WO2016075232A1

WO2016075232A1 - Gene signature associated with tolerance to renal allograft

Info

Publication number: WO2016075232A1
Application number: PCT/EP2015/076426
Authority: WO
Inventors: Daniel Baron; Gérard RAMSTEIN; Sophie Brouard; Jean-Paul Soulillou; Magali Giral; Rémi HOULGATTE
Original assignee: Institut National De La Sante Et De La Recherche Medicale; Universite De Nantes; Chu Nantes; Université De Lorraine
Priority date: 2014-11-12
Filing date: 2015-11-12
Publication date: 2016-05-19

Abstract

The present invention concerns a method for the in vitro diagnosis of a graft tolerant or non-tolerant phenotype, comprising determining from a grafted subject biological sample an expression profile comprising the 20 following genes: TCL1A, MZB1, CD22, BLK, MS4A1, CD79B, BLNK, FCRL2, IRF4, ID3, AKR1C3, HINT1, RFC4, ANXA2R, CD40, FCER2, CTLA4, AKIRIN2, EPS15 and PLBD1; comparing the obtained expression profile with at least one reference expression profile, and determining the graft tolerant or graft non-tolerant phenotype from said comparison.

Description

Gene signature associated with tolerance to renal allograft

The present invention concerns methods and tools for identifying patients tolerant to a kidney graft.

Transplantation is the treatment of choice for end-stage renal disease. Recent advances in immunosuppression have improved management of acute rejection and graft survival. However, due to their toxicity, these drugs have numerous deleterious side effects and only a marginal effect on long term rejection. Tolerance is thus increasing regarded as an ideal solution.

It has indeed been observed that some patients can maintain the tolerance for years to their graft without any immunosuppressive treatment, demonstrating that a state of operational tolerance can naturally occur. Current estimates report roughly 100 cases of such natural operational tolerance. However, the real proportion of tolerant grafted subjects is probably underestimated. Indeed, although the possibility to progressively stop the immunosuppressive treatment has never been investigated, a significant proportion of kidney grafted subjects accept their graft with a minimal dose of immunosuppressive drug (cortisone monotherapy <10 mg a day). In addition, among patients developing posttransplantation lymphoproliferative disorders, leading to the interruption of their immunosuppressive treatment, some does not reject their graft. Thus, a significant proportion of kidney grafted subjects might display an unsuspected, total or partial, operational tolerance state to their graft.

There is thus an important need of methods enabling identifying potentially graft tolerant patients, likely to support an immunosuppressive therapy weaning.

Currently, only a biopsy of the grafted kidney allows, through the analysis of the presence or absence of several histological lesion types, for the precise evaluation of said grafted kidney functionality. However, a biopsy is an invasive examination, which is not without danger for the grafted organ, and is thus usually avoided on grafted subjects that have stable biological parameters values. In addition, the variability of the diagnosis, due to the subjectivity of the analysis, is a drawback of the histological examination of biopsies.

A non-invasive accurate and reliable method for identifying potentially graft tolerant patients is thus needed.

Efforts have been devoted by the European and US transplant community to identify non-invasive biomarkers. These studies could report several gene lists, with evidences converging towards the potential implication of B lymphocytes as attested by the identification of numerous B-cell markers (Braud et al. (2008) J Cell Biochem 103: 1681 - 1692; Brouard et al. (2007) Proc Natl Acad Sci U S A 104: 15448-15453; Lozano et al. (201 1 ) Am J Transplant 11 : 1916-1926; Newell et al. (2010) J Clin Invest 120: 1836-1847; Sagoo et al. (2010) J C//n lnvest†20: 1848-1861 ). Although informative, these lists poorly overlapped, raising question about the pertinence of the results across technology, analyses, and investigated cohorts of patients as well as future clinical application.

The present invention arises from the unexpected finding by the inventors that a restricted set of 20 gene markers accurately discriminated tolerant grafted patients from stable grafted patients under immunosuppressive therapy with high sensitivity, specificity and reproducibility, both in the five cohorts used in the previous studies mentioned above and in a new independent cohort. The inventors thus identified a specific restricted set of gene markers which allowed the identification of grafted subject for whom a progressive, total or partial withdrawal of immunosuppressive drugs is possible. This set of gene markers has the advantage to be sufficiently small to enable a simple implementation. Moreover, the diagnosis could be performed from a blood sample, which is completely harmless for the tested grafted subject.

The present invention thus concerns a method for the in vitro diagnosis of a graft tolerant or non-tolerant phenotype, comprising:

a) determining from a grafted subject biological sample an expression profile comprising the 20 following genes: TCL1 A, MZB1 , CD22, BLK, MS4A1 , CD79B, BLNK,

FCRL2, IRF4, ID3, AKR1 C3, HINT1 , RFC4, ANXA2R, CD40, FCER2, CTLA4, AKIRIN2, EPS15 and PLBD1 ,

b) comparing the obtained expression profile with at least one reference expression profile, and

c) determining the graft tolerant or graft non-tolerant phenotype from said comparison.

Detailed description of the invention

Marker genes

The main features of the 20 genes used in the present invention are listed in the following table 1 . Table 1 : Main features of the 20 signature genes of the invention

were available on November 10, 2014.

Phenotype

According to the present invention, a "graft tolerant phenotype" is defined as a state of tolerance of a subject to his/her graft. A "state of tolerance" means that this subject (referred to as a "graft tolerant subject") does not reject his/her graft in the absence of an immunosuppressive treatment with a well-functioning graft.

In contrast, a "graft non-tolerant phenotype" refers to the absence in said subject of a state of tolerance, meaning that said subject (referred to as a "graft non-tolerant subject") would, at the time of the diagnosis, reject its graft if the immunosuppressive treatment was withdrawn.

While the population of graft tolerant subjects only includes subjects in a state of tolerance to their graft, the population of graft non-tolerant subjects thus includes all other subjects and is composed of a variety of different states: patients in acute rejection, patients already suffering from obvious chronic rejection, patients at the early non symptomatic stage of chronic rejection, but also stable patients, which cannot at this time be considered as tolerant but who may later develop a graft tolerant phenotype. Indeed, it must be understood that the mechanisms of tolerance are complex and still not elucidated, and the cellular and molecular processes of tolerance induction may require a prolonged lapse of time. Thus, while the population of graft tolerant subjects only includes subjects who have already reached a stable state of tolerance to their graft, the population of graft non-tolerant subjects is heterogeneous and includes all other subjects, i.e. both subjects in the process of developing acute or chronic rejection and subjects in the process of developing tolerance.

The present invention possesses two major interests:

- first, it permits to diagnose or prognose (i.e. to identify), among patients under immunosuppressive treatment, those who are tolerant to their graft and who could thus benefit from a progressive partial or total withdrawal of the immunosuppressive treatment while remaining tolerant to their graft. Due to the side effects of immunosuppressive treatments, this achievement is really crucial; and

- second, it further permits to diagnose or prognose (i.e. to identify), among patients without immunosuppressive treatment who are diagnosed by the method according to the invention as graft non-tolerant (i.e. patients that are not diagnosed as graft tolerant), those who present a risk of degradation of the graft and who would thus benefit from a reinstatement of an immunosuppressive treatment. Indeed, tolerance is likely not a stable situation for "entire life" and reinstatement of an immunosuppressive treatment may be needed in some cases to prevent acute or chronic rejection. Subject

In a preferred embodiment of the methods of the invention, said subject is a kidney transplanted subject.

According to the invention, a "kidney transplanted subject" is a subject that was grafted with a non-syngeneic, including allogenic or even xenogenic, kidney. Said kidney transplanted subject may further have been grafted with another organ of the same donor providing the kidney. In particular, said kidney transplanted subject may further have been grafted with the pancreas, and optionally a piece of duodenum, of the kidney donor.

Immunosuppressive therapy

Immunosuppressive drugs that may be employed in transplantation procedures include azathioprine, methotrexate, cyclophosphamide, FK-506 (tacrolimus), rapamycin, corticosteroids, and cyclosporins. These drugs may be used in monotherapy or in combination therapies.

In the case of kidney graft, the following immunosuppressive protocols are usually used.

Subjects with primary kidney graft generally receive an induction treatment consisting of 2 injections of basiliximab (Simulect^®, a chimeric murine/human monoclonal anti IL2-Ra antibody commercialized by Novartis), in association with tacrolimus (Prograf™, Fujisawa Pharmaceutical, 0.1 mg/kg/day), mycophenolate mofetil (Cellcept™, Syntex Laboratories, Inc, 2 g/day) and corticoids (1 mg/kg/day), the corticoid treatment being progressively decreased of 10 mg every 5 days until end of treatment, 3 months post transplantation.

Subjects with secondary or tertiary kidney graft, or subjects considered at immunological risk (percentage of anti-T PRA previously peaking above 25% or cold ischemia for more than 36 hours), generally receive a short course of anti-thymocyte globulin (ATG) (7 days), in addition from day 0 with mycophenolate mofetil (Cellcept™, Syntex Laboratories, Inc, 2 g/day), and corticosteroids (1 mg/kg/day), then the steroids are progressively tapered of 10 mg every 5 days until end of treatment and finally stopped around 3 months post transplantation. Tacrolimus (Prograf™, Fujisawa Pharmaceutical) is introduced in a delayed manner (at 6 days) at a dose of 0.1 mg/kg/day.

Biological sample

A "biological sample" may be any sample that may be taken from a grafted subject, such as a serum sample, a plasma sample, a urine sample, a blood sample, a lymph sample, or a biopsy. Such a sample must allow for the determination of an expression profile comprising or consisting of the 20 genes defined in the section "Gene markers". Preferred biological samples for the determination of an expression profile include samples such as a blood sample, a lymph sample, or a biopsy. Preferably, the biological sample is a blood sample, more preferably a peripheral blood sample comprising peripheral blood mononuclear cells (PBMC). Indeed, such a blood sample may be obtained by a completely harmless blood collection from the grafted patient and thus allows for a noninvasive diagnosis of a graft tolerant or non-tolerant phenotype. Expression profile

By "expression profile" is meant a group of at least 20 values corresponding to the expression levels of the 20 genes defined in the section " Marker genes" above, optionally with further other values corresponding to the expression levels of other genes. Preferably, the expression profile consists of a maximum of 200, preferably 100, 75, 50, more preferably 40, 35, 30, 25, even more preferably 20 distinct genes, 20 of which being the 20 genes defined in the section " Marker genes" .

In a most preferred embodiment, the expression profile consists of the 20 genes defined in the section "Marker genes" only, since this expression profile has been demonstrated to be particularly relevant for assessing graft tolerance/non-tolerance.

The expression profile may be determined by any technology known by a person skilled in the art. In particular, each gene expression level may be measured at the genomic and/or nucleic and/or proteic level. In a preferred embodiment, the expression profile is determined by measuring the amount of nucleic acid transcripts of each gene. In another embodiment, the expression profile is determined by measuring the amount of each gene corresponding protein.

The amount of nucleic acid transcripts can be measured by any technology known by a man skilled in the art. In particular, the measure may be carried out directly on an extracted messenger RNA (mRNA) sample, or on retrotranscribed complementary DNA (cDNA) prepared from extracted mRNA by technologies well-known in the art. From the mRNA or cDNA sample, the amount of nucleic acid transcripts may be measured using any technology known by a man skilled in the art, including nucleic microarrays, quantitative PCR, microfluidic cards, and hybridization with a labelled probe.

In a preferred embodiment, the expression profile is determined using quantitative PCR. Quantitative, or real-time, PCR is a well-known and easily available technology for those skilled in the art and does not need a precise description. In a particular embodiment, which should not be considered as limiting the scope of the invention, the determination of the expression profile using quantitative PCR may be performed as follows. Briefly, the real-time PCR reactions are carried out using the TaqMan Universal PCR Master Mix (Applied Biosystems). 6 μΙ cDNA is added to a 9 μΙ PCR mixture containing 7.5 μΙ TaqMan Universal PCR Master Mix, 0.75 μΙ of a 20X mixture of probe and primers and 0.75 μΙ water. The reaction consisted of one initiating step of 2 min at 50°C, followed by 10 min at 95°C, and 40 cycles of amplification including 15 sec at 95°C and 1 min at 60°C. The reaction and data acquisition can be performed using the ABI PRISM 7900 Sequence Detection System (Applied Biosystems). The number of template transcript molecules in a sample is determined by recording the amplification cycle in the exponential phase (cycle threshold or C_T), at which time the fluorescence signal can be detected above background fluorescence. Thus, the starting number of template transcript molecules is inversely related to C_T.

In another preferred embodiment, the expression profile is determined by the use of a nucleic microarray. Preferably, the expression profile is determined by the use of the nucleic microarray of the invention, as defined in the section "Nucleic microarray" below.

The amount of gene corresponding protein can be measured by any technology known by a man skilled in the art, for example by employing antibody-based detection methods such as immunohistochemistry, enzyme-linked immunosorbent assay or western blot analysis, protein microarray, flow cytometry or flow lateral dipstick.

In particular, the expression profile may be determined by the use of a protein microarray. In particular, antibodies, aptamers, or affibodies microarrays can be used, more particularly antibodies microarrays. The antibodies, aptamers, or affibodies are attached to various supports using various attachment methods, using a contact or non- contact spotter. Examples of suitable supports include glass and silicon microscope slides, nitrocellulose, microwells (for instance made of a silicon elastomer). For detection, two main technologies can be used: 1 ) direct labelling, single capture assays and 2) dual- antibody sandwich immunoassays. In direct labelling, single capture assays, proteins contained in one or more samples are labelled with distinct labels (generally fluorescent or radioisotope labels), hybridized to the microarray, and labelled hybridized proteins are directly detected. In dual-antibody sandwich immunoassays, the sample is hybridized to the microarray, and a secondary tagged antibody is added. A third labelled (generally fluorescent or radioisotope label) antibody specific for the tag of the secondary antibody is then used for detection). Comparison step

The determination of the presence of a graft tolerant or graft non-tolerant phenotype is carried out thanks to the comparison of the obtained expression profile with at least one reference expression profile in step (b).

A "reference expression profile" is a predetermined expression profile, obtained from a biological sample from a subject with a known particular graft state. In particular embodiments, the reference expression profile used for comparison with the test sample in step (b) may have been obtained from a biological sample from a graft tolerant subject ("tolerant reference expression profile"), and/or from a biological sample from a graft non- tolerant subject ("non-tolerant reference expression profile"). Preferably, a non-tolerant expression profile is an expression profile of a long-term stable grafted subject under classical immunosuppressive therapy.

Preferably, at least one reference expression profile is a tolerant reference expression profile. Alternatively, at least one reference expression profile may be a non- tolerant reference expression profile. More preferably, the determination of the presence or absence of a graft tolerant phenotype is carried out by comparison with at least one tolerant and at least one non-tolerant reference expression profiles. The diagnosis (or prognostic) may thus be performed using one tolerant reference expression profile and one non-tolerant reference expression profile. Advantageously, to get a stronger diagnosis, said diagnosis is carried out using several tolerant reference expression profiles and several non-tolerant reference expression profiles.

The comparison of a tested subject expression profile with said reference expression profiles can be done using the PLS regression (Partial Least Square) which aim is to extract components, which are linear combinations of the explanatory variables (the genes), in order to model the variable response (eg: 0 if STA, 1 if TOL). The PLS regression is particularly relevant to give prediction in the case of small reference samples. The comparison may also be performed using PAM (predictive analysis of microarrays) statistical method. A non supervised PAM 3 classes statistical analysis can thus be performed. Briefly, tolerant reference expression profiles, non-tolerant reference expression profiles, and the expression profile of the tested subject are subjected to a clustering analysis using non supervised PAM 3 classes statistical analysis. Based on this clustering, a cross validation (CV) probability may be calculated (CV,₀|), which represents the probability that the tested subject is tolerant. In the same manner, another cross validation probability may be calculated (CV_non._t0|), which represents the probability that the tested subject is non-tolerant. The diagnosis is then performed based on the CV,₀i and/or CVnon-toi probabilities. Preferably, a subject is diagnosed as a tolerant subject if the CV,₀i probability is of at least 0.5, at least 0.6, at least 0.7, at least 0.75, at least 0.80, at least 0.85, more preferably at least 0.90, at least 0.95, at least 0.97, at least 0.98, at least 0.99, or even 1 .00, and the CV_non-toi probability is of at most 0.5, at most 0.4, at most 0.3, at most 0.25, at most 0.20, at most 0.15, at most 0.10, at most 0.05, at most 0.03, at most 0.02, at most 0.01 , or even 0.00. Otherwise, said subject is diagnosed as a graft non- tolerant subject.

Typically, when compared to at least one non-tolerant reference expression profile, the expression profile of a graft tolerant phenotype is as follows: the levels of expression of the genes TCL1 A, MZB1 , CD22, BLK, MS4A1 , CD79B, BLNK, FCRL2, IRF4, ID3, AKR1 C3, HINT1 , RFC4, ANXA2R, CD40, FCER2 and CTLA4 are increased and the levels of expression of the genes AKIRIN2, EPS15 and PLBD1 are decreased. Preferably, the levels of expression of the genes are respectively significantly increased or decreased. Additional parameters useful for the diagnosis

In particular embodiments of the methods according to the invention, said methods may further comprise determining from a biological sample of the subject at least one additional parameter useful for the diagnosis. Such "parameters useful for the diagnosis" are parameters that cannot be used alone for a diagnosis but that have been described as displaying significantly different values between tolerant grafted subjects and subjects in chronic or acute rejection and may thus also be used to refine and/or confirm the diagnosis according to the above described method according to the invention. They may notably be selected from:

- standard biological parameters specific for said subject grafted organ type, - phenotypic analyses of peripheral blood mononuclear cells (PBMC), and

- qualitative and/or quantitative analysis of PBMC immune repertoire.

As defined herein, "standard biological parameters specific for said subject grafted organ type" means biological parameters that are usually used by clinicians to monitor the stability of grafted subjects status and to detect graft rejection. These standard biological parameters specific for said subject grafted organ type usually comprise serum or plasma concentrations of particular proteins, which vary depending on the grafted organ type. However, these standard biological parameters specific for said subject grafted organ type are, for each organ type, well known of those skilled in the art.

For instance, standard biological parameters specific for kidney include serum or plasma urea and creatinine concentrations. In a healthy subject, the serum creatinine concentration is usually comprised between 40 to 80 μηιοΙ/Ι for a woman and 60 to 100 μηιοΙ/Ι for a man, and the serum urea concentration between 4 to 7 mmol/l.

For instance, for liver transplantation, standard biological parameters include serum or plasma concentrations of gamma glutamyl transpeptidase (GGT), aspartate aminotransferase (AST), alanine aminotransferase (ALT), lactate dehydrogenase (LDH), and bilirubin (total or conjugated).

These standard biological parameters have the advantage of being easily measurable from a blood sample, but are not sufficient to establish a precise graft tolerant or non-tolerant diagnosis. However, when combined with the determination of an expression profile according to the present invention, the resulting method according to the invention makes it possible to detect graft tolerant subject whose immunosuppressive treatment could be progressively decreased.

The phenotypic analyses of peripheral blood mononuclear cells (PBMC) may comprise various types of phenotypic analysis. In particular they may comprise:

- measuring the percentage of CD4⁺ CD25⁺ T cells in peripheral blood lymphocytes, which may be performed by any technology known in the art, in particular by flow cytometry using labelled antibodies specific for the CD4 and CD25 molecules. Preferably, the percentage of CD4⁺ CD25⁺ T cells in peripheral blood lymphocytes of a tolerant subject is not statistically different from that of a healthy volunteer, whereas it is significantly lower (p < 0.05) in a non-tolerant grafted subject.

- determining the cytokine expression profile of T cells, which may be performed using any technology known in the art, including quantitative PCR and flow cytometry analysis. Preferably, the oligoclonal νβ families of a non-tolerant grafted subject express increased levels compared to a healthy volunteer of Th1 or Th2 effector molecules, including interleukin 2 (IL-2), interleukin 8 (IL-8), interleukin 10 (IL-10), interleukin 13 (IL- 13), transforming growth factor beta (TGF-β), interferon gamma (IFN-γ) and perforin, whereas oligoclonal νβ families of a tolerant grafted subject do not express increased levels of those effector molecules compared to a healthy volunteer.

The analysis of PBMC immune repertoire consists advantageously in the qualitative and quantitative analysis of the T cell repertoire, such as the T cell repertoire oligoclonality and the level of TCR transcripts or genes.

The T cell repertoire oligoclonality may be determined by any technology enabling to quantify the alteration of a subject T cell repertoire diversity compared to a control repertoire. Usually, said alteration of a subject T cell repertoire diversity compared to a control repertoire is determined by quantifying the alteration of T cell receptors (TCR) complementary determining region 3 (CDR3) size distributions. In a healthy subject, who can be considered as a control repertoire, such a TCR CDR3 size distribution displays a Gaussian form, which may be altered in the presence of clonal expansions due to immune response, or when the T cell repertoire diversity is limited and reaches oligoclonality.

The level of TCR expression at the genomic, transcriptional or protein level is preferably determined independently for each νβ family by any technology known in the art. For instance, the level of TCR transcripts of a particular νβ family may be determined by calculating the ratio between these νβ transcripts and the transcripts of a control housekeeping gene, such as the HPRT gene. Preferably, in a graft tolerant subject, a significant percentage of νβ families display an increase in their transcript numbers compared to a normal healthy subject.

An example of methods to analyze T cell repertoire oligoclonality and/or the level of TCR transcripts, as well as scientific background relative to T cell repertoire, is clearly and extensively described in WO 02/084567. Preferably, a graft tolerant subject, as well as a subject in chronic or acute rejection, displays a T cell repertoire with a significantly higher oligoclonality than a normal healthy subject.

Such additional parameters may be used to confirm the diagnosis obtained using the expression profile of the invention. For instance, when the subject is a kidney grafted subject, certain values of the standard biological parameters may confirm a graft non- tolerant diagnosis: if the serum concentration of urea is superior to 7 mmol/l or the serum concentration of creatinine is superior to 80 μηιοΙ/Ι for a female subject or 100 μηιοΙ/Ι for a male subject, then the tested subject is diagnosed as not tolerant to his/her graft.

Nucleic microarray

Another object of the invention concerns a nucleic acid microarray comprising nucleic acids specific for the 20 following genes: TCL1 A, MZB1 , CD22, BLK, MS4A1 ,

CD79B, BLNK, FCRL2, IRF4, ID3, AKR1 C3, HINT1 , RFC4, ANXA2R, CD40, FCER2,

CTLA4, AKIRIN2, EPS15 and PLBD1 .

According to the invention, a "nucleic microarray" consists of different nucleic acid probes that are attached to a substrate, which can be a microchip, a glass slide or a microsphere-sized bead. A microchip may be constituted of polymers, plastics, resins, polysaccharides, silica or silica-based materials, carbon, metals, inorganic glasses, or nitrocellulose. Probes can be nucleic acids such as cDNAs ("cDNA microarray") or oligonucleotides ("oligonucleotide microarray"), and the oligonucleotides may be about 25 to about 60 base pairs or less in length.

To determine the expression profile of a target nucleic sample, said sample is labelled, contacted with the microarray in hybridization conditions, leading to the formation of complexes between target nucleic acids that are complementary to probe sequences attached to the microarray surface. The presence of labelled hybridized complexes is then detected. Many variants of the microarray hybridization technology are available to the man skilled in the art, such as those described in patents or patent applications US 5,143,854; US 5,288,644; US 5,324,633; US 5,432,049; US 5,470,710; US 5,492,806; US 5,503,980; US 5,510,270; US 5,525,464; US 5,547,839; US 5,580,732; US 5,661 ,028; US 5,800,992; WO 95/21265; WO 96/31622; WO 97/10365; WO 97/27317; EP 373 203; and EP 785 280.

In a preferred embodiment, the nucleic microarray is an oligonucleotide microarray comprising, or consisting of, 20 oligonucleotides specific for the 20 genes defined in the section "Marker genes" above. Preferably, the oligonucleotides are about 50 bases in length.

Suitable microarray oligonucleotides specific for the 20 genes defined in the section "Marker genes" above may be designed, based on the genomic sequences of these genes (defined in Table 1 above), using any method of microarray oligonucleotide design known in the art. In particular, any available software developed for the design of microarray oligonucleotides may be used, such as, for instance, the OligoArray software (available at http://berry.engin.umich.edu/Oligoarray/), the GoArrays software (available at http://www.isima.fr/bioinfo/goarrays/), the Array Designer software (available at http://www.premierbiosoft.com/dnamicroarray/index.html), the Primer3 software (available at http://frodo.wi.mit.edu/primer3/primer3_code.html), or the Promide software (available at http://oligos.molgen.mpg.de/).

Kit

The present invention also relates to a kit for the in vitro diagnosis of a graft tolerant phenotype, comprising at least one reagent for the determination of an expression profile comprising the 20 following genes: TCL1 A, MZB1 , CD22, BLK, MS4A1 , CD79B, BLNK, FCRL2, IRF4, ID3, AKR1 C3, HINT1 , RFC4, ANXA2R, CD40, FCER2, CTLA4, AKIRIN2, EPS15 and PLBD1 .

By "a reagent for the determination of an expression profile" is meant a reagent which specifically allows for the determination of said expression profile, i.e. a reagent specifically intended for the specific determination of the expression level of the genes comprised in the expression profile. This definition excludes generic reagents useful for the determination of the expression level of any gene, such as taq polymerase or an amplification buffer, although such reagents may also be included in a kit according to the invention. Such a kit for the in vitro diagnosis of a graft tolerant or graft non-tolerant phenotype may further comprise instructions for determination of the presence or absence of a graft tolerant phenotype.

Such a kit for the in vitro diagnosis of a graft tolerant phenotype may also further comprise at least one reagent for the determining of at least one additional parameter useful for the diagnosis such as standard biological parameters specific for said subject grafted organ type, phenotypic analyses of PBMC (notably the percentage of CD4⁺ CD25⁺ T cells in peripheral blood lymphocytes and the cytokine expression profile of T cells), and quantitative and/or qualitative analysis of PBMC immune repertoire (such as the T cell repertoire oligoclonality and the level of TCR transcripts).

In any kit for the in vitro diagnosis of a graft tolerant phenotype according to the invention, the reagent(s) for the determination of an expression profile comprising, or consisting of, the 20 genes defined in the section "Marker genes" above, preferably include specific amplification primers and/or probes for the specific quantitative amplification of transcripts of the genes TCL1 A, MZB1 , CD22, BLK, MS4A1 , CD79B, BLNK, FCRL2, IRF4, ID3, AKR1 C3, HINT1 , RFC4, ANXA2R, CD40, FCER2, CTLA4, AKIRIN2, EPS15 and PLBD1 , and/or a nucleic microarray for the detection of the genes TCL1 A, MZB1 , CD22, BLK, MS4A1 , CD79B, BLNK, FCRL2, IRF4, ID3, AKR1 C3, HINT1 , RFC4, ANXA2R, CD40, FCER2, CTLA4, AKIRIN2, EPS15 and PLBD1 .

In addition, the instructions for the determination of the presence or absence of a graft tolerant phenotype preferably include at least one reference expression profile. In a preferred embodiment, at least one reference expression profile is a graft tolerant expression profile. Alternatively, at least one reference expression profile may be a graft non-tolerant expression profile.

Method of monitoring and method of treatment

The present invention also concerns a method of treatment of a grafted subject, comprising:

(i) determining from a subject biological sample the presence of a graft tolerant or graft non-tolerant phenotype using a method according to the invention, and

(ii) adapting the immunosuppressive treatment in function of the result of step (i). Said adaptation of the immunosuppressive treatment may consist in:

- a reduction or suppression of said immunosuppressive treatment if the subject has been diagnosed as graft tolerant, or

- a beginning or modification of said immunosuppressive treatment if the subject has been diagnosed as graft non-tolerant or as developing a chronic or acute rejection. Accordingly, the present invention also concerns a method for monitoring the suitability of an immunosuppressive treatment or absence of treatment in a kidney transplanted subject, comprising the steps of:

a) determining from a biological sample from the grafted subject an expression profile comprising the 20 following genes: TCL1 A, MZB1 , CD22, BLK, MS4A1 , CD79B,

BLNK, FCRL2, IRF4, ID3, AKR1 C3, HINT1 , RFC4, ANXA2R, CD40, FCER2, CTLA4, AKIRIN2, EPS15 and PLBD1 ,

c) optionally based on the comparison in step b), beginning, continuing or discontinuing an immunosuppressive therapy in said subject.

Preferably, when the comparison in step b) leads to the identification of the subject as a graft tolerant subject, the method further comprises starting an immunosuppressive therapy weaning, in particular if said subject was under immunosuppressive therapy, or continuing the absence of immunosuppressive therapy, in particular if said subject was not under immunosuppressive therapy.

Still preferably, when the comparison in step b) leads to the identification of the subject as a graft non-tolerant subject, the method further comprises starting an immunosuppressive therapy, in particular if said subject was not under immunosuppressive therapy, or continuing, modifying or increasing an immunosuppressive therapy, in particular if said subject was under immunosuppressive therapy.

Accordingly, the present invention further concerns an immunosuppressive therapy, as defined in the section "Immunosuppressive therapy" above for use for the treatment of a kidney transplanted subject identified as a graft non-tolerant subject, comprising identifying the subject as a graft non-tolerant subject by:

a) determining from a biological sample from the grafted subject an expression profile comprising the 20 following genes: TCL1 A, MZB1 , CD22, BLK, MS4A1 , CD79B, BLNK, FCRL2, IRF4, ID3, AKR1 C3, HINT1 , RFC4, ANXA2R, CD40, FCER2, CTLA4, AKIRIN2, EPS15 and PLBD1 ,

c) based on the comparison in step b), identifying the subject as a graft non- tolerant subject.

The present invention also concerns a method for treating a kidney transplanted subject with immunosuppressive therapy comprising the steps of: a) determining from a biological sample from the grafted subject an expression profile comprising the 20 following genes: TCL1 A, MZB1 , CD22, BLK, MS4A1 , CD79B, BLNK, FCRL2, IRF4, ID3, AKR1 C3, HINT1 , RFC4, ANXA2R, CD40, FCER2, CTLA4, AKIRIN2, EPS15 and PLBD1 ,

c) treating the subject with immunosuppressive therapy as defined in the section "Immunosuppressive therapy" above if the comparison in step b) indicates that the subject has a graft non-tolerant phenotype.

Preferably, the immunosuppressive therapy denotes an immunosuppressive composition comprising at least one immunosuppressive drug, such as azathioprine, methotrexate, cyclophosphamide, FK-506 (tacrolimus), rapamycin, a corticosteroid, a cyclosporine, basiliximab, mycophenolate mofetil or antithymocyte globulin; or an immunosuppressive combination comprising at least two immunosuppressive drugs, such as at least two immunosuppressive drugs selected from the group consisting in azathioprine, methotrexate, cyclophosphamide, FK-506 (tacrolimus), rapamycin, a corticosteroid, a cyclosporine, basiliximab, mycophenolate mofetil and antithymocyte globulin.

Another object of the invention concerns a method for identifying a kidney transplanted subject under immunosuppressive therapy as a candidate for immunosuppressive therapy weaning, comprising the steps of:

c) identifying the subject as eligible to immunosuppressive therapy weaning if the comparison in step b) indicates that the subject has a graft tolerant phenotype.

By "immunosuppressive therapy weaning" is meant herein the progressive reduction, and optionally eventually the suppression of an immunosuppressive therapy.

The present invention will be further illustrated by the figures and examples below.

Brief description of the figures Figure 1 : Cross-validation of the meta-signature of the invention.

In panel 1 , the classification performances through the six cross-validation folds described in the example are shown. Results are displayed with a histogram in which each bar (one of the six folds) represents the accuracy obtained on one dataset used as test while learning is performed on the 5 others. From left (first fold) to right (sixth fold), tests were datasets from Braud (A), Brouard (B), Lozano (C), Newell (D), Sagoo_a (European "Indices of Tolerance" (IOT) cohort, E) and Sagoo_b (American "Immune Tolerance Network" (ITN) cohort, F).

In panel 2, the influence of the origin of a dataset on the performances of classification is assessed. The accuracy from the six test datasets (big circle) is compared to the accuracy obtained on comparable test datasets (equal size and same sample composition) constructed by a random selection of tolerant (TOL) and stable (STA) samples from the total pool of samples (whatever the study of origin). Results are depicted by box plots (boxes: interquartile range (IQR); whiskers: 1 .5 x IQR) corresponding each to the values obtained after 1000 repeated random selections. Values beyond the range are considered outliers and shown as small circles

Figure 2: Relative expression of the 20 marker genes of the invention across samples from the meta-analysis described in the example. Values are depicted by histogram for STA (n=31 1 ), CR patients under dialysis (n=4), TOL (n=96) and HV (n=62). Each bar is a mean expression.

Figure 3: Expression of the 20 marker genes of the invention across samples from the experimental validation set described in the example. Values from PCR measurements are shown for HV (n=19), TOL (n=18) and STA (n=30) samples and are depicted by box plots with and arbitrary scale. Boxes correspond to IQR and whiskers denote 1 .5 ^χ IQR; values beyond this range are considered outliers and shown as black dots. P values (results from Student's t-tests) are shown for statistically significant differences.

Brief description of the sequences

SEQ ID NO: Description

1 -2 Nucleotide sequences of TCL1 A transcripts

3 Amino acid sequence of TCL1 A protein

4 Nucleotide sequence of MZB1 transcript

5-9 Amino acid sequences of MZB1 protein isoforms

10-13 Nucleotide sequences of CD22 transcripts

14-15 Amino acid sequences of CD22 protein isoforms 16 Nucleotide sequence of BLK transcript

17 Amino acid sequence of BLK protein

18-20 Nucleotide sequence of CD79B transcripts

21 -22 Amino acid sequences of CD79B protein isoforms

23-24 Nucleotide sequences of MS4A1 transcripts

25 Amino acid sequence of MS4A1 protein

26 Nucleotide sequence of AKR1 C3 transcript

27 Amino acid sequence of AKR1 C3 protein

28-29 Nucleotide sequences of BLNK transcripts

30-32 Amino acid sequences of BLNK protein isoforms

33 Nucleotide sequence of FCRL2 transcript

34-37 Amino acid sequences of FCRL2 protein isoforms

38 Nucleotide sequence of ID3 transcript

39 Amino acid sequence of ID3 protein

40 Nucleotide sequence of IRF4 transcript

41 -42 Amino acid sequences of IRF4 protein isoforms

43-45 Nucleotide sequences of HINT1 transcripts

46 Amino acid sequence of HINT1 protein

47-48 Nucleotide sequences of RFC4 transcripts

49 Amino acid sequence of RFC4 protein

50 Nucleotide sequence of ANXA2R transcript

51 Amino acid sequence of ANXA2R protein

52 Nucleotide sequence of PLBD1 transcript

53 Amino acid sequence of PLBD1 protein

54 Nucleotide sequence of AKIRIN2 transcript

55 Amino acid sequence of AKIRIN2 protein

56-57 Nucleotide sequences of CTLA4 transcripts

58-62 Amino acid sequences of CTLA4 protein isoforms

63 Nucleotide sequence of FCER2 transcript

64 Amino acid sequences of FCER2 protein

65-66 Nucleotide sequences of CD40 transcripts

67-68 Amino acid sequences of CD40 protein isoforms

69-70 Nucleotide sequences of EPS15 transcripts

71 -72 Amino acid sequences of EPS15 protein isoforms

Example

This example describes the identification of the expression profile of the graft tolerant phenotype and the confirmation of its potent use as a diagnosis marker of graft tolerant or non-tolerant subjects.

Material and methods

Data collection

The data used in this study were published in Braud et al. (2008) J Cell Biochem 103: 1681 -1692; Brouard et al. (2007) Proc Natl Acad Sci U S A 104: 15448-15453; Lozano et al. (201 1 ) Am J Transplant 11 : 1916-1926; Newell et al. (2010) J Clin Invest 120: 1836-1847 and Sagoo et al. (2010) J Clin Invest 120: 1848-1861 and are publicly available. The five microarray datasets and related information on samples were retrieved from the GEO database. They are referred to by the first author of the original publication and include studies from Braud (GSE47755) (Braud et al. (2008) J Cell Biochem 103: 1681 -1692), Brouard (GSE47683) (Brouard et al. (2007) Proc Natl Acad Sci U S A 104: 15448-15453), Lozano (GSE22707) (Lozano et al. (201 1 ) Am J Transplant 11 : 1916- 1926), Newell (GSE22229) (Newell et al. (2010) J Clin Invest 120: 1836-1847) and Sagoo (GSE14655) (Sagoo et al. (2010) J Clin Invest 20: 1848-1861 ) (Table 2).

Table 2: Summary of the five tolerance related studies used for the meta-analysis

For each study, the numbers of samples in each group - healthy volunteer (HV), tolerant (TOL), minimally immunosuppressed (MIS), stable under classical treatment (STA), chronic rejection (CR), acute rejection (AR) - are given (brackets: number of hybridizations). Technical information on the microarray platform (GPL ID) is also provided (a: two-channel; b: single-channel; c: dedicated; d: whole genome). Study from Sagoo comprises two independent cohorts (EU IOT: "Indices of Tolerance"; US ITN: "Immune Tolerance Network") sponsored respectively.

Altogether 596 samples were available (equivalent to 932 distinct hybridizations) gathering 62 samples from healthy volunteers (HV), 96 samples corresponding to 50 unique operationally tolerant kidney recipients (TOL), 32 samples from recipients under minimal immunosuppressive treatment (MIS), 31 1 samples from long-term stable recipients under classical immunosuppressive therapy (STA), 81 samples from patients with chronic rejection (CR) and 14 samples from patients undergoing acute rejection (AR). Some of the TOL patients were assessed in more than one study: according to extremely good inter-study correlations and as most of these samples were collected at different time points and processed on different platforms, they were thus analyzed as unique samples. The clinical definition of the patient groups was described in Braud et al. (2008) J Cell Biochem 103: 1681 -1692; Brouard et al. (2007) Proc Natl Acad Sci U S A 104: 15448-15453; Lozano et al. (201 1 ) Am J Transplant 11 : 1916-1926; Newell et al. (2010) J Clin Invest 120: 1836-1847 and Sagoo et al. (2010) J Clin Invest 120: 1848-1861 . Meta-analysis

To identify a gene signature of tolerance, the comparison was focused on the TOL group (n=96) and patients with stable graft function (n=343) either under standard (STA, n=31 1 ) or minimal immunotherapy (MIS, n=32). To this end, the inventors performed two types of meta-analyses. The first captures in each individual dataset the clusters of differential genes between the two groups and identifies the overlap as a consensus gene set. The second relies on the integration of the different datasets as a single corpus of data and identifies, after an analysis similar to the one performed on the individual datasets, the clusters of differentially expressed genes. Reprocessing, integration and analysis

Reprocessing: To ensure efficient removal of non-linear effects and scaling of the samples, the datasets were normalized using a LOWESS (locally weighted scatterplot smoothing) normalisation procedure. To be applicable to any type of datasets (mono and two-channels related studies), the inventors used the method described by Workman et al. (2002) Genome Biol 3: research0048. This method, based on a "channel by channel" procedure, uses for a specified dataset, a prototype sample (the median profile of all the samples) to normalize the samples analyzed. The LOWESS regression thus corrects the linear biases forcing array distributions to have the same central tendency and non linear biases locally fitting the distribution curve of the signal values to that of the median profile. In the present work, for one-channel studies (Lozano, Newell and Sagoo), the regression was thus performed on raw data (Sagoo) or Robust Multichip Average (RMA) pre- processed data (Newell, Lozano). For dual-channel studies (Braud and Brouard), the normalisation step was performed on raw data on each channel separately before performing ratios (i.e. sample vs. reference).

Information on genes was retrieved with MADGene to convert probes and match the genes between the different datasets. Signals from probes corresponding to the same gene (often technical replicates) were averaged. Similarly, duplicated hybridizations (technical replicates) performed on the same individual sample were averaged. Besides, data were log₂ transformed and median centered on genes, so that relative variations rather than absolute values were used for interpretation

Integration: Integration of heterogeneous datasets is especially a problem of data scales and distributions and variation due to probe effects is larger than the variation due to arrays. For that reason, for each dataset, to ensure that all genes lie within the same dynamic range, the inventors applied a per gene standardization to ensure that all genes lie within the same dynamic range (same mean, same variance). This location (mean)- scale (variance) adjustment of the genes is one of the generally advisable methods performing well to remove experiment effects and it is assumed that this transformation, while trivially making data more comparable, do not remove any biological signal of interest. One can also note that as the Pearson correlations between gene profiles are not impacted by this linear scaling, this natural transformation is also commonly used for classification of gene expression data. This conveniently fits the analysis of the meta- dataset using clustering (hierarchical and K-means) techniques. As the per gene distribution (mean and variance) could be artificially shifted regarding the different proportion of samples (especially tolerant and stable) from the different studies, the inventors used the standardization method proposed by Wang et al. (2006) Cancer Inform 2: 87-97. For a gene, instead of using measures from all samples, this approach relies on the utilization of a reference group. For the present study, the stable group (ST A) was chosen as reference as being the most represented in all studies.

For each dataset, standardization of gene expression was thus performed according to the following formula: g_iiS = [/W,,_s -

where g_iiS denotes the standardized expression measurements of gene /^' in sample S; M_iiS is the (log₂ transformed) expression level of gene /^' in sample S before being standardized; M_I:(STA₎ is the mean expression of gene /^' across STA samples; and SD_I:(STA₎ is the standard deviation of gene /^' computed across STA samples. After standardization, expression measurements from STA samples were normally distributed with mean_{fSL J} = 0 and SDfSTa; = 1 . Merging was performed on the consensus set of 1846 genes being present in each of the five examined datasets.

Analysis: Hierarchical clustering was performed to investigate relations between gene expression profiles and samples with the Cluster program (Eisen et al. (1998) Proc Natl Acad Sci U S A 95: 14863-14868). The clustering method employed was an average linkage with the uncentered correlation as a similarity metric. It was applied to the individual datasets (log₂ median centered data) and the consensus set (standardized data). Results were displayed (heatmaps and dendrograms) using the TreeView program.

Individualization of gene clusters and analysis

Individualization: A non-hierarchical K-means algorithm implemented in the

Cluster program was used to partition the datasets. The maximum number of iterations to reach stability was set to 1000 and the number of nodes was fixed to 10. The evaluation of this parameter was based on an empirical estimation of the number of clusters from hierarchical classification of the data sets, and also from a silhouette analysis of the same datasets. This estimation fits the mean number of clusters (n=10.5) generally observed in a dataset. This led to the individualization of 10 homogeneous clusters (K1 to K10) for each dataset displayed with TreeView. Besides, to investigate differences between the two considered groups (TOL and ST A), each individualized cluster was further restrained to TOL and STA samples and further classified using two-way (genes-against-samples) average-linkage hierarchical clustering with correlation distance.

Comparison of clusters: To identify a consensus gene set, the 50 clusters from the five studies were systematically compared to each other (pairwise comparisons). Considering the pool of common genes between two studies, the resemblance of two clusters (one cluster A from the first study and the other cluster B from the second study) was measured using the normalized intersection " as similarity metric. It is calculated following the formula: l=[2x(AnB)/(A B)] where "A" and "B" are the sizes of the two gene list A and B respectively, "AnB" denotes the intersection of the two lists and "A B" their union. The significance of the intersection was thus assessed using the Fisher's exact test.

Functional and biological interpretation: Functional annotation of the clusters was performed by Gene Ontology analysis (GOA) using Gene Ontology (GO) (Ashburner et al. (2000) Nat Genet 25: 25-29) and the GoMiner program (Zeeberg et al. (2003) Genome Biol 4: R28) to identify significant bias in GO terms. As the genes to be used as the reference set includes only the pool of genes assessed, the inventors used for each study the collection of genes represented on their related microarray platform as recommended by GoMiner developers. To account for multiple testing problems, reported q-values for each GO term, corresponding to FDR adjusted p-values, were used for the analysis. Only GO terms with q<0.01 were kept for interpretation. To gain biological insight into the clusters, the inventors also used the gene set analysis (GSA) approach (Subramanian et al. (2005) Proc Natl Acad Sci U S A 102: 15545-15550) to identify among a large collection of gene sets (MSigDB) those with a similar gene composition. Analyses were performed on the version 4 of the database using the collections C2 (curated gene sets) and C5 (GO gene sets). Finally, to highlight cell-specific gene subsets, the inventors performed a virtual microdissection analysis (VMDA) (Alizadeh et al. (2001 ) J Pathol 195: 41 -52), according to the fact that immune cell types exhibit specific transcriptional profiles, and that expression profiles in blood-based samples are strongly dependent on the predominant constituent cell types. Molecular assignment was thus achieved as shown in Alizadeh et al. (2000) Nature 403: 503-51 1 , confronting the gene expression profiles to those of tissues samples and cell populations from different lineages and at different stages of cell differentiation.

Statistics: Considering the two groups of patients, TOL and STA, the inventors tested the differential expression of each cluster (K1 to K10) on its own based on the dependence (i.e. correlation) assumption between genes. To this end, the inventors reduced the cluster to a meta-gene that is defined here as the median vector of its individual genes. This essentially shifts the level of analysis of the microarray experiment from thousands of single genes to 10 related meta-genes (M1 to M10). Significance was measured for each meta-gene by a Student's t-test, giving one p-value for the cluster. In addition, the ability of these genes to discriminate samples from the two groups (TOL and STA) was measured by analyzing the composition of the main separation on the sample dendrogram. Significance was calculated by Fisher's exact test as described in Feuerstein et al. (2012) PLoS One 7: e40449. For individual datasets, the threshold for the two tests was set to p<0.01. As statistical power deeply increased with the addition of samples, a more conservative threshold of p<0.00001 was used for the meta-dataset.

Reproducibility of the signature in external datasets

Data retrieval : In order to assess the functional and cellular components of the signature with external data, the inventors mined a large collection of public microarray studies stemming from the public repository GEO (Gene Expression Omnibus, NCBI, Bethesda, MD) (Barrett et al. (201 1 ) Nucleic Acids Res 39: D1005-D1010). Microarray platforms with a good coverage of the gene set (more than 85%) and a large utilization (more than 50 datasets) were retained. This filter retained five platforms (GPL570, GPL571 , GPL96, GPL201 and GPL8300) accounting for 46.3% of the available human transcriptome data and corresponding to 4658 GSE series. In this collection, the inventors systematically identified trends of expression in terms of recurrent co-expression and rank analysis.

Correlations: According to the conservation of pairs of co-expressed genes across studies, the inventors identified for each of the 595 genes the 200 best neighbors of co-expression (absolute counts) across the GSE datasets. The resulting matrix M(i,j) was a Boolean matrix that indicated that a gene i (one of the 8224 genes) is a neighbor of j (one of the 595 genes). This matrix was represented by a graph and a hierarchical clustering was used as a layout algorithm to highlight the structure of this graph, to filter and to reduce the number of visible elements, and to provide a condensed representation of strongly connected components (clusters). The intra-cluster connectivity corresponds to the proportion of edges inside the cluster. Connectivity corresponds to the number of distinct edges (or paths) that exist between each pair of genes. Accordingly, pairs of connected genes were iteratively joined to form dense nodes equivalent to clusters. Density of connected genes, defined as intra-cluster density, was then used as the cluster fitness measure and ranged from 0 [isolated genes] to 1 [fully connected genes]. Clusters with more than 2 genes and a density higher than a preset threshold of 0.5 were retained resulting in 284 good clusters gathering 1462 genes. Results were displayed as a network using Cytoscape (Shannon et al. (2003) Genome Res 13: 2498-2504). Each vertex of the graph corresponds to a cluster and its size is proportional to the number of genes it contains (3 to 101 genes). Edges represent inter-cluster densities greater or equal to 0.2. The graph was manually split into 6 meta-clusters which were interpreted using the plug-in BiNGO (Maere et al. (2005) Bioinformatics 21 : 3448-3449) to assess over-representation of GO categories in the biological network.

Rank: According to gene patterns supported by rank-based differences between two biological situations (Feng et al. (2009) BMC Genomics 10: 41 1 ), the inventors also tried to discover coordinated gene expression comparable to the one observed in tolerance. To this end, samples from each GEO study were preprocessed using rank- based normalization (Tsodikov et al. (2002) Bioinformatics 18: 251 -260). For each dataset, all pairs of samples c and d (i.e. all possible combinations) were considered. Let one call p_c and p_d their respective profile formed by the expression values of a set G of genes. Let one define G as the reference pattern of 251 genes identified by student's t- test as the most differentially expressed (p<0.005) between the TOL and STA groups of patients. G comprises two subsets A ('positive') and B ('negative') related to the 168 genes over-expressed and the 83 genes under-expressed in the TOL group respectively. One can thus compute for each pair (c,d) the vector V=p_d- p_c representing the difference of their profiles. Assuming sample c and d were close to STA and TOL samples respectively, the resulting vector V should possess positive (A) and negative (B) values and should tend to fit the reference pattern G=A+B. The area under the ROC curve (AUC) was then used to measure for each computed vector V, how well genes were related to their corresponding binary labels (A and B). For each dataset, the pair (c,d) with the best AUC value was retained among which those having a value greater than 0.80 were selected. This threshold is statistically highly significant and corresponds to a q-value of 10^~11 to observe such an AUC value when the labels are given at random. The resulting 215 pairs (c,d) were used to create the expression matrix on the set G of genes. For each pair (c,d), data were log₂ transformed and median centered. Results were displayed by a heat-map using Treeview, except for 7 genes with more than 20% missing values and which were removed from the visualization. In addition, a text mining approach was applied on the 215 datasets to identify, in titles and summaries, significant bias (Fisher's exact test) of key key-word frequencies compared to the rest of the datasets.

Cross-validation of the signature and classification performances

Gene Selection: In microarray analysis, gene selection is a crucial step for increasing the performances of classifiers. The inventors used the T-test to rank genes according to their p-value and keep the top ranked genes.

Choice of the classifier: As there is not a unique emerging classification method, the selection of a classifier is essential for prediction accuracy. Two advanced techniques are widely used in bioinformatics: SVM (Support Vector machine) and Random Forest. The inventors compared these two advanced techniques (SVM and Random Forest) with the more basic KNN (k-nearest neighbor) approach. This comparison was performed on the meta-dataset. For each classifier, a ten-fold cross-validation was performed with different selections of genes (n=20, 50, 100 or 200). For a fair comparison, the inventors used the same gene selection method for all classifiers: the "n" most discriminative genes were selected by a Student's t-test. Thanks to this analysis, SVM emerged as the best performer in the present situation: it outperformed KNN and Random Forest in term of accuracy, whatever the size of the gene list.

SVM classifier: Support Vector machine (SVM), also known as maximum margin classifier, is a supervised machine learning technique. SVM was retained for the analysis of the classification performances. The inventors thus defined the positive samples as those belonging to the TOL class and the negative as those belonging to the STA class of patients. Basically, SVM maps input data points to construct maximal-marginal hyperplanes in higher dimensional space to classify data with the two class labels TOL and STA. The hyperplane is constructed using only the support vectors (i.e., data that lie on the margin) simultaneously minimizing the empirical classification error and maximizing the geometric margin. In addition, to solve the problems of data imbalance (none equilibrated effectives in the considered classes), the inventors used an ensemble of under-sampled SVMs based on data resampling and a majority vote for decisions on the 100 folds.

Performance evaluation: The inventors imposed the independence between the learning procedure and the test through cross-validation. They first reduced all samples to the expression values relative to the gene selection. They then iteratively partitioned the dataset into two parts: samples from five studies were used as a whole for learning the SVM model; samples from the sixth study were predicted using the classifier. For each study, the inventors then obtained classification accuracy and other common performance metrics: sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) (as disclosed in Table 3).

Table 3: Confusion matrix (2x2 contingency table) and common performance metrics calculated from it

The mean accuracy determined the fitness of this gene selection. To evaluate the influence of a missing dataset on the learning procedures, the performances of classification obtained on test datasets were also compared to the ones obtained on 1000 equivalent datasets (same size, same sample composition) made of TOL and STA samples randomly picked from the different studies. Independent confirmation of the results by real time PCR using a new collection of samples

Selection of patients: The study was approved by the University Hospital Ethical Committee and the Committee for the Protection of Patients from Biological Risks. All participating patients gave written informed consent. Transplanted kidney recipients were recruited from a multi-centric French cohort centralized in Nantes Hospital. Selected patients (disclosed in Table 4) included 18 tolerant (TOL) and 30 stable patients under classical immunosuppressive regimen (STA).

Table 4: Clinical data of HV, TOL and STA patients

Their selection was performed on the basis of the same criterion of eligibility defined for the microarray studies. Briefly, from the 18 tolerant cases, 12 recipients, analyzed in at least one of the five microarray studies, were randomly picked from the cohort of 27 cases described in (Brouard et al. (2012) Am. J. Transplant. 12:3296-3307). They correspond to patients with a stable kidney graft function (creatinemia <150 μηιοΙ/Ι and proteinuria <1 g/24 h) in the absence of immunosuppression for at least 1 year. The 6 new cases, not analyzed in the microarray studies, have been included to the aforementioned cohort of tolerant recipients following the same criteria of inclusion. The 30 stable recipients were randomly selected from a large cohort of 131 transplant recipients (Garrigue et al. (2014) Transplantation 97: 168-175) who had received a first and unique kidney transplant from a deceased donor and displayed a stable graft function (creatinemia <150 μηηοΙ/Ι, proteinuria <1 g/24 h and GFR>40 ml/min) under standard immunosuppression (tacrolimus or cyclosporine A for maintenance therapy) for at least 5 years. For comparison, 19 healthy volunteers (HV) presenting normal blood formula and no infectious or other concomitant pathology for at least 6 months before the study, were also used as controls.

Corresponding samples and related information were retrieved from the DIVAT biocollection (Donnees Informatisees et Validees en Transplantation www.divat.fr, Inserm N8 02555, N8CNIL 891735). These samples consisted in Peripheral Blood Mononuclear Cells (PBMC) fractions separated from blood samples on a Ficoll layer (Eurobio, Les Ulis, France) and frozen at -80°C.

RNA preparation and Reverse-transcription: Total RNA was prepared from frozen PBMC samples using the Trizol method (Invitrogen, Cergy Pontoise, France) according to the manufacturer's instructions. RNA was quantified using a NanoDrop microvolume ND-1000 spectrophotometer (Thermo Scientific, Courtaboeuf, France). The integrity of the RNA samples was assessed using an Agilent 2100 Bioanalyzer (Agilent Technologies, Massy, France) with the Eukaryote Total RNA Nano assay. The RNA samples were stored at -80°C until needed.

Amplification and synthesis of complementary DNA (cDNA) from 1 μg total RNA in a 20 μΙ_ reaction volume was carried out using the Quantitect Whole Transcriptome Kit (Qiagen, Hilden, Germany) as per the manufacturer's instructions. Following reverse- transcription, cDNA was stored at -20°C.

Real Time Polymerase Chain Reaction (PCR): Expression levels of the 20 top genes linked to tolerance were measured using custom-made TaqMan Low Density Arrays (TLDA) in addition to controls (Applied Biosystems, Foster City, CA). The cDNA (9 ng) was mixed with 2x TaqMan Universal PCR Master Mix II (Applied Biosystems) with Uracil-N glycosylase (UNG), loaded on a TLDA card, and run on an ABI-Prism 7900 HT Sequence Detection System (Applied Biosystems) as per the manufacturer's instructions.

Analysis: For each sample data were then collected in SDS (Sequence Detection System) files, which were thereafter uploaded into RQ (Relative Quantification) Manager software (Applied Biosystems) for automated data analysis. For each gene, this program provides values known as the cycle threshold (Ct) ranging from 1 to 40 and unexpressed values labelled as undetermined. Values were treated and normalized against a set of several housekeeping genes as described in Livak et al. (2001 ) Methods 25: 402-408 and Schmittgen et al. (2008) Nat Protoc 3: 1 101 -1 108. For each gene, the significance of the change between groups (TOL vs. STA, TOL vs. HV and STA vs. HV) was evaluated using a student's t-test. Besides, the values of the 20 genes were used to classify samples using a leave one out strategy. Results

Comparison of gene lists from the different datasets failed to identify a robust gene signature of tolerance:

A k-means clustering identified 10 clusters (K1 to K10) in each study with clear functional annotations. A total of 19 clusters out of the 50 were found significant between TOL and STA with at least one of the two tests (Student and/or Fisher, p<0.01 ). The 50 clusters had relevant but limited similarities. Accordingly, a focus on the 19 clusters indicated that most of the differential genes do not replicate in another studies. Only 0.74% (14 genes: APOM, ARHGAP17, AURKB, IGBP1 , IL10RA, IL15RA, IL1 RL1 , INSM1 , IRF4, MAOA, MICB, SMAD3, TK1 , YPEL2) was in fact commonly identified across the five studies. Two of them (TK1 , IRF4) were present either in the footprint of 49 genes from Brouard or the list of the 30 top genes from Newell. These 14 genes did not accurately discriminate TOL from STA samples (Table 5).

Table 5: Performances of classification of the set of 14 best genes from the comparison of gene lists

Mean^ mean performances

Mean₂: overall performance mean

Sagoo_A: EU IOT cohort

Sagoo_B: US ITN cohort

Altogether, these data show that comparison of gene lists identified limited similarities with less than 1 % of genes common to the five studies. Such comparisons are thus not enough to derive a robust meta-signature.

Meta-analysis by integration of the datasets identified a statistically and functionally relevant gene signature of tolerance:

To ensure comparability across studies, the 1846 common genes were retained. This selection is highly skewed toward immune functions, but it covers as high as 65% of the biological functions from the gene ontology. After standardization, datasets were merged and analyzed as a single corpus of data comprising 596 samples and available in the Gene Expression Omnibus (GEO) repository (# GSE49198). Hierarchical clustering highlighted clear profiles correlated to groups of samples. To identify the ones associated to tolerance, the inventors further analyzed the 10 K-means clusters: K1 (224 genes) linked to proliferation, K2 (183 genes) to endocytosis and K10 (188 genes) to lymphocyte (B and T) activation and differentiation, were found as the most differential (p<0.00001) between TOL (n=96) and STA (n=343). They thus defined a highly discriminative (p=5.05E-15) 595 gene signature as measured by Fisher's exact test on the contingencies of the sample tree. It comprised 8 (APOM, AURKB, IGBP1 , IL10RA, IL1 RL1 ,IRF4,SMAD3,TK1 ) out of the 14 genes identified by comparing lists of genes and also 13 B-cell molecules (AFF3, BLK, BLNK, CD22, CD79B, FCER2, FCRL2, ID3, IGKC, IGLL1 , MS4A1 , MZB1 , TCL1 A) belonging to the top ranked gene lists from the different studies. Altogether, these data show that integrative meta-analysis can identify a robust gene meta-signature of tolerance.

Gene set analysis (GSA) corroborated functionality and revealed the possible involvement of specific cell populations in tolerance:

A screening of the compendium of gene sets retained the 100 best hits for each cluster (K1 , K2 and K10). Although significant (p<0.00001), these overlaps were partial (coverage: mean=8.89±4.82%, max=31 .15%) but displayed good functional congruence: the 100 bests hits converged to cell proliferation and cell adhesion for K1 , inflammatory response for K2 and lymphocyte activation for K10. Virtual microdissection analysis (VMDA) revealed the clear participation of B, CD4 T lymphocytes and monocytes in tolerance:

The TOL signature (K1 , K2 and K10) was compared to clusters from various tissue and blood cell samples. This comparison identified K1 as a proliferation cluster (e.g. AURKB, CCNB2, CDC20, CHEK1 , NEK2, PLK4) gathering 67% (p=6.31E-13) and 73% (p= 1.04E-13) respectively of the genes from proliferating tissues (e.g. testis, skin) or cells (early blood precursors). K2 and K10 gathered 90% (p=2.30E-20) of immune tissue markers (e.g. bone marrow, thymus, spleen, lymph nodes) showing their immunological specificity. K2 contained 82% (p=3.44E-85) of the granulocyte/monocyte lineage markers and corresponded to a CD14 monocyte cluster (e.g. CD14, CD163, CD68, ITGAM, ITGB2, PECAM1 ). K10 contained 86% {p=5. 12E-40) of the T lineage makers preferentially expressed in na^'ive and differentiated CD4 subsets (e.g. CD247, CD28, CD48, CD5, LAT, MAL). It also harboured 71 % (p=9.31E-10) of the B lineage markers expressed in na^'ive and differentiated subsets (e.g. BLNK, CD22, CD40, CD79B, FCER2, and MS4A1 ). Altogether these data reflect specific expansion and differentiation of B and CD4 T cells, whereas genes associated to CD14 monocyte functions were down- regulated, suggesting the involvement of different cell subsets in tolerance. Generalization of the results using external data confirmed the biological relevance (functionality and cellularity) of the tolerance gene signature and its reproducibility:

To assess the reproducibility of the signature with external data, similar trends of expression were searched in a collection of datasets using co-expression and rank based approaches. A clustering approach showed that recurrent partners of co-expression fell into 6 meta-clusters that were functionally reliant. The two most significant (M3 and M6) gathered 169 (p= 1. 1E-84) and 62 (p= 1. 1E-55) genes involved in immune and cell cycle functions. In the second analysis, 215 distinct pairs of samples significantly (AUC <0.8; Q- value<10^~11) harboured the same pattern in term of rank differences. A text-mining analysis revealed that 70% (p=3.40E- 14) of the matched studies were related to blood or other cell subsets (mononuclear cells: p=3.45E-10; lymphocytes: p=2.88E-07; B-cells: p=6.29E-03) analyses.

Altogether these data show that, across a wide range of conditions, genes from the signature are recurrently associated with the same functionally related neighbours of co- expression and display a TOL-like pattern in numerous blood cell related studies.

Cross-validation analysis yielded good prediction performances and the use of the 20 top ranked biomarkers enabled the experimental validation of the tolerance gene signature in new samples and new patients

To statistically validate the signature, a cross-validation (Figure 1 ) and an experimental validation were performed. A full 6-fold cross-validation was performed on the set of 1846 genes (Figure 1 ). Each initial dataset was used as external validation set while the five other served as training sets for the selection of the top discriminating genes (paneh).

A limited selection of 20 markers, mostly centred on B-cells and over-expressed in TOL (Figure 2), was sufficient to accurately discriminate TOL from STA samples (Table 6).

Table 6: Performances of classification of the set of 20 best genes from the meta-analysis

Sagooe

CV3 Braud- Lozano 83.3 66.7 100 100 75.0

Brouard-

Newell-

Sagoo_A-

Sagooe

CV4 Braud- Newell 84.8 73.7 92.6 87.5 83.3

Brouard-

Lozano-

Sagoo_A-

Sagooe

CV5 Braud- SagooA 87.8 70.0 91 .5 63.6 93.5

Brouard- Lozano- Newell-Sagooe

CV6 Braud- Sagoo B 89.6 90.9 88.9 80.0 95.2

Brouard- Lozano- Newell-Sagoo_A

Mean_! 86.6 69.9 93.4 79.0 86.6

Mean₂ 88.6 69.8 94.0 77.0 91.5

Mean^ mean performances

Mean₂: overall performance mean

Sagoo_A: EU IOT cohort

Sagoo_B: US ITN cohort

These top 20 gene markers are described in Table 7.

Table 7: Description of the 20 gene signature

Rank Order Symbol P-Value (TOL vs. STA) Fold Change (TOL vs. STA) Cluster

1 TCL1 A 7,5692E-23 3, 1 1 178126 K10

2 MZB1 5,202E-14 2,9855797 K1

3 CD22 1 .2589E-13 2,51545618 K10

4 BLK 1 ,7491 E-12 2, 16831984 K10

5 CD79B 3,5564E-12 2,3627036 K10

6 MS4A1 4,9599E-12 1 ,98528044 K10

7 AKR1 C3 6,0545E-12 2,412271 19 K1

8 BLNK 2.0513E-1 1 2,58295032 K10

9 FCRL2 3.004E-1 1 2,32472077 K10

1 1 I D3 6,9274E-1 1 2,46193333 K1

12 IRF4 2,4877E-10 1 ,74549718 K1

13 HINT1 2,7565E-10 1 ,98029813 K1

14 RFC4 1 .4926E-09 2,01222979 K1

15 ANXA2R 1 .7898E-09 1 ,74738186 K10

16 PLBD1 1 .8022E-09 0,49317839 K2

18 AKIRIN2 2,3252E-09 0,54130304

20 CTLA4 3,388E-09 1 ,77909461

21 FCER2 5,479E-09 1 ,94696909 K10

22 CD40 9,9968E-09 1 ,616941 13 K10

23 EPS15 1 .0662E-08 0,50854019 K2 Their quantitative measurement by real-time PCR (independent technique) on a new collection of 67 samples (Figure 3) yielded strong discrimination of TOL and STA (p=4.39E-9) samples: 16 markers (80%) were indeed altered (p<0.05) in the same sense. Leave-one-out prediction yielded excellent reproducibility (100% recall) on the 12 new time samples from already analyzed TOL cases and good external validation on the 6 new TOL cases (83.3% recall). 91 .7% of good classification was achieved (94.4% sensitivity, 90% specificity, 85% PPV, 96.4% NPV), one TOL and three STA being misclassified (Table 8).

Table 8: Classification performances from the experimental validation set

Altogether these data show that the signature could be revalidated whatever the study, the technology and the sample provided.

A "healthy" profile of the 20 top ranked biomarkers in blood from healthy volunteers

Of the 20 biomarkers, TOL and HV were closely related as no (0 gene, p<0.001) or minor difference (3 genes, p<0.05) could be detected in meta-analysis (Figure 2) and RT- PCR set (Figure 3). This similarity between TOL and HV was reinforced by the observation that only 68 genes out of the 1846 analyzed were differential (p<0.001). Conversely, 18 of the markers (90%) also displayed differential expression between STA and HV, both in the meta-analysis {p<0.001) and in the RT-PCR {p<0.05) sets (Figures 2 and 3). Altogether, these data show that TOL and HV display roughly the same "healthy" profile.

Conclusion

The present inventors defined a gene signature thanks to the meta-analysis of blood transcriptome studies comparing TOL group with the more related group of STA patients. This group of patients was chosen because they represent the most appropriate cohort to look at tolerance markers to identify the patients who may benefit from an IS weaning protocol in the future.

Interestingly, the inventors found that most of the genes differentially expressed in TOL vs. STA were unchanged between TOL vs. HV both through the 20 top ranked genes and the whole meta-analysis. To ascertain that these results are not the only reflect of the absence of treatment, the inventors also look at these markers in patients with chronic rejection and off IS available in study from Brouard. As these patients harbour a highly differential profile from TOL, they reasonably exclude an effect of the absence of treatment (Figure 2).

To assess the reliability of the signature, the inventors first performed a full cross- validation procedure. This analysis yielded good predictions and enabled to validate a selection of the top-20 markers, mostly centred on B cells, as accurately discriminating tolerant from stable recipients. In a second step, these 20 markers were experimentally revalidated in an independent cohort of new TOL samples, from which 6 corresponded to new cases. These results showed that the initial findings of the inventors were not dependant on the technology used or the analyzed set of samples. Both analyses yielded good prediction performances (more than 90%). Hence, the biomarkers of the invention could be reliably used to detect tolerance and stratify kidney recipients in clinics. First they may help for a better follow-up of the tolerant patients. Indeed, several lines of evidences suggest that tolerance is likely not a stable situation for "entire life" for most of the studied cases. In such situations, immunotherapy could be reinstated before degradation of the graft. Second, these biomarkers may help to monitor kidney transplanted recipients under classical IS. Such stable cases, presenting low risk of rejection, would thus be highly eligible for progressive IS weaning. In the present meta-analysis, 3% (8 cases) of the STA patients did express this signature and further examination of their clinical status revealed they were still stable without degradation of renal function, years after the test. This result agreed previous observations ranking from 3.5% to 15%.

Claims

1. Method for the in vitro diagnosis of a graft tolerant or non-tolerant phenotype, comprising:

a) determining from a grafted subject biological sample an expression profile comprising the 20 following genes: TCL1 A, MZB1 , CD22, BLK, MS4A1 , CD79B, BLNK, FCRL2, IRF4, ID3, AKR1 C3, HINT1 , RFC4, ANXA2R, CD40, FCER2, CTLA4, AKIRIN2, EPS15 and PLBD1 ,

2. The method according to claim 1 , wherein the obtained expression profile is compared to at least one reference expression tolerant and/or non-tolerant profile in step b).

3. The method according to claim 1 or 2, wherein the expression profile is determined by measuring the amount of nucleic acid transcripts of said gene(s).

4. The method according to claim 3, wherein the expression profile is determined using quantitative PCR or a nucleic microarray.

5. The method according to claim 1 or 2, wherein the expression profile is determined using a nucleic microarray or a protein microarray.

6. The method according to any one of claims 1 to 5, wherein the biological sample is a blood sample.

7. The method according to any one of claims 1 to 6, wherein the subject is a kidney transplanted subject.

8. The method according to any one of claims 1 to 7, further comprising determining at least one additional parameter selected from standard biological parameters specific for said subjects grafted organ type, phenotypic analyses of peripheral blood mononuclear cells (PBMC) and qualitative and/or quantitative analysis of PBMC immune repertoire.

9. A kit for the in vitro diagnosis of a graft tolerant phenotype, comprising at least one reagent for the determination of an expression profile comprising the 20 following genes: TCL1 A, MZB1 , CD22, BLK, MS4A1 , CD79B, BLNK, FCRL2, IRF4, ID3, AKR1 C3, HINT1 , RFC4, ANXA2R, CD40, FCER2, CTLA4, AKIRIN2, EPS15 and PLBD1 .

10. A nucleic acid microarray comprising nucleic acids specific for the 20 following genes: TCL1 A, MZB1 , CD22, BLK, MS4A1 , CD79B, BLNK, FCRL2, IRF4, ID3, AKR1 C3, HINT1 , RFC4, ANXA2R, CD40, FCER2, CTLA4, AKIRIN2, EPS15 and PLBD1 .

11. The nucleic acid microarray according to claim 10, which is an oligonucleotide microarray.

12. Immunosuppressive therapy for use for the treatment of a kidney transplanted subject identified as a graft non-tolerant subject, comprising identifying the subject as a graft non- tolerant subject by:

13. A method for treating a kidney transplanted subject with immunosuppressive therapy comprising the steps of:

c) treating the subject with immunosuppressive therapy if the comparison in step b) indicates that the subject has a graft non-tolerant phenotype.

14. A method for identifying a kidney transplanted subject under immunosuppressive therapy as a candidate for immunosuppressive therapy weaning, comprising the steps of: a) determining from a biological sample from the grafted subject an expression profile comprising the 20 following genes: TCL1 A, MZB1 , CD22, BLK, MS4A1 , CD79B, BLNK, FCRL2, IRF4, ID3, AKR1 C3, HINT1 , RFC4, ANXA2R, CD40, FCER2, CTLA4, AKIRIN2, EPS15 and PLBD1 ,

15. A method for monitoring the suitability of an immunosuppressive treatment or absence of treatment in a kidney transplanted subject, comprising the steps of:

BLNK, FCRL2, IRF4, ID3, AKR1 C3, HINT1 , RFC4, ANXA2R, CD40, FCER2, CTLA4, AKIRIN2, EPS15 and PLBD1 , and

b) comparing the obtained expression profile with at least one reference expression profile.