotide 45628393, a region on chromosome 17, from nucleotide 29663346 to nucleotide 29665822, a region on chromosome 2, from nucleotide 47630496 to nucleotide 47630511, and a region on chromosome 2, from nucleotide 47630511 to nucleotide 47630513; or a region on chromosome 14, from nucleotide 45628281 to nucleotide 45628393, a region on chromosome 17, from nucleotide 29663346 to nucleotide 29665822, a region on chromosome 2, from nucleotide 47630494 to nucleotide 47630496, a region on chromosome 2, from nucleotide 47630494 to nucleotide 47630496, and a region on chromosome 2, from nucleotide 47630511 to nucleotide 47630513.

An increase in copy number in at least one of these genomic regions is indicative of an individual being non-responsive for treatment with FOLFIRINOX. A no increase in copy number in copy number is indicative of an individual being responsive for treatment with FOLFIRINOX.

More preferably, a set of at least 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, or 26 genomic marker regions from the genomic marker regions listed in Table 1 is used, including the indicated genomic regions covering at least part of NF1 gene, FANCM gene and one or more of the regions on chromosome 2 in the vicinity of the MSH2 gene. Table 1: Genomic marker regions associated with FOLFIRINOX therapy response. Chromosome number, start and end coordinates are derived from Genome

Reference Consortium Human Build 37 (GRCh37; Hgl9). “Type” indicates the type of CNV identified within the genomic marker region. The column “Response vs Non-Response” indicated the correlation of the identified CNV to the FOLFIRINOX response or non-response groups. “LOH” = loss of heterozygosity. = Gene not within region; = no nearby gene.

4.4 Determining of copy number variation of genomic marker regions

The determination of a copy number of one or more genomic marker regions can be accomplished by any means known in the art such as Southern blotting, quantitative PCR (qPCR), paralog-ratio testing (PRT), molecular copy number counting (MCC), microarray analysis, and DNA sequencing such as next- generation sequencing (NGS). Preferably, the copy number of multiple genomic marker regions are assessed simultaneously on a genome-wide scale, by array- based and NGS-based methods as described by Li and Olivier (2013. Physiol Genomics 45: 1-16).

Quantitative PCR (qPCR), or real-time PCR (RT-PCR), is a technique which is used to amplify and simultaneously quantify a template nucleic acid molecule. qPCR can be used to compare threshold cycles (Ct) between a target gene and a reference sequence with normal copy numbers, to generate ACt values which are used for CNV calculation. The detection of the amplification product can in principle be accomplished by any suitable method known in the art. The amplified products may be directly stained or labelled with radioactive labels, antibodies, luminescent dyes, fluorescent dyes, or enzyme reagents. Direct DNA stains include for example intercalating dyes such as acridine orange, ethidium bromide, ethidium monoazide or Hoechst dyes. These intercalating dyes are non-specific and bind to all double stranded DNA in the PCR. An increase in DNA products during amplification, results in an increased fluorescence intensity being measured. Another direct DNA detection method includes the use of sequence specific DNA probes consisting of a fluorescent reporter and quencher. Upon binding of the probe to its complementary sequence, polymerases of the PCR break the proximity of the reporter and the quencher, resulting in the emission of fluorescence. Commonly used reporter dyes include FAM (Applied Biosystems), HEX (Applied Biosystems), ROX (Applied Biosystems), YAK (ELITech Group) or VIC (Life Technologies) and commonly used quenchers include TAMRA (Applied Biosystems), BHQ (Biosearch Technologies) and ZEN (Integrated DNA Technologies). Alternatively, the amplified product may be detected by incorporation of labelled dNTP bases into the synthesized DNA fragments. Detection labels which may be associated with nucleotide bases include, for example, fluorescein, cyanine dye and BrdUrd. For the simultaneous detection of multiple nucleic acid gene expression products, a multiplex qPCR can be used. In multiplex qPCRs, two or more template nucleic acid molecules are amplified and quantified in the same reaction. A commonly used method of achieving the simultaneously detection of multiple targets, is by using probes with different fluorescent dyes to distinguish distinct nucleic acid targets.

PRT uses a single pair of primers to explore the degree of similarity between sequence elements (often dispersed repeats) both in the target locus (with CNV) and a remote reference locus.

MCC is a simple assay that exploits the ability of PCR to detect single molecules of a target sequence (marker). In MCC, the number of samples that are positive for a target sequence can be counted and compared with those of other target sequences, thus allowing the relative copy number of multiple sequences in the original DNA sample to be inferred.

Further suitable means include multiplex PCR-based approaches, such as multiplex amplifiable probe hybridization, multiplex ligation- dependent probe amplification, multiplex PCR-based real-time invader assay, quantitative multiplex PCR of short fluorescent fragments, and multiplex amplicon quantification, may be used for targeted detection of CNVs.

The most commonly used array-based copy number determination methods are comparative genome hybridization (CGH) arrays and SNP arrays. CGH arrays have been developed for copy number comparison between differentially labeled target and reference DNAs by measuring the fluorescence ratio along the length of each nucleic acid region, indicating relative losses or gains in a target sample using fluorescence in situ hybridization (FISH). CGH arrays use arrays of long (synthetic) oligonucleotides to probe specific regions of interest for copy number assessment. SNP arrays, primarily supplied by Affymetrix (Santa Clara, CA) and Illumina (San Diego, CA), use short base-pair sequences to capture labelled fragments of genomic DNA and to infer copy number based on the captured hybridization intensities. Optionally, a reference may be cohybridized together with a target sample. A preferred array for copy number determination in a method according to the invention is the Infinium Global Screening Array from Illumina.

Next-generation sequencing (NGS) provides a base-by-base view of the genome and allows for the detection of small copy number variants and novel copy number variants that arrays often miss. NGS platforms, including Illumina® sequencing; Roche 454 pyrosequencing®, ion torrent and ion proton sequencing, and ABI SOLiD® sequencing, allow sequencing of multiple DNA fragments in parallel. Bioinformatics analyses are used to piece together these fragments by mapping the individual reads. Each base is sequenced multiple times, providing high depth to deliver accurate data and an insight into overall DNA variation, including unexpected DNA variation. NGS can be used to sequence, for example, a complete exome, i.e. substantially all genomic exon sequences, a limited number of individual genes including intronic sequences, and any variation thereof.

Pyrosequencing detects the release of inorganic pyrophosphate (PPi) as particular nucleotides are incorporated into the nascent strand (Ronaghi et al., 1996. Analytical Biochemistry 242: 84-9; Ronaghi, 2001. Genome Res 11: 3-11; Ronaghi et al., 1998. Science 281: 363; U.S. Patent No. 6,210,891 ; U.S. Patent No. 6,258,568 ; and U.S. Patent No. 6,274,320, which are all incorporated herein by reference. In pyrosequencing, released PPi can be detected by being immediately conversion to adenosine triphosphate (ATP) by ATP sulfurylase, and the level of ATP generated is detected via luciferase-produced photons.

NGS also includes so called third generation sequencing platforms, for example nanopore sequencing on an Oxford Nanopore Technologies platform, and single-molecule real-time sequencing (SMRT sequencing) on a PacBio platform, with or without prior amplification of the RNA expression products.

Further high throughput sequencing techniques include, for example, sequencing-by-synthesis. Sequencing-by-synthesis or cycle sequencing can be accomplished by stepwise addition of nucleotides containing, for example, a cleavable or photo-bleachable dye label as described, for example, in U.S. Patent No. 7,427,673; U.S. Patent No. 7,414,116; WO 04/018497; WO 91/06678; WO 07/123744; and U.S. Patent No. 7,057,026, all of which are incorporated herein by reference.

Sequencing techniques also include sequencing by ligation techniques. Such techniques use DNA ligase to incorporate oligonucleotides and identify the incorporation of such oligonucleotides and are inter alia described in U.S. Patent No 6,969,488 ; U.S. Patent No. 6,172,218 ; and U.S. Patent No. 6,306,597. Other sequencing techniques include, for example, fluorescent in situ sequencing (FISSEQ), and Massively Parallel Signature Sequencing (MPSS).

Multiple algorithms and platforms to analyse CGH and SNP array data and NGS data and identify CNVs have been developed and are described by Li and Olivier (2013. Physiol Genomics 45: 1-16).

4.5 Prediction of an individual's responsiveness to FOLFIRINOX therapy.

The invention provides a method for typing a non-cancerous tissue sample to predict an individuals’ responsiveness to FOLFIRINOX therapy. Typing of a tissue sample can be performed in various ways. In one method, the difference or similarity between an individual's copy numbers for a certain genomic marker region or multiple genomic marker regions and the copy number for said genomic marker region(s) of a previously established reference is determined. The reference's copy number for a certain genomic marker region is composed of the average copy number of said genomic marker region in a sample from a reference group. The reference group may comprise a single individual. Preferably the reference group comprises the copy numbers for said genomic marker region of at least 10, 25, 50, 100, 200 or 300 individuals. The reference group may include individuals with different responses following FOLFIRINOX therapy. The reference group may also include individuals that all are responsive to FOLFIRINOX therapy (i.e. responsive reference group) or individuals not responsive to FOLFIRINOX therapy (i.e. non-responsive reference group). Alternatively, an individual's sample can also be typed by comparing the individual's copy number(s) to multiple reference copy numbers. For example, the individual's copy number(s) can be compared to both reference's copy numbers identified above (i.e. the responsive reference group and the non-responsive reference group). If the copy number(s) of the individual's sample is substantially more similar to the responsive reference group, when compared to the non- responsive reference group, it will be predicted as responsive to FOLFIRINOX therapy.

The difference or similarity between an individual's copy number(s) and a reference copy number(s) for a certain genomic marker region or multiple genomic marker regions can be determined by determining a correlation of the copy numbers of the genomic marker regions. For example, one can determine whether the copy number of a subset of genomic marker regions in a tissue sample correlates to the copy numbers of the same subset of genomic marker regions in a reference profile. This correlation can be numerically expressed using a correlation coefficient. Several correlation coefficients can be used. Preferred methods are parametric methods which assume a normal distribution of the data. One of these methods is the Pearson product-moment correlation coefficient, which is obtained by dividing the covariance of the two variables by the product of their standard deviations.

Said correlations between the copy numbers of genomic marker regions in the individual's sample and the reference group, can be used to produce an overall similarity score for the set of marker genes used. A similarity score is a measure of the average correlation of copy numbers of a set of genomic marker regions in a tissue sample from an individual with gastrointestinal cancer and a reference profile. Said similarity score can, but does not need to be, a numerical value between +1, indicative of a high correlation between the copy numbers of the set of genomic marker regions in a sample of said individual and said reference profile, and -1, which is indicative of an inverse correlation. A threshold can be used to differentiate between responsive samples, and non-responsive. Said threshold is an arbitrary value that allows for discrimination between samples from responsive individuals, and samples of non-responsive individuals. If a similarity threshold value is employed, it is preferably set at a value at which an acceptable number of patients with response would score as false negatives, and an acceptable number of patients with no response would score as false positives.

The average copy number of a genomic marker region in the individual's sample will be about 2. An individual of which one or more of the genomic marker regions covering at least part of NF1, FANCM and MSH2, shows an increased copy number, will be typed by methods of the invention as being non-responsive to FOLFIRINOX therapy. Hence, said individual will benefit from treatment with an alternative to FOLFIRINOX, for example gemcitabine. On the other hand, an individual of which the genomic marker regions covering at least part of NF1, FANCM and MSH2, do not show an increased copy number, will be typed by methods of the invention as being responsive to FOLFIRINOX therapy. Hence, said individual will benefit from treatment with FOLFIRINOX.

Alternatively, or in addition to the determination of copy number variation for certain genomic regions (table 1), a loss of heterozygosity (LOH) may be used to type an individual according to the invention. As used herein, the term “loss of heterozygosity” refers to the loss of one copy of a segment of the genome of a diploid individual. LOH occurs when a cell that is originally heterozygous at a locus loses one of its two alleles at that locus, either by simple deletion of one allele (copy-loss LOH), or by deletion of one allele accompanied by duplication of the remaining allele (copy-neutral LOH).

Based on the predictions made by the methods of the invention, one can determine a course of treatment of the individual with gastrointestinal cancer. For example if the individual's copy numbers for certain genomic marker regions are not substantially different from the non-responsive group, or alternatively substantially different from the responsive group, this indicates that the individual is predicted not to be responsive to FOLFIRINOX therapy. In that case, it is not recommended to provide FOLFIRINOX therapy.

4.6 Methods of treating an individual with gastrointestinal cancer

For pancreatic cancer, primary treatment usually involves local treatment including surgery, mostly combined with pre-operative radiotherapy. Treatment of pancreatic cancer may also involve systemic treatment in addition to surgery. The main systemic therapy option used in pancreatic cancer is chemotherapy. Said chemotherapy may be administered pre- operative, post-operative, or both pre- operative and post-operative. Chemotherapy furthermore plays an important role in the treatment of pancreatic cancers that are not considered resectable. Commonly used chemotherapeutic options in pancreatic cancer include treatment with gemcitabine, a combination of gemcitabine and paclitaxel (gemcitabine + nab- paclitaxel, GnP) and FOLFIRINOX treatment.

FOLFIRINOX is a combination of 5-fluorouracil, leucovorin, irinotecan and oxaliplatin and has been shown to significantly improve overall survival in locally advanced pancreatic cancer patients, with a median survival of 14 months as compared to 9 months for gemcitabine monotherapy (Walma et al., 2021. Eur J Surg Oncol 47: 699-707). However, it has also been associated with more adverse side-effects, such as neutropenia, thrombocytopenia, diarrhoea, fatigue and neuropathy (Gresham et al., 2014. BMC Cancer 14: 471). Additionally, the percentage of patients responding to FOLFIRINOX treatment is quite low in metastatic disease and in locally advanced disease. Therefore, discriminating responding from non-responding patients beforehand could profoundly aid in selecting suitable treatment and prevent unnecessary toxicity and discomfort for the patient.

FOLFIRINOX therapy has proven to be effective in pancreatic cancer, and other gastrointestinal cancers such as colorectal cancer, gastric cancer, oesophagus cancer, biliary tract cancer and rectum cancer. When FOLFIRINOX is used to treat colorectal cancer, it is often referred to as “FOLFOXIRI”.

The invention provides a method of treating an individual with gastrointestinal cancer, comprising typing of a non-cancerous tissue sample from said individual according to the invention and treating the individual that is typed as responsive to FOLFIRINOX therapy with FOLFIRINOX therapy, optionally in combination with gemcitabine and/or radiotherapy and treating the individual that is typed non-responsive to FOLFIRINOX therapy with gemcitabine, optionally in combination with paclitaxel or capecitabine, and/or radiotherapy. Said individual is preferably treated with FOLFIRINOX therapy or with gemcitabine, alone or in combination with paclitaxel, capecitabine and/or radiotherapy as neoadjuvant therapy, i.e. before surgery.

The invention provides a use of FOLFIRINOX therapy for the treatment of an individual with gastrointestinal cancer that is typed according to the invention as responsive to FOLFIRINOX therapy.

The invention provides a use of a combination of FOLFIRINOX therapy and gemcitabine and/or radiotherapy for the treatment of an individual with gastrointestinal cancer that is typed according to the invention as responsive to FOLFIRINOX therapy. Said combination for treatment of an individual with gastrointestinal cancer that is typed according to the invention as responsive to FOLFIRINOX therapy may be FOLFIRINOX therapy and gemcitabine; FOLFIRINOX therapy and radiotherapy; FOLFIRINOX therapy, gemcitabine and radiotherapy. The invention provides a use of FOLFIRINOX therapy for the treatment of an individual with gastrointestinal cancer that is typed according to the invention as responsive to FOLFIRINOX therapy, wherein said treatment further comprises gemcitabine and/or radiotherapy.

The invention provides a combination of FOLFIRINOX therapy and gemcitabine and/or radiotherapy for use in the treatment of an individual with gastrointestinal cancer that is typed according to the invention as responsive to FOLFIRINOX therapy.

The invention provides FOLFIRINOX therapy for use in the treatment of an individual with gastrointestinal cancer that is typed according to the invention as responsive to FOLFIRINOX therapy, wherein said treatment further comprises gemcitabine and/or radiotherapy.

The invention provides a use of FOLFIRINOX therapy in the preparation of a medicament for the treatment of an individual with gastrointestinal cancer that is typed according to the invention as responsive to FOLFIRINOX therapy.

The invention provides a use of a combination of FOLFIRINOX therapy and gemcitabine in the preparation of a medicament for the treatment of an individual with gastrointestinal cancer that is typed according to the invention as responsive to FOLFIRINOX therapy.

The invention provides a use of FOLFIRINOX therapy in the preparation of a medicament for the treatment of an individual with gastrointestinal cancer that is typed according to the invention as responsive to FOLFIRINOX therapy, wherein said treatment further comprises gemcitabine and/or radiotherapy.

The invention provides a use of an alternative to FOLFIRINOX therapy for the treatment of an individual with gastrointestinal cancer that is typed according to the invention as non-responsive to FOLFIRINOX therapy. Said alternative to FOLFIRINOX therapy preferably is gemcitabine, optionally in combination with a taxane such as paclitaxel or an antimetabolite such as capecitabine, and/or radiotherapy. Examples of taxanes for the treatment of an individual with cancer include cabazitaxel (Sanofi), docetaxel (Sanofi), paclitaxel (Celgene) and tesetaxel (Odonate Therapeutics). A preferred taxane for use in a method according to the invention is paclitaxel. Examples of antimetabolites for the treatment of an individual with cancer include azacitidine (Pfizer), capecitabine (Roche), cytarabine (Pfizer), cladribine (Janssen Pharmaceutica), clofarabine (Sanofi), decitabine (Janssen Pharmaceutica), fludarabine (Bayer), (5-)fluorouracil (FivepHusion), 5- fluoro- 2 '-deoxyuridine (Sigma-Aldrich), gemcitabine (Eli Lilly and Company), (6- ) mercaptopurin (Aspen), methotrexate (Aldeyra Therapeutics), nelarabine (Novartis), pemetrexed (Eli Lilly and Company), pentostatin (Pfizer) and (6- )tioguanine (Aspen). A preferred antimetabolite for use in a method according to the invention is capecitabine.

Preferred combinations of alternatives to FOLFIRINOX therapy include gemcitabine alone; gemcitabine with paclitaxel; gemcitabine with capecitabine; gemcitabine with radiotherapy; gemcitabine with paclitaxel and radiotherapy; gemcitabine with capecitabine and radiotherapy.

The invention provides a use of gemcitabine for the treatment of an individual with gastrointestinal cancer that is typed according to the invention as non-responsive to FOLFIRINOX therapy, wherein said treatment further comprises a taxane such as paclitaxel or an antimetabolite such as capecitabine, and/or radiotherapy.

The invention provides a combination of gemcitabine and a taxane such as paclitaxel or an antimetabolite such as capecitabine, and/or radiotherapy for use in the treatment of an individual with gastrointestinal cancer that is typed according to the invention as non-responsive to FOLFIRINOX therapy.

The invention provides gemcitabine for use in the treatment of an individual with gastrointestinal cancer that is typed according to the invention as non- responsive to FOLFIRINOX therapy, wherein said treatment further comprises a taxane such as paclitaxel or an antimetabolite such as capecitabine, and/or radiotherapy.

The invention provides a use of gemcitabine in the preparation of a medicament for the treatment of an individual with gastrointestinal cancer that is typed according to the invention as non-responsive to FOLFIRINOX therapy.

The invention provides a use of a combination of gemcitabine and a taxane such as paclitaxel or an antimetabolite such as capecitabine in the preparation of a medicament for the treatment of an individual with gastrointestinal cancer that is typed according to the invention as non-responsive to FOLFIRINOX therapy. The invention provides a use of gemcitabine in the preparation of a medicament for the treatment of an individual with gastrointestinal cancer that is typed according to the invention as non-responsive to FOLFIRINOX therapy, wherein said treatment further comprises a taxane such as paclitaxel or an antimetabolite such as capecitabine and/or radiotherapy.

FOLFIRINOX used in the treatment of individuals with gastrointestinal cancer and typed according to the invention may comprise a therapeutically effective amount of FOLFIRINOX known to treat cancer patients. Said FOLFIRINOX may be administered at a dosage of 300-3600 mg/m² 5-fluorouracil, 100-1200 mg/m² leucovorin, 60-220 mg/m² oxaliplatin, and 120-220 mg/m² irinotecan on day 1 followed by 100-3600 mg/m² 5-fluorouracil every two weeks. For example, FOLFIRINOX may be administered at a dosage of about 400 mg/m² 5- fluorouracil, 400 mg/m² leucovorin, 85 mg/m² oxaliplatin, and 180 mg/m² irinotecan on day 1 followed by 2400 mg/m² 5-fluorouracil every two weeks. Said FOLFIRINOX is preferably administered intravenously.

Gemcitabine used in the treatment of individuals with gastrointestinal cancer and typed according to the invention may comprises a therapeutically effective amount of gemcitabine known to treat cancer patients. Said gemcitabine may be administered at a dosage of 500-3500 mg/m², such as about 1000 mg/m², every 1-4 weeks. Said gemcitabine is preferably administered intravenously.

Treatment of individuals with gastrointestinal cancer and typed according to the invention may in addition to gemcitabine further comprise a therapeutically effective amount of a taxane such as paclitaxel, for example as an albumin stabilized nanoparticle formulation of paclitaxel (called nαb-paclitaxel, or Abraxane®). Said taxane is preferably administered intravenously, preferably by infusion. Said taxane preferably is repeatedly administered, for example once every week, once every two weeks, or once every three weeks. Said paclitaxel may be administered at a dosage of 75-300 mg/m², such as about 125 mg/m², every 1-4 weeks. Said paclitaxel is preferably administered intravenously.

Treatment of individuals with gastrointestinal cancer and typed according to the invention may in addition to gemcitabine further comprise a therapeutically effective amount of capecitabine. Said capecitabine may be administered at a dosage of 250-1500 mg/m², such as about 825 mg/m², twice a day or every 1, 2, 3, 4, 5, 6 or 7 days for 1-4 weeks. Said capecitabine is preferably administered orally.

Treatment of individuals with gastrointestinal cancer and typed according to the invention may be treated in combination with radiotherapy. The term “radiotherapy” encompasses a variety of sources of radiation including but not limited to X-rays, gamma rays, particle beams, proton beam therapy, and high- energy photon radiation. In gastrointestinal cancer treatment, such as pancreatic cancer treatment, three types of radiation are typically used: external beam therapy (EBT), stereotactic body radiotherapy (SBRT) and proton therapy. During EBT, high-energy x-ray beams are delivered to the tumor. Beams are usually generated by a linear accelerator and targeted to destroy cancer cells while sparing surrounding normal tissues. Most pancreatic cancer patients receive a type of external beam therapy called Intensity-Modulated Radiation Therapy (IMRT). IMRT is a type of 3-D radiation that uses linear accelerators to safely and painlessly deliver precise radiation doses to a tumor while minimizing the dose to surrounding normal tissue. EBT typically requires daily treatment over a period of three to six weeks. SRBT is a type of radiation therapy which uses image-guidance to deliver precisely-targeted x-ray radiation in fewer high-dose treatments than traditional EBT. The total dose of radiation is divided into smaller "fractionated" doses given over several days instead of several weeks. This helps preserve healthy tissue. Proton beam radiation therapy delivers radiation to the tumor in a much more confined way than conventional radiation therapy. It allows the delivery of a higher dose to the tumor while minimizing side effects. This can be especially helpful in treating pancreatic cancer since the pancreas is located so closely to other essential organs. Proton beam radiation therapy mostly requires daily treatment over a period of four to five weeks.

A person skilled in the art will understand that the dosage in a combination according to the invention, may be at the low range of the indicated dosages, or even below the indicated dosages.

The invention will now be illustrated by the following examples, which are provided by way of illustration and it will be understood that many variations in the methods described and the amounts indicated can be made without departing from the spirit of the invention and the scope of the appended claims. 5 EXAMPLES

Example 1: Identification of marker regions for the prediction of response to FORFIRINOX

Materials and methods

Study design and data pre-processing

In a cohort-study of 96 pancreatic cancer patients treated with FOLFIRINOX, whole blood DNA samples were collected. These samples were sent to the Uitterlinden lab at Erasmus Medical Center for genotyping using Illumina's Global Screening Array VI containing 692,368 markers. The resulting Illumina Final Report file was subjected to a proprietary in-house quality control pipeline to correct for missing data and deviation from Hardy- Weinberg equilibrium using the Plink vl.9 program (Chang et al., 2015. GigaScience 4: 7).

Response of the tumour was measured by CT-scans after 8 cycles of FOLFIRINOX treatment. These measurements were classified using the RECIST criteria (Eisenhauer et al., 2009. Eur J Cancer 45: 228-247), which define four different types of responses, namely complete response, partial response, stable disease, and progressive disease. At the time of writing, this response assessment was made available for 53 out of 96 patients. From the RECIST classifications, patients were divided into two groups: responders and non-responders to FOLFIRINOX. Responders were defined as complete response, partial response or stable disease, while non-responders were defined as progressive disease.

A list of all known human genes with Entrez ID and Hgl9-based genomic locations was retrieved from the UCSC Genome Table Browser (Haeussler et al., 2019. Nucleic Acids Res 47:D853-D858). The genomic start and end positions were derived from the first and last exon of a gene, respectively. This gene list functioned as a genomic annotation resource throughout the study.

Copy Number Variations analysis

CNV regions were determined from the Log Likelihood Ratios (LRRs) and B- allele frequencies (BAFs) of the Illumina SNP data using the PennCNV algorithm (Wang et al., 2007. Genom Res 17: 1665-1674). The LRR is the relative difference in observed and expected signal intensity of the probes. The BAF is the proportion of intensity from the B-allele over the A-allele. Therefore, a SNP that is not within a CNV region would have a LRR around zero and a BAF around 0, 0.5 or 1 depending on the genotype (AA, AB, and BB respectively). CNV calling by PennCNV was applied using a generated GC model for correction of genomic waves. After CNV calling, close segments were merged. The resulting regions were tested for association with the response phenotype using the CNV specific T-test from CoNVaQ (Larsen et al., 2018. BMC Genomics 19: 369). This method uses a segmentation algorithm to define CNV segments from the CNVs that overlap between samples, which are then tested for the null hypothesis of no association of the CNV segment with FOLFIRINOX response.

Machine Learning-based Biomarker Discovery

To allow further analysis of the CNVs using machine learning methods, the CNVs were transformed into a suitable format. For this purpose, the CNVs were segmented into overlapping CNV regions between samples. These segments were defined as the features. For each of the samples, the occurrence of a CNV within that segment was then determined. This resulted in a feature table in which ‘0’ represents the absence of CNVs (a.k.a. copy number neutral), ‘-1’ the presence of a deletion and ‘+1’ the presence of an amplification.

To identify CNV biomarkers that can accurately predict FOLFIRINOX treatment response, multiple machine learning methods were evaluated. A linear regression model with LASSO penalisation was trained on the transformed CNVs, to reduce noise and allow only the high impact CNVs to be left. This method essentially simplifies the model by reducing the weights of low-impact CNVs. This means the model will perform slightly worse, but it reduces the likelihood of overfitting the model and allowing better interpretation of weights. The method was implemented in R using the glmnet package (Friedman et al., 2010. J Stat Softw 33: 1-22) using a λ of 10-2.4 and mean squared error as loss.

The tree gradient boosting method XGBoost was also implemented in R, using the xgboost R package (Chen et al., Proc 22nd ACM SIHKKD Int Conf Knowl Discov Data Min 785-794), and trained on the transformed CNVs. An advantage of this method is that multiple decision trees can be trained and the ensemble prediction is improved, especially if they complement each other, when compared to a single tree. The XGBoost model was developed using squared error and 11 regularisation, while the parameters were set as follows: alpha = 0.05, eta = 0.1, gamma = 1, max depth = 3 and max rounds = 10.

Performance of both methods was assessed using Leave-One-Out Cross- Validation (LOOCV). Like a standard Cross-Validation the method creates splits of the data to create K equal sized datasets d₁, d₂ ... d_k. In a loop over 1 . . . K the models are trained using data d_k ≠ i and validated by using d_k = i. In LOOCV, K is equal to the amount of samples, which means that 53 models were trained. This allowed for inspection of the weights of all models during the cross-validation and the mean weights of the CNV regions were calculated with their standard deviations.

Pathway Analysis-based Biomarker Discovery

To capture different genetic effects, as well as identify robust gene sets, it is recommended to apply multiple pathway analysis methods (Mooney and Wilmot, 2015. Neuropsychatr Genet 168: 517-527). For this reason, multiple different methods were evaluated through literature review to select the two most suitable methods. A variety of characteristics were taken into consideration, among which the frequency of usage in previous studies and the inclusion of CNV-specific properties.

The gene-set enrichment method implemented within Plink vl.07 was one of the selected methods, as it takes CNV specific features, such as CNV size and individual CNV burden, into account (Kirov et al., 2012. Mol Psychiatry 17: 142- 153; Martin et al., 2014. J Am Acad Child Adolesc Psychiatry 53: 761-770; Raychaudhuri et al., 2010. PLoS Genet 6). Gene sets were taken from the Molecular Signatures Database (MSigDB, Liberzon et al., 2011. Bioinformatics 27:1739-1740), which consists of 31,117 different gene sets. Each gene set was tested for enrichment relative to the whole genome in patients responding to FOLFIRINOX treatment as compared to patients of the non-responding group using Plinkvl.07's cnv-enrichment-test.

The second method was CCRET, which stands for CNV Collapsing Random Effects Test (Tzeng et al., 2015. PLoS Genet 11). This method was selected as it is robust to the etiological heterogeneity of CNVs caused by the difference in the effect of copy number gain and loss on the same genomic region. CCRET takes three different features of CNVs into account: dosage, length and intersection with genes from the provided gene set. Subsequently, it uses these features in a mixed model framework in order to evaluate either dosage or gene-intersection effects, representing burden and enrichment tests respectively. In this study, the CCRET method for evaluating the gene-intersection effect, using Davies statistical test (Davies, 1980. J R Stat Soc Ser C Appl Stat 29: 323-333) and gene sets from MSigDB, was adopted in R following the CCRET publication (Tzeng et al., 2015. PLoS Genet 11: 1005403).

Candidate Biomarker Evaluation

The results of previous analysis were used to define a subset of candidate biomarker regions. These candidate regions were further analyzed using generalized linear models (GLM) to calculate their individual and combinatory discriminative power (Nelder and Wedderburn, 1972. J R Stat Soc Ser Gen 135: 370-384). Each possible combination of candidate regions was modeled using the GLM implementation within the caret R package (Kuhn, 2008. J Stat Softw 28: 1- 26) and their predictive power was assessed by their LOOCV performance. The resulting AUC and accuracy scores were used to select the optimal list of candidates.

Results

Characteristics and clinical outcome of the patients

Data were collected and analyzed as described in the methods section. Genotyping resulted in a total of 692,368 markers of which 687,339 were kept after quality control and were included for further analysis. The clinical outcome and distribution of response is listed in Table 2.

Table 2: Characteristics and clinical outcome of patients included within the study. The RECIST scores were converted to binary classification of responders vs. non- responders. Patients that showed complete response, partial response and stable disease were defined as responders and patients with progressive disease were defined as non-responders.

Identification of seven germline CNV regions associated with FOLFIRINOX response

CNV calling using PennCNV resulted in 1,222 unique CNVs within all patients. Statistical testing of these CNVs was performed using CoNVaQ's CNV- specific adaptation of Fisher's exact test. CNVRs with a P-value below significance threshold 0.05 are listed in Table 3. In total, seven CNVRs were deemed to be associated with FOLFIRINOX response

Table 3: CNV regions with P-value < 0.05, as determined by CoNVaQ's Fisher's exact test. Chromosome, start and end coordinates are derived from Hgl9. P-values have not been corrected for multiple testing. LOH: Loss-of-Heterozygosity. Patients describe the total number of patients genotyped (n=96) in which the CNV region was identified. Response vs non-response was determined out of all patients with available response data (n=53).“-“ = no nearby gene.

Machine Learning-based Biomarker Discovery

A LASSO regression model was developed on the CNV data as described in the methods section. Leave-One-Out Cross-Validation resulted in an AUC score of 0.60, with a sensitivity of 0.833 and specificity of 0.273. The most important CNVs for the XGBoost model and their weights are listed in Table 4. Most notable, CNVs within NF1 and MSH2 are also of importance for the LASSO model.

Table 4: Most important LASSO features as derived using LOOCV. Abbreviations: Chr: chromosome no., Start / End: CNV start and end location on chromosome hgl9, # samples: number of samples CNV was called in, Weight: mean information gain of this feature over all CV folds, Gene: closest gene. * = Gene not within segment. “-“ = no nearby gene.

An XGBoost model was developed on the CNV data as described in the methods section. Leave-One-Out Cross-Validation resulted in an AUC score of 0.955, with a sensitivity of 0.952 and specificity of 0.727, making it the best performing model out of the 2. The most important CNVs for the XGBoost model and their weights are listed in Table 5. Comparing Table 3, 4 and 5, CNVs within the MSH2 gene are identified thrice to be associated with FOLFIRINOX responsiveness. Table 5: Most important XGBoost features as derived using LOOCV.

Abbreviations: Chr: chromosome no., Start / End: CNV start and end location on chromosome hgl9, # samples: number of samples CNV was called in, Gain: mean information gain of this feature over all CV folds, Gene: closest gene. * = Gene not within segment.“-“ = no nearby gene.

Pathway Analysis-based Biomarker Discovery

Pathway analysis was executed on gene sets retrieved from the MSigDB using Plink and CCRET as described in the methods section. Using a significance threshold of 0.05 for each of the methods, this resulted in a total of 108 and 844 pathways being differentially enriched between responding and non-responding patients identified by Plink and CCRET respectively. Of these pathways, 61 overlapped between methods. The enriched pathways were studied further by determining the most often affected genes shared between them. The occurrence of each individually affected gene within all enriched pathways was determined by calculating the occurrences of genes affected by CNVs using gene coordinates from UCSC genes. The twenty most occurring affected genes and the number of pathways they are identified within are shown in Figure 1. Notably, the CNVs within NF1 and MSH2 are most often present within all enriched pathways. Besides MSH2, other affected DNA Damage Repair genes included BRCA2, RAD51C, MUTYH, TP73, and MSH3.

One other notable enriched pathway that overlaps with a significantly associated CNV within FANCM, is the Fanconi Anemia (FA) pathway from Reactome (see //reactome.org/content/detail/R-HSA-6783310). This pathway is significantly enriched by both Plink and CCRET, with P-values of 0.0220 and 0.0104 respectively. The FA pathway is a DNA repair pathway, specifically for recognition and repair of interstrand cross-links (Niraj et al., 2019. Annu Rev Cancer Biol 3: 457-478). While deficiencies in the FA pathway, among which FANCM, are also known to make cancers more sensitive to drugs targeting DNA- damage repair mechanisms (Liu et al., 2020. Cell Biosci 10: 39). to our knowledge this is the first study to identify a somatic CNV within FANCM that is associated with FOLFIRINOX treatment response.

Candidate Biomarker Evaluation

Based on previous results, the CNVs within MSH2, NF1 and FANCM were selected for further evaluation. As described in the methods section, generalized linear models were built from either germline CNVs of the individual genes as well as all possible combinations. The resulting ROC-plots from the Leave-One-Out Cross-Validation statistics are depicted in Figure 2. As can be deduced from the AUC performance, the predictive value for each of the genes individually is around random. However, the combination of MSH2, NF1 and FANCM together vastly improves the predictive power. With an AUC of 0.67, the combination of somatic CNVs within MSH2, NF1 and FANCM can function as a biomarker signature for FOLFIRINOX responsiveness.

Claims

Claims

1. A method of typing a tissue sample of an individual suffering from gastrointestinal cancer, comprising:

(i) providing a non-cancerous tissue sample from the individual, wherein said tissue sample comprises genomic nucleic acids;

(ii) determining a copy number of at least three genomic marker regions in said tissue sample, wherein a genomic marker region is covering at least part of a neurofibromatosis 1 gene (NF1 ), a genomic marker region is covering at least part of FA Complementation Group M (FANCM) gene and a genomic marker region is covering at least part of MutS Homolog 2 (MSH2) gene;

(iii) comparing the copy number of the at least three genomic marker regions to a baseline or reference; thereby typing the sample for a response following FOLFIRINOX therapy, whereby an increase in copy number in at least one of the three marker regions is an indicator of FOLFIRINOX therapy non-responsiveness and whereby a no increase in copy number is an indicator of FOLFIRINOX therapy responsiveness.

2. The method according to claim 1, wherein a genomic marker region is the genomic marker region corresponding to the NF1 gene listed in Table 1, a genomic marker region is the genomic marker region corresponding to the FANCM gene listed in Table 1 and a genomic marker region is at least one of the genomic marker regions corresponding to the MSH2 gene listed in Table 1.

3 The method according to claim 1 or claim 2, comprising determining a copy number of at least one further genomic marker region selected from Table 1.

4. The method according to any one of claims 1-3, wherein the tissue sample comprises mucous membrane, provided by a swab, preferably a cheek swab, or wherein the tissue sample is a liquid tissue sample, preferably a blood sample.

5. The method according to any one of claims 1-4, wherein the individual who is typed as being responsive for FOLFIRINOX therapy is subsequently treated with FOLFIRINOX therapy.

6. The method according to any one of claims 1-5, wherein said gastrointestinal cancer preferably is a pancreatic cancer, more preferably a pancreatic ductal adenocarcinoma (PDAC).

7. A method of treating an individual with gastrointestinal cancer, comprising

- typing a non-cancerous tissue sample from said individual using the method according to any one of claims 1-6;

- treating an individual that is typed as responsive to FOLFIRINOX therapy with FOLFIRINOX therapy, and

- treating the individual that is typed as non-responsive to FOLFIRINOX therapy with gemcitabine.

8. The method of claim 7, wherein said individual that is typed as being responsive to FOLFIRINOX therapy is treated with FOLFIRINOX in combination with gemcitabine and/or radiotherapy.

9. The method of claim 7 or claim 8, wherein said individual that is typed as being non-responsive to FOLFIRINOX therapy is treated with gemcitabine in combination with a taxane such as paclitaxel or an antimetabolite such as capecitabine, and/or radiotherapy.

10. A use of a copy number of at least three genomic marker regions comprising at least part of a neurofibromatosis 1 gene (NF1), at least part of FA Complementation Group M (FAN CM) gene and at least part of MutS Homolog 2 (MSH2) gene, which copy number is determined in a non-cancerous tissue sample from an individual suffering from gastrointestinal cancer, to predict a response of the individual following FOLFIRINOX therapy.