Prognostic signature for patients suffering from retroperitoneal sarcoma

Abstract

Disclosed is a method for determining the probability of recurrence of retroperitoneal sarcoma in a patient, based on quantitative analysis of the expression of a group of genes (prognostic biomarkers or "gene signature") by a machine-learning process based on cross-validation. Further disclosed are a gene signature with predictive value for the recurrence of retroperitoneal sarcoma, a kit for determining the expression level of the biomarker genes, and a system for implementing the method according to the invention.

Description

[0001] PROGNOSTIC SIGNATURE FOR PATIENTS SUFFERING FROM

[0002] RETROPERITONEAL SARCOMA

[0003] FIELD OF INVENTION

[0004] The present invention relates to a method for improving prognostic predictions for patients with retroperitoneal sarcoma, based on identification of a set of prognostic biomarkers.

[0005] BACKGROUND TO THE INVENTION

[0006] Soft-tissue sarcomas (STS) are rare tumours. They can arise in any part of the body, the most frequent being the extremities / trunk wall and retroperitoneum (45% and 15% respectively). The various histological subtypes present a wide range of clinical behaviours, and high-risk STS involves a substantial risk of mortality and recurrence. The size, grade and histological subtype of the tumour are the main prognostic variables. Historically, STSs were classified according to the local extent and spread of the tumour (AJCC / UICC staging system), but this subdivision does not take sufficient account of the risk stratification.

[0007] Sarculator nomograms for retroperitoneal sarcoma (RPS) predict the individual prognosis of patients with primary RPS after surgery on the basis of clinical and pathological characteristics. Disease-free survival (DFS) nomograms have proved robust in various external validations, and were incorporated in the 8th edition of the AJCC staging system.

[0008] Recently, “omics” analyses have emerged strongly due to their potential prognostic role in oncological patients. In particular, transcriptomics, which allows molecular profiling of thousands of genes, leads to better understanding of tumour biology and the development of prognostic and predictive molecular models.

[0009] No effective, clinically practicable method for the prognosis of retroperitoneal sarcoma has been developed to date.

[0010] SUMMARY OF THE INVENTION

[0011] The inventors have found a gene expression signature that allows accurate, effective individual prognostic evaluation at the time of diagnosis of retroperitoneal sarcoma.

[0012] In a first aspect thereof, the invention provides a method for determining the risk of recurrence of retroperitoneal sarcoma in a patient suffering from said condition, by means of quantitative analysis of the expression of a group of genes (prognostic biomarkers or “gene signature”), attribution of a score processed with a linear regression model on the basis of specific coefficients for each gene, and comparison of the risk score with a reference value or threshold value determined from retroperitoneal sarcoma samples by a machine-learning process based on cross-validation.

[0013] Further aspects of the present invention are a gene signature with predictive value of the recurrence of retroperitoneal sarcoma, and a kit for determining the expression level of the biomarker genes that make up the gene signature by means of gene expression analysis techniques on suitable solid media.

[0014] A further aspect of the invention is a system for implementing the method according to the invention, comprising a gene expression analyser and a processing unit configured to calculate the risk score.

[0015] DESCRIPTION OF FIGURES

[0016] - FIGURE 1: Heatmap of expression of the signature in the training set on the basis of colours, wherein the squares represent the expression value of a given gene measured in a given sample, ranging from white, which denotes low expression values, to greyscale, denoting intermediate expression, and black, representing high expression.

[0017] - FIGURE 2: Bar chart showing risk score distribution.

[0018] - FIGURE 3: Graph illustrating signature performance in training set.

[0019] - FIGURE 4: Graph illustrating signature performance in TCGA data set.

[0020] DESCRIPTION OF THE INVENTION

[0021] The subject of the present invention is a method for determining the risk of recurrence of retroperitoneal sarcoma in a patient suffering from said condition, wherein said method comprises the following steps:

[0022] (a) determining, in a tumour sample from the patient, the expression level of genes TBX1, AP1S1, NPTX2, DDOST, PFN1, PAXX, LCNL1, ZNF593, AGMAT, ATG4D, MBD5, GMPPA, and ZSCAN29; and

[0023] (b) calculating a score by totalling the weighted expression levels for each gene listed in step (a), according to the formula genen expression x coefficientn, wherein said coefficient for each genenis listed in Table 1 below:

[0024] Table 1

[0025] (c) classifying the score described in step (b) vs a threshold value, so that:

[0026] - if the score is lower than the threshold value, the risk of recurrence of retroperitoneal sarcoma in the patient tested is low;

[0027] - if the score is higher than the threshold value, the risk of recurrence of retroperitoneal sarcoma in said patient is high.

[0028] In a preferred embodiment of the invention, the patient referred to in step (a) is undergoing chemotherapy or radiotherapy treatment.

[0029] In another preferred embodiment, the threshold value referred to in step (c) is determined from retroperitoneal sarcoma samples from patients with or without recurrence by a machine-learning process based on cross-validation (Hastie, 2001). Using cross-validation, the algorithm was trained with a 10-part resampling method to divide it into two subsets: in the first, containing 9 / 1 Oth of the samples, the model with the corresponding threshold is determined; in the second (the remaining 1 / 1 Oth of the samples), the predictive capacity of the model is estimated. The algorithm training was reiterated 10 times, choosing different combinations of the 10 parts every time. The threshold value calculated on the average of the 10 reiterated procedures amounted to 10.13. Said value can vary, depending on the numerical size of the learning dataset.

[0030] In a preferred embodiment, the gene expression level in step (a) is determined by:

[0031] (i) extraction of mRNA from the biological sample;

[0032] (ii) reverse transcription of the extracted mRNA to cDNA; and

[0033] (iii) determination of the amount of cDNA in each gene.

[0034] A good-quality biological sample requires material collection methods that guarantee the gene expression levels in the tumour cells. Fresh frozen samples and samples embedded in paraffin wax fixed in formalin are good options among said methods.

[0035] Any known technique can be used to extract nucleic acid (mRNA) from the biological sample. In a preferred aspect, the quantitation of cDNA in step (iii) is performed by one of the following methods: qPCR, digital molecular barcode technology, new-generation sequencing, microarray or chip analysis.

[0036] The set of 13 genes (TBX1, AP1S1, NPTX2, DDOST, PFN1, PAXX, LCNL1, ZNF593, AGMAT, ATG4D, MBD5, GMPPA and ZSCAN29) constitutes the gene signature according to the present invention. A training set consisting of 118 retroperitoneal sarcomas was used to identify the signature, while 98 samples taken from The Cancer Genome Atlas (TCGA) were used at the development stage to confirm its performance.

[0037] The signature score corresponds to the sum of the product of the gene expression level and its relative weight (coefficient). Table 2 lists the 13 genes, their annotation, and their coefficients.

[0038] Table 2

[0039] The algorithm used to allocate a risk score to a tumour sample is based on a linear regression model that includes the gene expressions and coefficients or weights shown in Table 2. The weights of the genes listed in Table 2 are both positive and negative, as demonstrated by the regression analysis of the model, wherein a positive coefficient indicates the genes whose expression increases in proportion to the prognostic risk, while a negative coefficient indicates the genes whose expression decreases in inverse proportion to the prognostic risk.

[0040] Another aspect of the present invention is the use of a gene signature comprising genes TBX1, AP1S1, NPTX2, DDOST, PFN1, PAXX, LCNL1, ZNF593, AGMAT, ATG4D, MBD5, GMPPA and ZSCAN29 to determine the risk of recurrence of retroperitoneal sarcoma in a patient suffering from said condition.

[0041] A further aspect of the invention relates to a kit for determining the expression level of genes TBX1, AP1S1, NPTX2, DDOST, PFN1, PAXX, LCNL1, ZNF593, AGMAT, ATG4D, MBD5, GMPPA and ZSCAN29 in a human tissue sample, which comprises: a support comprising or consisting of said genes, or portions or sequences thereof able to anneal to them, in separate areas of its surface, wherein said support is preferably in microarray format; and reagents to conduct quantitative analysis of said genes on the support.

[0042] In a preferred embodiment, the kit comprises at least one primer and / or probe able to anneal to each gene or a portion thereof, and a polymerase enzyme.

[0043] A further aspect of the present invention relates to a system for implementing the method according to the invention, comprising:

[0044] - a gene expression analyser

[0045] - a processing unit configured to calculate the risk score as defined in claim 1.

[0046] In a preferred embodiment, the system comprises a fully-automatic analyser and a kit according to the invention.

[0047] The processing unit is preferably a digital processing unit, such as a microprocessor programmed by means of software instructions. The processing unit can be incorporated in the gene expression analyser or constitute a separate processing unit. For example, processing can also be implemented by a personal computer, notebook, tablet or smartphone wherein an application configured to calculate the risk score on the basis of the gene expression (GE) levels is installed. In general, therefore, the processing unit is configured to receive the GE levels from the analyser. The GE levels can be stored in a volatile or non-volatile memory, such as a portable storage medium, or can be transmitted from the analyser to the processing unit.

[0048] The method according to the present invention enables the therapeutic approach to be customised by adapting it to the individual cancer biology. It can be used to predict the risk of recurrence, and represents a precision medical tool useful to improve the probability of a cure and reduce the toxicity associated with the treatment. The clinical validity of the signature is supported by various lines of evidence. Firstly, the signature is based on bioinformatic and statistical methods which have been widely validated in other contexts. Moreover, the signature was able to identify patients with an unfavourable survival prognosis in a set of independent external data, such as the TCGA (The Cancer Genome Atlas) cohort. Glossary of terms used

[0049] According to the present invention:

[0050] - the expression “unfavourable prognosis” denotes a high probability of disease recurrence after curative treatments;

[0051] - the expression “favourable prognosis” denotes a low probability of disease recurrence after curative treatments;

[0052] - the expression “signature” means the set of genes comprising at least the genes listed in Table 1;

[0053] “training set” means the gene expression dataset comprising the primary data and processed data together with the detailed clinical annotation, wherein the signature is developed.

[0054] Materials and methods

[0055] The gene expression data were collected from cancer samples with the following characteristics: primary tumour tissue (retroperitoneal sarcoma) on diagnosis (before treatment); significant clinical information; informed consent. The gene expression level (acquired as raw data and representative of the signal intensity) was normalised within each individual.

[0056] In the process of development of a multi-gene classifier as described herein, some key steps were identified: i) selection of characteristics; ii) construction of model; iii) internal validation; iv) estimate of model accuracy and sample size; v) design of external validation studies; vi) performance of classifier vs standard prognostic factors. The methodology applied to the present study follows the approach described above.

[0057] To train the model with the gene expression data, the Lasso and Ridge methods of the Glmnet package of project R were used for the statistical calculation. The procedure enables the following steps to be performed: i) selection of characteristics; ii) construction of model; iii) internal validation. BRB-array tools, an integrated package for display and statistical analysis using Excel (Microsoft, Redmond, WA) as front-end interface and including tools developed in the R statistical system, was used to apply Glmnet. A ten-fold cross-validation was used to train the prediction for high- and low-risk patients, as defined above. The prognostic model stratifies the samples into high- or low-risk classes, according to a 10-fold cross- validation. In each iteration, the data were divided between the 90% of cases used to predict the remaining 10% of omitted test cases. After repeating the entire procedure, excluding a different 10% of cases until each case had been omitted once, all the cases were stratified as high- (25%) or low-risk (75%) on the basis of quartiles of the entire cohort. The cut-off value of the signature to define low / high risk was calculated in the training set on the basis of cross-validation. The risk-score formula is calculated as follows: genenexpression x genencoefficient

[0058] The ability to forecast the survival rate was evaluated with the Kaplan-Meier curves, and the statistical significance was evaluated with the log-rank test. A test of 1,000 permutations was calculated to evaluate the significance of the log-rank statistic.

[0059] Results

[0060] Figure 1 illustrates the expression of the genes included in the signature in the training set, grouped on the basis of the signature scores. In the training set, the gene expression data originating from 118 FFPE samples allow the identification of a signature of 13 genes able to stratify patients with a low or high risk of recurrence on the basis of a 10-fold cross-validation. Figure 2 illustrates the distribution of the signature scores in the training set, showing a right- skewed Gaussian distribution. On the basis of the cross-validation in the training set, it was established that the threshold corresponded to a signature cut-off value of 10.13; thus the signature stratifies the cases as high-risk (n=29; 25%) and low-risk (n=89; 75%).

[0061] As shown in Figure 3, Kaplan-Meier analysis in the training set indicates that the signature led to separation of the patients as high- and low-risk (HR= 7.24; log-rank, p = 1. SEIS; grey = low-risk; black = high-risk) reaching a disease-free survival rate at the 72-month reference point of 59.3 and 0% respectively.

[0062] The signature was tested on the publicly available TCGA gene expression data (Figure 4); the data for 98 retroperitoneal sarcomas were recovered, and the signature was attributed with the result of separating the patients into high- and low-risk (HR= 4.29; log-rank, p = 4.9E- 06; grey line = low-risk; black = high-risk) reaching a disease-free survival rate at the 72-month reference point of 58.1 and 14.4% respectively.

Claims

CLAIMS1. In vitro method for determining the risk of recurrence of retroperitoneal sarcoma in a patient suffering from said condition, said method comprising the following steps:(a) determining, in a tumour sample of the patient, the expression level of genes TBX1, AP1S1, NPTX2, DDOST, PFN1, PAXX, LCNL1, ZNF593, AGMAT, ATG4D, MBD5, GMPPA and ZSCAN29; and(b) calculating a risk score by totalling the weighted expression levels for each gene referred to in step (a), according to the formula genenexpression x coefficientn, wherein said coefficient for each genenis reported in the table below:(c) classifying the risk score in step (b) against a threshold value, such that:- if the score is lower than the threshold value, the risk that the retroperitoneal sarcoma will recur in the patient in question is low;- if the score is higher than the threshold value, the risk that the retroperitoneal sarcoma will recur in the patient is high.

2. Method according to claim 1, wherein said threshold value in step (c) is determined from retroperitoneal sarcoma samples of patients with or without recurrence by an automated machine-learning procedure based on cross-validation.

3. Method according to claim 2, wherein said threshold value is 10.13.

4. Method according to claim 1, wherein the expression level of the genes in step (a) is determined by:(i) extraction of mRNA from the biological sample;(ii) reverse transcription of the extracted mRNA to cDNA; and(iii) determination of the amount of cDNA in each gene.

5. Method according to claim 4, wherein the quantitative determination of the cDNA in step (iii) is performed by one of the following methods: qPCR, digital molecular barcode technology, next-generation sequencing, microarray or chip analysis.

6. Method according to claim 1, wherein said patient is undergoing chemotherapy or radiotherapy treatment.

7. Kitto determine the expression level of genes TBX1, AP1S1, NPTX2, DDOST, PFN1, PAXX, LCNL1, ZNF593, AGMAT, ATG4D, MBD5, GMPPA and ZSCAN29 in a human tissue sample, which comprises or consists of:- a support carrying, in separate areas of its surface, said genes or portions or sequences thereof capable of annealing therewith, said support preferably being a microarray; and- reagents to carry out the quantitative analysis of said genes on the support.

8. Kit according to claim 7, comprising at least one primer and / or probe capable of annealing with each gene or a portion thereof, and a polymerase enzyme.

9. System suitable to carry out the method according to claims 1-6, comprising:- a gene expression analyser- a processing unit configured to calculate the risk score as defined in claim 1.

10. Use of a gene signature comprising genes TBX1, AP1S1, NPTX2, DDOST, PFN1, PAXX, LCNL1, ZNF593, AGMAT, ATG4D, MBD5, GMPPA and ZSCAN29 to determine the risk of recurrence of retroperitoneal sarcoma in a patient suffering from said condition.

Images