A method to improve outcomes by identifying patients with difficult-to-treat osteosarcoma at the time of diagnosis and providing new treatment options.

Abstract

The present invention provides a tool for stratifying patients suffering from osteosarcoma to identify those with tumors with poor prognosis.The present invention also relates to the use of inhibitors of the PPARγ pathway as antitumor treatments to improve the overall survival of patients with osteosarcoma with poor prognosis.

Description

[Technical Field]

[0001] Field of Invention This invention relates to the field of anti-cancer therapy. In particular, this invention relates to the prognosis of osteosarcoma in individuals, and provides an in vitro method for determining this prognosis, as well as a diagnostic kit for carrying out this method. This specification also describes the use of PPARγ pathway inhibitors as an antitumor therapy to improve overall survival in patients with osteosarcoma who have a poor prognosis. [Background technology]

[0002] Background of the Invention Osteosarcoma is the most common primary bone cancer in adolescents and young adults, and as a result of multiple chromosomal rearrangements, the genomic and transcriptome landscape is highly heterogeneous. 1,2 Such heterogeneity makes it a complex exploratory study to determine the major oncogenic drivers underlying disease onset or progression, and routinely used prognostic biomarkers or robust stratification have not yet been defined. As a result, there have been no major changes in treatment over the past 40 years, and one-third of osteosarcoma patients experience treatment failure, primarily with metastatic recurrence. 3-7 Over the past 20 years, comparative genome hybridization arrays (CGH) 8-12 From whole exome (WES) 1,2,13-18 Or genome sequencing (WGS) 17-19 The focus has been on the genetic exploration of diseases using various DNA analysis techniques. These have reported numerous genetic events (primarily copy number variations, CNVs) associated with increased mild / moderate tumor fitness without clear tumor stratification.

[0003] Recently, despite common phenotypic features, primary osteosarcoma has been shown to be highly polyclonal under near-neutral selection. 20These observations suggest that the major phenotypic osteosarcoma trait may be an urgent feature stemming from heterogeneous genomic rearrangements in the polyclonal community, and / or transcriptional reprogramming. In other words, progression is a consequence of tumor plasticity.

[0004] Furthermore, in the highly constrained bone environment of osteosarcoma, it is difficult to evaluate the co-evolution of tumor cells with respect to tumor microenvironment (TME) composition, stromal cell ratio, or activation state. 21,22 . [Overview of the Initiative]

[0005] Summary of the Invention The inventors hypothesized that, in order to better understand the underlying mechanisms of osteosarcoma and enhance our equipment for this disease, the transcriptome landscape could overcome the genetic complexity of the disease and provide more interpretable information. Using unsupervised machine learning strategies, they defined a repertoire of genetic components describing osteosarcoma tumor clones and TMEs. They observed that component interactions mediated by co-expression stratified the cohort into tumors with good prognosis and those with poor prognosis.

[0006] Functional characterization of the constituents links angiogenesis, osteoclast activity, and adipogenic activity between tumors with a good prognosis and tumors with specific congenital immune expression and poor prognosis, with each group having distinct CNVs. These distinct functional characteristics can be used to stratify treatment in osteosarcoma, for example, immunomodulation (e.g., mifamultide) for G1 tumors (good prognosis) and anti-osteoclast and anti-angiogenic therapies for G2 (difficult to treat) tumors.

[0007] The inventors have also identified underlying biological pathways for osteosarcoma, including PPARγ, piRNA, or CTA.

[0008] Finally, the researchers identified a panel of 37 genes involved in osteosarcoma and confirmed the predictive power of 15 genes for minimal prognostic features.

[0009] According to a first aspect, the present invention relates to an in vitro method for determining the prognosis of osteosarcoma in an individual, the method comprising the steps of: measuring the expression levels of a collection of signature genes from a biological sample taken from the individual; applying the measured expression levels to a predictive model that associates the expression levels of the collection of signature genes with the outcome of osteosarcoma; and evaluating the outcome of the predictive model to determine the prognosis of osteosarcoma in the individual.

[0010] The present invention also relates to a diagnostic kit for predicting the progression of osteosarcoma in a subject by measuring the expression levels of a collection of signature genes from a biological sample taken from an individual.

[0011] Another object of the present invention is the use of PPARγ inhibitors, particularly PPARγ antagonists, for treating patients with a poor prognosis for osteosarcoma.

[0012] In particular, the present invention provides an in vitro method for determining the prognosis of osteosarcoma in an individual, wherein (a) a collection of signature genes derived from a biological sample taken from the individual is measured, and the collection of signature genes is AMZ2P1, ASXL2, BIRC6, C11orf58, C1orf53, C5orf28, CCDC34, DCUN1D2, DNMT3A, ECI1, EIF3M, ESCO1, FAM35DP, GAGE12D, GAGE12G, HADHA, HIST2H2AA3, HIST2H2AA4, MAPK1, MEAF6, (b) comprising at least two genes selected from the group consisting of MYO1B, MYOM3, NDP, PAIP1, PCID2, PEG10, PEX26, POLR2C, PRR14L, RALGAPA2, SLC2A6, SLC7A4, ST7L, THAP9-AS1, TIMP3, TTPAL, and WNT4; (b) applying the expression levels measured in step (a) to a predictive model that correlates the expression levels of the signature gene collection with the outcome of osteosarcoma; and (c) evaluating the outcome of the predictive model to determine the prognosis of osteosarcoma in the individual. In some embodiments, the signature gene collection includes CCDC34 and MEAF6. In some embodiments, the signature gene collection includes at least five, preferably at least seven, and more preferably at least eight genes selected from the group consisting of AMZ2P1, ASXL2, BIRC6, C11orf58, C1orf53, C5orf28, CCDC34, DCUN1D2, DNMT3A, ECI1, EIF3M, ESCO1, FAM35DP, GAGE12D, GAGE12G, HADHA, HIST2H2AA3, HIST2H2AA4, MAPK1, MEAF6, MYO1B, MYOM3, NDP, PAIP1, PCID2, PEG10, PEX26, POLR2C, PRR14L, RALGAPA2, SLC2A6, SLC7A4, ST7L, THAP9-AS1, TIMP3, TTPAL, and WNT4. In some embodiments, the signature gene collection includes AMZ2P1, C5orf28, CCDC34, GAGE12D, MEAF6, SLC2A6, SLC7A4, and THAP9-AS1.In some embodiments, the signature gene collection includes at least 10, preferably at least 12, more preferably at least 15 genes selected from the group consisting of AMZ2P1, ASXL2, BIRC6, C11orf58, C1orf53, C5orf28, CCDC34, DCUN1D2, DNMT3A, ECI1, EIF3M, ESCO1, FAM35DP, GAGE12D, GAGE12G, HADHA, HIST2H2AA3, HIST2H2AA4, MAPK1, MEAF6, MYO1B, MYOM3, NDP, PAIP1, PCID2, PEG10, PEX26, POLR2C, PRR14L, RALGAPA2, SLC2A6, SLC7A4, ST7L, THAP9-AS1, TIMP3, TTPAL, and WNT4. In some embodiments, the signature gene collection includes AMZ2P1, C5orf28, CCDC34, ESCO1, GAGE12D, HADHA, MEAF6, RALGAPA2, SLC2A6, SLC7A4, THAP9-AS1, and TIMP3. In some embodiments, the signature gene collection includes AMZ2P1, C5orf28, CCDC34, ESCO1, FAM35DP, GAGE12D, HADHA, HIST2H2AA3, HIST2H2AA4, MAPK1, MEAF6, POLR2C, RALGAPA2, SLC2A6, SLC7A4, ST7L, THAP9-AS1, TIMP3, and TTPAL.

[0013] In some embodiments, the expression levels of the signature gene collection are measured at the time of diagnosis. In some embodiments, the expression levels of the signature gene collection are measured at the time of relapse.

[0014] In some embodiments, the gene expression level of the signature gene is combined with one or more other parameters to predict the progression of osteosarcoma in the individual. In some embodiments, one or more other parameters are selected from the group consisting of drugs administered to the patient, age, tumor height and stage at diagnosis, presence or absence of metastasis at diagnosis, and any combination thereof.

[0015] In some embodiments, the biological sample is an osteosarcoma biopsy sample derived from an individual.

[0016] In some embodiments, the method further includes developing the predictive model using stability selection. In some embodiments, the method further includes developing the predictive model using logistic regression. In some embodiments, the method further includes developing the predictive model by selecting genes using stability selection by elastic net normalized logistic regression.

[0017] The present invention also relates to a diagnostic kit for predicting the progression of osteosarcoma in a subject, the kit comprising at least one nucleic acid probe or oligonucleotide that can be used in a method for measuring the expression levels of a collection of signature genes derived from a biological sample taken from the subject, the collection of signature genes being AMZ2P1, ASXL2, BIRC6, C11orf58, C1orf53, C5orf28, CCDC34, DCUN1D2, D It contains at least two genes selected from the group consisting of NMT3A, ECI1, EIF3M, ESCO1, FAM35DP, GAGE12D, HADHA, HIST2H2AA3, HIST2H2AA4, MAPK1, MEAF6, MYO1B, MYOM3, NDP, PAIP1, PCID2, PEG10, PEX26, POLR2C, PRR14L, RALGAPA2, SLC2A6, SLC7A4, ST7L, THAP9-AS1, TIMP3, TTPAL, and WNT4.

[0018] Another aspect of the present invention is the use of a PPARγ inhibitor, particularly a PPARγ antagonist, as an antitumor treatment in osteosarcoma patients. In some embodiments, the subject is a teenager or young adult. In some embodiments, the subject is identified as a poor responder to chemotherapy by the method described above. In some embodiments, the inhibitor is used in combination with another antitumor treatment. In some embodiments, the PPARγ inhibitor is T0070907 (CAS number 313516-66-4).

Brief Description of the Drawings

[0019] [Figure 1] A. Overall survival curves of G1 (pink) and G2 (blue) patients. B. Progression-free survival curves of G1 (pink) and G2 (blue) patients. C. Barplot showing the proportion of the model including clinical variables and G1 / G2 stratification for predicting the overall survival period of the cohort patients. D. Principal component analysis of the independent component metagene vectors. The names of the ICs of the most contributing load variables are shown in the plot. Thick dots and transparent dots indicate dead patients and surviving patients, respectively. Dots are colored according to their membership in G1 (pink) or G2 (blue). E. Load of the most contributing IC estimated by partial least squares discriminant analysis to stratify patients between G1 or G2. [Figure 2] Gene expression distribution of each subnetwork and their predicted biological functions. Adjusted P-values calculated by gene set enrichment analysis using the log2 fold change (G1 vs G2) of genes from the network to detect significant enrichment in the subnetwork gene sets. [Figure 3A] A. GISTIC analysis of copy number changes revealed significant association with G1 or G2 tumors. [Figure 3B-3D]B. Overall survival curves for predicted G1 (pink) and G2 (blue) patients from an independent cohort of 82 osteosarcoma patients. C. Overall survival curves for predicted G1 (pink) and G2 (blue) patients using a custom Nanostring panel for 96 samples from OS2006 Cohort D. The bar plot on the left shows the proportion of G1 and G2 tumors in 39 out of 79 patients enrolled in the OS2006 trial, where tumors were submitted for RNA sequencing and relapsed. The bar plot on the right shows the proportion of G1 and G2 tumors collected at relapse from 42 osteosarcoma patients enrolled in the MAPPYACTS trial. [Figure 4A-1] A. Evaluation of cell viability of osteosarcoma cell lines at 48 and 72 hours under pressure with a PPAR antagonist (T0070907-circle dot) and two PPAR agonists (TGZ-square or RGZ-triangle). Concentrations were [0;0.01;0.1;1;5;10;17.5;20;25;30;50 or 100 μmol / L] as determined by MTS assay. At least three independent tests were performed, each with three replicates. Cells derived from normal lung tissue were used as a positive control for PPARγ expression, and β-actin was used as a housekeeping gene control. [Figure 4A-2] A. Evaluation of cell viability of osteosarcoma cell lines at 48 and 72 hours under pressure with a PPAR antagonist (T0070907-circle dot) and two PPAR agonists (TGZ-square or RGZ-triangle). Concentrations were [0;0.01;0.1;1;5;10;17.5;20;25;30;50 or 100 μmol / L] as determined by MTS assay. At least three independent tests were performed, each with three replicates. Cells derived from normal lung tissue were used as a positive control for PPARγ expression, and β-actin was used as a housekeeping gene control. [Figure 4B] B.PPARγ protein expression was measured by Western blotting in osteosarcoma cell lines. [Figure 5] This is a scatter plot showing the contribution of each IC sample and the logarithmic change in CNA, with the largest coefficient in the corresponding regression model. [Figure 6]A. Bar plot showing gene selection based on correlation between RNA-seq and Nanostring expression. Blue bars indicate correlated genes. p-values ​​are less than 10⁻⁵ and selected for Nanostring signature. The five italicized / bold genes correspond to housekeeping genes selected based on low variance in RNA-seq, normalized from Nanostring data B. Progression-free survival curves for predicted G1 (pink) and G2 (blue) patients using a custom Nanostring panel of 96 samples from the OS2006 cohort. [Figure 7] Genes differentially expressed in HOS cells after treatment with a PPARG antagonist or agonist (p-adj < 0.01). [Figure 8] PPARγ expression in osteosarcoma PDX as determined by immunohistochemistry (IHC). [Figure 9] In vivo effects of T0070907 on primary tumors (left) and lung metastases (right) in a paratibial PDX model of osteosarcoma, using MAP-217 alone (top figure) and in combination with methotrexate (bottom figure) (bottom figure). A. Delay (days) until tumor size reaches 2.5 times the initial size in control mice and T0070907-treated mice. B. Metastasis evaluation by IHC (Humain KI67 staining) in control and T0070907-treated mice. C. Tumor growth curves of primary tumors with monotherapy and combination therapy. [Modes for carrying out the invention]

[0020] Detailed description of preferred embodiments Unless otherwise specified, the methods and systems disclosed herein include prior art and apparatus commonly used in the fields of molecular biology, microbiology, protein purification, protein engineering, protein and DNA sequencing, and recombinant DNA, which are known to those skilled in the art and described in numerous publications and references.

[0021] Unless otherwise defined herein, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art. Various scientific terms, including those included herein, are well known and available to those skilled in the art. While some methods and materials similar to or equivalent to those described herein may be used in the implementation or testing of the embodiments disclosed herein, only a limited number of methods and materials are described. The terms defined below are explained in more detail by referring to the entire specification. It should be understood that this disclosure is not limited to the specific methods, protocols, and reagents described, as they may vary depending on the context in which they are used by those skilled in the art.

[0022] The following general definitions are used in this text.

[0023] Where used herein, the singular words “a,” “an,” and “the” include multiple references unless the context clearly indicates otherwise.

[0024] The grammatical equivalents of “nucleic acid sequence,” “expressed nucleic acid,” or the corresponding signature gene, as used in the context of these terms, mean a nucleic acid sequence whose quantity is measured as an indicator of the gene's expression level. A nucleic acid sequence may be a part of a gene, a regulatory sequence, genomic DNA, cDNA, RNA including mRNA and rRNA, or other. Certain embodiments utilize mRNA as the primary target sequence. As outlined herein, a nucleic acid sequence may be a sequence from a sample or a secondary target such as a reaction product, e.g., a PCR amplification product (e.g., “amplicon”). The nucleic acid sequence corresponding to the signature gene may be of any length, under the understanding that longer sequences are more specific. A probe is hybridized to the nucleic acid sequence to determine whether or not the signature gene is expressed in the sample.

[0025] As used herein, the term “contains” means that a named element is included, but other elements (e.g., an unnamed signature gene) may be added and still represent a composition or method within the scope of the claims.

[0026] As used herein, the term “signature gene” means a gene whose expression correlates, either positively or negatively, with disease progression or outcome, or with another predictor of disease progression or outcome. “Characteristic nucleic acid” means, in the case of cDNA, a nucleic acid containing, or corresponding to, the complete or partial sequence of the RNA transcript encoded by the signature gene, or a complement of such a complete or partial sequence. Characteristic proteins are encoded by, or correspond to, the signature genes disclosed.

[0027] The term “recurrence prediction” is used herein to refer to the prediction of the likelihood of osteosarcoma recurrence in patients who do not have apparent residual tumor tissue after treatment. The prediction methods of this disclosure can be used clinically to make treatment decisions by selecting the most appropriate treatment for any particular patient. The prediction methods of this disclosure can also provide a useful tool in predicting whether a patient is likely to respond favorably to treatment regimens such as surgical intervention, chemotherapy with a given drug or in combination with a drug, and / or radiotherapy.

[0028] In this specification, the terms "subject," "individual," or "patient" refer to human subjects.

[0029] In this specification, "osteosarcoma" refers to a type of bone cancer that originates in bone-forming cells. Osteosarcoma most commonly occurs in the legs, but can also occur in the arms, although it can occur in any bone. In very rare cases, it can occur in the soft tissues outside the bone.

[0030] As used herein, the term "antitumor therapy" refers to any treatment for cancer other than surgery. These include chemotherapy, hormone therapy, biological therapy, and radiation therapy.

[0031] As used herein, “to treat,” “to treat,” and “to treat” mean a reduction or improvement in any of the progression, severity, and / or duration of cancer, in particular solid tumors; for example, in osteosarcoma, a reduction in one or more symptoms resulting from the administration of one or more treatments.

[0032] Other definitions are provided below as needed.

[0033] According to a first aspect, the present invention relates to an in vitro method for determining the prognosis of osteosarcoma in a subject, comprising the following steps: a. A step of measuring the expression level of a collection of signature genes taken from the subject, wherein the collection of signature genes includes at least two genes selected from the group consisting of AMZ2P1, ASXL2, BIRC6, C11orf58, C1orf53, C5orf28, CCDC34, DCUN1D2, DNMT3A, ECI1, EIF3M, ESCO1, FAM35DP, GAGE12D, GAGE12G, HADHA, HIST2H2AA3, HIST2H2AA4, MAPK1, MEAF6, MYO1B, MYOM3, NDP, PAIP1, PCID2, PEG10, PEX26, POLR2C, PRR14L, RALGAPA2, SLC2A6, SLC7A4, ST7L, THAP9-AS1, TIMP3, TTPAL, and WNT4; b. Applying the expression levels measured in step (a) to a predictive model that correlates the expression levels of the above signature gene collection with osteosarcoma outcomes; and c. A step of evaluating the output of the above prediction model and determining the prognosis of osteosarcoma in the above subjects.

[0034] In the above method, the predictive model can be developed by any suitable method known to those skilled in the art. Machine learning (ML) approaches have been demonstrated to be useful in medicine and can be advantageously used to construct predictive methods. Examples of such methods are disclosed in the experimental section below. Other examples of computational protocols for identifying and improving prognostic features are described and can be used by those skilled in the art in the context of the present invention. 66-69 (Vey et al., Cancers, 2019; Xia et al., Nature Communications, 2019; Jiang et al., Cell Systems, 2018; Liu et al, 2020; EBioMedicine).

[0035] Predictive models are advantageously developed using data collected from patients known to have osteosarcoma. Those skilled in the art can obtain different predictive models to address different clinical situations. For example, a model could be developed using data collected only from patients at the time of diagnosis, and another model could be developed using data from patients who have undergone surgery for the primary tumor, for instance.

[0036] In some embodiments, gene expression data may be preprocessed by normalization, background correction, and / or batch effect correction. The preprocessed data can then be analyzed for differential gene expression to stratify patients into a good prognosis group (Group 1) and a poor prognosis group (Group 2).

[0037] In some embodiments, to develop a prediction model for the prognosis of osteosarcoma, the probes included in the final model are selected from the entire set of probes using stability selection. In some embodiments, the model is developed using logistic regression, such as elastic net logistic regression. Elastic net regression is a high-dimensional regression method that incorporates both LASSO (L1) and ridge regression (L2) regularization penalties. The exact mixture of penalties (LASSO vs. ridge) is controlled by the parameter α (α = 0 is pure ridge regression, α = 1 is pure LASSO). The degree of regularization is controlled by a single penalty parameter. Both LASSO and ridge regression shrink the model coefficients to zero for non-implemented regression, but LASSO can shrink the coefficients exactly to zero, and thus can effectively perform variable selection. However, LASSO alone tends to randomly select from among correlated predictors, and the addition of a ridge penalty helps prevent this. The ridge penalty is known to shrink the coefficients of correlated predictors towards each other, while LASSO tends to select one of them and discard the others. In some embodiments, the implementation of elastic net logistic regression in the R package "glmnet" can be used.

[0038] The idea behind stability selection is to find "stable" probes that consistently predict recurrence across multiple datasets obtained by "perturbing" the original data. Specifically, the perturbed version of the data is obtained by subsampling m < n subjects (n is the total number of subjects) without replacement. Next, regularized regression (or elastic net in some embodiments) is run on each subsample version of the data to obtain the full regularization path (i.e., the model coefficients as a function of the regularization penalty). The effect of the LASSO penalty is to shrink most of the probe coefficients exactly to zero, and probes with non-zero coefficients (predictors) over a substantial proportion of the subsample versions of the data are considered stable predictors.

[0039] In some embodiments, adjustment parameters can be calibrated using repeated cross-validation (e.g., using the R package for 10x cross-validation) to perform stability selection by elastic net regression. Those skilled in the art will select adjustment parameter α to include as many features as necessary while providing and maintaining good predictions. In some embodiments, stability selection may be performed using a different number of subsamples of data, each subsample having a portion of the total sample size (each having approximately the same ratio of cases and controls as the original) to identify robust predictors for the final model. In some embodiments, gene expression levels are not standardized by standard deviation (the default in glmnet to place all gene features on the same scale) because differential variability in gene expression levels may be biologically important. In some embodiments, such standardization can be performed. In some embodiments, clinical variables such as the presence of metastasis at diagnosis are forcibly included (i.e., unaffected by the elastic net regularization penalty).

[0040] According to the specific embodiment illustrated in the experimental section, the predictive model stratifies patients into two groups: G1 with a good prognosis (3-year OS = 100%) and G2 with a poor prognosis (3-year OS < 70%), requiring closer monitoring.

[0041] Patients classified as G1 may benefit from treatment with immunomodulatory agents (e.g., mifamultide), while patients in G2 may benefit from treatment with polytyrosine kinase inhibitors or monoclonal antibodies, anti-osteoclast agents such as biphosphonates or anti-RANKL, and anti-angiogenic agents such as PPARγ antagonists (as illustrated in Example 2).

[0042] Tables 4-8 in the experimental section below provide parameters for five different models that can be used to carry out the above method.

[0043] In a preferred embodiment, the signature gene collection (or panel) includes at least CCDC34 and MEAF6. Table 7, shown in the experimental section, provides model parameters based on the expression levels of these two genes. Naturally, these are provided as examples and are not limiting. Those skilled in the art can, for example, refine these parameters using different patient cohorts or adapt them to values ​​obtained using different techniques used to measure gene expression levels. Naturally, this applies to all models provided in this application.

[0044] In some embodiments, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, or all of the 37 genes shown in Table 4 can be used in the predictive model. For illustrative purposes only, Table 4 shown in the experimental section provides parameters for a model based on the expression levels of these 37 genes.

[0045] In some embodiments, the model includes at least five, preferably at least seven, and more preferably at least eight, genes from the list in Table 4. In some embodiments, it includes at least ten, preferably at least twelve, and more preferably at least fifteen, genes from the list in Table 4. In some embodiments, two or more of these genes may be selected based on their correlation with recurrence in the training dataset to develop a predictive model. In some embodiments, one or more genes may be selected based on their confidence rank. In some embodiments, two or more genes may be selected based on their predictive power ranking.

[0046] In some embodiments, the signature gene panel includes at least AMZ2P1, C5orf28, CCDC34, GAGE12D, MEAF6, SLC2A6, SLC7A4, and THAP9-AS1. For illustrative purposes only, Table 6, shown in the experimental section, provides parameters for a model based on the expression levels of these eight genes.

[0047] In some embodiments, the signature gene panel includes at least AMZ2P1, ASXL2, C11orf58, C5orf28, CCDC34, ESCO1, GAGE12D, HADHA, MEAF6, PAIP1, RALGAPA2, SLC2A6, SLC7A4, THAP9-AS1, and TIMP3. For illustrative purposes only, Table 5 shown in the experimental section provides parameters for a model based on the expression levels of these 15 genes.

[0048] In some embodiments, the signature gene panel includes at least AMZ2P1, C5orf28, CCDC34, ESCO1, FAM35DP, GAGE12D, HADHA, HISTH2H2A3, HIST2H2A4, MAPK1, MEAF6, POLR2C, RALGAPA2, SLC2A6, SLC7A4, ST7L, THAP9-AS1, TIMP3, and TTPAL. For illustrative purposes only, Table 8 shown in the experimental section provides parameters for a model based on the expression levels of these 20 genes.

[0049] In some embodiments, the signature gene panel includes at least AMZ2P1, C5orf28, CCDC34, ESCO1, GAGE12D, HADHA, MEAF6, RALGAPA2, SLC2A6, SLC7A4, THAP9-AS1, and TIMP3. These 12 genes are present in the models shown in both Tables 5 and 8.

[0050] In any of the above embodiments, the method of the present invention may further include combining the gene expression level of the signature gene with one or more other parameters to predict the progression of osteosarcoma in the subject. Non-limiting examples of such clinical parameters that can be combined with the expression level of the selected gene include the presence and amount of metastasis at diagnosis, the size, height and / or stage of the tumor at diagnosis, drugs administered to the patient, age, and any combination thereof.

[0051] In some embodiments, based on the expression levels and clinical variables of a set of probes determined by stability selection, the predictive model is obtained by fitting a logistic model using elastic net normalized logistic regression. Glmnet addresses the following challenges on a grid of λ values ​​covering the entire range:

[0052]

number

[0053] This resolves the issue. The tuning parameter λ controls the overall intensity of the penalty.

[0054] In some embodiments, instead of selecting a subset of genes from the panel disclosed in Table 4, the model may weight the genes differently in logistic regression. In some embodiments, determining the prognosis of osteosarcoma for an individual involves applying the expression levels of a collection of signature genes to a predictive model, the predictive model including weighting the expression levels according to the stability ranking of the collection of signature genes. In some embodiments, the method includes weighting the expression levels according to the predictive power ranking of the collection of signature genes.

[0055] The logistic regression models described above represent specific methods for obtaining individual scores by combining expression levels and clinical variables. In some embodiments, expression levels are weighted in elastic-net-normalized logistic regression. In some embodiments, expression levels are weighted when using LASSO. Here, weighting refers not to model coefficients (which can be considered weights of expression levels and clinical variables), but to an additional mechanism for differentially explaining the importance of variables in the logistic regression procedure. In this regard, another embodiment involves unweighted logistic regression, i.e., treating all genes equally, and weighted logistic regression, considering weighting by stability selection frequency.

[0056] In some embodiments, various clinical variables (e.g., drugs administered to the patient, age, tumor height and stage at diagnosis, presence of metastasis at diagnosis) are included in the same logistic model along with the signature gene. Coefficients are defined for each variable (gene expression and clinical value). This logistic regression model provides the probability of clinical recurrence, given gene expression scores and clinical variables. This probability is a numerical value between 0 and 1 and indicates the prognosis of the disease for each patient.

[0057] In some embodiments, in addition to identifying coefficients in the predictive model, the disclosure identifies the most useful specificity and sensitivity that the user would like to have for a particular risk probability. Based on the desired specificity and sensitivity levels, the method reports the risk status of each patient. For example, those skilled in the art will find that, considering the specificity and sensitivity of the inventors' model, a patient with a 45% probability of being classified into the good-prognosis responder group (Group 1) is better classified into the poor-prognosis responder group (Group 2) than into Group 1, or vice versa. In other words, more user-friendly criteria can be selected based on a more detailed analysis of further datasets, determining the most practical interpretation of the risk probability depending on how much risk the clinician is willing to have of having false positives or false negatives.

[0058] Individuals suspected of having any of the following bone cancers, including osteosarcoma, bone cancer, or cancers of the bone, joints, and soft tissues, as well as various bone cancers such as pediatric sarcoma, may be evaluated using the methods disclosed. Illustrative cancers that may be evaluated using the methods disclosed include, but are not limited to, osteosarcoma, pediatric sarcoma, and recurrent adult and pediatric bone cancers.

[0059] Exemplary clinical outcomes that can be determined from the models of this disclosure include, for example, responses to specific courses of treatment such as surgical removal of a tumor, radiation, or chemotherapy.

[0060] Those skilled in the art will understand that patient tissue samples containing osteosarcoma cells may be used in the methods of this disclosure, including, but not limited to, those intended to determine the prognosis of the disease. In these embodiments, the expression level of the signature gene can be assessed by evaluating the amount, for example, absolute amount or concentration, of the signature gene product, e.g., the protein and RNA transcript encoded by the signature gene, as well as fragments of the protein and RNA transcript, in the sample obtained from the patient. The sample may, of course, be subjected to various well-known post-collection preparation and preservation techniques (e.g., fixation, preservation, freezing, thawing, homogenization, DNA or RNA extraction, ultrafiltration, concentration, evaporation, centrifugation, etc.) before evaluating the amount of signature gene product in the sample.

[0061] Useful tissue samples for preparing models to determine the prognosis of osteosarcoma include, for example, paraffin and polymer-embedded samples, ethanol-embedded samples, and / or formalin and formaldehyde-embedded tissues, but any suitable sample can be used. In general, nucleic acids isolated from preserved samples can be highly degraded, and the quality of nucleic acid preparations can depend on several factors, including the sample storage period, immobilization technique, and isolation method.

[0062] If necessary, nucleic acid samples containing signature gene sequences are prepared using known techniques. For example, the samples can be treated to lyse cells using known lysis buffers, sonication, electroporation, etc., and, if necessary, can be purified and amplified as understood by those skilled in the art. Furthermore, this reaction can be achieved in various ways as understood by those skilled in the art, and the components of this reaction can be added simultaneously or sequentially, in any order, in the preferred embodiments outlined below. Furthermore, the reaction may include various other reagents that may be useful in the assay. These include reagents such as salts, buffers, neutral proteins, e.g., albumin, and surfactants, which may be used to facilitate optimal hybridization and detection and / or reduce nonspecific or background interactions. Reagents that improve the efficiency of the assay, such as protease inhibitors, nuclease inhibitors, and antimicrobial agents, may also be used, depending on the sample preparation method and purity.

[0063] In some embodiments, the expression levels of signature genes can be determined using biological samples in addition to, or instead of, osteosarcoma tissue. In some embodiments, suitable biological samples include, but are not limited to, blood, urine from a patient or other bodily fluid, exosomes, and circulating tumor cells isolated from circulating tumor nucleic acids.

[0064] According to a particular embodiment, the expression levels of the signature gene collection described above are measured at the time of diagnosis.

[0065] According to another specific embodiment, the expression levels of the signature gene collection described above are measured at the time of relapse.

[0066] In some embodiments, the gene expression level of a signature gene may be measured multiple times. In some embodiments, the dynamics of the expression level may be used in combination with the signature gene expression levels to better predict clinical outcomes. Those skilled in the art will understand that various approaches may be used to combine the effects of the expression levels and dynamics of the signature gene to determine the prognosis of osteosarcoma.

[0067] The methods of this disclosure rely on the detection of differentially expressed genes for expression profiling across heterogeneous tissues. Therefore, the methods rely on profiling genes whose expression is activated at higher or lower levels in specific tissues, for example, in individuals with cancer such as osteosarcoma, compared to their expression in non-cancerous tissues or controls. Gene expression can be activated at high or low levels at different stages under the same conditions, and differentially expressed genes can be activated or inhibited at the nucleic acid or protein level.

[0068] Differential signature gene expression can be identified or confirmed using methods known in the art, such as qRT-PCR (quantitative reverse transcription polymerase chain reaction) and microarray analysis. In certain embodiments, differential signature gene expression can be identified or confirmed using microarray technology or any similar technology (e.g., NanoString technology). Thus, signature genes can be measured using microarray technology in either fresh or paraffin-embedded tumor tissue. In this method, the polynucleotide sequence of interest is plated or arrayed on a microchip substrate. The array sequence is then hybridized with a specific DNA probe derived from the cells or tissue of interest.

[0069] The expression level of a signature gene in a tissue sample can be determined by contacting a set of probes with nucleic acid molecules derived from the tissue sample, under conditions that the fully complementary probes form a hybridization complex with the nucleic acid sequence corresponding to the signature gene, each probe containing at least two universal priming sites and a signature gene target-specific sequence; the probes forming the hybridization complex are amplified to produce an amplicon; and the detection of the amplicon indicates the presence of the nucleic acid sequence corresponding to the signature gene in the tissue sample; and the expression level of the signature gene is determined.

[0070] The expression levels of nucleic acid sequences corresponding to a set of signature genes in a tissue sample are determined by contacting nucleic acid molecules from the tissue sample with a pair of probes under conditions in which complementary probes form a hybridization complex with signature gene-specific nucleic acid sequences containing at least two universal priming sites and signature gene-specific nucleic acid sequences; amplifying the probes that form the hybridization complex; detecting the amplicons, where detection of the amplicons indicates the presence of nucleic acid sequences corresponding to a set of signature genes in the tissue sample; and determining the expression levels of target sequences, where the expression of at least two, at least three, and at least five signature gene-specific sequences is detected.

[0071] The present invention also relates to a collection of isolated osteosarcoma signature genes comprising at least two genes selected from the group consisting of AMZ2P1, ASXL2, BIRC6, C11orf58, C1orf53, CDC34, CCDC1C34, DNMT3A, DNMT3A, ECI1, EIF3M, EIF1, EIF3M, ESCO1, FAM35DP, GAGE12D, GAGE12G, HADHA, HISTH2H2A3, HIST2H2A4, MAPK1, MEAF6, MYO1B, MYOM3, NDP, PAIP1, PCID2, PEG10, PEX26, POLR2C, PRR14L, RALGAPA2, SLC2A6, SLC7A4, ST7L, THAP9-AS2, TIMP3, and WNT4. In particular, the present invention provides a suitable collection of probes for each of the above collections of signature genes. Examples of such probes are provided in Table 3 below.

[0072] The present invention includes compositions, kits, and methods for determining the prognosis of osteosarcoma in an individual from which a sample is obtained. The kit is any manufacture (e.g., package or container) comprising at least one reagent, e.g., a probe or a set of such probes (e.g., nucleic acid probes of SEQ ID NOs. 1-41 or a subset thereof) for specifically detecting the expression of a gene sign, as described herein.

[0073] As shown in Example 2 below, the inventors demonstrated that a PPARγ antagonist reduced cell viability in seven human osteosarcoma strains. This is evidence in principle that inhibitors of the PPARγ pathway may constitute a groundbreaking innovation in the equipment for treating osteosarcoma.

[0074] In another embodiment of the present invention, the present invention thus relates to the use of PPARγ inhibitors as an antitumor therapy in subjects having osteosarcoma, particularly for the treatment of adolescents or young adults.

[0075] Examples of PPARγ inhibitors that can be used in accordance with the present invention include T0070907 (CAS number 313516-66-4), BADGE, FH535, GW9662, SR16832, SR202, and combinations thereof.

[0076] According to certain embodiments, the PPARγ inhibitor is used, as described above, to treat patients identified as poor responders to chemotherapy by the method of the present invention.

[0077] In certain embodiments, the PPARγ inhibitor can be combined with other treatments such as treatment with anti-angiogenic agents (e.g., multi-tyrosine kinase inhibitors and monoclonal antibodies), anti-osteoclasts (e.g., bisphosphonates and anti-RANKL), etc.

[0078] Other features of the present invention will become apparent in the course of the following description of biological assays, which are carried out within the framework of the present invention and provide the necessary experimental support without limiting its scope.

Example

[0079] The inventors inferred that in osteosarcoma, in addition to providing access to the TME composition, gene expression at the RNA level should better characterize osteosarcoma tumors as the ultimate readout of epigenetic and transcriptional regulation of a highly reconfigured genetic landscape rather than phenotypic features.

[0080] To test this hypothesis, the authors performed RNA sequencing (RNA-seq) of 79 primary osteosarcoma tumors sampled at diagnosis from patients enrolled in the first-choice OS2006 trial (NCT00470223) and uniformly treated. These samples summarized the distribution of the main clinical characteristics of the entire cohort (Table 1). To study the transcriptome landscapes of osteosarcoma cells and TME cells, the respective transcriptome programs were first dissociated. For that purpose, the ICASSO stabilization procedure 23Independent component analysis using (ICA, BIODICA implementation) 23 This process broke down the RNA gene expression matrix into 50 independent components (ICs), also known as gene modules.

[0081] [Table 1]

[0082] Example 1: The immunoinfiltration and tumor microenvironment transcription program stratifies pediatric osteosarcomas into prognostic groups at diagnosis. Materials and methods Groups and materials Biological samples were prospectively collected from patients (up to 50 years) enrolled in the French OS2006 / sarcoma09 trial (NCT00470223), which was approved as a therapeutic drug. This trial was conducted in accordance with the ethical principles of the Declaration of Helsinki and the standards for conducting clinical trials of medicinal products. If a patient was under 18 years of age at enrollment, special informed consent for blood and tumor samples was obtained from the patient or their parents / guardians. The information provided to patients, the written consent used, the sample collection, and the research project were approved by an independent ethics committee and institutional review board. As part of supplementary biological research, RNA-seq, CGH arrays, and WES were performed at the Gustave Roussy Cancer Campus.

[0083] Between 2009 and 2014, 79 frozen osteosarcoma biopsy samples were collected at diagnosis and analyzed by RNA-seq. The clinical and survival characteristics of the 79 patients (Table 1) were comparable to those of the overall osteosarcoma population enrolled in the OS2006 trial. The sex ratio M / F was 0.55, and the median age was 15.43 years (4.71–36.83 years). Primary tumors were located in the extremities (94.9%), and 20.2% had metastases at diagnosis. Chemotherapy was administered according to the OS2006 trial, with 86% of patients receiving the MTX-etoposide / ifosfamide (M-EI) regimen, 12.6% receiving either the adriamycin / platinum-based / ifosfamide (API-AI) regimen, and 34.1% of the randomized group receiving zoledronic acid. 77 patients underwent surgery for the primary tumor, and 24% had poor histological response. Twenty-nine patients experienced relapse, with a median delay of 1.55 years (range 0.09 to 3.62). Twenty-two patients died, with a median delay of 2.25 years (range 0.07 to 6.9). The median follow-up period was 4.8 years. Overall, the 3-year progression-free survival (PFS) and overall survival (OS) rates were 65.82% and 82.27%, respectively.

[0084] Nucleic acid extraction DNA and RNA were isolated using the AllPrep DNA / RNA mini-kit (Qiagen, Courtabouf, France) according to the manufacturer's instructions. Quantitative analysis and qualification were performed using a Nanodrop 2000 spectrophotometer (Thermo Fisher Scientific, Illkirch, France) and a Bioanalyzer DNA 7500 (Agilent Technologies, Les Ulis, France).

[0085] RNA sequencing RNA sequencing libraries were prepared using the TrueSeq Stranded mRNA Kit according to the following recommendations: poly(A) mRNA capture with oligo(dT) beads using 1 μg total RNA, fragmentation to approximately 400 pb, DNA double-strand synthesis, and ligation of the library by Illumina adapter amplification of the library by PCR for sequencing. Library sequencing was performed using an Illumina sequencer (NextSeq 500 or Hiseq 2000 / 2500 / 4000) in 75 bp paired-end mode for both techniques, and data sequencing was processed by bioinformatics analysis. A proprietary meta-caller approach was used for optimal detection of potential fusion transcripts by RNA-seq.

[0086] Gene expression analysis using RNA-seq The quality of the strand-paired-end RNA-seq libraries was evaluated using fastqc (a quality control tool for high-throughput sequence data available on the Babraham Bioinformatics page of the Babraham Institute website). Reads were mapped using RSEM with the GRCh37 ENSEMBL mRNA dataset as the reference sequence.

[0087] ICA was performed using BIODICA, one of the most powerful implementations of fastICA, including icasso stability analysis. Genes represented by fewer than 100 reads were filtered from the ICA, the gene expression matrix was logarithmically scaled, then scaled by gene, and finally decomposed into 50 components. To evaluate the relationships between ICs, Pearson correlations between ICs were calculated from the metagene matrix. For each IC, contributing genes were defined as genes within 3 standard deviations of the mean of the corresponding IC metagene vector. For each component, gene sets were enriched using msigdb for positional enrichment and g:Profiler for functional enrichment. For each component, only the most significant enrichment detected in the GO, REACTOME, or TF categories is reported. For positional enrichment, only cell bands with a minimum FDR of 1e-4 are reported. If several cit bands were enriched from the same chromosome arm, only p-values ​​with low significance are reported in Table 2. From the gene expression matrix, only contributing genes were selected within 2.5 standard deviations of the mean of the corresponding IC metagene vector. Network inference was performed using Pearson distance to calculate edge lengths. Edges with distances greater than 0.25 were discarded. Network and subnetwork diagrams and analyses were created using Cytoscape v3.6.1. The inventors used the REACTOME FI viz plugin to detect clusters within the network using a spectral partitioning-based network clustering algorithm and estimated the functional enrichment of each cluster in GO biological process terms.

[0088] Using Pearson distance and Ward construction methods, a hierarchical classification of the metagene matrix and corresponding heatmap was created using the Pheatmap R package. The Pls-da function from the MixOmics R package was used to select the independent component that contributed most to the G1 / G2 patient classification.

[0089] Kaplan-Meier OS and PFS curves were generated using the Servminer and Survival R packages. Log-rank tests were used to compare survival distributions between groups, and corresponding p-values ​​were provided.

[0090] [Table 2]

[0091] Differential mRNA expression was estimated from the raw read count table using the DESeq2 R package.

[0092] Using a Gaussian regression model with the glmnet R package (parameters: type.measure="mse", α=0.8, family="gaussian"), we selected CNAs (estimated from CGH arrays paired with RNA-seq samples) and optimally modeled each IC. Similarly, we detected the dependence between ICs corresponding to large chromosomal transcriptional regulation and copy number changes from the same chromosomal region. The chromosomal regions written in bold in Table 2 correspond to IC-rich regions and overlap with the CNAs that contribute most to the model, suggesting a dose-dependent relationship. Figure 5 shows the case of the copy number change with the highest coefficient in the regression model, based on a scatter plot of IC signals versus a logarithmic change of the logarithm of the copy number.

[0093] Using a logistic regression model with the glmnet R package, we defined a minimal genetic signature that could distinguish G1 tumors from a G2 RNA sequencing library (parameters: type.measure="mse", α=0.35, family="binomial"). A cross-validation strategy was used to select the best λ. Next, we predicted these characteristic G1 and G2 tumors from 82 osteosarcoma RNA-seq datasets created by the TARGET consortium and compared the signatures with those in 42 recurrent osteosarcoma samples from the MAPPYACTS trial (NCT02613962). For TARGET and MAPPYACT, gene expression was scaled by the mean and variance of gene expression in our 79 samples. For TARGET samples, Kaplan-Meier OS and PFS curves were generated using the Survminer and Survival R packages. Log-rank tests were used to compare survival distributions between groups and provide corresponding p-values.

[0094] Oligonucleotide array comparative genomic hybridization (aCGH) assay In all experiments, sex-matched normal DNA from pooled human females or males (Promega, Madison, WI, USA) was used as a reference. Oligonucleotide aCGH processing was performed as detailed in the manufacturer's protocol (version 7.5; http: / / www.agilent.com). Equal amounts (500 ng) of tumor and normal DNA were fragmented with AluI and RsaI (Fermentas, Euromedex, France). The fragmented DNA was labeled with cyanine Cy3-deoxyuridine triphosphate (dUTP) or Cy5-dUTP. Hybridization was performed on a SurePrint G3 Human CGH Microarray 4x180K (Agilent Technologies, Santa Clara, CA, USA) array in a 20 rpm rotating oven (Robbins Scientific, Mountain View, CA) at 65°C for 24 hours. Appropriate washing steps were performed after hybridization.

[0095] Oligonucleotide aCGH mixture (joint + solid) pretreatment Glass microarray scanning was performed using an Agilent G2505C DNA microarray scanner at 100% PMT with a resolution of 3 μm at 20°C in a low-ozone environment. Data was extracted from the scanned TIFF images along with protocol CGH 1105 Oct12 using Feature Extraction software (v11.5.1.1, Agilent). In v3.4 (http: / / cran.r-project.org), all additional data processing was performed under R statistical analysis. Acquired raw intensity was converted to log2 (test / reference). Joint normalization of the entire cohort of raw CGH profiles was performed using the "cghseg" package (v1.0.2.1) with default parameters: a common "wave effect" track was calculated and then subtracted from all individual profiles through Lowess regression. As a second step, an individual normalization step was performed for each profile by subtracting a pre-calculated GC content track through Lowess regression. Joint segmentation of the entire cohort of normalized CGH profiles was performed using penalized least squares regression implemented in the "copy number" package (v1.16.0). To set eigenvalues ​​for the gamma (penalty) parameter across the entire cohort, the optimal gamma value was calculated for each profile using 100 cross-validation loops, and the median of the optimal gamma was selected (median = 18, sd = 4.2, range = 11:41). The joint segmentation resulted in 1604 segments. Individual profile centering was performed using our proprietary method of selecting the mode that best centered the distribution density of the probe's log2(ratio) values. No abnormal calls were performed.

[0096] Oligonucleotide aCGH analysis All genomic coordinates were established on the UCSC Human Genome Construction hg19. Hierarchical clustering of samples was performed under R on segmented data using the Euclidean / Pearson / Spearman distance and Ward's construction method. Clustering of samples based on non-negative matrix factorization and spherical k-means was performed using the "NMF" (v0.20.6) and "skmeans" (v0.2.10) packages, respectively. Minimum Common Region (MCR) analysis was performed using GISTIC2 (v2.0.22). Difference analysis was performed using nonparametric statistical tests (Wilcoxon for 2 classes, Kruskal-Wallis for N classes), and all p-values ​​were adjusted for FDR using the Benjamini-Hochberg method.

[0097] Multivariate proportional hazards model To model the association between overall survival and major clinical factors (sex, histological response, metastatic status, tumor size, treatment, chemotherapy-induced adolescent status) and G1 / G2 signatures, we investigated stability selection through boosting using a Cox model (Hofner et al., 2015: Controlling false discoveries in high-dimensional situations: boosting with stability selection; BMC Bioinformatics volume 16, Article number: 144).

[0098] The selection frequencies for these predictors were calculated from 1,000 bootstrap samples. The number of selection variables per boosting was fixed at 2, and the threshold defining stable variables was fixed at 0.9 (this can be loosened, increasing the risk of false-positive predictor selection). The family-per-family error rate (PFER) and significance level corresponding to the expected maximum number of false-positive selection predictors were calculated from two previously defined quantities, and from the definitions provided by Meinshausen and Buhlmann (Meinshausen and Buhlmann (2010); Stability selection; Royal Statistical Society 1369-7412 / 10 / 72417 J. R. Statist. Soc. B (2010) 72, Part 4, pp. 417-473). These analyses were performed using the mboost R package.

[0099] Targeted RNA expression analysis Direct gene expression analysis was performed using the Nanostring standard custom approach, following supplier recommendations. Following Nanostring's guidelines, a specific panel of 41 probes of interest was designed, including five housekeeping genes (CNOT1, EIF4G2, SF1, SLC39A1, and SURF4) (Table 3: Target List). RNA integrity was controlled using a fragment analyzer system (RNA concentration, RNA quality number, RNA fragment percentage > 300 nt), RNA concentration, and purity using a Nanodrop ND8000.

[0100] [Table 3]

[0101] Due to the number of targets (<400), 100 ng of total RNA was hybridized to Nanostring probes. Our cohort consisted of 176 samples in two batches of hybridization, each containing a No Template Control (NTC, water) sample and a Universal RNA sample (Agilent Technologies, P / N: 7400000). Prior to hybridization, Nanostring positive and negative controls were added to the samples as spikes in the controls. After hybridizing the probes and mRNA at 65°C for 16 hours, processing was performed on a NanoString nCounter preparation station to remove excess probe, and the biotinylated hybrids were immobilized on streptavidin-coated cartridges. The cartridges were scanned at the maximum scanning resolution (555 fields of view (FOV)) on an nCounter Digital Analyzer (NanoString Technologies), and individual fluorescent barcodes were counted to quantify RNA molecules. Raw data were controlled and normalized using Nanostring nSolver 4.0 software. Image QC criteria exceeded 93% (threshold: 75% of FOV). Binding density was monitored as positive and negative control signals. Counts obtained for each target in the NTC sample (1–9 counts) were estimated for the corresponding target in the control sample. RNA content was normalized using the geometric mean of housekeeping gene signals. Universal RNA comparison showed platform agreement (r2: 0.98792). Of the 176 samples, three underperforming samples were rejected from the primary analysis: one sample had low binding density (low threshold: 0.1; EX148 ARN010: 0.09), and two samples had high mRNA content normalization factors (threshold: 20; EX148 ARN071: 25.7 and EX148 ARN165: 870.6).

[0102] Prediction of G1 / G2 layers from a target RNA expression panel To detect suspicious discrepancies, we first compared the RNA expression nanostrings and RNA-seq estimates of 35 genes and 5 housekeeping genes. Similarly, we excluded 7 genes with correlation p-values ​​(Parson's method) higher than 10⁻⁵ (not shown in Figure 6A). From 166 samples, 70 samples already analyzed by RNA-seq and thus classified as G1 or G2 were selected as a training set to retrain our logistic model using glmnet on a 28 target RNA expression panel with elastic net normalization. Cross-validation strategies were used to select α to 0.8 and λ to λ. Next, we predicted the G1 / G2 layers of the remaining 96 samples and generated Kaplan-Meier OS and PFS curves using the Servminer and Survival R packages. We compared survival distributions between groups using the log-rank test and provided corresponding p-values.

[0103] result 1.1. Independent components summarize biological functions and their interactions. First, we examined whether specific ICs correlated individually with clinical variables. Except for IC2 (not shown in the figure), which was associated with sex and obvious gene enrichment from the Y chromosome, other ICs were not significantly associated with clinical items after p-value adjustment.

[0104] Next, we characterized each IC by functional enrichment analysis using the most contributing gene. We observed that 90% of the ICs were significantly associated with either large chromosomal regions of transcriptional regulation according to the msigdb database or specific cellular / molecular functions according to g:Profiler (Table 2). Several ICs appeared to be specific to tumor clones with gene expression changes involving statistically significant large chromosomal regions. Some of these regions summarized known CNVs observed in osteosarcoma, confirmed by comparison with 70 CGH arrays paired with our RNA-seq samples (Table 2, Figure 5). Contributing genes from the remaining 10 ICs summarize diverse TME cell transcription programs or interactions with tumor cells. We identified osteosarcoma functional gene modules belonging to the bone microenvironment (osteoclast / bone resorption IC25; osteoblast / ossification IC44), angiogenesis (IC13), and immune response (IC41), but also identified neuronal projection (IC48) and muscle (IC1). Pearson correlation studies of components revealed proximity between specific ICs (International Coherence) with two main component groups (not shown in figure). Such IC clustering indicates that osteosarcoma tumors share two specific IC patterns, confirming strong relationships across TME, cancer cells, and the regulation of large chromosomal regions. To elucidate these potential functional interactions, the authors inferred co-expression networks using only genes that significantly contribute to ICs (greater than 3 standard deviations from the mean). Genes belonging to the same IC showed strong interconnections within the network, confirming the existence of continuity between latent IC variables and actually measured gene expression (not shown in figure). In most cases, agreement between ICs and functional subnetwork annotations performed using REACTOME FI viz was identified (not shown), confirming the efficacy of ICA in decomposing the gene expression matrix into functionally related gene modules from heterogeneous tumors. Furthermore, co-expression network annotation helped classify genes from ICs into more precise subnetworks. IC41 was initially flagged as a general immune response and has since been subdivided into "adaptive immune response / neutrophil degranulation," "type 1 interferon response," and "adaptive immune response."IC9 is subdivided into "neutrophil degranulation / defensin" and "adipogenesis / peroxisome proliferator-activated receptor (PPAR) signaling pathways." On a larger scale, the network is structured around several hubs of highly connected IC / subnetworks, likely reflecting active crosstalk between TMEs and tumor cells.

[0105] 1.2. A stable and independent component for stratifying patients by prognosis. The inventors questioned whether IC / subnetwork interconnects reflecting TME composition were associated with clinical variables. Tumors were classified hierarchically using stable IC (stability index > 0.5) reflecting clinical annotations (not shown in figure). In this unsupervised analysis, two tumor groups associated with different mean “survival status” were defined, and the cohort was classified from high-risk group G2 (43.1% DR; not shown in figure) to 35 low-risk tumors, granulation G1 (8.5% mortality, DR). The inventors confirmed significant differences in overall survival (OS) between the G1 / G2 groups, estimated by the Kaplan-Meier method, using a log-rank test (p-value = 0.00042; Figure 1A). At a median follow-up of 4.8 years, 3-year OS was 100% and 67.8% in G1 and G2, respectively. A difference was also observed in progression-free survival (PFS), but the significance was low (p-value = 0.042; Figure 1B). Next, we challenged the prognostic value of stratification using a multivariate proportional hazards model that included known osteosarcoma prognostic factors. The model tested in this stability analysis identified G1 / G2 stratification at diagnosis as the most contributing factor to OS (68.5% inclusion rate; Figure 1C). To refine our analysis, we selected the ICs that most contributed to prognostic G1 / G2 stratification by projecting them in the first two principal component spaces (Figure 1D). As confirmed by supervised partial least squares discriminant analysis (Figure 1E), IC39 best represents G1 / G2 stratification, but several other ICs are involved in distinguishing between the two groups (Figures 1D, 1E). The first principal component (PCA1) showed a high correlation with the prognosis of several ICs when the second principal component (PCA2) was not significantly associated with any clinical item. Therefore, while PCA2 may be associated with interesting conditions of tumor cells (IC48: neurites, IC49: EGR1 regulatory gene, IC50: KIT and YES1 expression, IC42: E2F1 regulatory gene), we focused further on PCA1 function related to survival. These observations support our hypothesis that complex traits associated with our stratification and patient prognosis emerged from the interaction of several gene modules.

[0106] 1.3. Stratified ICs are related to specific biological functions. To understand this hidden, complex tumor phenotype, detectable at diagnostic time, at a biological level, we functionally characterized two prognostic groups at a network scale. On the inferred network, we overlaid log-2-fold gene expression for G1 vs. G2 tumors (data not shown).

[0107] The inventors identified that the G1 group expresses genes from two IC41 subnetworks at higher levels, significantly richer in genes involved in "innate immune response / interferon-1 response" and "inflammatory response" (Figure 2). Interestingly, IC39, a functionally complex component and the most contributing gene to the G1 group, appears to be related to immune TME (e.g., epigenetic reprogramming). HDAC11 / HDAC6 negatively modulates IL10, and KDM6A positively modulates IL6. These observations support conclusions from previous literature regarding the favorable prognostic role of osteosarcoma antitumor immunity. Furthermore, some G1 tumors express a specific cancer-testis antigen (CTA; IC26), known as a neogenic antigen source. In contrast, the poor-prognosis G2 group exhibited associated subsystems that re-establish other aspects of TME associated with poor prognosis and premetastatic phenotype, such as osteoclast differentiation (IC25), tumor angiogenesis / VEGFR (IC13), and neutrophil degranulation (IC9). Two other isolated subsystems were strongly upregulated in the G2 group (Figure 2): i) the fibroblast network (IC1) likely reflects myofibroblast reprogramming of cancer-associated fibroblasts (CAFs) that contribute to lung metastasis or mesenchymal phenotype and metastasis; and ii) the adipocyte / PPAR signaling pathway (IC9) is involved in metabolic reprogramming that promotes inflammatory status and cancer progression. Combining these G2 characteristics paints a picture of poor-prognosis osteosarcoma TMEs that are prone to metastasis and represent early events that fix TME and tumor fate in a vicious cycle.

[0108] To complement this functional approach and test the validity of our findings, we returned to the gene expression matrix and performed differential gene expression analysis between G1 and G2 tumors (Figure 2). Consistent with the above results, in G1 tumors, we observed re-expression of specific CTAs, including those of the GAGE ​​family clustered at Xp11.23, and others (FMR1NB, PASD1, NLRP4, CT45A7, MAGEC3, PAGE2, MAEL, TDRD1, IL13RA2, etc.), in addition to genes involved in spermatogenesis (e.g., RHOXF2B, GTSF1, RHOXF2, DAZ3). This re-expression induces several key epigenetic changes at methylation levels necessary for pre-meiotic or meiotic regulation. Accordingly, the inventors observed significant upregulation of several genes (adjusted p-value < 1.3 × 10⁻⁴) that are known to be involved in RNA gene silencing pathways (e.g., MAEL, DDX4, TDRD1) and to suppress metastatic factor expression during meiosis via the piRNA pathway. Similarly, the functional features of G2 tumors were confirmed by differential gene expression analysis with upregulation of genes involved in adipocyte differentiation, osteoclast formation, and angiogenesis, including some association with other G2 functions and poor prognosis in osteosarcoma (e.g., PLA2G2A, PAQR3, HP).

[0109] Thus, all three complementary functional analyses were consistent and underpinned the major contribution of TME to osteosarcoma progression. Innate immune responses were associated with favorable prognosis G1 tumors, while angiogenesis, osteoclast formation, and adipogenesis were associated with unfavorable prognosis G2 tumors. This is consistent with the observed clinical efficacy of anti-angiogenic multi-tyrosine kinase inhibitors in recurrent osteosarcoma, but the observed ineffectiveness of zoledronate in first-line osteosarcoma treatment is thought to be partly related to its effects on the immune system.

[0110] 1.4. Stratified ICs are associated with different large chromosomal regions. In addition to these different TME compositions, changes in gene expression were observed, including large chromosomal regions likely associated with the dose-effect relationship of CNAs (not shown in the figure, Table 2), and are thought to be associated with each of these G1 / G2 stratified groups. The most influential IC contributors in the G1 group (not shown in the figure, Table 2) reflected known osteosarcoma cell features, such as the site band 12q14.1 (CDK4, OS9) associated with 4qter (IC34) or 6p21.1-22.1 (IC6:RUNX2, CDC5L, UBR2), which includes genes involved in osteosarcoma tumorigenesis and response to chemotherapy. Other G1 group contributing ICs were largely characterized by dysregulated expression of telomere regions (IC24:15qter, 21qter, 12qter; IC34:4qter; IC35:13qter). This feature was not detected in the unfavorable G2 group, but the most actively contributing gene to IC19 is DAXX, which, as part of the ATRX / DAXX / HistoneH3.3 complex, regulates telomere maintenance through surrogate telomere extension. In the poor-prognosis G2 group, three major chromosomal regions were dysregulated: chromosome 6p (IC19), 8q (IC23), and 22q (IC33). The first two are well known to be high copy gain or amplification events associated with osteosarcoma carcinogenesis, more frequent in recurrent / metastatic osteosarcoma than in primary osteosarcoma and previously associated with poor prognosis. Site band 6p is a repeat amplification region in osteosarcoma that does not regulate the expression level of the oncogene CDCL5. Site band 8q copy number increase is involved in tumorigenesis via MYC-driven super-enhancer signaling and is strongly suspected to be a prognostic factor for osteosarcoma. Surprisingly, chromosome 22q has not been previously reported as a prognostic factor for osteosarcoma and may require further investigation. These different chromosomal imbalances, specific to each group, may reflect different tumorigenetic pathways.

[0111] Overall, these ICs with dysregulated chromosomal regions highlighted the major contribution of CNVs to G1 / G2 stratification, even though we cannot rule out the possibility that some general regulation arises from epigenetic alterations. To estimate such potential contributions, we integrated the results with 70 CNV profiles (CGH arrays) paired with our RNA sequencing samples. Differential analysis of aCGH profiles using GISTIC2.0 characterized four chromosomes with significantly altered regions that differed between prognostic groups (adjusted p-value < 0.1, Figure 3A). Thus, increases in the 2p21-23 and 22q11.22 site bands and disappearance of the 11p12-p15 site band were detected in the G2 group, while disappearance of the 13q12 region was more pronounced in the G1 group than in the G2 group. Of these regions, only the chr22:21859431-22318144(8.27E-03) region was identified at an expression level (IC33) involved in G1 / G2 stratification. This region contained the genes PI4KAP2, RIMBP3B, RIMBP3C, UBE2L3, YDJC, CCDC116, SDF2L1, MIR301B, MIR130B, PPIL2, YPEL1, MAPK1, PPM1F, and TOP3B. While the MAPK signaling pathway has been previously proposed as a key driver in the metastatic stage, the PI3K-Akt signaling pathway may be involved in the early and late stages of osteosarcoma evolution. Furthermore, we noticed that the chr8q site band containing MYC was close to significant in aCGH profile analysis, supporting multiple reports on the involvement of MYC amplification in osteosarcoma. Considering the number of ICs associated with chromosomal region and copy number changes, our analysis suggests that a relatively small number of them are associated with G1 / G2, supporting the idea that tumor cells contribute less to disease progression than TME composition.

[0112] 1.5. Verification of prognostic characteristics of osteosarcoma RNA-seq Gene expression in oncology often exhibits high variance between samples, frequently leading to overfitted models or classifications based on noise rather than signal, raising questions about the reproducibility of results in other cohorts and the suitability of introducing such models for patient use.

[0113] To verify the robustness of our stratification in independent cohorts, we identified four gene signatures based on 37, 15, 8, and 5 genes, respectively. These signatures were learned from the initial gene expression matrix, and G1 / G2 tumors were predicted using 20 gene signatures (Table 8) that were normalized by elastic network (Tables 4-7) and LASSO-normalized logistic regression, respectively, by machine learning from the initial gene expression matrix.

[0114] [Table 4]

[0115] [Table 5]

[0116] [Table 6]

[0117] [Table 7]

[0118] [Table 8]

[0119] The inventors examined 15 gene signatures shown in Table 5 and predicted G1 / G2 tumors from an independent cohort of 82 pediatric osteosarcoma tumors for which gene expression tables and paired clinical data were available through open access to the osteosarcoma project page on the Office of Cancer Genomics (National Cancer Institute) website. To confirm the validity of the prediction, OS was compared between the predicted G1 / G2 and the generated associated log-rank p-values ​​(Figure 3B). Consistent with observations in the OS2006 cohort, the results in this independent cohort demonstrated that predicted G2 tumors were significantly associated with a worse prognosis than predicted G1 tumors (p-value: 0). The predictive power of this gene signature was supported regardless of the first-line treatment used. In fact, the first-line chemotherapy used in this independent cohort differed from that in the OS2006 cohort, mainly based on MAP regimens. Another valid criticism of RNA-seq-based features concerns reproducibility in laboratories with different technical settings. Therefore, a custom Nanostring panel was designed based on RNA-seq signatures applied to 166 samples from the OS2006 cohort. The inventors' elastic net model was retrained using 70 samples that had already been analyzed by RNA-seq. G1 / G2 predictions for the remaining 96 tumors reaffirmed that this stratification was associated with significantly different overall survival and progression-free survival rates (Figures 3C and 6B).

[0120] Finally, we used RNA-seq-based features to estimate the validity of stratification at relapse and examined the reversibility of this prognostic feature throughout disease progression. The proportion of G1 / G2 tumors collected at diagnosis was similar among patients in the OS2006 RNA-seq cohort who experienced relapse in the MAPPYACTS trial (NCT02613962) (Figure 3D). This result supports the persistence of this stratification throughout disease progression.

[0121] 1.6. Consideration Using an unsupervised machine learning strategy, the authors defined a repertoire of genetic components describing osteosarcoma tumor clones and TMEs. They observed that component interactions via co-expression stratified the cohort into good-prognosis and poor-prognosis tumors. Functional characterization of the components linked good-prognosis tumors with specific innate immune expression accompanied by angiogenesis, osteoclasts, and adipogenic activity, and poor-prognosis tumors, each group associated with distinct CNVs.

[0122] These distinct functional characteristics enable therapeutic stratification in osteosarcoma, for example, immunomodulation for G1 tumors (e.g., mifamultide) and anti-osteoclast therapy and anti-angiogenic therapy for G2 tumors.

[0123] The inventors also identified underlying biological pathways in osteosarcoma, including PPARγ, piRNA, or CTA, which highlighted novel viable targets. The inventors' data suggest that early, dramatic genetic / transcriptome perturbations in specific clones may influence not only tumor evolution, response to treatment, and metastasis potential, but also TME (transient mesenteric malformation).

[0124] Finally, the predictive power of 15 genes for minimal prognostic features was confirmed within an independent cohort of 82 primary osteosarcoma tumors, using a reproducible nanostring assay. This study paves the way for the development of prognostic tests that will lead to personalized treatment in osteosarcoma, particularly by helping clinicians identify patients who are difficult to treat at diagnosis.

[0125] Our data highlight that personalized treatment in osteosarcoma should be based not only on the genetic abnormalities of the tumor itself (e.g., mutations / CNVs), but also on RNA expression profiling that takes into account DNA abnormalities transcribed at the RNA level and the TME landscape.

[0126] Example 2: Modification of the PPARγ pathway, a novel therapeutic target in osteosarcoma. Materials and methods MTS cell proliferation assay HOS, 143B, MG-63, U2OS, and IOS18 cell lines were seeded at 5,000 cells / well, and Saos-2, Saos-2-B, and IOR / OS14 cell lines were seeded at 10,000 cells per well in a 96-well plate with a final volume of 100 μl / well. The cells were then allowed to settle overnight in DMEM containing 10% fetal bovine serum.

[0127] Cells were treated with drugs at varying concentrations ranging from 0 μmol / L to 100 μmol / L (troglitazone (TGZ) or rosiglitazone (RGZ) PPARγ agonists dissolved in DMSO, or the PPARγ antagonist T0070907). Controls (untreated cells) received the same volume of DMSO as treated cells (10 μl per 1 ml). Cell viability was measured at 48 and 72 hours post-exposure. Old medium was removed, and MTS / new medium (20 μl MTS solution - final concentration 0.33 mg / ml) (CellTiter 96 Aqueous One Solution cell proliferation assay; Promega Corporation, Charbonnieres, France) was added. One set of wells was prepared simultaneously with only MTS / new medium for background subtraction. The cells were incubated at 37°C for 1 to 7 hours (cell line metabolism dependent), and colorimetric measurements were performed at 490 nm using an automated plate reader (Elx808; Fisher Bioblock Scientific SAS, Illkirch, France).

[0128] Western blot method HOS, 143B, MG-63, U2OS, Saos-2, Saos-2-B, IOR / OS18, and IOR / OS14 osteosarcoma cell lines were seeded in 60 mm plates and collected at approximately 80% confluence. Cells were collected in lysis buffer (10 ml of TNEN 5 mM buffer, 1 / 2 protease inhibitor pill, 50 μl of NaF, and 50 μl of ortho-vanadate (phosphatase inhibitor)), and then obtained by alternating 5 cycles of freezing in nitrogen and thawing in a 37°C water bath. After centrifugation at 13,200 rpm for 20 minutes at 4°C, the protein supernatant was collected.

[0129] Protein quantification was performed using a series of bovine serum albumin concentrations (BSA, Euromedex, 04-100-812-E, Souffelyersheim, France) with the ThermoFisher Scientific (Thermoscific Pierce™ BCA Protein Assay Kit) kit. Absorbance was read at 570 nm using an automated microplate reader (Elx808; Fisher Bioblock Scientific SAS, Illkirch, France).

[0130] Proteins (30 μg / well) were separated by 4-20% polyacrylamide gel electrophoresis with a tris-glycine extension (Mini-Protean TGX, Bio-Rad, California, USA), and transferred to a polyvinylidene fluoride (PVDF) membrane (Trans-Blot Bio-Rad, California, USA) using the Trans-Blot Turbo transfer system (Bio-Rad Laboratories, California, USA).

[0131] The membrane was saturated with 5% BSA at room temperature for 45 minutes, and antibody against PPARγ (5 mg / mL; P5505-05B; US Biological, USA) or AdipoQ (1:1000; ab75989; Abcam, Cambridgeshire, UK) was added and incubated overnight at 4°C. Subsequently, the membrane was washed several times with buffer (1% TBS and 0.1% Tween®) and incubated with secondary antibody (goat anti-rabbit IgG, 1:5000; A9169; Sigma-Aldrich, St. Louis, Missouri, USA) for at least 2 hours.

[0132] The membrane was identified using the Clarity Western ECL substrate kit (Bio-Rad Laboratories, California, USA) and protein bands detected by chemiluminescence using the Bio-Rad ChemiDoc imaging system (Bio-Rad Laboratories, California, USA).

[0133] Next, stripping was performed to incubate the membrane with a β-actin antibody directly conjugated to HRP (HRP conjugate; 1:1000; #5125; Cell Signaling, MA, USA) (stripping buffer 62.5 mL Tris 0.5 M, 50 mL SDS 20%, 3.5 mL β-mercaptoethanol, 500 mL H2O).

[0134] result The therapeutic potential of the PPARγ pathway was investigated in vitro by evaluating cell proliferation under the effects of PPARγ agonists and PPARγ antagonists (MTS trial).

[0135] The PPARγ antagonist T0070907 reduces cell viability in seven human osteosarcoma lines with an IC50 ranging from 9 to 18 μM (median IC50 20 μM). The agonist showed little effect at the highest concentration tested, which can be explained by a negative feedback loop of PPARγ mediated by the expression of PPARγ-negative dominant isoforms.

[0136] conclusion Modulation of the PPARγ pathway is a new therapeutic target in osteosarcoma.

[0137] Example 2: Potential of PPAR therapy in osteosarcoma: Preclinical in vitro and in vivo experiments Materials and methods cell culture Human osteosarcoma cell lines HOS, HOS R / MXT, HOS R / DOXO, 143B, U2OS, Saos-2B, Saos-2B, MG-63, IOR / OS14, and IOR / OS18 were cultured at 37°C under mycoplasma-free conditions in Dulbecco's modified Eagle medium (DMEM, Invitrogen, St. Aubin, France) with various genetic backgrounds, supplemented with 10% (v / v) fetal bovine serum (FBS, Invitrogen, St. Aubin, France). shRNA and overexpression were performed using bioluminescent cell lines (luciferase / mKate2).

[0138] PPARg expression qRT-PCR PPARg expression was evaluated by qRT-PCR. ARN extraction was performed using the AllPrep DNA / RNA mini-kit (Cat number / ID: 80204 Qiagen) according to the manufacturer's instructions. Next, RNA (1 μg) was reversed using MLV reverse transcriptase (Invivogen ref 28025.013). Finally, 5 μL of cDNA was mixed with 6 μL of nuclease-free water, 1.5 μM primer, 10 μM (PPARY-s TTGACTTCTCCAGCATTTCTAC (SEQ ID NO: 42) + PPARY-as CTTTATCTCCACAGACACGAC (SEQ ID NO: 43)) and 12.5 μL of Cyber ​​Green Master Mix (reference K0223 Thermo Fisher Science). Amplification of PPARY and GAPDH was performed for 1 cycle, 2 minutes at 50°C and 10 minutes at 95°C, followed by 40 cycles of 15 seconds at 95°C and 1 minute at 60°C. To identify unique PCR products, melting curves were performed at the end of PCR (15 seconds at 95°C, 1 minute at 60°C, and 15 seconds at 95°C). Amplification was monitored using a ViiA 7 real-time PCR system. GAPDH was used as a housekeeping gene. The relative expression of each transcript was calculated using the 2-ΔΔCt method.

[0139] Western blot method Osteosarcoma cell lines were incubated overnight at 37°C as described above, and either treated the following day or not treated. The cell pellets were resuspended in 100 μl of lysis buffer (10 ml of TNEN 5 mM buffer with 1 / 2 protease inhibitor pill, 50 μl of NaF, and 50 μl of ortho-vanadate) after 6, 24, 48, or 72 hours. Proteins were extracted using a frozen cell suspension in nitrogen, thawed in a 37°C (5×) water bath, and then centrifuged at 4°C for 20 minutes at 13,200 rpm. Protein quantification was performed using a BCA protein quantification kit (Thermoscientific Pierce™ BCA protein quantification kit) according to the manufacturer's instructions.

[0140] In short, proteins (20 μg–30 μg) were separated on a 4–15% Mini-Proteam TGX-stained free gel (ref 4568086 Bio-Rad) and then transferred to a nitrocellulose membrane (Ref 1704156 Bio-Rad) using a trans-blot turbo transport system (Bio-Rad). The membrane was then incubated overnight with PPARY primary antibody (anti-PPARg polyclonal P5505-05B; US Biological). The membrane was washed (5-fold with wash buffer) and incubated for 2 hours with secondary antibody (goat anti-rabbit a9169 sigma, whole immunoblot imaging performed using a 1 / 5000 Bio-Rad ChemiDoc imaging system). The membrane was then incubated in striping solution for bactin characterization (same procedure). Relative band intensity was calculated using ImageJ software.

[0141] compound Doxorubicin (DOXO) and methotrexate (MTX) were purchased from Clinisiences under the names Sigma Aldrich (Lyon, France), mafosfamide (MAF), T0070907, Rosiglitazone (RGZ), and troglitazone (TGZ) (Nantelle, France). All compounds were solubilized in dimethyl sulfoxide (DMSO; Sigma Aldrich, Lyon, France) in a 10 mM stock solution and stored at -20°C. In vivo, all compounds used were solubilized in PBS, except for T0070907, which was solubilized in 2.5% DMSO + 47.5% PEG400 + 50% PBS.

[0142] In vitro cell viability assay Parental HOS, 143B, MG-63, IOR / OS18, derived HOS R / MTX, and derived PPAR shRNA or overexpressing PPAR cell lines were seeded at 5,000 cells / well. Parental cells, Saos2, Saos-2B, IOR / OS14, resistant HOS R / DOXO, and derived PPAR shRNA or overexpressing PPAR cell lines were seeded at 10,000 cells / well in 96-well plates in DMEM supplemented with 10% FBS for both assays.

[0143] The day after seeding, cells were incubated for 72 hours in the presence of a range of drug concentrations (0–100 μmol / L for DOXO and MAF; 0–500 μmol / L for MTX, T0070907, rosiglitazone (RGZ), and troglitazone (TGZ)). Cell viability was evaluated using the CellTiter 96 aqueous one-solution cell proliferation assay (MTS assay) (Promega, Charbonnieres, France) according to the manufacturer's instructions. The semi-maximal inhibitory concentration (IC50) was measured using GraphPad Prism5 software (Graphpad Software Inc., California, USA).

[0144] Combined testing Cells were treated simultaneously with increasing concentrations of drugs, either individually or in equivalent molar ratios. The effect on cell number was measured by MTS assay according to the manufacturer's instructions. The results were analyzed using median effects analysis, deriving binding indexes (CIs) calculated at the equipotentially bound drug concentration (ED50) that inhibits growth at 50%. Combinations were analyzed using exclusive CI values.

[0145] PDX Model The experiments were validated by CEEA26, the Ethics Committee (Approval No.: APAFIS#27183-200914229028 v3), and conducted under conditions defined by the European Community (Directive 2010 / 63 / UE). Animals were purchased from Gustave Roussy (Villejuif, France) and maintained in their respective animal facilities in accordance with standard animal regulations, health management, and ethical management. Osteosarcoma PDXs were established from relapsed patients by transplantation in immunodeficient NSG mice and were considered established if tumor growth was maintained after at least two in vivo passages. Further transplantation was performed by direct transplantation of tumor fragments from previous passages, either fresh or preserved by soft congeners (frozen in FBS, +10% DMSO). Under anesthesia (3% isoflurane, 1.5 L / min air), after a 0.5 cm skin incision and gentle activation of the periosteum (periosteal detachment), the tumor sample is placed sympatrically between the muscle and bone of the tibia, paratibially (approximately 2 mm). 3 The tumor was transplanted to the paratibial region. To avoid bone pain, in addition to general anesthesia, or if symptoms occurred, an analgesic (buprenorphine 0.3 mg / kg) was administered. Clinical status, tumor uptake, and tumor growth were evaluated 1-3 times per week. Paratibial tumors were detected by palpation and macroscopic appearance of the tumor (caliper measurement). The experiment continued until the tumor reached a specific tumor volume of approximately 1500 mm3 and significant weight loss or difficulty walking occurred. Next, the mice were anesthetized and bone structural changes were analyzed by CT scan imaging. At the endpoint, the mice were euthanized, samples were harvested, and processed as described below.

[0146] In vivo T0070907 Treatment of an orthotopic model of osteosarcoma PDX Eight animals with the same OTS orthotopic PDX model were treated daily from day 14 after tumor transplantation with either T0070907 (10 mg / kg / injection) or a vehicle (saline, control group) via intraperitoneal (IP) injection (volume 10 ml / kg). In combination therapy with T0070907 and methotrexate, four groups—the control group, the combo group, and the saline group—were administered T0070907 (10 mg / injection) at J1-J4 or J1-J4, J6-J7 and methotrexate (10 mg / injection) at J5. Clinical status, tumor uptake, and tumor growth were evaluated every two days. Paratibial tumors were detected by palpation and macroscopic examination of the tumor (caliper measurement). Tumor CT scans were performed weekly. Tumor-affected legs, normal legs, lungs, spleen, and liver were collected at the time of slaughter and preserved for future analysis (RNA-seq, WES, histology, single-cell analysis, spatial transcriptome analysis, etc.).

[0147] In vivo CT scan imaging An IVIS SpectrumCT (Perkin Elmer, Courtabouf, France) system was used for image acquisition. This system enables the detection of the primary tumor using X-ray tomography. CT scans were performed under 3% (v / v) isoflurane anesthesia.

[0148] histology Organs were fixed in 4% (v / v) paraformaldehyde and embedded in paraffin. Tissues were morphologically stained with hematoxylin-eosin-safranin (HES). Paraffin sections were treated at 100°C for 20 minutes for heat-induced antigen recovery (ER2-compatible EDTA buffer pH 9). Slides were incubated at room temperature for 1 hour with mouse monoclonal anti-human Ki67 antibody (clone MIB1; 1:20; Agilent Dako) or anti-PPAR polyclonal (P5505-05B; US Biological). Nuclear signals were revealed using the Klear mouse kit (GBI Laboratory). Slides were examined using a light microscope (Zeiss, Marly-Le-Roy, France), and a single representative whole tumor tissue section from each animal was digitized using a slide scanner NanoZoomer 2.0-HT (C9600-13, Hamamatsu Photonics). A bone pathologist performed a histological examination.

[0149] background Outcomes for adolescents / young adults with osteosarcoma have not improved over the decades.

[0150] RNA sequencing of 79 biopsy samples from osteosarcoma diagnosis identified stable, independent components that reproducible the tumor microenvironment and clones (see Marchais et al. Cancer Research 2022, Example 1 in the publication). Metagene unsuppressed classification stratified this cohort into good-prognosis (G1) and poor-prognosis (G2) tumors with respect to overall survival. In multivariate survival analysis, this stratification was ranked as the most influential variable. Functional characterizations were associated with good-prognosis G1 tumors and poor-prognosis G2 tumors, as well as upregulation of PPAR pathways, with angiogenesis, osteoclast and adipogenic activity, and congenital immunity.

[0151] PPARs (gamma isotypes of peroxisome proliferator-activated receptors) are involved in various biological processes, including adipogenesis, angiogenesis, and immunity (macrophage polarization). 70,71It is a nuclear receptor involved in various aspects of cancer.

[0152] Proactive and anti-cancer effects are described depending on the tumor type. As a result, both antagonists and agonists have been explored as potential anti-cancer therapies. -PPARs are generally considered tumor suppressor genes through their roles in antiproliferation, apoptosis promotion, and redifferentiation in some cancers. Loss of expression in epithelial cancers. 72 , mutations that impair ligand binding, and consequently, their transcriptional activity 73 These have been reported. In these cases, synthetic agonists of PPARs may have anticancer activity. The oncogenic fusion PAX8-PPARγ has been described in thyroid cancer, and the PPARγ ligand pioglitazone induces transdifferentiation into adipocyte-like cells. Negative dominant isoforms can also be expressed in several cancers (e.g., ORF4 in colic cancer). 74 . Conversely, PPAR activation in some cancer cells may induce an energetic conversion toward fatty acid utilization, potentially contributing to cell proliferation and survival. 35 Several target genes of PPARs are also involved in tumor invasiveness, such as angiogenesis. Activation of PPARs in a fatty acid-rich microenvironment may favor metastasis in some cancers (breast cancer, melanoma). 75 In that case, T0070907, a PPAR-specific synthetic antagonist, is metastatic 75 It may lower [something].

[0153] In osteosarcoma, the PPAR agonist troglitazone (5 μM) promotes cell line survival in vitro by suppressing AKT-dependent spontaneous apoptosis. 76 At high concentrations, troglitazone (100 μM) exhibited the opposite effects to those of troglitazone, which has antiproliferative activity in vitro and antitumor activity in vivo through apoptosis-promoting and differentiation-promoting effects. This is likely due to a negative retrocontrol loop with a dominant-negative isotype. 78,79 .

[0154] The known effects of PPARγ activation are summarized in Figure 2 by Ahmadian et al. (2013 Nature Medecine; 19(5): PPARγ signaling and metabolism: the good, the bad and the future).

[0155] PPAR-γ and targets in osteosarcoma biopsy at diagnosis: OS2006 cohort Based on RNA-seq studies conducted on the OS2006 osteosarcoma cohort at the time of diagnosis, the PPAR signaling pathway is associated with the G2 poor prognosis group identified by our signature. RNA expression of several PPAR targets correlated with PPAR activation, as well as other components identified in this cohort that are associated with poor prognosis, such as pro-angiogenicity and osteoclast and adipocyte activity. All of these suggest the pro-tumor and pro-metastatic activity of PPARs in osteosarcoma, and the potential therapeutic role of PPAR antagonists in these patients.

[0156] PPAR-γ in osteosarcoma: In vitro preclinical data To evaluate the in vitro antiproliferative activity (MTS assay) of two PPAR agonists, troglitazone and rosiglitazone, and one synthetic specific antagonist, T0070907, both alone and in combination with chemotherapy used for osteosarcoma patients (methotrexate, doxorubicin, maphosphamide), PPARs (Western blot; Figure 4B) and their methotrexate or doxorubicin were used. 79 Eight osteosarcoma cell lines expressing resistance countermeasures to the drug were used (median effect analysis method). 80 .

[0157] In vitro antiproliferative activity of T0070907 in osteosarcoma cell lines The median IC50 of T0070907 for all eight osteosarcoma parent cell lines was 20 μM (range 9.3–37.4; MTS assay), but two agonists did not reach IC50 even at a maximum concentration of 100 μM (Figure 4A). The IC50 of T0070907 in osteosarcoma cells was similar to that of various histological types of adult cancer. 81 RNA sequencing of HOS cells exposed to T0070907 at 25 μM (IC50) for 24 hours confirmed pro-apoptosis and induction of the osteoblast differentiation program (Figure 7). As expected, the PPAR agonist (rosiglitazone) significantly upregulated osteosarcoma cells treated only with the adipocyte differentiation program, indicating that PPAR was the major regulator of adipogenesis.

[0158] In vitro synergistic effects of chemotherapy and T0070907 in parental and doxorubicin-resistant HOS osteosarcoma cell lines Next, the combined efficacy of T0070907, routinely used for osteosarcoma, and chemotherapy (methotrexate, doxorubicin, maphosphamide) was tested using the MTS assay, and a median action analysis was performed on osteosarcoma cell lines HOS and their counterparts resistant to methotrexate and doxorubicin. 78 The following was performed. Synergistic effects with all drugs were observed in all cell lines and at all timings of T0070907 administration (24 hours before, simultaneously with, or 24 hours after chemotherapy), and the resistance index increased when T0070907 was administered first (Table 1).

[0159] [Table 9]

[0160] In vivo preclinical data in the PPARγ:PDX model for osteosarcoma The initial hypothesis was that PPAR antagonists may modify the composition of the tumor microenvironment in osteosarcomas with a poor prognosis. To test this hypothesis, patients were treated with T0070907 alone or in combination with methotrexate (xenograft (PDX) model derived from orthotopic (paratibial) osteosarcoma patients) derived from metastatic samples of recurrent MAP-217-PT patients. This model was selected for its ability to express PPARs and form lung metastases. PPAR expression induced by IHC was observed in several PDX models (Figure 8).

[0161] After initial dose studies to determine the tolerable concentration of T0070907 in mice, the effects of T0070907 administered intraperitoneally (IP) at a dose of 10 mg / kg / injection daily (Figure 9A), or T0070907 administered intraperitoneally (IP) at a dose of 10 mg / kg / injection on days 1-4 or 1-4 and 6-7 in combination with methotrexate on day 5 (Figure 9B), were evaluated in vivo by analysis of primary tumor growth and lung metastases at mouse sacrifice.

[0162] T0070907 induced delayed tumor growth in the MAP-217 PT PDX model (the delay to reaching 2.5 times the initial leg volume, reflecting tumor volume, was 20% higher in T0070907-treated mice compared to controls; Figure 9A) and reduced the number and size of lung metastases (quantification in progress by pathologists; Figure 9B). Furthermore, T0070907 in combination with methotrexate appears to have a greater effect on primary tumor growth in vivo than each drug alone (Figure 9C). The effect of the combination on metastases is currently being quantified by pathologists.

[0163] Spatial transcriptome studies are being performed to further understand the effects of T0070907 alone and in combination with methotrexate on tumor cells and the tumor microenvironment.

[0164] Abbreviations used in the text: API-AI Adriamycin / Platinum / Ifosfamide Area under the AUC curve BCA bicinchoninate assay BSA (Bovine Serum Albumin) CAF cancer-associated fibroblasts CGH comparative genomic hybridization CNA copy number abnormality CNV (Copier NV) copy count variation CTA (Cancer Testicular Antigen) Annealing, selection, and ligation via DASL cDNA DMEM Dulbecco / Voigt Modified Eagle Minimum Essential Medium DMSO (Dimethyl Sulfoxide) DNA (Deoxyribonucleic Acid) DR mortality rate dUTP Deoxyuridine Triphosphate FOV field of view HRP Umarajish peroxidase IC Independent Components ICA independent component analysis IL Interleukin MAP: Methotrexate, doxorubicin, and cisplatin MCR Minimum common area M-EI MTX - Etoposide / Ifosfamide OLA Oligonucleotide Ligation Assay OS overall survival OTS Osteosarcoma PCA principal component analysis Error rate by PFER family PFS (Progression-Free Survival) PPARγ (Peroxisome Proliferator-Activated Receptor) PVDF (Polyvinylidene Fluoride) qRT-PCR (Quantitative Reverse Transcription Polymerase Chain Reaction) RGZ Rosigritazone RNA (ribonucleic acid) RNA sequencing (RNA-seq) TGZ Troglitazone TME (Tumor Microenvironment) WES Whole Exome Sequence Determination WGS Whole Genome Sequencing

[0165] References

[0166] [ka]

[0167] [ka]

[0168] [ka]

[0169] [ka]

[0170] [ka]

Claims

1. An in vitro method for determining the prognosis of osteosarcoma in a subject, (a) A step of measuring the expression level of a collection of signature genes derived from a biological sample taken from the subject, wherein the collection of signature genes includes AMZ2P1, ASXL2, C11orf58, C5orf28, CCDC34, ESCO1, GAGE12D, HADHA, MEAF6, PAIP1, RALGAPA2, SLC2A6, SLC7A4, THAP9-AS1 and TIMP3; (b) A step of applying the expression levels measured in step (a) to a predictive model relating to the expression levels of the collection of signature genes associated with an osteosarcoma outcome, wherein the predictive model is obtained by implementing elastic net logistic regression in the R package "glmnet"; and (c) A step of evaluating the outcome of the prediction model and determining the prognosis of the target osteosarcoma. In vitro methods, including

2. The method according to claim 1, wherein the expression levels of the signature gene collection are measured at the time of diagnosis.

3. The method according to claim 1 or 2, wherein the expression level of the signature gene collection is measured at the time of relapse.

4. The method according to any one of claims 1 to 3, further comprising combining the gene expression level of the signature gene with one or more other parameters in order to predict the progression of osteosarcoma in the subject.

5. The method according to any one of claims 1 to 4, wherein the biological sample is an osteosarcoma biopsy sample derived from the subject.

6. A diagnostic kit for predicting the progression of osteosarcoma in a subject, wherein the kit comprises at least one nucleic acid probe or oligonucleotide that can be used in a method defined in any one of claims 1 to 5 to measure the expression level of a collection of signature genes derived from a biological sample taken from the subject, the collection of signature genes comprising AMZ2P1, ASXL2, C11orf58, C5orf28, CCDC34, ESCO1, GAGE12D, HADHA, MEAF6, PAIP1, RALGAPA2, SLC2A6, SLC7A4, THAP9-AS1, and TIMP3.

Images