WO2020005068A2

WO2020005068A2 - Gene signatures and method for predicting response to pd-1 antagonists and ctla-4 antagonists, and combination thereof

Info

Publication number: WO2020005068A2
Application number: PCT/NL2019/050401
Authority: WO
Inventors: Oscar KRIGJGSMAN; Daniel Simon Peeper
Original assignee: Stichting Het Nederlands Kanker Instituut-Antoni van Leeuwenhoek Ziekenhuis
Priority date: 2018-06-29
Filing date: 2019-06-28
Publication date: 2020-01-02
Also published as: WO2020005068A8; WO2020005068A3

Abstract

The present invention relates to the field of cancer, more particularly to the field of immunotherapy and gene signatures. Provided are two specific and distinct gene signatures, namely a Response Immune Signal (RIS) gene signature and a Stromal Immune Signal (SIS) gene signature, which can be used as biomarkers to accurately predict the response of a cancer subject to treatment with a PD-1 antagonist (e.g., PD-1 antibody) and/or a CTLA-4 antagonist (e.g., CTLA-4 antibody). In particular, it was found that the RIS and SIS gene signatures of the invention may be used in combination to predict the response of a cancer subject to treatment with a combination therapy consisting of a PD-1 antagonist (e.g., PD-1 antibody) and a CTLA-4 antagonist (e.g., CTLA-4 antibody). The gene signatures of the invention may be advantageously used in methods for treating cancer, such as melanoma, and to help devise treatment strategies best suited to individual patients (e.g., to achieve personalized therapy, and spare patients from undesired side effects, e.g., toxicity).

Description

Title: Gene signatures and method for predicting response to PD-1 antagonists and CTLA-4 antagonists, and combination thereof.

DESCRIPTION

FIELD OF THE INVENTION

The present invention relates to the field of cancer, more particularly to the field of immunotherapy and gene signatures. Provided are two specific and distinct gene signatures, namely a Response Immune Signal (RIS) gene signature and a Stromal Immune Signal (SIS) gene signature, which can be used as biomarkers to accurately predict the response of a cancer subject to treatment with a PD-1 antagonist (e.g., PD- 1 antibody) and/or a CTLA-4 antagonist (e.g., CTLA-4 antibody). In particular, it was found that the RIS and SIS gene signatures of the invention may be used in combination to predict the response of a cancer subject to treatment with a combination therapy consisting of a PD-1 antagonist (e.g., PD-1 antibody) and a CTLA-4 antagonist (e.g., CTLA-4 antibody). The gene signatures of the invention may be advantageously used in methods for treating cancer, such as melanoma, and to help devise treatment strategies best suited to individual patients (e.g., to achieve personalized therapy, and spare patients from undesired side effects, e.g., toxicity).

BACKGROUND

Innovations in both targeted therapy and immunotherapy (IT) of (metastatic) cancer (e.g., melanoma) have led to improved responses in a considerable number of patients. However, for both types of therapies, resistance remains a formidable challenge. In addition, for IT, knowledge on the mechanisms causing resistance is an urgent and unmet need for better patient stratification.

One active field of research in this respect is the field of (metastatic) melanoma. Melanoma is a very heterogeneous disease, mostly due to the number of mutations that it harbors (Berger et al., (2013), Nature 485 502-506; Pleasance et al., (2010), Nature 463 191-196) mainly caused by UV-irradiation. Driver mutations like BRAF^V600E and NRAS⁰⁶¹ have been identified more than a decade ago (Davies et al., (2002), Nature 417 949-954), which has led to the development and clinical use of inhibitors specific for BRAF^V600E, like dabrafenib and vemurafenib, and for MEK, like trametinib and cobimetinib (Chapman et al., (2011), N Engl J Med 364, 2507-2516; Larkin et al., (2014), N Engl J Med 371, 1867-1876; Long et al., (2014), N Engl J Med 371, 1877-1888; Robert et al., (2015), N Engl J Med 372, 30-39). Despite good initial clinical responses to these targeted inhibitors, almost all patients will eventually acquire resistance to these agents (Hugo et al., (2015), Cell 1-14; Shi et al., (2014), Cancer Discovery 4, 80-93; Wagle et al., (2011), Journal of Clinical Oncology 29, 3085-3096), urging the development of additional and new therapeutics.

The high number of UV-induced mutations, which increases the chance for the development of neo-antigens (Schumacher and Schreiber, (2015), Science 348, 69-74), has made this tumor type very eligible for immunotherapy, such as inhibiting checkpoint regulators like CTLA-4 and PD-1 by ipilimumab and nivolumab, respectively (Hodi et al., 2010, N Engl J Med 363, 711-723). Combining these two therapeutics resulted in long-term responses in a subset of patients (Larkin et al., (2015), N Engl J Med 371, 1867-1876; Wolchok et al., (2013), N Engl J Med 369, 122- 133). However, despite this impressive clinical result, a subset of patients is intrinsically resistant, or acquires resistance, to these (combination) immunotherapies.

For instance, AXL expression has been associated with poor response to anti-PD1 therapy(Hugo et al., (2016), Cell 1-14), indicating that the concomitant invasive phenotype might also result in resistance to immunotherapy (Falletta et al., (2017), Genes & Development 1-17).

Despite the discovery of potential biomarkers like AXL and MITF (Boshuizen et al., (2018), Nat. Med. 24, 203-212; MOIIer et al., (2014), Nat Commun 5, 5712), accurately predicting which patients will respond to targeted or immunotherapy remain a difficult task, as current biomarkers are not optimal (e.g., are not accurate).

Other biomarkers, such as gene signatures, are currently being developed for the purpose of predicting response prior to onset of treatment with MAPK pathway inhibitors or immunotherapy. For instance, the innate anti-PD-1 signature (IPRES) has been developed for the purpose of predicting or classifying responders and non- responders to anti-PD1 blockade in melanomas before start of therapy(Hugo et al., (2016), Cell 1-14). However, such biomarkers are not always optimal (e.g., lack accuracy). Further, there is a lack in the art of gene signatures which can be used for accurately predicting response to treatment with a CLTA-4 antagonist (e.g., CTLA-4 antibody such as ipilimumab) and in particular to a combination therapy consisting of a PD-1 antagonist and a CLTA-4 antagonist (e.g., nivolumab and ipilimumab). The availability of such gene signatures would not only enrich or deepen the knowledge on the mechanism underlying treatment resistance but would also allow the design of new or alternative methods for predicting response to a particular therapeutic agent as well as methods for treating cancer (e.g., melanoma).

It is therefore an object of the present invention to provide alternative or improved gene signatures, which can be used to accurately predict the response to a subject suffering from cancer to treatment with a PD-1 antagonist (e.g., a PD-1 antibody such as nivolumab) or to a CLTA-4 antagonist (e.g., a CTLA-4 antibody such as ipilimumab) or in particular to combination therapy consisting of such PD-1 antagonist and CLTA- 4 antagonist. It is a further object of the present invention to provide new of alternative methods for treating cancer (e.g., melanoma) (relying on the use of gene signature(s)), which are better tailored / adapted to the particular need(s) or characteristics of individual cancer patients as well as methods (relying on the use of gene signature(s)) for accurately predicting the response of a cancer subject to a treatment PD-1 antagonist (e.g., a PD-1 antibody such as nivolumab) or to a CLTA-4 antagonist (e.g., a CTLA-4 antibody such as ipilimumab) or to a combination therapy consisting of a PD-1 antagonist and a CLTA4 antagonist.

SUMMARY

The present inventors have uncovered two new gene signatures, namely the Response Immune Signal (RIS) gene signature (involving two or more of the genes in Table 1 , preferably two or more of the genes in Table 1A) and the Stromal Immune Signal (SIS) gene signature (involving two or more, preferably at least 5, 10, 15, or 22, of the genes in Table 2). The present inventors have found that the RIS gene signature and the SIS gene signature as disclosed herein can accurately predict the response of a subject suffering from cancer (for example skin cancer, in particular melanoma), to a treatment with a PD-1 antagonist (e.g., a PD-1 antibody such as nivolumab) or a CTLA-4 antagonist (e.g., CTLA-4 a CTLA-4 antibody such as ipilimumab), respectively.

The present inventors in particular found that both the RIS and SIS gene signatures can also be used to accurately predict the response of a cancer subject (e.g. melanoma) to treatment with a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist (e.g., a PD-1 antibody and a CTLA-4 antibody such as nivolumab and ipilimumab in combination).

In other words, with the gene signatures as disclosed herein, it is possible to predict the response of a cancer patient to treatment with a PD-1 antagonist or to treatment with a CTLA-4 antagonist or to treatment with a combination of a PD-1 antagonist and a CTLA-4 antagonist. In this way, cancer patients can be classified as (likely) responders or non-responders to the indicated treatments.

In order to identify the gene signatures of the present invention (i.e. , the RIS and SIS gene signatures as taught herein), the present inventors have used a specific and innovative methodology, which allowed to obtain better characterization of the gene expression profile in tumor tissues or samples that is characteristic to the disease state (cancer, e.g. melanoma). Specifically, the present inventors have identified the gene expression profile that is characteristic to the intra-tumoral cells as well as the gene expression profile that is characteristic of the stromal cells invading the tumor.

Specifically, the present inventors performed gene expression profiling on patient derived tumor xenografts (PDX panel (n=95) derived from metastatic melanoma (Kemper et al., 2016)) to investigate to what extent stromal signals influence gene expression signatures and subgroup classifications. The present inventors also took advantage of the possibility to discriminate between (murine) stromal and (human) tumor cell-intrinsic gene expression signals to identify melanoma gene expression profiles based on tumor cell-intrinsic signals only (as shown in the examples herein). Based on this methodology, the present inventors developed two genes signatures, i.e. , one based on the tumor cell-intrinsic gene expression signature (RIS gene signature) and the other based on the stromal gene expression signature (SIS gene signature) herein). This was possible because the PDX platform allowed the present inventors to distinguish between gene expression signals that came from the stromal cells (vessels, immune cells etc., which were from mouse origin) and the gene expression signals that came from the tumor cells itself (which were from human origin). In other words, the PDX platform allowed the present inventors to identify the tumor cell-intrinsic signals (gene expression profile) in vivo without having it mixed the stromal signals (gene expression profile). Therefore, the two gene signatures of the present invention (RIS and SIS gene signatures) could not have been easily found from the mere analysis of tumor samples.

The present inventors used the RIS and/or SIS gene signature(s) to determine their ability to predict (predictive power) the response of a cancer subject to treatment with an immune check point blocker, such as anti-PD-1 (e.g. nivolumab) or anti-CTLA-4 (e.g. ipilimumab) or a combination thereof (as shown in the examples herein).

The present inventors found that the specific gene signatures of the invention, i.e. , the RIS gene signature (derived from the gene expression profile of intratumoral cells) and the SIS gene signature (derived from the gene expression profile of stromal cells) surprisingly represents unique sets of gene signatures with high predictive power, which can be used to reliably predict the response of a cancer subject to a treatment with a PD-1 antagonist or a CTLA-4 antagonist or a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist (e.g., a PD-1 antibody and a CTLA-4 antibody such as nivolumab and ipilimumab in combination) (as shown herein).

Further advantages associated with the two gene signatures of the invention (RIS and SIS gene signatures, as taught herein) will be discussed in more details below. DETAILED DESCRIPTION

Definitions

A portion of this disclosure contains material that is subject to copyright protection (such as, but not limited to, diagrams, device photographs, or any other aspects of this submission for which copyright protection is or may be available in any jurisdiction). The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or patent disclosure, as it appears in the Patent Office patent file or records, but otherwise reserves all copyright rights whatsoever.

Various terms relating to the methods, compositions, uses and other aspects of the present invention are used throughout the specification and claims. Such terms are to be given their ordinary meaning in the art to which the invention pertains, unless otherwise indicated. Other specifically defined terms are to be construed in a manner consistent with the definition provided herein. Although any methods and materials similar or equivalent to those described herein can be used in the practice for testing of the present invention, the preferred materials and methods are described herein. For purposes of the present invention, the following terms are defined below.

As used herein, the singular forms "a," "an" and "the" include plural referents unless the context clearly dictates otherwise. For example, a method for administrating a drug includes the administrating of a plurality of molecules (e.g. 10's, 100's, 1000's, 10's of thousands, 100's of thousands, millions, or more molecules).

As used herein, the term“and/or” indicates that one or more of the stated cases may occur, alone or in combination with at least one of the stated cases, up to with all of the stated cases.

As used herein, the term "at least" a particular value means that particular value or more. For example, "at least 2" is understood to be the same as "2 or more" i.e., 2, 3, 4, 5, 6, 7, 8, 9, 10, 1 1 , 12, 13, 14, 15, ... , etc. The term "to comprise" and its conjugations as used herein is used in its non-limiting sense to mean that items following the word are included, but items not specifically mentioned are not excluded. It also encompasses the more limiting“to consist of”.”

The terms“cancer” and“tumor” (used interchangeably), as used herein, refer to or describe the physiological condition in humans that is typically characterized by unregulated cell growth. The terms“cancer” and“tumor” also refer to cells that have undergone a malignant transformation that makes them pathological to the host organism. Cancer cells (e.g. melanoma cells) can be distinguished from non- cancerous cells by techniques known to the skilled person.

The term "tumor sample" or“tumor biopsy sample” as used herein, refers to biological material or tissue that has/have been removed from a tumor, including following a surgical tumor resection or tumor biopsy. The tumor sample can be subjected to a variety of well-known post-collection preparative and storage techniques prior performing, for example, analysis of gene expression profiles. Preferably the tumor sample is not obtained from a lymph node.

The term "subject" or“patient” (used interchangeably) as used herein refers to a subject (a mammal, preferably human) male or female, adult, child or infant, suffering from a cancer (e.g. melanoma), regardless of the stage or state of the cancer.

The terms "treat," treating", "treatment",“therapy” and the like as used herein refer to reducing or ameliorating a disorder (e.g. cancer, e.g. melanoma) and/or symptoms associated therewith. It is appreciated that treating a disorder or condition (e.g. cancer, e.g. melanoma) does not require that the disorder, condition or symptoms associated therewith be completely eliminated. It is further understood that the terms "treat," treating", "treatment", “therapy” and as used herein may be a first or first line of treatment (i.e. patient is naive to any cancer treatment) or a second or third line treatment and so on (i.e. the first treatment or second treatment and so on was not effective or has failed). In the context of the present invention,“treat” or“treating” a cancer or cancer patient or subject means to administer a PD-1 antagonist or a CLTA- 4 antagonist CTLA-4 or a PD-1 antagonist and a CTLA-4 antagonist , possibly with other therapeutic agents, to a subject having a cancer (e.g. a melanoma patient or subject), or diagnosed with a cancer (e.g. melanoma), to achieve at least one positive therapeutic effect, such as for example, reduced number of cancer cells , reduced tumor size, reduced rate of cancer cell infiltration into peripheral organs, or reduced rate of tumor metastasis or tumor growth or increased survival rate (progression free survival), reduce or prevent side effects or toxicity, reduce or prevent risk of relapsing, reduce or prevent treatment resistance, and others. Positive therapeutic effects in cancer can be measured in a number of ways (See, W. A. Weber, (2009) J. Null. Med. 50: 1 S-10S; Eisenhauer E. A. et al. , (2009) Eur. J Cancer 45:228-247). In some preferred embodiments, response to a PD-1 antagonist (e.g., PD-1 antibody) and/or CLTA-4 antagonist (e.g., CTLA-4 antibody) is assessed using RECIST 1.1 criteria (Eisenhauer E. A. et al., (2009) Eur. J Cancer 45:228-247). The dosage regimen of a therapy described herein that is effective to treat a cancer patient (e.g., melanoma patient) may vary according to factors such as the disease state, age, and weight of the patient.

The term “Programmed Death-1 (PD-1)” receptor, as used herein, refers to an immune-inhibitory receptor belonging to the CD28 family. In humans, PD-1 is encoded by the PDCD1 gene. PD-1 is expressed predominantly on previously activated T cells in vivo, and binds to two ligands, PD-L1 and PD-L2. The term“PD-1” as used herein includes human PD-1 (hPD-1), variants, isoforms, and species homologs of hPD-1 , and analogues having at least one common epitope with hPD-1. The complete hPD-1 sequence can be found under GENBANK Accession No. U64863) PD-1 is expressed on immune cells such as activated T cells (including effector T cells), B cells, myeloid cells, thymocytes, and natural killer (NK) cells (Suya Dai et al., (2014) Cellular Immunology, Vol:290, pages 72-79; Gianchecchi et al., (2013), Autoimmun. Rev. 12 1091-1 100).

The term“PD-1 antagonists” as used herein refers to any chemical compound or agent or biological molecule (e.g. antibody) that blocks binding of PD-L1 expressed on a cancer cell (e.g., melanoma cells) to PD-1 expressed on an immune cell (T cell, B cell or NKT cell) and preferably also blocks binding of PD-L2 expressed on a cancer cell (e.g. melanoma cell) to the immune-cell expressed PD-1. In the context of the present invention, it is understood that when a subject (e.g., human individual) is being treated with a PD-1 antagonist (e.g., a PD-1 antibody such as nivolumab), the PD-1 antagonist blocks the binding of (human) PD-L1 to (human) PD-1 , and preferably blocks binding of both (human) PD-L1 and PD-L2 to (human) PD-1. Human PD-1 amino acid sequences can be found in NCBI Locus No.: NP_1005009. Human PD-L1 and PD-L2 amino acid sequences can be found in NCBI Locus No.: NP_054862 and NP_079515, respectively. Non-limiting examples of PD-1 antagonists are antibodies against PD-1 (also referred to as PD-1 antibodies or anti- PD-1 antibodies) such as for instance PD-1 monoclonal antibody (mAb), or antigen binding fragment thereof, which specifically binds to PD-1 , and preferably specifically binds to human PD-1. The mAb may be a human antibody, a humanized antibody or a chimeric antibody, and may include a human constant region. Non-limiting examples of PD-1 antagonist compounds include PD-1 antibodies such as nivolumab (Opdivo®, Bristol-Myers Squibb), pembrolizumab (Keytruda®, Merck), BGB-A317, and others such as PDR001 (Novartis). Other non-limiting examples of PD-1 antagonists include pidilizumab (Cure Tech), AMP-224 (GlaxoSmithKline), AMP-514 (GlaxoSmithKline), PDR001 (Novartis), and cemiplimab (Regeneron and Sanofi). Further PD-1 antagonists also include any anti-PD-1 antibody described in US8008449, US7521051 and US8354509.

Other non-limiting examples of PD-1 antagonists include immunoadhesins (also known as fusion proteins), which are compounds capable of specifically binding to PD- 1 and block its binding to PD-L1. Examples of immunoadhesion molecules that specifically bind to PD-1 are described in W02010/027827, US2016/0304969, and WO2011/066342. For instance, a non-limiting example of a fusion proteins that may be used as PD-1 antagonist in the present invention is AMP-224 (which is recombinant B7-DC Fc-fusion protein composed of the extracellular domain of the PD-1 ligand programmed cell death ligand 2 (PD-L2, B7-DC) and the Fc region of human immunoglobulin (Ig) G1).

The term“cytotoxic T-lymphocyte-associated protein 4 (abbreviated“CTLA-4” and also known as cluster of differentiation 152 (CD152)), as used herein, refers to a protein receptor that functions as an immune checkpoint. CTLA-4 is a member of the immunoglobulin superfamily that is expressed by activated T cells and transmits an inhibitory signal to T cells. CTLA-4 is homologous to the T-cell co-stimulatory protein, CD28, and both molecules bind to CD80 and CD86, also called B7-1 and B7-2 respectively, on antigen-presenting cells. CTLA-4 binds CD80 and CD86 with greater affinity and avidity than CD28 thus enabling it to outcompete CD28 for its ligands. CTLA-4 transmits an inhibitory signal to T cells, whereas CD28 transmits a stimulatory signal. CTLA-4 is also found in regulatory T cells (Tregs) and contributes to their inhibitory function. T cell activation through the T cell receptor and CD28 leads to increased expression of CTLA-4. The CTLA-4 protein is encoded by the CTLA-4 gene in humans (Ensembl ref: ENSG00000163599). Normally, after T-cell activation, CTLA- 4 is upregulated on the plasma membrane where it functions to downregulate T-cell function through a variety of mechanisms, including preventing co-stimulation by outcompeting CD28 for its ligand, B7, and also by inducing T-cell cycle arrest Postow et al (2015) J. Clinical oncology, Vol. 33, pages 1974-1983; Pardoll, D. et al (2012), Nature Reviews Cancer 12, 252-264.

The term“CTLA-4 antagonist” as used herein refers to any chemical compound or agent or biological molecule that blocks binding of CTLA-4 with its ligands B7-1 and/or B7-2. In the context of the present invention, it is understood that when a subject (e.g. human individual) is being treated with a CTLA-4 antagonist (e.g. CTLA-4 antibody such as ipilimumab), the CTLA-4 antagonist blocks the binding of (human) CTLA-4 to (human) B7-1 and/or B7-2.

Non-limiting examples of CTLA-4 antagonist compounds currently considered for clinical use in the treatment of cancer (e.g. melanoma) include antagonistic antibodies against CTLA-4 such as ipilimumab ((Yervoy®, MDX-010, Bristol-Myers Squibb, FDA approved for melanoma in 201 1) as a means of inhibiting immune system tolerance to tumors and thereby providing a potentially useful immunotherapy strategy for patients with cancer. A further example of CTLA-4 antagonist compounds is tremelimumab (Medimmune (Postow et al (2015) J. Clinical oncology, Vol. 33, pages 1974-1983; Pardoll, D. et al (2012), Nature Reviews Cancer, Vol. 12, pages 252-264). Other non-limiting examples of CTLA-4 antagonists include immunoadhesins (also known as fusion proteins), which are compounds capable of specifically binding to CTLA-4 and block its binding to B7-1 and/or B7-2.

The term“antibody” (e.g. PD-1 antibody and CTLA-4 antibody) as used herein refers to any form of antibody, and fragment(s) thereof, which exhibits the desired biological or binding activity (e.g., block the binding of PD-1 to its ligands or block binding of CTLA-4 to its ligands, as discussed above). Thus, it is used in the broadest sense and specifically covers, but is not limited to, monoclonal antibodies (including full length monoclonal antibodies) and fragments thereof, polyclonal antibodies and fragments thereof, multispecific antibodies (e.g., bispecific antibodies) and fragments thereof, humanized, fully human antibodies and fragment thereof, chimeric antibodies and fragments thereof, and camelized single domain antibodies, and fragments thereof.

The terms“(not) likely to respond to treatment with” and“(not) likely to benefit from treatment with” as used herein refer to a situation where the likelihood or chances that a subject treated with a therapeutic agent such as a PD-1 antagonist or a CTLA-4 antagonist or a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist (e.g., nivolumab and ipilimumab) will (or will not) respond to or experience benefits from said therapeutic agent(s). For example, response CTLA-4 can be determined by assessing whether a subject experiences toxicity or side effects in response to treatment or whether the treatment increases the survival rate or whether the treatment causes the tumor size to decrease, etc. (as already described herein elsewhere).

As will be shown herein, it was found that the gene signatures as disclosed herein predict the chances or likelihood of relapsing after onset of the drug treatment. In other words, the SIS gene signature as disclosed herein can be used to predict the chances or likelihood that a patient will relapse from treatment with CTLA-4 antagonist. In other words, the RIS gene signature as disclosed herein can be used to predict the chances or likelihood that a patient will relapse from treatment with PD-1 antagonist. Moreover, it was found that the ratio of the SIS gene signature score over the RIS gene signature score can predict the chances or likelihood that a patient will relapse from treatment with an PD-1 antagonist and a CTLA-4 antagonist (e.g. a combination therapy). Specifically, Figure 4g in the examples show the relationship between the SIS gene signature over the RIS gene signature and the likelihood that a patient will relapse (Ratio SIS/RIS below 0 in Figure 4g) or not relapse (Ratio SIS/RIS above 0 in Figure 4g). Therefore, within the context of the current invention, treatment effects may also be assessed by measuring or observing whether a subject has relapsed or not after onset of the treatment with the therapeutic agent, for example 6 months, 12 months, 18 months, 24 months, or more months after onset of the treatment.

The term“relapse” or“relapsed” or“relapsing” as used herein refers to a situation where a disease (e.g. cancer such as melanoma) or the signs and symptoms of said disease return (e.g. the tumor continues to grow or metastases, etc.) after a period of improvement, e.g. after weeks or months following onset of a treatment. The reasons for relapsing following onset of a particular cancer treatment (e.g. treatment with a PD-1 antagonist or a CTLA-4 antagonist CTLA-4 or a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist (e.g., a PD-1 antibody and a CTLA-4 antibody such as nivolumab and ipilimumab in combination) are diverse, e.g., it may be because the subject developed resistance to a particular therapeutic agent or because the particular therapeutic agent is toxic or because the particular therapeutic agent has not biological or has minimal biological activity in a particular subject, etc.

The term“stratifying patients for treatment with” as used herein refers to separating or classifying subjects or patients into subgroups, e.g. responders to treatment with a particular therapeutic agent (e.g., a PD-1 antagonist (e.g., a PD-1 antibody such as nivolumab) or a CTLA-4 antagonist (e.g., a CTLA-4 antibody such as ipilimumab) or a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist (e.g., a PD-1 antibody and a CTLA-4 antibody such as nivolumab and ipilimumab ) and non responders to said therapeutic agent, or subjects that will (likely) benefit from a particular treatment and subject that will not (likely) benefit from a particular treatment. Stratifying patients is advantageous in certain situations, as it contributes to provide treatment therapies directed to or best appropriate for such patient subgroups. The term“gene signature” or“signature” as used herein, refers to a combined group of genes in a cell or tissue with a uniquely characteristic pattern of gene expression that occurs as a result of an altered or unaltered biological process or pathogenic medical condition (e.g. cancer such as melanoma) with validated specificity in terms of diagnosis, prognosis or prediction of therapeutic response. The present invention provides two new gene signatures, namely a Response Immune Signal (RIS) gene signature (genes that may form the signature are shown in Table 1 , and a preferred list is provided in Table 1A) and a Stromal Immune Signal (SIS) gene signature (genes that may form the signature are shown in Table 2) which can be used to predict response to treatment with a PD-1 antagonist, with a CTLA-4 antagonist, or with the combination as taught herein. Such gene signature may also be referred to as a biomarker.

Method for predicting response

In a first aspect, the present invention relates to a method for predicting the response of a subject suffering from a cancer to treatment with a PD-1 antagonist or for stratifying a subject for treatment with a PD-1 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

(c) Measuring the RNA expression level in the sample of step (b) for each gene in a Response Immune Signal (RIS) gene signature;

(d) Calculating the arithmetic mean of the expression level for each of the genes in the RIS gene signature to generate a RIS signature score, wherein the RIS signature comprises at least two of the genes in T able 1 , preferably at least two of the genes in Table 1A;

(e) Comparing the calculated RIS signature score of step (d) to a reference RIS score; and

(f) Predicting the response of said subject to treatment with a PD-1 antagonist or stratifying said subject for treatment with a PD-1 antagonist, wherein:

-(i) if the calculated RIS score is higher than the RIS reference score, then it indicates that said patient is not likely to benefit from treatment with the PD-1 antagonist (non-responder) or is not likely to respond to treatment with the PD-1 antagonist; or

-(ii) if the calculated RIS score is lower than the RIS reference score, then it indicates that said patient is likely to benefit from treatment with the PD-1 antagonist (responder) or is likely to respond to treatment with the PD-1 antagonist.

The term“Response Immune Signal” (abbreviated as“RIS”) signature as used herein refers to a biomarker, i.e. a gene signature as defined above, that can be used to predict the response (e.g., identify a responder or a non-responder of a subject suffering from a cancer (e.g., melanoma) to a PD-1 antagonist (e.g., a PD-1 antibody such as nivolumab) or that can be used to stratify subjects for treatment with a PD-1 antagonist (e.g., to classify subjects as responders and non-responders to PD-1 antagonist treatment so as to separate them into distinct subgroups) or that can be used to identify which cancer subjects or patients are most likely to achieve or experience a clinical benefit from treatment with a PD-1 antagonist or that can be used to decide whether a patient can be treated with a PD-1 antagonist. In the context of the present invention, the RIS signature comprises at least 2 of the genes in Table 1 below, for instance at least 2, 3, 4, 5, 6, 7, 8, 9, 10, 1 1 , 12, 13, 14, 15, 16, 17, 18, 19, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, or 89 of the genes in Table 1. For instance, the skilled person may select 10 of the genes in Table 1 to determine the RIS signature score (as taught herein), which in turn will be useful to, e.g. establish or determine whether a cancer subject will respond to or will benefit from treatment with a PD-1 antagonist.

In a preferred embodiment, the RIS signature comprises at least 2 of the genes in Table 1A below, for instance at least 2, 3, 4, 5, 6, 7, 8, 9, 10, 1 1 , 12, 13, or 14 of the genes in Table 1A. For instance, the skilled person may select a certain number, for example 1 1 or all of the genes in Table 1A to determine the RIS signature score (as taught herein), which in turn will be useful to, e.g. establish or determine whether a cancer subject will respond to or will benefit from treatment with a PD-1 antagonist. In a preferred embodiment, the RIS signature comprises all of the genes in TablelA // Table 1. RIS gene Signature

Tables 1A - Preferred RIS gene signature

The term“RIS signature score” as used herein refers to a numerical value that is obtained by calculating the arithmetic mean of the expression level of each of the genes in the RIS signature for one tumor sample. For instance, if the RIS signature consists of 3 of the genes in Table 1 or in a preferred embodiment Table 1 A, e.g. gene 1 , gene 2 and gene 3 or gene 10, gene 21 , and gene 40, etc., then the RIS signature score is the mean average of the expression levels for each of these genes. The mean average (arithmetic mean) can be calculated as follows: for instance, if gene 1 has an expression level value of 1.5, and gene 2 has an expression level value of 1.8, and gene 3 has an expression level value of 1.2, then the RIS signature score is: 1.5 + 1.8+ 1.2 = 4.5 divided by 3 = 1.5 for that particular tumor sample. It is understood that the expression levels (RNA levels) of the genes in the RIS signature can be measured by any suitable means, e.g., PCR analysis, in situ hybridization, next generation sequencing, RNA sequencing, nanostring, and the like, and may be expressed as optical density values, amounts (e.g. micrograms), and the like. The method for measuring gene expression (RNA levels) and expressing the data (e.g. optical density data, amounts, etc.) does not influence the way the RIS signature score is calculated, as shown above.

Persons skilled in the art are aware of several methods useful for detecting and quantifying the level of RNA transcripts within a sample. Quantitative detection methods include, but are not limited to, RNA sequencing, arrays (i.e., microarrays), quantitative real time PCR (RT-PCR), multiplex assays, nuclease protection assays, and Northern blot analyses. Generally, such methods employ labeled probes that are complimentary to a portion of each transcript to be detected. Probes for use in these methods can be readily designed based on the known sequences of the genes and the transcripts expressed thereby. In certain embodiments, the probes are designed to hybridize to each of the gene signature transcripts identified in Table 1 , or preferably Table 1A. Suitable labels for the probes are well-known and include, e.g., fluorescent, chemiluminescent and radioactive labels.

The term“reference RIS score” as used herein to a numerical value that is derived from a suitable control, such as a group of (different) tumor samples (e.g., 10, 20, 30, 50, 100, 200, 500 or more tumor samples. The reference RIS score may be obtained as follows e.g. according to situation 1 or 2 below

Situation 1 : 10 tumor samples from cancer patients

Step 1 : calculating the arithmetic mean of the expression level of each of the genes in the RIS signature in each of the tumor sample. For instance, if the RIS signature consists of 3 of the genes in Table 1 , or, in a preferred embodiment, Table 1A, e.g. gene 1 , gene 2 and gene 3, then the RIS signature score is the mean average of the expression levels for each of these genes. The mean average (arithmetic mean) can be calculated as follows: if in sample 1 , gene 1 has an expression level value of 1.5, and gene 2 has an expression level value of 1.8, and gene 3 has an expression level value of 1.2, then the RIS signature score is: 1.5 + 1.8+ 1 .2 = 4.5 divided by 3 = 1.5 for tumor sample 1. The same can be repeated for tumor sample 2, 3, 4, and so on. For instance, if in sample 2, gene 1 has an expression level value of 1.1 , and gene 2 has an expression level value of 1.2, and gene 3 has an expression level value of 1.9, then the RIS signature score is: 1.1 + 1.2+ 1.9 = 4.2 divided by 3 = 1.4 for tumor sample 2.

Step 2: calculating the arithmetic mean (mean average) of the RIS signature score for each tumor sample in the group. For instance, in a group of 10 tumor samples, the mean average of the RIS signature scores is calculated as follows:

Tumor sample 1 = 1.5

Tumor sample 2 = 1.4

Tumor sample 3 = 1.3

Tumor sample 4 = 1.6

Tumor sample 5 = 1.7

Tumor sample 6 = 1.1

Tumor sample 7 = 1.2

Tumor sample 8 = 1.4

Tumor sample 9 = 1.5

Tumor sample 10 = 1.8

The arithmetic average is: 1.5 + 1.4 + 1.3 + 1.6 + 1.7 + 1.1 + 1.2 + 1.4 + 1.5 + 1.8 = 14.5 divided by 10 = 1.45. Therefore, the reference RIS score is 1.45. In the context of situation 1 , the tumor samples used to calculate the reference RIS score are preferably of the same tumor type as the test tumor sample, e.g. a melanoma tumor sample. Further, the tumor samples used to calculate the reference RIS score are preferably obtained from cancer patients who have been treated with a PD-1 antagonist. In the context of situation 1 , it is not essential to know which patients responded or not to the PD-1 antagonist.

In the context of the present invention, the reference RIS score is used as a cut off value or a threshold value for determining whether a cancer subject is likely to respond to or benefit from treatment with a PD-1 antagonist (e.g., a PD-1 antibody such as nivolumab, as explained above). Specifically, this is done by comparing the RIS signature score associated with a tumor sample to be tested (referred to as a test sample) for said patient with the reference RIS value.

The present inventors have found that if the RIS signature score from the test sample is higher than the reference RIS value, then it indicates that said subject is not likely to respond to a PD-1 antagonist (e.g. a PD-1 antibody such as nivolumab, as explained above) or is not likely to benefit from treatment with a PD-1 antagonist. The term “higher” as used herein refers to, for example at least 5% above, 10% above, 20% above, 30% above, 40% above, or 50% above the reference RIS score or more or for example at least 1 fold above, 2 fold above, 3 fold above, 4 fold above or 5 fold above the reference RIS score or more. On the other hand, if the RIS signature score from the test sample is lower than the reference RIS value, then it indicates that said subject is likely to respond to a PD-1 antagonist or is likely to benefit from treatment with a PD-1 antagonist. The term“lower” as used herein refers to, for example at least 5% below, 10% below, 20% below, 30% below, 40% below, or 50% below the reference RIS score or less or for example at least 1 fold below, 2 fold below, 3 fold below, 4 fold below or 5 fold below the reference RIS score or less.

Situation 2: 10 tumor samples from cancer patients

Step 1 : calculating the arithmetic mean of the expression level of each of the genes in the RIS signature in each of the tumor sample. For instance, if the RIS signature consists of 3 of the genes in Table 1 , or in a preferred embodiment Table 1A, e.g. gene 1 , gene 2 and gene 3, then the RIS signature score is the mean average of the expression levels for each of these genes. The mean average (arithmetic mean) can be calculated as follows: if in sample 1 , gene 1 has an expression level value of 1.5, and gene 1 has an expression level value of 1.8, and gene 3 has an expression level value of 1.2, then the RIS signature score is: 1.5 + 1.8+ 1.2 = 4.5 divided by 3 = 1.5 for tumor sample 1. The same can be repeated for tumor sample 2, 3, 4, and so on. For instance, if in sample 2, gene 1 has an expression level value of 1.1 , and gene 2 has an expression level value of 1.2, and gene 3 has an expression level value of 1.9, then the RIS signature score is: 1.1 + 1.2+ 1.9 = 4.2 divided by 3 = 1.4 for tumor sample 2. Step 2: Classifying the tumor samples into 1) a responder group and 2) a non responder group.

Step 3: calculating the arithmetic mean (mean average) of the RIS signature score for each tumor sample in each group (i.e. the responder group and the non-responder group). For instance, in a group of 10 tumor samples, the mean average of the RIS signature scores is calculated as follows:

Tumor sample 1 = 1.5

Tumor sample 2 = 1.4

Tumor sample 3 = 1.3

Tumor sample 4 = 1.6

Tumor sample 5 = 1.7

Tumor sample 6 = 1.1

The arithmetic average of the responder group is: 1.5 + 1.4 + 1.3 + 1.6 + 1 .7 + 1.1 = 8.6 divided by 6 = 1.43. Therefore, the reference RIS score for the responder group is 1.43

Non-responder group (n=4)

Tumor sample 7 = 1.2

Tumor sample 8 = 1.4

Tumor sample 9 = 1.5

Tumor sample 10 = 1.8

The arithmetic average is: 1.2 + 1.4 + 1.5 + 1.8 = 5.9 divided by 4 = 1.48. Therefore, the reference RIS score for the non-responder group is 1.48.

In the context of situation 2, the tumor samples used to calculate the reference RIS score for the responder group and the non-responder group are preferably of the same tumor type as the test tumor sample, e.g., a melanoma tumor sample. In the context of the present invention, the reference RIS score for the responder group and the reference RIS score for the non-responder group are used as cut off values or threshold values for determining whether a cancer subject is likely to respond to or benefit from treatment with a PD-1 antagonist (as explained above). Specifically, this is done by comparing the RIS signature score associated with a tumor sample (test sample) for said patient with the reference RIS score for the responder group and the reference RIS score for the non-responder group. In the context of situation 2, if the RIS signature score associated with the test tumor sample is closer to the reference RIS score of the responder group relative to the reference RIS score of the non responder group, then it indicates that said subject is likely to respond to a PD-1 antagonist or is likely to be benefit from treatment with a PD-1 antagonist. On the other hand, if the RIS signature score associated with the test tumor sample is closer to the reference RIS score of the non-responder group relative to the reference RIS score of the responder group, then it indicates that said subject is not likely to respond to a PD- 1 antagonist or is not likely to benefit from treatment with a PD-1 antagonist.

The term“closer” as used herein means“being the closest to a given number”. For instance, if the RIS signature score of a test sample is 1.5 and the reference RIS score of the responder group is 1.6 and the reference RIS score of the non-responder group is 1.1 , then the RIS signature score (i.e., 1.5) of the test sample is closer to the reference RIS score of the responder group (i.e., 1.6) relative to the reference RIS score of the non-responder group (i.e., 1.1 ). In that case, it indicates that the subject is likely to respond to a PD-1 antagonist or is likely to benefit from treatment with a PD-1 antagonist.

In the context of the present invention, the reference RIS score can be determined or calculated according to situation 1 or 2 above, without affecting the outcome, e.g., without affecting the predictive value of the methods as taught herein. In a preferred embodiment, the reference RIS score is calculated according to situation 1.

It is understood that the RIS signature score and the Reference RIS score may be calculated using other methods (other than arithmetic means, e.g. as in Tibshirani et al (2002), Proceedings of the National Academy of Sciences, Vol: 99 (10): pages 6567-6572) without changing the predicting power of the RIS gene signature, e.g. for predicting the response of a subject to treatment with a PD-1 antagonist (as taught herein).

In a second aspect, the present invention relates to a method for predicting the response of a subject suffering from a cancer to a treatment with a CTLA-4 antagonist or for stratifying a subject for treatment with a CTLA-4 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

(c) Measuring the RNA expression level in the sample of step (b) for each gene in a Stromal Immune Signal (SIS) gene signature;

(d) Calculating the arithmetic mean of the expression level for each of the genes in the SIS gene signature to generate a SIS signature score, wherein the SIS signature comprises at least two, preferably at least 5, 10, 15 or 22, of the genes in Table 2;

(e) Comparing the calculated SIS signature score of step (d) to a reference SIS score; and

(f) Predicting the response of said subject to a CTLA-4 antagonist or stratifying said subject for treatment with a CTLA-4 antagonist, wherein

-(i) if the calculated SIS score is higher than the SIS reference score, then it indicates that said patient is likely to benefit from treatment with the CLTA4 antagonist (responder) or is likely to respond to treatment with the CTLA-4 antagonist; or

-(ii) if the calculated SIS score is lower than the SIS reference score, then it indicates that said patient is not likely to benefit from treatment with the CLTA4 antagonist (non-responder) or is not likely to respond to treatment with the CLTA4 antagonist.

The term“Stromal Immune Signal” (abbreviated as“SIS”) signature as used herein refers to a biomarker, i.e. a gene signature as defined above or biomarker that can be used to predict the response (e.g., identify a responder or a non-responder) of a subject suffering from a cancer (e.g., melanoma) to a CTLA-4 antagonist (e.g., a CTLA-4 antibody such as ipilimumab or that can be used to stratify subjects for treatment with a CTLA-4 antagonist (e.g., to classify subjects as responders and non responders to CTLA-4 antagonist treatment so as to separate them into distinct subgroups) or that can be used to identify which cancer subjects or patients are most likely to achieve or experience a clinical benefit from treatment with a CTLA-4 antagonist or that can be used to decide whether a patient can be treated with a CTLA- 4 antagonist. In the context of the present invention, the SIS signature comprises at least 2 of the genes in Table 2 below, for instance at least 2, 3, 4, 5, 6, 7, 8, 9, 10, 1 1 , 12, 13, 14, 15, 16, 17, 18, 19, 20, 30, 40, 50, 60, 70, 80, 90, 100, 1 10, 120, 130, 140, 150, 160, 170, 180, 190, 200, 210, 220, 230, 240, 250, 260, 270, 280, 290, 300, 310, 320, 330, 340, 350, 360, or 361 of the genes in Table 2. It was shown, see the examples, that with a signature comprising randomly selected genes from the genes mentioned in Table 2, the response of a patient to the treatment can be predicted with high accuracy. For example, in case of selecting 5 random genes from the set presented in Table 2 (verified in 1000 different iterations) and testing the number of times the classification is identical as to the full set of genes in Table 2, it was shown that in nearly all case, i.e. 95% of cases with random sets of 5 genes the classification is identical to the set of all genes. In a preferred embodiment, the SIS signature comprises at least 5, 10, 15 or 22 of the genes in Table 2. In a preferred embodiment the at least 5 genes or at least 10 genes are selected from the combination of 5 genes or combination of 10 genes shown in Figure 6 or 7 (V1-V5 or V1-V10 indicates the different genes in the set). For instance, the skilled person may select 5 or 10 of the genes in Table 2 to determine the SIS signature score (as taught herein), which in turn will be useful to, e.g. establish or determine whether a cancer subject will respond to or will benefit from treatment with a CTLA-4 antagonist (e.g., a CTL4A antibody such as ipilimumab). Table 2. SIS Gene Signature

The term“SIS signature score” as used herein refers to a numerical value that is obtained by calculating the arithmetic mean of the expression level of each of the genes in the SIS signature for one tumor sample. For instance, if the SIS signature consists of 3 of the genes in Table 2, e.g., gene 1 , gene 2 and gene 3 or gene 10, gene 21 , and gene 40, etc., then the SIS signature score is the mean average of the expression levels for each of these genes. The mean average (arithmetic mean) can be calculated as follows: for instance, if gene 1 has an expression level value of 1.5, and gene 2 has an expression level value of 1.8, and gene 3 has an expression level value of 1.2, then the SIS signature score is: 1.5 + 1 .8+ 1.2 = 4.5 divided by 3 = 1.5 for that particular tumor sample.

It is understood that the expression levels (RNA levels) of the genes in the SIS signature can be measured by any suitable means, e.g. PCR analysis, in situ hybridization, next generation sequencing, nanostring and the like, and may be expressed as optical density values, amounts (e.g., micrograms), and the like. The method for measuring gene expression (RNA levels) and expressing the data (e.g. optical density data, amounts, etc.) does not influence the way the SIS signature score is calculated, as shown above.

Persons skilled in the art are aware of several methods useful for detecting and quantifying the level of RNA transcripts within a sample. Quantitative detection methods include, but are not limited to, arrays (i.e., microarrays), quantitative real time PCR (RT-PCR), multiplex assays, nuclease protection assays, and Northern blot analyses. Generally, such methods employ labeled probes that are complimentary to a portion of each transcript to be detected. Probes for use in these methods can be readily designed based on the known sequences of the genes and the transcripts expressed thereby. In certain embodiments, the probes are designed to hybridize to each of the gene signature transcripts identified in Table 2. Suitable labels for the probes are well-known and include, e.g., fluorescent, chemiluminescent and radioactive labels.

The term“reference SIS score” as used herein to a numerical value that is derived from a suitable control, such as a group of (different) tumor samples (e.g., 10, 20, 30, 50, 100, 200, 500 or more tumor samples). The reference SIS score may be obtained as follows e.g. according to situation 1 or 2 below.

Situation 1 : 10 tumor samples from cancer patients

Step 1 : calculating the arithmetic mean of the expression level of each of the genes in the SIS signature in each of the tumor sample. For instance, if the SIS signature consists of 3 of the genes in Table 2, e.g. gene 1 , gene 2 and gene 3, than the SIS signature score is the mean average of the expression levels for each of these genes. The mean average (arithmetic mean) can be calculated as follows: if in sample 1 , gene 1 has an expression level value of 1.5, and gene 2 has an expression level value of 1.8, and gene 3 has an expression level value of 1.2, then the SIS signature score is: 1.5 + 1.8+ 1.2 = 4.5 divided by 3 = 1.5 for tumor sample 1. The same can be repeated for tumor sample 2, 3, 4, and so on. For instance, if in sample 2, gene 1 has an expression level value of 1.1 , and gene 2 has an expression level value of 1.2, and gene 3 has an expression level value of 1.9, then the SIS signature score is: 1.1 + 1.2+ 1.9 = 4.2 divided by 3 = 1.4 for tumor sample 2.

Step 2: calculating the arithmetic mean (mean average) of the SIS signature score for each tumor sample in the group. For instance, in a group of 10 tumor samples, the mean average of the SIS signature scores is calculated as follows:

Tumor sample 1 = 1.5 Tumor sample 2 = 1.4

Tumor sample 3 = 1.3

Tumor sample 4 = 1.6

Tumor sample 5 = 1.7

Tumor sample 6 = 1.1

Tumor sample 7 = 1.2

Tumor sample 8 = 1.4

Tumor sample 9 = 1.5

Tumor sample 10 = 1.8

The arithmetic average is: 1.5 + 1.4 + 1.3 + 1.6 + 1.7 + 1.1 + 1.2 + 1.4 + 1.5 + 1.8 = 14.5 divided by 10 = 1.45. Therefore, the reference SIS score is 1.45. In the context of situation 1 , the tumor samples used to calculate the reference SIS score are preferably of the same tumor type as the test tumor sample, e.g., a melanoma tumor sample. Further, the tumor samples used to calculate the reference SIS score are preferably obtained from cancer patients who have been treated with a CTLA-4 antagonist. In the context of situation 1 , it is not essential to know which patients responded or not to the CTLA-4 antagonist.

In the context of the present invention, the reference SIS score is used as a cut off value or a threshold value for determining whether a cancer subject is likely to respond to or benefit from treatment with a CTLA-4 antagonist (e.g. a CTLA-4 antibody such as ipilimumab, as explained above). Specifically, this is done by comparing the SIS signature score associated with a tumor sample to be tested (referred to as a test sample) for said patient with the reference SIS value.

The present inventors have found that if the SIS signature score from the test sample is higher than the reference SIS value, then it indicates that said subject is likely to respond to treatment with a CTLA-4 antagonist (e.g. a CTLA-4 antibody such as ipilimumab, as explained above) or is likely to benefit from treatment with a CTLA-4 antagonist. The term“higher” as used herein refers to, for example at least 5% above, 10% above, 20% above, 30% above, 40% above, or 50% above the reference SIS score or more or for example at least 1 fold above, 2 folds above, 3 folds above, 4 folds above or 5 folds above the reference SIS score or more. On the other hand, if the SIS signature score from the test sample is lower than the reference SIS value, then it indicates that said subject is not likely to respond to treatment with a CTLA-4 antagonist or is not likely to benefit from treatment with a CTLA-4 antagonist. The term “lower” as used herein refers to, for example at least 5% below, 10% below, 20% below, 30% below, 40% below, or 50% below the reference SIS score or less or for example at least 1 fold below, 2 fold below, 3 fold below, 4 fold below or 5 fold below the reference SIS score or less.

Situation 2: 10 tumor samples from cancer patients

Step 1 : calculating the arithmetic mean of the expression level of each of the genes in the SIS signature in each of the tumor sample. For instance, if the SIS signature consists of 3 of the genes in Table 2, e.g. gene 1 , gene 2 and gene 3, then the SIS signature score is the mean average of the expression levels for each of these genes. The mean average (arithmetic mean) can be calculated as follows: if in sample 1 , gene 1 has an expression level value of 1.5, and gene 1 has an expression level value of 1.8, and gene 3 has an expression level value of 1.2, then the SIS signature score is: 1.5 + 1.8+ 1.2 = 4.5 divided by 3 = 1.5 for tumor sample 1. The same can be repeated for tumor sample 2, 3, 4, and so on. For instance, if in sample 2, gene 1 has an expression level value of 1.1 , and gene 2 has an expression level value of 1.2, and gene 3 has an expression level value of 1.9, then the SIS signature score is: 1.1 + 1.2+ 1.9 = 4.2 divided by 3 = 1.4 for tumor sample 2.

Step 2: Classifying the tumor samples into 1) a responder group and 2) a non responder group.

Step 3: calculating the arithmetic mean (mean average) of the SIS signature score for each tumor sample in each group (i.e. the responder group and the non-responder group). For instance, in a group of 10 tumor samples, the mean average of the SIS signature scores is calculated as follows: Responder group (n=6):

Tumor sample 1 = 1.5

Tumor sample 2 = 1.4

Tumor sample 3 = 1.3

Tumor sample 4 = 1.6

Tumor sample 5 = 1.7

Tumor sample 6 = 1.1

The arithmetic average of the responder group is: 1.5 + 1.4 + 1.3 + 1.6 + 1.7 + 1.1 = 8.6 divided by 6 = 1.43. Therefore, the reference SIS score for the responder group is 1.43

Non-responder group

Tumor sample 7 = 1.2

Tumor sample 8 = 1.4

Tumor sample 9 = 1.5

Tumor sample 10 = 1.8

The arithmetic average is: 1.2 + 1.4 + 1.5 + 1.8 = 5.9 divided by 4 = 1.48. Therefore, the reference SIS score for the non-responder group is 1.48.

In the context of situation 2, the tumor samples used to calculate the reference SIS score for the responder group and the non-responder group are preferably of the same tumor type as the test tumor sample, e.g. a melanoma tumor sample.

In the context of the present invention, the reference SIS score for the responder group and the reference SIS score for the non-responder group are used as cut off values or threshold values for determining whether a cancer subject is likely to respond to or benefit from treatment with a CTLA-4 antagonist (as explained above). Specifically, this is done by comparing the SIS signature score associated with a tumor sample (test sample) for said patient with the reference SIS score for the responder group and the reference SIS score for the non-responder group. In the context of situation 2, if the SIS signature score associated with the test tumor sample is closer to the reference SIS score of the responder group relative to the reference SIS score of the non-responder group, then it indicates that said subject is likely to respond to a CTLA- 4 antagonist or is likely to be benefit from treatment with a CTLA-4 antagonist. On the other hand, if the SIS signature score associated with the test tumor sample is closer to the reference SIS score of the non-responder group relative to the reference SIS score of the responder group, then it indicates that said subject is not likely to respond to a CTLA-4 antagonist or is not likely to benefit from treatment with a CTLA-4 antagonist.

The term“closer” as used herein means“being the closest to a given number”. For instance, if the SIS signature score of a test sample is 1.5 and the reference SIS score of the responder group is 1.6 and the reference SIS score of the non-responder group is 1.1 , then the SIS signature score (i.e. 1.5) of the test sample is closer to the reference SIS score of the responder group (i.e. 1.6) relative to the reference SIS score of the non-responder group (i.e. 1.1 ). In that case, it indicates that the subject is likely to respond to a CTLA-4 antagonist or is likely to benefit from treatment with a CTLA-4 antagonist.

In the context of the present invention, the reference SIS score can be determined or calculated according to situation 1 or 2 above, without affecting the outcome, e.g. without affecting the predictive value of the methods as taught herein. In a preferred embodiment, the reference SIS score is calculated according to situation 1.

It is understood that the SIS signature score and the Reference SIS score may be calculated using other methods (other than arithmetic means, e.g. as in Tibshirani et al (2002), Proceedings of the National Academy of Sciences, Vol: 99 (10): pages 6567-6572)) without changing the predicting power of the RIS gene signature, e.g. for predicting the response of a subject to treatment with a CTLA-4 antagonist (as taught herein).

In a further aspect, the present invention relates to a method for predicting the response of a subject suffering from a cancer to a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist or for stratifying a subject for treatment with a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

(c) Measuring the RNA expression level in the sample of step (b) for each gene in Response Immune Signal (RIS) gene signature;

(d) Measuring the RNA expression level in the sample of step (b) for each gene in a Stromal Immune Signal (SIS) gene signature;

(f) Calculating the arithmetic mean of the expression level for each of the genes in the RIS gene signature of step (c) to generate a RIS signature score, wherein the RIS signature comprises at least two of the genes in Table 1 , preferably at least two of the genes in Table 1A;

(g) Calculating the arithmetic mean of the expression level for each of the genes in the SIS gene signature of step (d) to generate a SIS signature score, wherein the SIS signature comprises at least two, preferably at least 5, 10, 15, or 22, of the genes in Table 2; and

(h) Predicting the response of said subject to the combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist or stratifying said subject for treatment with a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist wherein:

(i) if the calculated RIS score is lower than the RIS reference score and if the calculated SIS score is higher than the SIS reference score, then it indicates that said patient is likely to benefit from treatment with the combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist (responder) or is likely to respond to treatment with the combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist; or

(ii) if the calculated RIS score is higher than the RIS reference score and if the calculated SIS score is lower than the SIS reference score, then it indicates that said patient is not likely to benefit from treatment with the combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist (non-responder) or is not likely to respond to treatment with the combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist. The advantages, methodology and terminology described above apply to the embodiments descried herein.

Method of treatment

In a further aspect, the present invention relates to a method for treating a subject suffering from a cancer with a PD-1 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

(d) Calculating the arithmetic mean of the expression level for each of the genes in the RIS gene signature to generate a RIS signature score, wherein the RIS signature comprises at least two of the genes in Table 1 , preferably at least two of the genes in Table 1A;

(f) Treating said subject with a PD-1 antagonist if the calculated RIS score is lower than the RIS reference score.

The advantages, methodology and terminology described above apply to the embodiments descried herein.

In a further aspect, the present invention relates to a method for treating a subject suffering from a cancer with a CTLA-4 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

(d) Calculating the arithmetic mean of the expression level for each of the genes in the SIS gene signature to generate a SIS signature score, wherein the SIS signature comprises at least two, preferably at least 5, 10, 15, or 22, of the genes in Table 2;

(f) Treating said subject with a CTLA-4 antagonist if the calculated SIS score is higher than the SIS reference score.

In a further aspect, the present invention relates to a method for treating a subject suffering from a cancer with a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

(f) Calculating the arithmetic mean of the expression level for each of the genes in the RIS gene signature o step (c) to generate a RIS signature score, wherein the RIS signature comprises at least two of the genes in Table 1 , preferably at least two of the genes in Table 1A;

(g) Calculating the arithmetic mean of the expression level for each of the genes in the SIS gene signature of step (d) to generate a SIS signature score, wherein the SIS signature comprises at least two, preferably at least 5, 10, 15, or 22, of the genes in; and

(h) Treating said subject with a combination therapy consisting of a PD- 1 antagonist and a CTLA-4 antagonist if the calculated RIS score is lower than the RIS reference score and if the calculated SIS score is higher than the SIS reference score. The advantages, methodology and terminology described above apply to the embodiments descried herein.

Method for detecting biomarkers

In a further aspect, the present invention relates to a method for testing a tumor from a subject for the presence or absence of a biomarker that predicts the response of said subject to treatment with a PD-1 antagonist, said method comprises:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

(f) Predicting the response of said subject to a PD-1 antagonist, wherein: -(i) if the calculated RIS score is higher than the RIS reference score, then it indicates that said patient is not likely to benefit from treatment with the PD-1 antagonist (non-responder) or is not likely to respond to treatment with the PD-1 antagonist; or

In a further aspect, the present invention related to a method for testing a tumor from a subject for the presence or absence of a biomarker that predicts the response of said subject to treatment with a CTLA-4 antagonist, said method comprises: (a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

(f) Predicting the response of said subject to a CTLA-4 antagonist, wherein

-(i) if the calculated SIS score is higher than the SIS reference score, then it indicates that said patient is likely to benefit from treatment with the CTLA-4 antagonist (responder) or is likely to respond to treatment with the CTLA-4 antagonist; or

-(ii) if the calculated SIS score is lower than the SIS reference score, then it indicates that said patient is not likely to benefit from treatment with the CLTA4 antagonist (non-responder) or is not likely to respond to treatment with the CTLA-4 antagonist.

In a further aspect, the present invention related to a method for testing a tumor from a subject for the presence or absence of a biomarker that predicts the response of said subject to a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

(c) Measuring the RNA expression level in the sample of step (b) for each gene in Response Immune Signal (RIS) gene signature; (d) Measuring the RNA expression level in the sample of step (b) for each gene in a Stromal Immune Signal (SIS) gene signature;

(h) Predicting the response of said subject to the combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist wherein:

(ii) if the calculated RIS score is higher than the RIS reference score and if the calculated SIS score is lower than the SIS reference score, then it indicates that said patient is not likely to benefit from treatment with the combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist (non-responder) or is not likely to respond to treatment with the combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist.

In an embodiment relating to the methods as thought herein, the cancer is melanoma.

In an embodiment relating to the methods as thought herein, the PD-1 antagonist is an antibody. In an embodiment relating to the methods as thought herein, the PD-1 antibody may be selected from the group of nivolumab, pembrolizumab, pidilizumab cemiplimab, PDR001 , AMP-224, AMP-514, preferably nivolumab.

In a preferred embodiment, the PD-1 antibody is nivolumab.

In an embodiment relating to the methods as thought herein, the CTLA-4 antagonist is an antibody.

In an embodiment relating to the methods as thought herein, the CTLA-4 antibody is selected from the group of ipilimumab, and tremelimumab, preferably ipilimumab.

In a preferred embodiment, the CTLA-4 antibody is ipilimumab.

In an embodiment relating to the methods as thought herein, the PD-1 antagonist is an antibody and the CTLA-4 antagonist is an antibody.

In an embodiment relating to the methods as thought herein, the PD-1 antibody is selected from the group of nivolumab, pembrolizumab, pidilizumab cemiplimab, PDR001 , AMP-224, AMP-514, preferably nivolumab, and the CTLA-4 antibody is selected from the group of ipilimumab, and tremelimumab, preferably ipilimumab.

In a preferred embodiment relating to the combination therapy, the PD-1 antibody is nivolumab and the CLTA4 antibody is ipilimumab.

In an embodiment relating to the methods as thought herein, the subject is a human subject.

Kit

In a further aspect, the present invention relates to a kit for assaying a tumor sample from a subject to determine a RIS gene signature score and/or a SIS gene signature score for said tumor sample, wherein the kit comprises a set of probes for detecting the expression of each gene in a RIS gene signature and/or a SIS gene signature, wherein the RIS gene signature comprises at least two of the genes of Table 1 , preferably at least two of the genes in Table 1A,and wherein the SIS gene signature comprises at least two, preferably at least 5, 10, 15, or 22, of the genes in Table 2.

The term “probe” as used herein means an oligonucleotide that is capable of specifically hybridizing under stringent hybridization conditions to a transcript expressed by a gene of interest listed in Table 1 , preferably Table 1A, and/or Table 2, and in some preferred embodiments, specifically hybridizes under stringent hybridization conditions to the particular transcript listed in Table 1 , preferably Table A1 , and/or Table 2 for the gene of interest (i.e. the genes of the RIS signature and SIS signature.

Uses

In a further aspect, the present invention relates to the use of the RIS gene signature as taught herein for predicting the response of a subject suffering from a cancer to a PD-1 antagonist or for stratifying patients for treatment with a PD-1 antagonist; wherein the RIS gene signature comprises at least two of the genes in Table 1 , preferably at least two of the genes in Table 1A.

In a further aspect, the present invention relates to the use of the SIS gene signature as taught herein for predicting the response of a subject suffering from a cancer to a CTLA-4 antagonist or for stratifying patients for treatment with a CTLA-4 antagonist; wherein the SIS gene signature comprises at least two, preferably at least 5, 10, 15, or 22, of the genes in Table 2.

In a further aspect, the present invention relates to the use of the RIS signature as taught herein and the SIS gene signature as taught herein for predicting the response of a subject suffering from a cancer to a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist or for stratifying a subject for treatment with a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist; wherein the RIS gene signature comprises at least two of the genes in Table 1 , preferably at least two of the genes in Table 1A, and wherein the SIS gene signature comprises at least two, preferably at least 5, 10, 15, or 22, of the genes in Table 2. The advantages, methodology and terminology described above apply to the embodiments descried herein.

Medical uses

In a further aspect, the present invention relates to a PD-1 antagonist for use in the treatment of cancer, wherein treatment comprises a method for predicting the response of a subject suffering from a cancer to a PD-1 antagonist or a method for stratifying a subject for treatment with a PD-1 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

(d) Calculating the arithmetic mean of the expression level for each of the genes in the RIS gene signature to generate a RIS signature score, wherein the RIS signature comprises at least two of the genes in T able 1 , preferably at least two of the genes in Table 1 A;

(f) Predicting the response of said subject to a PD-1 antagonist or stratifying said subject for treatment with a PD-1 antagonist, wherein:

In a further aspect, the present invention relates to a CTLA-4 antagonist for use in the treatment of cancer, wherein the treatment comprises a method for predicting the response of a subject suffering from a cancer to a CTLA-4 antagonist or a method for stratifying a subject for treatment with a CTLA-4 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

In a further aspect, the combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist for use in the treatment of cancer, wherein the treatment comprises a method for predicting the response of a subject suffering from a cancer to a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist or a method for stratifying a subject for treatment with a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a); (c) Measuring the RNA expression level in the sample of step (b) for each gene in Response Immune Signal (RIS) gene signature;

The advantages, methodology and terminology described above apply to the embodiments described herein. Also disclosed is a device or kit suitable for measuring the SIS and/or RIS signature in the methods as disclosed herein, for example the kit or device is suitable for measuring a (preselected) gene signature with RT-PCR, with nanostring, or with MiSeq technology.

BRIEF DESCRIPTION OF THE FIGURES

Figure 1. Melanoma PDX characteristics

A. Pie chart illustrating the genotypes of the PDX (BRAF^V600E, NRAS^Q61 or BRAF^WT/NRAS^WT) and the treatment that the patients received (pre-vemurafenib treatment, post-vemurafenib treatment, post-MAPKi, on-vemurafenib treatment or intrinsic resistant). B. Number of mapped read pairs per PDX sample. Blue represents mapped human reads, red represents mapped mouse reads. C. Cartoon illustrating the difference between analyzing patient tumors, where the stroma and tumor cells are mixed and both of human origin, and PDX, where the tumor cells are of human origin and the stromal cells are of mouse origin.

Figure 2. Stromal-derived gene signatures largely determine subgroup classification

A. Cluster analysis and associated heatmap based on all melanoma (SKCM) samples from the TCGA dataset and the three TCGA gene expression profiles (“MITF-low”, “Keratin” and“Immune”, indicated in yellow, green and red, respectively). B. Cluster analysis and associated heatmap based on 95 PDX samples and the three TCGA gene sets. C. Scatter plots of the number of normalized read counts in patient data and the PDX data. Genes in the “MITF-low”, “Keratin” and “Immune” profiles are super imposed. Box plots indicate the average expression for the genes in the profiles in patient data and PDX (*** = p-value <0.001).

Figure 3. Gene signatures based on tumor-intrinsic gene expression can be identified in the PDX.

A. Heatmap of 95 PDX samples for the PCA derived MITF-high gene expression signature. B. Gene sets significantly enriched in the MITF-low samples (EMT) and enriched in the MITF-high samples (OXPHOS) as revealed by Gene Set Enrichment Analysis. C. Scatter plot of AXL and MITF expression in PDX samples ordered according to the MITF-high gene expression signature. D. Immunoblotting for phenotype switching genes, like AXL, MITF and Melan-A, in the order of the MITF- high gene expression signature. Five samples of the left part of the heatmap, five samples of the right part and five samples in the middle of the heatmap were loaded. Vinculin was used as a loading control. E. Heatmap of 95 PDX samples for the PCA derived RIS gene expression signature. F. Gene sets significantly enriched in the RIS high samples as revealed by Gene Set Enrichment Analysis. G. Heatmap of 95 PDX samples for the PCA derived Mitotic-high gene expression signature. H. Gene sets significantly enriched in the Mitotic-high samples as revealed by Gene Set Enrichment Analysis. I. Growth speed, measured in days till a tumor reached 250mm³. J. Six representative PDX samples ordered according to the Mitotic gene expression signature stained for Ki67. K. Ki67 staining of PDX samples high and low for the Mitotic gene signature (n=10 per side). Scale bars indicate 400 and 40 pm. The stainings were scored for percentage of staining by a pathologist. Graph represents the scorings data. An unpaired t-test was performed to analyze statistical significance. ** = p < 0.01.

Figure 4. Tumor intrinsic and stromal signatures predict outcome to immune checkpoint blockade

A. Gene Set Enrichment Analysis (GSEA) on anti-PD-1 -treated patient samples for the tumor intrinsic RIS signature and the Stromal Immune signature B. Heatmap of the anti-PD-1-treated patient samples for the RIS signature. Bar below the heatmap represents the average expression of the RIS signature. C. Gene Set Enrichment Analysis (GSEA) on anti-CTLA-4-treated patient samples for the tumor intrinsic RIS signature and the Stromal Immune signature. D. Heatmap of the anti-CTLA-4-treated patient samples for the Stromal Immune signature. Bar below the heatmap represents the average expression of the Stromal Immune signature. E. Heatmap of the patients from the OpACIN trial, both for the RIS signature and the Stromal Immune signature. Samples are ordered according to clinical outcome and average gene expression of the signature. Relapsed and non-relapsed samples are indicated with red and green respectively. G. Heatmap of the patients from the OpACIN trial, both for the RIS signature and the Stromal Immune signature. Samples are ordered according to the average gene expression of the signature. Relapsed and non-relapsed samples are indicated with red and green respectively. H. Barplot and heatmap of the ratio Stromal Immune (SI) and RIS signature. Relapsed patients are indicated in red, non-relapsed in green.

Figure 5: Patient stratification based on stromal and tumor-cell intrinsic gene signatures

A. Heatmap of the IFNg related signature disclosed in Avers et al (Ayers, M. et al. IFN- g-related mRNA profile predicts clinical response to PD-1 blockade. J. Clin. Invest. 127, 2930-2940 (2017). Relapsed patients are indicated in red, non-relapsed in green. B. Scatterplot of the ratio SIS/RIS and the IFNg-related signature. C. The tumor cell-intrinsic RIS and stromal SIS signature used in conjunction on the samples from the OpACIN trial (NCT02437279). Circles indicate patients who did not relapse within 24 months after start of combination ICB therapy, the stars indicate patients who did relapse within 24 months after start of combination ICB. The three quadrants (left column and bottom right column) indicate the RIS and SIS signature values for samples that are predicted as responders to either anti-CTLA-4, anti-PD-1 , or both (anti-CTLA-4 (top left), anti-PD1 (bottom right) or both (bottom left)). The right column top quadrant indicates the RIS and SIS signature values for samples that are predicted as non-responders to anti-CTLA-4 as well as anti-PD-1. D. Kaplan-Meier plot comparing recurrence free survival (RFS) for patients predicted as responders (blue) to patients predicted as non-responders (orange) for the OpACIN trial.

Figure 6 and Figure 7: Different sets of genes (5 genes in Figure 6 and 10 genes in Figure 7) of the SIS signature.

Figure 8 and Figure 9: Analysis to establish minimal number of random genes from the genes in Table 2 (SIS) for prediction.

It will be understood that all details, embodiments and preferences discussed with respect to one aspect of embodiment of the invention is likewise applicable to any other aspect or embodiment of the invention and that there is therefore not need to detail all such details, embodiments and preferences for all aspect separately. Having now generally described the invention, the same will be more readily understood through reference to the following examples which is provided by way of illustration and is not intended to be limiting of the present invention. Further aspects and embodiments will be apparent to those skilled in the art.

EXAMPLES

Example 1 : Passaging patient tumor material in immune deficient mice induces complete replacement of human stroma by mouse stroma within one passage

We have previously generated a large PDX panel derived from either excisions or fine- needle biopsies of stage IV metastatic melanoma (Kemper et al., (2016), CellReports 1- 16). Patient tumors have been coded, e.g. M001. When passaged into immunodeficient NOD.Cg-Prkdcscid N2rgtm1 Wjl/SzJ (NSG) mice, the addition of X1 was added to the nomenclature (e.g. M001.X1 , data not shown). In total, we now have obtained a total of 95 PDX, comprising of BRAFWT/NRASWT, NRASQ61 or BRAFV600E mutated melanomas. The BRAFV600E PDX were derived either before start of treatment with vemurafenib or resistance was acquired to this BRAFV600E inhibitor (see Figure 1A).

The complete panel of 95 PDX samples were analyzed by paired-end RNA- sequencing. Contaminating sequence reads of mouse origin were separated from the human sequence reads using XenofilteR (Kluin et al. (2018) BMC Bioinformatics, 19(1 ): 366). We observed a wide range in the percentage of mouse reads per PDX (1 - 36%) with an average of 6.1 % (see Figure 1 B) which suggested that these PDX do indeed contain some murine stromal infiltration. Therefore, the presence of the stromal compartment in the PDX was studied by immunohistochemistry (IHC) by staining PDX material for either the human stromal marker vimentin or mouse stromal marker podoplanin (data not shown). While human vimentin was expressed in the tumor cells, no expression was found in the stromal compartment of the PDX. Mouse podoplanin expression however was solely expressed in the stromal areas (data not shown), suggesting that human stroma was completely replaced by mouse stroma within one passage in mice. The ability to separate mouse sequence reads from sequence reads of human origin, presented the opportunity to investigate tumor gene expression without the stromal compartment. This is impossible when RNA expression profiles of patient tumors are analyzed, due to the fact that a part of the gene expression is derived from the stromal compartment (see Figure 1C).

Example 2: Previously published melanoma transcriptomic gene signatures are partially based on stromal infiltration

As we can now study the gene expression profiles of the tumor-intrinsic (and not stromal) genes in our PDX samples, we questioned to what extent previously published gene signatures, developed by using expression data from patient samples, could segregate our PDX tumor-intrinsic gene expression data into subgroups. To this end, we performed hierarchical clustering on the melanoma TCGA dataset (SKin Cutaneous Melanoma: SKCM) using the three melanoma gene expression signatures developed by the Cancer Genome Atlas Network (Akbani, R et al., (2015), Cell 161 , 1681-1696).

As expected, this provided the three different subgroups previously identified by the Cancer Genome Atlas Network, (see Figure 2A). When the same gene signatures were applied to our PDX gene expression data, only expression of the genes that identify the MITF-low subgroup and expression of a limited number of genes that identify the keratin subgroup were detected (see Figure 2B), suggesting that a large number of genes that determine the keratin and the immune subgroups are stromal-derived.

To corroborate these findings, we plotted the average normalized read count for each gene expressed in our PDX samples and compared these to the average normalized read count for each gene of a second set of patient samples (Hugo et al., (2015), Cell 1-14). Genes that classify the MITF-low subgroup were expressed in both PDX and patient samples, indicating that these genes are expressed by tumor cells (see Figure 2C, left panel). Genes that determined the immune and keratin subgroups were significantly lower expressed in PDX compared to patient samples, suggesting that these genes originate primarily from stromal cells (see Figure 2C, middle and right panel). Next, we analyzed these potential stromal genes further by selecting the genes that were solely expressed in the patient samples but not or very lowly expressed in the PDX (n=767; see Figure 1 D-E). First, we performed a gene ontology analysis on this gene list, which revealed pathways involved in immune response and in keratinocyte signaling (data not shown), supporting the potential stromal origin of these genes. Second, we checked if these genes were indeed expressed in stromal cell types using previously published melanoma single cell-sequencing data (Tirosh et al., (2016), Science 352 189-196). Indeed, genes that were only found in patient samples but not in PDX, were mainly expressed in stromal cell types, like T-cells, B-cells, endothelial cells, NK-cells and cancer-associated fibroblasts (data not shown). Therefore, we refer to these genes as‘stromal-derived gene set’.

Example 3: Stromal-derived gene signatures determine subgroup classification

Next, we applied the above generated stromal-derived gene set to the SKCM TCGA samples in order to segregate these melanoma samples into subgroups (data not shown). Performing cluster analysis based only on the stromal-derived gene set, we were able to discriminate three subgroups, namely immune^hi9h, keratin^high and the immune^lowkeratin^low group. Surprisingly, the segregation based on the stromal-derived gene set largely overlapped with the subgroups identified using the TCGA gene expression signatures (data not shown). This finding illustrated that the classification of melanoma samples into subgroups is mainly based on stromal-derived gene expression and independent of tumor cell intrinsic gene expression. Of note, keratinocytes are predominantly present in the skin, which correlates with the observation that most samples in the keratin^high subgroup are primary tumors (data not shown).

In addition to the classification of SKCM TCGA samples, we applied the stromal- derived gene signature to a different cohort, which contained 57 stage IV melanomas (Jonsson et al., (2010), Clin. Cancer Res. 16 3356-3367). Again, we were able to segregate these melanomas into the immune^hi9h, keratin^high and the immune^lowkeratin^low subgroups (data not shown). This classification based on the stromal-derived genes highly overlaps with the classification from the original publication based on hierarchical clustering based on global gene expression data (Jonsson et al. , (2010), Clin. Cancer Res. 16 3356-3367). These results furthermore corroborate the finding that classification of patient samples based on gene expression is highly influenced by stromal-derived genes.

Example 4: Gene expression signatures based on tumor-intrinsic gene expression can be identified in the PDX

Previously published gene expression signatures to determine subgroups in melanoma seem to be dependent largely on gene expression from stromal cells. To generated gene expression profiles specifically for tumor-cell intrinsic genes, we used the human component of our PDX gene expression data. To that end, we performed a principle component analysis (PCA) on these human PDX gene expression data, resulting in three Gene expression Profiles (GP1 , GP2 and GP3) (data not shown).

We performed Gene Set Enrichment Analysis (GSEA) on the GP1 ordered PDX samples (see Figure 3A) which demonstrated enrichment of genes involved in “Epithelial to Mesenchimal Transition (EMT)”,“KRAS signaling”,“UV response” and “Angiogenesis” in the PDX samples with a low GP1 signature score and enrichment of genes involved on“Oxidative Phosphorylation” in the PDX samples with a high GP1 signature score (see Figure 3B).

Enrichment of these gene sets are a strong indication of phenotype switching (Boshuizen, J., et al. (2018), Nat. Med. 24, 203-212.). Indeed, expression of MITF, a commonly used marker for phenotype switching, correlates with the GP1 signature score (R=0.635) whereas AXL is anti-correlated with the GP1 signature score (R=- 0.639, see Figure 3C).

We validated the expression of AXL, MITF and downstream target Melan-A in PDX with different GP1 signature scores by immunoblotting. Expression of AXL was mainly found in the samples with a low GP1 signature score, whereas MITF and downstream Melan-A were mainly expressed in the PDX with a high GP1 signature score. The PDX samples with an intermediate GP1 score expressed both MITF and AXL, illustrating that this group is quite heterogeneous (as shown previously for the Hoek signature(Hoek et al., (2006), Pigment Cell Res. 19 290-302), see Figure 3D). Furthermore, ordering of the PDX samples based on GP1 was highly correlated to the ordering based on the Hoek signature (R=0.81). From now on, we refer to this gene expression signature as‘MITF^high’.

Next, we performed GSEA on the GP2-ordered PDX samples (see Figure 3E), which revealed enrichment of “TGFb signaling”,“IL2/STAT5 signaling” and“TNF signaling via_NFKB” in the PDX with a high GP2 signature score (see Figure 4F). These enriched pathways suggest a tumor-cell intrinsic inflammatory response in a subset of the PDX samples. To further test this inflammatory response like signal we tested the ADDITIONAL ANALYSIS. We refer from now on to this gene expression signature as the‘Chronic Inflammation’ signature (also referred herein as the Response Immune Signal (RIS)).

Lastly, we performed GSEA on the GP3 ordered PDX samples (Figure 3G), which demonstrated enrichment for the gene sets“Spermatogenesis”,“G2M checkpoint” and “Mitotic spindle” in the PDX with a high GP3 signature score (see Figure 3H). Enrichment of these gene sets suggested a higher proliferation rate of the PDX with a high GP3 signature score. Indeed, the GP3 signature score was correlated with the growth speed of these PDX (R=0.28; see Figure 3I) as measured in the days until a PDX reached the tumor size of 250mm². Also, we validated this correlation of GP3 signature score to increased growth speed by performing Ki67 staining on PDX with different GP3 signature scores. Ki67 expression of PDX with a high GP3 signature score was indeed significantly higher compared to the PDX with a low GP3 signature score (see Figure 3J-K), confirming that the GP3 signature was associated with higher growth speed. This gene signature will from now on be referred to as the‘Mitotic’ signature.

Thus, we identified three tumor-intrinsic gene expression signatures that could be linked to the following biological processes; GP1 to phenotype switching; GP2 to a tumor intrinsic inflammatory response (RIS); and GP3 to proliferation.

Example 5: PDX-derived cell lines maintain tumor-intrinsic gene expression signatures After generating the PDX, we have established cell lines after the first passage of these PDX in mice, resulting in the addition. CL to the nomenclature, e.g. M001.X1.CL (Kemper, K., (2016), Cell Reports 1-16). In total, 22 low passage cell lines (up to ~12 passages) were generated, of which RNA was send for paired-end RNA-sequencing. These PDX-derived cell lines were completely devoid of mouse reads, as they were preselected to not contain any murine cells.

Next, we analyzed if our PDX-derived tumor-intrinsic gene signatures were also represented and even maintained, in our PDX-derived cell lines. To that end, we compared the tumor intrinsic gene expression signatures (GP1-3) between each PDX and its matching PDX-derived cell line. The gene expression pattern for the MITF^hi9h signature was largely maintained in the PDX cell lines (data not shown). We validated the similarity between the PDX and the corresponding cell lines by analyzing the expression of AXL, MITF and its downstream target Melan-A by immunoblotting (data not shown). The protein expression of AXL, MITF and Melan-A remained similar between PDX and its derived cell lines, although some PDX were heterogeneous for MITF and AXL expression (e.g., M032R6.X1), while the derived cell lines seemed to have shifted to either high AXL or high MITF expression (e.g. , M032R6.X1.CL).

Similarly, the gene expression pattern for the RIS signature was also largely maintained in the PDX cell lines (data not shown). This was confirmed when we validated the gene expression by qPCR of two genes specific for this signature, namely NGFR and ABCB5\ which showed the same expression levels of the genes in PDX and matched PDX-derived cell lines (data not shown).

Lastly, we analyzed the gene expression pattern for the mitotic signature, which was strongly maintained in the PDX cell lines (data not shown). Validation of two genes specific for the mitotic signature, namely SPOCK1 and NAT8L (data not shown), by qPCR showed high consistency in the level of expression between PDX and PDX- derived cell lines. Altogether, these data illustrate that these PDX-derived tumor- intrinsic gene signatures are maintained in the low passage PDX-derived cell lines. Example 6: Tumor-intrinsic and stromal gene signatures differentially predict response to anti-PD1 and anti-CTLA-4

The gene expression signatures identified in our PDX samples, and which are retained in PDX cell lines, are of tumor cell intrinsic origin. However, also the stromal signals have been shown to affect the prognosis of melanoma patients(Akbani, R et al., (2015), Cell 161, 1681-1696)). To investigate the relevance of gene expression signatures in patient melanoma samples treated with immune checkpoint blockade (ICB), we included the tumor-cell intrinsic signatures (‘MITF-high’, ‘Chronic inflammation’ and ‘Mitotic-high’) and the two stromal gene signatures (‘Keratin-high’ and ‘Stromal Immune’) These two signatures were used to identify the stromal immune-high samples and the keratin-high samples in the TCGA and Jonsson data (data not shown).

To test the predictive value of the tumor intrinsic gene sets (‘MITF-high’, ‘Chronic inflammation’ (referred to herein as RIS) and‘Mitotic-high’) and the stromal-derived gene sets (‘Keratin-high’ and‘Stromal Immune’ (referred to herein as Stromal Immune Signal (SIS)), we applied GSEA using these five gene sets to baseline gene- expression data derived from patients treated with anti-PD1 antibody (nivolumab). This analysis showed an enrichment in the non-responder group for the ‘chronic inflammation’ signature (i.e. the RIS gene signature) (FDR = 0.097, see Figure 4A-B). However, no significant enrichment was observed for the stromal immune signature (i.e. the SIS gene signature) in the baseline samples of anti-PD1 treated patients (FDR = 0.26, see Figure 4A-B), nor for any of the other signatures.

Next, we applied GSEA using the same five gene signatures to baseline gene- expression data derived from patients treated with anti-CTLA-4 antibody (ipilimumab). In contrast to the anti-PD1 treated samples, the anti-CTLA-4 treated samples did not show an enrichment for the ‘Chronic Inflammation’ signature (i.e. the RIS gene signature, see Table 1 , preferably Table 1A) (FDR = 0.40, see Figure 4C-D). However, a high ‘Stromal Immune’ signature (i.e. SIS gene signature, see Table 2) was significantly associated with response in these patients (FDR = 0.02, see Figure 4C- D). This suggests different transcriptional signals related to response for the two ICB therapies: response to anti-PD1 therapy seems to be more dependent on tumor- intrinsic factors while the response to anti-CTLA-4 therapy seem to be more dependent on stromal factors.

The predictive value of our gene expression signatures in the single ICB (anti-PD1 (nivolumab) and anti-CTLA-4(ipilimumab)) prompted us to investigate the response in the settings of combination treatment. Therefore, we analyzed the gene expression data of 18 patient samples from the OpACIN’ trial (Blank et al. (2019) Nature Medicine, in press). Patients in this randomized phase 1 b trial for high risk stage III melanoma (NCT02437279), received the combination of anti-PD1 (nivolumab) and anti-CTLA-4 (ipilimumab) either adjuvant or neo-adjuvant. Out of these 18 patients, six had a relapse within 24 months after start of treatment, suggesting intrinsic or acquired resistance.

First, we measured the gene expression levels of the‘Chronic Inflammation’ signature both in the relapsed and non-relapse patient samples. Indeed, similar to the anti-PD1 treatment patient samples the‘Chronic Inflamed’ signature (i.e., RIS gene signature, see Table 1 or Table 1A) expression levels were higher in the samples of patients that relapsed (see Figure 4E). Second, we measured the gene expression levels of the “Stromal Immune” signature (i.e., SIS gene signature, see Table 2) both in the relapsed and the non-relapsed patient samples. Similar to the anti-CTLA-4 mono therapy treated patient samples the“Stromal Immune” signature (SIS gene signature) expression levels were higher in the patient samples that did not relapse (see Figure 4E). However, neither signature alone could predict relapse in samples of patients treated with the combination of anti-PD1 and anti-CTLA-4 (Figure 4F).

Because the expression levels of the signature ‘Chronic inflamed’ (i.e. RIS gene signature) and‘Stromal Immune’ (i.e., SIS gene signature) to some extent correlate with relapse we hypothesized that a combination of the two signatures might better predict which patients relapsed. Patients with a low‘Chronic Inflamed’ (i.e. RIS gene signature) and high‘Stromal Immune’ signature (i.e. SIS gene signature) are expected to respond to at least one of the therapies while patients with a high‘Chronic Inflamed’ (i.e. RIS gene signature) and low‘Stromal Immune’ signature (i.e. SIS gene signature) are expected not to respond to either anti-PD1 or anti-CTLA-4. Indeed, by taking the ratio of the‘Stromal Immune’ (SIS gene signature) and‘Chronic Inflamed’ (RIS gene signature), we observed an enrichment of relapsed patients in samples with a high value (5/8) and an enrichment of patients with no relapse in the samples with a low ratio value (1/10) (see Figure 4G).

Example 7: Patients can be stratified based on gene signature expression in baseline samples

Previously, we showed that samples with a high risk of relapse could be identified with a IFN gamma related signature developed to stratify response to anti-PD1 on baseline samples (Blank et al. ( Ayers, M., et al. (2017), J. Clin. Invest. 127, 2930-2940.). Indeed, the average expression of the genes in the signature correlate well with relapse (Figure 5A). Moreover, the IFN gamma based signature highly correlated with the ratio of ‘Stromal Immune’ (i.e. SIS gene signature) and‘Chronic Inflamed’ signatures (i.e. RIS gene signature, see Figure 5B). Both signatures equally well identify the samples with a high risk of relapse in baseline and neo-adjuvant samples. However, our proposed ratio score or the IFN gamma-related signature can predict the response to anti-PD1 or anti-CTLA-4 individually. Improved patient stratification to the best predicted mono-therapy might decrease the high adverse side effects observed with the ICB combination of anti-PD1 and anti-CTLA-4 while retaining the high response rate.

Since the combination of the RIS gene signature and SIS gene signature is a strong predictor of response to anti-PD1 (nivolumab) and anti-CTLA-4 (ipilimumab), respectively, we hypothesized that our signatures could be used to identify in baseline samples to which specific treatment each patient likely responded. Therefore, we plotted the average expression of the samples for both the SIS gene signature and the RIS gene signature (see Figure 5C). Indeed, many of the patient samples are predicted to be a responder to either anti-PD1 (nivolumab)(right bottom quadrant) or anti-CTLA- 4 (ipilimumab)(left top quadrant). Three samples were predicted to be a possible responder to both anti-PD1 and anti-CTLA-4. Furthermore, the combined signature prediction showed a significant difference in progression-free survival (PFS; logrank, p=0.0007) between predicted responders and non-responders (see Figure 5D)

In addition, we tested the SIS and RIS signatures in conjunction in a dataset of Stage IV melanoma patients treated with the combination of anti-CTLA-4 and anti-PD-1 inhibitors. Again, and similar to the first melanoma dataset, the signatures in conjunction accurately separates the patients that did responder to the combination treatment (anti-CTLA-4 + anti-PD-1 ) from the patients that did not respond to the combination treatment (anti-CTLA-4 + anti-PD-1). This analysis again validated the signatures when used in conjunction for their predictive power in patients treated with anti-CTLA-4 and anti-PD-1 in combination.

In conclusion, the combination of the RIS and SIS gene signatures can be used to predict response to the combination treatment consisting of an anti-PD1 (nivolumab) and an anti-CTLA-4 (ipilimumab). Furthermore, our results suggest the two signatures can be used for predict to which of the ICB a patient will respond before treatment.

Example 8: Tumor exclusive genes of the RIS signature

For each gene mentioned in Table 1 we calculated the average expression in tumor cells and the stromal cells (cancer associated fibroblasts, T-cells, B-cells, endothelial cells and NK-cells) using scSEQ data (see Tirosh et al. 2016 science. sciencemag.org/content/352/6282/189. long

). Samples were normalized to 1 M reads per sample and three samples with high number of tumor cells as well as stromal cells were used for the analysis.

For each gene the average expression in tumor cells and stromal cells was calculated. Out of the 86 genes that comprises the RIS signature 14 genes had a 10-fold higher expression in tumor compared to stromal cells. These 14 genes were selected as the “tumor exclusive” RIS signature and validated as described herein. Example 9: Analysis to establish minimal number of random genes from the genes in Table 2 (SIS) for prediction.

The minimal number of random genes that can accurately classify samples into, e.g. a SIS-high or SIS-low class (e.g. higher or lower than the SIS reference score) was tested by random selection of 2 genes, 3 genes, 4 genes and further up to the full list of genes from the SIS signature genes of Table 2 .

Each random selection of the n-number of genes (n= 2:361) was iterated 1 ,000 times. Results are plotted in figure 8.

An additional plot shows examples of specified sizes of tested genes sets (including 5 random genes from Table 2, 10 random genes from Table 2, 15 random genes from Table 2, 22 random genes from Table 2, 40 random genes from Table 2 and 168 random genes from Table 2) and with the y-axis changed to percentages. See Figure 9.

The data shows that with a random set of 5 genes from Table 2 the classification is 100% identical to the full 361 genes in 92% of times.

With a random set of 10 genes from Table 2 the classification is 100% identical to the full 361 genes in 96% of times.

With a random set of 15 genes from Table 2 the classification is 100% identical to the full 361 genes in 97% of times.

With a random set of 22 genes from Table 2 the classification is 100% identical to the full 361 genes in 98% of times.

With a random set of 40 genes from Table 2 the classification is 100% identical to the full 361 genes in 99% of times.

With a random set of 168 genes from T able 2 the classification is 100% identical to the full 361 genes in 100% of times. 150 random sets of 5 and 10 genes from Table 2 are presented in Figures 6 and 7 All the presented datasets classify the samples 100% identical to the full set of 361 genes from the SIS signature.

In other words, the data shows that selecting a low number of genes from the set of genes of Table 2 already allows to classify a patient as a responder or a non- responder. For example, in case of selecting 5 random genes from the set presented in Table 2 (verified in 1000 different iterations) and testing the number of times the classification is identical as to the full set of genes in Table 2, it was shown that in nearly all case, i.e. 95% of cases with a random sets of 5 genes the classification is identical to the set of 361 genes. So, a patient can be classified as a responder or as a non-responder with high accuracy by selecting at least 2, in this example (at least) 5 genes from the genes mentioned in Table 2.

Having now fully described this invention, it will be appreciated by those skilled in the art that the same can be performed within a wide range of equivalent parameters, concentrations, and conditions without departing from the spirit and scope of the invention and without undue experimentation.

All references cited herein, including journal articles or abstracts, published or corresponding patent applications, patents, or any other references, are entirely incorporated by reference herein, including all data, tables, figures, and text presented in the cited references. Additionally, the entire contents of the references cited within the references cited herein are also entirely incorporated by references.

Reference to known method steps, conventional methods steps, known methods or conventional methods is not in any way an admission that any aspect, description or embodiment of the present invention is disclosed, taught or suggested in the relevant art.

The foregoing description of the specific embodiments will so fully reveal the general nature of the invention that others can, by applying knowledge within the skill of the art (including the contents of the references cited herein), readily modify and/or adapt for various applications such specific embodiments, without undue experimentation, without departing from the general concept of the present invention. Therefore, such adaptations and modifications are intended to be within the meaning and range of equivalents of the disclosed embodiments, based on the teaching and guidance presented herein. It is to be understood that the phraseology or terminology herein is for the purpose of description and not of limitation, such that the terminology or phraseology of the present specification is to be interpreted by the skilled artisan in light of the teachings and guidance presented herein, in combination with the knowledge of one of ordinary skill in the art.

Claims

1. A method for predicting the response of a subject suffering from a cancer to treatment with a PD-1 antagonist or for stratifying a subject for treatment with a PD- 1 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

(d) Calculating the arithmetic mean of the expression level for each of the genes in the RIS gene signature to generate a RIS signature score, wherein the RIS signature comprises at least two of the genes in Table 1 , preferably at least two of the genes in Table 1 A;

-(i) if the calculated RIS score is higher than the RIS reference score, then it indicates that said patient is not likely to benefit from treatment with the PD-1 antagonist or is not likely to respond to treatment with the PD-1 antagonist; or

-(ii) if the calculated RIS score is lower than the RIS reference score, then it indicates that said patient is likely to benefit from treatment with the PD-1 antagonist or is likely to respond to treatment with the PD-1 antagonist.

2. A method for predicting the response of a subject suffering from a cancer to a treatment with a CTLA-4 antagonist or for stratifying a subject for treatment with a CTLA-4 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

-(i) if the calculated SIS score is higher than the SIS reference score, then it indicates that said patient is likely to benefit from treatment with the CLTA4 antagonist or is likely to respond to treatment with the CTLA-4 antagonist; or

-(ii) if the calculated SIS score is lower than the SIS reference score, then it indicates that said patient is not likely to benefit from treatment with the CLTA4 antagonist or is not likely to respond to treatment with the CLTA4 antagonist.

3. A method for predicting the response of a subject suffering from a cancer to a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist or for stratifying a subject for treatment with a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

(i) if the calculated RIS score is lower than the RIS reference score and if the calculated SIS score is higher than the SIS reference score, then it indicates that said patient is likely to benefit from treatment with the combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist or is likely to respond to treatment with the combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist; or

(ii) if the calculated RIS score is higher than the RIS reference score and if the calculated SIS score is lower than the SIS reference score, then it indicates that said patient is not likely to benefit from treatment with the combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist or is not likely to respond to treatment with the combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist.

4. A method for treating a subject suffering from a cancer with a PD-1 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

5. A method for treating a subject suffering from a cancer with a CTLA-4 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

(c) Measuring the RNA expression level in the sample of step (b) for each gene in a Stromal Immune Signal (SIS) gene signature; (d) Calculating the arithmetic mean of the expression level for each of the genes in the SIS gene signature to generate a SIS signature score, wherein the SIS signature comprises at least two, preferably at least 5, 10, 15, or 22, of the genes in Table 2;

6. A method for treating a subject suffering from a cancer with a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

(h) Treating said subject with a combination therapy consisting of a PD- 1 antagonist and a CTLA-4 antagonist if the calculated RIS score is lower than the RIS reference score and if the calculated SIS score is higher than the SIS reference score.

7. A method for testing a tumor from a subject for the presence or absence of a biomarker that predicts the response of said subject to treatment with a PD-1 antagonist, said method comprises:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a); (c) Measuring the RNA expression level in the sample of step (b) for each gene in a Response Immune Signal (RIS) gene signature;

(f) Predicting the response of said subject to a PD-1 antagonist, wherein:

8. A method for testing a tumor from a subject for the presence or absence of a biomarker that predicts the response of said subject to treatment with a CTLA-4 antagonist, said method comprises:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

(f) Predicting the response of said subject to a CTLA-4 antagonist, wherein

-(i) if the calculated SIS score is higher than the SIS reference score, then it indicates that said patient is likely to benefit from treatment with the CTLA-4 antagonist or is likely to respond to treatment with the CTLA-4 antagonist; or -(ii) if the calculated SIS score is lower than the SIS reference score, then it indicates that said patient is not likely to benefit from treatment with the CLTA4 antagonist or is not likely to respond to treatment with the CTLA-4 antagonist.

9. A method for testing a tumor from a subject for the presence or absence of a biomarker that predicts the response of said subject to a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

10. Method according to any one of the previous claims, wherein the cancer is melanoma.

1 1. Method according to any one of the previous claims, wherein the PD-1 antagonist is an antibody.

12. Method according to claim 1 1 , wherein the PD-1 antibody is selected from the group of nivolumab, pembrolizumab, pidilizumab cemiplimab, PDR001 , AMP- 224, AMP-514, preferably nivolumab.

13. Method according to any one of the previous claims, wherein the CTLA- 4 antagonist is an antibody.

14. Method according to claim 3 wherein the antibody is selected from the group of ipilimumab, and tremelimumab, preferably ipilimumab.

15. Method according to any one of the previous claims, wherein the subject is a human subject.

16. Method according to claims 3 and 6 and 9, wherein the PD-1 antagonist is an antibody and the CLTA4 antagonist is an antibody.

17. Method according to claim 16, wherein the PD-1 antibody is selected from the group of nivolumab, pembrolizumab, pidilizumab cemiplimab, PDR001 , AMP- 224, AMP-514, preferably nivolumab, and the CTLA-4 antibody is selected from the group of ipilimumab, and tremelimumab, preferably ipilimumab.

18. A kit for assaying a tumor sample from a subject to determine a RIS gene signature score and/or a SIS gene signature score for said tumor sample, wherein the kit comprises a set of probes for detecting the expression of each gene in a RIS gene signature and/or a SIS gene signature, wherein the RIS gene signature comprises at least two of the genes of Table 1 , preferably at least two of the genes in Table 1A, and wherein the SIS gene signature comprises at least two, preferably at least 5, 10, 15, or 22, of the genes in Table 2.

19. Use of a RIS gene signature for predicting the response of a subject suffering from a cancer to a PD-1 antagonist or for stratifying patients for treatment with a PD-1 antagonist; wherein the RIS gene signature comprises at least two of the genes in Table 1 , preferably at least two of the genes in Table 1A.

20. Use of a SIS gene signature for predicting the response of a subject suffering from a cancer to a CTLA-4 antagonist or for stratifying patients for treatment with a CTLA-4 antagonist; wherein the SIS gene signature comprises at least two, preferably at least 5, 10, 15, or 22, of the genes in Table 2.

21. Use of a RIS signature and a SIS gene signature for predicting the response of a subject suffering from a cancer to a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist or for stratifying a subject for treatment with a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist; wherein the RIS gene signature comprises at least two of the genes in Table 1 , preferably at least two of the genes in Table 1A, and wherein the SIS gene signature comprises at least two, preferably at least 5, 10, 15, or 22, of the genes in Table 2.

22. A PD-1 antagonist for use in the treatment of cancer, wherein treatment comprises a method for predicting the response of a subject suffering from a cancer to a PD-1 antagonist or a method for stratifying a subject for treatment with a PD-1 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

23. A CTLA-4 antagonist for use in the treatment of cancer, wherein the treatment comprises a method for predicting the response of a subject suffering from a cancer to a CTLA-4 antagonist or a method for stratifying a subject for treatment with a CTLA-4 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);

24. A combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist for use in the treatment of cancer, wherein the treatment comprises a method for predicting the response of a subject suffering from a cancer to a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist or a method for stratifying a subject for treatment with a combination therapy consisting of a PD-1 antagonist and a CTLA-4 antagonist, said method comprising the steps of:

(a) Providing a tumor sample from said subject;

(b) Isolate RNA from the sample of step (a);