Method of predicting prognosis and treatment response

Abstract

The present invention provides a computer-implemented method for predicting the treatment response of a mammalian cancer patient to anti-cancer immunotherapy, the method comprising: a) providing a plurality of features comprising gene expression levels of at least 8 of the following cancer promoting genes PTGS2, CXCL1, CXCL2, CXCL5, CXCL6, IL1A, HAS2, IL11, IL1B, EREG, IL6, CXCL8 (IL8), MMP12, NRG1, VCAN, AREG previously measured in a sample obtained from the tumour of a mammalian cancer patient; b) computing a score using the features provided in a); and c) predicting the treatment response of the mammalian cancer patient based on the score computed in step b). Also provided are related methods and systems for predicting treatment response.

Description

[0001] METHOD OF PREDICTING PROGNOSIS AND TREATMENT RESPONSE

[0002] Field of the Invention

[0003] The present invention relates to materials and methods for predicting response to cancer therapy and overall survival among cancer patients, and particularly, although not exclusively, to patients undergoing immunotherapy.

[0004] Background

[0005] The concept that cancer induces inflammation and that inflammatory cells at the tumour site can support cancer progression, is well established (Coussens et al., 2013; Hanahan and Weinberg, 2011 ; Mantovani et al., 2008). Several cellular and molecular inflammatory mediators commonly found in clinically apparent tumours are well known for having pro-tumorigenic effects and to be associated with many features of aggressive and invasive tumours in both preclinical models and cancer patients. These include prevalent tumour-infiltrating leukocytes such as macrophages, neutrophils, immature myeloid cells or regulatory T cells and molecules produced by these and other leukocytes, stromal cells, or directly by cancer cells. IL-6, IL-8, CCL2, CXCL1 or vascular endothelial growth factor (VEGF) are classic examples of soluble factors with pleiotropic effects that can foster cancer growth and spread (Coussens et al., 2013; Mantovani et al., 2008).

[0006] Inflammation at the tumour site can also play an anti-cancer role, partly by contributing to immune recognition and elimination of cancer cells. Cytotoxic T cells (CTLs), in particular, are recognised anti-tumour effectors in preclinical cancer models and their intratumoural abundance associates with improved patient outcome and response to cancer therapy (Binnewies et al., 2018; Fridman et al., 2012; Galon, 2006; Thorsson et al., 2018). Accordingly, more favourable prognosis has been also linked with high intratumoural levels of CTL chemoattractants, like CXCL9 or CXCL10, or cytokines that promote type I immunity, CTL differentiation and effector function, such as IL-12 or type I and II interferons (IFNs) (Gajewski et al., 2013; Spranger and Gajewski, 2018; Vesely et al., 2011). The levels of these factors within the tumour microenvironment (TME) can vary depending on the overall systemic and / or local inflammatory status and their integrated effect contributes to the strength and extent of the anti-tumour immune response. In addition to conventional helper CD4+and cytotoxic CD8+T cells, further evidence indicates that tumour infiltration by other inflammatory cells equally favours immune attack and correlates with good prognosis. Natural killer (NK) cells, gamma delta T cells, innate like lymphocytes and the Batf3-dependent conventional dendritic cells type I (cDC1) constitute some of the immune subsets often associated with improved outcome (Bottcher et al., 2018;

[0007] Broz et al., 2014; Gentles et al., 2015; Mittal et al., 2017; Morvan and Lanier, 2016; Ruffell et al., 2014; Sanchez-Paulete et al., 2016; Spranger et al., 2015; 2017). This is true both in spontaneous and therapy-induced anti-tumour responses like those ensuing from administration of immune-checkpoint inhibitors, a treatment modality that has revolutionised cancer treatment eliciting beneficial responses, including long-term remissions, in a plethora of cancer types (Ribas and Wolchok, 2018). Especially in immune checkpoint blockade (ICB) therapy, the abundance of select immune cells or inflammatory mediators has been associated positively or negatively with treatment response (Ayers et al., 2017; De Henau et al., 2016; Herbst et al., 2014; Roh et al., 2017; Rooney et al., 2015; Tumeh et al., 2014)’.

[0008] Yet, how antagonistic cancer promoting or inhibitory inflammatory TME are established during cancer development and progression is not understood. Moreover, the signals and pathways that regulate the quality and quantity of the different elements of the inflammatory infiltrate are poorly defined. Improving our understanding of these processes is of great clinical relevance as it could strengthen our ability to predict response from therapy and further provide attractive therapeutic targets to improve the efficacy of anti-cancer treatment.

[0009] The notion of immune-based control of cancer has historically been controversial but has gained irrefutable support over the past decade from the remarkable clinical successes of cancer treatments that harness the cancer-restraining functions of the immune system (Alfaro et al ., 2016; Van Allen et al., 2015). In particular, so-called immune checkpoint blockade (ICB) therapies targeting the CTLA-4 and PD(L)-1 axis have transformed the landscape of cancer treatment (Andrea et al., 2012). Despite their widespread use in clinical practice, including as first-line treatment across multiple cancer types, many patients derive little benefit from ICB and some experience serious side effects (Auslander et al., 2018). There is thus an urgent unmet need for sensitive and accurate biomarkers to predict response to ICB and guide treatment selection, especially in settings in which alternative therapies may also lead to patient benefit (Ayers et al., 2017; Azizi et al., 2018).

[0010] Established and emerging immune biomarkers for ICB therapy fall under two categories which broadly evaluate T cell activity or cancer cell (neo)antigenicity. Tumors described as ‘hot’ or ‘T cell-inflamed’ are those which show signs of T cell-infiltration, particularly of CD8+T cells, and / or increased levels of inflammatory mediators linked to cytotoxic T cell effector differentiation and function. Patients harbouring tumors with these characteristics tend to respond better to ICB. Likewise, measures of cancer cell antigenicity such as tumor mutational burden (TMB) and neoantigen clonality also associate with ICB response. Importantly, T cell-inflammation and TMB are poorly correlated, and when combined, exhibit joint predictive power (Cristescu et al., 2018). Yet, even when combined, these biomarkers lack sufficient accuracy with abundant examples of highly CD8+T cell-infiltrated tumors failing to respond to ICB and, conversely, cold and poorly antigenic tumors sometimes respond vigorously. Furthermore, in some cancer types such as clear cell renal cell cancer (ccRCC), a malignancy with one of the highest response rates to ICB, none of the currently available immune biomarkers have shown meaningful predictive utility (Bhattacharya et al., 2020; Bisawas et al., 2019). A common feature of past and current tumor immune profiling efforts and of established and emerging immunotherapy biomarkers is the focus on T cell-inflammation. Inflammation, however, is arguably a feature of all solid tumors. Even in so-called ‘immune-desert’ tumors, which broadly lack immune cell infiltration, oncogene-driven inflammatory signalling is nevertheless commonly upregulated in cancer or stromal cells. However, in contrast to T cell- mediated inflammation, the type of inflammatory response most frequently present in clinically- apparent tumors is associated with malignant cancer growth and therapy resistance. Thus, rather than being inflamed or not, tumors exhibit quantitatively and qualitatively different inflammatory profiles with cancer promoting or suppressive effects.

[0011] Multiple studies have highlighted the intimate link between oncogenic transformation and tumor progression with inflammation and immune evasion (van Weverwij Pellyk and de Visser, 2023; Spranger and Gajewski, 2018). Among these studies, recent work has uncovered a central role for the cyclooxygenase (COX)-2 / prostaglandin E2 (PGE2) pathway, which is upregulated in numerous cancers and implicated in various aspects of malignant growth, as a major pro- tumorigenic inflammatory axis that enables cancer immune escape and drives resistance to immunotherapy (Zelenay et al., 2015; Bonavita et al., 2020; Pelly et al., 2021 ). Accordingly, antiinflammatory drugs or antagonists that inhibit PGE2 synthesis or signalling augment the efficacy of ICB and chemoimmunotherapy combinations in preclinical models (Pelly et al., 2021 ; Bell et al., 2022). Moreover, by interrogating large cancer patient datasets, previous work by the inventors demonstrated the prognostic and predictive power of a murine-derived inflammatory gene signature that integrated aspects of both COX-2 associated pro-tumorigenic and T cellactivation. Thus, the level of activity of the COX-2 / PGE2 pathway, upregulated in numerous cancers and implicated in multiple aspects of malignant tumor growth, constitutes a candidate for molecular stratification of the tumor inflammatory phenotype.

[0012] Whilst previously described predictive and prognostic models of cancer show promise, there remains an unmet need for further human-centred models able to predict treatment response and / or survival of cancer patients. The present invention seeks to fulfil these needs and provides further related advantages.

[0013] Summary of the Invention

[0014] The present inventors conducted a systemic meta-analysis of thousands of patient samples and cancer cell lines, to study the value of monitoring tumor-promoting inflammation as a prognostic and predictive immune-relevant biomarker. Using COX-2 expression as a putative nodal indicator of intratumoral pro-tumorigenic inflammatory activity, the inventors defined a human- derived pro-tumorigenic inflammatory (PTI) gene signature of extracellular mediators that shows profound and consistent prognostic value across molecularly-defined immune subtypes, pancancer and in individual malignancies. Through interrogation of genomic, transcriptomic and proteomic data, the present inventors uncovered the cellular source, oncogenic correlates and the impact of PTI on T cell-inflammation. Notably, PTI is shown to exhibit joint prognostic and predictive utility when combined with the current strongest predictors of ICB response. Moreover, in clear cell renal cell carcinoma (ccRCC), a cancer type in which established immune biomarkers do not show predictive utility, PTI is shown to selectively predict immunotherapy outcome. Molecular profiling of archival ccRCC samples and validation through analysis of two independent ICB-treated patient cohorts indicates that patient stratification based on pre-treatment PTI levels has the potential to guide immunotherapy decisions. PTI is established as an independent prognostic and predictive immune biomarker with potential to guide treatment selection in the clinic. Overall, pro-tumorigenic inflammation is established as a key determinant of cancer patient outcome and the PTI signature as a powerful prognostic and predictive biomarker of response to immune-checkpoint inhibitor therapy.

[0015] In a first aspect of the invention, there is provided a computer-implemented method for predicting the treatment response of a mammalian cancer patient to anti-cancer immunotherapy, the method comprising: a) providing a plurality of features comprising gene expression levels of at least 8, at least 9, at least 10, at least 11 , at least 12 or more (such as all) of the following cancer promoting genes PTGS2, CXCL1 , CXCL2, CXCL5, CXCL6, IL1 A, HAS2, IL11 , IL1 B, EREG, IL6, CXCL8 (IL8), MMP12, NRG1 , VCAN, AREG previously measured in a sample obtained from the tumour of a mammalian cancer patient; b) computing a score using the features provided in a); and c) predicting the treatment response of the mammalian cancer patient based on the score computed in step b).

[0016] In some embodiments, the features further comprise Clonal_TMB and CXCL9.

[0017] In some embodiments, the features comprise the gene expression levels of all of: PTGS2, CXCL1 , CXCL2, CXCL5, CXCL6, IL1 A, HAS2, IL11 , IL1 B, EREG, IL6, CXCL8 (IL8), MMP12, NRG1 , VCAN, AREG.

[0018] In some embodiments, the gene expression level is a normalised gene expression level.

[0019] In some embodiments, the score is computed by z-score normalising the individual gene expression values of the features and calculating the mean of said normalised measurements.

[0020] In some embodiments, the score is computed according to the formula: score = — Tip E"=-> G''os(e) wherein npis the number of said gene expression features, Gosis the positive correlated genes, within an (i) interval of unitary values, and (e) represents the gene expression values expressed as Iog2 counts per million (CPM); optionally wherein Gipos(e) is the mean expression of Iog2 transformed counts per million (reads per kilobase million (FPKM)) of positive genes.

[0021] In other embodiments, the score is a probabilistic score calculated using the formula: wherein Po is an intercept weight, is a vector of weights for each of k variables and % is a vector of variables associated with the sample, wherein the variables comprise said plurality of features.

[0022] In some embodiments, the computer-implemented method further comprises assessing the tumour burden and / or neoantigen prevalence of the cancer patient.

[0023] In some embodiments, predicting the treatment response of the mammalian cancer patient comprises: comparing the computed score from step b) with a reference median score derived from a sample cohort of responders and a reference median score derived from a sample cohort of non-responders, wherein the sample cohort of responders and non-responders are of i) cancer patients having the same type of cancer as said cancer patient, or ii) cancer patients which are tumour type agnostic.

[0024] In some embodiments, the score computed in step b) indicates that the cancer is predicted to respond to anti-cancer immunotherapy, and the method further comprises selecting the mammalian cancer patient for anti-cancer immunotherapy.

[0025] In some embodiments, the gene expression levels of the at least 8 cancer promoting genes are acquired by measuring the gene expression levels of at least 8, at least 9, at least 10, at least 11 , at least 12 or more (such as all) of the following cancer promoting genes: PTGS2, CXCL1 , CXCL2, CXCL5, CXCL6, IL1 A, HAS2, IL11 , IL1 B, EREG, IL6, CXCL8 (IL8), MMP12, NRG1 , VCAN, AREG in a sample obtained from the tumour of a mammalian cancer patient.

[0026] In a second aspect of the invention, there is provided a method of predicting the treatment response of a mammalian cancer patient to anti-cancer immunotherapy, the method comprising: a) measuring a plurality of features comprising measuring the gene expression of at least 8, at least 9, at least 10, at least 11 , at least 12 or more (such as all) of the following cancer promoting genes: PTGS2, CXCL1 , CXCL2, CXCL5, CXCL6, IL1 A, HAS2, IL11 , IL1 B, EREG, IL6, CXCL8 (ILS), MMP12, NRG1 , VCAN, AREG in a sample obtained from the tumour of a patient; b) determining a score using the features measured in a); and c) predicting the treatment response of the patient based on the score determined in step b).

[0027] In some embodiments, the features further comprise Clonal_TMB and CXCL9.

[0028] In some embodiments, determining the score to predict the treatment response of a subject comprises: inputting data representing the features comprising the gene expression levels of the at least 8 cancer promoting genes of the subject into a machine learning classifier, wherein said machine learning classifier has been trained on a training data set comprising the gene expression levels of at least the same cancer promoting genes as those forming the score of the subject, and wherein the training data set is derived from at least 2 sample cohorts; wherein one cohort comprises a plurality of samples derived from a plurality of mammalian cancer patients known to have responded to anti-cancer immunotherapy (responders), and the other cohort comprises a plurality of samples derived from mammalian cancer patients known not to respond to anti-cancer immunotherapy (non-responders); and causing the machine learning classifier to classify the score of the subject as belonging to the response group or the nonresponse group.

[0029] In some embodiments, the training data set comprises gene expression levels of the at least 8 cancer promoting genes of at least 50, at least 100, at least 150, at least 200, at least 250, at least 300 samples derived from mammalian cancer patients known to have responded to anticancer immunotherapy (responders) and gene expression levels of the at least 8 cancer promoting genes of at least 50, at least 100, at least 150, at least 200, at least 250, at least 300, at least 350, at least 400, at least 450, at least 500, at least 550 samples derived from mammalian cancer patients known not to have responded to anti-cancer immunotherapy (non- responders).

[0030] In some embodiments, the machine learning classifier has been trained on a plurality of scores derived from at least three sample cohorts, wherein one cohort comprises a plurality of samples derived from a plurality of mammalian cancer patients known to have completely responded to anti-cancer immunotherapy (complete responders), one cohort comprises a plurality of samples derived from a plurality of mammalian cancer patients known to have partially responded to anti-cancer immunotherapy (partial responders), and the other cohort comprises a plurality of samples derived from mammalian cancer patients known not to respond to anti-cancer immunotherapy (non-responders). In some embodiments, the machine learning classifier is a logistic regression model, elastic net regression model or a tree-based model such as a random forest classifier.

[0031] In a third aspect of the invention, there is provided a method of classifying a sample obtained from a mammalian cancer subject, said sample comprising nucleic acid and / or protein, the method comprising: a) analysing the sample to obtain the gene expression levels of the at least 8 cancer promoting genes selected from: PTGS2, CXCL1 , CXCL2, CXCL5, CXCL6, IL1 A, HAS2, IL11 , IL1 B, EREG, IL6, CXCL8 (IL8), MMP12, NRG1 , VCAN and AREG; and b) carrying out the method of the first or second aspect of the invention using the gene expression levels obtained in step a).

[0032] In some embodiments, said sample is derived from one or more cancer cells of the subject.

[0033] In some embodiments, the sample obtained from the tumour of the patient is a formalin-fixed- paraffin embedded (FFPE) sample.

[0034] In a fourth aspect of the invention, there is provided a method of treating a cancer in a mammalian cancer patient, the method comprising: a) carrying out the method according to of any one of the preceding aspects of the invention to predict the treatment response of the mammalian cancer patient to anti-cancer immunotherapy; b) determining that the mammalian cancer patient is predicted to respond to said anti-cancer immunotherapy in step a); and c) administering a therapeutically effective amount of said anti-cancer immunotherapy to the subject in need thereof.

[0035] In a fifth aspect of the invention, there is provided a method of treatment of a cancer in a mammalian cancer patient, comprising: a) carrying out the method of any one of the first to fourth aspects of the invention; b) determining that the score of the mammalian cancer patient indicates that the mammalian cancer patient is predicted to respond to anti-cancer immunotherapy; and c) administering immunotherapy to the mammalian cancer patient in need thereof.

[0036] In some embodiments, the method further comprises analysing one or more additional markers of anti-cancer immunotherapy response derived from the mammalian cancer patient to supplement and / or corroborate the prediction of anti-treatment response or non-response.

[0037] In a sixth aspect of the invention, there is provided a system for predicting the treatment response of a mammalian cancer patient to an anti-cancer immunotherapy, the system comprising: at least one processor; and at least one non-transitory computer readable medium containing instructions that when executed by the at least one processor, cause the at least one processor to perform the method of the first or second aspect of the invention.

[0038] In a seventh aspect of the invention, there is provided one or more computer readable media comprising instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the method of the first or second aspect of the invention.

[0039] In an eighth aspect of the invention, there is provided a pharmaceutical composition for use in the treatment of a mammalian cancer patient comprising an anti-cancer immunotherapy, wherein the patient has been predicted to respond to said anti-cancer immunotherapy by a method of any one of the first to third aspects of the invention.

[0040] In some embodiments of the first to eighth aspect of the invention, the mammalian cancer patient has and / or the cancer is: renal cancer, such as renal cell carcinoma or kidney renal cell carcinoma (KIRC) ; gastroesophageal tumour; oesophageal cancer (ECA); melanoma; bladder cancer (BLCA); gastric cancer; lung cancer, such as lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC) or non-small cell lung cancer (NSCLC); pancreatic cancer (PAAD); liver cancer (LIHC); glioblastoma (GBM); adrenocortical carcinoma (ACC);cervical cancer (CESC); head and neck cancer such as head and neck squamous cell carcinoma (HNSC); thyroid cancer (THCA); uterine cancer (UCEC); Tenosynovial giant cell tumor (TGCT); melanoma; BRCA; sarcomas and pan-cancer. In some embodiments, the renal cell carcinoma is clear cell renal cell carcinoma (ccRCC). In some embodiments, the immune subtype of the cancer is one of: wound healing, IFNy-dominant, TGFp-dominant, inflammatory and / or lymphocyte depleted subtype.

[0041] In some embodiments, said anti-cancer immunotherapy comprises IL-2 immunotherapy or immune checkpoint blockade therapy alone, or in combination with anti-VEGF therapies. In some embodiments, the immune checkpoint blockade therapy comprises programmed death-1 (PD-1 ) blockade, programmed death-ligand 1 (PD-L1) blockade and / or cytotoxic T-lymphocyte- associated protein 4 (CTLA-4) blockade. In some embodiments, said immune checkpoint blockade therapy comprises treatment with Nivolumab, Pembrolizumab, Atezolizumab and / or Ipilimumab. In some embodiments, the anti-VEGF therapy is bevacizumab.

[0042] The invention includes the combination of the aspects and preferred features described except where such a combination is clearly impermissible or expressly avoided. Summary of the Figures

[0043] Embodiments and experiments illustrating the principles of the invention will now be discussed with reference to the accompanying figures in which:

[0044] Figure 1. Derivation of a pan-cancer pro-tumorigenic inflammatory signature linked to COX-2. (A) COX-2 mRNA (PTGS2) expression across 31 TCGA (the cancer genome atlas) datasets. (B) Bioinformatic pipeline used to discover a gene signature of COX-2-centred PTI.

[0045] (C) Protein-protein interaction network for genes upregulated in COX-2hightumors pan-cancer.

[0046] (D) Significantly enriched gene ontology biological processes in the COX-2highprotein-protein interaction network. (E) Kaplan-Meier (KM) plot of pan-solid tumor TCGA patients stratified into quantiles by PTI signature score with associated risk table.

[0047] Figure 2. Pro-tumorigenic inflammatory mediators are powerful indicators of poor prognosis. (A) HRs and 95% confidence intervals for continuous PTI and IFNy signatures, as well as their interaction term, from a multivariate CoxPH model with cancer type and disease stage strata pan-cancer. (B) A simulation of the hazard rate for exemplar high and low IFNy patients across a range of PTI values. Lines indicate median hazard rates. Darker shaded areas are quartiles and lighter shaded areas indicate minimum and maximum hazard rates. (C) HRs and 95% confidence intervals from univariate Cox regression survival analysis for biomarkers of interest in six immune subtypes. (D) KM survival curve (left panel) for ‘IFNy- dominant subtype patients stratified into four groups by PTI and IFNy signatures. Asterisks refer to FDR values. Boxplots (right panel) comparing CD8+T cell scores between groups. All P values are represented by asterisks unless otherwise stated (*<0.05, **<0.01 , ***<0.001). (E) HRs, 95% confidence intervals and Wald test P values from multivariate CoxPH models of OS in ccRCC (n= 531 ), LIHC (n= 374), PAAD (n= 176), glioblastoma (GBM, n= 168), CESC (n= 306), adrenocortical carcinoma (ACC, n= 79), LICEC (n= 277), THCA (n= 513) and HNSC (n= 240).

[0048] Figure 3. MEK inhibition attenuates secretion of pro-tumorigenic inflammatory mediators by cancer cells. (A) Pairwise Spearman correlation coefficients of PTI genes in non- haematological, non-fibroblast cancer cell lines from the CCLE. Non-significant (P> 0.05) relationships are marked by ’X’. (B) PTI status coefficients from linear mixed effects models in which PTI status (low / high) was used to predict cell line drug sensitivity adjusted for cancer type. Drug names (black) and targets (light grey / dark grey) are labelled on the axis. All bars shown are from statistically significant associations (P< 0.05). (C) Correlation of cell line sensitivity to refametinib by the natural log of the drug IC50 value, and of the cancer-intrinsic PTI transcriptional program for all cell lines (top), and for NSCLC lines (bottom). (D) Concentration of secreted proteins detected by Luminex 6-, 8- and 24-hours following treatment of NCI H 1339 (top panel) and NCIH3255 (bottom row) cells with trametinib at 1pM and 10pM.

[0049] Figure 4. NFKB links MAPK pathway mutations to cancer cell-intrinsic PTI expression. (A) Boxplot showing the association of the cancer cell-intrinsic PTI transcriptional program and KRAS mutations in pan-cancer cell lines. (B) RELA gene copy number estimates (0= diploid) for cancer cell lines with low, intermediate and high cancer cell-intrinsic PTI expression. (C) Levels of pQ5 / RELA pS536 in RPPA data from KRAS mutated cancer cell lines stratified according to cancer-intrinsic PTI status. (D) Table showing the top ten significantly enriched motifs in the extended, cancer cell-intrinsic PTI gene list alongside predicted transcription factor binders. (E) Sequence logos from multiple sequence alignments of NFKB1 and NFKB2 binding motifs from chromatin immunoprecipitation sequencing (ChlPseq) experiments showing consensus nucleotide sequences for the associated motifs. (F) Odds ratios and 95% confidence intervals from multivariate linear regression models using mutation status to predict PTI scores in the pan-cancer TCGA dataset adjusted for cancer type and TMB. (G) The proportion of significant mutation-to-PTI links that are in different oncogenic signalling pathways. Breakdown of significant MAPK pathway mutation-to-PTI links in individual cancers. (H) Comparison of intratumoral PTI signature scores between pan-cancer TCGA tumors with diploid copies of EGFR and RELA (WT) versus those with chromosomal amplifications. Wilcoxon test P values are shown.

[0050] Figure 5. Pro-tumorigenic inflammatory mediators are derived from heterogeneous cell types and affect the transcriptional phenotype of tumor-infiltrating T cells. (A) Dimensionality reduction of single cell transcriptomic data from primary HNSC tumors colored by (A) major cell type or (B) PTI signature module score. (C) PTI module scores of individual cancer cells from primary HNSCs. Patients defined as having high cancer cell-intrinsic PTI expression are shown in light grey, and low in dark grey. Asterisks represent adjusted P values from a non-parametric T test comparing cancer cells from PTIhigh(light grey) versus PTIlow(dark grey) groups (*<0.05, **<0.01 , ***<0.001 ). The Kruskal-Wallis test P value was <0.001 . (D) Heatmap with circle size and colour representing the mean expression levels of individual PTI genes in individual clusters of immune and stromal cells in the HNSC scRNAseq dataset. Grey squares highlight selected groups of co-expressed genes. (E) The mean proportion of CIBERSORT-inferred immune cell populations in ‘ I FNy -dominant subtype tumors with high, intermediate and low levels of PTI. (F) Correlation of PTI signature score with a signature of T cell anergy including contours to illustrate point density. Spearman correlation coefficient and P value are shown. (G) CD8 ICB response signature scores in individual CD8+T cells from MEL and NSCLC tumors separated by PTI status. Adjusted P values from multiple non-parametric T tests are shown (*<0.05, **<0.01 , ***<0.001 ).

[0051] Figure 6. PTI predicts lack of benefit from checkpoint inhibitors and outperforms other signatures of resistance. (A) Pie chart showing the proportion of samples (n=1032) in the pancancer immunotherapy meta-dataset from five cancer types. (B) A table showing the individual cohorts, sample numbers and treatment in the pan-cancer immunotherapy meta-dataset. (C) Barplots showing the proportion of responders (PR / CR) in patient groups stratified according to median levels of PTI, TIS, IFNy signalling, CD8+T cells and CD274 expression. (D) The proportion of responders in four groups of patients stratified simultaneously by PTI and biomarkers of ‘hot’ tumors including left to right TIS, IFNy signalling, CD8+T cells and CD274 expression. Selected x2test P values are shown (*<0.05, **<0.01 , ***<0.001). (E) OS of patients from the pan-cancer ICB dataset in each PTI / TIS quadrant with associated risk table. Wald test P values are shown. (F) HRs, 95% confidence intervals and -log P values from multivariate CoxPH models for resistance signatures and randomly generated signatures using a median cut-off value, with TIS status and dataset as covariates. The graph shows only those signatures correlated with poor prognosis (HR>1). (G) Comparison of PTI scores between patients that had PD or PR / CR following ICB in the CPU 000 dataset. P value refers to a non-parametric T test. (H) Effect sizes and standard errors from meta-analysis of the PTI signature and other putative biomarkers across seven individual ICB datasets from the CPU 000 cohort. Error bars are capped at an odds ratio of 5 for visualisation purposes. CYT = cytolytic score. (I) OS of CPU 000 patients with high and low levels of clonal TMB (top panel), stratified by PTI signature. Logrank test P values are shown.

[0052] Figure 7. Pro-tumorigenic inflammation is a putative predictive biomarker in clear cell renal cell carcinoma. (A) Heatmap of pairwise Spearman correlation coefficients between individual PTI genes in the in-house Nanostring dataset. (B) Heatmap showing immune, stromal and TME gene signature scores in samples ordered left to right by increasing PTI score. On the right of the heatmap asterisks refer to Spearman correlation P values for signatures against the PTI signature (*<0.05, **<0.01 , ***<0.001 ). (C) Correlation of the PTI signature with CD8+T cell and IFNy signalling gene signatures. Points are coloured according to PTI score. Spearman Rho values (R) and P values are shown. (D) Heatmap of individual PTI gene expression levels in the Manchester Nanostring cohort for patients evaluable for prognosis. High and low PTI groups are indicated by the top annotation (light grey = high, dark grey = low). (E) OS of Manchester cohort patients stratified by PTI with associated risk table. P value refers to the log rank test. (F) HRs and 95% confidence intervals from univariate Cox regression models predicting OS of patients with selected gene signatures. Patients were stratified by median cutoff values. (G) KM plot with associated risk tables showing the OS of Checkmate 009, 010 and 025 patients treated with nivolumab (left) or everolimus (right) stratified by median PTI expression level. (H) Comparison of PTI signature scores between PD, and PR / CR patients treated with atezolizumab, atezolizumab plus bevacizumab or sunitinib on the IMMotion150 trial. P values from non-parametric T tests comparing PD and PR / CR patients are shown.

[0053] Figure 8. COX-2 expression and PGE2 signalling associate with expression of PTI mediators by multiple cell types in vitro. (A) Normalised enrichment scores from GSEA with Hallmark gene signatures using a ranked list of genes from the comparison of PTGS2 high and low cancers. (B) Log2FC values and associated FDRs for PTI genes from differential gene expression analysis of PTGS2 high and low human cancer cell lines. (C) Log2FC of PTI genes from differential gene expression (DGE) analysis of mouse BMDMs, human fibroblasts and CD4+T cells treated in vitro synthetic PGE2. Significantly upregulated genes (FDR< 0.05, log2FC> 1 ) are in light grey and downregulated genes are in dark grey (FDR< 0.05, log2FC< -1 ).

[0054] Figure 9. PTI is a powerful and independent prognostic indicator that correlates with oncogenic transformation and advancing disease stage. (A) Boxplots showing scores for PTI, CD8+T cell, T cell inflammation (TIS) and IFNy signalling signatures across Thorsson et al (2018) pan-cancer immune subtypes. Non-parametric T test results are shown. (B) OS of ‘IFNy - dominant’ patients stratified by PTI and clockwise by TIS and CD8 signatures, TIL fraction and CD274 expression. For all survival analyses individual FDR values from Wald tests are shown (*FDR<0.05, **FDR<0.01 , ***FDR<0.001 ). (C) HRs, 95% confidence intervals and Wald test P values from multivariate CoxPH models of OS in KIRK / ccRCC (n= 531 ), LIHC (n= 374), PAAD (n= 176), glioblastoma (GBM, n= 168), CESC (n= 306), adrenocortical carcinoma (ACC, n= 79), LICEC (n= 277), THCA (n= 513) and HNSC (n= 240). (D) Mean PTI signature scores (±SEM) for samples from different stages of LUSC. (E) Correlation of disease stage with PTI score (mean ± SEM) in LUSC (n= 498), LUAD (n= 509), PAAD (n= 176), ccRCC (n= 531 ), and TGCT (n= 139) TCGA datasets.

[0055] Figure 10. Cancer cell-intrinsic PTI expression is linked to MEK activity. (A) Heatmap illustrating the stratification of PTI high and low cancer cell lines from the Garnett et al (2012) drug screen dataset, which comprises microarray gene expression data for more than 600 cell lines. These cell lines partially overlap those in the CCLE dataset. (B) Dose response curves of ten PTIhighand ten PTIlowrefametinib-treated NSCLC cell lines. Shapes represent the different cell lines. (C) Change from baseline confluence percentage over time for cell lines NCIH1339 (top) and NCIH3255 (bottom) treated with two concentrations of trametinib in 96 well plates. Lines indicate timepoints of supernatant analysis. (D) Representative images from confluence readouts after 8, 24 and 72 hours of 10pM trametinib treatment. Confluence percentage of the entire well is shown in the bottom right of each image. (E) Concentration of four proteins in the supernatant of NCIH1339 lung cancer cells treated with 1 or 10p.M of selumetinib for 6- 8- and 24-hours.

[0056] Figure 11. Cancer cell lines with high expression of PTI mediators have distinct copy number profiles and protein expression. (A) Copy number estimates (0= diploid) for four different genes separated by cancer cell-intrinsic PTI status. FDR values from non-parametric T tests are shown (*<0.05, **<0.01 , ***<0.001 ). (B) Volcano plot showing differential expression of proteins and phosphoproteins measured by RPPA comparing PTIhighand PTIlowKRAS mutant cell lines.

[0057] Figure 12. PTI mediators derive from heterogeneous cell types. (A) A heatmap of Spearman correlations for the PTI signature against cell types inferred by the MCPcounter algorithm across TCGA datasets. Significant (p< 0.05) correlations are shown by an asterisk. (B) t-SNE clustering of single cells from two independent scRNAseq patient datasets from NSCLC and MEL coloured on the top row by cell type and beneath by PTI module score. (C) and (D) Heatmaps showing mean expression of individual PTI genes by size and colour in individual clusters of immune and stromal cells across NSCLC and MEL scRNAseq datasets respectively. Grey squares highlight co-expressed genes. (E) Scatter plots with contours showing point density for the correlation of PTI signature score with gene set scores for CD8+T cell-related gene signatures in ‘IFNy -dominant' subtype tumors (Singer etal., 2016; Guo et al., 2018). Spearman correlation coefficients (E) and associated P values are shown.

[0058] Figure 13. PTI adds predictive and prognostic value to current biomarkers of ICB response. (A) AUC values from multivariate logistic regression models predicting patient ICB response (PD versus PR / CR) across 16 datasets for TIS combined with resistance signatures. Asterisk refers to P value (*P<0.05) from a non-parametric T test. Grey lines indicate mean values with upper and lower confidence intervals from 1000 bootstraps. (B) Heatmap of pairwise Spearman correlation coefficients for continuous immune biomarkers including the PTI signature. Shapes illustrate the direction and strength of the correlation. SDI = Shannon diversity index. Boxes have been placed around the four main clusters by hierarchical clustering. (C) Boxplots showing the association of PTI scores with binary clinical and genomic biomarkers in the CPU 000 dataset. Wilcoxon test P values are shown. B1501 is an HLA type, and B44 and B62 refer to HLA supertypes. DDR = DNA damage response pathway mutations. RTK = receptor tyrosine kinase mutation. Max het = maximum HLA heterozygosity. HLA-LOH = HLA locus loss of heterozygosity. (D) OS of PTI high, intermediate and low CPU 000 patients (Mariathasan dataset n= 331 , Riaz n= 38, VanAllen n= 34, Hugo n= 27, Snyder PLOS Med n= 25, Snyder NEJM n= 7). Asterisks refer to the Wald test P value (**< 0.01). (E) Barplots showing deviance estimates from univariate and multivariate logistic regression models to predict patient response to ICB (PD v PR / CR) in the CPU 000 cohort. Green dots beneath the x axis indicate the variables included in each model. Lines between green dots indicate interaction terms. Statistical significance between nested models is judged according to the likelihood ratio test. (F) KM plots showing OS of CPU 000 cohort patients with high and low CXCL9 (top panel) or CD274 (bottom panel) expression stratified by PTI status. Logrank test P values are shown. (G) OS of clonal TMBhi9hCXCL9h'9hCPU 000 patients stratified by PTI signature status with logrank P value shown.

[0059] Figure 14. Experimental workflow and clinical characteristics of the Manchester RCC cohort. (A) A workflow illustrating the methodology applied to the in-house ccRCC cohort. (B) Table summarising patient clinical characteristics with number and percentage of total samples. aPD(L)1 refers to treatment with either a PD1 or PDL1 blocking antibody. TKI = tyrosine kinase inhibitor. Here, aVEGF refers to bevacizumab.

[0060] Figure 15. Receiver operating characteristic curve for tumour-type specific logistic regression model A (Table 3) showing false positive rate (x axis) against true positive rate (y axis). Figure 16. AUC values in the training (x axis) and test (y axis) CPU 000 datasets for the elastic net classification model trained on PTI genes, clonal_TMB and CXCL9 where gene expression levels were normalised (A) according to the tumour type (tumour type specific dataset model) or (B) agnostic to the tumour type (model B). Points indicate models trained with different values of the a hyperparameter between 0 and 1 . (C) Training and test AUC values from models trained to predict immunotherapy response in the CPU 000 dataset with only PTI genes as features. The optimal model had an a hyperparameter of 0.2 and 7 important features (AREG, CXCL1 , CXCL2, EREG, HAS2, IL1 A and PTGS2). Values of a are shown next to the corresponding point (trained model).

[0061] Figure 17. Protocol for RNA extraction and gene expression profiling of formalin-fixed paraffin- embedded (FFPE) surgical resected samples.

[0062] Detailed Description of the Invention

[0063] Aspects and embodiments of the present invention will now be discussed with reference to the accompanying figures. Further aspects and embodiments will be apparent to those skilled in the art. All documents mentioned in this text are incorporated herein by reference.

[0064] In describing the present invention, the following terms will be employed, and are intended to be defined as indicated below.

[0065] The features and advantages of the present disclosure will be more readily understood, by those of ordinary skill in the art from reading the following detailed description. It is to be appreciated that certain features of the disclosure, which are, for clarity, described above and below in the context of separate embodiments, may also be provided in combination in a single element. Conversely, various features of the disclosure that are, for brevity, described in the context of a single embodiment, may also be provided separately or in any sub-combination. It will be understood that in the following, embodiments referred to in relation to one broad aspect of the invention are equally applicable to each of the other broad aspects of the present invention described above. It will be further understood that, unless the context dictates otherwise, the embodiments described below may be combined.

[0066] “and / or” where used herein is to be taken as specific disclosure of each of the two specified features or components with or without the other. For example “A and / or B” is to be taken as specific disclosure of each of (i) A, (ii) B and (iii) A and B, just as if each is set out individually herein. Samples

[0067] A “test sample” as used herein may be a cell or tissue sample (e.g. a biopsy), a biological fluid, an extract (e.g. a protein or DNA extract obtained from the subject). In particular, the sample may be a tumour sample, e.g. a solid tumour such as a gastroesophageal tumour, a melanoma, a bladder tumour or a renal tumour. The sample may be one which has been freshly obtained from the subject or may be one which has been processed and / or stored prior to making a determination (e.g. frozen, fixed or subjected to one or more purification, enrichment or extractions steps).

[0068] In some embodiments, the sample is a formalin-fixed-paraffin-embedded (FFPE) sample. In some embodiments, the sample is a surgical resected sample. In some embodiments, the sample is H&E (haematoxylin and eosin) stained. In preferred embodiments, the sample is a FFPE surgical resected sample. Preferably, the FFPE sample, for example a FFPE surgical resected sample, is H&E stained.

[0069] Gene expression

[0070] Reference to “determining the expression level” refers to determination of the expression level of an expression product of the gene. Expression level may be determined at the nucleic acid level and / or the protein level.

[0071] The gene expression levels determined may be considered to provide an “expression profile”. As used herein, “expression profile” is understood to mean a data set relating to the level of expression of one or more of the relevant genes in an individual, in a form which allows comparison with comparable expression profiles, for example from individuals from whom the prognosis is already know, in order to assist in the determination of prognosis and in the selection of suitable treatment for the individual patient.

[0072] The determination of gene expression levels may involve determining the presence or amount of mRNA in a sample of cancer cells. Methods for doing this are well known to the skilled person. Gene expression levels may be determined in a sample of cancer cells using any conventional method, for example using nucleic acid microarrays or using nucleic acid synthesis (such as quantitative PCR). For example, gene expression levels may be determined using a NanoString nCounter Analysis system (see, e.g., US 7,473,767).

[0073] In some embodiments, the gene expression levels of the at least 8 cancer promoting genes is acquired by measuring the gene expression levels of the at least 8 cancer promoting genes in a sample obtained from the tumour of a mammalian cancer patient.

[0074] Alternatively or additionally, the determination of gene expression levels may involve determining the protein levels expressed from the genes in a sample containing cancer cells obtained from an individual. Protein expression levels may be determined by any available means, including using immunological assays. For example, expression levels may be determined by immunohistochemistry (IHC), Western blotting, ELISA, immunoelectrophoresis, immunoprecipitation, flow cytometry, mass cytometry and immunostaining. Using any of these methods it is possible to determine the relative expression levels of the proteins expressed from the genes listed in Table 1 .

[0075] In some embodiments, the gene expression level of each of said cancer promoting genes is a normalised gene expression level.

[0076] Gene expression levels and the PTI score derived therefrom as detailed herein may be compared with the expression levels and corresponding score of the same genes in cancers from a group of patients whose survival time and / or treatment response is known. The patients to which the comparison is made may be referred to as the ‘control group’. Accordingly, the determined gene expression levels and PTI score may be compared to the expression levels in a control group of individuals having cancer. The comparison may be made to expression levels determined in cancer cells of the control group. The comparison may be made to expression levels determined in samples of cancer cells from the control group. The cancer in the control group may be the same type of cancer as in the individual. For example, if the expression is being determined for an individual with renal cancer, the expression levels and PTI score may be compared to the expression levels and PTI score in the cancer cells of patients also having renal cancer.

[0077] Other factors may also be matched between the control group and the individual and cancer being tested. For example, the stage of cancer may be the same, the immune subtype (as determined by Thorsson et al. 2018) of the cancer may be the same, the subject and control group may be age-matched and / or gender matched.

[0078] Additionally, the control group may have been treated with the same form of surgery and / or same anti-cancer treatment, for example a chemotherapeutic treatment.

[0079] Clonal TMB

[0080] Clonal tumour mutational burden (TMB) is a measure of the somatic mutations estimated to be present across every tumour cell. Clonal TMB is advocated to be a pan-cancer predictor of checkpoint blockage outcome. Clonal TMB measurements were obtained from Litchfield et al., 2021.

[0081] PTI Signature

[0082] A bioinformatic pipeline was used to discover the PTI gene signature (Fig. 1 B). TCGA (the cancer genome atlas) gene expression data in the form of raw counts was downloaded from the SAGE synapse database (syn4311114), alongside clinical data (syn12026747). Raw counts were TMM-normalised using calcNorm Factors in edgeR (3.24.3 / Genes that failed to reach 0.25 Counts-Per-Million (CPM) in 10% of samples were removed from the analysis. Patients belonging to individual TCGA studies were stratified into PTGS2 high and low sub-groups using the top and bottom quartiles of Iog2-normalised expression levels. Non-solid tumor types e.g., LAML (acute myeloid leukaemia) and DLBC (diffuse large B cell lymphoma) were excluded from further analysis. A model matrix was designed to compare PTGS2 high and low patients whilst adjusting for cancer type. This was done using the top ten cancer types by median PTGS2 levels (LUAD, BLCA, LUSC, HNSC, PRAD, PAAD, ESCA, STAD, CESC, KICH). DGE analysis was carried out with limma-trend (P< 0.05, FC> 1.2) (Ritchie etal., 2015). The 54 upregulated genes were filtered according to a list of ligand:receptor interactions (Ramilowski et al., 2015), and the resulting 16 genes were used as the final PTI signature (Table 1). *NCBI Gene ID (version as of 31 January 2023). Available at https: / / www.ncbi.nlm.nih.gov / gene . The nucleotide for each gene is disclosed at the NCBI gene ID number on 31 January 2023 is expressly incorporated herein by reference.

[0083] PTI Score

[0084] As used herein, the terms “gene expression score”, “score” and “PTI score” are used interchangeably. The terms “cancer promoting genes” and “pro-tumorigenic inflammatory genes” and “PTI genes”, as used herein, can be used interchangeably.

[0085] The cancer promoting genes or pro-tumorigenic inflammatory (PTI) genes as described in Table 1 were found to have tumour gene expression which positively correlates with PTGS2 expression, tumour growth, and poor treatment response to immunotherapy. The present inventors found that computing a score of the gene expression of said cancer promoting genes enhances the predictive power relative to that of previous prediction signatures under the particular condition assessed (see Figure 6F for example), including the earlier COX-IS ratio as disclosed in WO 2019 / 243567. As used herein, the term “score” such as in “PTI score” is intended to have a broad meaning, not only encompassing the additive value of each individual gene expression value, but to also include any relationship between features, for example that combines the gene expression values of said cancer promoting genes.

[0086] In some embodiments, the at least 8 cancer promoting genes comprise HAS2 and / or EREG. In some embodiments, the at least 8 cancer promoting genes comprise HAS2, EREG, CXCL8 (IL8) and / or VCAN. In some embodiments, the at least 8 cancer promoting genes comprises HAS2, EREG, CXCL1 , CXCL2 and / or AREG.

[0087] In some embodiments, the gene expression of at least 8, at least 9, at least 10, at least 11 , at least 12, at least 13, at least 14, at least 15, and / or or all (i.e. all 16) of the cancer promoting genes listed in Table 1 may be measured in a sample obtained from the tumour of a patient. The gene expression levels of the at least 8 PTI genes listed in table 1 may be utilised to compute a PTI score of gene expression in a sample obtained from the tumour of a patient. Said PTI score may be used to make a prediction of the treatment response and / or posttreatment prognosis of the patient based on the gene expression levels of the at least 8 PTI genes.

[0088] PTI scores calculated with gene expression data from a reduced subset of the PTI gene set may have an equal, or even greater predictive power than a PTI score calculated with gene expression data from all 16 of the PTI genes. The optimal reduced PTI gene set may vary between clinical settings. In some embodiments, the PTI score (also known as the “score”) is calculated by z-score normalising individual gene expression values of the at least 8 PTI genes and computing or calculating the mean of the normalised measurements.

[0089] In some embodiments, the PTI score is calculated or computed according to the formula:

[0090] PTI score wherein npis the number of said cancer promoting genes, Gosis the positive correlated genes, within an (i) interval of unitary values, and (e) represents the gene expression values expressed as Iog2 counts per million (CPM).

[0091] In some embodiments Gipos(e) is the mean expression of Iog2 transformed counts per million (reads per kilobase million (FPKM)) of positive genes.

[0092] In other embodiments the PTI score is calculated or computed as the weighted sum of z-score normalised gene expression values of the at least 8 PTI genes plus the intercept. The probability of response to immunotherapy can then be computed by using the logistic (sigmoid) function as per the equations below where Po is the intercept and i-nrefers to the coefficients for the at least 8 PTI genes.

[0093] PTI score = o+ Pi Xi + ...+ Pr

[0094] Pr(response to ICB) = 1 / 1 + e_pTl score

[0095] Expression values may be expressed in, for example, any of RPKM (reads per kilobase million), FPKM (fragments per kilobase million), CPM (counts per million) and / or nanostring counts.

[0096] Gene expression-based clarification methods

[0097] In some embodiments, the present invention provides methods for classifying, prognosticating, or monitoring cancer in subjects. In particular, data obtained from analysis of gene expression may be evaluated using one or more pattern recognition algorithms. Such analysis methods may be used to form a predictive model, which can be used to classify test data. For example, one convenient and particularly effective method of classification employs multivariate statistical analysis modelling, first to form a model (a "predictive mathematical model") using data ("modelling data") from samples of known subgroup (e.g., from subjects known to have a particular cancer prognosis subgroup: high risk and low risk), and second to classify an unknown sample (e.g., "test sample") according to subgroup.

[0098] Pattern recognition methods have been used widely to characterise many different types of problems ranging, for example, over linguistics, fingerprinting, chemistry and psychology. In the context of the methods described herein, pattern recognition is the use of multivariate statistics, both parametric and non-parametric, to analyse data, and hence to classify samples and to predict the value of some dependent variable based on a range of observed measurements. There are two main approaches.

[0099] The first method of pattern recognition is termed "unsupervised"; these simply reduce data complexity in a rational way and also produce display plots which can be interpreted by the human eye. However, this type of approach may not be suitable for developing a clinical assay that can be used to classify samples derived from subjects independent of the initial sample population used to train the prediction algorithm.

[0100] The second method of pattern recognition is termed "supervised" whereby a training set of samples with known class or outcome is used to produce a mathematical model which is then evaluated with independent validation data sets. Here, a "training set" of gene expression data is used to construct a statistical model that predicts correctly the "subgroup" of each sample. This training set is then tested with independent data (referred to as a “test set” or “validation set”) to determine the robustness of the computer-based model. These models are sometimes termed "expert systems," but may be based on a range of different mathematical procedures such as support vector machine, decision trees, k-nearest neighbour and naive Bayes. Supervised methods can use a data set with reduced dimensionality (for example, the first few principal components), but typically use unreduced data, with all dimensionality. In all cases the methods allow the quantitative description of the multivariate boundaries that characterise and separate each subtype in terms of its intrinsic gene expression profile. It is also possible to obtain confidence limits on any predictions, for example, a level of probability to be placed on the goodness of fit. The robustness of the predictive models can also be checked using cross- validation, by leaving out selected samples from the analysis.

[0101] As used herein, the term “training set” refers to a portion of a data set which is used to fit or train a model for prediction or classification of values that are known in the training set, but unknown in (future) test data. In some embodiments, the training set comprises gene expression levels of the at least 8 cancer promoting genes of at least 50, at least 100, at least 150, at least 200, at least 250, at least 300, at least 350, at least 400, at least 500, at least 600, at least 700, at least 800, at least 900 samples derived from mammalian cancer patients. In some embodiments, a training set may be a data set from the cancer genome atlas (TCGA). In some embodiments, the training set may the CPU 000 dataset (Litchfield et al., 2021 ) which comprises a collection of datasets from patients treated with ICB (anti-PD(L) 1 , anti-CTLA4) across multiple cancer types with whole exome sequencing (WES) and RNA sequencing (RNAseq) data.

[0102] After stratifying the training samples according to subtype, a centroid-based prediction algorithm may be used to construct centroids based on the expression profile of the gene sets described in Table 1. "Translation" of the descriptor coordinate axes can be useful. Examples of such translation include normalization and mean-centering. "Normalization" may be used to remove sample-to- sample variation. Some commonly used methods for calculating normalization factor include: (i) global normalization that uses all genes on the microarray or nanostring codeset; (ii) housekeeping genes normalization that uses constantly expressed housekeeping / invariant genes; and (iii) internal controls normalization that uses known amount of exogenous control genes added during hybridization.

[0103] In one embodiment, the expression level of each of the PTI cancer promoting genes listed in Table 1 can be normalised, for example to one or more control housekeeping genes. Exemplary housekeeping genes include ACTB (60), GAPDH (2597) and TBP (6908), the numbers in brackets following each gene name being the NCBI Gene ID number for that gene; the nucleotide sequence for each gene as disclosed at that NCBI Gene ID number on 31 January 2023 is expressly incorporated herein by reference. It will be understood by one skilled in the art that the methods disclosed herein are not bound by normalization to any particular housekeeping genes, and that any suitable housekeeping gene(s) known in the art can be used.

[0104] Many normalization approaches are possible, and they can often be applied at any of several points in the analysis. In one embodiment, microarray data is normalised using the LOWESS method, which is a global locally weighted scatterplot smoothing normalization function. In another embodiment, qPCR and NanoString nCounter analysis data is normalised to the geometric mean of set of multiple housekeeping genes. Moreover, qPCR can be analysed using the fold-change method.

[0105] "Mean-centering" may also be used to simplify interpretation for data visualisation and computation. Usually, for each descriptor, the average value of that descriptor for all samples is subtracted. In this way, the mean of a descriptor coincides with the origin, and all descriptors are "centered" at zero. In "unit variance scaling,” data can be scaled to equal variance.

[0106] Usually, the value of each descriptor is scaled by 1 / StDev, where “StDev” is the standard deviation for that descriptor for all samples. In another embodiment, z-score normalisation can be used to normalise every value in the dataset such that the mean of all values is zero and the standard deviation is equal to 1 z-score. “Pareto scaling" is, in some sense, intermediate between mean centering and unit variance scaling. In pareto scaling, the value of each descriptor is scaled by l / sqrt(StDev), where “StDev” is the standard deviation for that descriptor for all samples and “sqrt” denotes square root. In this way, each descriptor has a variance numerically equal to its initial standard deviation. The pareto scaling may be performed, for example, on raw data or mean centered data.

[0107] "Logarithmic scaling" may be used to assist interpretation when data has a positive skew and / or when data spans a large range, e.g., several orders of magnitude. Usually, for each descriptor, the value is replaced by the logarithm of that value. In "equal range scaling,” each descriptor is divided by the range of that descriptor for all samples. In this way, all descriptors have the same range, that is, 1 . However, this method is sensitive to presence of outlier points. In "autoscaling,” each data vector is mean-centered and unit variance scaled. This technique is a very useful because each descriptor is then weighted equally, and large and small values are treated with equal emphasis. This can be important for genes expressed at very low, but still detectable, levels.

[0108] When comparing data from multiple analyses (e.g., comparing expression profiles for one or more test samples to the centroids constructed from samples collected and analysed in an independent study), it will be necessary to normalise data across these data sets. In one embodiment, Distance Weighted Discrimination (DWD) is used to combine these data sets together (Benito et al. (2004) Bioinformatics 20(1): 105-114, incorporated by reference herein in its entirety). DWD is a multivariate analysis tool that is able to identify systematic biases present in separate data sets and then make a global adjustment to compensate for these biases; in essence, each separate data set is a multi-dimensional cloud of data points, and DWD takes two points clouds and shifts one such that it more optimally overlaps the other.

[0109] In some embodiments described herein, the prognostic performance of the gene expression score (PTI score) may be assessed utilizing a Cox Proportional Hazards (CoxPH) Model Analysis, which is a regression method for survival data that provides an estimate of the hazard ratio and its confidence interval. The Cox model is a well-recognised statistical technique for exploring the relationship between the survival of a patient and particular variables. This statistical method permits estimation of the hazard (i.e., risk) of individuals given their prognostic variables (e.g., gene expression ratio as described herein). The "hazard ratio" is the risk of death at any given time point for patients displaying particular prognostic variables. CoxPH regression analysis may be performed using the PTI score as an input value and disease stage and / or cancer type as strata.

[0110] Logistic regression models may be calculated using the PTI score as an input value to predict patient’s treatment response outcomes. Said logistic regression models may be calculated for both tumour type agnostic and tumour type specific data.

[0111] In some embodiments, logistic regression models to predict binary patient treatment response outcomes for tumour type agnostic and tumour type specific approaches are determined by: measuring the individual gene expression values for the at least 8 PTI genes, adding a pseudocount of 1 before Iog2-normalising each value, then either z-score normalising all samples in the dataset (“tumour type agnostic”), or grouping samples by cancer type (“Tum_Type”) and z-score normalising by cancer type (“tumour type specific”), then taking the mean of the scaled expression data for the at least 8 PTI genes in the “tumour type agnostic” or “tumour type specific” groups to calculate a reference gene expression score. In some embodiments, patients who achieved stable disease as best RECIST response (metadata feature: Response_RECIST) are removed from the PTI score calculation. In some embodiments, ‘on treatment’ samples (metadata feature: PriorJO) are removed from the PTI score calculation.

[0112] The PTI scores predictive power outperforms that of known resistance signatures. The claimed PTI scores power to predict poor overall survival in ICB-treated patients is greater than that of know resistance signatures, for example the earlier COX-IS ratio as disclosed in WO 2019 / 243567. As exemplified in Figure 6F and discussed in Example 11 , the pan-cancer survival analysis of the PTI score (HR 1.38 [1.15-1.65], P=0.0004) outperforms the COX-IS ratio (HR 1.18 [0.96-1.45], P=0.12).

[0113] Predicting Treatment Response

[0114] An individual grouped with the good prognosis group, may be identified as having a cancer that is sensitive to immunotherapy, e.g. immune checkpoint blockade (ICB) therapy. They may also be referred to as responders, an individual that responds well to immunotherapy, or an individual that has partial response (PR) or complete response (CR) to immunotherapy (collectively termed PR / CR), such as immune checkpoint blockade (ICB) therapy.

[0115] An individual grouped with the poor prognosis group, may be identified as having a cancer that is resistant to immunotherapy, such as immune checkpoint blockade (ICB) therapy. They may be referred to as non-responders, an individual that responds poorly to immunotherapy, an individual that experienced no benefit from immunotherapy, or an individual that has progressive disease (PD) following immunotherapy, such as immune checkpoint blockade (ICB) therapy.

[0116] Where the individual is grouped with the good prognosis group, the individual may be selected for treatment with suitable immunotherapy (e.g. immune checkpoint blockade therapy) as described in further detail below. Where the individual is grouped with the poor prognosis group, the individual may be deselected for treatment with the aforementioned immunotherapy and may, for example, receive surgical treatment, radiotherapy and / or another form of anticancer agent (e.g. one or more non-immune chemotherapeutic agents or anti-angiogenic agents).

[0117] In some embodiments, a computer-implemented method for predicting the treatment response of a mammalian cancer patient to anti-cancer immunotherapy, the method comprising: a) providing the gene expression data comprising gene expression levels of at least 8, at least 9, at least 10, at least 11 , at least 12 or more (such as all) of the following cancer promoting genes PTGS2, CXCL1 , CXCL2, CXCL5, CXCL6, IL1 A, HAS2, IL11 , IL1 B, EREG, IL6, CXCL8 (IL8), MMP1 2, NRG1 , VCAN, AREG previously measured in a sample obtained from the tumour of a mammalian cancer patient; b) computing a score (PTI score) of the gene expression of the at least 8 or more cancer promoting genes provided in a); and c) predicting the treatment response of the mammalian cancer patient based on the PTI score computed in step b). In some embodiments, the PTI score is computed based on the gene expression levels of all 16 of the cancer promoting genes.

[0118] In some embodiments, the computer-implemented method further comprises assessing the tumour burden and / or neoantigen prevalence of the cancer patient.

[0119] In some embodiments, predicting the treatment response of a mammalian cancer patient comprises: comparing the computed PTI score with a reference median gene expression score derived from a sample cohort of responders and a reference median gene expression score derived from a sample cohort of non-responders, wherein the sample cohort of responders and non-responders are of i) cancer patients having the same type of cancer as said cancer patient, or ii) cancer patients which are tumour type agnostic.

[0120] In some embodiments, where the PTI score indicates that the cancer is predicted to respond to anti-cancer immunotherapy, the mammalian cancer patient is selected for anti-cancer immunotherapy.

[0121] In some embodiments, a method of predicting treatment response of a mammalian cancer patient to anti-cancer immunotherapy, the method comprising: a) measuring the gene expression of at least 8 cancer promoting genes selected from: PTGS2, CXCL1 , CXCL2, CXCL5, CXCL6, IL1 A, HAS2, IL11 , IL1 B, EREG, IL6, CXCL8 (IL8), MMP12, NRG1 , VCAN, AREG in a sample obtained from the tumour of a patient; b) determining a PTI score of the gene expression of said at least 8 cancer promoting genes based on the gene expression score measured in a); and c) predicting the treatment response of the patient based on the gene expression levels determined in step b).

[0122] In some embodiments, determining the PTI score to predict treatment response of a subject comprises: inputting data representing the gene expression levels of the at least 8 cancer promoting genes of the subject into a machine learning classifier, wherein said machine learning classifier has been trained on a training data set comprising the gene expression levels of at least the same cancer promoting genes as those forming the PTI score of the subject, and wherein the training data set is derived from at least 2 sample cohorts; wherein one cohort comprises a plurality of samples derived from a plurality of mammalian cancer patients known to have responded to anti-cancer immunotherapy (responders), and the other cohort comprises a plurality of samples derived from mammalian cancer patients known not to respond to anticancer immunotherapy (non-responders); and causing the machine learning classifier to classify the PTI score of the subject as belonging to the response group or the non-response group. In some embodiments, the training data set comprises gene expression levels of the at least 8 cancer promoting genes of at least 50, at least 100, at least 150, at least 200, at least 250, at least 300 samples derived from mammalian cancer patients known to have responded to anticancer immunotherapy (known as responders) and gene expression levels of the at least 8 cancer promoting genes of at least 50, at least 100, at least 150, at least 200, at least 250, at least 300, at least 350, at least 400, at least 450, at least 500, at least 550 samples derived from mammalian cancer patients known not to have responded to anti-cancer immunotherapy (known as non-responders).

[0123] In some embodiments, the machine learning classifier has been trained on a plurality of PTI scores derived from at least three sample cohorts, wherein one cohort comprises a plurality of samples derived from a plurality of mammalian cancer patients known to have completely responded to anti-cancer immunotherapy (known as complete responders), one cohort comprises a plurality of samples derived from a plurality of mammalian cancer patients known to have partially responded to anti-cancer immunotherapy (known as partial responders), and the other cohort comprises a plurality of samples derived from mammalian cancer patients known not to respond to anti-cancer immunotherapy (known as non-responders).

[0124] Methods of classifying a sample comprising nucleic acid and / or protein from a mammalian cancer patient comprise; analysing the sample to obtain the gene expression levels of the at least 8 cancer promoting genes of the mammalian cancer patient sample, analysing the gene expression levels of the at least 8 cancer promoting genes to determine a PTI score, and analysing the sample PTI score to classify the sample as matching the PTI score of an anticancer immunotherapy response score or an anti-cancer immunotherapy non-response score, wherein the sample is characterised as being derived from an anti-cancer immunotherapy responsive subject if the PTI score is classified as matching the PTI score of the anti-cancer immunotherapy response score and is characterised as being derived from an anti-cancer immunotherapy non-responsive subject if the PTI score is classified as matching the PTI score of the anti-cancer immunotherapy non-response score. Optionally, the sample may be derived from one or more cancer cells of the mammalian cancer patient.

[0125] A system for predicting the treatment response of mammalian cancer patient to an anti-cancer immunotherapy is also provided. The system comprises at least one processor and at least one non-transitory computer readable medium containing instructions that when executed by the at least one processor, cause the at least one processor to: predict the treatment response of the mammalian cancer patient to anti-cancer immunotherapy and / or classify a sample obtained from a mammalian cancer subject as a PTI score of a responder or non-responder to anticancer immunotherapy. In some embodiments, the invention comprises one or more computer readable media comprising instructions that, when executed by one or more processors, cause the one or more processors to predict the treatment response of the mammalian cancer patient to anti-cancer immunotherapy and / or classify a sample obtained from a mammalian cancer subject as a PTI score of a responder or non-responder to anti-cancer immunotherapy. Whether a prognosis is considered good or poor may vary between cancers and stage of disease. In general terms a good prognosis is one where the overall survival (OS) and / or progression-free survival (PFS) is longer than average for that stage and cancer type. A prognosis may be considered poor if PFS and / or OS is lower than average for that stage and type of cancer. The average may be the median survival OS or PFS. As used herein, the term “overall survival” (OS) may be interpreted as the length of time, either from the date of diagnosis or the start of treatment for a disease, such as cancer, that a patient diagnosed with said disease is still alive. As used herein, the term “progression free survival” (PFS) may be understood to mean the length of time during and after the treatment of a disease that a patient lives without a deterioration in said disease, such as cancer.

[0126] In some embodiments, the machine learning classifier is a logistic regression model.

[0127] For example, a prognosis may be considered good if the PFS is > 6 months and / or OS > 18 months. Similarly, PFS of < 6 months or OS of < 18 months may be considered poor. In particular, PFS of > 6 months and / or OS of > 18 months may be considered good for advanced cancers.

[0128] In general terms, a “good prognosis” is one where survival (OS and / or PFS) of an individual patient can be favourably compared to what is expected in a population of patients within a comparable disease setting, such as a similar stage and cancer type. This might be defined as better than median survival (i.e. survival that exceeds that of 50% of patients in population).

[0129] In general terms, a “poor prognosis” is one where survival (OS and / or PFS) of an individual patient can be unfavourably compared to what is expected in a population of patients within a comparable disease setting, such as a similar stage and cancer type. This might be defined as worse than median survival (i.e. survival that does not exceed that of 50% of patients in population).

[0130] “Predicting the response of a cancer patient to a selected treatment” is intended to mean assessing the likelihood that a patient will experience a positive or negative outcome with a particular treatment.

[0131] As used herein, “indicative of a positive treatment outcome” refers to an increased likelihood that the patient will experience beneficial results from the selected treatment (e.g. reduction in tumour size, ‘good’ prognostic outcome, improvement in disease-related symptoms and / or quality of life).

[0132] “Indicative of a negative treatment outcome” is intended to mean an increased likelihood that the patient will not receive the aforementioned benefits of a positive treatment outcome (e.g. no reduction in tumour size, ‘poor’ prognostic outcome, no improvement in disease-related symptoms and / or quality of life). In some embodiments, the PTI score of a cancer patient may be referenced to or compared with a reference median gene expression score derived from a sample cohort of cancer patients having the same type of cancer as said cancer patient (and optionally age-matched, matched for the time since diagnosis and / or disease stage) to predict treatment response and / or prognosis of said cancer patient (“tumour type specific” prediction).

[0133] In other embodiments, the PTI score of a cancer patient may be compared with a reference median gene expression score derived from a sample cohort of cancer patients with the same and different types of cancer as said patient (“tumour type agnostic” prediction) (and optionally age-matched, matched for the time since diagnosis and / or disease stage).

[0134] Treatment

[0135] Methods of treating a cancer in a mammalian cancer patient may comprise: carrying out a method of predicting the treatment response of the mammalian cancer patient to anti-cancer immunotherapy according to the methods of the present invention, determining that the mammalian cancer patient is predicted to respond to said anti-cancer immunotherapy and administering a therapeutically effective amount of anti-cancer immunotherapy to the subject in need thereof. Methods of treatment of a cancer in a mammalian cancer patient may comprise: carrying out the method of predicting the treatment response of the mammalian cancer patient to anti-cancer immunotherapy according to methods of the present invention, determining that the PTI score of the mammalian cancer patient indicates that the mammalian cancer patient is predicted to respond to anti-cancer immunotherapy; and administering immunotherapy to the mammalian cancer patient in need thereof. Some methods of treating a mammalian cancer patient may further comprise analysing one or more additional markers of anti-cancer immunotherapy response derived from the mammalian cancer patient to supplement and / or corroborate the prediction of anti-treatment response or non-response.

[0136] A pharmaceutical composition is provided, comprising an anti-cancer immunotherapy for use in the treatment of a mammalian cancer patient, wherein the patient has been predicted to respond to said anti-cancer immunotherapy.

[0137] The mammalian cancer patient may have, or the cancer may be, any one of the following cancer types: renal cancer, such as renal cell carcinoma; gastroesophageal tumour; melanoma; bladder cancer; gastric cancer; lung cancer, such as lung adenocarcinoma (LUAD); pancreatic cancer (PAAD); liver cancer (LIHC); cervical cancer (CESC); head and neck cancer (HNSC); thyroid cancer (THCA); uterine cancer (UCEC); and Tenosynovial giant cell tumor (TGCT). The renal cell carcinoma may be clear cell renal cell carcinoma (ccRCC). The mammalian cancer may have, or the cancer may be, any one of the following immune subtypes as defined by Thorsson et al (2018): wound healing, IFNy-dominant, TGFp-dominant, inflammatory and / or lymphocyte depleted subtype.

[0138] The anti-cancer immunotherapy may comprise IL-2 immunotherapy or immune checkpoint blockade therapy alone, or in combination with anti-VEGF therapies. Optionally, the anti-VEGF therapy may be bevacizumab. The immune checkpoint blockade therapy may comprise any one of: programmed death-1 (PD-1) blockade, programmed death-ligand 1 (PD-L1) blockade and / or cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) blockade. The immune checkpoint blockage therapy may comprise treatment with Nivolumab (an anti-PD-1 blockade), Pembrolizumab (an anti-PD-1 blockade), Atezolizumab (an anti-PD-L1 blockade and / or Ipilimumab (an anti-CTLA-4 blockade).

[0139] The features disclosed in the foregoing description, or in the following claims, or in the accompanying drawings, expressed in their specific forms or in terms of a means for performing the disclosed function, or a method or process for obtaining the disclosed results, as appropriate, may, separately, or in any combination of such features, be utilised for realising the invention in diverse forms thereof.

[0140] While the invention has been described in conjunction with the exemplary embodiments described above, many equivalent modifications and variations will be apparent to those skilled in the art when given this disclosure. Accordingly, the exemplary embodiments of the invention set forth above are considered to be illustrative and not limiting. Various changes to the described embodiments may be made without departing from the spirit and scope of the invention.

[0141] For the avoidance of any doubt, any theoretical explanations provided herein are provided for the purposes of improving the understanding of a reader. The inventors do not wish to be bound by any of these theoretical explanations.

[0142] Any section headings used herein are for organizational purposes only and are not to be construed as limiting the subject matter described.

[0143] Throughout this specification, including the claims which follow, unless the context requires otherwise, the word “comprise” and “include”, and variations such as “comprises”, “comprising”, and “including” will be understood to imply the inclusion of a stated integer or step or group of integers or steps but not the exclusion of any other integer or step or group of integers or steps.

[0144] It must be noted that, as used in the specification and the appended claims, the singular forms “a,” “an,” and “the” include plural referents unless the context clearly dictates otherwise. Ranges may be expressed herein as from “about” one particular value, and / or to “about” another particular value. When such a range is expressed, another embodiment includes from the one particular value and / or to the other particular value. Similarly, when values are expressed as approximations, by the use of the antecedent “about,” it will be understood that the particular value forms another embodiment. The term “about” in relation to a numerical value is optional and means for example + / - 10%.

[0145] The following is presented by way of example and is not to be construed as a limitation to the scope of the claims.

[0146] Examples

[0147] Materials and Methods

[0148] Data and code availability

[0149] Transcriptomic data from treatment of mouse bone marrow-derived macrophages with synthetic PGE2 in vitro will be made available via GEO (https: / / www.ncbi.nlm.nih.gov / geo / ). Suitable code for carrying out analyses is available at https: / / gitlab.com / christian.bromley / protumorigenic-inflammation.

[0150] Statistical methods

[0151] R version 4.0.3 was used for all in silico analysis. All statistical tests were carried out using R or, in some cases, within GraphPad Prism (7.0d). Unless otherwise stated, P values less than 0.05 (*), 0.01 (**) and 0.001 (***) are labelled. The Benjamini-Hochberg method was used to adjust P values for multiple hypothesis testing. These are referred to throughout as FDR values. Two-sided, non-parametric tests were used where it was not reasonable to assume a normal distribution, x2tests were used to compare categorical variables. Where there were more than two groups, Kruskal-Wallis tests were used with Dunn’s multiple comparisons test including P value adjustment for multiple hypothesis testing using FDR. For univariate survival analysis, we used Wald test P values and log rank test P values to assess statistical significance. For multivariate survival analyses, we harvested Wald test P values for individual covariates using the survival R package (3.1-12) coxph function.

[0152] TGCA differential gene expression and network analysis

[0153] TCGA (the cancer genome atlas) gene expression data in the form of raw counts was downloaded from the SAGE synapse database (syn4311114), alongside clinical data (syn 12026747). Raw counts were TMM-normalised using calcNormFactors in edgeR (3.24.3 / Genes that failed to reach 0.25 Counts-Per-Million (CPM) in 10% of samples were removed from the analysis. Patients belonging to individual TCGA studies were stratified into PTGS2 high and low sub-groups using the top and bottom quartiles of Iog2-normalised expression levels. Non-solid tumor types e.g., LAML (acute myeloid leukaemia) and DLBC (diffuse large B cell lymphoma) were excluded from further analysis. A model matrix was designed to compare PTGS2 high and low patients whilst adjusting for cancer type. This was done using the top ten cancer types by median PTGS2 levels. DGE analysis was carried out with / / mma-trend^Ritchie et al., 2015). The upregulated genes (P< 0.05, FC> 1 .2) were filtered according to a list of ligands (Ramilowski et al., 2015), and the resulting 16 genes were used as the final PTI signature (Table 1).

[0154] For gene set enrichment analysis (GSEA) we ranked genes based on directional log P values. Ranks were then used as input to the fgsea (1 .8.0) package (500 iterations). Hallmark gene signatures were downloaded from the Molecular Signature Database (https: / / www.gsea- msigdb.org / gsea / msigdb / ). Upregulated genes from the DGE analysis were imported into the STRING database (https: / / string-db.org) to construct protein-protein interaction networks. Nodes without any edges were removed. A table with significantly enriched gene ontology biological processes was exported from STRING. Networks were also exported and analyzed in R using the igraph package (1 .2.7).

[0155] TCGA survival analysis

[0156] Signature scores for gene signatures of interest were calculated using the mean expression of signature genes with Z-score normalised expression data in all cases. The Hmisc package (4.3.0) rcorr function was used to calculate pairwise Spearman correlation coefficients. Survival analysis was carried out in R using the survival (3.1 -12) package and KM graphs plotted using survminer (0.4.6). For survival analysis we leveraged OS days and OS status information from the pan-TCGA clinical data table that was downloaded from SAGE synapse (syn 12026747). For the pan-cancer survival analysis, we used continuous PTI scores, as well as stratifying into quartiles or by median values within each pan-cancer immune subtype (Thorsson et al., 2018). HRs were adjusted by using cancer type and disease stage as strata in CoxPH models. We used the supplementary file (Supp1 a.txt) from Thorsson et a / (2018) to gain detailed immunological and genomic annotations of TCGA patients. For cancer type-specific survival analyses we used continuous signature scores in multivariate CoxPH models adjusted by available clinical parameters. Patient stage was converted to a continuous variable (for example: stage la and stage lb = 1). To infer TMB in the TCGA data we used ‘non-silent mutation rate’ annotated by Thorsson et a / (2018). For HNSC we extracted the final HPV status of patients from the supplementary data of the TCGA study (Lawrence et al., 2015).

[0157] Analysis of PGE2-treated cells

[0158] Mouse BMDMs were generated as previously described (Zelenay et al., 2015). Macrophages were treated with synthetic PGE2 for 24 hours before RNA was isolated and sequenced.

[0159] Reads from the RNA sequencing experiments were quality checked using FastQC (http: / / www.bioinformatics.bbsrc.ac.uk / projects / fastqc / ) and bases were trimmed. Next, we aligned reads to the mouse genome (mm10) using STAR (Dobin et al., 2013) with default parameters. The featureCounts function within the R package Flsubread (Liao, Smyth and Shi, 2014) was then used to generate gene level count data using the Ensembl GTF annotation (Mus_musculus.GRCm38.75.gtf). The DESeq2 (1 .24.0) package was utilized for DGE analysis, and DEGs were defined as those with an FDR < 0.05 and 1 < log2FC < -1 . Publicly available microarray datasets from the gene expression omnibus (GEO) were downloaded for human fibroblasts (GSE63659), and CD4+T cells (GSE17200) treated with PGE2. We performed DGE analysis using the Umma package to fit linear models with array weights. We defined significant DEGs as FDR < 0.05 and 1 < log2FC < -1 .

[0160] TCGA mutation analysis

[0161] A pan-cancer, mutation annotation file (mc3.v0.2.8. PUBLIC. maf.gz) was downloaded from SAGE synapse (syn7214402). R package maftools (2.2.10) was used for plotting and data exploration. Mutation data was filtered to exclude non-solid tumor types, and to exclude variants classified as RNA, Silent, 3’Flank or 5’Flank. For each patient sample we calculated the total number of mutations, and performed a Iog2-normalisation. Multivariate linear regression with the following formula: lm(PTI ~ mutation + Iog2-total mutations + cancer type) was used to identify pan-cancer predictors of PTI expression. The same formula was applied to individual cancer types for a selection of known cancer driver genes(Lawrence et al., 2014). P values from linear models were adjusted by the Benjamini-Hochberg method. For the KRAS gene we downloaded pan-cancer TCGA mutation, structural variant and copy number estimates from cBioPortal (https: / / www.cbioportal.org).

[0162] CCLE PTI expression analysis

[0163] Gene expression data (CCLE_RNAseq_rsem_genes_tpm_20180929.txt), and cell line annotations (Cell_lines_annotations_20181226.txt) were downloaded from the cancer cell line encyclopedia (CCLE) (https: / / portals.broadinstitute.org / ccle / data). EnsDb.Hsapiens.v79 and ensembldb R packages were used to convert transcript IDs to gene symbols. Cell lines named either as “FIBROBLAST” or “HAEM” were discarded. TPM values were transformed to the Iog2 scale with a pseudocount of 1 . Pairwise Spearman correlation values were computed between PTI genes, and mean Spearman values were calculated for the pairwise correlation coefficients of PTI genes, and separately for 1000 random gene signatures of equal length to the PTI signature (n= 16). MMP12 was not present in the CCLE dataset. To compare cell lines with high and low cancer cell-intrinsic PTGS2 expression we compared the top and bottom quartiles using edgeR (3.28.1) for DGE analysis. Genes with greater than 1 CPM in 10% of samples were retained for analysis. Cancer type, as defined by the TCGA code variable in the cell line annotations file, was adjusted for in the modelling. We defined upregulated DEGs as those with adjusted P value < 0.05 and log2FC> 2 Using the same pipeline as for PTGS2, we compared PTIhighand PTIlowcell lines. Highly upregulated genes from the PTI comparison (FC>5, FDR<0.05) were filtered to remove long non-coding RNAs and pseudogenes. With the final list of upregulated genes (n= 500) we performed motif enrichment analysis using the Rcistarget (1 .6.0) package. We searched for conserved motifs in the 500 base pairs immediately upstream of the TSSs of our genes of interest using the motif rankings file hg19-500bp- upstream-10species.mc9nr. feather downloaded from https: / / resources.aertslab.org / cistarget / .

[0164] CCLE mutation and somatic copy number analysis

[0165] Mutation calls (CCLE_DepMap_18q3_maf_20180718.txt) and copy number by gene estimates (CCLE_copynumber_byGene_2013-12-03.txt) were downloaded from the DepMap portal (https: / / depmap.org / portal / download), Information regarding cell line aneuploidy and chromosomal arm-level copy number were extracted from Supplementary Table 1 of a recent study (Cohen-Sharir et al., 2021 ). For analysis of cell line mutations we discarded the following variant classes: RNA, Silent, 3’Flank and 5’Flank, and filtered to include 568 published cancer driver genes (Martinez-Jimenez et al., 2020). Signature scores for the PTI signature were modelled by multivariate linear regression utilising driver gene mutation status, mutation rate and cancer type as covariates. For cancer type, we converted each level of the TCGA code categorical variable into a series of binary variables (for example: HNSC = 1 , Other = 0). To plot individual mutations in the KRAS gene (RefSeq ID: NM 004985) we used maftools (2.2.10). For gene level SCNA estimates, we fitted multivariate logistic regression models to predict high and low cancer cell-intrinsic PTI score utilising gene copy number, aneuploidy and cancer type covariates. Significant predictors were defined as those with FDR< 0.25.

[0166] CCLE RPPA data analysis

[0167] RPPA data (CCLE RPPA 20181003.csv) was downloaded from the CCLE database. Cell lines were filtered to include only those with a KRAS mutation. PTI high (n= 44) cell lines were then compared to PTI low (n= 18) cell lines using the Umma package. Upregulated proteins and phosphoproteins were defined as those with FC> 1 .5 and FDR< 0.05, whereas downregulated proteins had FC< -1.5, FDR< 0.05.

[0168] Drug screen data analysis

[0169] The R data file (E-MTAB-783.eSet.r) containing gene expression data from cell lines screened by Garnett et al (2012) was downloaded. Utilizing the affy (1.62.0) package rma function microarray data was normalized, and IDs (affy_hg_u133a) were converted to gene symbols using biomaRt (2.40.5). For observations that mapped to a single gene, we selected the row with the highest variance. Individual cell line cancer cell-intrinsic PTI signature scores were calculated as in the CCLE. The top 25% highest expressing cell lines were defined as PTIh'9h. Cell line drug screening data was downloadedfrom the supplementary file of Garnett et al (2012). As in Jerby-Arnon et al (2018), sensitive cell lines were defined for each drug as the top 25% lowest by IC50. Drug cell line sensitivities were modelled as a function of cell line PTI status (high / low) whilst adjusting for cancer type as a covariate within a linear mixed effects model using the Ime4 package (1 .1 -23). The following formula was used: glmer(sensitivity ~ PTIstatus + (1 + cancertype), family = binomial(link=”logit”), control = glmerControl(optimizer=”Nelder_Mead”, optCtrl=list(maxfun=20000))). Drugs that preferentially kill PTIhighlines were inferred by filtering models in which PTI status was a significant predictor of cell line sensitivity (coefficient > 0 and P < 0.05). IC25, IC50, IC75 and IC90 concentrations of the drug refametinib for cell lines of interest were extracted from the supplementary data of Garnett et al (2012).

[0170] TCGA immune deconvolution

[0171] The MCPcounter (Becht et al., 2016) immune deconvolution algorithm was applied to the pancancer TCGA expression dataset. The resulting immune cell scores were correlated with the PTI signature within individual cancer type datasets and Spearman correlation coefficients against the PTI signature were plotted using corrplot (0.9).

[0172] Single cell RNA sequencing data analysis

[0173] All analysis of scRNAseq data was carried out using the Seurat package (3.1 .5). The sctransform package (0.2.1 ) and SCTransform function were used to regress out unwanted sources of variation. The AddModuleScore function was used to compute signature scores in individual cells. Where possible original cluster annotations were utilised. The FindMarkers function was used for DGE analysis between clusters or groups with a cut-off of 25% expression across cells in the cluster of interest. To plot the expression of individual PTI genes in specific clusters, we calculated the mean Iog2-normalised TPM or CPM expression levels across each cluster. Hierarchical clustering was performed on mean expression matrices in order to cluster PTI genes by cell type-specific expression.

[0174] Lambrechts et al (2018) NSCLC:

[0175] We downloaded an R data file containing log2CPM data and t-SNE coordinates from ArrayExpress under accession code E-MTAB-6149. sctransform (0.2.1) was used to regress out total number of counts per cell and mitochondrial content. We re-annotated clusters from the original publication into broad cell types. Dendritic cells were a combination of cells from clusters annotated as "cross-presenting dendritic cells (14)", "Langerhans cells (3)”, "plasmacytoid dendritic cells (24)", "monocyte-derived dendritic cells (19)". Clusters containing “fibroblast” were grouped together as fibroblasts, and the same principle was applied for endothelial cells and cancer cells. Cell types native to the lung were discarded from the analysis. Cells in the CD8+T cell cluster from the tumor of patient 4 ( PTIhigh) were compared with those cells from all other tumors.

[0176] Puram et al (2017) HNSC:

[0177] Expression data and annotations were downloaded as an R data file from Zenodo (https: / / zenodo.org / record / 3260758 / files / hnscc_expression.rds) utilising data collated in a recent study. Cells from lymph node samples (HN10, HN12, HN13, HN24, HN7, HN8, HN23) were discarded. We used identical cell filtering steps as in the original study (Puram etal., 2017). Number of counts per cell, and a variable describing the enzyme used for reverse transcription were regressed out using sctransform. Six PCs were utilised for graph-based clustering and the resulting clusters were annotated using marker genes and annotations from the original study. CD8+T cells were compared between tumors with high and low cancer cell-intrinsic PTI expression using the FindMarkers function for DGE analysis.

[0178] Jerby-Arnon et al (2018) melanoma:

[0179] Gene expression (GSE115978_tpm.csv) and cellular annotations (GSE115978_cell.annotations.csv) were downloaded from GEO (GSE115978). Cells with a minimum of 500 genes were retained, and genes expressed by two cells or more passed filtering. 20 PCs were selected for graph-based clustering. 25 new clusters were defined and annotated using known marker genes and annotations from the original study. Clusters were grouped into broad cellular categories to reflect those in other scRNAseq datasets. All T cells were sub-clustered to define CD4+T cells, naive-like T cells and CD8+T cells, so that CD8+T cells could be compared between tumors with high and low cancer cell-intrinsic PTI expression.

[0180] TCGA CIBERSORT and CD8+T cell signature analysis

[0181] Utilising the pan-solid tumor TCGA expression dataset, we first selected so-called ‘hot’ tumors by filtering to include only those patients with ‘IFNy-dominant immune subtype tumors (Thorsson etal., 2018). We then calculated the PTI signature and several signatures related to CD8+T cell phenotype including signatures of dysfunction (Singer etal., 2016), anergy (Azizi et al., 2018) which is a gene set derived from initial studies of tolerant self-antigen specific CD8+T cells (Schietinger et al., 2012) and exhaustion (Guo et al., 2018) . Spearman correlation coefficients were calculated between CD8+T cell phenotype signatures and the intratumoral PTI score. We harnessed CIBERSORT data from Thorsson et al (2018) to compare the proportion of different immune populations out of total immune cells in PTI high, intermediate and low tumors. We compared the mean proportion of each individual population between PTI high and low tumors using non-parametric T tests with P values adjusted for multiple hypothesis testing.

[0182] Immune checkpoint blockade dataset analysis

[0183] We collated transcriptomic data from patients treated with ICB across 12 publications. For datasets that required normalisation from counts to TPM values, we first filtered out genes that were expressed above 0.25 CPM in less than 10% of samples throughout the entire cohort. For three datasets ((Kim et al., 2018; Mcdermott et al., 2018; Gide et al., 2019) we downloaded FASTQ files. RNAseq reads were quality checked using FastQC software (http: / / www.bioinformatics.bbsrc.ac.uk / projects / fastqc / ) and trimming of bases was performed. Reads were then aligned to the human genome assembly (Homo_sapiens.GRCh38) using the STAR aligner (Dobin et al., 2013) with default parameter settings. Mapped data were converted to transcript level integer read counts using featureCounts (Liao, Smyth and Shi, 2014) and the Ensemble GTF annotation (Homo_sapiens.GRCh38.85.gtf). Transcript counts were converted to TPM.

[0184] For each cohort, patients were stratified into groups of high and low signature score based on median cut-off values. The proportion of responders (PR / CR) and non-responders (PD) was then compared between stratified patient groups using / _2test. SD patients were discarded from this analysis. To model patient response to ICB in each dataset, we fitted multivariate logistic regression models using the TIS signature alone as a baseline predictive model. Each resistance signature was combined with the TIS signature in a multivariate logistic regression model that included the linear multiplicative interaction term of the two variables in the model (e.g. glm(response ~ TIS + PTI + TIS:PTI)). The pROC (1 .17.0.1 ) R package was used to compute AUC values for each model, and AUCs across 16 datasets were compared by nonparametric T tests. We also compared the OS of patients in each of the quadrants using baseline, pre- ICB datasets for which OS data was available. For this pan-cancer immunotherapy meta-dataset, we fitted a multivariate Cox regression model including dataset and TIS status as covariates. 1000 random gene signatures and published signatures of resistance to ICB were subjected to the identical analysis. Wald test P values for each resistance / random signature from each multivariate CoxPH model were adjusted by the Benjamini-Hochberg method. Analysis of the developmental stages of lung squamous cell carcinoma

[0185] Microarray data from different developmental stages of LUSC were downloaded from the GEO database (GSE33479) (Mascaux etal., 2019). Where multiple probes mapped to the same gene, the probe with the highest variance across the cohort of samples was selected. The PTI signature score was calculated as in the TCGA analysis.

[0186] Lung cancer cell line in vitro assays

[0187] Human lung cancer cell lines NCIH1339, and NCIH3255 were cultured in RPMI 1640 medium with L-Glutamine (Lonza: LZBE12-702F) with 10% foetal bovine serum (Merck: F7524) and 5% penicillin / streptomycin (Thermo: 11548876) at 37°C in Nunc T75 cell culture flasks (Life Technologies: 156499). Cells were washed with Dulbecco’s phosphate buffer saline (Fisher: 12559069) suspended using trypsin-EDTA and plated at 1x105cells per well in 12 well plates or 2x104cells per well in 96 well flat-bottomed plates (Starlab: CC7682-7596). After 24 hours, media was removed from plates, and replaced with fresh media containing 0.1% dimethyl sulfoxide as a control or with 1pM or 10pM trametinib or selumetinib. At 1 , 2, 4-, 6-, 8- and 24- hour time-points thereafter, supernatants were taken and stored at -20°C. Cells were suspended with trypsin-EDTA and centrifuged at 1500rpm for five minutes. Supernatants were discarded and cell pellets were stored at -80°C. To monitor and infer cell proliferation, drug- treated 96 well plates were imaged at two-hour intervals for three days using the Incucyte instrument. Percentage confluence, and change in confluence relative to baseline, were used as surrogate markers for cell proliferation.

[0188] To measure soluble cytokines and chemokines in cell culture supernatants we used the Luminex assay protocol. 96-well, flat-bottomed Luminex plates were washed with purified H20. Lyophilised protein standards were made up to SOpd in cell culture medium and centrifuged for 10 minutes at 15000rpm, and left on ice for 10 minutes. One in four serial dilution of protein standards were made in cell culture medium. Beads were diluted 1 :100 in assay wash buffer and 50p.l was added to each well. The plate was then placed on the magnet for two minutes, before liquid was flicked off and samples / standards were plated in 50pJ with a 1 :2 dilution in media. The plate was left to shake for two hours and then washed with assay wash buffer three times. Antibodies were diluted 1 :200 in detection antibody diluent and plated 25pJ per well. The plate was then left to shake for 30 minutes at room temperature, before three more washes were carried out with wash buffer. 30pd streptavidin-PE was added to each well and the plate was left on the shaker for 30 minutes. The plate was then washed four more times in wash buffer, and 75pl reading buffer was added to each well. The plate was then read using the Luminex MAGPIX machine. Protein concentration were extrapolated from the standard curve that was fit by plotting expected concentrations against net MFI levels within GraphPad Prism v8.

[0189] CPU 000 dataset analysis

[0190] We were granted access to individual PTI gene expression, clinical and genomic biomarker data from the CPU 000 dataset (Litchfield et al., 2021 ). Expression data was available for a total of 523 patients. Logs- normalised TPM values (log2(TPM+1 )) for each gene were Z-score normalised, and the PTI signature was calculated as previously. MMP12 was missing from the data leaving 15 / 16 PTI genes present. Continuous variables were co-correlated and Spearman Rho values were used for hierarchical clustering using the default parameters within the corrplot package (0.84) corrplot function. SD patients were again discarded, as well as on-treatment samples. Univariate logistic regression models for individual biomarkers were fitted to predict PD versus PR / CR patients. Effect sizes, standard errors and Wilcoxon test P values from six studies were extracted, and submitted for meta-analysis using the R meta package (4.18-0) and the metagen function. To compare the effect of measuring PTI in combination with other biomarkers we fitted multivariate logistic regression models with 4 variables (tumor type, clonal TMB, CXCL9, PTI) including linear multiplicative interaction terms of the numeric variables. Nested models were then compared using the likelihood ratio test. Variance in patient response explained by each model was estimated by calculating the deviance with the anova function. For survival analysis in the CPU 000 dataset, patients were stratified according to a 75% cut-off value for PTI, and by median values for clonal TMB, CXCL9 and CD274. To determine the optimal combination of individual PTI genes for predicting ICB response in the CPU 000 dataset, we trained several elastic net regression models using the glmnet R package. With the loga- normalised TPM values of the CPU 000 dataset we created two scaled expression matrices. For the first we scaled the data across all of the samples (tumour type agnostic), and for the second we grouped by cancer type and then scaled the data (tumour type specific). Patients with SD were removed, as well as those with on treatment samples. The respective matrices were split into 70% training and 30% test datasets. Using 10-fold cross validation we trained elastic net regression models to optimise the 1 hyperparameter for increasing values of the a hyperparameter between 0 and 1 with increments of 0.1 . For each trained model, we used AUG as the metric to select the optimal model. Each trained model was then used to predict patient ICB response in the test dataset.

[0191] Manchester ccRCC cohort patients

[0192] We gained ethical approval for the retrospective analysis of RCC patient tumor FFPE and blood plasma under the Manchester Cancer Research Centre Biobank Research Tissue Bank Ethics (ref: 18 / NW / 0092). 48 surgical blocks were identified from patients that received immunotherapy at some stage in their treatment course including those treated with high dose IL-2. No tumor tissue was observed upon pathological assessment of H&E-stained sections from three tumors. There were two blocks from the same patient that had two surgeries to remove brain metastases. Another patient had separate blocks from surgeries to remove adrenal gland and lung metastases. A pathologist marked regions of tumor, and healthy tissue, on H&E sections, and defined histologic grade and morphology.

[0193] Formalin-fixed-paraffin embedded (FFPE) tissue processing, RNA extraction and gene expression profiling

[0194] FFPE surgical resected tissues are the most common clinically available specimen and can be used as an input material for molecular inflammatory profiling, for example using the NanoString nCounter platform. H&E (haematoxylin and eosin) stained FFPE ccRCC blocks were sectioned and placed on glass slides. Samples were macro-dissected to exclude undesired tissue such as large parts of necrotic or normal tissue, blood clots and cartilage, and then scraped into 1 .5 mL Eppendorfs. On average 15 x 5p.m tissue sections were used. Smaller samples that did not require macro-dissection were cut straight from the block into 1 .5 mL Eppendorfs. RNA was extracted from a maximum of 10 slides using the commercially available Qiagen RNeasy FFPE kit (RNeasy FFPE Kit for RNA Extraction; RSP-54) (Figure 17). Total RNA concentration was first determined using the NanoDrop instrument (Thermo Fisher). Qubit 2.0 Fluorometer (Invitrogen) RBA quantification measurements were also taken, and quality estimates were calculated using the Bioanalyser (Agilent) platform. RNA RIN values ranged from 1 to 6.4 with a mean of 2.2. The extracted RNA was then used as an input material for molecular inflammatory profiling systems, for example the Nanostring nCounter platform, which are compatible with samples exhibiting low RNA quality and high fragmentation, such as that derived from FFPE tissues.

[0195] Nanostring gene expression analysis

[0196] Gene expression profiling was performed using the Nanostring nCounter protocol and the commercial Pan-Immune codeset including an additional 30 custom genes. For the nCounter workflow, appropriate housekeeper genes were selected and added to the codeset of the PTI genes. To quantify gene expression, 150 ng total RNA was hybridised to capture and reporter probes specific to PTI and housekeeper genes overnight at 65°C. The subsequent reactions, which included hybridised samples and master mix reagents were then supplied to the Nanostring nCounter preparation station which aligns the samples for imaging. The nCounter digital analyzer was then used to scan samples and count fluorescent barcodes. This involved quantifying the target RNA transcripts by directly counting the fluorescent signal from the retained reporter-capture probe complexes on the cartridge surface. Nanostring output data (RCC files) was analyzed within R using the NanoStringQC function within the nanostringr package (0.1.4.9000). This analysis included exporting the resultant gene expression counts into the NanoString nSolver 4.0 software for quality control analysis and normalisation (to internal possible controls and / or housekeeper genes). NanoString nSolver 4.0 software enables rapid and straight forward analysis and correction for batch effects using reference signals without the need for advanced computational skills. This analysis flagged samples that failed to meet the requisite thresholds for data quality (signal-to-noise ratio > 50, percentage of genes detected > 50%). Altogether three samples failed the quality control steps due to low signal-to-noise ratio. A published pipeline for normalisation of Nanostring counts data (Bhattacharya et al., 2020) was followed. Firstly, we used the MASS package (7.3-53) to fit a negative binomial model to each housekeeper gene expression using a binary variable as predictor (primary renal cancer v other). Nine housekeepers were excluded on the basis of P< 0.05. The RUVseq package (1.24.0) was then used to identify and remove unwanted sources of variation with parameter k=1 . Limma (3.46.0) and DESeq2 (1 .30.0) packages were used for DGE analysis. Batch-normalised expression data was then used for downstream correlative analyses. Gene signature scores were then calculated as the mean expression of Z-score normalised, batch-corrected expression data.

[0197] EXAMPLE 1 : PTI: a human-derived COX-2-centered signature of pro-tumorigenic inflammation

[0198] Given the vast amount of literature implicating the COX-2 / PGE2 pathway in fuelling multiple cancer hallmarks including the inventors previous work establishing this inflammatory axis as central to immune escape (Zelenay et al., 2015; Bonavita et al., 2020; Pelly et al., 2021 ), it was hypothesized that transcripts associated with high COX-2 mRNA levels would capture pan-cancer conserved features of a pro-tumorigenic inflammatory signalling network.

[0199] Mining TCGA data, it was found that transcript levels of PTGS2, encoding for COX-2, were heterogeneous between and within cancer types (Figure 1A). Consequently, the inventors devised a computational pipeline that compared PTGS2 high and low tumors from the top ten FTGS2-expressing cancer types (Figure 1 A, Figure 1 B). Among the 54 genes overexpressed in PTGS2 high cancers were classic secreted factors implicated in local and systemic inflammation such as IL6, CXCL8, IL1B and CXCL2, cell surface receptors (CSF3R, IL7R, CD69), neutrophil markers (G0S2) as well as EGFR ligands and growth factors such as AREG and EREG, in addition to ECM modifiers HAS2and MMP12 (Figure 1 B). Gene set enrichment analysis (GSEA) using Hallmark pathway signatures revealed that the TNF-NFKB pathway and other inflammatory pathways (IL6-STAT3) were upregulated in COX-2hi9hcancers (Figure 8A). Interestingly, epithelial-mesenchymal transition (EMT), complement, hypoxia, KRAS signalling and apoptosis pathways were also overexpressed (FDR< 0.05, NES > 1 ) (Figure 8A). Thus, COX-2 expression is a relevant, pan-cancer node for inflamed tumors exhibiting several other cancer hallmarks. Next, it was reasoned that genes encoding extracellular molecules, including secreted factors, would be strong candidate instructive modulators of the cellular and molecular immune TME profile. The upregulated genes in Ptgs2n ntumors were filtered by a list of ‘predicted ligands’ (Ramilowski et al., 2015). The resulting genes were cytokines (IL1A, IL1B, IL6), chemokines (CXCL1, CXCL2, CXCL5, CXCL6, CXCL8), and growth factors (NRG1). This was named the 15 gene signature, together with PTGS2, the protumourigenic inflammatory (PTI) signature.

[0200] Human cancer cell lines with high PTGS2 expression had significantly elevated expression of 13 / 15 of evaluable PTI genes (Figure 8B). Two of the most abundant, non-cancerous cell types occupying the TME are fibroblasts and macrophages. To investigate whether PGE2 signalling towards fibroblasts and macrophages might explain the association of these ligands with COX-2 in tumors, transcriptomic data from cells treated with synthetic PGE2 in vitro were analysed. Notably, each ligand was modulated by PGE2 either in fibroblasts or macrophages (Figure 1 D). A number of ligands were also induced by PGE2 treatment of human CD4+T cells (Figure 1 D). These findings validate our bioinformatic pipeline and suggest that the PTI signature of ligands represents nodes hierarchically downstream of the COX-2 / PGE2 axis that functions as a ‘hub’ in the network.

[0201] To address whether the genes upregulated in COX-2h'9htumors were members of a biologically- meaningful and cohesive signalling cascade, a protein-protein interaction network was constructed using the curated STRING database (Szklarczyk etal., 2019) (Figure 1 C). Consistent with their shared involvement in signalling pathways, proteins encoded by genes upregulated in COX-2h'9htumors had significantly more interactions than would be expected by chance (P< 1 E- 16) with an average of 9.7 edges per node. These proteins were overrepresented in gene ontology biological processes such as the inflammatory response, myeloid cell chemotaxis, response to cell death and stress (Figure 1 D). This data highlights how genes identified in the pipeline are involved in shared processes and signalling cascades with COX-2 at the epicentre of the network.

[0202] EXAMPLE 2: PTI signature as a powerful indicator of poor prognosis

[0203] Next, it was questioned whether the PTI signature was associated with poor prognosis. Using The Cancer Genome Atlas (TCGA) data, a pan-cancer Cox Proportional Hazards (CoxPH) regression analysis was performed utilising disease stage and cancer type as strata. The PTI signature correlated strongly with shortened survival of cancer patients both as a continuous variable (HR= 1.23, P= 1 E-05) and stratifying according to quartiles (PTIhighversus PTIlow, HR= 1.25, P= 0.02) (Figure 1 E).

[0204] EXAMPLE 3: PTI signature is a powerful indicator of poor prognosis across cancer immune subtypes

[0205] The interaction of PTI with signatures of cancer-inhibitory inflammation in predicting cancer patient OS was studied using multivariate CoxPH models including interaction terms (Ayers etal., 2017; Danaher et al., 2017). By including linear multiplicative interaction terms (for example the product of PTI and TIS signatures) in the models, it was estimated whether the effect of PTI on survival depends on the extent of IFNy signaling and T cell inflammation. As anticipated, PTI generally associated with increased mortality (HR= 1 .25, P< 0.0001 ) (Figure 2A). Therefore, for a given patient with mean levels of IFNy signalling, a one value increase in PTI correlates with a 25% increase in risk. By contrast, for a patient with mean levels of PTI, an incremental increase in IFNy signalling corresponded to a 16% decrease in the risk of death (HR= 0.84, P< 0.0001 ) (Figure 2A). There was a significant interaction term for PTI and IFNy signatures (HR= 1.13, P= 0.004). To illustrate this interaction, the hazard rate at high and low expression levels of IFNy signalling was simulated (Figure 2B). At low levels of IFNy signalling the hazard rate was consistently elevated regardless of PTI level, whereas for patients with high IFNy signalling the risk of death was lower increased strongly with PTI expression (Figure 2B). Indeed, at a high level of PTI (for example PTI score= 1.5, Figure 2B) a patient with high IFNy signalling in the TME, that is typically a marker of better prognosis tumors, would have a similar risk of death to a patient with a non-T cell-inflamed tumor. Thus, high intratumoral IFNy signalling correlates with better outcome in patients where PTI levels are simultaneously low. Similar interaction effects were observed between PTI and CD8+T cell and TIS signatures (Table 2). Collectively, these findings uncover an intimate cross-talk between pro-tumorigenic and anti-tumorigenic inflammatory phenotypes highlighting the profound antagonistic effect of the PTI signature towards T cell-inflamed tumors.

[0206]

[0207] The immune infiltrate and TME varies greatly across tumor types, yet previous pan-cancer analysis has defined six broad immune subtypes: ‘Wound healing (C1)’, ‘IFNy-dominant (C2) ‘Inflammatory (C3)’, ‘Lymphocyte-depleted (C4)’, ‘ Immunologically-quiet (C5)’, and ‘TGF / 3- dominant (C6)’ groups (Thorsson et al., 2018). These subtypes were found to have unique genomic, transcriptomic and clinical characteristics (Thorsson et al., 2018). It was hypothesised that PTI would retain prognostic power, and its antagonism towards anti-tumor immunity, across diverse immune contextures. PTI was highest amongst ‘Wound healing’, ‘IFNy-dominant and ‘TGFfi-dominant’ subtypes (Figure 9A),and was significantly prognostic in all 3 subtypes as well as ‘Inflammatory’ and ‘Lymphocyte-depleted’ subtypes (Figure 2C). TIS, CD8+T cell, and IFNy signalling signatures, on the other hand, were more heterogeneous, and highest amongst the ‘IFNy-dominant subtype (Figure 8B). In addition, these biomarkers of so-called “hot” T cell- inflamed tumors, as well as biomarkers of antigenicity such as non-silent mutation rate, referred to here as tumor mutational burden (TMB), and single nucleotide variant neoantigen burden (SNV NAB) were inconsistently and infrequently prognostic across the subtypes (Figure 2C) (Thorsson et al., 2018). Therefore, PTI can be a prognostic indicator across tumor-immune subtypes with altogether very contrasting features, whereas typical markers of so-called “hot” tumors such as IFNy signalling correlated with prolonged OS in a single immune subtype. EXAMPLE 4: PTI signature is a powerful indicator of poor prognosis across IFNy- dominant subtypes

[0208] To extend this analysis, the inventors focussed on the ‘IFNy dominant subtype which encompasses patients with highly T cell-inflamed tumors and for whom CD8+T cells demonstrate unequivocal association with better outcome, as is often the case for immunotherapy-treated patients. In these highly infiltrated tumors, PTI was powerfully associated with poor survival at high and low IFNy levels (Figure 2D). Indeed, PTIh'9hIFNyh'9hpatients had a median OS ~5 years compared with >15 years in the PTIlowIFNyh'9hgroup despite equivalently high levels of IFNy and CD8+T cell infiltration (Figure 2D). These results suggest that the benefits of high levels of IFNy signalling are strongly antagonized by concomitant PTI expression. Similar findings were observed when substituting the IFNy signaling signature with different established immune signatures indicative of T cell inflammation (TIS, IFNy, CD8, PDL1 ) or with the fraction of tumor infiltrating lymphocyte (TIL) determined by immunohistochemistry (Thorsson etal., 2018) and PD- L1 (CD274) mRNA expression all of which are reported indicators of highly immunogenic, ICB- responsive tumors (Figure 9B). As such, COX-2 centered PTI is antagonistic towards diverse indicators of more immunogenic tumors.

[0209] To account for potential confounding effects of organ-specific immune features (Salmon et al., 2019), the interaction between PTI and IFNy signalling in individual cancer types was examined using multivariate CoxPH models. In nine cancer types, it was found that the PTI signature either as a single variable or through its interaction with IFNy signalling was a statistically significant prognostic factor when adjusted for clinical factors such as age, sex and disease stage (Figure 2E (ccRCC (n= 531 ), LIHC (n= 374), PAAD (n= 176), glioblastoma (GBM, n= 168), CESC (n= 306), adrenocortical carcinoma (ACC, n= 79), UCEC (n= 277), THCA (n= 513) and HNSC (n= 240))). Increasing PTI was associated with higher mortality regardless of the level of IFNy signalling in pancreatic cancer (PAAD), ccRCC, lung adenocarcinoma (LUAD), liver cancer (LIHC) and cervical cancer (CESC) (Figure 2E). In thyroid cancer (THCA), head and neck cancer (HNSC), and uterine cancer (UCEC) there was a significant interaction of PTI and IFNy signalling, thereby mimicking observations in the pan-cancer dataset (Figure 2E). In conclusion, the prognostic power of PTI extends to individual malignancies, and crucially is frequently linked to its suppressive effect on IFNy signalling.

[0210] EXAMPLE 5: The prognostic ability of individual PTI genes within the PTI signature

[0211] It was assessed whether there was a dominant factor explaining the prognostic power of the PTI signature by performing survival analysis for individual PTI genes. The COX-2 gene (PTGS2) itself indicated poor survival in ccRCC patients (Figure 9C). EGFR ligands EREG and AREG were associated with poor survival in cancers with prevalent KR S mutations like PAAD or a high prevalence of receptor tyrosine kinase-RAS pathway alterations like GBM and THCA. Cytokines IL11 and IL1A as well as chemokines CXCL5 and CXCL1 were prognostic in CESC and ccRCC. This data highlights how individual inflammatory ligands heterogeneously contribute towards the association of PTI with poor survival across multiple malignancies, and no single gene is as consistently prognostic as the signature itself.

[0212] EXAMPLE 6: The association between PTI levels and cancer stage

[0213] It was questioned whether there is an association between PTI levels and cancer stage. For this, PTI levels in healthy, pre-neoplastic and were compared to malignant diseased states mining a lung squamous cell carcinoma (LUSC) dataset comprising gene expression data from nine progressive cancer developmental stages (Mascaux et al., 2019). PTI levels were lowest in normal lung tissue and progressively increased (Figure 9D). Elevated expression of the PTI signature was first notable in moderately dysplastic lesions and peaked in carcinoma samples (Figure 9D). These findings suggest that highly proliferative, pre-cancerous lesions begin to coopt an inflammatory transcriptional network prior to full malignancy.

[0214] Finally, using the TNM cancer staging criteria from TCGA that take into account the size of the primary tumor, lymph node spread and the presence of metastases in other organs, the correlation between advancing disease stage and PTI levels was investigated. Pan-cancer there was a modest but statistically significant association of PTI expression with tumor stage independent of cancer type (P= 0.005). In select cancer types there was a marked increase in PTI levels alongside disease progression (Figure 9E). In most malignancies, PTI increased in pre-metastatic tumor states (l-lll) while in LUSC, PTI levels were comparable between stages I- III, and elevated in stage IV tumors with distant metastases (Figure 9E). To conclude, PTI correlates with oncogenic transformation, neoplasia and progression to more advanced and metastatic disease stages. Collectively this data supports the notion that inflammation is integral towards tumor progression and establishes the PTI signature as a broad and consistent indicator of poor prognosis.

[0215] EXAMPLE 7: MEK inhibition attenuates secretion of pro-tumoriqenic inflammatory mediators by cancer cells

[0216] Considering the researcher’s previously reported observations that numerous cancer cell lines express COX-2 and secrete PGE2 (Zelenay etal., 2015; Bell etal. 2022) it was evaluated whether COX-2 was co-expressed with other PTI genes in cancer cells. Mining transcriptomic data from more than 800 human cancer cell lines from the cancer cell line encyclopaedia (CCLE), it was found that PTI genes were expressed heterogeneously amongst cell lines but positively cocorrelated with one another (Figure 3A). To search for ways of targeting cancer cells that express PTI factors, data from a pan-cancer drug screen in which over 600 human cancer cell lines, with matched transcriptional data, were treated with a panel of chemotherapies and targeted agents was analysed (Garnett et al., 2012). Grouping cell lines by PTI expression and screening for associations between PTI status and drug sensitivity whilst adjusting for cancer type, it was found that PTIhighlines were selectively sensitive to four different MEK inhibitors (refametinib, Cl 1040, PD0325901 , AZD6244) (Figure 3B, Figure 10A). By contrast, PTIlowcell lines were sensitive to inhibitors of PARP1 / 2 and receptor tyrosine kinase (RTK) inhibitors that target VEGFR, PDGFR and FGFR (Figure 3B). Therefore, PTIhighcell lines are not susceptible to blanket inhibition of RTKs, the majority of which are thought to signal via ERK. In fact, more targeted inhibition of the MAPK cascade with MEK1 / 2 inhibitors led to potent growth inhibition of PTIhighbut not PTIlowcell lines. As an example, for cell lines treated with refametinib, a MEK1 / 2 inhibitor, there was a significant anti-correlation between the PTI and refametinib half-maximal inhibitory concentration (IC50) (Figure 3C, top panel). Refametinib-sensitive PTIhighlines in the top left quadrant (Figure 3C, top panel) encompassed several cancer types including 27% NSCLC, 18% oesophageal, 8% pancreas and 8% bladder. The refametinib-sensitive, PTIhighgroup contained significantly more NSCLCs compared to all others (21 % vs 7%, FDR= 0.006). Focusing on NSCLC cell lines only, an even more pronounced anti-correlation of refametinib IC50 and PTI was found (Figure 3C, bottom panel).

[0217] To determine whether pharmacological inhibition of MEK suppresses the production of PTI factors by cancer cells, two human lung cancer cell lines (NCIH1339, NCIH3255) with high PTI levels according to the CCLE transcriptional data were studied. Trametinib, an approved and selective MEK1 / 2 inhibitor, suppressed proliferation of these cells 24 hours post-treatment (Figure 10C, Figure 10D). In line their transcriptome, analysis of their supernatants by multiplex cytokine / chemokine assays showed that both cell lines secreted PTI factors such as IL- 1a, CXCL1 , IL-6, and IL-8 (Figure 3C). Notably, as early as 6 hours and for as long as 24 hours posttreatment prior to any clear impact on cell growth, trametinib inhibited the production of all these inflammatory mediators by both the KF?ASG12R(NCI H 1339) and KRASnEGFR-amplified cell lines (NCIH3255) (Figure 3C). Similar results were obtained with a second MEK1 / 2 inhibitor selumetinib (Figure 10E). Of note, selumetinib did not alter the production of osteopontin, an inflammatory mediator which is not part of the PTI, arguing that MEK inhibition does not indiscriminately inhibit cytokine production (Figure 10E). Collectively this data suggests that inhibition of MEK kinase activity can rapidly blunt the production of PTI factors by cancer cells. EXAMPLE 8: NFKB links MAPK pathway mutations to cancer cell-intrinsic PTI expression

[0218] The differential sensitivity of PTIhighcell lines to MEK inhibitors suggested that oncogenic alterations that lead to increase MAPK activity might correlate with PTI levels in cancer cell lines and tumors. Consistent with this hypothesis, KRAS mutant cell lines had higher PTI expression (Figure 4A). Next, screens for cancer driver gene copy number estimates (Martinez-Jimenez et al., 2020) that predict high cancer cell-intrinsic PTI expression independent of cancer type and aneuploidy were carried out. Amongst the genes positively associated with PTI levels were HRAS, RASA1, a RAS GTPase activating protein (GAP), RTKs PDGFRBand FL 74 that signal via MAPK signalling and RELA, the gene encoding for NFKB subunit p65 that can be trans-activated by MEK (Figure 4B, Figure 11 A) (Chang et al., 2003). Altogether, these results suggest that KRAS mutations and / or amplifications in MAPK-NFKB signalling often drive PTI expression by cancer cells.

[0219] Given that MAPK and NFKB pathways intersect, it was reasoned that convergence around NFKB activation was likely to be the dominant driver of PTI levels. Indeed, analysis of reverse phase protein array (RPPA) data showed that KRAS mutant cell lines with high PTI had significantly increased levels of activating p65 RELA) serine 536 phosphorylation, in addition to higher protein levels of EGFR, ECM components fibronectin, transglutaminase, and annexin-1 that regulates NFKB activity (Moraes, Ampomah and Lim, 2018) (Figure 11C). Levels of p65 serine 536 phosphorylation progressively increased correlating with PTI expression (Figure 4C). Moreover, motif enrichment analysis showed that NFKB binding motifs are significantly enriched in the promoter regions of an extended set of PTI genes (see methods). There were three NFKB transcription factors, including highly conserved motifs bound both by canonical (NFKB1 ) and non-canonical forms (NFKB2) of NFKB, amongst the top ten motifs significantly enriched in the promoter regions of PTI genes (Figure 4D, Figure 4E). Collectively, these findings at the gene and protein level suggest that NFKB activity promotes expression of the PTI transcriptional program in cancer cells bearing KRAS mutations.

[0220] To test whether intratumoral PTI levels correlate with genomic alterations that lead to MAPK and NFKB pathway overactivation, mutation calls from the pan-cancer TCGA dataset were analysed. In line with the cancer cell line data, mutations in MAPK pathway component (MAP2K1, that encodes MEK1 , NF1 and BRAF), were significant predictors of PTI expression pan-cancer independent of mutation burden and cancer type (FDR< 0.25). Of note, TP53 mutations also strongly positively correlated with PTI levels (Figure 4F) whereas mutations in the oncogene NFE2L2 or tumor- suppressor gene mutations in NRF2 (AXIN2) and WNT (KEAP1) pathways were enriched in PTIlowtumors (Figure 4F). The same analysis in individual cancer types yielded >30 significant mutational links to high PTI expression (FDR< 0.25). Of these mutations, a large fraction were MAPK pathway genes including KRAS in LLIAD, ccRCC and PAAD, NF1 in LLISC and GBM, and BRAF'in THCA (Figure 4G). Thus, MAPK activation is a hallmark of PTIhightumors in multiple tissue types. Moreover, it was found that EGFR and RELA amplifications in solid tumors were positively correlated with intratumoral PTI signature scores (Figure 4H). Therefore, analysis of cancer cell line and bulk tumor genomic and transcriptomic data indicated that both mutations and SCNAs in the MAPK / NFKB pathway drive expression of PTI factors.

[0221] EXAMPLE 9: Diverse malignant and non-maliqnant cell types underscore high PTI signature expression

[0222] To investigate the infiltrate composition of PTIhighversus PTIlowthe researchers computed pairwise Spearman correlations between PTI levels and multiple immune and non-immune cell types inferred with the MCPcounter algorithm to bulk tumor transcriptomes (Becht et al., 2016). PTI signature levels were significantly positively correlated with multiple cell types including endothelial cells (26 / 33), fibroblasts (30 / 33), neutrophils (31 / 33), and monocytic-lineage cells (28 / 33), in most cancer types. By contrast, cytotoxic cells, such as CD8+T cells (X / 33) or NK cells (X / 33) were much less consistently and poorly correlated with PTI expression (Figure 12A). This data indicates that the predominant tumor-infiltrating cellular component of PTIhightumors comprises various myeloid cell subsets, mesenchymal cells like fibroblasts suggesting that cancer cell-intrinsic PTI expression leads to expansion and / or recruitment of diverse cell types.

[0223] To identify the cellular source of PTI factors, three independent scRNAseq datasets from three different cancer types were interrogated (HNSC, MEL, NSCLC) (Puram et al., 2017; Jerby-Arnon et al., 2018; Lambrechts et al., 2018). Single cells from tumor samples were re-clustered using unsupervised, graph-based clustering and assigned states guided by meta-data from the original studies. T-distributed stochastic neighbour embedding (t-SNE) identified multiple and diverse cell types as PTI sources (Figure 5A, Figure 12B). Measuring the mean expression of each PTI gene in each original cluster revealed high levels of expression amongst neutrophils, macrophages, fibroblasts, mast cells and endothelial cells (Figure 5B, Figure 12C). Some PTI genes were expressed higher by specific cell types. For example, IL6 and HAS2 were predominantly expressed by fibroblasts, whereas CXCL5 was unique to macrophages (Figure 5B, Figure 12C). Other PTI genes, including VCAN, AREG, PTGS2 and CXCL2 were expressed across different immune and non-immune cell types (Figure 5B, Figure 12C). Together, this data indicates that, in addition to being directly expressed by cancer cells, PTI factors can be heterogeneously produced by multiple immune and non-immune cell types.

[0224] Next, PTI levels within single cancer cells isolated from ten independent HNSC patients were compared. There was heterogenous PTI expression between patients (Figure 5D). The researchers compared PTI levels within cancer cells from individual patients to define PTIhighand PTIlowtumors based on cancer cell-intrinsic expression (Figure 5D). Collectively, these findings support a model whereby inflammatory mediators, many of which are coordinately expressed by cancer cells and can be induced by PGE2, are also expressed by numerous cell types within the TME.

[0225] EXAMPLE 10: The PTI signature affects the proportion and transcriptional phenotype of tumor-infiltrating cytotoxic lymphocytes

[0226] To address whether high PTI levels are associated with a unique leukocyte infiltrate composition in ‘hot’ tumors, CIBERSORT estimates of immune cell proportion in ‘ IFNy dominant subtype tumors were analysed (Newman et al., 2015; Thorsson et al., 2018). Comparing the mean cell type proportion across patients with different PTI levels, it was found that the proportion of leukocytes was largely consistent across the groups (Figure 5E). Lymphocytes made up a significantly lower proportion of total immune cells in PTIh'9htumors (FDR= 0.00003, Figure 5E). PTIhightumors had a statistically significant lower proportion of CD8+T cells (FDR= 0.002), and activated NK cells (FDR= 8.6E-08), yet a higher proportion of activated memory CD4+T cells (FDR= 1 .2E-19) and resting NK cells (FDR= 1 .2E-30, Figure 5E). These results indicate that the overall immune infiltrate of ‘hot’ tumors with concomitant high levels of PTI is subtly altered with fewer effectors of anti-tumor immunity.

[0227] CD8+T cells that infiltrate tumors exhibit a range of transcriptional states from naive-like to dysfunctional and cytotoxic (van der Leun, Thommen and Schumacher, 2020). Based on the previous analysis, it was hypothesised that PTI expression promotes CD8+T cell dysfunction and thus would positively correlate with sub-optimal CD8+T cell transcriptional signatures. In accordance with this hypothesis, a strong positive correlation of PTI with a signature of dysfunctional self-antigen specific CD8+T cells was found (Schietinger et al., 2012), referred to as the T cell anergy signature (Azizi et al., 2018) (Figure 5F). Thus, PTI is strongly associated with the presence of CD8+T cells that transcriptionally resemble cells incapable of mounting an effective anti-tumor immune response. In addition, PTI positively correlated with signatures of CD8+T cell dysfunction and exhaustion (Singer et al., 2016; Guo et al., 2018) (Figure 12D). A previous study that analyzed CD8+T cells from tumors that responded favourably to ICB in MEL revealed an ICB response associated, single cell, naive-like CD8+T cell signature composed of genes involved in memory, survival and activation (Sade-Feldman et al., 2018). It was found that CD8+T cells infiltrating PTIlowMELs and NSCLCs had significantly higher expression of this CD8 ICB response signature than those from their PTIh'9hcounterparts. Thus, CD8+T cells in PTIhightumors did not resemble the naive / memory like cells that promote ICB responses in MEL. Analysis of tumor-derived CD8+T cell single cell transcriptomes supports our bulk tumor findings, and establishes a strong connection between expression of pro-tumorigenic inflammatory mediators and sub-optimal T cell transcriptional states. EXAMPLE 11 : PTI signature predicts lack of benefit from checkpoint inhibitors and outperforms other signatures of resistance

[0228] Thus far, powerful predictors of ICB resistance have been elusive. Given the powerful opposing prognostic power of the PTI signature against current immune biomarkers of T cell-inflammation, many of which associate with ICB response, the utility of PTI as a biomarker of ICB resistance was investigated. 16 transcriptomic datasets comprising >1000 pre-treatment samples (n=1032) from patients treated with ICB were collated (Figure 6A, Figure 6B). This pan-cancer ICB metadataset contained patients with five cancer types (ccRCC, mllC, MEL, GCA and NSCLC), who were treated with anti-CTLA-4, anti-PD1 or anti-PD-L1 antibodies as monotherapy and in combination. 25% of samples were from patients that responded to the therapy (partial or complete responses (PR / CR)) while 43% derived from patients that experienced no benefit (progressive disease (PD)). By stratifying patients according to median pre-treatment PTI levels in each dataset, a significant enrichment in PR / CR was found among PTIlowpatients (x2, P= 0.004). Conversely, in line with the reported value of markers of T cell-inflammation, patients with low TIS (x2, p= 4x10-7), CD8+T cell (x2, p= 4x10-6), IFNy signalling (x2, p= 3x106) signatures, or PD-L1 gene expression (CD274, x2, p= 0.003) were enriched in PD patients (Figure 6C) (Tumeh et al., 2014; Ayers et al., 2017). Crucially, PTI exhibited joint predictive power when combined with all these current biomarkers of ‘hot’ T cell-inflamed tumors (TIS, CD8, IFNy and / or CD274) such that the fraction of CR / PR was lowest among PTIh'9hTISlowpatients, and progressively increased being highest within PT IlowT I Shi9hpatients (Figure 6D). Likewise, the combination of PTI and TIS also exhibited improved predictive value for OS in these ICB-treated patients (Figure 6E). PTIh'9hTISh'9hpatients had significantly worse OS, with about a year difference in median survival, than PTIlowTISh'9hpatients. Notably, PTIh'9hTISh'9hand PTI|OWTIS|OWpatients had comparable survival in line with our earlier TCGA data analysis indicating an antagonistic interaction between PTI and anti-tumour immune mediators (Figure 6E). Together, this analysis demonstrates that the PTI signature aids distinguishing ICB responders from non-responders, including among patients with equivalent high levels of intratumoral T cell inflammation.

[0229] Next, the researchers sought to benchmark the PTI signature against other reported signatures linked to ICB resistance. These included a WNT / pcatenin signalling signature previously shown to correlate with ‘cold’ poorly T cell-inflamed tumors (Gajewski et al., 2019), gene signatures related to immunosuppressive TGFp signaling via fibroblasts (F-TBRS, Chakravarthy etal., 2018) and ECM remodelling (TGFp-ECM, Mariathasan et al., 2018), a signature of EMT linked to reduced response rates in T cell-inflamed mUCs (Stroma-EMT, Wang et al., 2019), and a signature of angiogenesis-related factors highly expressed during the inflammation-to-cancer (I2C) transition associated with melanoma ICB resistance (Chen et al., 2019). To evaluate the added predictive value of each of these signatures in combination with a T cell-inflamed signature, the TIS signature was used as a baseline predictive model, one of the most powerful and widely- validated gene expression biomarker for ICB (Ayers et al., 2017; Cristescu et al., 2018). Each continuous signature score was combined with the TIS in a joint predictive model where the area- under-the-curve (AUC) was calculated as an indicator of model performance for each one of the 16 ICB-treated patient cohorts. Notably, the PTI / TIS model had the highest mean AUC in comparison with the other resistance signature models and the TIS alone (Figure 13A). Importantly, the mean AUC for the PTI plus TIS model (0.79) was significantly higher than the TIS alone (mean AUC= 0.69, P= 0.02) highlighting the predictive utility of combining PTI and TIS values across pan-cancer ICB datasets. The researchers also compared the performance of all resistance signatures for predicting patient OS following ICB therapy which does not necessarily follow the response rate. Patients were stratified by a median cut-off value for each signature and fit into a multivariate CoxPH model adjusted for TIS status and ICB dataset. 1000 randomly generated signatures of equal length to the PTI were also included as a control. Consistent with the objective response data, PTI was by far the most powerful of the resistance signatures at predicting poor OS in ICB-treated patients (Figure 6F). Together, this data further demonstrated the unique and superior utility of combining the PTI with immune biomarkers of ‘hot’ tumors to predict patient response and survival following ICCB.

[0230] The presently claimed PTI signature has improved predictive power over the COX-IS ratio disclosed in the inventor’s earlier application, WO 2019 / 243567. Figure 6F shows the predictive power of resistance signatures at predicting poor overall survival in ICB-treated patients. The PTI signature (HR 1.38 [1.15-1.65], P=0.0004) outperforms the inventor’s earlier COX-IS ratio (HR 1.18 [0.96-1.45], P=0.12) and therefore exemplifies improved predictive power.

[0231] Next the CPU 000 dataset (Litchfield et al., 2021 ) which comprises a collection of datasets, substantially overlapping those in our above meta-dataset, from patients treated with ICB (anti- PD(L)1 , anti-CTLA4) across multiple cancer types with whole exome sequencing (WES) and RNA sequencing (RNAseq) data was examined. This unique dataset underwent extensive harmonisation of WES, RNAseq and clinical data across several cohorts. Firstly, the analysis focussed on 523 patients from seven studies covering three cancer types (MEL, BLCA, ccRCC) with annotations for multiple genomic and transcriptomic tumor characteristics previously linked to ICB response (Snyder et al., 2014, 2017; Van Allen et al., 2015; Hugo et al., 2017; Riaz et al., 2017; Mariathasan etal., 2018; Mcdermott etal., 2018). Besides modest positive correlations with the APOBEC signature (Spearman= 0.26, P= 9x107), TP53 mutational status (Spearman= P= 0.0004) and CD274 expression (Spearman= 0.22, P= 3x1 O'7), PTI levels were not strongly associated with any single parameter (Figure 13B, 13C) supporting the notion that PTI captures a distinct independent feature of tumor inflammation, and its potential use as an independent immune biomarker. To evaluate the predictive power of PTI in this harmonised dataset benchmarking against multiple signatures and parameters of ICB resistance such as F-TBRS and Stroma-EMT (Mariathasan et al., 2018; Wang et al., 2019) or response such as the IMPRES algorithm, TIS score, CXCL13 expression, TLS signature, CYT score and different TMB measures was carried out (Rizvi et al., 2015; Rooney et al., 2015; McGranahan et al., 2016; Ayers et al., 2017; Auslander et al., 2018; Mariathasan et al., 2018; Cabrita et al., 2020; Litchfield, Reading, et al., 2020). Remarkably, PTI levels in pre-treatment samples of PD patients were higher than in PR / CR patients (n= 552, Figure 6G). Next the inventors fitted univariate logistic regression models for the PTI, and other biomarkers, in the seven individual datasets that made up the wider pan-cancer cohort (Van Allen et al., 2015; Hugo et al., 2017; Riaz et al., 2017; Snyder et al., 2017; Mariathasan et al., 2018; Mcdermott et al., 2018; Liu et al., 2019). To assess the consistency of each putative biomarker, a meta-analysis to compute overall effect sizes was performed. The PTI was the best predictor of ICB resistance (odds ratio= 0.67, P= 0.03, Figure 6H). PTIhighpatients in the entire CPU 000 cohort including CR / PR, PD and also stable disease (SD) patients had significantly worse OS than PTI intermediate and low groups (HR= 1.41 [1.1 -1.8], P= 0.007) (Figure 13D). Moreover, PTI remained a significant prognostic factor after adjusting for cancer type and prior exposure to anti-CTLA-4 therapy (HR= 1.29 [1.006-1.65], P= 0.04) further demonstrating the independent predictive value of the PTI signature as an immune biomarker.

[0232] A model combining clonal TMB with CXCL9 expression was shown to be the best for predicting patient responses to ICB in the CPI1000 dataset (Litchfield etal., 2021 ). It was tested if PTI levels would add significant predictive utility to the clonal TMB and CXCL9 combination by fitting predictive models covering cancer type, clonal TMB, CXCL9, PTI and their combinations. Notably, the PTI added significant predictive value on top of cancer type, clonal TMB and CXCL9 (Figure 13E, likelihood ratio test, P= 0.01 ). Clonal TMBhi, or CXCL9hior TMBhiCXCL9hipatients containing the highest fraction of ICB responders (Litchifield et al) were further stratified by the PTI signature (Figure 61, Figure 13F, Figure 13G). These findings further demonstrate the value of measuring PTI in pre-treatment samples concomitantly with other immune biomarkers. Moreover, this data positions PTI as a useful tool for identifying patients that should receive more aggressive combination treatment, such as with COX-2 inhibitors or other drugs that dampen PTI, in order to maximise their benefit from ICB.

[0233] EXAMPLE 12: Models to predict binary patient response outcomes for tumour type agnostic and tumour type specific approaches

[0234] Logistic regression models were trained on the CPU 000 dataset. Firstly, TPM values from the CPU 0000 dataset were taken for the PTI genes. A pseudocount of 1 was added before Iog2- normalising the values. Then a) data was scaled (z-score normalised) across all of the samples in the data set (tumour type agnostic), or b) data was grouped by cancer type (metadata feature: “Tum_Type”) and expression values were scaled by tumour type (tumour type specific). The mean of the PTI genes of the scaled expression data was then determined to compute the PTI score. Patients which achieved stable disease as best RECIST response (metadata feature: Response_RECIST) were removed, alongside ‘on treatment’ samples (metadata feature: Prior lO).

[0235] Three different logistic regression models (discussed in more detail below) were trained to predict binary patient response outcomes (metadata feature: Response_Responder) for both tumour type agnostic and tumour type specific approaches.

[0236] The three logistic regression models are as shown in Table 3. The results of logistic regression models A to C are shown in Table 4. For tumour specific model A, coefficients for PTGS2, CXCL2, CXCL6, IL1A, AREG, EREG, CXCL1 , CXCL5, HAS2, IL11 , IL1 B, IL6, CXCL8, NRG1 , VCAN, Clonal_TMB and CXCL9 were extracted as per Table 5. Area under the curve (ALIC) was calculated as an indicator of model performance, with a larger AUC indicating better predictive utility. The receiver operating characteristic curve for this logistic regression model A (tumour-type specific) is shown in Figure 15.

[0237] Model C, which combines PTGS2, CXCL1 , CXCL2, CXCL5, CXCL6, CXCL8, IL1 A, IL1 B, IL6, IL11 , NRG1 , HAS2, AREG, EREG and VCAN expression in the absence of TMB or CXCL9, was shown to predict binary patient response outcomes for tumour type agnostic, and tumour type specific approaches as indicated in Table 4 below.

[0238] Table 3: Summary of logistic regression model A, model B and model C for predicting binary patient response outcomes for both tumour type agnostic and tumour type specific approaches.

[0239] Table 4: Results of logistic regression model A, model B and model C for tumour type agnostic and tumour type specific approaches.

[0240] Table 5: coefficients extracted from logistic regression model A (tumour-type specific).

[0241] EXAMPLE 13: The clinical value of the PTI signature as an immune biomarker of ICB outcome

[0242] Lastly, the predictive utility of the PTI signature and its potential clinical value as an immune biomarker analyzing formalin-fixed paraffin embedded (FFPE) tissue from patients treated with immunotherapy was evaluated. ccRCC, a cancer type in which ICB has become a mainstream treatment but for which current immune biomarkers fail to anticipate outcome was focussed on. As a proof-of-concept the researchers obtained archival FFPE samples from 42 patients (91% ccRCC) that underwent treatment for advanced disease with either high dose IL-2 immunotherapy (n=23), or ICB as monotherapy (n=1 1 ) or in combination with anti-CTLA-4 (n=4) or anti-VEGF (n=1 ) therapies for whom overall survival data had been matched (Figure 14A). This diverse, real-world clinical cohort contained nephrectomy samples, as well as samples from surgical procedures to excise brain and pulmonary metastases (Figure 14B). RNA was extracted and multiplex gene expression analysis on the Nanostring nCounter platform was performed using a pan-cancer immune profile panel that contained 800 immune-related genes including most of the PTI signature genes (12 / 16). 83% of the PTI genes present were expressed above the limit-of- detection in more than 50% of samples and were generally co-correlated with one another (mean Spearman= 0.3, Figure 7A). Therefore, irrespective of quantification method, PTI represents a conserved TME transcriptional program. The PTI signature correlated with the presence of different myeloid cell subsets (neutrophils, DCs, macrophages), as well as signatures of stroma and complement (Figure 7B). Therefore, in line with earlier bulk RNAseq findings, the PTI signature was associated with the presence of multiple immune populations and molecular features of the disease. Crucially, PTI levels inversely correlated with the CD8+T cell signature (CD8A, CD8B, FLT3LG, GZMM, PRF1) and the IFNy signalling signature (CXCL9, CXCL10, IDO1, STAT1, IFNG, HLA-DRA) (Figure 7C).

[0243] From these 42 RCC tumor biopsies, 25 were pre-treatment samples from patients that received IL-2 immunotherapy, anti-PD(L)1 or anti-PD1 / anti-CTLA-4 blockade. Stratifying patients according to PTI levels in these samples identified patients with significantly different survival outcomes whereby PTIhighpatients had a significantly shortened OS (Figure 7D, Figure 7E). Crucially, none of the immune signatures of ‘hot’ T cell-inflamed tumors showed a significant association with patient survival in this patient cohort (Figure 7F). A ‘myeloid cell-signature’ previously suggested to predict ICB outcome in ccRCC patients (Mcdermott et al., 2018) also failed to show a significant association. These striking findings in a small and heterogenous cohort of real-word immunotherapy treated ccRCC patients, using a clinically-compatible assay, further highlights the potential power of monitoring PTI to anticipate ICB outcome and guide patient treatment.

[0244] EXAMPLE 14: PTI as a predictor of ccRCC patient outcome following immunotherapy

[0245] Finally, to further validate the specific utility of PTI in predicting patient outcomes following immunotherapy in ccRCC, transcriptomic data from Checkmate trials 009, 010 and 025 was analysed. In all this trials nivolumab, an anti-PD-1 blocking agent demonstrated clinical benefit for patients with previously-treated, advanced ccRCC, including when compared with the MTOR inhibitor everolimus in the phase III study (Checkmate 025) (Braun et al., 2020). Remarkably, patient stratification based on pre-treatment PTI expression levels negatively associated with survival in nivolumab-treated patients (HR= 1.46 [1.02-2.08], P= 0.04), but not in everolimus- treated patients (HR= 1 .03 [0.71 -1 .50], P= 0.87) (Figure 7G). To test the predictive power of PTI in patients treated with ICB as first line therapy, data from the phase II IMMotion150 study in which previously-untreated advanced ccRCC patients were placed on one of three trial arms to receive atezolizumab (anti-PD-L1) and bevacizumab (anti-VEGF), monotherapy atezolizumab or sunitinib (Mcdermott etal., 2018) was analysed. Consistent with the inventor’s analysis in second line ICB-treated patients, PTI levels were higher at baseline in non-responder patients following first line atezolizumab or atezolizumab plus bevacizumab combination therapy compared to responders (Figure 7H). By contrast, responders and non-responders to sunitinib; a multi-target tyrosine kinase inhibitor, showed no difference in baseline PTI expression (Figure 7H). Collectively, these findings indicate that PTI selectively identifies ccRCC patients resistant to current immunotherapy regimens and demonstrate that its predictive utility from ICB extends beyond its prognostic value.

[0246] EXAMPLE 15: Minimum number of PTI genes required to calculate a score which is predictive of treatment response and / or prognosis

[0247] Elastic net regression is a statistical model that can help to reduce the feature space and select the most predictive features in classification tasks such as the prediction of ICB response. To illustrate this, we trained elastic net regression models firstly using PTI genes together with clonal TMB and CXCL9 expression level as the input features. We did this using normalised expression matrices. In the first of these (model D), we normalised expression levels separately for each tumour type (tumour type specific), and for the second (model E) we normalised expression levels in a tumour agnostic fashion (tumour type agnostic). In addition, we trained elastic net regression models on members of the PTI gene set in the absence of clonal_TMB and CXCL9 as covariates in a tumour agnostic fashion (model F) (tumour type agnostic).

[0248] Elastic net regression models were trained with different values of the a hyperparameter between 0 (ridge regression) and 1 (lasso regression). The key difference is that an increasing a shrinks the coefficients of less important features towards 0, thereby allowing us to perform a stricter selection of important features.

[0249] Method

[0250] First, tumour type agnostic (model E) and tumour type specific (model D) data sets were divided into train (70%) and test (30%) sets. Elastic net classifier models were trained to predict response (CR / PR) using 10-fold cross validation, and with AUC as the performance metric. Models were trained at different values of the a hyperparameter between 0 and 1 with increments of 0.1 . The value of 1 corresponding to the best training performance was selected as the best model for each training. Patient response to ICB was then predicted in the test dataset using each of the optimised models, and the AUC was computed. AUCs from the training and test sets were then plotted (Figure 16A-B for models D and E respectively). An alpha value of 0 indicates where all the available features (all 16 PTI genes, clonal_TMB and CXCL9) were used.

[0251] An elastic net regression model was also trained on members of the PTI gene set in the absence of clonal_TMB and CXCL9 as covariates (model F) (Figure 16C).

[0252] Results

[0253] The plots in Figure 16A and 16B show test AUCs and training AUCs for tumour type specific (model D) and tumour type agnostic (model E) models respectively, where AUC was calculated using clonal_TMB and gene expression levels of all 16 PTI genes and CXCL9. Points indicate models trained with different values of the a hyperparameter between 0 and 1 . The corresponding a values are shown next to each point. Typically, the higher the a value, the closer to lasso regression, and the fewer features the model utilises. As is clear from Figure 16, models trained with a values greater than 0 tended to have improved training and test performance. Notably, many of the models had a reduced number of important features compared to the number of features the model was trained on. An important feature is defined as any feature with a non-zero regression coefficient after applying regularisation.

[0254] Table 6 exemplifies the important features and coefficients of the best tumour type agnostic dataset model (model E) with an alpha value of 0.3. This best model E has a reduced feature set. Four genes from the PTI signature (EREG, HAS2, CXCL8 and VCAN) were identified as important features alongside Clonal_TMB and CXCL9.

[0255] Table 6: Important features and their corresponding coefficients for best model E (tumour type agnostic) with an alpha hyperparameter of 0.3.

[0256] Figure 16C exemplifies training and test AUC values from a tumour type agnostic model (model F) with only PTI genes as features (CXCL9 and Clonal_TMB are not included as covariates). Values of a are shown next to the corresponding point (trained model). The best model had an a hyperparameter of 0.2 and 7 important features (AREG, CXCL1 , CXCL2, EREG, HAS2, IL1 A and PTGS2) which correspond to 7 of the 16 PTI genes. The coefficients of each of these 7 features for the best model F (a=0.2) are exemplified in Table 7.

[0257] Table 7: Important features and their corresponding coefficients for the best model F with an alpha hyperparameter of 0.2.

[0258] Whilst elastic regression models trained on members of the PTI gene set alone (in the absence of clonal_TMB and CXCL9) do not perform as well as those which are trained on PTI genes in the presence of clonal_TMB and CXCL9, they still explain a minor proportion of the variance in patient ICB response.

[0259] DISCUSSION

[0260] Determining the likelihood of patient benefit from ICB is a significant challenge for the field (Hegde and Chen, 2020). The inventors reasoned that biological mechanisms beyond current biomarkers must underpin divergence in patient outcomes following ICB. Subsequently the inventors showed that COX-2 centred pro-tumorigenic inflammatory ligands frequently govern such divergence including even in patients with high clonal TMB and CXCL9 expression, more likely to benefit from ICB (Litchfield et al., 2021 ). These findings demonstrate the need for multi-combinatorial biomarkers requiring both WES, and a gene expression assay that should crucially include PTI genes besides those that measure T cell inflammation. An important consideration of gene signature-based biomarker development is intratumoral heterogeneity (ITH). The extent to which sampling biases may confound a PTI-based patient classifier is unclear. However, one method of circumventing these issues would be to select PTI genes with overall lower ITH that are less prone to sampling biases as previously carried out for a prognostic signature in NSCLC (Biswas etal., 2019). Utilising a published method (Jerby-Arnon et al., 2018) to analyse cell line drug screen data (Garnett et al., 2012), it was shown that PTIh'9hcell lines are sensitive to MEK inhibitors. The dataset was, however, not supplemented with next generation inhibitors of historically challenging targets such as KRAS (Canon et al., 2019). In addition, identifying a core targetable node to disrupt the PTI signalling network will require experiments in vivo with the full complement of interacting cell types. Nevertheless, this work indicates that the COX-2 / PGE2 axis is a strong candidate (Pelly et al., 2021 ). Numerous studies have reported the favourable, immunomodulatory properties of MEK inhibitors (Hu-Lieskovan et al., 2015; Ebert et al., 2016; Bommareddy et al., 2018; Baumann et al., 2020). While future work is required to test if MEK inhibitors specifically dampen the PTI transcriptional program in vivo, the present data further supports the notion that they have broader scope not just as cytostatic agents (Templeton and Musib, 2015; Hamarsheh et al., 2020).

[0261] Leveraging bulk tumour RNAseq, immune deconvolution algorithms and scRNAseq data (Becht et al., 2016; Puram et al., 2017; Jerby-Arnon et al., 2018; Lambrechts et al., 2018) it is shown that, in addition to cancer cells themselves, diverse intratumoural cell types expressed PTI genes and infiltrated PTIh'9htumours. Neutrophils, in particular, were highly correlated with levels of PTI and, consistent with the working model, have been shown to be elevated in KRAS mutant settings (Hamarsheh et al., 2020; Polidoro et al., 2020). It remains unclear, however, whether these cells play an active role in the maintenance of cancer-promoting inflammation. Furthermore, it remains ambiguous whether cancer cell-intrinsic production of inflammatory mediators propagates the PTI signature throughout non-cancer cells that lie in close proximity or vice versa. Future mechanistic studies with spatially-resolved transcriptomics data from human tumours should help to address these questions.

[0262] The underlying basis for the association of PTI ligands with ICB resistance was investigated. PTI factors have diverse roles within the TME contributing to all cancer hallmarks.

[0263] A profound interaction of PTI with T cell infiltration and IFNy signalling was uncovered. Indeed, direct modulation of the activation state of intratumoural lymphocytes likely contributes towards the ICB resistance phenotype. Patients with CD8+T cell-inflamed tumours or high levels of IFNy signalling had divergent outcomes depending on PTI levels; a trend that was also noted in patients with high plasma levels of the pro-tumourigenic chemokine IL-8 (Yuen et al., 2020). The COX- 2 / PGE2 axis, a central node in the PTI phenotype, has previously been associated with T cell exhaustion (Harris et al., 2002; Chen et al., 2015). Using a CD8 ICB response signature (Sade- Feldman et al., 2018), as well as signatures of CD8+T cell dysfunction (Azizi et al., 2018; Guo et al., 2018) it is shown that CD8+T cells in PTIhightumours had sub-optimal activation states.

[0264] In conclusion, a 16-gene COX-2-centred inflammatory gene signature is defined that is associated with molecularly distinct T cell infiltrates and dismal outcome including following treatment with immune checkpoint inhibitors. References

[0265] A number of publications are cited above in order to more fully describe and disclose the invention and the state of the art to which the invention pertains. Full citations for these references are provided below. The entirety of each of these references is incorporated herein.

[0266] Alfaro, C. et al. (2016) ‘Tumor-Produced lnterleukin-8 Attracts Human Myeloid-Derived Suppressor Cells and Elicits Extrusion of Neutrophil Extracellular Traps (NETs)’, Clinical Cancer Research, 22(15), pp. 3924 LP - 3936. doi: 10.1158 / 1078-0432.CCR-15-2463.

[0267] Van Allen, E. M. etal. (2015) ‘Genomic correlates of response to CTLA-4 blockade in metastatic melanoma’, Science, 350(6257), pp. 207-211. doi: 10.1126 / science.aad0095.

[0268] Andrea, S. etal. (2012) ‘Rescued Tolerant CD8 T Cells Are Preprogrammed to Reestablish the Tolerant State’, Science, 335(6069), pp. 723-727. doi: 10.1126 / science.1214277.

[0269] Auslander, N. etal. (2018) ‘Robust prediction of response to immune checkpoint blockade therapy in metastatic melanoma’, Nature Medicine, doi: 10.1038 / s41591 -018-0157-9.

[0270] Ayers, M. etal. (2017) ‘IFN- y - related mRNA profile predicts clinical response to PD-1 blockade’, The Journal of Clinical Investigation, 127(15), pp. 1-11. doi: 10.1172 / JCI91190.

[0271] Azizi, E. etal. (2018) ‘Single-Cell Map of Diverse Immune Phenotypes in the Breast Tumor Microenvironment’, Cell, 174(5), pp. 1293-1308. e36. doi: https: / / doi.Org / 10.1016 / j.cell.2O18.05.060.

[0272] Baumann, D. etal. (2020) ‘Proimmunogenic impact of MEK inhibition synergizes with agonist anti-CD40 immunostimulatory antibodies in tumor therapy’, Nature Communications, 11 (1 ). doi: 10.1038 / S41467-020-15979-2.

[0273] Becht, E. et al. (2016) ‘Erratum to Estimating the population abundance of tissue-infiltrating immune and stromal cell populations using gene expression [Genome Biol., 17 (2016), (218)]’, Genome Biology, 17(1 ), pp. 1-20. doi: 10.1186 / s13059-016-1113-y.

[0274] Bhattacharya, A. et al. (2020) ‘An approach for normalization and quality control for NanoString

[0275] RNA expression data’, Briefings in Bioinformatics, 00(June), pp. 1-13. doi: 10.1093 / bib / bbaa163.

[0276] Biswas, D. etal. (2019) ‘A clonal expression biomarker associates with lung cancer mortality’, Nature Medicine, 25(October). doi: 10.1038 / s41591 -019-0595-z.

[0277] Bommareddy, P. K. etal. (2018) ‘MEK inhibition enhances oncolytic virus immunotherapy through increased tumor cell killing and T cell activation’, Science Translational Medicine, 10(471 ). doi: 10.1126 / scitranslmed.aau0417.

[0278] Bonavita, E. et al. (2020) ‘Antagonistic Inflammatory Phenotypes Dictate Tumor Fate and Response to Immune Checkpoint Blockade’, Immunity, 53(6), pp. 1215-1229. e8. doi: 10.1016 / j.immuni.2020.10.020.

[0279] Braun, D. A. etal. (2020) ‘Interplay of somatic alterations and immune infiltration modulates response to PD-1 blockade in advanced clear cell renal cell carcinoma’, Nature Medicine, doi: 10.1038 / S41591 -020-0839-y.

[0280] Browaeys, R., Saelens, W. and Saeys, Y. (2020) ‘NicheNet: modeling intercellular communication by linking ligands to target genes’, Nature Methods, doi: 10.1038 / s41592-019- 0667-5.

[0281] Cabrera, R. M. etal. (2018) ‘A novel neuregulin - jaggedl paracrine loop in breast cancer transendothelial migration’, Breast Cancer Research, 20(1), pp. 1-10. doi: 10.1186 / s13058-018- 0960-8.

[0282] Cabrita, R. et al. (2020) ‘Tertiary lymphoid structures improve immunotherapy and survival in melanoma’, Nature, 577(7791), pp. 561-565. doi: 10.1038 / s41586-019-1914-8.

[0283] Calon, A. etal. (2012) ‘Dependency of Colorectal Cancer on a TGF-p-Driven Program in Stromal Cells for Metastasis Initiation’, Cancer Cell, 22(5), pp. 571-584. doi:

[0284] 10.1016 / j.ccr.2O12.08.013.

[0285] Canon, J. et al. (2019) ‘The clinical KRAS(G12C) inhibitor AMG 510 drives anti-tumour immunity’, Nature, 575(7781 ), pp. 217-223. doi: 10.1038 / s41586-019-1694-1.

[0286] Chang, F. et al. (2003) ‘Signal transduction mediated by the Ras / Raf / MEK / ERK pathway from cytokine receptors to transcription factors: Potential targeting for therapeutic intervention’, Leukemia, 17(7), pp. 1263-1293. doi: 10.1038 / sj.leu.2402945.

[0287] Chen, J. H. et al. (2015) ‘Prostaglandin E2 and programmed cell death 1 signaling coordinately impair CTL function and survival during chronic viral infection’, Nature Medicine, 21 (4), pp. 327- 334. doi: 10.1038 / nm.3831 .

[0288] Chen, X. etal. (2019) ‘Immune Cell Types and Secreted Factors Contributing to Inflammation- to-Cancer Transition and Immune Therapy Response’, Cell Reports, pp. 1965-1977. doi: 10.1016 / j.celrep.2O19.01.080.

[0289] Chittezhath, M. et al. (2014) ‘Molecular Profiling Reveals a Tumor-Promoting Phenotype of Monocytes and Macrophages in Human Cancer Progression’, Immunity, 41 (5), pp. 815-829. doi: 10.1016 / j.immuni.2014.09.014.

[0290] Cohen-Sharir, Y. etal. (2021) ‘Aneuploidy renders cancer cells vulnerable to mitotic checkpoint inhibition’, Nature, 590(7846), pp. 486-491. doi: 10.1038 / s41586-020-03114-6.

[0291] Coussens, L. M., Zitvogel, L. and Palucka, A. K. (2013) ‘Neutralizing tumor-promoting chronic inflammation: A magic bullet?’, Science, 339(6117), pp. 286-291. doi: 10.1126 / science.1232227.

[0292] Cristescu, R. etal. (2018) ‘Pan-tumor genomic biomarkers for PD-1 checkpoint blockade-based immunotherapy’, Science, 362(6411 ), p. eaar3593. doi: 10.1126 / science.aar3593.

[0293] Danaher, P. etal. (2017) ‘Gene expression markers of Tumor Infiltrating Leukocytes’, pp. 1-15. doi: 10.1186 / S40425-017-0215-8.

[0294] Danaher, P. etal. (2018) ‘Pan-cancer adaptive immune resistance as defined by the Tumor Inflammation Signature ( TIS ): results from The Cancer Genome Atlas ( TCGA )’, pp. 1-17. Dannenberg, A. J. and Subbaramaiah, K. (2003) ‘Targeting cyclooxygenase-2 in human neoplasia: Rationale and promise’, Cancer Cell, 4(6), pp. 431-436. doi: https: / / doi.Org / 10.1016 / S1535-6108(03)00310-6.

[0295] Davis, A. A. and Patel, V. G. (2019) ‘The role of PD-L1 expression as a predictive biomarker: An analysis of all US food and drug administration (FDA) approvals of immune checkpoint inhibitors’, Journal for ImmunoTherapy of Cancer, 7(1), pp. 1-8. doi: 10.1186 / s40425-019-0768- 9.

[0296] Dobin, A. etal. (2013) ‘STAR: Ultrafast universal RNA-seq aligner’, Bioinformatics, doi: 10.1093 / bioinformatics / bts635.

[0297] Ebert, P. J. R. etal. (2016) ‘MAP Kinase Inhibition Promotes T Cell and Anti-tumor Activity in Combination with PD-L1 Checkpoint Blockade’, Immunity, 44(3), pp. 609-621. doi: 10.1016 / j.immuni.2O16.01 .024.

[0298] Feng, B. et al. (2020) ‘Integrative analysis of multi-omics data identified EGFR and PTGS2 as key nodes in a gene regulatory network related to immune phenotypes in head and neck cancer’, Clinical Cancer Research, doi: 10.1158 / 1078-0432. ccr-19-3997.

[0299] Fridman, W. H. et al. (2012) ‘The immune contexture in human tumours: impact on clinical outcome’, Nature Reviews Cancer, 12(4), pp. 298-306. doi: 10.1038 / nrc3245.

[0300] Gajewski, T. F. etal. (2019) ‘WNT / p-catenin pathway activation correlates with immune exclusion across human cancers’, Clinical Cancer Research, p. clincanres.1942.2018. doi: 10.1158 / 1078-0432. ccr-18-1942.

[0301] Garnett, M. J. etal. (2012) ‘Systematic identification of genomic markers of drug sensitivity in cancer cells’, Nature, 483(7391), pp. 570-575. doi: 10.1038 / nature11005.

[0302] Georgouli, M. et al. (2019) ‘Regional Activation of Myosin II in Cancer Cells Drives Tumor Progression via a Secretory Cross-Talk with the Immune Microenvironment’, Cell, 176(4), pp. 757-774. e23. doi: 10.1016 / j.cell.2O18.12.038.

[0303] Gerlinger, M. et al. (2014) ‘Genomic architecture and evolution of clear cell renal cell carcinomas defined by multiregion sequencing’, Nature Genetics, doi: 10.1038 / ng.2891 . Gide, T. N. etal. (2019) ‘Distinct Immune Cell Populations Define Response to Anti-PD-1 Monotherapy and Anti-PD-1 / Anti-CTLA-4 Combined Therapy’, Cancer Cell, 35(2), pp. 238- 255. e6. doi: 10.1016 / j.ccell.2O19.01.003.

[0304] Green, J. A. etal. (2017) ‘A nonimmune function of T cells in promoting lung tumor progression’, Journal of Experimental Medicine, 214(12), pp. 3565-3575. doi: 10.1084 / jem.20170356.

[0305] Guo, X. etal. (2018) ‘Global characterization of T cells in non-small cell lung cancer by singlecell sequencing’, Nature Medicine, 24(July), pp. 1-18. doi: 10.1038 / s41591 -018-0045-3. Hamarsheh, S. etal. (2020) ‘Immune modulatory effects of oncogenic KRAS in cancer’, Nature Communications, 11 (1 ), p. 5439. doi: 10.1038 / s41467-020-19288-6. Harris, S. G. et al. (2002) ‘Prostaglandins as modulators of immunity’, Trends in Immunology. doi: 10.1016 / S1471 -4906(01 )02154-8.

[0306] Havel, J. J. (2019) ‘The evolving landscape of biomarkers for checkpoint inhibitor immunotherapy’, Nature Reviews Cancer, 19(March). doi: 10.1038 / s41568-019-0116-x. Hegde, P. S. and Chen, D. S. (2020) ‘Perspective Top 10 Challenges in Cancer Immunotherapy’, Immunity, 52(1), pp. 17-35. doi: 10.1016 / j.immuni.2019.12.011 .

[0307] Hennequart, M. et al. (2017) ‘Constitutive IDO1 Expression in Human Tumors Is Driven by Cyclooxygenase-2 and Mediates Intrinsic Immune Resistance’, Cancer Immunology Research, 5(8), pp. 695 LP - 709. doi: 10.1158 / 2326-6066.CIR-16-0400.

[0308] Hu-Lieskovan, S. et al. (2015) ‘Improved antitumor activity of immunotherapy with BRAF and MEK inhibitors in BRAFV600E melanoma’, Science Translational Medicine, 7(279). doi: 10.1126 / scitranslmed.aaa4691 .

[0309] Hugo, W. et al. (2017) ‘Genomic and transriptomic features of anti-PD1 response’, 165(1), pp. 35-44. doi: 10.1016 / j.cell.2016.02.065.Genomic.

[0310] Jerby-Arnon, L. etal. (2018) ‘A Cancer Cell Program Promotes T Cell Exclusion and Resistance to Checkpoint Blockade’, Cell, 175(4), pp. 984-997.e24. doi: 10.1016 / j.cell.2O18.09.006.

[0311] Keenan, T. E., Burke, K. P. and Van Allen, E. M. (2019) ‘Genomic correlates of response to immune checkpoint blockade’, Nature Medicine, doi: 10.1038 / s41591-019-0382-x.

[0312] Kim, S. T. et al. (2018) ‘Comprehensive molecular characterization of clinical responses to PD-1 inhibition in metastatic gastric cancer’, Nature Medicine, doi: 10.1038 / s41591 -018-0101 -z. Lambrechts, D. etal. (2018) ‘Phenotype molding of stromal cells in the lung tumor microenvironment’, Nature Medicine, 24(8), pp. 1277-1289. doi: 10.1038 / s41591 -018-0096-5. Lawrence, M. S. et al. (2014) ‘Discovery and saturation analysis of cancer genes across 21 tumour types’, Nature. 2014 / 01 / 05, 505(7484), pp. 495-501. doi: 10.1038 / nature12912.

[0313] Lawrence, M. S. et al. (2015) ‘Comprehensive genomic characterization of head and neck squamous cell carcinomas’, Nature, 517(7536), pp. 576-582. doi: 10.1038 / nature14129. van der Leun, A. M., Thommen, D. S. and Schumacher, T. N. (2020) ‘CD8+ T cell states in human cancer: insights from single-cell analysis’, Nature Reviews Cancer, 20(4), pp. 218-232. doi: 10.1038 / S41568-019-0235-4.

[0314] Liao, Y., Smyth, G. K. and Shi, W. (2014) ‘FeatureCounts: An efficient general purpose program for assigning sequence reads to genomic features’, Bioinformatics, doi:

[0315] 10.1093 / bioinformatics / btt656.

[0316] Liberzon, A. etal. (2015) ‘The Molecular Signatures Database Hallmark Gene Set Collection’, Cell Systems, 1 (6), pp. 417-425. doi: 10.1016 / j.cels.2O15.12.004.

[0317] Litchfield, K., Reading, J. L., et al. (2020) ‘Escape from nonsense-mediated decay associates with anti-tumor immunogenicity’, Nature Communications, 11 (1 ), pp. 1-11. doi: 10.1038 / S41467-020-17526-5. Litchfield, K., Stanislaw, S., et al. (2020) ‘Representative Sequencing: Unbiased Sampling of Solid Tumor Tissue’, Cell Reports, 31 (5). doi: 10.1016 / j.celrep.2020.107550.

[0318] Litchfield, K. etal. (2021) ‘M eta-analysis of tumor- and T cell-intrinsic mechanisms of sensitization to checkpoint inhibition’, Cell, pp. 1-19. doi: 10.1016 / j. cell.2021 .01 .002. Liu, David et al. (2019) ‘Integrative molecular and clinical modeling of clinical outcomes to PD1 blockade in patients with metastatic melanoma’, Nature Medicine, 25(December). doi: 10.1038 / S41591 -019-0654-5.

[0319] Mantovani, A. et al. (2008) ‘Cancer-related inflammation’, Nature, 454(7203), pp. 436-444. doi: 10.1038 / nature07205.

[0320] Mariathasan, S. et al. (2018) ‘TGFp attenuates tumour response to PD-L1 blockade by contributing to exclusion of T cells’, Nature, 554(7693), pp. 544-548. doi: 10.1038 / nature25501 . Markosyan, N. etal. (2019) ‘Tumor cell-intrinsic EPHA2 suppresses antitumor immunity by regulating PTGS2 (COX-2)’, Journal of Clinical Investigation, doi: 10.1172 / JC1127755.

[0321] Martinez-Jimenez, F. etal. (2020) ‘A compendium of mutational cancer driver genes’, Nature Reviews Cancer, 20(10), pp. 555-572. doi: 10.1038 / s41568-020-0290-x.

[0322] Mascaux, C. etal. (2019) ‘Immune evasion before tumour invasion in early lung squamous carcinogenesis’, Nature, 571 (7766), pp. 570-575. doi: 10.1038 / s41586-019-1330-0.

[0323] McCarthy, D. J., Chen, Y. and Smyth, G. K. (2012) ‘Differential expression analysis of multifactor RNA-Seq experiments with respect to biological variation’, Nucleic Acids Research, 40(10), pp. 4288-4297. doi: 10.1093 / nar / gks042.

[0324] Mcdermott, D. F. et al. (2018) ‘Clinical activity and molecular correlates of response to atezolizumab alone or in combination with bevacizumab versus sunitinib in renal cell carcinoma David’, Nature medicine, 24(June). doi: 10.1038 / s41591 -018-0053-3.

[0325] McGranahan, N. et al. (2016) ‘Clonal neoantigens elicit T cell immunoreactivity and sensitivity to immune checkpoint blockade’, Science, 351 (6280), pp. 1463 LP - 1469. Available at: http: / / science.sciencemag.org / content / 351 / 6280 / 1463.abstract.

[0326] McLane, L. M., Abdel-Hakeem, M. S. and Wherry, E. J. (2019) ‘CD8 T Cell Exhaustion During Chronic Viral Infection and Cancer’, Annual Review of Immunology, doi: 10.1146 / annurev- immunol-041015-055318.

[0327] Moraes, L. A., Ampomah, P. B. and Lim, L. H. K. (2018) ‘Annexin A1 in inflammation and breast cancer: a new axis in the tumor microenvironment’, Cell Adhesion & Migration, 12(5), pp. 417- 423. doi: 10.1080 / 19336918.2018.1486143.

[0328] Motzer, R. J. etal. (2020) ‘Avelumab plus axitinib versus sunitinib in advanced renal cell carcinoma: biomarker analysis of the phase 3 JAVELIN Renal 101 trial’, Nature Medicine, 26(11), pp. 1733-1741. doi: 10.1038 / s41591 -020-1044-8.

[0329] Newman, A. M. etal. (2015) ‘Robust enumeration of cell subsets from tissue expression profiles’, Nature Methods, 12(5), pp. 453-457. doi: 10.1038 / nmeth.3337. Okuda, H. etal. (2012) ‘Hyaluronan Synthase HAS2 Promotes Tumor Progression in Bone by Stimulating the Interaction of Breast Cancer Stem-Like Cells with Macrophages and Stromal Cells’, Cancer Research, 72(2), pp. 537 LP - 547. doi: 10.1158 / 0008-5472.CAN-11-1678.

[0330] Pelly, V. S. et al. (2021) ‘Anti-inflammatory drugs remodel the tumor immune environment to enhance immune checkpoint blockade efficacy’, Cancer Discovery, p. candisc.1815.2020. doi: 10.1158 / 2159-8290.CD-20-1815.

[0331] Polidoro, M. A. etal. (2020) ‘Impact of RAS mutations on the immune infiltrate of colorectal liver metastases: A preliminary study’, Journal of Leukocyte Biology, 108(2), pp. 715-721. doi:

[0332] 10.1002 / JLB.5AB0220-608R.

[0333] Puram, S. V. et al. (2017) ‘Single-Cell Transcriptomic Analysis of Primary and Metastatic Tumor Ecosystems in Head and Neck Cancer’, Cell, pp. 1-14. doi: 10.1016 / j. cell.2017.10.044.

[0334] Ramilowski, J. A. et al. (2015) ‘A draft network of ligand-receptor-mediated multicellular signalling in human’, Nature Communications, 6. doi: 10.1038 / ncomms8866.

[0335] Riaz, N. et al. (2017) ‘Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab’, Cell, 171 (4), pp. 934-949.e15. doi: 10.1016 / j.cell.2O17.09.028.

[0336] Ritchie, M. E. etal. (2015) ‘limma powers differential expression analyses for RNA-sequencing and microarray studies’, Nucleic Acids Research, 43(7), pp. e47-e47. doi: 10.1093 / nar / gkv007. Rizvi, N. A. etal. (2015) ‘Mutational landscape determines sensitivity to PD-1 blockade in nonsmall cell lung cancer’, Science, 348(6230), pp. 124 LP - 128. Available at: http: / / science.sciencemag.org / content / 348 / 6230 / 124.abstract.

[0337] Rooney, M. S. etal. (2015) ‘Molecular and genetic properties of tumors associated with local immune cytolytic activity’, Cell, 160(1-2), pp. 48-61. doi: 10.1016 / j. cell.2014.12.033.

[0338] Rosenthal, R. etal. (2019) ‘Neoantigen-directed immune escape in lung cancer evolution’, Nature, doi: 10.1038 / s41586-019- 1032-7.

[0339] Sade-Feldman, M. et al. (2018) ‘Defining T Cell States Associated with Response to Checkpoint Immunotherapy in Melanoma’, Cell, doi: 10.1016 / J.CELL.2018.10.038.

[0340] Salmon, H. etal. (2019) ‘Host tissue determinants of tumour immunity’, Nature Reviews Cancer. doi: 10.1038 / S41568-019-0125-9.

[0341] Shalapour, S. and Karin, M. (2019) ‘Pas de Deux: Control of Anti-tumor Immunity by Cancer- Associated Inflammation’, Immunity, doi: 10.1016 / j.immuni.2019.06.021.

[0342] Singer, M. et al. (2016) ‘A Distinct Gene Module for Dysfunction Uncoupled from Activation in Tumor-Infiltrating T Cells’, Cell, 166(6), pp. 1500-1511.e9. doi: 10.1016 / j.cell.2O16.08.052.

[0343] Snyder, A. et al. (2014) ‘Genetic Basis for Clinical Response to CTLA-4 Blockade in Melanoma’, New England Journal of Medicine, 371 (23), pp. 2189-2199. doi: 10.1056 / NEJMoa1406498.

[0344] Snyder, A. et al. (2017) ‘Contribution of systemic and somatic factors to clinical response and resistance to PD-L1 blockade in urothelial cancer: An exploratory multi-omic analysis’, PLoS Medicine, 14(5), pp. 1-24. doi: 10.1371 / journal.pmed.1002309. Szklarczyk, D. etal. (2019) ‘STRING v11: protein-protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets’, Nucleic Acids Research, 47(D1 ), pp. D607-D613. doi: 10.1093 / nar / gky1131 .

[0345] Templeton, I. E. and Musib, L. (2015) ‘MEK inhibitors beyond monotherapy: Current and future development’, Current Opinion in Pharmacology, doi: 10.1016 / j.coph.2015.05.012.

[0346] Thorsson, V. et al. (2018) ‘The Immune Landscape of Cancer’, Immunity, pp. 1-19. doi: 10.1016 / j.immuni.2018.03.023.

[0347] Todoric, J. and Karin, M. (2019) ‘The Fire within: Cell-Autonomous Mechanisms in Inflammation-Driven Cancer’, Cancer Cell, 35(5), pp. 714-720. doi: 10.1016 / j.ccell.2O19.04.001 . Tumeh, P. C. et al. (2014) ‘PD-1 blockade induces responses by inhibiting adaptive immune resistance’, Nature, 515(7528), pp. 568-571. doi: 10.1038 / natu re 13954.

[0348] Wang, D. and Dubois, R. N. (2010) ‘Eicosanoids and cancer’, Nature Reviews Cancer, 10(3), pp. 181-193. doi: 10.1038 / nrc2809.

[0349] Wang, L. etal. (2019) ‘EMT- A nd stroma-related gene expression and resistance to pd-1 blockade in urothelial cancer’, Journal of Urology, 202(3), p. 458. doi: 10.1038 / s41467-018- 05992-x.

[0350] Wang, Y. etal. (2019) ‘Epiregulin reprograms cancer-associated fibroblasts and facilitates oral squamous cell carcinoma invasion via JAK2-STAT3 pathway’, Journal of Experimental & Clinical Cancer Research, 38(1), p. 274. doi: 10.1186 / s13046-019-1277-x.

[0351] Yuen, K. C. etal. (2020) ‘High systemic and tumor-associated IL8 correlates with reduced clinical benefit of PD-L1 blockade’, Nature medicine, 26(May). doi: 10.1038 / s41591 -020-0860-1. Zelenay, S. et al. (2015) ‘Cyclooxygenase-Dependent Tumor Growth through Evasion of Immunity’, Cell, 162(6), pp. 1257-1270. doi: 10.1016 / j.cell.2O15.08.015.

[0352] For standard molecular biology techniques, see Sambrook, J., Russel, D.W. Molecular Cloning, A Laboratory Manual. 3 ed. 2001 , Cold Spring Harbor, New York: Cold Spring Harbor Laboratory Press

Claims

Claims:1 . A computer-implemented method for predicting the treatment response of a mammalian cancer patient to anti-cancer immunotherapy, the method comprising: a) providing a plurality of features comprising gene expression levels of at least 8, at least 9, at least 10, at least 11 , at least 12 or more (such as all) of the following cancer promoting genes PTGS2, CXCL1 , CXCL2, CXCL5, CXCL6, IL1 A, HAS2, IL11 , IL1 B, EREG, IL6, CXCL8 (ILS), MMP12, NRG1 , VCAN, AREG previously measured in a sample obtained from the tumour of a mammalian cancer patient; b) computing a score using the features provided in a); and c) predicting the treatment response of the mammalian cancer patient based on the score computed in step b).

2. The computer-implemented method of claim 1 , wherein the features further compriseClonal_TMB and CXCL9.

3. The computer-implemented method of claim 1 or claim 2, wherein the features comprise the gene expression levels of all of: PTGS2, CXCL1 , CXCL2, CXCL5, CXCL6, IL1 A, HAS2,IL11 , IL1 B, EREG, IL6, CXCL8 (IL8), MMP12, NRG1 , VCAN, AREG.

4. The computer-implemented method of any one of the preceding claims, wherein the gene expression level is a normalised gene expression level.

5. The computer-implemented method of any one of the preceding claims, wherein the score is computed by z-score normalising the individual gene expression values of the features and calculating the mean of said normalised measurements.

6. The computer-implemented method of any one of the preceding claims, wherein the score is computed according to the formula: scorewherein npis the number of said gene expression features, Gosis the positive correlated genes, within an (i) interval of unitary values, and (e) represents the gene expression values expressed as Iog2 counts per million (CPM); optionally wherein Gipos(e) is the mean expression of Iog2 transformed counts per million (reads per kilobase million (FPKM)) of positive genes.

7. The method of any one of claims 1 to 5, wherein the score is a probabilistic score calculated using the formula:wherein Po is an intercept weight, is a vector of weights for each of k variables and % is a vector of variables associated with the sample, wherein the variables comprise said plurality of features.

8. The method of any one of the preceding claims, wherein the computer-implemented method further comprises assessing the tumour burden and / or neoantigen prevalence of the cancer patient.

9. The method of any one of the preceding claims, wherein predicting the treatment response of the mammalian cancer patient comprises: comparing the computed score from step b) with a reference median score derived from a sample cohort of responders and a reference median score derived from a sample cohort of non-responders, wherein the sample cohort of responders and non-responders are of i) cancer patients having the same type of cancer as said cancer patient, or ii) cancer patients which are tumour type agnostic.

10. The method of any one of the preceding claims, wherein the score computed in step b) indicates that the cancer is predicted to respond to anti-cancer immunotherapy, and the method further comprises selecting the mammalian cancer patient for anti-cancer immunotherapy.11 . The method of any one of the preceding claims, wherein the gene expression levels of the at least 8 cancer promoting genes are acquired by measuring the gene expression levels of at least 8, at least 9, at least 10, at least 11 , at least 12 or more (such as all) of the following cancer promoting genes: PTGS2, CXCL1 , CXCL2, CXCL5, CXCL6, IL1 A, HAS2, IL11 , IL1 B, EREG, IL6, CXCL8 (ILS), MMP12, NRG1 , VCAN, AREG in a sample obtained from the tumour of a mammalian cancer patient.

12. A method of predicting the treatment response of a mammalian cancer patient to anti-cancer immunotherapy, the method comprising: a) measuring a plurality of features comprising measuring the gene expression of at least 8, at least 9, at least 10, at least 1 1 , at least 12 or more (such as all) of the following cancer promoting genes: PTGS2, CXCL1 , CXCL2, CXCL5, CXCL6, IL1 A, HAS2, IL11 ,IL1 B, EREG, IL6, CXCL8 (IL8), MMP12, NRG1 , VCAN, AREG in a sample obtained from the tumour of a patient; b) determining a score using the features measured in a); and c) predicting the treatment response of the patient based on the score determined in step b).

13. The method of claim 12, wherein the features further comprise Clonal_TMB and CXCL9.

14. The method of any one of the preceding claims, wherein determining the score to predict the treatment response of a subject comprises: inputting data representing the features comprising the gene expression levels of the at least 8 cancer promoting genes of the subject into a machine learning classifier, wherein said machine learning classifier has been trained on a training data set comprising the gene expression levels of at least the same cancer promoting genes as those forming the score of the subject, and wherein the training data set is derived from at least 2 sample cohorts; wherein one cohort comprises a plurality of samples derived from a plurality of mammalian cancer patients known to have responded to anti-cancer immunotherapy (responders), and the other cohort comprises a plurality of samples derived from mammalian cancer patients known not to respond to anti-cancer immunotherapy (nonresponders); and causing the machine learning classifier to classify the score of the subject as belonging to the response group or the non-response group.

15. The method of claim 14, wherein the training data set comprises gene expression levels of the at least 8 cancer promoting genes of at least 50, at least 100, at least 150, at least 200, at least 250, at least 300 samples derived from mammalian cancer patients known to have responded to anti-cancer immunotherapy (responders) and gene expression levels of the at least 8 cancer promoting genes of at least 50, at least 100, at least 150, at least 200, at least 250, at least 300, at least 350, at least 400, at least 450, at least 500, at least 550 samples derived from mammalian cancer patients known not to have responded to anticancer immunotherapy (non-responders).

16. The method of claims 14 and 15, wherein the machine learning classifier has been trained on a plurality of scores derived from at least three sample cohorts, wherein one cohort comprises a plurality of samples derived from a plurality of mammalian cancer patients known to have completely responded to anti-cancer immunotherapy (complete responders), one cohort comprises a plurality of samples derivedfrom a plurality of mammalian cancer patients known to have partially responded to anticancer immunotherapy (partial responders), and the other cohort comprises a plurality of samples derived from mammalian cancer patients known not to respond to anti-cancer immunotherapy (non-responders).

17. The method of claims 14 to 16, wherein the machine learning classifier is a logistic regression model, elastic net regression model or a tree-based model such as a random forest classifier.

18. A method of classifying a sample obtained from a mammalian cancer subject, said sample comprising nucleic acid and / or protein, the method comprising: a) analysing the sample to obtain the gene expression levels of the at least 8 cancer promoting genes selected from: PTGS2, CXCL1 , CXCL2, CXCL5, CXCL6, IL1 A, HAS2, IL11 , IL1 B, EREG, IL6, CXCL8 (IL8), MMP12, NRG1 , VCAN and AREG; and b) carrying out the method of any one of claims 1 to 17 using the gene expression levels obtained in step a).

19. The method of claim 18, wherein said sample is derived from one or more cancer cells of the subject.

20. A method of treating a cancer in a mammalian cancer patient, the method comprising: a) carrying out the method of any one of the preceding claims to predict the treatment response of the mammalian cancer patient to anti-cancer immunotherapy; b) determining that the mammalian cancer patient is predicted to respond to said anticancer immunotherapy in step a); and c) administering a therapeutically effective amount of said anti-cancer immunotherapy to the subject in need thereof.21 . A method of treatment of a cancer in a mammalian cancer patient, comprising: a) carrying out the method of any one of claims 1 to 20; b) determining that the score of the mammalian cancer patient indicates that the mammalian cancer patient is predicted to respond to anti-cancer immunotherapy; and c) administering immunotherapy to the mammalian cancer patient in need thereof.

22. The method of any one of the preceding claims, wherein the method further comprises analysing one or more additional markers of anti-cancer immunotherapy response derivedfrom the mammalian cancer patient to supplement and / or corroborate the prediction of antitreatment response or non-response.

23. The method according to any one of the preceding claims, wherein the sample obtained from the tumour of the patient is a formalin-fixed-paraffin embedded (FFPE) sample.

24. A system for predicting the treatment response of a mammalian cancer patient to an anticancer immunotherapy, the system comprising: at least one processor; and at least one non-transitory computer readable medium containing instructions that when executed by the at least one processor, cause the at least one processor to perform the method of any one of claims 1 to 17.

25. One or more computer readable media comprising instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the method of any one of claims 1 to 17.

26. A pharmaceutical composition for use in the treatment of a mammalian cancer patient comprising an anti-cancer immunotherapy, wherein the patient has been predicted to respond to said anti-cancer immunotherapy by a method of any one of claims 1 to 19.

27. The method, system and / or pharmaceutical composition according to any one of the preceding claims, wherein the mammalian cancer patient has and / or the cancer is: renal cancer, such as renal cell carcinoma or kidney renal cell carcinoma (KIRC) ; gastroesophageal tumour; oesophageal cancer (ECA); melanoma; bladder cancer (BLCA); gastric cancer; lung cancer, such as lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC) or nonsmall cell lung cancer (NSCLC); pancreatic cancer (PAAD); liver cancer (LIHC); glioblastoma (GBM); adrenocortical carcinoma (ACC);cervical cancer (CESC); head and neck cancer such as head and neck squamous cell carcinoma (HNSC); thyroid cancer (THCA); uterine cancer (UCEC); Tenosynovial giant cell tumor (TGCT); melanoma; BRCA; sarcomas and pan-cancer.

28. The method, system and / or pharmaceutical composition according to claim 27, wherein the renal cell carcinoma is clear cell renal cell carcinoma (ccRCC).

29. The method, system and / or pharmaceutical composition according to any one of the preceding claims, wherein the immune subtype of the cancer is one of: wound healing, IFNy- dominant, TGFp-dominant, inflammatory and / or lymphocyte depleted subtype.

30. The method, system and / or pharmaceutical composition according to any one of the preceding claims, wherein said anti-cancer immunotherapy comprises IL-2 immunotherapy or immune checkpoint blockade therapy alone, or in combination with anti-VEGF therapies.31 . The method, system and / or pharmaceutical composition according to claim 30, wherein the immune checkpoint blockade therapy comprises programmed death-1 (PD-1 ) blockade, programmed death-ligand 1 (PD-L1 ) blockade and / or cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) blockade.

32. The method, system and / or pharmaceutical composition according to claim 30 or claim 31 , wherein said immune checkpoint blockade therapy comprises treatment with Nivolumab, Pembrolizumab, Atezolizumab and / or Ipilimumab.

33. The method, system and / or pharmaceutical composition according to claim 30, wherein the anti-VEGF therapy is bevacizumab.

Images