In vitro method for predicting the response of patients suffering from rheumatoid arthritis to a treatment with tumor necrosis factor (TNF) inhibitors

Abstract

The present invention refers to a method for the identification of biomarkers or biomarker signatures for predicting the response of patients suffering from rheumatoid arthritis to a treatment with tumor necrosis factor (TNF) inhibitors, or for classifying patients suffering from rheumatoid arthritis into responder or non-responder patients to a treatment with TNF inhibitors, or for monitoring patients suffering from rheumatoid arthritis to assess whether they are responding to a treatment with TNF inhibitors, or for deciding or recommending whether to treat patients suffering from rheumatoid arthritis with TNF inhibitors.

Description

[0001]IN VITRO METHOD FOR PREDICTING THE RESPONSE OF PATIENTSSUFFERING FROM RHEUMATOID ARTHRITIS TO A TREATMENT WITH TUMOR NECROSIS FACTOR (TNF) INHIBITORS FIELD OF THE INVENTION The present invention refers to the medical field. Particularly, the present invention refers to a method for the identification of biomarkers or biomarker signatures for predicting the response of patients suffering from rheumatoid arthritis to a treatment with tumor necrosis factor (TNF) inhibitors, or for classifying patients suffering from rheumatoid arthritis into responder or non-responder patients to a treatment with TNF inhibitors, or for monitoring patients suffering from rheumatoid arthritis to assess whether they are responding to a treatment with TNF inhibitors, or for deciding or recommending whether to treat patients suffering from rheumatoid arthritis with TNF inhibitors. STATE OF THE ART Rheumatoid arthritis (RA) is a chronic, systemic, inflammatory disease affecting approximately 0.5-1% of the global population, having more prevalence in females than in males. Is characterized by the inflammation of the synovial joints, which leads to progressive joint destruction, disability, and reduced quality of life. In particular, 80% of insufficient treated patients develops misaligned joints and a half of them are unable to work within ten years after the diagnosis. Treatment strategy at the diagnosis begins with lifestyle counselling and non-steroid anti-inflammatory drugs (NSAIDs). Next step are the glucocorticoids and since nineties the denominated DMARDs (disease modifying anti-rheumatic drugs), in whichstands out the anti- tumour necrosis factor (TNF)-α therapy: the TNF inhibitors and the TNF-receptor inhibitors. Anti-TNF agents have revolutionized the treatment of RA, significantly improved clinical outcomes and inhibited radiographic progression in a substantial proportion of patients. However, although anti-TNF therapy is effective in most cases, up to 30% of treated patients show an inadequate response. Despite all the recent knowledge about the RA, there are some challenges that need to be solved. Specifically, due to the heterogeneity of the disease and because of the elevated risk of unfavourable health problems, such as infections, there is an urge to predict the therapy response. Therefore, there is an unmet medical need of finding reliable tool, particularly biomarkers,aimed at predicting the response of RA patients to TNF inhibitors treatment. The presentinvention is focused on solving this problem, thus providing a reliable method for the identification of biomarkers / biomarker signatures for predicting the response to TNFinhibitors in RA patients, for classifying RA patients into responders or non-responders tothis treatment and, finally, for monitoring whether RA patients are responding to the therapyin order to decide or recommend whether TNF inhibitors treatment should be used.DESCRIPTION OF THE INVENTION Brief description of the inventionThe present invention refers to an in vitro method for the identification of biomarkers orbiomarker signatures for predicting the response of patients suffering from RAto a treatment with TNF inhibitors. Moreover, the present invention refers to the use of the identifiedbiomarkers or biomarker signatures for classifying patients suffering from RA into responderor non-responder patients to a treatment with TNF inhibitors, or for monitoring patientssuffering from RA to assess whether they are responding to a treatment with TNF inhibitors,or for deciding or recommending whether to treat patients suffering from RA with TNFinhibitors.Particularly, the inventors of the present invention initially performed the discovery phasedemonstrating that transcriptomic data could serve as a prognostic and predictive biomarkerin RA patients under anti-TNF therapy. So, a bioinformatic analysis was performed in whicha transcriptomic signature was reported, as proof of concept, with high capacity to predict theresponse to anti-TNF treatment in patients with RA, prior to its initiation. Although, a explained above, a specific transcriptomic signature comprising the genesKCNK17, GLS2, GTPBP2, DNTTIP1, IL18R1 and COMTD1 was reported as proof ofconcept in the discovery phase (see Example 2), the following tables (Table 1, Table 2 andTable 3) show several signatures comprising at least two genes selected from the listconsisting of: KCNK17, GLS2, GTPBP2, DNTTIP1, IL18R1, COMTD1, FEM1C, MRPL24and / or MYPOP, or combinations thereof comprising between 2 and 9 of these genes, asreliable signatures for performing the method of the invention in different data sets, all of them showing AUC ≥ 0.8. Table 1 4 KCNK17+GLS2+DNTTIP1+FEM1C 0.804252658KCNK17+DNTTIP1+IL18R1+FEM1C 0.8136335214287741857264454411 6Dataset GSE138746 (39 R vs.41 NR) Table 2 GLS2+MYPOP+GTPBP2+IL18R1+COMTD1+FEM1C 0.80952381Dataset GSE224842 (21 R vs.9 NR) Table 3 Gene combination AUC666166696646961664166549641946 KCNK17+GTPBP2+IL18R1+COMTD1+FEM1C 0.819444444GLS2+MRPL24+GTPBP2+DNTTIP1+IL18R1 0.803240741546549115434694415465 9Dataset GSE33377 (18 R vs.24 NR)So, the first embodiment of the present invention refers to an in vitro method for theidentification of biomarkers or biomarker signatures for predicting the response of patientssuffering from RA to a treatment with TNF inhibitors, or for classifying patients sufferingfrom RA into responder or non-responder patients to a treatment with TNF inhibitors, or formonitoring patients suffering from RA to assess whether they are responding to a treatmentwith TNF inhibitors, or for deciding or recommending whether to treat patients sufferingfrom RA with TNF inhibitors which comprises: a) Assessing the level of expression of atleast a gene selected from the list consisting of: KCNK17, GLS2, GTPBP2, DNTTIP1,IL18R1, COMTD1, FEM1C, MRPL24 and / or MYPOP, or any combination thereof, in abiological sample obtained from patients suffering from RA who are responder patients totreatment with TNF inhibitors, b) assessing the level of expression of at least a gene selectedfrom the list consisting of KCNK17, GLS2, GTPBP2, DNTTIP1, IL18R1, COMTD1, FEM1C, MRPL24 and / or MYPOP, or any combination thereof, in a biological sampleobtained from patients suffering from RA who are non-responder patients to treatment withTNF inhibitors, and c) wherein the identification of a deviation of the level of expression of atleast a gene selected from the list consisting of KCNK17, GLS2, GTPBP2, DNTTIP1, IL18R1, COMTD1, FEM1C, MRPL24 and / or MYPOP, or any combination thereof, across responder and non-responder subjects, is an indication that the gene / s can be used as biomarkers or biomarker signatures for predicting the response of patients suffering from RAto treatment with TNF inhibitors, or for classifying patients suffering from RA into responderor non-responder to a treatment with TNF inhibitors, or for monitoring patients sufferingfrom RA to assess whether they are responding to a treatment with TNF inhibitors, or fordeciding or recommending whether to treat patients suffering from RA with TNF inhibitors.The second embodiment of the present invention refers to an in vitro method for predictingthe response of patients suffering from RA to a treatment with TNF inhibitors, or forclassifying patients suffering from RA into responder or non-responder patients to a treatmentwith TNF inhibitors, for monitoring patients suffering from RA to assess whether they areresponding to treatment with TNF inhibitors, or for deciding or recommending whether totreat patients suffering from RA with TNF inhibitors, which comprises: a) Assessing the levelof expression of at least a gene selected from the list consisting of: KCNK17, GLS2,GTPBP2, DNTTIP1, IL18R1, COMTD1, FEM1C, MRPL24 and / or MYPOP, or anycombination thereof, in a biological sample obtained from patients suffering from RA, b)wherein the identification of a higher expression level of the genes: KCNK17, IL18R1,FEM1C and / or MYPOP as compared with a pre-established threshold value measured in non- responder patients, is an indication that the patient will respond to treatment with TNFinhibitors, or that the patient is responding to treatment with TNF inhibitors, or c) wherein theidentification of a higher expression level of the genes: GLS2, GTPBP2, DNTTIP1,COMTD1 and / or MRPL24 as compared with a pre-established threshold value measured innon-responder patients, is an indication that the patient will not respond to treatment with TNF inhibitors, or that the patient is not responding to treatment with TNF inhibitors. In a preferred embodiment the methods described above comprise: a) Assessing the level ofexpression of a combination of genes selected from Table 1, Table 2 or Table 3, in abiological sample obtained from patients suffering from RA, b) wherein the identification ofa higher expression level of the genes: KCNK17, IL18R1, FEM1C and / or MYPOP ascompared with a pre-established threshold value measured in non-responder patients, is an indication that the patient will respond to treatment with TNF inhibitors, or that the patient isresponding to treatment with TNF inhibitors, or c) wherein the identification of a higherexpression level of the genes: GLS2, GTPBP2, DNTTIP1, COMTD1 and / or MRPL24 ascompared with a pre-established threshold value measured in non-responder patients, is an indication that the patient will not respond to treatment with TNF inhibitors, or that the patient is not responding to treatment with TNF inhibitors.The third embodiment of the present invention refers to the in vitro use of at least a geneselected from the list consisting of: KCNK17, GLS2, GTPBP2, DNTTIP1, IL18R1,COMTD1, FEM1C, MRPL24 and / or MYPOP, or of a kit comprising reagents for theidentification of the level of expression of at least a gene selected from the list consisting of:KCNK17, GLS2, GTPBP2, DNTTIP1, IL18R1, COMTD1, FEM1C, MRPL24 and / orMYPOP, or any combination thereof, for predicting the response of patients suffering fromRA to a treatment with TNF inhibitors, or for classifying patients suffering from RA intoresponder or non-responder patients to a treatment with TNF inhibitors, or for monitoringpatients suffering from RA to assess whether they are responding to a treatment with TNFinhibitors, or for deciding or recommending whether to treat patients suffering from RA withTNF inhibitors.In a preferred embodiment the present invention refers to the in vitro use of a combination ofgenes selected from Table 1, Table 2 or Table 3, for predicting the response of patientssuffering from RA to a treatment with TNF inhibitors, or for classifying patients sufferingfrom RA into responder or non-responder patients to a treatment with TNF inhibitors, or formonitoring patients suffering from RA to assess whether they are responding to a treatmentwith TNF inhibitors, or for deciding or recommending whether to treat patients sufferingfrom RA with TNF inhibitors.The fourth embodiment of the present invention refers to TNF inhibitors for use in a methodof treatment of RA, wherein the method comprises assessing whether the patient would be a responder patient to a treatment with TNF inhibitors by assessing the level of expression of at least a gene selected from the list consisting of: KCNK17, GLS2, GTPBP2, DNTTIP1,IL18R1, COMTD1, FEM1C, MRPL24 and / or MYPOP, or any combination thereof, in abiological sample obtained from patients suffering from RA, wherein the identification of ahigher expression level of the genes: KCNK17, IL18R1, FEM1C and / or MYPOP ascompared with a pre-established threshold value measured in non-responder patients, is an indication that the patient will respond to treatment with TNF inhibitors; or wherein the identification of a higher expression level of the genes: GLS2, GTPBP2, DNTTIP1,COMTD1 and / or MRPL24 as compared with a pre-established threshold value measured innon-responder patients, is an indication that the patient will not respond to treatment with TNF inhibitors, or that the patient is not responding to treatment with TNF inhibitors. Alternatively, the present invention refers to a method for treating a patient suffering fromRA with TNF inhibitors, wherein the method comprises a first step of determining whetherthe patient would be a responder patient to a treatment with TNF inhibitors by assessing the level of expression of at least a gene selected from the list consisting of: KCNK17, GLS2,GTPBP2, DNTTIP1, IL18R1, COMTD1, FEM1C, MRPL24 and / or MYPOP, or anycombination thereof, in a biological sample obtained from patients suffering from RA,wherein the identification of a higher expression level of the genes: KCNK17, IL18R1,FEM1C and / or MYPOP as compared with a pre-established threshold value measured in non- responder patients, is an indication that the patient will respond to treatment with TNF inhibitors; or wherein the identification of a higher expression level of the genes: GLS2,GTPBP2, DNTTIP1, COMTD1 and / or MRPL24 as compared with a pre-establishedthreshold value measured in non-responder patients, is an indication that the patient will not respond to treatment with TNF inhibitors, or that the patient is not responding to treatment with TNF inhibitors.In a preferred embodiment the present invention refers to TNF inhibitors for use in a methodof treatment of RA, wherein the method compromises assessing whether the patient would bea responder patient to a treatment with TNF inhibitors by assessing the level of expression ofa combination of genes selected from Table 1, Table 2 or Table 3, in a biological sampleobtained from patients suffering from RA, wherein the identification of a higher expressionlevel of the genes: KCNK17, IL18R1, FEM1C and / or MYPOP as compared with a pre-established threshold value measured in non-responder patients, is an indication that the patient will respond to treatment with TNF inhibitors, or that the patient is responding to treatment with TNF inhibitors; or wherein the identification of a higher expression level ofthe genes: GLS2, GTPBP2, DNTTIP1, COMTD1 and / or MRPL24 as compared with a pre-established threshold value measured in non-responder patients, is an indication that the patient will not respond to treatment with TNF inhibitors, or that the patient is not respondingto treatment with TNF inhibitors. Alternatively, the present invention refers to a method fortreating a patient suffering from RA with TNF inhibitors, wherein the method comprises afirst step of determining whether the patient would be a responder patient to a treatment withTNF inhibitors by assessing whether the patient would be a responder patient to a treatment with TNF inhibitors by assessing the level of expression of a combination of genes selectedfrom Table 1, Table 2 or Table 3, in a biological sample obtained from patients sufferingfrom RA, wherein the identification of a higher expression level of the genes: KCNK17,IL18R1, FEM1C and / or MYPOP as compared with a pre-established threshold value measured in non-responder patients, is an indication that the patient will respond to treatment with TNF inhibitors, or that the patient is responding to treatment with TNF inhibitors; or wherein the identification of a higher expression level of the genes: GLS2, GTPBP2,DNTTIP1, COMTD1 and / or MRPL24 as compared with a pre-established threshold valuemeasured in non-responder patients, is an indication that the patient will not respond to treatment with TNF inhibitors, or that the patient is not responding to treatment with TNF inhibitors. In a preferred embodiment, the biological sample is blood, serum or plasma. In a preferred embodiment, the TNF inhibitors is selected from: infliximab, etanercept, golimumab, certolizumab and adalimumab. On the other hand, the inventors of the present invention carried out a validation phasewherein an analysis of peripheral blood mononuclear cells (PBMC) transcriptomic profilesfrom RA patients was performed, utilizing data from the Gene Expression Omnibus (GEO)database (GSE138746). An additional dataset (GSE33377), profiled by microarray, was alsoadded to define a transcriptomic signature based on highly discriminatory genes forpredicting response to anti-TNF treatment. Two external validations were then conductedusing data from the COMBINE and the Reina Sofia Hospital cohort, contributing to the studywith a total of 260 individuals, 160 responders and 100 non-responders.The initial RNA-seq analysis (GSE138746) identified 53 differentially expressed genesbetween responders and non-responders; however, none remained significant after p-value adjustment with the Benjamini–Hochberg method. A small-scale genetic signature comprising the 18 most discriminatory genes was then developed, achieving a leave-one-outcross-validation (LOOCV) predictive accuracy of 88.75%. The list was further defined toseven genes (COMTD1, MRPL24, DNTTIP1, GLS2, GTPBP2, IL18R1 and KCNK17) that effectively predicted response to anti-TNF treatment, with an area under the receiver operating characteristic (ROC) curve (AUC) of 0.84 in the GSE33377 dataset. Internal validation in GSE138746 dataset showed an AUC of 0.89. Finally, external validation in theCOMBINE and the Reina Sofia Hospital cohorts confirmed the robustness of the seven-genemodel, yielding an AUC of 0.85.Although, as it is explained above, a specific transcriptomic signature comprising the genesCOMTD1, MRPL24, DNTTIP1, GLS2, GTPBP2, IL18R1 and KCNK17 was reported asproof of concept in the validation phase (see Example 2), the following tables (Table 4 andTable 5) show several signatures comprising at least two genes selected from the list consisting of: COMTD1, MRPL24, DNTTIP1, GLS2, GTPBP2, IL18R1 and KCNK17, as reliable signatures for performing the method of the invention in different data sets. Table 4. Data set GSE3337769888172648653 MRPL24, COMTD1 0.583333333333333GTPBP2, DNTTIP1 0.775462962962963629287268863323169216768719186 MRPL24, GTPBP2, IL18R1 0.805555555555556MRPL24, GTPBP2, COMTD1 0.7893518518518526536797617732641864748162971676 GLS2, MRPL24, DNTTIP1, COMTD1 0.678240740740741GLS2, MRPL24, IL18R1, COMTD1 0.6736111111111117244626142461229842841755594 KCNK17, GLS2, MRPL24, GTPBP2, DNTTIP1, COMTD1 0.784722222222222KCNK17, GLS2, MRPL24, GTPBP2, IL18R1, COMTD1 0.83333333333333344K 3 The results of the data set GSE33377 show a high AUC in the 7 gene signature, with a valueof 0.837962962962963. Combinations including at least 3 genes already exhibited a good accuracy (slightly lower in comparison with GSE138746 dataset, but anyway worthy and >0.8) when it comes to distinguish between the Responder and Non-responders’ patientsfrom RA under TNF inhibitors treatment (for example, KCNK17, GTPBP2, IL18R1, AUC=0.803240740740741). Table 5. Data set GSE138746 G , 8 0.808005003 695GTPBP2, COMTD1 0.768605378361476 MRPL24, DNTTIP1, IL18R1 0.84677923702314MRPL24, DNTTIP1, COMTD1 0.813633520950594 GLS2, GTPBP2, DNTTIP1, IL18R1 0.873671044402752GLS2, GTPBP2, DNTTIP1, COMTD1 0.862414008755472 KCNK17, GLS2, MRPL24, DNTTIP1, IL18R1, COMTD1 0.931207004377736KCNK17, GLS2, GTPBP2, DNTTIP1, IL18R1, COMTD1 0.924953095684803 This results in data set GSE138746 show a high AUC in the 7 gene signature, with a value of almost 0.95 (AUC = 0.948717948717949). Combinations including at least 3 genes already exhibited a good accuracy when it comes to distinguish between the Responder and Non- responders’ patients from rheumatoid arthritis under TNF inhibitors treatment (for example, KCNK17, IL18R1, COMTD1, AUC= 0.868042526579112).In conclusion, during the validation phase, a transcriptomic signature was identified thatenables prediction of response to anti-TNF treatment in RA patients. These findings underscore the potential of transcriptomic profiling to enhance the selection of optimal treatments for individual RA patients. So, the present invention also refers to:An in vitro method for predicting the response of patients suffering from RA to a treatmentwith TNF inhibitors, or for classifying patients suffering from RA into responder or non-responder patients to a treatment with TNF inhibitors, for monitoring patients suffering fromRA to assess whether they are responding to treatment with TNF inhibitors, or for deciding orrecommending whether to treat patients suffering from RA with TNF inhibitors, wherein themethod comprises assessing the level of expression of the gene combination: MRPL24, COMTD1, DNTTIP1, GLS2, GTPBP2, IL18R1 and KCNK17, in a biological sample obtained from the patient. In a preferred embodiment, the identification of a higher expression level of the genes: KCNK17 and / or IL18R1, as compared with a pre-established threshold value measured in non-responder patients, is an indication that the patient will respond to treatment with TNF inhibitors, or that the patient is responding to treatment with TNF inhibitors, and / or wherein the identification of a higher expression level of the genes: GLS2, GTPBP2, DNTTIP1,COMTD1 and / or MRPL24 as compared with a pre-established threshold value measured in non-responder patients, is an indication that the patient will not respond to treatment with TNF inhibitors, or that the patient is not responding to treatment with TNF inhibitors.The present invention also refers to the in vitro use of the gene combination: MRPL24,COMTD1, DNTTIP1, GLS2, GTPBP2, IL18R1 and KCNK17, or of a kit comprising reagents for the identification of the level of expression of the gene combination: MRPL24, COMTD1, DNTTIP1, GLS2, GTPBP2, IL18R1 and KCNK17, for predicting the response ofpatients suffering from RA to a treatment with TNF inhibitors, or for classifying patientssuffering from RA into responder or non-responder patients to a treatment with TNFinhibitors, or for monitoring patients suffering from RA to assess whether they areresponding to a treatment with TNF inhibitors, or for deciding or recommending whether totreat patients suffering from RA with TNF inhibitors.The present invention also refers to TNF inhibitors for use in a method of treatment of RA,wherein the method comprises assessing whether the patient would be a responder patient to a treatment with TNF inhibitors by assessing the level of expression of the gene combination: MRPL24, COMTD1, DNTTIP1, GLS2, GTPBP2, IL18R1 and KCNK17, in a biological sample obtained from patients suffering from RA, wherein the identification of a higherexpression level of the genes: KCNK17 and / or IL18R1as compared with a pre-establishedthreshold value measured in non-responder patients, is an indication that the patient will respond to treatment with TNF inhibitors; and / or wherein the identification of a higherexpression level of the genes: GLS2, GTPBP2, DNTTIP1, COMTD1 and / or MRPL24 ascompared with a pre-established threshold value measured in non-responder patients, is an indication that the patient will not respond to treatment with TNF inhibitors, or that the patient is not responding to treatment with TNF inhibitors. In a preferred embodiment, the biological sample is blood, serum or plasma.In a preferred embodiment, the TNF inhibitors are selected from: infliximab, etanercept,golimumab, certolizumab and adalimumab. For the purpose of the present invention the following terms are defined: ●The term "comprising" means including, but it is not limited to, whatever follows theword "comprising". Thus, use of the term "comprising" indicates that the listed elements are required or mandatory, but that other elements are optional and may or may not be present. ●By "consisting of” means including, and it is limited to, whatever follows the phrase“consisting of”. Thus, the phrase "consisting of” indicates that the listed elements are required or mandatory, and that no other elements may be present. ●According to the present invention, a reference value can be a “pre-establishedthreshold value” or a “cut-off” value. Typically, a "threshold value" or "cut-off value" can be determined experimentally, empirically, or theoretically. According to the present invention, the “pre-established threshold” value refers to a value previously determined in patients suffering for rheumatoid arthritis who are non-responders to a treatment with TNF inhibitors. A “threshold value” can also be arbitrarily selected based upon the existing experimental and / or clinical conditions, as would be recognized by a person of ordinary skilled in the art. The “threshold value” has to be determined in order to obtain the optimal sensitivity and specificity according to the function of the test and the benefit / risk balance (clinical consequences of false positive and false negative). Typically, the optimal sensitivity and specificity (and so the “threshold value”) can be determined using a Receiver Operating Characteristic (ROC) curve based on experimental data. Description of the figures Figure 1. Flowchart utilized to search of a transcriptomic signature that predicts response to anti-TNF treatment. Figure 2. Differential expression analysis of the GSE138746 dataset.Figure 3. Predictive model consisting of 6 genes was selected to effectively discriminate R(responders) and NR (non-responders), where the results showed high area under the curve(AUC) values using three different datasets.Figure 4. Receiver operating characteristic curves of each gene using the GSE224842 dataset. Figure 5. Receiver operating characteristic curves of each gene using the GSE33377 dataset. Figure 6. Flowchart utilized to search for a transcriptomic signature that predicts response toanti-TNF treatment. Abbreviations: DEG, differentially expressed genes; PBMC, peripheralblood mononuclear cells; RA, rheumatoid arthritis.Figure 7. Differential expression analysis of the GSE138746 dataset. (A) Volcano plot ofdifferentially expressed genes (Fold Change > |1| and p-value <0.05); (B) Heatmap and (C) Principal Component Analysis of 53 differentially expressed genes at baseline (Fold Change> |1| and p-value <0.05). Figure 8. ROC Curve for predicting response using the transcriptomic signature in theGSE33377 dataset.Figure 9. Validation of the 7-gene transcriptomic signature for predicting response to anti-TNF therapy. (A) Boxplots showing the expression levels of each gene in the 7-genesignature (KCNK17, GLS2, MRPL24, GTPBP2, DNTTIP1, IL18R1, and COMTD1) between responders (R) and non-responders (NR) in the GSE138746 dataset. Statisticalsignificance was assessed using the Wilcoxon rank-sum test (*p-value < 0.05; **p-value <0.005). The ROC curve displays the discriminatory power of the 7-gene signature in theGSE138746 dataset; (B) ROC curve for the 7-gene transcriptomic signature applied to anindependent validation cohort and (C) ROC curve for the COMBINE cohort. Detailed description of the invention The present invention is illustrated by means of the Examples set below without the invention of limiting its scope of protection. Example 1. Material and methods Discovery phase Example 1.1. Data source The study scheme for uncovering and testing the capacity of a transcriptomic signature topredict anti-TNF response is shown in Figure 1. We performed a retrospective analysis ofwhole peripheral blood mononuclear cells (PBMC) transcriptomic profiling for patients with RA, to define the small-scale signature genes involved in the response to anti-TNF treatment. The Gene Expression Omnibus (GEO) database (https: / / www.ncbi.nlm.nih.gov / geo / ) numbered GSE138746 was obtained [Tao W, Concepcion AN, Vianen M, Marijnissen ACA, Lafeber FPGJ, Radstake TRDJ, et al. Multiomics and Machine Learning Accurately Predict Clinical Response to Adalimumab and Etanercept Therapy in Patients With Rheumatoid Arthritis. Arthritis Rheumatol. 2021;73(2):212-22]. This dataset contained 80 samples with gene expression profiling on PBMC, including 38 and 42 samples from RA patients treatedwith adalimumab and etanercept, respectively. Two more publicly available datasets(GSE224842 and GSE33377) profiled by microarray was downloaded from GEO database todefine, based on the small-scale signature genes, a transcriptomic signature capable ofpredicting response prior to anti-TNF treatment. GSE224842 dataset contained expressionprofiling on PBMC for 30 RA patients initiating treatment with abatacept, whereas theGSE33377 dataset contained expression profiling on whole blood for 42 RA patients startingtreatment with infliximab or adalimumab [Toonen EJM, Gilissen C, Franke B, Kievit W, Eijsbouts AM, den Broeder AA, et al. Validation study of existing gene expression signatures for anti-TNF treatment in patients with rheumatoid arthritis. PLoS One.2012;7(3):e33199].Example 1.2. Differentially expressed genes analysisTo perform an initial analysis of differential expression, the statistical analyses were performed using R version 4.1.1. After processing the expression matrix of the GSE138746 dataset using the DESeqDataSetFromMatrix R function, principal component study and differential expression analysis were performed with the DESeq function [Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014;15(12):550]. Differential gene expression analysis was done using the parametric Wald test with the Benjamini-Hochberg adjustment method. Those genes with counts mean lower than 5 were removed from the DESeq table before the p-value adjustment. The GSEA algorithm was used for enrichment analysis of Wikipathways and GO terms (cellular component, biological processes and molecular function) [Yu G, Wang LG, Han Y, He QY. clusterProfiler: an R package for comparing biological themes among gene clusters. OMICS.2012;16(5):284-7]. Example 1.3. Genetic small-scale signature The identification of a transcriptomic signature capable of predicting response prior to anti- TNF treatment constitutes a phenotype prediction problem. This category of problems is characterized by a high degree of under-determinacy, given that the number of monitored genes far exceeds the number of samples employed. It is imperative to prioritize genes based on their discriminatory power in predicting the phenotype and to focus on sampling highly predictive networks anticipated to be involved in genetic pathways underlying treatment response. Mathematically, these networks inhabit the uncertainty space associated with the corresponding classifier used for phenotype discrimination. The mathematical structure of uncertainty spaces in the context of linear and nonlinear inverse problems has been previously elucidated [Fernández Martínez JL, Fernández Muñiz MZ, Tompkins MJ. On the topography of the cost functional in linear and nonlinear inverse problems. GEOPHYSICS. 2012;77(1):W1-15][Fernández-Martínez JL, Fernández-Muñiz Z, Pallero JLG, Pedruelo- González LM. From Bayes to Tarantola: New insights to understand uncertainty in inverse problems. Journal of Applied Geophysics. 2013;98:62-72]. Within a given classifier, the smallest-scale signature is defined as one comprising the fewest discriminatory genes with the highest predictive accuracy. However, the presence of noise in genetic data implies that some highly discriminatory signatures may not directly correspond to relevant genetic pathways [deAndrés-Galiana EJ, Fernández-Martínez JL, Sonis ST. Sensitivity analysis of gene ranking methods in phenotype prediction. J Biomed Inform. 2016;64:255- 64][Deandrés-Galiana EJ, Fernández-Martínez JL, Saligan LN, Sonis ST. Impact of Microarray Preprocessing Techniques in Unraveling Biological Pathways. J Comput Biol. 2016;23(12):957-68]. In addressing parameter identification problems, the significance of noise necessitates its integration into the evaluation of identified solutions, underscoring the importance of appraising the discriminatory power of key genes related to the phenotype using sampling methodologies. In this research, the Fisher's Ratio Sampler has been applied [Cernea A, Fernández-Martínez JL, deAndrés-Galiana EJ, Fernández-Ovies FJ, Fernández-Muñiz Z, Alvarez-Machancoses O, et al. Sampling Defective Pathways in Phenotype Prediction Problems via the Fisher’s Ratio Sampler. En: Rojas I, Ortuño F, editores. Bioinformatics and Biomedical Engineering. Cham: Springer International Publishing; 2018. p. 15-23], involving the sampling of defective pathways based on the discriminatory power of individual genes as measured by the Fisher's ratio. This entails identifying genes with the highest Fisher's ratio within the set of genes exhibiting the highest fold change. Specifically, genes that are differentially expressed in both tails (over and under-expressed) are identified, and these genes are ranked based on their Fisher's ratio, which seeks genes that effectively separate the classes and exhibit low intra-class variance. The set of discriminatory genes is defined as those that are differentially expressed and possess a Fisher's ratio greater than 0.8. The Fisher's ratio cutoff value is a tunable parameter in this procedure, which can be adjusted, if necessary, to a minimum value of fr = 0.5 when the number of discriminatory genes is limited within this set. To establish a small-scale genetic signature, the most discriminatory genes are ranked in decreasing order according to their discriminatory power. The algorithm subsequently determines the small-scale signature that optimally discriminates between classes through recursive feature elimination. Predictive accuracy estimation is conducted through Leave-One-Out-Cross-Validation (LOOCV) using a nearest neighbour classifier. The small-scalesignature provides an approximate representation of the typical length (number of genes) of high discriminatory networks. The random sampler is employed to identify alternative networks of highly discriminatory genes, utilizing a prior sampling probability for any individual gene proportional to its Fisher's ratio. Following the random construction of a network based on the Fisher's probability distribution, its LOOCV predictive accuracy is established. This sampling process adheres to Bayes' rule, incorporating a prior probability contingent on the Fisher's ratio of the selected genes and a likelihood probability function dependent on the LOOCV predictive accuracy of the network. Ultimately, considering the most discriminatory networks with a predictive accuracy exceeding a predefined threshold (typically greater than 85%), subsequent sampling frequencies of the primary prognostic genes involved in these networks are determined. Validation phase Example 1.4. Data source and patients The study scheme for uncovering and testing the capacity of a transcriptomic signature topredict anti-TNF response is shown in Figure 6. We perform a retrospective analysis ofwhole peripheral blood mononuclear cells (PBMC) transcriptomic profiling for patients with RA to define the small-scale signature genes involved in the response to anti-TNF treatment. The Gene Expression Omnibus (GEO) database (https: / / www.ncbi.nlm.nih.gov / geo / ) numbered GSE138746 was obtained. This dataset contained 80 samples with gene expression profiling on PBMC, including 38 and 42 samples from RA patients treated with adalimumab(ADA) and etanercept (ETN), respectively. One more publicly available dataset (GSE33377)profiled by microarray was downloaded from the GEO database to define, based on thesmall-scale signature genes, a transcriptomic signature capable of predicting response prior to anti-TNF treatment. A small-scale signature gene is the smallest set of genes capable ofpredicting with the highest possible accuracy. The GSE33377 dataset contained expressionprofiling on whole blood for 42 RA patients starting treatment with infliximab (IFX) orADA. For independent validation of the predictive signature, we used two independent sets ofsamples. The Reina Sofia Hospital cohort including 94 RA patients recruited at the ReinaSofia Hospital from Cordoba, Spain, with moderate to high disease activity. These patients started TNF therapy after failing standard DMARDs treatment, and the cohort included 72responders and 22 non-responders (Table 6).Table 6 RF, IU / mL (mean ± SD) 175.88 ± 257.89 Treatments Additionally, we included the COMBINE cohort, comprising 37 individuals who previouslydid not respond to methotrexate (MTX) treatment and initiated TNFi therapy. This cohortincluded 26 TNFi responders (R) and 11 non-responders (NR). In both cases, COMBINA andReina Sofia Hospital cohorts, Good and Moderate EULAR responders were classified as “responders” and compared to EULAR “non-responders.” Example 1.5. Differentially expressed genes analysis To perform an initial analysis of differential expression, the statistical analyses were performed using R version 4.1.1. After processing the expression matrix of the GSE138746 dataset using the DESeqDataSetFromMatrix R function, principal component study and differential expression analysis were performed with the DESeq function. Differential gene expression analysis was done using the parametric Wald test with the Benjamini-Hochberg adjustment method. Genes with counts mean lower than 5 were removed from the DESeq table before the p-value adjustment. The GSEA algorithm was used for enrichment analysis of Wikipathways and GO terms (cellular component, biological processes and molecular function). Example 1.6. Genetic small-scale signature The identification of a transcriptomic signature capable of predicting response prior to anti- TNF treatment constitutes a phenotype prediction problem. This category of problems is characterized by a high degree of under-determinacy, given that the number of monitored genes far exceeds the number of samples employed. It is imperative to prioritize genes based on their discriminatory power in predicting the phenotype and to focus on sampling highly predictive networks anticipated to be involved in genetic pathways underlying treatment response. These networks occupy the uncertainty space linked to the classifier used for phenotype discrimination, a structure previously discussed in the context of linear and nonlinear inverse problems. The smallest-scale signature within a classifier includes the fewest genes with optimal predictive accuracy. However, noise in genetic data means that some high-accuracy signatures may not align with relevant genetic pathways. Addressing parameter identification thus requires accounting for noise, highlighting the need to evaluate the discriminatory power of key phenotype-related genes using sampling methods. In this study, we applied the Fisher's Ratio Sampler to identify defective pathways based on genes with high discriminatory power, as indicated by the Fisher's ratio. Genes with the highest Fisher's ratios among those with significant fold changes are selected, focusing ondifferentially expressed genes in both over- and under-expressed tails. A cutoff of 0.8 for theFisher's ratio defines the set of discriminatory genes, adjustable to 0.5 if needed when the number of such genes is limited. To establish a small-scale genetic signature, genes are ranked by discriminatory power, and recursive feature elimination identifies the smallest optimal signature. Predictive accuracy is assessed via Leave-One-Out Cross-Validation (LOOCV) with a nearest neighbor classifier, approximating the typical size of high- discriminatory networks. The random sampler identifies alternative networks of highly discriminatory genes, assigning each gene a sampling probability proportional to its Fisher's ratio. Networks are randomly constructed based on this distribution, and their LOOCV predictive accuracy is evaluated. This process follows Bayes' rule, incorporating a prior probability based on the Fisher's ratio of selected genes and a likelihood function dependent on the network's LOOCV accuracy. Ultimately, considering the most discriminatory networks with a predictive accuracy exceeding a predefined threshold (typically greater than 85%), subsequent sampling frequencies of the primary prognostic genes involved in these networks are determined. Example 1.7. Predictive model for TNF inhibitors treatment response To construct a predictive model for TNF inhibitors treatment response, logistic regressionwas employed, incorporating the normalized expression levels of seven genes (KCNK17,DNTTIP1, IL18R1, GLS2, MRPL24, GTPBP2, and COMTD1), interaction terms, and demographic variables (sex and age) as predictors. For the GSE138746 dataset, predicted probabilities of response were then calculated for each patient, and the model’s discriminative performance was evaluated using receiver operating characteristic (ROC) curve analysis and the area under the curve (AUC). The same methodology was applied to the independent validation cohorts. Example 1.8. Statistical analysis Statistical analyses were performed using R version 4.2.1. The following R packages were used: tidyverse, pROC, ggplot2, dplyr, caret, leaps, patchwork and ggpubr. For boxplotcomparisons of gene expression between R and NR, p-values were calculated using theWilcoxon rank-sum test. ROC curves were generated using the pROC package, and AUC values were compared using DeLong’s test for statistical significance. Example 2. Results Discovery phase Example 2.1. Analysis of differential expression genes of the GSE138746 dataset Based on an initial analysis of the RNA-seq results, a total of 53 differentially expressed genes were identified between responders (N) and non-responders (NR), including 28 up-regulated genes and 25 down-regulated genes in R versus NR, according to the criteria: FoldChange > |1| and p-value <0.05. To visualize the differential expression genes, we constructed a volcano-plot (Figure 2A). No significant differences were found between both groups when the p-value was adjusted by the Benjamini–Hochberg adjustment method. In fact, the representation of the heatmap and the Principal Component Analysis (PCA) analysis show a poor separation of the samples from responders (yellow) and non-responders (red) toanti-TNF therapy according to the 53 differentially expressed genes found by RNA-seqanalysis (Figure 2B and Figure 2C, respectively). Functional enrichment analysis does notreport any significant relation between the analyzed genes and the response (all the pathways show p-values>0.05). Example 2.2. A transcriptomic signature was associated with the prediction of response to anti-TNF therapyTable 7 shows the list of most discriminatory genes of the phenotypes of the responderpatients compared with no-responders (most discriminatory genes using the Fisher's RatioSampler). Table 7 Gene MedianC1 StdC1 MedianC2 StdC2 FC FR LOOCVKCNK17 1.0 13.28 15.0 13.04 -0.92 0.57 67.50LOC100506235 29.0 15.90 13.0 14.49 0.64 0.55 68.75TRBV6-4 12.0 12.45 0.0 11.71 0.93 0.49 70.00GLS2 68.0 37.01 35.0 30.20 0.53 0.48 68.75MRPL24 456.0 107.41 356.0 110.06 0.20 0.42 72.50DNTTIP1 639.0 197.64 464.0 183.20 0.27 0.42 72.50MYPOP 101.0 51.91 150.0 56.29 -0.33 0.41 78.75LPIN3 12.0 16.64 0.0 9.49 1.22 0.39 73.75OR2A9P 16.0 12.90 4.0 15.38 0.20 0.36 75.00GTPBP2 1824.0 341.08 1539.0 337.47 0.12 0.35 70.00ZFTA 141.0 56.53 189.0 58.65 -0.20 0.35 76.25COMTD1 199.0 65.24 150.0 51.84 0.37 0.35 77.50ZFP57 7.0 67.25 72.0 89.25 -0.59 0.34 81.25IL18R1 377.0 136.53 508.0 181.10 -0.36 0.33 86.25KRBOX5 324.0 81.78 398.0 99.33 -0.20 0.33 86.25FEM1C 597.0 166.85 764.0 242.94 -0.20 0.32 86.25ENSG00000278627 4.0 9.79 13.0 12.51 -0.78 0.32 86.25THBS4-AS1 48.0 27.61 27.0 25.11 0.52 0.32 88.75ECRG4 16.0 15.97 5.0 11.49 1.02 0.31 86.25ENSG00000215447 120.0 48.23 84.0 43.71 0.32 0.31 82.50LRRC75A 98.0 39.62 68.0 37.10 0.28 0.31 78.75NDUFA8 340.0 102.98 262.0 97.07 0.22 0.30 83.75RPS24P19 15.0 6.14 10.0 6.68 0.31 0.30 82.50COPG2 526.0 150.26 419.0 124.08 0.29 0.30 81.25The small-scale genetic signature found was composed of the 18 most discriminatory genes with an LOOCV predictive accuracy of 88.75%. Next, we evaluated the discriminatory power of these genes to create a transcriptomic signature that could accurately distinguish between these patient groups with a minimal number of genes using ROC analyses. Thus, a predictive model composed of 6 genes was selected to effectively discriminate responders and non-responders, where the results showed high area under the curve (AUC) values usingthe GSE224842 and GSE33377 datasets. Figure 3A and Figure 3B, respectively. Singlegenes showed a low to moderate discriminatory power to distinguish between responders andnon-responders (AUC= 0.44-0.78, Figure 4 and Figure 5). Then, we performed an internalvalidation using the first dataset (GSE138746). ROC analyses were conducted to combine the 6 genes of the model, which showed a high AUC (AUC= 0.81, Figure 3C). Validation phase Example 2.3. Analysis of differential expression genes of the GSE138746 dataset An initial analysis of the RNA-seq results identified 53 differentially expressed genesbetween responders and non-responders, including 28 up-regulated and 25 down-regulatedgenes in responders versus non-responders, according to the criteria: Fold Change > |1| and p-value <0.05. To visualize the differential expression genes, we constructed a volcano-plot(Figure 7A). No significant differences were found between both groups when the p-value was adjusted by the Benjamini–Hochberg adjustment method. In fact, the representation of the heatmap and the Principal Component Analysis (PCA) analysis show a poor separation ofthe samples from R and NR to anti-TNF therapy according to the 53 differentially expressedgenes found by RNA-seq analysis (Figure 7B and 7C, respectively). Functional enrichmentanalysis did not report any significant relation between the analyzed genes and the response(all the adjusted p-values>0.1). Example 2.4. Identification of a highly discriminatory gene setTable 5 shows the list of the most discriminatory genes (calculated with Fisher's RatioSampler) of the phenotypes of the responders patients compared with non-respondersobtained through Fisher’s Ratio Sampler Once. The small-scale genetic signature found wascomposed of the 18 most discriminatory genes (KCNK17, LOC100506235, TRBV6-4, GLS2, MRPL24, DNTTIP1, MYPOP, LPIN3, OR2A9P, GTPBP2, ZFTA, COMTD1, ZFP57, IL18R1, KRBOX5, FEM1C, ENSG00000278627, THBS4-AS1) with an LOOCV predictive accuracy of 88.75%. Table 8. Most discriminatory genes using the Fisher's Ratio Sampler. GVKLTGMDMLOGZCZI . . . . . . . MedianC1 and MedianC2 represent the median values of each group (C1 as the response to anti-TNF treatment and C2 as non-response). FC: Fold change value; FR: Fisher's Ratio value; LOOCV: Mean accuracy in Leave-One-Out-Cross- Validation; Std: standard deviation of each group. Example 2.5. A transcriptomic signature was associated with the prediction of response to anti-TNF therapy Next, we evaluated the discriminatory power of the 18 most discriminatory genes to create atranscriptomic signature capable of accurately distinguishing responders from non-respondersusing ROC analyses. Using a logistic regression model, we selected a predictive model of 7genes (MRPL24, COMTD1, DNTTIP1, GLS2, GTPBP2, IL18R1, and KCNK17) thateffectively discriminated responders and non-responders (AUC=0.84; Figure 8A). Singlegenes showed a low to moderate discriminatory power to distinguish between responders andnon-responders (AUC= 0.43-0.78).Then, we performed a validation using the first dataset (GSE138746), the same dataset we used to establish the genetic small-sale signature. Boxplots for each gene revealed significantdifferences in expression levels between responders and non-responders (p-value<0.05 forGLS2, MRPL24, GTPBP2, and DNTTIP1, and p-value<0.005 for KCNK17, IL18R1, and COMTD1). ROC analysis combining the 7 genes demonstrated a high predictive performance with an AUC=0.949 (Figure 9A). To validate the signature, we applied the model to two independent datasets. The first validation cohort achieved an AUC=0.85,supporting the robustness of the 7-gene model in distinguishing R from NR to anti-TNFtherapy (Figure 9B). The second dataset, the COMBINE cohort, confirmed the predictive value of the signature with an AUC=0.939 (Figure 9C).

Claims

CLAIMS 1. In vitro method for predicting the response of patients suffering from rheumatoidarthritis to a treatment with tumor necrosis factor (TNF) inhibitors, or for classifyingpatients suffering from rheumatoid arthritis into responder or non-responder patients to a treatment with TNF inhibitors, for monitoring patients suffering from rheumatoid arthritis to assess whether they are responding to treatment with TNF inhibitors, or fordeciding or recommending whether to treat patients suffering from rheumatoidarthritis with TNF inhibitors, wherein the method comprises assessing the level ofexpression of the gene combination: MRPL24, COMTD1, DNTTIP1, GLS2,GTPBP2, IL18R1 and KCNK17, in a biological sample obtained from the patient.

2. In vitro method, according to claim 1, wherein the identification of a higherexpression level of the genes: KCNK17 and / or IL18R1, as compared with a pre-established threshold value measured in non-responder patients, is an indication thatthe patient will respond to treatment with TNF inhibitors, or that the patient isresponding to treatment with TNF inhibitors, and / or wherein the identification of ahigher expression level of the genes: GLS2, GTPBP2, DNTTIP1, COMTD1 and / or MRPL24 as compared with a pre-established threshold value measured in non-responder patients, is an indication that the patient will not respond to treatment with TNF inhibitors, or that the patient is not responding to treatment with TNF inhibitors.

3. In vitro use of the gene combination: MRPL24, COMTD1, DNTTIP1, GLS2,GTPBP2, IL18R1 and KCNK17, or of a kit comprising reagents for the identification of the level of expression of the gene combination: MRPL24, COMTD1, DNTTIP1, GLS2, GTPBP2, IL18R1 and KCNK17, for predicting the response of patients suffering from rheumatoid arthritis to a treatment with TNF inhibitors, or for classifying patients suffering from rheumatoid arthritis into responder or non- responder patients to a treatment with TNF inhibitors, or for monitoring patients suffering from rheumatoid arthritis to assess whether they are responding to a treatment with TNF inhibitors, or for deciding or recommending whether to treat patients suffering from rheumatoid arthritis with TNF inhibitors.

4. TNF inhibitor for use in a method of treatment of rheumatoid arthritis, wherein themethod comprises assessing whether the patient would be a responder patient to atreatment with TNF inhibitors by assessing the level of expression of the gene combination: MRPL24, COMTD1, DNTTIP1, GLS2, GTPBP2, IL18R1 and KCNK17, in a biological sample obtained from patients suffering from rheumatoid arthritis, wherein the identification of a higher expression level of the genes: KCNK17and / or IL18R1as compared with a pre-established threshold value measured in non-responder patients, is an indication that the patient will respond to treatment with TNF inhibitors; and / or wherein the identification of a higher expression level of the genes: GLS2, GTPBP2, DNTTIP1, COMTD1 and / or MRPL24 as compared with a pre-established threshold value measured in non-responder patients, is an indication thatthe patient will not respond to treatment with TNF inhibitors, or that the patient is not responding to treatment with TNF inhibitors.

5. In vitro method, according to any of the claims 1 or 2, in vitro use, according to claim3, or TNF inhibitor for use according to claim 4, wherein the biological sample is blood, serum or plasma.

6. In vitro method, according to any of the claims 1 or 2, in vitro use, according to claim3, or TNF inhibitor for use according to claim 4, wherein the TNF inhibitors is selected from: infliximab, etanercept, golimumab, certolizumab and adalimumab.

Images