Patent Application
20120040848
The present invention relates to a method to in vitro determine the grade of a liver tumor in a sample previously obtained from a patient, using a molecular signature based on the expression of a set of genes comprising at least 2, especially has or consist of 2 to 16 genes, preferably a set of 16 genes. In a particular embodiment, the method focuses on hepatoblastoma (HB) or hepatocellular carcinoma (HCC), in adults or in children. The invention is also directed to sets of primers, sets of probes, compositions, kits or arrays, comprising primers or probes specific for a set of genes comprising at least 2 genes, especially has or consists of 2 to 16 genes, preferably exactly 16 genes. Said sets, kits and arrays are tools suitable to determine the grade of a liver tumor in a patient.
The liver is a common site of metastases from a variety of organs such as lung, breast, colon and rectum. However, liver is also a site of different kinds of cancerous tumors that start in the liver (primary liver cancers). The most frequent is the Hepatocellular Carcinoma (HCC) (about 3 out of 4 primary liver cancers are this type) and is mainly diagnosed in adults. In the United States approximately 10,000 new patients are diagnosed with hepatocellular carcinoma each year. Less frequent liver tumours are cholangiocarcinoma (CC) in adults and hepatoblastoma (HB) in children.
The prognosis and treatment options associated with these different kinds of cancers is difficult to predict, and is dependent in particular on the stage of the cancer (such as the size of the tumor, whether it affects part or all of the liver, has spread to other places in the body or its aggressiveness). Therefore, it is important for clinicians and physicians to establish a classification of primary liver cancers (HCC or HB) to propose the most appropriate treatment and adopt the most appropriate surgery strategy. Some factors are currently used (degree of local invasion, histological types of cancer with specific grading, tumour markers and general status of the patient) but have been found to not be accurate and sufficient enough to ensure a correct classification.
As far as the HB is concerned, the PRETEXT (pre-treatment extent of disease) system designed by the International Childhood Liver Tumor Strategy Group (SIOPEL) is a non invasive technique commonly used by clinicians, to assess the extent of liver cancer, to determine the time of surgery and to adapt the treatment protocol. This system is based on the division of the liver in four parts and the determination of the number of liver sections that are free of tumor (Aronson et al. 2005; Journal of Clinical Oncology; 23 (6): 1245-1252). A revised staging system taking into account other criteria, such as caudate lobe involvement, extrahepatic abdominal disease, tumor focality, tumor rupture or intraperitoneal haemorrhage, distant metastases, lymph node metastases, portal vein involvement and involvement of the IVC (inferior vena cava) and/or hepatic veins, has been recently proposed (Roebuck; 2007; Pediatr Radiol; 37: 123-132). However, the PRETEXT system, even if reproducible and providing good prognostic value, is based on imaging and clinical symptoms, making this system dependent upon the technicians and clinicians. There is thus a need for a system, complementary to the PRETEXT system, based on genetic and molecular features of the liver tumors.
The present invention concerns a method or process of profiling gene expression for a set of genes, in a sample previously obtained from a patient diagnosed for a liver tumor. In a particular embodiment said method is designed to determine the grade of a liver tumor in a patient.
By “liver tumor” or “hepatic tumor”, it is meant a tumor originating from the liver of a patient, which is a malignant tumor (comprising cancerous cells), as opposed to a benign tumor (non cancerous) which is explicitly excluded. Malignant liver tumors encompass two main kinds of tumors: hepatoblastoma (HB) or hepatocellular carcinoma (HCC). These two tumor types can be assayed for the presently reported molecular signature. However, the present method may also be used to assay malignant liver tumors which are classified as unspecified (non-HB, non-HCC).
The present method may be used to determine the grade of a liver tumor or several liver tumors of the same patient, depending on the extent of the liver cancer. For convenience, the expression “a liver tumor” will be used throughout the specification to possibly apply to “one or several liver tumor(s)”. The term “neoplasm” may also be used as a synonymous of “tumor”.
In a particular embodiment, the tumor whose grade has to be determined is located in the liver. The presence of the tumor(s) in the liver may be diagnosed by ultrasound scan, x-rays, blood test, CT scans (computerised tomography) and/or MRI scans (magnetic resonance imaging).
In a particular embodiment, the tumor, although originating from the liver, has extended to other tissues or has given rise to metastasis.
In a particular embodiment, the patient is a child i.e., a human host who is under 20 years of age according to the present application. Therefore, in a particular embodiment, the liver tumor is a paediatric HB or a paediatric HCC. In another embodiment, the liver tumor is an adult HCC.
A grade is defined as a subclass of the liver tumor, corresponding to prognostic factors, such as tumor status, liver function and general health status. The present method of the invention allows or at least contributes to differentiating liver tumors having a good prognosis from tumors with a bad prognosis, in terms of evolution of the patient's disease. A good prognosis tumor is defined as a tumor with good survival probability for the patient (more than 80% survival at two years for HB and more than 50% survival at two years for HCC), low probability of metastases and good response to treatment for the patient. In contrast, a bad prognosis tumor is defined as a tumor with an advanced stage, such as one having vascular invasion or/and extrahepatic metastasis, and associated with a low survival probability for the patient (less than 50% survival in two years).
The method of the invention is carried out on a sample isolated from the patient who has previously been diagnosed for the tumor(s) and who, optionally, may have been treated by surgery. In a preferred embodiment, the sample is the liver tumor (tumoral tissue) or of one of the liver tumors identified by diagnosis imaging and obtained by surgery or a biopsy of this tumor. The tumor located in the liver tumor is called the primary tumor.
In another embodiment, the sample is not the liver tumor, but is representative of this tumor. By “representative”, it is meant that the sample is regarded as having the same features as the primary tumors, when considering the gene expression profile assayed in the present invention. Therefore, the sample may also consist of metastatic cells (secondary tumors spread into different part(s) of the body) or of a biological fluid containing cancerous cells (such as blood).
The sample may be fixed, for example in formalin (formalin fixed). In addition or alternatively, the sample may be embedded in paraffin (paraffin-embedded) or equivalent products. In particular, the tested sample is a formalin-fixed, paraffin-embedded (FFPE) sample.
One advantage of the method of the present invention is that, despite the possible heterogeneity of some liver tumors (comprising epithelial tumor cells at different stages of liver differentiation within the same tumor), the assay has proved to be reproducible and efficient on liver tumor biopsies obtained from any part of the whole tumor. Therefore, there is no requirement for the isolation of cells presenting particular features except from the fact that they are obtained from a liver tumor or are representative thereof, to carry out the gene expression profile assay.
In a particular embodiment, the tumor originates from a patient having a Caucasian origin, in particular European, North American, Australian, New-Zealander or Afrikaners.
In a first step, the method or process of the invention comprises assaying the expression level of a set of genes in a sample, in order to get an expression profile thereof.
By “expression of a set of genes” (or “gene expression”), it is meant assaying, in particular detecting, the product or several products resulting from the expression of a gene, this product being in the form of a nucleic acid, especially RNA, mRNA, cDNA, polypeptide, protein or any other formats. In a particular embodiment, the assay of the gene expression profile comprises detecting a set of nucleotide targets, each nucleotide target corresponding to the expression product of a gene encompassed in the set.
The expression “nucleotide target” means a nucleic acid molecule whose expression must be measured, preferably quantitatively measured. By “expression measured”, it is meant that the expression product(s), in particular the transcription product(s) of a gene, are measured. By “quantitative” it is meant that the method is used to determine the quantity or the number of copies of the expression products, in particular the transcription products or nucleotide targets, originally present in the sample. This must be opposed to the qualitative measurement, whose aim is to determine the presence or absence of said expression product(s) only.
A nucleotide target is in particular a RNA, and most particularly a total RNA. In a preferred embodiment, the nucleotide target is mRNA or transcripts. According to the methods used to measure the gene expression level, the mRNA initially present in the sample may be used to obtain cDNA or cRNA, which is then detected and possibly measured.
In an embodiment, the expression of the gene is assayed directly on the sample, in particular in the tumor. In an alternative embodiment, the expression products or the nucleotide targets are prepared from the sample, in particular are isolated or even purified. When the nucleotide targets are mRNA, a further step comprising or consisting in the retro-transcription of said mRNA into cDNA (complementary DNA) may also be performed prior to the step of detecting expression. Optionally, the cDNA may also be transcribed in vitro to provide cRNA.
During the step of preparation, and before assaying the expression, the expression product(s) or the nucleotide target(s) may be labelled, with isotopic (such as radioactive) or non isotopic (such as fluorescent, coloured, luminescent, affinity, enzymatic, magnetic, thermal or electrical) markers or labels.
It is noteworthy that steps carried out for assaying the gene expression must not alter the qualitative or the quantitative expression (number of copies) of the expression product(s) or of the nucleotide target(s), or must not interfere with the subsequent step comprising assaying the qualitative or the quantitative expression of said expression product(s) or nucleotide target(s).
The step of profiling gene expression comprises determining the expression of a set of genes. Such a set is defined as a group of genes that must be assayed for one test, and especially performed at the same time, on the same patient's sample. A set comprises at least 2 and has especially from 2 to 16 genes, said 2 to 16 genes being chosen from the 16 following genes: alpha-fetoprotein (AFP), aldehyde dehydrogenase 2 (ALDH2), amyloid P component serum (APCS), apolipoprotein C-IV (APOC4), aquaporin 9 (AQP9), budding uninhibited by benzimidazoles 1 (BUB1), complement componant 1 (C1S), cytochrome p450 2E1 (CYP2E1), discs large homolog 7 (DLG7), dual specificity phosphatase 9 (DUSP9), E2F5 transcription factor (E2F5), growth hormone receptor (GHR), 4-hydroxyphenylpyruvase dioxygenase (HPD), immunoglogulin superfamily member 1 (IGSF1), Notchless homolog 1 (NLE1) and the ribosomal protein L10a (RPL10A) genes.
A complete description of these 16 genes is given in Table 1. This table lists, from left to right, the symbol of the gene, the complete name of the gene, the number of the SEQ ID provided in the sequence listing, the Accession Number from the NCBI database on June 2008, the human chromosomal location and the reported function (when known).
A set of genes comprises at least 2 out the 16 genes of Table 1, and particularly at least or exactly 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 or 15 out of the 16 genes of Table 1. In a particular embodiment, the set comprises or consists of the 16 genes of Table 1 i.e. the set of genes comprises or consists of AFP, ALDH2, APCS, APOC4, AQP9, BUB1, C1S, CYP2E1, DLG7, DUSP9, E2F5, GHR, HPD, IGSF1, NLE1 and RPL10A genes. Accordingly, unless otherwise stated when reference is made in the present application to a set of 2 to 16 genes of Table 1, it should be understood as similarly applying to any number of genes within said 2 to 16 range.
In other particular embodiments, the set of genes comprises or consists of one of the following sets: (a) the E2F5 and HPD genes, (b) the APCS, BUB1, E2F5, GHR and HPD genes, (c) the ALDH2, APCS, APOC4, BUB1, C1S, CYP2E1, E2F5, GHR and HPD genes, (d) the ALDH2, APCS, APOC4, AQP9, BUB1, C1S, DUSP9, E2F5 and RPL10A genes, or (e) the ALDH2, APCS, APOC4, AQP9, C1S, CYP2E1, E2F5, GHR, IGSF1 and RPL10A genes.
As indicated by the expression “comprises from 2 to 16 genes of Table 1”, the set may, besides the specific genes of Table 1, contain additional genes not listed in Table 1. This means that the set must comprises from 2 to 16 genes of Table 1, i.e. 2 to 16 genes of Table 1 (in particular 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, or 16 genes), and optionally comprises one or more additional genes. Said set may also be restricted to said 2 to 16 genes of Table 1.
Additional genes may be selected for the difference of expression observed between the various grades of liver cancer, in particular between a tumor of good prognosis and a tumor of poor prognosis.
The invention also relates to a set of genes comprising or consisting of the 16 genes of Table 1 (i.e., AFP, ALDH2, APCS, APOC4, AQP9, BUB1, C1S, CYP2E1, DLG7, DUSP9, E2F5, GHR, HPD, IGSF1, NLE1 and RPL10A genes), in which 1, 2, 3, 4 or 5 genes out of the 16 genes are substituted by a gene presenting the same features in terms of difference of expression between a tumor of a good prognosis and a tumor of poor prognosis.
In a particular embodiment, the number of genes of the set does not exceed 100, particularly 50, 30, 20, more particularly 16 and even more particularly is maximum 5, 6, 7, 8, 9 or 10.
When considering adding or substituting a gene or several genes to the disclosed set, the person skilled in the art will consider one or several of the following features:
Examples of genes which can be added or may replace genes of the set may be identified in following Table 2.
The Accession Numbers of the genes of Table 2. as found in NCBI database in June 2008, are the following: IPO4 (BC136759), CPSF1 (NM_013291), MCM4 (NM_005914.2; NM_182746.1; two accession numbers for the same gene correspond to 2 different isoforms of the gene), EIF3S3 (NM_003756.2), NCL (NM_005381.2), CDC25C (NM_001790.3), CENPA (NM_001809.3; NM_001042426.1), K1F14 (BC113742), IPW
In a particular embodiment of the invention, the set of genes of the invention is designed to determine the grade of hepatoblastoma, in particular paediatric hepatoblastoma. In another embodiment, the set of genes is designed to determine the grade of hepatocellular carcinoma, in particular paediatric HCC or adult HCC.
The expression of the genes of the set may be assayed by any conventional methods, in particular any conventional methods known to measure the quantitative expression of RNA, preferably mRNA.
The expression may be measured after carrying out an amplification process, such as by PCR, quantitative PCR (qPCR) or real-time PCR. Kits designed for measuring expression after an amplification step are disclosed below.
The expression may be measured using hybridization method, especially with a step of hybridizing on a solid support, especially an array, a macroarray or a microarray or in other conditions especially in solution. Arrays and kits of the invention, designed for measuring expression by hybridization method are disclosed below.
The expression of a gene may be assayed in two manners:
In the present invention, the expression which is assayed is preferably the relative expression of each gene, calculated with reference to at least one (preferably 1, 2, 3 or 4) invariant gene(s). Invariant genes, suitable to perform the invention, are genes whose expression is constant whatever the grade of the liver tumors, such as for example ACTG1, EFF1A1, PNN and RHOT2 genes, whose features are summarized in Table 3. In a particular embodiment preferred, the relative expression is calculated with respect to at least the RHOT2 gene or with respect to the RHOT2 gene.
In another advantageous embodiment, the relative expression is calculated with respect to at least the PNN gene or with respect to the PNN gene. It may be calculated with respect to the RHOT2 and PNN genes.
The calculation of the absolute expression or of the relative expression of each gene of the set and of each invariant gene being assayed with the same method from the same sample, preferably at the same time, enables to determine for each sample a gene expression profile.
An additional step of the method or process comprises the determination of the grade of said liver tumor, referring to the gene expression profile that has been assayed. In a particular embodiment of the invention, the method is designed to determine the grade of hepatoblastoma, in particular paediatric hepatoblastoma. In another embodiment, the method is designed to determine the grade of hepatocellular carcinoma, in particular paediatric HCC or adult HCC.
According to a particular embodiment of the invention, in the step of the method which is performed to determine the grade of the liver tumor, a gene expression profile or a signature (preferably obtained after normalization), which is thus specific for each sample, is compared to the gene expression profile of a reference sample or to the gene expression profiles of each sample of a collection of reference samples (individually tested) whose grade is known, so as to determine the grade of said liver tumor. This comparison step is carried out with at least one prediction algorithm. In a particular embodiment, the comparison step is carried out with 1, 2, 3, 4, 5 or 6 prediction algorithms chosen in the following prediction algorithms: Compound Covariate Predictor (CCP), Linear Discriminator Analysis (LDA), One Nearest Neighbor (1NN), Three Nearest Neighbor (3NN), Nearest Centroid (NC) and Support Vector Machine (SVM). These six algorithms are part of the “Biometric Research Branch (BRB) Tools” developed by the National Cancer Institut (NCI) and are available on http://linus.nci.nih.gov/BRB-ArrayTools.html. Equivalent algorithms may be used instead of or in addition to the above ones. Each algorithm classifies tumors within either of the two groups, defined as tumors with good prognosis (such as C1) or tumors with bad prognosis (such as C2); each group comprises the respective reference samples used for comparison, and one of these two groups also comprises the tumor to be classified.
Therefore, when 6 algorithms are used, the grade of a tumor sample may be assigned with certainty to the class of good prognosis or to the class of bad prognosis, when 5 or 6 of the above algorithms classified the tumor sample in the same group. In contrast, when less than 5 of the above algorithms classify a tumor sample in the same group, it provides an indication of the grade rather than a definite classification.
Reference samples which can be used for comparison with the gene expression profile of a tumor to be tested are one or several sample(s) representative for tumor with poor prognosis (such as C2), one or several sample(s) representative of tumor with good prognosis (such as C1), one or several sample(s) of a normal adult liver and/or one or several sample(s) of a fetal liver.
Table 4 lists the level of expression of each gene of Table 1 depending upon the status of the reference sample i.e., robust tumor with poor prognostic and robust tumor with good prognostic. Examples of methods to identify such robust tumors are provided in the examples. The present invention provides a new classification method in this respect, which is based on discretization of continuous values.
Reference samples usually correspond to so-called “robust tumor” for which all the marker genes providing the signature are expressed (either under expressed or overexpressed) as expected i.e., in accordance with the results disclosed in Table 5, when tested in similar conditions, as disclosed in the examples hereafter.
A robust tumor having an overexpression of one or several gene(s) selected among ALDH2, APCS, APOC4, AQP9, C1S, CYP2E1, GHR and HPD genes (these genes belong to the so-called group of differentiation-related genes), and/or an underexpression of one or several gene(s) selected among AFP, BUB1, DLG7, DUSP9, E2F5, IGSF1, NLE1 and RPL10A genes (these genes belong to the so-called group of proliferation-related genes), is an indicator of a robust liver tumor, in particular of a hepatoblastoma, with a good prognosis. A robust tumor having an overexpression of one or several gene(s) selected among AFP, BUB1, DLG7, DUSP9, E2F5, IGSF1, NLE1 and RPL10A genes, and/or an underexpression of one or several gene(s) among ALDH2, APCS, APOC4, AQP9, C1S, CYP2E1, GHR and HPD genes, is an indicator of a robust liver tumor, in particular of a hepatoblastoma, with a poor prognosis. In the present application, a gene is said “underexpressed” when its expression is lower than the expression of the same gene in the other tumor grade, and a gene is said “overexpressed” when its expression is higher than the expression of the same gene in the other tumor grade.
In a particular embodiment, Table 5 provides the gene expression profiles of the 16 genes of Table 1 in 13 samples of hepatoblastoma (HB) including 8 samples that have been previously identified as rC1 subtype and 5 samples that have been previously identified as rC2 subtype. This Table can therefore be used for comparison, to determine the gene expression profile of a HB tumor to be classified, with the robust tumors disclosed (constituting reference samples), for a set of genes as defined in the present application. Said comparison involves using the classification algorithms which are disclosed herein, for both the selected reference samples and the assayed sample.
Normalized qPCR data of 16 genes in 13 HB samples including 8 samples of the rC1 subtype and 5 samples of the rC2 subtype (in grey). The qpCR values have been obtained by measuring the expression of the 16 genes in 8 samples of the rC1 subtype and 5 samples of the rC2 subtype by the SYBR green method using the primers as disclosed in Table 6 below and in the conditions reported in the examples, and normalized by the ROTH2 gene (primers in Table 7).
The method of the present invention is also suitable to classify new tumor samples, and to use them as new reference samples. Therefore, the gene expression values of these new reference samples may be used in combination or in place of some of the values reported in Table 5.
In another embodiment of the invention, the step of determining the tumor grade comprises performing a method of discretization of continuous values of gene expression obtained on the set of genes the tested patients' samples. Discretization is generally defined as the process of transforming a continuous-valued variable into a discrete one by creating a set of contiguous intervals (or equivalently a set of cutpoints) that spans the range of the variable's values. Discretization has been disclosed for use in classification performance in Lustgarten J. L. et al, 2008.
The inventors have observed that discretization can be effective in determining liver tumor grade, especially for those tumors described in the present application, including Hepatoblastoma (HB) or Hepatocellular carcinoma (HCC).
The discretization method is especially disclosed in the examples where it is illustrated by using data obtained on tumor samples wherein these data are those obtained from profiling the 16 genes providing the large set of genes for expression profiling according to the invention. It is pointed out that the discretization method may however be carried out on a reduced number of profiled genes within this group of 16 genes, starting from a set consisting of 2 genes (or more genes) including one (or more) overexpressed proliferation-related genes chosen among AFP, BUB1, DLG7, DUSP9, E2F5, IGSF1, NLE1 and RPL10A and one down-regulated differentiation-related gene chosen among ALDH2, APCS, APOC4, AQP9, C1S, CYP2E1, GHR, HPD, said genes being thus classified as a result of gene profiles observed on robust tumors with poor prognosis (according to the classification in Table 4 above). In particular embodiments of the discretization method, the number of assayed gene for expression profiling is 2, 4, 6, 8, 10, 12, 14 or 16 and the same number of genes in each category (either the group of overexpressed proliferation-related genes or the group of downregulated differentiation-related gene) is used to perform the method.
The invention thus relates to a method enabling the determination of the tumor grade on a patient's sample, which comprises a classification of the tumor through discretization according to the following steps:
The above defined ratio of average values may be alternatively calculated as the ratio of the average for the discresized values for the differentiation-related genes on the average for the discretized values for the proliferation-related genes, to obtain a score. If this calculation made is adopted the cut-offs values are inversed, i.e., are calculated as 1/xxx.
In order to carry out the discretization method of the invention, the data obtained on the assayed genes for profiling a patient's sample are preferably normalized with respect to one or more invariant gene(s) of the present invention, in order to prevent detrimental impact on the results that may arise from possible inaccuracy in the quantification of initial nucleic acid, especially RNA, in the sample.
Normalization with respect to one invariant gene only, especially when said invariant gene is RHOT2 gene has proved to be relevant in the results obtained by the inventors. Similarly normalization with respect to PNN gene would be an advantageous possibility because the gene does also not vary in expression.
In order to design a discretization method for the determination of tumor grade of an individual sample of a patient, according to the invention, cut-offs values have to be determined to allow the determination of the tumor grade. The cut-offs values can be determined experimentally by carrying out the following steps on expression profiling results obtained on a determined number of tumor samples:
Therefore the invention provides cut-offs values as reference cut-offs, in order to carry out the determination of tumor grade in particular testing conditions as those disclosed below and in the examples.
In a particular embodiment of the method of discretization, the cut-off for each gene is the value corresponding to a determined percentile, which can be different for each of the considered two groups of genes (proliferation-related genes on the one hand and differentiation-related genes on the other hand). The selected percentile (or quantile) is determined with respect to the fraction of tumors (such as ⅓ or more) harbouring some chosen features such as overexpression of proliferation-related genes and/or dowregulation of differentiation-related genes, in the two groups of genes of the set of genes. Especially, when one intends to assign more weight to tumors displaying strong overexpression of proliferation-related genes and/or strong downregulation of differentiation-related genes, the cut-off corresponds to a high quantile (above the 50th, preferably the 60th, or even above the 65th, such as the 67th and for example within the range of 55th and 70th) for said proliferation-related genes and the cut-off corresponds to a low quantile (below the 50th, preferably equal to or below the 40th for example the 33rd, and for example within the range of between 20th and 40th) of the differentiation-related genes. The cut-off for each group of genes and the cut-off for the sample may be determined with respect to the same percentile(s) or may be determined with respect to different percentile.
According to a particular embodiment of the invention, for HB tumors, the percentile which is chosen for the overexpressed proliferation-related genes is the 67th and the percentile which is chosen for the downregulated differentiation-related genes is the 33rd. According to a particular embodiment of the invention, for HC tumors, the percentile which is chosen for the overexpressed proliferation-related genes is the 60th and the percentile which is chosen for the downregulated differentiation-related genes is the 40rd.
Each percentile (or cut-off value corresponding to the percentile) defines a cutpoint and the discretized values for each gene are either “1” or “2” below or above said percentile. The values “1” and “2” are distributed with respect to the percentiles so as to create the highest difference in the values of the calculated ratio for the most different tumor grades. This is illustrated in the examples for the selected percentiles.
It has been observed that in a preferred embodiment of the invention, the relative values of the profiled genes are determined by real-time PCR (qPCR).
Conditions to carry out the real-time PCR are disclosed herein, especially in the examples, as conditions applicable to analyzed samples.
PCR primers and probes suitable for the performance of RT-PCR are those disclosed herein for the various genes.
In a particular embodiment of the invention, the analysed tumor is a hepatoblastoma and its grade is determined by discretization as disclosed above and illustrated in the examples, taking into account that:
In another embodiment of the invention, the tumor is an hepatocellular carcinoma and its grade is determined by discretization as disclosed above and illustrated in the examples, taking into account that:
the cut-offs value for each gene based on −dCt (minus delta Ct) measures) are:
It is observed that the refinement of the results which are between the cut-offs of the samples is advantageous for hepatocellular carcinoma in order to increase the relevancy of the information on the tumor grade.
Generally said refinement of the classification of the intermediate results in the HCC is obtained by performing the following steps:
a modified score is determined which corresponds to the average of the discretized values of the “proliferation-related genes” only for the sample. A new cut-off value is determined for said genes, which is the cut-off value for the modified score (in the present case it is 1.3). This cut-off can be determined via a percentile (here the 60th) of the distribution of the modified scores, using the samples of the intermediate class. A sample (initially classified in the intermediate class) with a modified score below the “proliferation cut-off” (for example 1.3) can be re-classified into the C1 class, and a sample with a modified score above the “proliferation cut-off” (for example 1.3) can be re-classified into the C2 class.
From the 16 genes expressed in liver cells listed in Table 1, a set comprising from 2 to 16 genes (or more generally a set as defined herein) may be used to assay the grade of tumor cells in a tumor originating from the liver. The results obtained, after determining the expression of each of the genes of the set, are then treated for classification according to the steps disclosed herein. The invention relates to each and any combination of genes disclosed in Table 1, to provide a set comprising from 2 to 16 of these genes, in particular a set comprising or consisting of 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 or 16 of these genes. In the designed set, one or many genes of Table 1 may be modified by substitution or by addition of one or several genes as explained above, which also enable to determine the grade of the liver tumor, when assayed in combination with the other genes.
In a preferred embodiment, the liver tumor is a paediatric HB, and the method or process of the invention enables to distinguish a first class, called C1, qualifying as a good prognosis tumor and a second class, called C2, qualifying as a poor prognosis tumor. The C1 grade is predominantly composed of fetal histotype cells (i.e., well differentiated and non proliferative cells). In contrast, the C2 grade presents cells other than the fetal histotype such as embryonic, atypic (crowded fetal), small cell undifferiantiated (SCUD) and/or macrotrabecular cells.
The present invention also relates to a kit suitable to determine the grade of a liver tumor from the sample obtained from a patient. This kit is appropriate to carry out the method or process described in the present application.
In a particular embodiment, the kit comprises a plurality of pairs of primers specific for a set of genes to be assayed, said set comprising from 2 to 16 genes, said 2 to 16 genes being chosen in the group consisting of AFP, ALDH2, APCS, APOC4, AQP9, BUB1, C1S, CYP2E1, DLG7, DUSP9, E2F5, GHR, HPD, IGSF1, NLE1 and RPL10A genes.
By “plurality”, it is mean that the kit comprises at least as many pairs of primers as genes to enable assaying each selected gene, and in particular the nucleotide target of this gene. Accordingly, each gene and in particular its nucleotide target is specifically targeted by a least one of these pairs of primers. In a particular embodiment, the kit comprises the same number of pairs of primers as the number of genes to assay and each primer pair specifically targets one of the genes, and in particular the nucleotide targets of one of these genes, and does not hybridize with the other genes of the set.
The kits of the invention are defined to amplify the nucleotide targets of the sets of genes as described in the present invention. Therefore, the kit of the invention comprises from 2 to 16 pairs of primers which, when taken as a whole, are specific for said from 2 to 16 genes out of the 16 genes of Table 1. In particular, the kit comprises or consists of 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 or 15 pairs of primers specific for 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 or 15 out of the 16 genes of Table 1. In a particular embodiment, the kit comprises or consists of 16 pairs of primers specific for the 16 genes of Table 1 i.e., a primer pair specific for each of the following genes: AFP, ALDH2, APCS, APOC4, AQP9, BUB1, C1S, CYP2E1, DLG7, DUSP9, E2F5, GHR, HPD, IGSF1, NLE1 and RPL10A genes.
When the set of genes has been modified by the addition or substitution of at least one gene as described above, the kit is adapted to contain a pair of primers specific for each added or substituted gene(s). As indicated by the term “comprises”, the kit may, besides the pairs of primers specific for the genes of Table 1, contain additional pair(s) of primers.
In a particular embodiment, the kit comprises at least one pair of primers (preferably one) for at least one invariant gene (preferably one or two) to be assayed for the determination of the expression profile of the genes, by comparison with the expression profile of the invariant gene.
The number of pairs of primers of the kit usually does not exceed 100, particularly 50, 30, 20, more particularly 16, and even more particularly is maximum 5, 6, 7, 8, 9 or 10.
In the kits of the invention, it is understood that, for each gene, at least one pair of primers and preferably exactly one pair, enabling to amplify the nucleotide targets of this gene, is present. When the kits provide several pairs of primers for the same gene, the gene expression level is measured by amplification with only one pair of primers. It is excluded that amplification may be performed using simultaneously several pairs of primers for the same gene.
As defined herein, a pair of primers consists of a forward polynucleotide and a backward polynucleotide, having the capacity to match its nucleotide target and to amplify, when appropriate conditions and reagents are brought, a nucleotide sequence framed by their complementary sequence, in the sequence of their nucleotide target.
The pairs of primers present in the kits of the invention are specific for a gene i.e., each pair of primers amplifies the nucleotide targets of one and only one gene among the set. Therefore, it is excluded that a pair of primers specific for a gene amplifies, in a exponential or even in a linear way, the nucleotide targets of another gene and/or other nucleic acids contained in sample. In this way, the sequence of a primer (whose pair is specific for a gene) is selected to be not found in a sequence found in another gene, is not complementary to a sequence found in this another gene and/or is not able to hybridize in amplification conditions as defined in the present application with the sequence of the nucleotide targets of this another gene.
In a particular embodiment, the forward and/or backward primer(s) may be labelled, either by isotopic (such as radioactive) or non isotopic (such as fluorescent, biotin, fluororochrome) methods. The label of the primer(s) leads to the labelling of the amplicon (product of amplification), since the primers are incorporated in the final product.
The design of a pair of primers is well known in the art and in particular may be carried out by reference to Sambrook et al. (Molecular Cloning, A laboratory Manual, Third Edition; chapter 8 and in particular pages 8.13 to 8.16). Various softwares are available to design pairs of primers, such as Oligo™ or Primer3.
Therefore, each primer of the pair (forward and backward) has, independently from each other, the following features:
In a particular embodiment, when the pairs of primers are used in a simultaneous amplification reaction carried out on the sample, the various primers have the capacity to hybridize with their respective nucleotide targets at the same temperature and in the same conditions.
Conventional conditions for PCR amplification are well known in the art and in particular in Sambrook et al. An example of common conditions for amplification by PCR is dNTP (200 mM), MgCl2 (0.5-3 mM) and primers (100-200 nM).
In a particular embodiment, the sequence of the primer is 100% identical to one of the strands of the sequence of the nucleotide target to which it must hybridize with, i.e. is 100% complementary to the sequence of the nucleotide target to which it must hybridize. In another embodiment, the identity or complementarity is not 100%, but the similarity is at least 80%, at least 85%, at least 90% or at least 95% with its complementary sequence in the nucleotide target. In a particular embodiment, the primer differs from its counterpart in the sequence of the sequence of the nucleotide target by 1, 2, 3, 4 or 5 mutation(s) (deletion, insertion and/or substitution), preferably by 1, 2, 3, 4 or 5 nucleotide substitutions. In a particular embodiment, the mutations are not located in the last 5 nucleotides of the 3′ end of the primer.
In a particular embodiment, the primer, which is not 100% identical or complementary, keeps the capacity to hybridize with the sequence of the nucleotide target, similarly to the primer that is 100% identical or 100% complementary with the sequence of the nucleotide target (in the hybridization conditions defined herein). In order to be specific, at least one of the primers (having at least 80% similarity as defined above) of the pair specific for a gene can not hybridize with the sequence found in the nucleotide targets of another gene of the set and of another gene of the sample.
In a particular embodiment, the pairs of primers used for amplifying a particular set of genes are designed, besides some or all of the features explained herein, in order that the amplification products (or amplicons) of each gene have approximately the same size. By “approximately” is meant that the difference of size between the longest amplicon and the shortest amplicon of the set is less than 30% (of the size of the longest amplicon), preferably less than 20%, more preferably less than 10%. As particular embodiments, the size of the amplicon is between 100 and 300 bp, such as about 100, 150, 200, 250 or 300 bp.
The nucleotide sequences of the 16 genes of Table 1 are provided in the Figures, and may be used to design specific pairs of primers for amplification, in view of the explanations above.
Examples of primers that may be used to measure the expression of the genes of Table 1, in particular to amplify the nucleotide targets of the genes of Table 1, are the primers having the sequence provided in Table 6 or variant primers having at least 80% similarity (or more as defined above) with the sequences defined in Table 6.
The kit of the invention may further comprise one or many pairs of primers specific for one or many invariant genes, in particular specific for ACTG1, EFF1A1, PNN and/or RHOT2 genes. The pair of primers specific for invariant gene(s) may be designed and selected as explained above for the pair of primers specific for the genes of the set of the invention. In a particular embodiment, the pairs of primers of the invariant genes are designed in order that their amplification product (or amplicon) has approximately the same size as the amplicon of the genes of the set to be assayed (the term approximately being defined as above, with respect to the longest amplicon of the set of genes). Examples of primers that may be used to amplify the particular invariant genes are primers having the sequence provided in Table 7 or primers having at least 80% similarity (or more as defined above) with the sequences defined in Table 7.
The kits of the invention may also further comprise, in association with or independently of the pairs of primers specific for the invariant gene(s), reagents necessary for the amplification of the nucleotide targets of the sets of the invention and if any, of the nucleotide targets of the invariant genes.
The kits of the invention may also comprise probes as disclosed herein in the context of sets of probes, compositions and arrays. In particular, the kits also comprise the four dNTPs (nucleotides), amplification buffer, a polymerase (in particular a DNA polymerase, and more particularly a thermostable DNA polymerase) and/or salts necessary for the activity of the polymerase (such as Mg2+).
Finally, the kits may also comprise one or several control sample(s) i.e., at least one sample(s) representative of tumor with bad (i.e., poor) prognosis (in particular a HB C2 grade), at least one sample(s) representative of tumor with good prognosis (in particular a HB C1 grade), at least one sample of a normal adult liver and/or at least one sample of a fetal liver.
The kits may also comprise instructions to carry out the amplification step or the various steps of the method of the invention.
The invention is also directed to a set of probes suitable to determine the grade of a liver tumor from the sample obtained from a patient. This set of probes is appropriate to carry out the method or process described in the present invention. It may also be part of the kit.
This set of probes comprises a plurality of probes in particular from 2 to 16 probes, these 2 to 16 probes being specific for genes chosen in the group consisting of AFP, ALDH2, APCS, APOC4, AQP9, BUB1, C1S, CYP2E1, DLG7, DUSP9, E2F5, GHR, HPD, IGSF1, NLE1 and RPL10A genes.
By “plurality”, it is mean that the set of probes comprises at least as many probes as genes to assay. In a particular embodiment, the array comprises the same number of probes as the number of genes to assay.
The probes of the sets of the invention are selected for their capacity to hybridize to the nucleotide targets of the sets of genes as described in the present invention. Therefore, the set of probes of the invention comprise from 2 to 16 probes specific for 2 to 16 genes out of the 16 genes of Table 1. In particular, the sets of probes comprise or consist of 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 or 15 probes specific of 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 or 15 out of the 16 genes of Table 1. In a particular embodiment, the sets of probes comprise or consist of 16 probes specific for the 16 genes of Table 1 i.e., a probe specific of each of the following genes: AFP, ALDH2, APCS, APOC4, AQP9, BUB1, C1S, CYP2E1, DLG7, DUSP9, E2F5, GHR, HPD, IGSF1, NLE1 and RPL10A genes.
The specificity of the probes is defined according to the same parameters as those applying to define specific primers.
When the set of genes has been modified by the addition or substitution of at least one gene as described above, the set of probes is adapted to contain a probe specific for the added or substituted gene(s). As indicated by the term “comprises”, the set of probes may, besides the probes specific for the genes of Table 1, contain additional probe(s).
The number of probes of the set does usually not exceed 100, particularly 50, 30, 20, more particularly 16, and even more particularly is maximum 5, 6, 7, 8, 9 or 10.
In the set of probes of the invention, it is understood that for each gene corresponds at least one probe to which the nucleotide target of this gene hybridize to. The set of probes may comprise several probes for the same gene, either probes having the same sequence or probes having different sequences.
As defined herein, a probe is a polynucleotide, especially DNA, having the capacity to hybridize to the nucleotide target of a gene. Hybridization is usually carried out at a temperature ranging from 40 to 60° C. in hybridization buffer (see example of buffers below). These probes may be oligonucleotides, PCR products or cDNA vectors or purified inserts. The size of each probe is independently to each other from 15 and 1000 bp, preferably 100 to 500 bp or 15 to 500 bp, more preferably 50 to 200 bp or 15 to 100 bp. The design of probes is well known in the art and in particular may be carried out by reference to Sambrook et al. (Molecular Cloning, A laboratory Manual, Third Edition; chapters 9 and 10 and in particular pages 10.1 to 10.10).
The probes may be optionally labelled, either by isotopic (radioactive) or non isotopic (biotin, fluororochrome) methods. Methods to label probes are disclosed in Sambrook et al. (Molecular Cloning, A laboratory Manual, Third Edition; chapter 8 and in particular page 9.3). In a particular embodiment, the probes are modified to confer them different physicochemical properties (such as by methylation, ethylation). In another particular embodiment, the probes may be modified to add a functional group (such as a thiol group), and optionally immobilized on bead (preferably glass beads).
In a particular embodiment, the sequence of the probe is 100% identical to a part of one strand of the sequence of the nucleotide target to which it must hybridize, i.e. is 100% complementary to a part of the sequence of the nucleotide target to which it must hybridize. In another embodiment, the identity or complementarity is not 100% and the similarity is at least 80%, at least 85%, at least 90% or at least 95% with a part of the sequence of the nucleotide target. In a particular embodiment, the probe differs from a part of one strand of the sequence of the nucleotide target by 1 to 10 mutation(s) (deletion, insertion and/or substitution), preferably by 1 to 10 nucleotide substitutions. By “a part of”, it is meant consecutive nucleotides of the nucleotide target, which correspond to the sequence of the probe.
In a particular embodiment, the probe, which is not 100% identical or complementary, keeps the capacity to hybridize, in particular to specifically hybridize, to the sequence of the nucleotide target, similarly to the probe which is 100% identical or 100% complementary with the sequence of the nucleotide target (in the hybridization conditions defined herein).
In a particular embodiment, the size of the probes used to assay a set of genes is approximately the same for all the probes. By “approximately” is meant that the difference of size between the longest probe and the shortest probe of the set is less than 30% (of the size of the longest probe), preferably less than 20%, more preferably less than 10%.
The set of probes of the invention may further comprise at least one (preferably one) probe specific for at least one invariant gene (preferably one or two), in particular specific for ACTG1, EFF1A1, PNN and/or RHOT2 genes. The probes specific for invariant gene(s) may be designed and selected as explained above for the probes specific for genes of the sets of the invention. In a particular embodiment, the probes specific of the invariant genes have approximately the same size as the probes specific of the genes of the set of be assayed (the term approximately being defined as above, with respect to the longest probes of the set of genes).
The invention is also directed to an array suitable to determine the grade of a liver tumor from the sample obtained from a patient. This array is appropriate to carry out the method or process described in the present application.
An array is defined as a solid support on which probes as defined above, are spotted or immobilized. The solid support may be porous or non-porous, and is usually glass slides, silica, nitrocellulose, acrylamide or nylon membranes or filters.
The arrays of the invention comprise a plurality of probes specific for a set of genes to be assayed. In particular, the array comprises, spotted on it, a set of probes as defined above.
The invention also relates to a composition comprising a set of probes as defined above in solution.
In a first embodiment, the probes (as defined above in the set of probes) may be modified to confer them different physicochemical properties (such as methylation, ethylation). The nucleotide targets (as defined herein and prepared from the sample) are linked to particles, preferably magnetic particles, for example covered with ITO (indium tin oxide) or polyimide. The solution of probes is then put in contact with the target nucleotides linked to the particles. The probe/target complexes are then detected, for example by mass spectrometry.
Alternatively, probes may be modified to add a functional group (such as a thiol group) and immobilized on beads (preferably glass beads). These probes immobilized on beads are put in contact with a sample comprising the nucleotide targets, and the probe/target complexes are detected, for example by capillary reaction.
The invention is also directed to kits comprising the sets of probes, the compositions or the arrays of the invention and preferably the primer pairs disclosed herein. These kits may also further comprise reagents necessary for the hybridization of the nucleotide targets of the sets of genes and/or of the invariant genes, to the probes (as such, in the compositions or on the arrays) and the washing of the array to remove unbound nucleotides targets.
In a particular embodiment, the kits also comprise reagents necessary for the hybridization, such as prehybridization buffer (for example containing 5×SSC, 0.1% SDS and 1% bovine serum albumin), hybridization buffer (for example containing 50% formamide, 10×SSC, and 0.2% SDS), low-stringency wash buffer (for example containing 1×SSC and 0.2% SDS) and/or high-stringency wash buffer (for example containing 0.1×SSC and 0.2% SDS).
The kits may also comprise one or several control sample(s) i.e., at least one sample(s) representative for tumor with poor prognosis, at least one sample(s) representative of tumor with good prognosis, at least one sample of a normal adult liver and/or at least one sample of a fetal liver. Alternatively, it may comprise the representation of a gene expression profile of such tumors.
Finally, the invention provides a kit as described above further comprising instructions to carry out the method or process of the invention.
The arrays and/or kits (either comprising pairs of primers or probes or arrays or compositions of the invention or all the components) according to the invention may be used in various aspects, in particular to determine the grade of a liver tumor from a patient, especially by the method disclosed in the present application.
The arrays and/or kits according to the invention are also useful to determine, depending upon the grade of the liver tumor, the risk for a patient to develop metastasis. Indeed, the classification of a liver tumor in the class with poor prognosis is highly associated with the risk of developing metastasis.
In another embodiment, the arrays and/or kits according to the invention are also useful to define, depending upon the grade of the liver tumor, the therapeutic regimen to apply to the patient.
The invention also relates to a support comprising the data identifying the gene expression profile obtained when carrying out the method of the invention.
The colour version of the drawings as filed is available upon request to the European Patent Office.
(A) Schematic overview of the approach used to identify robust clusters of samples, including two tumor clusters (rC1 and rC2) and one non-tumor cluster (NL) (B) Expression profiles of 982 probe sets (824 genes) that discriminate rC1 and rC2 samples (p<0.001, two-sample t test). Data are plotted as a heatmap where red and green correspond to high and low expression in log2-transformed scale. (C) Molecular classification of 25 HB samples and status of CTNNB1 gene and β-catenin protein. C1 and C2 classification was based on rC1 and rC2 gene signature by using six different statistical predictive methods (CCP, LDA, 1NN, 3NN, NC and SVM) and the leave-one-out cross-validation. Black and gray squares indicate mutations of the CTNNB1 and AXIN1 genes. Immunohistochemical analysis of β-catenin in representative C1 and C2 cases is shown. (D) Expression of representative Wnt-related and β-catenin target genes (p<0.005, two-sample t test) in HB subclasses and non-tumor livers (NL). (E) Classification of hepatoblastoma by expression profile of a 16-gene signature. (F) Classification of normal human livers of children with HB (from 3 months to 6 years of age) (NT) or fetal livers at 17 to 35 weeks of gestation (FL) by expression profile of a 16-gene signature.
Among the total series of 114 patients, survival data were not available for one case.
Patients were treated either by partial hepatectomy (PH) or by orthotopic liver transplantation (OLT). Unless specified, the follow-up was closed at 146 months.
A: HCC cases were classified into 3 classes by the discretization method using as cut-offs the 33rd and the 67th percentiles.
B: 47 HCC cases previously classified into the intermediate class (33<p<67, see pannel A) were subdivided into 2 new subclasses using the 60th percentile of proliferation-related genes.
C: 92 HCC cases treated by partial hepatectomy (PH) were classified into 3 classes as in pannel A.
D: 21 HCC cases treated by orthotopic liver transplantation (OLT) were classified into 2 classes using as cut-off the 67th percentile.
E: HCC cases were classified into 2 classes using different combinations of scores as described in Table F.
F: HCC cases were classified into 2 classes using as cut-off the 33rd percentile.
G: HCC cases were classified into 2 classes using as cut-off the 50th percentile.
H: HCC cases were classified into 2 classes using as cut-off the 67th percentile.
I: 92 HCC cases treated by partial hepatectomy (PH) were classified into 2 classes using as cut-off the 33rd percentile.
J: 92 HCC cases treated by partial hepatectomy (PH) were classified into 2 classes using as cut-off the 50th percentile.
K: 92 HCC cases treated by partial hepatectomy (PH) were classified into 2 classes using as cut-off the 67th percentile.
L: Disease-free survival of 113 HCC cases after classification into 2 classes using as cut-off the 67th percentile. Follow-up was closed at 48 months. Data were not significant when the follow-up was closed at 146 months.
M: Disease-free survival of 92 HCC cases treated by PH, after classification into 2 classes using as cut-off the 67th percentile. Follow-up was closed at 48 months. Data were not significant when the follow-up was closed at 146 months.
A. Patients and Tissue Samples.
Sixty-six tumor specimens and biopsies from 61 patients with hepatoblastoma were collected from different hospitals in France (52 cases), Italy (6 cases), United Kingdom (1 case), Switzerland (1 case) and Slovakia (1 case). Forty-eight patients received chemotherapy treatment prior to surgery, most being enrolled in clinical trials of the International Childhood Liver Tumour Strategy Group (SIOPEL) (Perilongo et al., 2000). Samples from fresh tumors avoiding fibrotic and necrotic areas and from adjacent non tumor livers were snap frozen at the time of surgery and stored at −80° C.
Patients were children with median age of 2 years, and male:female ratio of 1.5. The median follow-up was 32 months; during this period, 15 patients died from disease. The histology of all tumor specimens was centrally reviewed by expert pathologist according to previously described criteria (Finegold et al., 2007; Zimmermann, 2005). Twenty-five tumors were analyzed on oligonucleotide microarrays and 24 of them, for which DNA was available, were subjected to aCGH analysis, while a second set of 41 tumors was analyzed by qPCR (
B. Oligonucleotide Microarrays and Gene Expression Data Analysis
Twenty-five HB samples and 4 non-tumor samples including a pool of livers from 3 males and a second from 3 females were analyzed using Affymetrix HG-U133A oligonucleotide arrays. Total RNA was prepared using FastPrep® system (Qbiogene, Strasbourg, France) and RNeasy mini Kit (Qiagen, Courtaboeuf, France). RNA quality was checked with the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, Calif.). Microarray experiments were performed according to the manufacturer's instructions. Affymetrix microarray data were normalized using RMA method (Irizarry et al., 2003). Class discovery was done as described elsewhere (Lamant et al., 2007). Pathway and Gene Ontology enrichment analyses were performed using GSEA method (Subramanian et al., 2005) and hypergeometric tests. For supervised tests and class prediction, we used Biometric Research Branch (BRB) ArrayTools v3.2.2 software, developed by R. Simon and A. Peng. Permutations of the measurements are then used to estimate the FDR (the percentage of genes identified by chance). Additionally, mouse fetal livers at E18.5 and postnatal livers at 8 days of birth were profiled on Affymetrix MG-U74A, B v2 arrays. Data were processed and analyzed as aforementioned.
Except when indicated, transcriptome analysis was carried out using either an assortment of R system software packages (http://www.R-project.org, v2.3.0) including those of Bioconductor v1.8 (Gentleman et al., 2004) or original R code.
B.1. Normalization
Raw data from Affymetrix HG-U133A 2.0 GeneChip™ microarrays were normalized in batch using robust multi-array average method (R package affy, v1.10.0) (Irizarry et al., 2003). Probe sets corresponding to control genes or having a “_x_” annotation were masked yielding a total of 19,787 probe sets available for further analyses.
B.2. Class Discovery
Step 1
Variance Test
The variance of each probe set across samples was tested and compared to the median variance of all the probe sets, using the model: ((n−1)×Var(probe set)/Varmed), where n refers to the number of samples. By using the same filtering tool of BRB ArrayTools software, the P-value for each probe set was obtained by comparison of this model to a percentile of Chi-square distribution with (n−1) degrees of freedom.
Robust Coefficient of Variation (rCV)
The rCV was calculated for each probe set as follows. After ordering the intensity values of n samples from min to max, we eliminated the min and max values and we calculated the coefficient of variation (CV) for the remaining values.
Unsupervised Probe Sets Selection
Unsupervised selection of probe set lists was based on the two following criteria:
(i) variance test at P<0.01,
(ii) rCV less than 10 and superior to a given rCV percentile. We used eight rCV percentile thresholds (60%; 70%; 80%; 90%; 95%; 97.5%; 99%; 99.5%), which yielded 8 probe set lists.
Step 2: Generation of a Series of 24 Dendrograms
Hierarchical clustering was performed by using the 8 rCV-ranked probe sets lists, 3 different linkage methods (average, complete and Ward's), and 1-Pearson correlation as a distance metric (package cluster v1.9.3). This analysis generated 24 dendrograms.
Step 3:
Stability Assessment
The intrinsic stability of each of the 24 dendrograms was assessed by comparing each dendrogram to the dendrograms obtained after data “perturbation” or “resampling” (100 iterations). Perturbation stands for the addition of random gaussian noise (μ=0, σ=1.5×median variance calculated from the data set) to the data matrix, and resampling for the random substitution of 5% of the samples by virtual sample's profiles, generated randomly. The comparison between dendrograms across all iterations yielded a mean ‘similarity score’ (see below). The overall stability was assessed by calculating a mean similarity score, using all pairs of the 24 dendrograms.
Similarity Score
To compare two dendrograms, we compared the two partitions in k clusters (k=2 to 8) obtained from these two dendrograms. To compare a pair of partitions, we used a similarity measure, which corresponds to the symmetric difference distance (Robinson and Foulds, 1981).
Step 4: Identification of Robust Clusters
We identified groups in which any pair of samples was co-classified in at least 22 of the 24 partitions, and considered only groups made of 4 samples or more. Then, for any pair of these groups, we calculated the mean number of co-classification of any sample in the first group with any sample in the second group. We aggregated the groups for which this score was at least 18 (over the 24 partitions).
B.3. Supervised Tests
We compared gene expression between two classes of samples by using the Student's t test with random variance model option (BRB ArrayTools software, version 3.4.0a, developed by Dr. Richard Simon and Amy Peng Lam, http://linus.nci.nih.gov/BRB-ArrayTools.html). False Discovery Rates were assessed by using 1000 random permutations of labels (Monte Carlo approach).
B.4. Classification
To classify samples according to gene expression profile, we used the Class prediction tool of BRB ArrayTools software using all 6 following algorithms: Compound Covariate Predictor (CCP), Linear Discriminant Analysis (LDA), 1-Nearest Neighbor (1NN), 3-Nearest Neighbors (3NN), Nearest Centroid (NC) and Support Vector Machines (SVM). Each sample was classified according to the majority of the 6 algorithms. Samples classified as C2 by at least 3 algorithms were classified accordingly.
B.5. Gene Ontology and Pathway Analysis
We used a hypergeometric test to measure the association between a gene (probe set) list and a gene ontology term (GO term), as in GO stats R package (R. Gentleman). To this end, we mapped the gene list and the GO terms to non-redundant Entrez Gene identifiers by using the annotation file HG-U133_Plus—2.annot.csv (http://www.affymetrix.com, Dec. 14, 2006). GO terms and their relationships (parent/child) were downloaded from http://www.geneontology.org (version Dec. 31, 2006). The list of proteins associated to GO terms (table gene_association.goa_human) and mapping the Entrez Gene ids (table human.xrefs) were downloaded from ftp://ftp.ebi.ac.uk/pub/databases/GO/goa.
KEGG pathway annotation was done by Onto-tools software (http://vortex.cs.wayne.edu/ontoexpress/servlet/UserInfo). We designated a significance threshold of each hypergeometric test at P<0.001, and the condition that a GO term or pathway be represented by at least 3 Entrez Gene identifiers.
B.6. Gene Set Enrichment Analysis (gsea)
GSEA (Subramanian et al., 2005) was used to evaluate the correlation of a specific gene list with two different sample groups (phenotypes). Briefly, this method calculates an enrichment score after ranking all genes in the dataset based on their correlation with a chosen phenotype and identifying the rank positions of all the members of a defined gene set. We used the signal2noise ratio as a statistic to compare specific and random phenotypes in order to evaluate statistical differences.
C. Array-Based Comparative Genomic Hybridization (aCGH)
Genomic DNA from 24 HBs and 3 non-tumor liver samples was analyzed using aCGH chips designed by the CIT-CGH consortium. This array contains 3400 sequence-verified PAC/BAC clones spaced at approximately 1 Mb intervals, spotted in triplicate on Ultra Gaps slides (Corning Inc, Corning, N.Y.).
The aCGH chip was designed by CIT-CGH consortium (Olivier Delattre laboratory, Curie Institute, Paris; Charles Theillet laboratory, CRLC Val d'Aurelle, Montpellier; Stanislas du Manoir laboratory, IGBMC, Strasbourg and the company IntegraGen™). DNAs were labeled by the random priming method (Bioprime DNA labelling system; Invitrogen, Cergy-Pontoise, France) with cyanine-5 (Perkin-Elmer, Wellesley, Mass.). Using the same procedure, we labeled control DNAs with cyanine-3. After ethanol-precipitation with 210 μg of Human Cot-1 DNA (Invitrogen), resuspension in hybridization buffer (50% formamide), denaturation at 95° C. for 10 minutes and prehybridization at 37° C. for 90 minutes, probes were cohybridized on aCGH. The aCGH slides were previously preblocked with a buffer containing 2.6 mg succinic anhydride/118 ml N-methyl-2-pyrrolidinone/32 ml sodium tetraborate decahydrate, pH 8.0 (Sigma-Aldrich, Lyon, France). After washing, arrays were scanned using a 4000B scan (Axon, Union City, Calif.). Image analysis was performed with Genepix 5.1 software (Axon) and ratios of Cy5/Cy3 signals were determined. The aCGH data were normalized using lowess per block method (Dudoit et al., 2002). Comparison between groups was done using chi-square test or Fisher's exact test, as appropriate.
Status assignment (Gain/Loss) was performed using R package GLAD v1.6.0. Computation of recurrent minimal genomic alterations was done using slight modification of a previously described method (Rouveirol et al., 2006). For comparison between groups, we used the Fischer exact test. Complete aCGH data will be published elsewhere.
D. Mouse Microarray Analysis
Murine Genome Affymetrix U74v2 A and B arrays were used to investigate liver expression at embryonic day 18.5 (E18.5) and at 8 days after birth (PN8). Each time point consisted of a pool of livers from 3-5 animals analyzed in triplicate. Microarray experiments were performed according to the manufacturer's instructions.
Publicly available Affymetrix Mouse Genome (MG) 430 2.0 array liver expression data at embryonic time points E11.5, E12.5, E13.5, E14.5, and E16.5 days of gestation (Otu et al., 2007), were downloaded from the Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE6998).
MG-U74v2, MG-430 2.0 and HG-133A 2.0 array intra- and cross-species probeset comparison was achieved by using the Affymetrix NetAffx analysis center and by choosing “Good Match” degree of specificity. Unification of sample replicates, multiple array data standardization and Heatmap visualization was done by using dCHIP v1.6 software. Comparison of fetal liver stages by supervised analysis was performed using BRB ArrayTools software as previously described, by classing E11.5 and E12.5 as “Early” and E14.5 and E16.5 as “Late” fetal liver stage. Supervised signature was applied to HB array data, and intensity cut-off=60 was chosen in order to remove probesets that did not reach such intensity level in at least one sample.
E. Quantitative PCR Analysis (qPCR)
For qPCR analysis, we used RNA from 52 tumor samples (including 11 samples analyzed on microarrays, see
RNA was extracted by using either Trizol, RNeasy kit (QIAGEN) or miRvana kit (Ambion), then quantified and quality-checked by Agilent technology. For each cDNA preparation, 1 μg of RNA was diluted at the final concentration of 100 ng/μl, and reverse transcribed with the Superscript RT kit (Invitrogen, Carlsbad, Calif.) following the manufacturer's protocol. Random primers (Promega, Charbonnières-les-Bains, France) were added at the final concentration of 30 ng/μl and the final volume was 20 μl.
The cDNA was diluted 1:25, and 5 μl were used for each qPCR reaction. We added 5 μl of 2×Sybr Green Master mix (Applied Biosystems) and 0.3 μl of each specific primer (final concentration 300 nM). Each reaction was performed in triplicate. qPCR reactions were run on the Applied Biosystems 7900HT Fast Real-Time PCR System with a 384-well thermo-block, in the following conditions: 2 min at 50° C. to activate Uracil-N-glycosylase (UNG)-mediated erase of aspecific reaction; 10 min at 95° C. to activate the polymerase and inactivate the UNG; 40 cycles (15 sec at 95° C. denaturation step and 1 min at 60° C. annealing and extension); and final dissociation step to verify amplicon specificity.
The lists of primers used for qPCR are provided in Table 6 and Table 7 above.
F. Immunohistochemistry (IHC)
IHC was carried out as reported previously (Wei et al., 2000). For antigen retrieval at 95° C., we used 1 mM EDTA (pH 8) for β-catenin and Ki-67 IHC, and 10 mM citrate buffer (pH 6) for AFP and GLUL IHC. We used monoclonal antibodies against β-catenin and GLUL (Cat. Nos. 610154 and 610517; BD Biosciences, Le Pont de Claix, France) and Ki-67 (M7240, Dako, Trappes, France) and polyclonal antibody against AFP (N1501, Dako). Reactions were visualized using the ChemMate Dako Envision Detection kit (Dako) and diaminobenzidine. Subcellular distribution and quantitative evaluation of immunostaining in the different histotypes were assessed by examining at least ten random high-power fields.
G. Clinical Data Analysis
We used the Chi-square test for comparisons between groups. Survival curves were calculated according to the Kaplan-Meier method, using the log-rank test to assess differences between curves. Variables independently related to survival were determined by stepwise forward Cox regression analysis. Follow-up was closed at February 2007 or at time of death. Statistical analysis was done with SPSS software v10.0 (SPSS Inc., Chicago, Ill.).
H. Examples of Other Pairs of Primers and Probes for the 16 Genes of Table 1 and the 4 Invariant Genes (Table 3) that can be Used in the Taqman® Method.
Results
Identification of Two HB Subclasses by Gene Expression Profiling
For robust unsupervised classification, we generated and screened a series of 24 dendrograms to identify samples that co-clustered whatever the method and the gene list. We obtained two robust subgroups of tumors named robust Cluster 1 (rC1, n=8) and robust Cluster 2 (rC2, n=5) (
The remaining tumors were classified into C1 (rC1-related) and C2 (rC2-related) subclasses by applying a predictive approach based on the rC1/rC2 gene signature and using robust samples as training set (
We observed differential expression of a number of Wnt members and targets between subclasses. C2 tumors showed increased expression of MYCN, BIRC5 that encodes the anti-apoptotic factor Survivin, NPM1 (encoding nucleophosmin) and HDAC2. By contrast, most C1 tumors prominently expressed the Wnt antagonist DKK3, BMP4, and genes previously found to be activated in liver tumors carrying mutant β-catenin (Boyault et al., 2007; Renard et al., 2007; Stahl et al., 2005). Remarkably, most genes related to liver functions are expressed in the perivenous area of adult livers, such as GLUL, RHBG, and two members of the cytochrome p450 family: CYP2E1 and CYP1A1 (Benhamouche et al., 2006; Braeuning et al., 2006) (
Further evidence that the rC1 subclass was enriched in genes assigned to the hepatic perivenous program was provided by Gene Set Enrichment Analysis (GSEA), a computational method for assessing enrichment of a predefined gene list in one class as compared with another (Subramanian et al., 2005). Thus, Wnt/β-catenin signaling appears to activate different transcriptional programs in HB subtypes, likely reflecting different cellular contexts.
HB Subclasses Evoke Distinct Phases of Liver Development
Next, we sought to determine whether HB subclasses were associated with specific histological phenotypes. Mixed epithelial-mesenchymal tumors that represented 20% of cases were not significantly associated with C1 and C2 subclasses. By contrast, a tight association was found with the main epithelial component, which defines the cell type occupying more than 50% of tumor cross-sectional areas. Sixteen out of 18 C1 tumors displayed a predominant fetal phenotype, including 4 ‘pure fetal’ cases, whereas all C2 tumors showed a more immature pattern, with prevailing embryonal or crowded-fetal histotypes associated with high proliferation (Finegold, 1994) (p<0.0001) (
To better define the relationships between HB subclasses and phases of hepatic differentiation, we first generated a liver development-related gene signature by making use of publicly available mouse fetal and adult liver data sets (Otu et al., 2007). When applied to HB samples, this signature was able to distinguish by hierarchical clustering two HB groups closely matching the C1/C2 classification. Next, we integrated HB gene expression data with the orthologous genes expressed in mouse livers at embryonic days (E) 11.5 to 18.5, and at 8 days of birth. In unsupervised clustering, most C2 tumors co-clustered with mouse livers at early stages of embryonic development (E11.5 and E12.5), whereas C1 tumors gathered with mouse livers at late fetal and postnatal stages. Together, these data comfort the notion that tumor cells in C2 and C1 subtypes are arrested at different points of the hepatic differentiation program.
Identification of a 16-Gene Signature as HB Classifier
To investigate the relevance of molecular HB classification in an independent set of tumors, we defined a HB classifier signature derived from the top list of genes differentially expressed between rC1 and rC2 clusters. After qPCR assessment, a list of 16 top genes at p≦10−7 was selected to form a class predictor (Table 1). Most of these genes show drastic variations in expression level during liver development, and among them, BUB1 and DLG7 have been repeatedly identified as hESC markers (Assou et al., 2007). The 16-gene expression profile was first investigated in rC1 and rC2 samples used as training set, and it predicted classification with 100% of accuracy in these samples, using either microarray or qPCR data. The robustness of this signature was confirmed by correct classification into C1 and C2 subclasses of all 13 remaining tumors analyzed by microarray (
Extending our previous observation, C1/C2 classification in this new set of tumors was unrelated to CTNNB1 mutation rate. Using qPCR, we also confirmed enhanced expression in C2 tumors of liver progenitor markers such as AFP, Ep-CAM, and KRT19, as well as MYCN (
NL, non-tumor liver; C1, good prognosis hepatoblastomas; C2, bad prognosis hepatoblastomas. Shown are the median values of 46 hepatoblastomas from 41 patients, the minimal and maximal values in each class, and the fold changes between classes. Data are presented in arbitrary units after normalization of the raw quantitative PCR values with genes (ACTG1,
The 16-Gene Signature as a Strong Independent Prognostic Factor
In a First Set of 61 Patients
The clinical impact of HB molecular classification was addressed in a first set of 61 patients (
We further assessed the prognostic validity of the 16-gene signature for all patients in multivariate analysis, using a Cox proportional hazards model with pathological and clinical variables associated to patients' survival. This analysis identified the signature as an independent prognostic factor, with better performance than tumor stage defined by PRETEXT stage, vascular invasion and extrahepatic metastases (
In a Second Set of 86 Patients
The clinical impact of HB molecular classification was addressed in a second set of patients (comprising the sample of the first set), comprising 53 (61%) C1 and 33 (39%) C2 cases. Besides strong association with predominant immature histotypes, HBs of the C2 subclass were tightly associated with features of advanced tumor stage, such as vascular invasion and extrahepatic metastasis (
Survival analysis (Kaplan Mejer, log rank test); DES: disease-free survival; Others: dead or alive with recurrent disease.
Next, we examined whether pre-operative chemotherapy treatment given to 73 patients could affect tumor classification. These cases were evenly distributed among HB subclasses, with no significant association with molecular classification. We examined the performance of the 16-gene signature on the 73 tumors resected after chemotherapy, and found significant difference in outcome between patients with C1 and C2 type HBs (p=0.0002, log rank test) (
We further assessed the prognostic validity of the 16-gene signature for all patients in multivariate analysis, using a Cox proportional hazards model with pathological and clinical variables associated to patients' survival. This analysis identified the signature as an independent prognostic factor, with better performance than tumor stage defined by PRETEXT stage, vascular invasion and extrahepatic metastases (
Finally, various clinical elements of 103 HB samples from 86 patients were compared with respect to their classification as C1 or C2 grade using the 16-gene signature (Table 10).
STD: standard risks (cisplatine) - High:high risk (cisplatine/doxorubicine, intensified treatment); Tumor stage (defined as Vasc. Inv and/or metastasis and/or PRETEXT stage IV); metastasis: extrahepatic metastasis (mainly lung); vascular invasion is determined by imagery; Pretext IV (involved an intrahepatic extent of the tumor to all hepatic sections);
The above results carried out on a first set of 61 patients, and on a second completed set of 86 patients, demonstrate that the 16-gene signature, identified in the present application, is a strong prognostic relevance when compared to current clinical criteria.
Discussion
The present application demonstrates that, using integrated molecular and genetic studies, hepatoblastoma encompass two major molecular subclasses of tumors that evoke early and late phases of prenatal liver development. Aberrant activation of the canonical Wnt pathway represented a seminal event in both tumor types, with cumulated mutation rates of β-catenin, APC and AXIN over 80%. However, depending on tumor differentiation stage, Wnt signaling activated distinct transcriptional programs involved in tumor growth and invasiveness or in liver metabolism. Further comparisons of immature, embryonal-type HBs with the bulk of more differentiated, fetal-type tumors revealed a tight correlation between stage of hepatic maturation arrest and clinical behavior, notably vascular invasion and metastatic spread, and patients' survival.
Molecular Hb Subclasses are Determined by Liver Differentiation Stages
In this study, expression-based classification of HB was achieved through a highly reliable statistical method combining different unsupervised hierarchical clustering approaches. This method led to the selection of two robust tumor subgroups, and this robustness was confirmed using a new, independent set of samples and 16 relevant genes discriminating these tumor subgroups. These results demonstrated that the most significant differences between HB subclasses can be ascribed to distinct hepatic differentiation stages, as defined by comparison with expression profiles of mouse livers at early (E11.5-E12.5) and late (E14.5-E18.5) embryonic stages. These studies also provide biological relevance to early histologic classification that distinguished fetal and embryonal cells as major HB components (Weinberg and Finegold, 1983). The C1 subclass recapitulates liver features at the latest stage of intrauterine life, both by expression profile and by mostly fetal morphologic patterns, while in the C2 subclass, transcriptional program and predominant embryonal histotype resemble earlier stages of liver development. Thus, despite frequent morphological heterogeneity in HB, these expression-based subclasses closely matched the histologic types found to be prevailing after microscopic examination of the entire tumor mass.
These results, showing that childhood liver tumors recapitulate programs of their developing counterpart, are in line with recent studies using cross-species comparisons. It has been demonstrated that clinically distinct medulloblastoma subtypes can be identified by their similarity with precise stages of murine cerebellar development (Kho et al., 2004). Evidence for conserved mechanisms between development and tumorigenesis was also obtained in Wilms' tumor, the embryonic kidney malignancy, which shares expression of sternness and imprinted genes with murine metanephric blastema (Dekel et al., 2006). It was noticed that HBs, like Wilms' tumors, exhibit robust overexpression of a number of paternally expressed genes like DLK1, IGF2, PEG3, and PEG10 that are involved in growth induction processes and downregulated with differentiation during development.
Previous studies using stem cell markers and markers of hepatocytic and biliary lineages have described differential patterns among HB components that reflect sequential stages of liver development (Schnater et al., 2003). The present data extent these observations, and indicate that immature C2-type tumor cells evoke hepatic cancer progenitor cells, with distinctive overexpression of highly relevant markers such as cytokeratin 19 and Ep-CAM (Roskams, 2006). Recently, embryonic stem/progenitor cells have been isolated from human fetal livers, either by enrichment of blast-like cells in primary hepatoblast cultures or by immunoselection of Ep-CAM-positive epithelial cells (Dan et al., 2006; Schmeizer et al., 2007). These cell lines have self-renewal capacity and can differentiate into mature hepatocytes and cholangiocytes, and one of them also gives rise to various mesenchymal lineages (Dan et al., 2006). Whether HBs arise from transformation of these cell types is presently unknown. As malignant mesenchymal derivatives are frequently admixed with epithelial tissues in HB, it is tempting to speculate that this tumor occurs from a multipotent progenitor harboring characteristics of mesenchymal-epithelial transitional cells. Moreover, since no significant differences in gene expression profiles was noted here between pure epithelial and mixed epithelial-mesenchymal HBs, tumor cells likely kept intrinsic capacities to undergo epithelial-mesenchymal transition.
A salient feature of immature HBs is the characteristic interplay of sternness and proliferation found in aggressive tumors (Glinsky et al., 2005). The C2-type expression profile was significantly enriched in hESC markers, including the mitotic cell cycle and spindle assembly checkpoint regulators cyclin B1, BUB1, BUB1B, and Aurora kinases. These mitotic kinases are centrosomal proteins that ensure proper spindle assembly and faithful chromosome segregation in mitosis. Overexpression of these kinases or other components of the spindle checkpoint induces centrosome amplification and defects in chromosome segregation leading to chromosome number instability and aneuploidy (Marumoto et al., 2005; Zhou et al., 1998). Non-disjunctional events are involved in developmental syndromes (Hassold and Hunt, 2001), and might be responsible for increased rate of chromosomal imbalances evidenced here in C2-type HBs.
Context-Dependent Transcriptional Programs Driven by Wnt Signalling
Mutational activation of β-catenin is a hallmark of HB, and accordingly, we found intracellular accumulation and nuclear localization of the protein in virtually all tumors, albeit with variable frequencies and intensities. Both immature and differentiated tumors overexpressed AXIN2 and DKK1, reflecting an attempt to activate a negative feedback loop aimed at limiting the Wnt signal. However, the two HB subtypes showed significant differences in β-catenin immunoexpression, illustrated by concomitant nuclear accumulation and decreased membranous localization of the protein in poorly differentiated, highly proliferative HBs. Heterogeneous distribution of nuclear β-catenin within colorectal tumors has been linked to different levels of Wnt signaling activity, resulting from differential combinations of autocrine and paracrine factors (Fodde and Brabletz, 2007). Similarly, nuclear β-catenin might be related to the absence of membranous E-cadherin in immature HBs, as we reported previously (Wei et al., 2000), and to cross-talks with growth-stimulating pathways in less differentiated cells. In this context, increased dosage of Wnt signaling might induce migratory and invasive phenotype.
Major differences between the two HB subtypes were found here in expression levels of Wnt targets involved in liver functions. Recent studies have demonstrated that Wnt/β-catenin signaling governs liver metabolic zonation by controlling positively the perivenous gene expression program and negatively the periportal program (Benhamouche et al., 2006). In our study, overexpression of hepatic perivenous markers such as GLUL was prominent in differentiated HBs, while genes encoding periportal functions like GLS2 were downregulated. This profile is highly similar to those of human and murine HCCs expressing mutant β-catenin (Boyault et al., 2007; Stahl et al., 2005), and corresponds to an hepatic signature of Wnt target genes. Accordingly, the zonation-related profile was lessened in poorly differentiated HBs, and mutant β-catenin was found to activate a different, muscle-related expression program in the pediatric Wilms' tumor (Zirn et al., 2006).
Clinical Implications
The clinical behavior of many human solid tumors has been related to their differentiation status and proliferative rate. We show that HB does not depart from this rule, with strong correlation of molecular subclasses linked to hepatic differentiation with clinical tumor stage and patient's outcome. This correlation was mainly determined by differences in invasive and metastatic phenotypes between the two subclasses, but not by differences in tumor localization or tumor extension across liver sections, which defines the preoperative staging (PRETEXT) utilized to evaluate tumor resectability (Perilongo et al., 2000). Major differences in expression profiles of the two molecular HB subtypes led us to elucidate a 16-gene signature that proved highly efficient in stratification of HBs as well as normal livers according to hepatic developmental stage. Most importantly, this classifier also discriminated aggressive tumors, exhibited powerful survival predictor capacities in pre-treatment biopsies and surgical specimens, and demonstrated strong prognostic relevance when confronted to current clinical criteria in multivariate analysis. Although immature HBs have been associated to worse clinical outcome as opposed to differentiated HBs (Weinberg and Finegold, 1983), frequent cellular heterogeneity has hampered the use of histopathologic criteria for defining risk groups, excepted for a minority of cases showing ‘pure fetal’ or SCUD types. The expression signature afforded here enables direct appraisal of the global degree of tumor cell maturation, allowing to bypass these difficulties. Thus, it can improve the outcome prediction and clinical management of hepatoblastoma, by identifying cases with increased risk of developing metastasis, or conversely, by avoiding unnecessary over-treatment.
In conclusion, the present application identifies a 16-gene signature that distinguishes two HB subclasses and that is able to discriminate invasive and metastatic hepatoblastomas, and predicts prognosis with high accuracy. The identification of this expression signature with dual capacities may be used in recognizing liver developmental stage and in predicting disease outcome. This signature can be applied to improve clinical management of pediatric liver cancer and develop novel therapeutic strategies, and is therefore relevant for therapeutic targeting of tumor progenitor populations in liver cancer.
Analysis of 64 Hepatocellular Carcinoma (HCC) from 64 Patients
Real time RT-PCR (Taqman methodology) was performed on 67 HCC samples, as disclosed for HB samples above. The clinical characteristics of the 67 patients diagnosed with HCC as well as the features of the HCC samples are disclosed in Tables 11 and 12 below.
Amplification was carried out with primers of the 16-gene signature disclosed in Table 6. Data were normalized to the expression of the ROTH2 gene (primers disclosed in Table 7) and analyzed by the ΔCt method. Quantitative PCR data are disclosed in Table 13.
N.A: non available;
macro: macrovacular invasion;
micro: microvacular invasion
N.A: non available; HBV: hepatitis B virus; HCV; hepatitis C virus; hemochro: hemochromatosis; NASH non alcoholic steatohepatitis.
NA: non available
Data were then analyzed by unsupervised clustering (dCHIP software) using 2 methods: average and centroid. Tumors were clustered into 2 groups, C1 and C2. Most of the samples have been attributed the same classification using the 2 methods, except for 6 samples (9%) that have been attributed a different classification (Table 15).
Clinical Parameters Associated to the C1 and C2 Molecular Subclasses
The clinico-pathological parameters of patients and tumors were compared between the two groups C1 and C2, using student's t test and Kaplan-Meier estimates. Since some data are not available for 3 patients, the following statistical studies were performed on 64 tumors.
Survival Analysis
There is a strong correlation of the molecular classification into C1 and C2 with patient's survival by using both classifications (Log rank: Centroid p=0.020 and Average p=0.024) (
Association of HCC Classification with Clinical Variables
Table 14 shows the correlation between some clinical variable and the classification of the tumors.
(ns: non-significant)
In a second analysis, the global set of 64 tumors was analyzed independently of the C1/C2 classification, for parameters associated to survival. Results are presented in Table 16.
(ns: non-significant)
These results demonstrate that the methods and the signatures of the invention are able to determine the grade not only of HB tumors but also of HCC tumors. The inventors have shown that hierarchical clustering is an efficient method for classification of tumor grade especially for HB. For HCC, this method may be less sufficient (less robust) when the amplitude of variation of expression results of the genes is less important than for HB.
Classification of Hepatoblastomas and Hepatocellular Carcinomas Using the Method of Discretization of Continuous Values.
85 hepatoblastomas (HBs) and 114 hepatocellular carcinomas (HCCs) including to the samples used in the above examples have been analyzed by quantitative PCR using the 16-gene signature and have been classified by the method of discretization of continuous values in order to determine their tumor grade.
Description of the Methodology for Classification
The inventors have designed a methodology for classification based on the principle of discretization of continuous values which refers to the process of converting continuous variables to “discretized” or nominal sets of values.
The major advantage of the discretization method relies on the definition of a cut-off for codification of each qPCR value (either by the Taqman or by the SybrGreen method), which provides an intrinsic score to directly classify an individual sample. There is hence no requirement to compare a sample to a large series of samples. In contrast, in other classification methods, the assigned subclass (such as C1 or C2 disclosed herein) is relative to the values obtained in a large number of cases. Moreover, the use of the average discretized values allows to tolerate missing values when analyzing the qPCR results (i.e. missed amplification of one of the genes for technical reasons).
Using the qPCR data of the 16 genes normalized to the reference RHOT2 gene (−deltaCt values), a cut-off (or threshold) has been defined for each gene. The −deltaCt values are converted into discrete values “1” or “2” depending on an assigned cut-off. In order to privilege the identification of samples that display strong overexpression of proliferation-related genes and/or strong downregulation of differentiation-related genes, the cut-offs have been defined as follows:
for the 8 proliferation-related genes (AFP, BUB1, DLG7, DUSP9, E2F5, IGSF1, NLE1, RPL10A), all −DeltaCts with a value above the 67th percentile have been assigned discretized value “2”, otherwise the assigned value was “1”.
for the 8 differentiation-related genes (ALDH2, APCS, APOC4, AQP9, C1S, CYP2E1, GHR, HPD), all −deltaCts with a value below the 33rd percentile have been assigned discretized value “1”, otherwise the assigned value was “2”.
Classification of 85 Hepatoblastomas (HB)
RNA Preparation and Quantitative PCR
RNA was extracted by using either Trizol, RNeasy kit (QIAGEN) or miRvana kit (Ambion), then quantified and quality-checked by Agilent technology.
For quantitative PCR analysis, the Sybr Green approach was used as described in point E. above. For each cDNA preparation, 1 μg of RNA was diluted at the final concentration of 100 ng/μl, and reverse transcribed with the Superscript RT kit (Invitrogen) following the manufacturer's protocol. Random primers were added at the final concentration of 30 ng/μl and the final volume was 20 μl. The cDNA was diluted 1:25, and 5 μl were used for each qPCR reaction. We added 5 μl of 2×Sybr Green Master mix (Applied Biosystems) and 0.3 μl of each specific primer (disclosed in point H. above) (final concentration 300 nM). Each reaction was performed in triplicate. qPCR reactions were run on the Applied Biosystems 7900HT Fast Real-Time PCR System with a 384-well thermo-block, and the conditions were the following:
2 min at 50° C. to activate Uracil-N-glycosylase (UNG)-mediated erase of a specific reaction
10 min at 95° C. to activate the polymerase and inactivate the UNG
40 cycles:
15 sec at 95° C. denaturation step
1 min at 60° C. annealing and extension
Final dissociation step to verify amplicon specificity.
The normalized qPCR (deltaCt) values of the 85 HB samples are given in Table A.
Analysis of qPCR Data.
Assignment of a discretized value for the 8 proliferation-related genes (“AFP” “BUB1” “DLG7” “DUSP9” “E2F5” “IGSF1” “NLE” “RPL10A”) was based on the 67th quantile (i.e. percentile), given that around ⅓ of HB cases overexpress proliferation genes, which is correlated with tumor aggressiveness and poor outcome. Assignment of a discretized value for the 8 differentiation-related genes (“ALDH2” “APCS” “APOC4” “AQP9” “C1S” “CYP2E1” “GHR” “HPD”) was based on the 33rd quantile, given that around ⅓ of HB cases underexpress differentiation genes, which is correlated with tumor aggressiveness and poor outcome.
The cut-offs (or thresholds) selected for the −deltaCT value of each gene were determined after considering said chosen percentiles for each group of genes are as follows:
AFP: 3.96139596; ALDH2: 4.3590482; APCS: 4.4691582; APOC4: 2.03068712; AQP9: 3.38391456; BUB1: −1.41294708; C1S: 4.24839464; CYP2E1: 6.70659644; DLG7: −3.3912188; DUSP9: 2.07022648; E2F5: −0.72728656; GHR: −0.1505569200; HPD: 2.27655628; IGSF1: 0.1075015200; NLE: −0.02343571999; RPL10A: 6.19723876
For the sample, the relative expression value is determined for each gene of the set of profiled genes. Each value is compared to the cut-off for the corresponding gene and is then discretized as a result of its position with respect to said cut-off.
The next step consisted in assigning a discretized score to each sample as follows:
1—the average of the “discretized” values of the 8 proliferation-related genes was determined. The 8 proliferation-related genes are the following: AFP, BUB1, DLG7, DUSP9, E2F5, IGSF1, NLE, and RPL10A.
2—the average of the “discretized” values of the 8 differentiation-related genes was determined. The 8 differentiation-related genes are the following: ALDH2, APCS, APOC4, AQP9, C1S, CYP2E1, GHR, and HPD.
3—The score for each sample was determined as the ratio between the average of proliferation-related genes and the average of differentiation-related genes.
According to this calculation, a score of 2 is the maximal score for highly proliferating and poorly differentiated tumors, whereas well differentiated and slowly proliferating tumors will have a minimal score of 0.5.
Based on the scores assigned to the 85 HB samples analyzed, cut-offs were identified to separate the samples into relevant subclasses. Two different cut-offs that correspond to the 33rd (0.615), and 67th percentile (0.91) have been assessed, leading to the definition of either 2 or 3 subclasses. These data together with the clinical data of 85 HB cases are given in the Table B.
Statistical Analysis of Clinical Correlations
All statistical correlations were analyzed using the discrete classification into 2 subclasses with the 67th percentile (see 3rd column of the table given in Table B).
*Other = embryonal, macrotrabecular, crowded fetal
The best correlation of the discrete classification was observed with the previous classification into C1 and C2 classes, followed by the main epithelial histological component. The correlation with patients' survival is also excellent, as shown by using the Kaplan-Meier estimates and the log-rank test. Illustrative Kaplan-Meier curves are given in
In conclusion, this study shows that the discretization method allows to classify hepatoblastoma as efficiently as the previously described method.
A similar approach was therefore applied to the analysis of hepatocellular carcinoma.
Analysis of 114 Hepatocellular Carcinomas (HC)
RNA Preparation
RNA was extracted by using either Trizol, RNeasy kit (QIAGEN) or miRvana kit (Ambion), then quantified and quality-checked by Agilent technology.
For each cDNA preparation, 1 μg of RNA was diluted at the final concentration of 100 ng/μl, and reverse transcribed with the Superscript RT kit (Invitrogen) following the manufacturer's protocol. Random primers were added at the final concentration of 30 ng/μl and the final volume was 20 μl. The cDNA was diluted 1:25, and 5 μl were used for each qPCR reaction. We added 5 μl of 2×Sybr Green Master mix or the Taqman Master mix (Applied Biosystems) and specific primers (and probes when using Taqman chemistry) at the concentration indicated by the manufacturer. Each reaction was performed in triplicate. qPCR reactions were run on the Applied Biosystems 7900HT Fast Real-Time PCR System with a 384-well thermo-block, and the conditions were the following:
2 min at 50° C. to activate Uracil-N-glycosylase (UNG)-mediated erase of aspecific reaction (omit if using the Taqman approach)
10 min at 95° C. to activate the polymerase and inactivate the UNG
40 cycles:
15 sec at 95° C. denaturation step
1 min at 60° C. annealing and extension
Final dissociation step to verify amplicon specificity (omit if using the Taqman approach)
Quantitative PCR
Real time RT-PCR was performed for 16 genes on 114 HCC samples using two different technologies:
Sybr Green as described above for hepatoblastoma (26 samples).
Taqman methodology (88 samples) using primers and probes designed and publicly released by Applied Biosystems company.
Each reaction was performed in triplicate for Sybr Green protocol and in duplicate for the taqman protocol. qPCR reactions were run on the Applied Biosystems 7900HT Fast Real-Time PCR System with a 384-well thermo-block.
Raw data for each gene were normalized to the expression of the ROTH2 gene, providing the deltaCt values that were then used for tumor classification into subclasses using the discretization method.
The normalized qPCR values (deltaCt) of the 16 genes in 26 HCC samples analyzed by the Sybr Green approach is given in Table C. The deltaCt values for 88 HCCs analyzed by the Taqman approach are given in Table D.
Analysis of qPCR Data.
The −deltaCt values for each gene in each sample was used. The cut-offs (or thresholds) selected for each gene using the Taqman method or the SybrGreen method are as follows:
For the sample, the relative expression value is determined for each gene of the set of profiled genes. Each value is compared to the cut-off for the corresponding gene and is then discretized as a result of its position with respect to said cut-off.
The next step consisted in assigning a score to each sample as follows:
1—the average of the “discretized” values of the 8 proliferation-related genes was determined. The 8 proliferation-related genes are the following: AFP, BUB1, DLG7, DUSP9, E2F5, IGSF1, NLE, and RPL10A.
2—the average of the “discretized” values of the 8 differentiation-related genes was determined. The 8 differentiation-related genes are the following: ALDH2, APCS, APOC4, AQP9, C1S, CYP2E1, GHR, and HPD.
3—The score for each sample was determined as the ratio between the to average of proliferation-related genes and the average of differentiation-related genes.
According to this calculation, a score of 2 is the theoretical maximal score for highly proliferating and poorly differentiated tumors, whereas well differentiated and slowly proliferating tumors will have a theoretical minimal score of 0.5.
Based on the scores assigned to the 114 samples analyzed, cut-offs are identified to separate the samples into relevant subclasses. Three different cut-offs that correspond to the 30rd (0.66), 50th (0.8125) and 67th percentile (0.925) have been assessed, leading to 4 different classification methods.
Samples were separated into the corresponding subgroups, and subsequent analysis was carried out using the 4 classification methods. Survival for each group was determined using the Kaplan-Meier estimates and the log-rank test.
Statistical Analysis of Clinical Correlations with the Subclasses for 114 HCCs
A complete table with all clinical and pathological data collected for 114 HCC patients is given in Table G. The different parameters are represented as follows:
Abbreviations:
HCV, hepatitis C virus;
HBV, hepatitis B virus;
NASH, Nonalcoholic steatohepatitis;
SR, surgical resection;
OLT, orthotopic liver transplantation;
W, well differentiated;
M, moderately differentiated;
P, poorly differentiated;
NA, not available;
DOD, dead of cancer.
*12 cases have more than one etiological agent and data were not available for 2 Gases.
†Data were not available for all cases. Percentages were deduced from available data.
In a second step, the intrinsic parameters of the tumors correlated with patients' survival were analyzed. In this series of tumors, only tumor grade (Edmonson) and vascular invasion were significantly correlated with survival.
Classification by Discretization of Continuous Values
The clinico-pathological parameters were compared between the tumor groups using student's t test and chi-square test. Survival was analyzed by using Kaplan-Meier curves and log rank test. A special attention was given to the classification with the 67th percentile. Follow-up was closed at 146 months for overall survival (OS) and at 48 months for disease-free survival (DFS).
In conclusion, these data show significant correlations between molecular classification using the 3 methods and the following parameters: Tumor grade (Edmonson), tumor differentiation (OMS), proliferation rate, vascular invasion and survival. In contrast, the classifications were not correlated with etiological factors (viral hepatitis, alcohol, etc. . . . ), with the state of the disease in adjacent, non tumoral livers or with tumor recurrence.
The data suggest that classification using the 67th percentile seems to be the most adequate and is strongly recommended to classify HCCs.
Multivariate Analysis
To further determine the efficiency of the molecular classification using the 67th percentile, we performed multivariate analysis with the Cox regression test on two sets of patients for which all data were available:
91 patients that received either surgical resection or orthoptic liver transplantation (OLT)
71 patients that received surgical resection.
Different variables associated to survival in the clinical settings have been included in the multivariate analysis: 1) Edmonson grade, 2) microvascular invasion and 3) Molecular classification using the 67th percentile.
Correlation of the Molecular Classifications with Survival
For overall survival (OS) and disease-free survival (DFS), we compared the efficiency of the 3 methods of discretization that separate the samples into 2 subclasses. Independent studies were made for patients that received surgical resection and for patients that received orthoptic liver transplantation (OLT). The ability of the 16-gene signature to discriminate between recurrent and non-recurrent tumors was also assessed.
Abbreviations: OS, overall survival; DFS, disease free survival
*There is a trend but it is not significant and it is lost in the P60 analysis
The different analyses are illustrated in the Kaplan-Meier plots shown in
In conclusion, the method described herein is able to classify HCC cases according to tumor grade and patient's survival, and represents a powerful tool at diagnosis to stratify the tumors according to the prognosis, and for further clinical management of HCC. In particular, it may be an excellent tool for the decision of orthotopic liver transplantation, since the criteria used currently are limited and often poorly informative of the outcome.
Protocol for Applying the Method to a New Sample
The following protocol is designed according to the invention:
1—extract total RNA from the tumor specimen using well established technologies.
2—synthesize cDNA synthesis (suggested conditions: 1 μg RNA and 300 ng of random hexamers for a 20 μl-reaction)
3—amplify the selected genes said genes being in equal number of each of the groups defined as overexpressed proliferation-related genes group and downregulated differentiation-related genes group (profiled genes within the group of 2 to 16 genes) and the reference gene (invariant gene) such as for example the RHOT2 gene 1:5 cDNA dilution, using either Taqman or SybrGreen qPCR technology.
4—determine the Delta Ct (DCt) value for each gene
5—compare the value with the threshold of reference (for HB or for HC) in order to assign a discretized value of “1” or “2”.
5—determine the average of discretized values in each group, i.e., for the selected proliferation-related genes (up to 8) separately for and the selected differentiation-related genes (up to 8) and determine the ratio of these 2 average values which is the score of the sample.
6—compare the result with the reference scores corresponding to the following cut-offs:
C1
|30rd=0.6667
|50th=0.8125
|67th=0.925
C2
For patient X having an HC tumor a Taqman qPCR is performed.
Step one: assignment of discretized values to each selected gene among proliferation-related genes and differentiation-related genes.
The DCt of AFP is −4.0523
The cut-off for AFP for qPCR using Taqman technology is −1.2634010 Given that −4.0523 is lower than the cut-off, the assigned discretized value is 2.
Step two: Determination of the average of discretized values for the 2 sets of 8 genes:
AFP=2; BUB1=1; DLG7=2; DUSP9=2; E2F5=2; IGSF1=1; NLE=2; RPL10A=1;
Average of Proliferation-Related Genes: (2+1+2+2+2+1+2+1)/8=1.625
ALDH2=1; APCS=1; APOC4=1; AQP9=1; C1S=2; CYP2E1=2; GHR=1; HPD=2;
Average of Differentiation-Related Genes: (1+1+1+1+2+2+1+2)/8=1.375
Step Three: calculate the ratio proliferation/differentiation score.
In this example: 1.625/1.375=1.18182
Step 4: compare the result with the reference scores:
C1
|30rd percentile=0.6667
|50th percentile=0.8125
|67th percentile=0.925
C2
Classification based on the value of the ratio=1.18182.
As the value is above the 67th percentile, the assigned class is C2.
| Number | Date | Country | Kind |
|---|---|---|---|
| 08290628.0 | Jun 2008 | EP | regional |
| 09151808.4 | Jan 2009 | EP | regional |
| Filing Document | Filing Date | Country | Kind | 371c Date |
|---|---|---|---|---|
| PCT/IB2009/006450 | 6/26/2009 | WO | 4/7/2011 |
| Number | Date | Country | |
|---|---|---|---|
| 20110183862 A1 | Jul 2011 | US |
