NOVEL PREDICTION METHOD AND GENE SIGNATURES FOR THE TREATMENT OF CANCER

Abstract

The present invention relates to methods for predicting the clinical outcome or the response of a patient suffering from cancer to an anti-cancer therapy based on novel gene signatures.

Claims

1-15. (canceled)

16. A method for selecting a gene signature being indicative of a therapeutic benefit of an anti-cancer therapy in a patient or of the clinical outcome of a patient, wherein the method comprises: providing, for several patients having a cancer and being treated with the same anti-cancer therapy and for which the clinical outcome is known, mRNA expression level in a tumor sample and a histologically matched normal sample of each gene of a set of n key genes known to be involved in the anti-cancer therapy; and wherein the tumor and normal histologically matched samples are from the same subject; determining a mRNA fold change of Tumor versus Normal (Fc.sub.n TvN) for each gene of the set of n key genes; multiplying the Fc.sub.n TvN by expression intensity of the gene (I.sub.n), either in the tumor sample (I.sub.n T) or in the histologically matched normal sample (I.sub.n N) or both, thereby determining a function (Fg.sub.n) for each gene of the set of n key genes; determining, for each gene of the set of n key genes, a correlation between the Fg.sub.n and the clinical outcome, and ranking the genes based on the correlation from the most correlated to the less correlated to the clinical outcome; combining 2 to n genes of the set of n key genes with an increment of 1, the genes being selected in the ranking order from the most correlated to the less correlated, and determining the correlation between each of the combination and the clinical outcome; and selecting the combination of genes best correlated with the clinical outcome, thereby providing a gene signature indicative of a therapeutic benefit with the anti-cancer therapy or of the clinical outcome of the patient.

17. The method according to claim 16, comprising: identifying a set of n key genes known to be involved in a mechanism of the anti-cancer therapy; selecting a group of patients with available transcriptomics data and clinical outcome under treatment with said anti-cancer therapy; providing mRNA expression level in a tumor sample and a histologically matched normal sample from the same patient for each gene of the set of n key genes identified in (1), and determining a mRNA Fc.sub.n TvN for each gene of the set of n key genes; for each gene of the n key genes, multiplying the Fc.sub.n TvN by expression intensity of the gene (I.sub.n) in the tumor sample (I.sub.n T) and multiplying the Fc.sub.n TvN by expression intensity of the gene (I.sub.n) in the histologically matched normal sample (I.sub.n N), thereby determining a Fg.sub.n with I.sub.n T and a Fg.sub.n with I.sub.n N; determining a correlation between Fg.sub.n with I.sub.n T and the clinical outcome of the selected patients and between Fg.sub.n with I.sub.n N and the clinical outcome of the selected patients; selecting the best correlation between Fg.sub.n with I.sub.n T and Fg.sub.n with I.sub.n N and keeping it as Fg.sub.n for the following steps; ranking the genes of the n key genes according to the correlation of Fg.sub.n with the clinical outcome, the highest correlation (F.sub.g) being ranked first; adding to the F.sub.g ranked first, one by one, the F.sub.g of other genes of the set of n key genes until all key genes are present thereby obtaining genes combinations, and determining the correlation between the combinations of 2 to n genes of the set of n key genes and the clinical outcome; selecting the combination having the best correlation, thereby providing a gene signature indicative of a therapeutic benefit with the anti-cancer therapy or of the clinical outcome of the patient treated with the anti-cancer therapy.

18. An in vitro method for predicting the clinical outcome of a subject suffering from cancer and being treated with an anti-cancer therapy or for predicting the response of a subject suffering from cancer to an anti-cancer therapy, wherein the method comprises: providing mRNA expression level of each gene of a gene signature selected from the group consisting of: i) a gene signature comprising AKT2, TSC1, FKB-12, TSC2, RPTOR, RHEB, PIK3CA and PIK3CB; ii) a gene signature comprising KITLG and KIT; iii) a gene signature comprising ERK2, ARAF, CRAF, MIK1, MEK2, HRAS, ERK1, MAPK10 and KSR1; iv) a gene signature comprising NRG4 and NRG2; v) a gene signature comprising FGF10, FGF16, FGF5, FGF2 and FGF13; and vi) a gene signature comprising TLR4, PDL2, PDL1, CD16(NK), CTLA4 and CD28; in a tumor sample and a histologically matched normal sample, wherein the tumor and histologically matched normal samples are both from the same subject suffering from cancer, determining a mRNA fold change of Tumor versus Normal (Fc.sub.n TvN) for each gene of the signature, multiplying the Fc.sub.n TvN by expression intensity of the gene (I.sub.n) in the tumor sample (I.sub.n T) and/or in the normal histologically matched sample (I.sub.n N) for each gene of the signature, determining a score based on the Fc.sub.n TvN multiplied by I.sub.nT and/or I.sub.n N, for the genes of the gene signature, this score being indicative of the clinical outcome of the patient suffering from cancer and having an anti-cancer therapy or the responsiveness of the patient to the anti-cancer therapy.

19. The method of claim 18, wherein the anti-cancer therapy is a Tyrosine kinase inhibitor; or a FGFR inhibitor; or a PD-1/PD-L1 inhibitor.

20. The method of claim 18, wherein (i) when the anti-cancer therapy is a mTOR inhibitor, the gene signature comprises AKT2, TSC1, FKB-12, TSC2, RPTOR, RHEB, PIK3CA and PIK3CB; (ii) when the anti-cancer therapy is a VEGFR inhibitor, the gene signature comprises KIT and KITLG; (iii) when the anti-cancer therapy is a pan-HER inhibitor, the gene signature comprises NRG4 and NRG2; (iv) when the anti-cancer therapy is a MEK inhibitor, the gene signature comprises ERK2, ARAF, CRAF; MEK1, MEK2, HRAS, ERK1, MAPK10 and KSR1; (v) when the anti-cancer therapy is a PD-1/PD-L1 inhibitor, the gene signature comprises TLR4, PDL2, PDL1, CD16, CTLA4 and CD28; and (vi) when the anti-cancer therapy is a FGFR inhibitor, the gene signature comprises FGF10, FGF16, FGF5, FGF2 and FGF13.

21. The method of claim 20, wherein the anti-cancer therapy is everolimus, and the method comprises: (a) providing mRNA expression level of each gene of the following gene signature: AKT2, TSC1, FKB-12, TSC2, RPTOR, RHEB, PIK3CA and PIK3CB, in a tumor sample and a histologically matched normal sample, wherein the tumor and histologically matched normal samples are both from the same subject suffering from cancer, and determining a mRNA fold change of Tumor versus Normal (Fc.sub.n TvN) for each of the 8 genes of the gene signature; (b) determining the intensity of gene expression in the tumor sample (I.sub.n T) for each of the 8 genes of the gene signature; (c) determining the absolute value of the fold of log 2(Fc.sub.n TvN) multiplied by log 1.1 (I.sub.n T) for each of the 8 genes of the gene signature and, based on a linear regression, determining a predicted progression-free survival (PFS), the predicted PFS being indicative of the clinical outcome or of the response of the subject to everolimus.

22. The method of claim 20, wherein the anti-cancer therapy is a PD-1/PD-L1 inhibitor or an anti-PD-1 or anti-PD-L1 antibody, and the method comprises: (a) providing mRNA expression level of each gene of the following gene signature: TLR4, PDL2, PDL1, CD16(NK), CTLA4 and CD28 in a tumor sample and a histologically matched normal sample, wherein the tumor and histologically matched normal samples are both from the same subject suffering from cancer, and determining the mRNA fold change of Tumor versus Normal (Fc.sub.n TvN) for each of the 6 genes of the gene signature; (b) determining the intensity of gene expression in the histologically matched normal sample (I.sub.n N) for each of the 6 genes of the gene signature; (c) determining the fold of log 2(Fc.sub.n TvN) multiplied by log 1.1 (I.sub.n N) for each of the 6 genes of the gene signature and, based on a linear regression, determining a predicted progression-free survival (PFS), the predicted PFS being indicative of the clinical outcome or of the response of the subject to the PD-1/PD-L1 inhibitor.

23. The method of claim 20, wherein the anti-cancer therapy is axitinib and the method comprises: (a) providing mRNA expression level of each gene of the following gene signature: KIT and KITLG in a tumor sample and a histologically matched normal sample, wherein the tumor and histologically matched normal samples are both from the same subject suffering from cancer, and determining the mRNA fold change of Tumor versus Normal (Fc.sub.n TvN) for each of the 2 genes of the gene signature; (b) determining the intensity of gene expression in the histologically matched normal sample (I.sub.n N) for each of the 2 genes of the gene signature; (c) determining the sum of log 2(Fc.sub.n TvN) multiplied by log 1.1 (I.sub.n N) for each of the 2 genes of the gene signature and, based on a linear regression, determining a predicted PFS, the predicted PFS being indicative of the clinical outcome or the response of the subject to axitinib.

24. The method of claim 20, wherein the anti-cancer therapy is afatinib and the method comprises: (a) providing mRNA expression level for each gene of the following gene signature: NRG4 and NRG2 in a tumor sample and a histologically matched normal sample, wherein the tumor and histologically matched normal samples are both from the same subject suffering from cancer, and determining the mRNA fold change of Tumor versus Normal (Fc.sub.n TvN) for each of the 2 genes of the gene signature; (b) determining the intensity of gene expression in the tumor sample (I.sub.n T) for each of the 2 genes of the gene signature; (c) determining the sum of log 2(Fc.sub.n TvN) multiplied by log 1.1 (I.sub.n T) for each of the 2 genes of the gene signature and, based on a linear regression, determining a predicted PFS, the predicted PFS being indicative of the clinical outcome or the response of the subject to afatinib.

25. The method of claim 20, wherein the anti-cancer therapy is trametinib and the method comprises: (a) providing mRNA expression level of each gene of the following gene signature: ERK2 ARAF, CRAF, MEK1, MEK2, HRAS, ERK1, MAPK10 and KSR1 in a tumor sample and a histologically matched normal sample, wherein the tumor and histologically matched normal samples are both from the same subject suffering from cancer, and determining mRNA fold change of Tumor versus Normal (Fc.sub.n TvN) for each of the 9 genes of the gene signature; (b) determining the intensity of gene expression in the tumor sample (I.sub.n T) for each of the 9 genes of the gene signature; (c) determining the fold of log 2(Fc.sub.n TvN) multiplied by log 1.1 (I.sub.n T) for each of the 9 genes of the gene signature and, based on a linear regression, determining a predicted PFS, the predicted PFS being indicative of the clinical outcome or the response of the subject to trametinib.

26. The method of claim 20, wherein the anti-cancer therapy is a FGFR inhibitor and the method comprises: (a) providing mRNA expression level of each gene of the following gene signature: FGF10, FGF16, FGF5, FGF2 and FGF13 in a tumor sample and a histologically matched normal sample, wherein the tumor and histologically matched normal samples are both from the same subject suffering from cancer, and determining the mRNA fold change of Tumor versus Normal (Fc.sub.n TvN) for each gene of the 5 genes of the gene signature; (b) determining the intensity of gene expression in the histologically matched normal sample (I.sub.n N) for each of the 5 genes of the gene signature; (c) determining the sum of log 2(Fc.sub.n TvN) multiplied by log 1.1 and sum of log 2(Fc.sub.n TvN) multiplied by log 1.1 (I.sub.n N) for each of the 5 genes of the gene signature, and based on a linear regression, determining a predicted PFS, the predicted PFS being indicative of the clinical outcome or the response of the subject to the FGFR inhibitor.

27. The method of claim 18, wherein the cancer is selected from the group consisting of prostate cancer, bladder cancer, breast cancer, colon cancer, colorectal cancer, Esophagus cancer, hypopharynx cancer, gastric cancer, rectum cancer, head and neck cancer, liver cancer, brain cancer, hepatocarcinoma, kidney cancer, ovarian cancer, cervical cancer, pancreatic cancer, Leiomyosarcoma, Liposarcoma, lung cancer, lymphoma, osteosarcoma, melanoma, neuroendocrine cancer, pleural cancer, Rhabdomyosarcoma, Small Intestine neuroendocrine cancer, endometrial cancer, soft tissue cancer, non-small cell lung carcinomas (NSCLC), metastatic non-small cell lung cancer, muscle cancer, adrenal cancer, thyroid cancer, uterine cancer, advanced renal cell carcinoma (RCC), and sub ependymal giant cell astrocytoma (SEGA) associated with tuberous sclerosis (TS).

28. The method of claim 20, wherein (i) when the anti-cancer therapy is a mTOR inhibitor and is everolimus, the gene signature comprises AKT2, TSC1, FKB-12, TSC2, RPTOR, RHEB, PIK3CA and PIK3CB; (ii) when the anti-cancer therapy is a VEGFR inhibitor and is axitinib, the gene signature comprises KIT and KITLG; (iii) when the anti-cancer therapy is a pan-HER inhibitor and is afatinib, the gene signature comprises NRG4 and NRG2; (iv) when the anti-cancer therapy is a MEK inhibitor and is trametinib, the gene signature comprises ERK 2, ARAF, CRAF, MEK1, MEK2, HRAS, ERK1, MAP10 and KSR1; and (v) when the anti-cancer therapy is a PD-1/PD-L1 inhibitor and is an anti-PD-1 or anti-PD-L1 antibody, the gene signature comprises TLR4, PDL2, PDL1, CD16, CTLA4 and CD28.

29. The method of claim 19, wherein the Tyrosine kinase inhibitor is a mTOR inhibitor, a VEGFR inhibitor, a MEK inhibitor, or a pan-HER inhibitor and is selected from the group consisting of everolimus, axitinib, trametinib and afatinib; or a FGFR inhibitor and is BGJ398 and TAS-120; or the PD-1/PD-L1 inhibitor is an anti-PD-1 or anti-PD-L1 antibody.

30. The method of claim 29, wherein the anti-PD-1 or anti-PD-L1 antibody is selected from the group consisting of Pembrolizumab, Nivolumab and Atezolizumab.

Description

BRIEF DESCRIPTION OF THE DRAWING

[0379] FIG. 1: DDPP tumor/normal intensity plots profiles of differential gene expression of the everolimus key genes for the 6 patients treated with everolimus in monotherapy. Clinical outcome is described in Table 1. A: ID 090, head and neck carcinoma, PFS=1.3 months (progressive disease), 3.sup.rd therapy line. B: ID 227, liposarcoma, PFS=1.7 months (progressive disease) in 5.sup.th therapy line. C: ID 117, head and neck carcinoma, PFS=1.9 months (progressive disease), 3.sup.rd therapy line. D: ID 006, unknown primary origin, PFS=8.1 months (stable disease) in 2.sup.nd therapy line. E: ID 148, neuroendocrine tumor of small gut PFS=11.6 months (stable disease) in 3.sup.rd therapy line. F: ID 203, neuroendocrine tumor of small intestine, PFS=60+ months (partial response disease) in 2.sup.nd therapy line. Y axis: intensity of the expression in tumors. X axis: intensity of the expression in normal matched tissue. Intensities are measured as relative fluorescence unit (RFU) signal as assessed with Agilent microarray technology. Overexpression for a given mRNA in the tumor as compared to the normal is denoted with a square (.square-solid.), underexpression is denoted with a triangle (.Math.) and no change is denoted with a black point (•).

[0380] FIG. 2: DDPP correlation with PFS for patients treated with everolimus. A: Pearson correlation plots of the 8 gene predictor: AKT2, TSC1, FKB-12, TSC2, RPTOR, RHEB, PIK3CA and PIK3CB with the PFS of 6 patients treated with everolimus as monotherapy: One patient out of these 6 had censored PFS (ID 203). Y axis: absolute fold value of log 2 based fold-changes tumor versus normal multiplied by log 1.1 based of the intensities in tumor values for each of the 8 genes selected; X axis: PFS under treatment with everolimus as monotherapy in months. B: Pearson correlation plots of the 8 gene predictor: AKT2, TSC1, FKB-12, TSC2, RPTOR, RHEB, PIK3CA and PIK3CB with the PFS of 6 patients treated with everolimus as monotherapy when only the tumor biopsy is investigated: Y axis: absolute fold value of log 1.1 based of the intensities in tumor values for each of the 8 genes selected; X axis: PFS under treatment with everolimus as monotherapy in months. C: Shuffle experiment: Pearson correlation plots of the 8 gene specific predictor for everolimus (AKT2, TSC1, FKB-12, TSC2, RPTOR, RHEB, PIK3CA and PIK3CB) with the PFS of 5 patients treated axitinib as monotherapy (Table 2); Y axis: absolute fold value of log 2 based fold-changes tumor versus normal multiplied by log 1.1 based of the intensities in normal values for each of the 8 genes selected, X axis: PFS under treatment with axitinib as monotherapy in months. D: Leave-one-out experiments: Each reiteration generates a predictor used to calculate PFS of the patient discarded. E: Pearson correlation between leave-one-out predicted PFS and the observed PFS. Y axis: Predicted PFS as defined by leave-one-out (in months), X axis: PFS under treatment with everolimus as monotherapy in months observed in WINTHER trial.

[0381] FIG. 3: DDPP intensity plots and correlation with PFS for patients treated with axitinib. Examples of DDPP profiles for two patients treated with axitinib in monotherapy, with different outcomes. A: ID 088, NSCLC, PFS=2.9 months in 2.sup.nd therapy line. B: ID 083, HN, PFS=8.8 months in 5.sup.th therapy line. Y axis: intensity of the expression in tumors; X axis: intensity of the expression in normal matched tissue. Overexpression for a given mRNA in the tumor as compared to the normal is denoted with a square (.square-solid.), underexpression is denoted with a triangle (.Math.) and no change is denoted with a black point (•). C: Plots representing the Pearson correlation of the 2 gene-predictor (KIT, KITLG) with the PFS of the 5 patients treated with axitinib as monotherapy (one patient NSCLC and four patients HN). Y axis: sum value of log 2 based fold changes tumor versus normal multiplied by the log.1.1 based intensity values of the normal for each of the 2 genes selected; X axis: PFS under treatment in months.

[0382] FIG. 4: DDPP profiles and correlations with PFS for patients treated with trametinib, afatinib and experimental FGFR inhibitors. Examples of DDPP profiles for two patients treated in monotherapy with trametinib (A, B, C), afatinib (D, E, F) and FGFR inhibitors (G, H, 1), with different outcomes. Overexpression for a given mRNA in the tumor as compared to the normal is denoted with a square (.square-solid.), underexpression is denoted with a triangle (.Math.) and no change is denoted with a black point (•). A; ID 118, NSCLC, PFS=3.1 months in 4.sup.th therapy line; B; ID 149, Rectum carcinoma, PFS=7.4 months in 6.sup.th therapy line. D: ID 136, NSCLC, PFS=0.4 months in 4.sup.th therapy line; E: ID 156, NSCLC, PFS=14.3 months in 3.sup.rd therapy line G: ID 228, Colon carcinoma, PFS=0.7 months in 6.sup.th therapy line; H: ID 237, Head and Neck carcinoma, PFS=19.3 months in 7.sup.th therapy line. Y axis: intensity of the expression in tumors; X axis: intensity of the expression in normal matched tissue. C: Plot representing the Pearson correlation of the 9 gene-predictor ERK2, ARAF, CRAF, MEK1, MEK2, HRAS, ERK1, MAPK10 and KSR1) with the PFS of the 3 patients treated with trametinib as monotherapy Y axis: fold value of log 2 based fold changes tumor versus normal multiplied by the log.1.1 based intensity values of the tumor for each of the 9 genes selected X axis: PFS under treatment in months. F: Plot representing the Pearson correlation of the 2 gene-predictor NRG4 and NRG2, with the PFS of the 3 patients treated with afatinib; Y axis: the sum of log 2(Fold change tumor versus normal) multiplied by log 1.1 (Intensity_Tumor) of each values for each of the 2 genes; X axis=PFS in months; 1′ Plot representing the Pearson correlation of the 5 gene-predictor FGF10, FGF16, FGF5, FGF2 and FGF13); Y axis: the sum of log 2(Fold change tumor versus normal) multiplied by log 1.1 (Intensity_Normal) of each values for each of the 5 genes; X axis=PFS in months.

[0383] FIG. 5: DDPP intensity plots and correlation with PFS for patients treated with anti-PD1 therapies. A, B, C: DDPP profiles of 3 patients treated with anti PDL1/PD1 antibodies, with different PFS under treatment. Data source WINTHER trial. A: colon cancer: PFS=0.9 months under treatment with atezolizumab, in 4.sup.th therapy line; B: head and neck cancer: PFS=1.7 months under treatment with nivolumab in 2.sup.nd therapy line; C: colon cancer, PFS: 61+ months under treatment with pembrolizumab in 3.sup.rd therapy line; Y axis: intensity of the expression in tumors; X axis: intensity of the expression in normal matched tissue. D: Pearson correlation between a vectorial summation of 6 genes: TLR4, PDL2, PDL1, CD16 (specific marker of NK), CTLA4 and CD28; Y axis: fold value of Log 2(Fold Change tumor vs. normal) multiplied by intensity in normal values for each of the 6 genes selected; Y axis: PFS in months. E: Pearson correlation between the vectorial summation of the 6 genes specific anti-PD1: TLR4, PDL2, PDL1, CD16 (specific marker of NK), CTLA4 and CD28 and the PFS of 3 patients treated with 10 when only tumor biopsy information is investigated; Y axis: fold value of Log 2 (intensity in tumor) values for each of the 6 genes selected; Y axis: PFS in months. F: Shuffle experiment: Pearson correlation between the vectorial summation of the 6 genes specific anti-PD1: TLR4, PDL2, PDL1, CD16 (specific marker of NK), CTLA4 and CD28 and the PFS of 3 patients treated with afatinib; Y axis: fold value of Log 2(Fold Change tumor vs. normal) multiplied by log 1.1 (intensity in normal) values for each of the 6 genes selected; Y axis: PFS of 3 patients under afatinib treatment (months).

[0384] FIG. 6: Definition of the optimal gene signature for PD-1/PD-L1 therapies based on the DDPP correlation with PFS for patients treated with anti-PD1 therapies. Each graph represents the Pearson correlation between the vectorial summation of the genes mentioned at the top of each graph.

EXAMPLES

[0385] The DDPP is a combinatorial biomarker based on transcriptomics of tumor versus normal tissue used for simultaneous assessment of the steady-state level of many key genes/mRNAs in tumor vs. organ-matched normal tissue in order to assess interactions between genes and pathways that govern sensitivity or resistance to treatments with targeted TKI. The same global method applies for each drug investigated, and starts with the selection, based on literature and the Food and Drug Administration US prescribing information (FDA USPI) of target(s) of the drug and of other key genes involved in modulating tumor sensitivity or resistance to the drug. The technology was developed using the transcriptomic database obtained from the international WINTHER trial. This database included tumor compared to normal RNA analysis of whole transcriptome in patients who were treated in the study.

Materials and Method

[0386] The full methodology of transcriptomic assessment and patient treatment is described in the WINTHER trial published in Nature Medicine. A total of 101 patients out of the 107 treated in WINTHER were available for data analysis herein; six patients could not be analyzed because transcriptomics data were not available. Detailed clinical and biological information for each patient is available in Table 1; further biological data is available on www.winconsortium.org containing: (i) tumor mutations data in XML format; and (ii) expression data in a table format (providing information about tumor/normal fold change and tumor intensity alone for all the cases for which mRNA was analyzed).

[0387] Application of Euclidian Hyperspace Mathematical Model to Precision Oncology

[0388] The fundamentals concepts behind the DDPP methodology: [0389] 1. A straight-line segment can be drawn joining any two points (Euclid's first postulate). [0390] 2. If a third (or fourth, etc.) points are aligned on the same line, they are linked by a function of linearity. [0391] 3. The most adequate way to identify a function of linearity is the linear regression through the Pearson correlation between two types of variables (in this case the differential expression tumor versus normal of key genes governing the mechanism of action of a specific drug, and the PFS of the patients treated with the specific drug). [0392] 4. Definition of transcriptomic variable was performed with a ‘step-in’ method. [0393] 5. Among existing methods, the most adequate appeared the application of Euclidian hyperspace mathematical model that integrates a structure of linearity. Multiple Linear regression and Cox regression methods were tested but could not be adapted for the purpose of the study

[0394] The Euclidian coordinate hyperspace Rn forms an n-dimensional vector space over the field of real numbers with the addition of the structure of linearity, and is often still denoted Rn. The aim of DDPP was to determinate the optimal number of ‘n’ coordinates. The operations on Rn are typically defined by a vector space (also called a linear space); The vector space is a collection of objects called vectors, which may be added together and multiplied (“scaled”) by numbers, called scalars. Scalars are often taken to be real numbers.

[0395] DDPP adapted the model to precision oncology, defining the biologic hyperspace as being the dynamics of the networks regulating normal biological systems and their disturbances in cancer, and the specific mechanisms of actions of drugs investigated. The coordinates needed to define correlate with the clinical outcome under treatment, are the key genes governing drug's mechanism of action. The vector space is constituted by the fold changes between tumor and analogous organ matched normal tissue of the same patient, and the scalars are the intensities of expression in tumor and normal tissues. The use of scalars is mandatory, as the same fold change can be obtained at different levels of intensities (reflecting the steady state-levels of mRNA of each specific key gene, in tumor and normal tissues).

[0396] The DDPP Methodology that Apply to any Type of Drugs is Based on the Following Steps: [0397] 1. Identification of key genes involved in drug's mechanisms of action, based on recent literature and based on the FDA US Prescribing Information (USPI) [0398] a. Everolimus: key genes: PIK3CA, PIK3CB, AKT1, MTOR, FKBP1A, RPS6KB1, EIF4EBP1, HIF1A, TSC1, TSC2, AKT2, RPTOR, PTEN, RHEB, MLST8, RICTOR, VEGFA [0399] b. Axitinib: key genes: VEGFA, VEGFB, VEGFC, PDGFA, PDGFB, FLT1(VEGFR1), KDR (VEGFR2), FLT4 (VEGFR3), PDGFRA, PDGFRB, KIT, KITLG, FIGF [0400] c. Trametinib: key genes: MEK1 (MAP2K1), MEK2 (MAP2K2), ARAF, BRAF, RAF1, ERK1 (MAPK3), ERK2 (MAPK1), MAPK10, KRAS, HRAS, NRAS, KSR1, RAP1A [0401] d. Afatinib: key genes: EGFR, ERBB2, ERBB3, ERBB4 and their ligands EGF, TGFA, AREG, EREG, HBEGF, BTC, NRG1, NRG2, NRG4 [0402] e. FGFR inhibitors: FGFR1, FGFR2, FGFR3, FGFR4 and the FGF ligands 1, 2, 3, 4, 5 etc. [0403] f. Anti-PD1/PDL1: key genes: PDL1, PDL2, PD1, CTLA4, CD28, CD80, CD86, LAG3, TLR4, together with specific markers of the presence of effector tumor infiltrating immune cells: CD8A (cytotoxic lymphocytes T), CD16 (Natural Killer cells) and FOXP3 (T-regs cells) [0404] 2. Selection of the patients with available transcriptomics data and clinical outcome (PFS) under treatment with each drug available. Minimum three patients are required. Everolimus (N=6); axitinib (N=5); [0405] a. Preferably, patients with non-censored PFS were selected for investigations [0406] b. Only two patients had censored PFS under treatment (ID 203 treated with everolimus and ID 183 treated with pembrolizumab). They had exceptional PFS>60 months, continuing the treatment). The inventors did not discard them from analysis, but considered de-censored. [0407] 3. “Step-in” analysis, to define the optimal ‘n’ genes investigated interrogated the correlation between gene expression and the PFS: For each drug, a Pearson correlation test was performed between Fg, the fold change multiplied by the intensity (of tumor and normal) of a single gene g (gene from the list of key genes of the drug) with the PFS for all the patients treated with the drug.

F.sub.g=log 2(fch tumor vs.normal)*log 1.1(tumor) or F.sub.g=log 2(fch tumor vs.normal)*log 1.1(normal)

[0408] The F.sub.g with the most significant correlated gene was driven to decision whether to continue with fold change multiplied by the intensity of the tumor or fold change multiplied by the intensity of the normal matched tissue. The key genes were then ranked based on the Pearson test's p value such that the gene with the highest correlation between F.sub.g and the PFS was ranked first. Then, the inventors added single genes by the following manner: the 2.sup.nd most ranked gene was added to the 1.sup.st most ranked and the F.sub.g1,g2 of the 2 genes was calculated by 5 different methods: mean, median, sum and fold (both the absolute and non-absolute values) of F.sub.g1 and F.sub.g2. Similarly, the 3.sup.rd most ranked gene was added to the 2 highest ranked genes. The addition of single genes described above was continued until all key genes were added. Then, a Pearson correlation test was performed between the various F.sub.g . . . gn with the PFS of the patients treated with the drug. The results were ranked again by the Pearson test's p value. The number of genes in the set which was the most correlated with the PFS was indicated as the optimal ‘n’ coordinates. In order to assess the likelihood of getting a significant correlator by ‘n’ genes, the inventors run an analysis with 100K random ‘n’ genes and tested how the F.sub.g . . . gn of these genes were correlated with the PFS. Significant results were considered by a threshold of absolute R value of 0.9 or above and p value of 0.05 and below. [0409] 4. Selecting of best correlators with PFS for each drug and computing a linear regression model to transform the best correlator into a predictor for a single drug. [0410] 1. Availability of the data: Population characteristics: Male or female patients above age 18 with advanced cancers that had progressed on standard treatment. Participating principal investigators (PIs) were located at Institute Gustave Roussy (IGR) (France), Centre Leon Berard (France), Vall d'Hebron Institute of Oncology (VHIO) (Spain), the Chaim Sheba Medical Center (CSM) (Israel), Segal Cancer Centre, MCGill University (Canada), University of Texas MD Anderson Cancer Center and University of California San Diego Moores Cancer Center. The protocol was approved at all sites that recruited patients. If the patient was navigated to an investigational clinical trial, the patient signed consent for that trial as well. Detailed clinical and biological information for each patient is available in Table 1. Further biological data available upon request on www.winconsortium.org containing: (i) tumor mutations data in XML format; and (ii) expression data in a table format (providing information about tumor/normal fold change and tumor intensity alone for all the cases for which mRNA was analyzed).

Results

[0411] The differential tumor versus analogous normal tissue expression of these genes was used for elaborating the prototype of the DDPP decision support tool. The inventors explored the fold changes, measuring the differential tumor versus normal gene expression of the key genes selected for each drug, which created different vectors/coordinates and correlated these data with progression-free survival (PFS) in patients treated in WINTHER trial. However, as the same fold change can be obtained with different intensity levels, the inventors explored, the fold changes in tumor versus normal multiplied by the intensity of the expression in tumor or in normal tissues (scalars). Details are provided in the Materials and Methods section.

[0412] The DDPP algorithm generates two types of results: 1) a digital visualization through tumor versus normal tissue expression intensity plots enabling an understanding of the interactions between the key genes and an estimate of their contributive weight; and 2) an outcome predictor generating, for each drug, the vectorial summation of the contributive genes and a regression model for the correlation between differential tumor to normal gene expression and PFS under treatment. The inventors investigated the DDPP profiles of key genes and examined the correlations with PFS for patients who received monotherapy with everolimus (n=6) and axitinib (n=5) for whom transcriptomic and PFS data were available. Similar work was performed for patients treated with other therapies in the WINTHER trial such as trametinib (MEK inhibitor); afatinib (pan-HER inhibitor), two experimental FGFR inhibitors (BGJ398 and TAS-120) with a similar mechanism of action, as well as for patients treated with anti-PD1/PDL1 monoclonal antibodies (pembrolizumab, nivolumab and atezolizumab).

[0413] All patients were heavily pretreated prior to receiving the treatment in the WINTHER trial. The main clinical and outcome characteristics of the patients treated are described in Table 1, together with next generation sequencing Foundation One test (Foundation Medicine) performed during the WINTHER study. One patient treated with everolimus (ID 203) and one patient treated with pembrolizumab (ID 183) had exceptional responses lasting in excess of 60 months and PFS was therefore censored at this timepoint. Both patients were included in the analyses.

[0414] DDPP Investigations of Patients Treated with Everolimus

[0415] Table 2 describes the currently recognized 17 key genes of the mTOR pathway (O'Reilly T. et al. Biomarker. Translational Oncology (2010) 3, 65-79, FDA-USPI everolimus: https://www.accessdata.fda.gov/drugsatfda_docs/label/2010/022334s61bl.pdf). Upstream regulators of MTOR: PIK3CA, PIK3CB, AKT1, AKT2, PTEN, TSC1, TSC2, RHEB; FKB-12 (FKBP1A) play a key role as it binds to everolimus and interacts with MTOR resulting in the formation of inhibitory complexes MTORC1 (MTOR, MLST8 and RPTOR) and MTORC2 (MTOR, MLST8 and RICTOR); Downstream effectors are: S6K1 (RPS6KB1), 4EBP1 (EIF4EBP1), HIF1 (HIF1A) and VEGFA.

[0416] It should be noted that the genomic alteration profile could not explain the variation in PFS observed in the everolimus monotherapy group; indeed, the two patients with the longest PFS, both with GI tract neuroendocrine tumors, had no mutations (ID 203) or no genomic alterations in the PI3K/AKT/mTOR pathway (ID 148), respectively. In contrast, the patients with much shorter PFS (ID 227, ID 6, ID 90, and ID 117) did have alterations in the PI3K/AKT/mTOR, albeit accompanied by co-alterations that might have driven resistance. Since DNA biomarkers could not explain variations in clinical outcome, the inventors further investigated whether transcriptomics and DDPP could provide a deeper insight.

[0417] FIG. 1 shows the DDPP profile of the 17 key genes of the 6 patients treated with everolimus as monotherapy. Whilst MTOR has the same steady state level in both tumor and normal tissues, FIGS. 1E and 1F suggest a trend of greater overexpression of upstream regulators of MTORC1 complex (in particular TSC1 and AKT2) in patients with longer PFS (ID 148 and ID203) as compared with the patients with shorter PFS (ID 090, ID 227, ID 117 and ID 006).

[0418] The inventors evaluated the relative contribution of each of the 17 genes, by correlating their differential expression with the PFS in patients treated with everolimus. Pearson correlations between differential gene expression and PFS for each of the 17 genes were: AKT2 (R=0.75, p=0.087; TSC1(R=0.74, p=0.094), FKB-12 (R=−0.67, p=0.149), TSC2 (R=0.63, p=0.178), RPTOR (R=0.61, p=0.198), RHEB (R=−0.49, p=0.325), PIK3CA (R=0.43, p=0.4), PIK3CB (R=−0.41, p=0.414), AKT1 (R=−0.35, p=0.496), MLST8 (R=−0.34, p=0.509), VEGFA (R=0.27, p=0.604), HIF1 (R=0.27, p=0.606), PTEN (R=−0.16, p=0.759), 4EBP1 (R=−0.16, p=0.77), RICTOR (R=0.14, p=0.788), MTOR (R=0.13, p=0.807) and S6K1 (R=−0.04, p=0.936).

[0419] The inventors further explored the combined differential expression in tumor versus normal tissues of the most contributive key genes involved in the everolimus pathway. For each of the correlations with PFS, the inventors built a vectorial summation using a ‘step-in’ method, starting with AKT2 and adding successively a gene in the order of their significance: AKT2-TSC1; AKT2-TSC1-FKB12, then AKT2-TSC1-FKB12-TSC2 and so forth, obtaining in total 17 different vector summations. Each combined vector was correlated with PFS. FIG. 2A shows that the optimal performance was obtained by the vectorial summation of the 8 most contributive genes: AKT2, TSC1, FKB-12, TSC2, RPTOR, RHEB, PIK3CA and PIK3CB, to obtain the most significant correlation with PFS, among the 17 possibilities: (R=0.99, p=5.67E-05). The higher the relative expression of these key genes, the longer the PFS is under treatment with everolimus. The linear regression model for the correlation with PFS is: Y=6588874000820.68X—19753182063978.6, where Y=the absolute value of the fold of log 2(Fold change tumor versus normal) multiplied by log 1.1 (Intensity_Tumor) of each value for each of the 8 genes, and X=PFS in months. FIG. 2B shows that when only the tumor biopsy was investigated (as usually done in the current biomarker studies in oncology practice or translational research), the significance of the correlation dropped to R=0.24, p=0.65), suggesting the importance of the new strategy of tumor versus normal analogous organ matched dual biopsy.

[0420] In order to assess the prognostic versus the predictive value of the DDPP data in these analyses, the inventors tested the specific predictor of the PFS for everolimus (n=6 patients) generated by 8 genes (AKT2, TSC1, FKB-12, TSC2, RPTOR, RHEB, PIK3CA and PIK3CB) and cross correlated their combined differential expression with the PFS of patients under axitinib treatment (n=5, Table 1). FIG. 2C demonstrates that the subset of genes identified for everolimus was highly specific to this everolimus regimen, as the cross correlation with PFS of patients treated with axitinib was not significant (R=0.29, p=0.637) suggesting that the DDPP findings are consistent with the known biology of the MTOR pathway and the mechanism of action of everolimus. Whether the DDPP biomarkers have predictive value to guide treatment selection must be confirmed in an independent cohort of patients.

[0421] To further explore the potential predictive value of DDPP, the inventors performed a full shuffle ‘step-in’ analysis and cross-correlated the differential expression of all of the 17 genes specific for everolimus with the PFS of the 5 patients treated with axitinib. None of the 17 genes correlated significantly with the PFS under axitinib: RICTOR (R=0.8, p=0.106); HIF1 (R=−0.77, p=0.131); RPTOR (R=−0.73, p=0.159); 4EBP1 (R=−0.73, p=0.163); S6K1 (R=−0.65, p=0.237) etc. Exploring the combined differential expression of these genes, there was no significant correlation in any of the 17 possible combinations: RICTOR-HIF1 (R=−0.27, p=0.656); RICTOR-HIF1-RPTOR (R=0.74, p=0.905); RICTOR-HIF1-RPTOR-4EBP1 (R=0.02, p=0.969); RICTOR-HIF1-RPTOR-4EBP1-S6K1 (R=0.37, p=0.54) etc. These data support the hypothesis that DDPP predictors for everolimus are specific for that therapeutic regimen.

[0422] In order to assess the robustness of DDPP method and to determine whether the model is over-fitting the correlations, the inventors performed both random selections of 8 genes (number corresponding to the optimal number of genes of the specific everolimus predictor) and of 17 genes (corresponding to the full set of key genes involved in everolimus mechanism of action) across the whole transcriptome (around 22,000 genes) and correlated their vectorial summation with PFS of the 6 patients who received everolimus monotherapy treatment. This analysis was repeated 100,000 times, randomly selecting a different set of 8 genes at each reiteration. Setting the threshold of significance at R≥=0.9 and p≤0.05, the percentage of random significant correlations with PFS was 16.587%. Setting the threshold of significance at the same value as the one observed for the predictor (R=0.99, p=5.67E-05), the percentage of random significant correlations with PFS was 1.018%. When the random selection involved 17 genes at each re-iteration (repeated 100,000 times), the percentage of the random significant correlations at the two different thresholds were 23.59% and 0.851% respectively.

[0423] The observation that randomly selected sets of genes generated significant correlations with PFS suggests that with only 6 patients in the cohort, a certain degree of overfitting of the correlations cannot be excluded in these analyses. The specificity of correlations could be increased only with a larger number of patients used as training and test datasets. Nevertheless, the biological understanding of the MTOR pathway and the effects of everolimus are consistent with the DDPP findings. Indeed, the most contributive genes, AKT2, TSC1, FKB-12, TSC2, RPTOR, RHEB are key for direct interaction with MTOR and its upstream regulation (TSC1, TSC2, RHEB). Furthermore, FKB-12 binds everolimus and associates to MTOR forming together with RPTOR the MTORC1 complex;

[0424] Leave-one-out experiments: To interrogate whether the findings of these analyses could be used as predictors, the inventors performed leave-one-out analyses, reiterating 6 combinatorial analyses. At each investigation, one patient was discarded, and a correlator/predictor was identified based on the remaining 5 patients applying the same methodology. The correlator was then used as a predictor to predict the PFS of the patient left out. FIG. 2D depicts the 6 correlations obtained in the re-iterative experiments, and FIG. 2E shows the correlation between the observed PFS and the leave-one-out predicted PFS. At each re-iteration, the predictor is generated by different subsets of genes, suggesting a high impact of the composition of the cohort when small number of patients investigated. This observation pinpoints the need to validate the DDPP on a larger cohort of patients. Nevertheless, the concordance between real PFS of the patients left out, and the predicted PFS using the correlator obtained at each reiteration on the remaining 5 patients is significant, with R=0.96, p=0.0022. These preliminary observations suggest the realistic possibility of obtaining a stable predictor, using a higher number of patients, to obtain a validated tool that may be useful to accurately estimate the PFS in a prospective clinical setting.

[0425] DDPP Investigations for Patients Treated with Axitinib

[0426] At nanomolar concentrations, axitinib specifically inhibits VEGFR1, VEGFR2 and VEGFR3. Thirteen key genes involved in the control of the angiogenesis were selected and investigated with DDPP methodology: FLT1 (VEGFR1), KDR (VEGFR2), FLT4 (VEGFR3) and their ligands VEGFA, VEGFB, VEGFC and FIGF, PDGFRA, PDGFRB, PDGFA, PDGFB, KIT and KITLG (FDA-USPI axitinib: https://www.accessdata.fda.gov/drugsatfda_docs/label/2012/2023241bl.pdf). Four patients had head and neck carcinoma, and one patient had a lung adenocarcinoma. Table 1 shows the different PFS under treatment with axitinib.

[0427] The differential tumor versus normal expression of KIT and of its ligand KITLG was identified as being the major driver of the correlation with the PFS of the patients treated with axitinib FIG. 3A, 3B), and their combined vector: KIT-KITLG generated the optimal performance in the correlation with PFS: R=0.99, p=4.68E-04 (FIG. 3C). The linear regression model is Y=49.13X-213.31, where Y=the sum of log 2(Fold change tumor versus normal) multiplied by log 1.1 (Intensity_Normal) of each value for each of the 2 genes, and X=PFS in months.

[0428] To examine the possibility of overfitting the correlations the inventors performed both random selections of 2 genes (number corresponding to the optimal number of genes of the specific optimal axitinib predictor) and of the 13 genes (corresponding to the full set of key genes involved in axitinib mechanism of action) across the whole transcriptome. Random selections of 2 genes across the whole transcriptome and correlation of their vectorial summation with PFS of the 5 patients treated with axitinib, repeated 100,000 times, show that the percentage of random significant correlations with PFS, at the threshold abs R≥0.9 and p≤0.05 is 5.957%. Using the same threshold as the specific predictor (R=0.99, p=4.68E-04) the percentage of significant correlations was 0.059%. Random selection of 13 genes (number corresponding to the full set of key genes) and correlation with PFS under axitinib, repeated 100,000 times, showed that the percentage of random significant correlations (at the same two thresholds) were 5.671% and 0.061% respectively.

[0429] Leave one out experiments: The inventors performed (using the same ‘step-in’ vectorial summation methodology) 5 leave-one-out re-iterations, discarding at each experiment one patient and building a predictor on the remaining 4. The inventors observed again an instability of the predictors and dependence on the compositions of the cohorts at each re-iteration. The concordance between real PFS of the patients left out, and the predicted PFS using the correlator obtained at each reiteration was lower than for the everolimus example (R=−0.81, p=0.1) likely related to a lower number of patients in each re-iteration. These data suggest again that performance and accuracy of the prediction of the PFS could be increased only with a higher number of patients in the training and validation datasets.

[0430] DDPP and Other Examples of TKI

[0431] Trametinib—Thirteen key genes were investigated: MEK1 (MAP2K1), MEK2 (MAP2K2), ARAF, BRAF, RAF1, ERK1 (MAPK3), ERK2 (MAPK1), MAPK10, KRAS, HRAS, NRAS, KSR1, RAP1A. The combined differential tumor versus normal tissue expression of 9 genes and their vectorial summation correlated with the PFS of 3 patients treated with trametinib as monotherapy Table 1 and FIG. 4A, 4B, 4C): ERK2, ARAF, CRAF, MEK1, MEK2, HRAS, ERK1, MAPK10 and KSR1 (R=−0.99, p=0.026); The linear regression model is Y=−145283210513695X+1125665952686062, where Y=the fold of log 2(Fold change tumor versus normal) multiplied by log 1.1 (Intensity_Tumor) of each values for each of the 9 genes, and X=PFS in months.

[0432] Afatinib—Thirteen key genes were investigated: EGFR, ERBB2, ERBB3, ERBB4 and their ligands EGF, TGFA, AREG, EREG, HBEGF, BTC, NRG1, NRG2, NRG4. The combined differential tumor versus normal tissue expression of 2 genes and their vectorial summation correlated with the PFS of 3 patients treated with afatinib as monotherapy (Table 1 and FIG. 4D, 4E, 4F): NRG4 and NRG2 (R=−1 p=1.83E-04). The linear regression model is Y=−21.7724X+55.93, where Y=the sum of log 2(Fold change tumor versus normal) multiplied by log 1.1 (Intensity_Tumor) of each value for each of the 2 genes, and X=PFS in months.

[0433] FGFR inhibitors—Nineteen key genes investigated: FGFR1, FGFR2, FGFR3, FGFR4 and the FGF ligands 1, 2, 3, 4, 5 etc.). The differential expression and vectorial summation of 5 genes correlated with the PFS of 3 patients treated with FGFR inhibitors BGJ398 or TAS-120 as monotherapy (Table 1 and FIG. 4G, 4H, 4I): FGF10, FGF16, FGF5, FGF2 and FGF13 (R=−1, p=1.6E-04). The linear regression model is Y=−19.011X+98.36, where Y=the sum of log 2(Fold change tumor versus normal) multiplied by log 1.1 (Intensity_Normal) of each value for each of the 5 genes, and X=PFS in months.

[0434] DDPP and Prediction of Outcome after IO Treatment:

[0435] Although the most advanced knowledge has been generated around the therapies targeting PD1/PDL1 or CTLA-4, there are multiple other important pathways that may impact the immune response to cancer involving many genes, in particular LAG3, TLR4, VISTA, TIM3, TIGIT, ICOS, OX40, GITR, TIM3 (Spencer C. et al. Cancer Discovery, 2018 |1069). Among them LAG3 and TLR4 may have a particular importance, as described in Table 3. Given the current status of knowledge, the 10-specific DDPP gene-set focuses on: PDL1, PDL2, PD1, CTLA4, CD28, CD80, CD86, LAG3, TLR4, together with specific markers of the presence of effector tumor infiltrating immune cells: CD8A (cytotoxic lymphocytes T), CD16 (Natural Killer cells) and FOXP3 (T-regs cells). Many types of immune cells are involved in the activation and regulation of the immune system attack against tumor cells (APC, LyT CD4+ etc.) but the inventors focused on specific markers for infiltrating LyTc, NK and Tregs that have the ability to recognize directly the tumor cells' neoantigens coupled with major histocompatibility complex 1 (CMH1) and are directly targeting tumor cells.

[0436] A correlation analysis between differential gene expression of the selected genes and the PFS was performed for the three patients treated with anti-PD1 antibodies (Table 1) in the WINTHER trial. The example provided in FIGS. 5A, 5B and 5C shows the DDPP intensity plots of 3 patients treated with IO pembrolizumab (anti-PD1), atezolizumab (anti-PDL1) and nivolumab (anti-PD1); two patients had colon cancer (CRC) and one head and neck (HN) cancer. Their clinical outcomes were: ID 183 (CRC) PFS 61+ months (the patient is in complete clinical remission and no longer receiving pembrolizumab); ID 294 (HN) PFS 1.7 months and ID 270 (CRC) PFS 0.9 months. DDPP profiling identified that the higher level of TLR4 fold change in tumor versus normal tissues, the shorter the survival under treatment with anti-PD1. Considering that all three treatments are directed to leverage the PDL1/PD1 negative blockade, and considering that the mechanisms may be agnostic of tumor type, the inventors performed the investigations aiming to explain the variations PFS.

[0437] The relative contribution of each of the key 12 genes was evaluated by correlating their differential expression with the PFS in patients treated with IO. Pearson correlations between differential gene expression and PFS for each of the 17 genes were: TLR4 (R=−0.99, p=0.103; PDL2 (R=0.97, p=0.143), PDL1 (R=0.90, p=0.294), CD16(NK) (R=0.77, p=0.445), CTLA4 (R=0.60, p=0.588), CD28 (R=−0.50, p=0.665), CD80 (R=−0.49, p=0.67), CD86 (R=−0.42, p=0.721), LAG3 (R=0.34, p=0.776), CD8A (LyTCD8+) (R=−0.30, p=0.803), FOXOP3 (Tregs) (R=−0.21, p=0.862) and PD1 (R=−0.18, p=0.882).

[0438] The inventors further explored the combined differential expression in tumor versus normal tissues of the most contributive key genes involved in the IO pathway. For each of the correlations with PFS, the inventors built a vectorial summation using a ‘step-in’ method, starting with TLR4 and adding successively a gene in the order of their significance: TLR4-PDL2, TLR4-PDL2-PDL1, then TLR4-PDL2-PDL1-CD16 and so forth, obtaining in total 12 different vector summations. Each combined vector was correlated with PFS. FIG. 5D shows that the optimal performance was obtained by the summation of the 6 most contributive genes: TLR4, PDL2, PDL1, CD16(NK), CTLA4 and CD28, to obtain the most significant correlation with PFS, among the 12 possibilities: (R=1, p=8.15E-04). The linear regression model for the correlation with PFS is: Y=1272862482.4864X−2015211477.55, where Y=the value of the fold of log 2(Fold change tumor versus normal) multiplied by log 1.1 (Intensity_Normal) of each value for each of the 6 genes, and X=PFS in months. These data suggest that the main confounders explaining differences in PFS under anti-PD1 therapy are the degree of activation of TLR4, and the balance between PDL1, PDL2 and CTLA4 activation of the negative immune-blockade. FIG. 5E shows that when only the tumor biopsy was investigated (as usually in the current biomarker studies in oncology practice or translational research), the significance of the correlation dropped to R=−0.45, p=0.704, suggesting the importance of the new strategy of tumor versus normal analogous organ matched dual biopsy.

[0439] In order to assess the prognostic versus the predictive value of the DDPP data for 10 in these analyses, the inventors tested the specific predictor of the PFS (with the 6 genes (TLR4, PDL2, PDL1, CD16(NK), CTLA4 and CD28)) for anti-PD1 treatments (n=3 patients) and cross correlated their combined differential expression with the PFS of patients under afatinib treatment (n=3, Table 1). FIG. 5F demonstrates that the subset of genes identified for anti-PD1 did not correlate with the PFS of the patients treated with afatinib (R=0.61, p=0.579) suggesting high specificity for IO regimen.

[0440] Robustness of DDPP method was tested through random selections of 6 genes (number corresponding to the optimal number of genes of the specific anti-PD1 predictor) across the whole transcriptome (around 22,000 genes) and correlated their vectorial summation with PFS of the 3 patients who anti-PD1 treatment. The analysis was repeated 100,000 times, randomly selecting a different set of 6 genes at each reiteration. Setting the threshold of significance at the same value as the one observed for the predictor (R=1, p=8.15E-04), the percentage of random significant correlations with PFS was 0.356%.

[0441] Based on the 6 genes identified, the inventors assessed ‘in silico’ the predicted PFS of the 82 patients (for whom no information was missing), agnostic of tumor type and independent of the number of prior lines of therapy, if they were treated with anti-PD1 therapies. For 57 patients (59.5%) the predicted PFS under anti PD1 treatment was ≤6 months (with a majority less than 3 months); for 25 patients (30.5%) the predicted PFS under anti-PD1 treatment was ≥6 months (of which 16 (19.5%) with PFS>24 months). These data are concordant with clinical trial data that IO benefits around 20% of patients for a prolonged period of time.

DISCUSSION

[0442] The actual biomarkers used in current translational research and clinical practice illustrate a paradox: On one hand, all models require large cohorts of patients for their validation, but often lack precision when applied to an individual patient, because of the complex portfolio of confounders found in the individual tumors. On the other hand, physicians are compelled to offer personalized therapies to unique individuals, without a systematic accurate system for treatment selection, using by default what is available, mainly companion diagnostics (https://www.fda.gov/medical-devices/vitro-diagnostics/list-cleared-or-approved-companion-diagnostic-devices-vitro-and-imaging-tools). This paradox reveals that the current therapeutic approach, using one-dimensional biologic coordinates (e.g., companion diagnostic provided by specific DNA aberrations, tumor mutation burden (TMB) or PDL1 status, microsatellite instability status) to select therapies, that predict potential responders versus non-responders (binary categories) is inadequate.

[0443] To address this paradox, the inventors shifted the paradigm of statistical analysis placing the patient's tumor in a multi-dimensional space instead of using one-dimensional coordinates. Such a shift was possible by exploring transcriptomics beyond DNA sequencing, and by adapting the Euclidian hyperspace mathematical model (Solomentsev, E. D. (2001) [1994], “Euclidean space”, in Hazewinkel, Michiel (ed.), Encyclopedia of Mathematics). Applied to oncology, the hyperspace refers to the biology of each patient's tumor and analogous organ matched normal tissue. The multi-dimensional coordinates (key genes) that may define a patient's clinical outcome in the biological space R.sup.n, forms an n-dimensional vector space over the field of real numbers with the addition of the structure of linearity. The PFS transcriptomics vector space associated with specific drugs comprised n-vectors (each vector in DDPP is defined by fold changes of the differential specific gene expression between tumor and normal for each specific gene). Vector interactions were obtained by their summation. Determination of the subset of coordinates that best correlates with PFS was performed by a step-in combinatorial investigation. To increase accuracy, the vectors have been multiplied/“scaled” by the intensities/steady state levels of transcripts of each specific gene (in DDPP). These data led to the selection of an optimal number of key genes/transcriptomic variables that correlate with PFS observed with treatment with a specific drug (everolimus, axitinib, trametinib, afatinib, FGFR, and anti-PD1/PDL1). It should be noted that other existent analytical methods such as Multiple Linear Regression (MLR) or COX Regression were tested but could not be adapted for the purpose of the study.

[0444] This methodology based on vectorial summation of the differential expression of the most contributive genes, differentiates DDPP from other methodologies that use one-dimensional biologic coordinates (e.g., specific DNA aberrations, TMB, PDL1 expression, microsatellite instability status). Another unique feature of DDPP that contrasts with the companion diagnostic test concept is that the predictors are non-binary, providing only categories of patients who will potentially benefit or not from specific therapies, but continuous, aiming to estimate the duration of the PFS.

[0445] The investigation of tumor and analogous organ-matched normal tissue biopsies from the same patient is of crucial importance for accurate interpretation of the transcriptomic data as it discards the transcriptomic genetic variability background noise in each patient, and lowers significantly the variance of transcriptomic measurements (Koscielny, S. Sci. Transl. Med. 2, 14ps2 (2010)). Today there are very few clinical applications of biomarkers based on transcriptomics (all focused on investigating only tumor biopsies), such as Oncotype and Mamaprint (loannidis, J. P. PLoS Medicine. 2, e124 (2005)), but they are not used for the purpose of predicting the PFS of patients receiving specific targeted therapies. The inventors were able to use such transcriptomic and PFS data from the WINTHER trial database. The WINTHER trial remains the only clinical trial that used transcriptomics in a prospective clinical setting in addition to conventional DNA sequencing to help inform the treatment decision for patients with advanced cancer. WINTHER is also the first and only trial that used the dual biopsy strategy, investigating both tumor and analogous normal tissue from the same patient, across a variety of solid tumors.

[0446] The inventors explored the DDPP to assess correlations with PFS associated with the drugs everolimus, axitinib, trametinib, afatinib, experimental FGFR inhibitors and anti-PD1/PDL1 therapies observed in the WINTHER trial. Remarkably, for all drugs tested, DDPP enabled identification of significant correlations between the differential expression of subsets of key genes and the PFS for each drug investigated. Preliminary observations show that the DDPP biomarkers seem to be specific to the therapeutic regimens. It should be noted that the DDPP was agnostic of tumor type and independent of the number of prior lines of therapy and could also provide important insight in better understanding the clinical outcomes by identifying the genes with the highest contributing weight driving the correlations.

[0447] Random testing performed for all drugs suggest that DDPP data are not likely to be statistical artefacts. Moreover, these data suggest that the subsets of genes selected, and the correlations obtained by their combined differential tumor versus normal tissues expression (vector summation) with the PFS for each therapeutic regimen may be specific for each drug and have a predictive value rather than a prognostic value, although this requires confirmation in larger studies.

[0448] Taken together, the data suggest the possibility that using a larger number of patients will allow to generate a validated tool that may be useful to estimate with accuracy the PFS in a prospective clinical setting. Indeed, many drugs investigated in this report, have a narrow spectrum of approved clinical uses, given the prevalence of their pathways in tumor growth and spread, and the reason is probably related to the lack of reliable biomarkers to select patients who might have a therapeutic benefit.

[0449] As representative examples, everolimus is approved today for the treatment of advanced renal cell carcinoma (RCC) after failure of treatment with sunitinib or sorafenib, for the treatment of sub-ependymal giant cell astrocytoma (SEGA) associated with tuberous sclerosis (TS) in patients who require therapeutic intervention but are not candidates for curative surgical resection and for the treatment of pancreatic and GI neuroendocrine tumors. Axitinib, (alone or in combination with avelumab), is indicated for the treatment of advanced renal cell carcinoma. In the WINTHER trial the experimental treatment with everolimus, axitinib, afatinib, trametinib and 10 resulted in significant responses in patients with other types of tumors who had been heavily pretreated and had exhausted standard therapeutic options: GI tract neuroendocrine tumors, head and neck adenocarcinomas, colorectal carcinomas and lung non-small cell adenocarcinomas.

[0450] To the inventor knowledge potential biomarkers based on transcriptomics do not exist for the clinical use of everolimus or axitinib. Indeed, DDPP may provide for the first time a methodology and tools that would enable prediction of PFS for any drug (10, or non IO targeted therapeutics) or tumor type and in any therapy line. DDPP predictors could be used (pending further validation) to identify the patients who could have clinical benefit from the treatment with everolimus and axitinib that was not predicted by genomic alterations in the WINTHER trial.

[0451] The DDPP concept and methodology was tested also on other drugs: trametinib, afatinib and two experimental FGFR inhibitors (BGJ398 and TAS-120) with a similar mechanism of action and similar trends obtained in the everolimus and axitinib examples.

[0452] The inventors investigated a cohort (n=3) of patients that received anti-PDL1 therapies. Data suggest that the main confounders explaining differences in PFS under anti PD1 therapy are the degree of activation of TLR4, and the balance between PDL1, PDL2 and CTLA4 activation of the negative immune-blockade, together with the level of infiltration of the tumor by Natural Killer cells. Both DDPP intensity plots and vectorial summation correlative analyses identified TLR4 as the most contributive gene to explain variations in PFS. Observations suggest that the current panel of biomarkers used in clinical practice (tumor mutation burden, microsatellite instability and PDL1 status) could be complemented with other potential biomarkers such as TLR4. Indeed, TLR4 signaling in immune and inflammatory cells of the tumor microenvironment may lead to production of pro-inflammatory cytokines (TNF, IL-1ρ, IL-6, IL-18, etc.), immunosuppressive cytokines (IL-10, TGF-β, etc.) and angiogenic mediators (VEGF, EGF, TGF-β) that influence the immune response to tumor cells. Furthermore, the exploration of the association of antiTLR4 with anti-PDL1 treatments could be of interest with the aim to increase the fraction of patients who could benefit from 10 treatments.

[0453] The methodology envisioned here can be applied in the earliest stages of clinical development, such Phase I clinical trials, as exemplified by investigation of experimental FGFR inhibitors BGJ398 and TAS-120 tested in the clinical trials NCT01004224 and NCT052778 respectively, as part of the navigational WINTHER trial.

[0454] In conclusion, the unique transcriptomic dataset obtained from tumor and organ matched normal tissue biopsies were essential to enable correlations with clinical outcome under treatment with TKI inhibitors and IO. The DDPP is potentially a new global biomarker model that can apply to any type of drug (IO or non IO targeted drugs) alone or in combination, agnostic of tumor type, and can lead, pending further prospective validation, to a new approach to optimal treatment selection for patients with cancer, in particular for those that have exhausted therapeutic and biomarker options.

TABLE-US-00001 TABLE 1 Characteristics of the patients treated in WINTHER trial, investigated with DDPP Study Cancer Prior PFS DNA - List of molecular ID Age Sex site* lines month alterations (FM report)** Drug_given 203 67 F GI/NE 1 60.0+ No mutation Everolimus 148 82 M GI/NE 2 11.6 BCOR N1652fs*34; CDKN1B E126fs*1 Everolimus 6 64 F UP 1 8.1 TSC1 splice site 913 + 1G > T; Everolimus BRCA1 truncation, intron 11; CDKN2A/B loss; DNMT3A R882H; LRP1B loss 117 34 M HN 2 1.9 TSC2 S1431L; TP53 G245S; BCOR Everolimus K374fs*19; SMARCA4 R1135W 227 56 M LS 4 1.7 STK11 F354L; STK11 F354L; TERT Everolimus promoter - 124 C > T 90 74 M HN 2 1.3 PIK3CA Q546R; EP300 D1154fs*30; Everolimus NOTCH1 L1746fs*40 83 59 M HN 4 8.8 MTOR L2209V; ETV6 trunc intron 5; Axitinib CIC S333fs*36; MLL2 G3698 fs*51 223 65 F HN 3 7.1 CCND1 T2861 Axitinib 259 53 F HN 4 6.2 PDGFRA amp Axitinib 25 65 M HN 2 5.3 TP53 I195F; KDM6A L725fs*4; MSH6 Axitinib K1358fs*2; NFE2L2 R18Q 88 56 M Lung 1 2.9 DNMT3A R635P; KRAS G12C; TP53 Axitinib Y220C; MLL2 T1246M 149 54 F CRC 5 7.4 KRAS G12V; ARID1A SPLICE SITE Trametinib 2733-1G > A 100 43 M Lung 2 6.6 BRAF A598_T599insT; IDH1 R132C Trametinib 118 78 F Lung 3 3.1 KRAS G12C; CDKN2A/B loss; TP53 Trametinib V157F, Y220 fs*27; MUTYH G382D 156 71 F Lung 2 14.3 EGFR E746_A750del, T790M; Afatinib + CDKN2A/B loss; CTNNB1 S33F; MYC Cetuximab amplification; SMAD4 P186fs*6; STAG2 splice site 1535-12_1630del108 235 60 F Lung 1 11.3 ERBB2 A775_G776insYVMA Afatinib 136 79 M Lung 3 0.4 ERBB3 amp; MET splice site Afatinib 3028 + 1G > A; STK11 Q100* ATM L2450fs*11; BRCA1 E23fs*17; CDK4 amp; CDKN2A/B loss; MDM2 amp; APC I1307K; KDM5C truncation; MAP3K1 S1475* 237 47 M HN 6 19.3 CCND1 amp; FGFR2 amp; CDKN2A/B NCT01004224 loss; FGF19 amp; FGF4 amp; BAP1 BGJ398 trunc exon 3; FGF3 amp; MAGI2 Q1077*; PBRM1 E1155fs*17 247 67 M Esophagus 2 1.6 FGFR2 amp; CDKN2A/B loss; TP53 NCT02052778 W91*; ASXL1 splice site 472 - TAS-120 2A > G 228 38 M CRC 5 0.7 FGFR1 amp; TP53 C176F; APC E1322*, NCT02052778 R213*; SMAD4 loss; SOX9 V163fs*21 TAS-120 183 66 M CRC 2 61.0+ RBB3; V104M; MAP2K1; E203K; Pembrolizumab CDKN2A/B loss; FBXW7 R465C; (TMB: 74.8) PIK3CA E39K; PIK3R1 R348*, R639*; (MSI: +) PTEN R233*, splice site 801 + 2T > G; TP53 R158H, R273H; APC R1450*, R499*; ARID1A P1115fs*46, Q1306fs*17; ATRX Q2422*; CDH1 D433N; EP300 R2263*; FAM123B R631*; FAT1 A4305V; FLCN H429fs*39; MSH6 L1330fs*12, S279fs*12 294 57 M HN 1 1.7 BRCA2 K3408* Nivolumab (TMB: 0) (MSI: −) 270 76 F CRC 3 0.9 FLT4 amp; FLT3 amp equivocal; Atezolizumab BARD1 C53fs*5; MYC amp; PARK2 loss (TMB: 10.4 exons 3-5; TP53 R175H; APC (MSI: −) T1556fs*3; BCL2L1 amp; CDK8 amp; ETV6 rearrangement intron 5; FAM123B R497*; GATA6 amp equivocal; KDM6A Y215*; MUTYH Y165C; NOTCH1 Q2123* Abbreviations: *GI = gastrointestinal; NE = neuroendocrine; HN = head and neck; UP = Unknown primary; LS = liposarcoma; ID: 203, PFS 60+ and OS60+ are censored values; FM = Foundation Medicine; TMB= tumor mutation burden; MSI = microsatellite instability; amp = amplification; del = deletion; trunc = truncation

TABLE-US-00002 TABLE 2 Everolimus mechanism of action and the key genes for everolimus pathway Everolimus is an inhibitor of mammalian target of rapamycin (mTOR), a serine-threonine kinase, downstream of the PI3K/AKT pathway. The mTOR pathway is dysregulated in several human cancers. Everolimus binds to an intracellular protein, FKBP-12, resulting in formation of an inhibitory complex (mTORC1) and thus inhibition of mTOR kinase activity. Everolimus reduces the activity of S6 ribosomal protein kinase (S6K1) and eukaryotic elongation factor 4E-binding protein (4E-BP1), downstream effectors of mTOR, involved in protein synthesis. In addition, everolimus inhibits the expression of hypoxia-inducible factor (e.g., HIF-1) and reduces the expression of vascular endothelial growth factor (VEGF). Inhibition of mTOR by everolimus has been shown to reduce cell proliferation, angiogenesis, and glucose uptake. Gene Symbol Role Phosphatidyl- PIK3CA Generates phosphatidylinositol 3,4,5-trisphosphate (PIP3). Involved in inositol- the activation of AKT1 upon stimulation by receptor tyrosine kinases Bisphosphate ligands such as EGF, insulin, IGF1, VEGFA and PDGF. Essential in Kinase Catalytic endothelial cell migration during vascular development through VEGFA Subunit Alpha signaling, possibly by regulating RhoA activity. Phosphatidyl- PIK3CB Generates PIP3. Involved in the activation of AKT1 upon stimulation by inositol- G-protein coupled receptors (GPCRs) ligands such as CXCL12, kinases. Bisphosphate Plays a role in platelet activation signaling triggered by GPCRs, alpha- Kinase Catalytic IIb/beta-3 integrins (ITGA2B/ITGB3) and ITAM. Subunit Beta AKT Serine AKT1 Plays a key in regulating cell survival, insulin signaling, angiogenesis and Threonine Kinase 1 tumor formation. AKT1 is a downstream mediator of the PI 3-K pathway, which results in the recruitment of Akt to the plasma membrane. AKT Serine AKT2 Plays a key in regulating cell survival, insulin signaling, angiogenesis and Threonine Kinase 2 tumor formation. AKT2 is a downstream mediator of the PI 3-K pathway, which results in the recruitment of Akt to the plasma membrane. Phosphatase And PTEN Tumor suppressor. It negatively regulates intracellular levels of Tensin Homolog phosphatidylinositol-3,4,5-trisphosphate in cells and functions as a tumor suppressor by negatively regulating AKT/PKB signaling pathway. Tuberous sclerosis TSC1 Inhibits the nutrient-mediated or growth factor-stimulated Complex Subunit 1 phosphorylation of S6K1 and EIF4EBP1 by negatively regulating mTORC1 signaling. Seems not to be required for TSC2 GAP activity towards RHEB. Involved in microtubule-mediated protein transport. Tuberous sclerosis TSC2 Inhibits the nutrient-mediated or growth factor-stimulated Complex Subunit 2 phosphorylation of S6K1 and EIF4EBP1 by negatively regulating mTORC1 signaling. Acts as a GTPase-activating protein (GAP) for the small GTPase RHEB, a direct activator of the protein kinase activity of mTORC1. Ras Homolog, RHEB Vital in regulation of growth and cell cycle progression due to its role in MTORC1 Binding the insulin/TOR/S6K signaling pathway. Activates the protein kinase activity of mTORC1, and thereby plays a role in the regulation of apoptosis. Stimulates the phosphorylation of S6K1 and EIF4EBP1 through activation of mTORC1 signaling. Has low intrinsic GTPase activity. FKBP Prolyl FKB-12 Play a role in immunoregulation and basic cellular processes involving Isomerase 1A protein folding and trafficking. Binds the immunosuppressants FK506 and rapamycin. It interacts with several intracellular signal transduction proteins including type I TGF-beta receptor. Mechanistic Target MTOR Target for the cell-cycle arrest and immunosuppressive effects of the Of Rapamycin FKBP12-rapamycin complex. Functions as part of 2 structurally and Kinase functionally distinct signaling complexes mTORC1 and mTORC2. Activation of MTORC1 trigger phosphorylation of EIF4EBP1 and release of its inhibition toward the elongation initiation factor 4E (eiF4E). Phosphorylates and activates RPS6KB1 that promote protein synthesis. MTOR Associated MLST8 Subunit of both mTORC1 and mTORC2. Within mTORC1, LST8 interacts Protein, LST8 directly with MTOR and enhances its kinase activity. In nutrient-poor Homolog conditions, stabilizes the MTOR-RPTOR interaction and favors RPTOR- mediated inhibition of MTOR activity. mTORC2 is also activated by growth factors, but seems to be nutrient-insensitive. Regulatory RPTOR Forms a stoichiometric complex with the mTOR kinase (MTORC1), and Associated Protein also associates with eukaryotic initiation factor 4E-binding protein-1 and Of MTOR Complex 1 ribosomal protein S6 kinase. The protein positively regulates the downstream effector ribosomal protein S6 kinase, and negatively regulates the mTOR kinase. RPTOR RICTOR Subunit of mTORC2: regulates cell growth and survival in response to Independent hormonal signals. mTORC2 is activated by growth factors, but, in contrast to mTORC1, is nutrient-insensitive. mTORC2 seems to function upstream Companion of of Rho GTPases to regulate the actin cytoskeleton, probably by activating MTOR Complex 2 one or more Rho-type guanine nucleotide exchange factors. mTORC2 promotes the serum-induced formation of stress-fibers or F-actin. Ribosomal Protein S6K1 Acts downstream of mTOR signaling in response to growth factors and S6 Kinase B1 nutrients to promote cell proliferation, cell growth and cell cycle progression. Regulates protein synthesis through phosphorylation of EIF4B, RPS6 and EEF2K, and contributes to cell survival by repressing the pro-apoptotic function of BAD. Eukaryotic 4EBP1 Translation repressor protein. Directly interacts with eukaryotic Translation translation initiation factor 4E (eIF4E), which is a limiting component of Initiation Factor 4E the multi-subunit complex that recruits 40S ribosomal subunits to the 5′ Binding Protein 1 end of mRNAs. Hypoxia Inducible HIF1 Master regulator of cellular and systemic homeostatic response to Factor 1 Subunit hypoxia by activating transcription of many genes, including those Alpha involved in energy metabolism, angiogenesis, apoptosis, and other genes whose protein products increase oxygen delivery or facilitate metabolic adaptation to hypoxia. Everolimus inhibits the expression of hypoxia- inducible factor (e.g., HIF-1). Vascular VEGFA Growth factor active in angiogenesis, vasculogenesis and endothelial cell Endothelial Growth growth. Induces endothelial cell proliferation, promotes cell migration, Factor A inhibits apoptosis and induces permeabilization of blood vessels. Binds to the FLT1/VEGFR1 and KDR/VEGFR2 receptors, heparan sulfate and heparin. Everolimus reduces the expression of VEGFA and reduce cell proliferation, angiogenesis, and glucose uptake.

TABLE-US-00003 TABLE 3 Description and rationale for the selection of key genes of immune blockade Usual Official Name names Role in the negative immune-blockade The T lymphocytes (LyT) that infiltrate the tumor (TILS) recognize the presented tumor neo-antigens. The neo-antigens are recognized as “non-self” as they are modified proteins because of mutations. The clone of LyT that recognizes specifically the neo-antigen is activated and proliferates. The recruitment of activated LyT that recognize specifically the tumor is a complex process that involves different antigen presentation mechanisms. Professional Antigen Presenting Cells (APC) present the neo-antigen associated to the major histocompatibility complex II (CMH2) recognized by LyT CD4+ that differentiate in LyT Helper 1 (Ly Th1) and Helper 2 (Ly Th2). Ly Th1 are key in recruitment of naïve LyT CD8+ and induce their activation. Lymphocytes T Cytotoxic (CD8+) and Natural Killer lymphocytes (NK) also recognize the neo-antigen restricted to CMH1 (Histocompatibility complex 1) and are subsequently activated, and can destroy directly the tumoral cells presenting the neo-antigen. The process of recruitment and activation of cytotoxic lymphocytes T CD8 is controlled by different mechanisms of negative blockade. PD-1 PDCD1 PDL1 and PDL2 binds to PD1 and directly inhibit the T receptor. CTLA4 has a PD-L1 CD274 high affinity and avidity for BF1 and BF2 ligands that binds to the co- stimulatory PD-L2 PDCD1LG2 molecule CD28. In this competitive manner, CTLA4 blocks CD28 and has a negative blockade effect. PD-1 and CTLA-4 are highly expressed on TILs in metastatic melanoma, NSCLC, UBC, and squamous cell carcinoma of the head CTLA4 CTLA4 and neck. PD-1 and CTLA-4 modulate effector T cell activation, proliferation, B7-1 CD80 and function through distinct, complementary mechanisms. The expression of B7-2 CD86 PD-1 and CTLA-4 on tumor-infiltrating T cell populations contributes to CD28 CD28 suppression and immunological escape. In vivo studies have shown that tumor- infiltrating lymphocytes and not peripheral T cells have been shown to be the major contributor to tumor control following anti-PD-L1 + anti-CTLA-4 mAb therapy. LAG3 LAG3 LAG3-Delivers inhibitory signals upon binding to ligands, such as FGL1 (responsible for LAG3 T-cell inhibitory function). Following TCR engagement, LAG3 associates with CD3-TCR in the immunological synapse and directly inhibits T-cell activation (may inhibit antigen-specific T-cell activation in synergy with PDCD1/PD-1, possibly by acting as a co-receptor for PDCD1/PD-1). LAG3 negatively regulates the proliferation, activation, effector function and homeostasis of both CD8(+) and CD4(+) T-cells. Also mediates immune tolerance. LAG3 is constitutively expressed on a subset of regulatory T-cells (Tregs) and contributes to their suppressive function. TLR4 TLR4 TLR4- Its over expression and activation by LPS activates MAPK and NF-κB pathways, implicating cell-autonomous TLR4 signaling in regulation of carcinogenesis, in particular, through increased proliferation of tumor cells, apoptosis inhibition and metastasis. TLR4 signaling in immune and inflammatory cells of tumor microenvironment may lead to production of pro- inflammatory cytokines (TNF, IL-1β, IL-6, IL-18, etc.), immunosuppressive cytokines (IL-10, TGF-β, etc.) and angiogenic mediators (VEGF, EGF, TGF-β). CD8 CD8A The level of infiltration of the tumor by Cytotoxic lymphocytes T CD8 (LyTc) - can be assessed by investigating the specific marker CD8. CD16 FCGR3 The level of infiltration of the tumor by Natural Killers cells (NK)- can be assessed by investigating the specific marker is CD16. FOXP3 FOXP3 The level of the infiltration of the tumor by a specific population of Lymphocytes T called regulatory (T-regs)- can be assessed by investigating the specific marker FOXP3.