Patent Application
20090297506
The present invention relates to a method for classification of cancer in an individual, wherein the microsatellite status and a prognostic marker are determined by examining gene expression patterns. The invention also relates to various methods of treatment of cancer. Additionally, the present invention concerns a pharmaceutical composition for treatment of cancer and uses of the present invention. The invention also relates to an assay for classification of cancer.
Studies of differential gene expression in diseased and normal tissues have been greatly facilitated by the building of large databases of the human genome sequences. Gene expression alterations are important factors in the progression from normal tissue to diseased tissue. In order to obtain a profile of transcriptional status in a certain cell type or tissue, array-based screening of thousands of genes simultaneously is an invaluable tool. Array-based screening even allows for the identification of key genes that alone, or in combination with other genes, regulate the behaviour of a cell or tissue. Candidate genes for future therapeutic intervention may thus also be identified.
Colorectal cancer generally occurs in 1 out of every 20 individuals at some point during their lifetime. In the United States alone about 150,000 new cases are diagnosed each year which amount to 15% of the total number of new cancer diagnoses. Unfortunately, colorectal cancer causes about 56,000 deaths a year in the United States.
The malignant transformation from normal tissue to cancer is believed to be a multistep process. Two molecular pathways are known to be involved in the development of colorectal cancer (Lengauer C, Kinzler K W, Vogelstein B., 1998) namely the microsatellite stable (MSS) pathway and the microsatellite instable (MSI) pathway. MSS is associated with high frequency of allelic losses, abnormalities of cytogenetic nature and abnormal tumor content of DNA. MSI however is associated with defects in the DNA mismatch repair system which leads to increased rate of point mutations and minor chromosomal insertions or deletions.
MSI tumors can be of hereditary or sporadic nature. Ninety percent of MSI tumours are of sporadic origin. Sporadic tumours are presumably MSI due to epigenetic hypermethylation of the MLH1 gene promoter. The hereditary tumours account for 10% of the MSI tumors. Mutations of for example the MLH1 or MSH 2 genes are often the cause of hereditary tumor development.
The ability of being able to determine the sporadic or hereditary nature of a MSI tumor is highly valuable. In case a tumor is characterized as being MSI, and certain clinical criteria are fulfilled such as age below 50 or three first degree relatives with colon cancer, a screening programme of family members for early diagnosis and treatment of potential colon or endometrial cancer development is initiated. The human and economic costs in relation to screening programmes are severe. Consequently, a need for identifying colon cancers with a hereditary character exists. Further, these patients have a poor prognosis, as they have an increased risk of metachronous colon tumors and a highly increased risk of getting cancer in the endometrium (females), upper urinary tract and a number of other organs. Thus, one may regard the determination of a colon tumor as being sporadic or hereditary as determination of a prognostic factor.
Tumors appearing to be similar—morphologically, histochemically or microscopically—can be profoundly different. They can have different invasive and metastasizing properties, as well as respond differently to therapy. There is thus a need in the art for methods which distinguish tumors and tissues on different bases than are currently in use in the clinic. Determination of microsatellite status using an array-based methodology is faster than conventional DNA based methods, as it does not require microdissection, and forms a set of genes that can be combined with other sets of genes on a colon cancer array that can be used to determine microsatellite status as well as e.g. predict disease course by identifying hereditary cases or other prognostic important factors, and finally predict therapy response.
In one aspect the present invention relates to a method of classifying cancer in an individual having contracted cancer comprising
in a sample from the individual having contracted cancer determining the microsatellite status of the tumor and
in a sample from the individual having contracted cancer, said sample comprising a plurality of gene expression products the presence and/or amount which forms a pattern, determining from said pattern a prognostic marker, wherein the microsatellite status and the prognostic marker is determined simultaneously or sequentially
classifying said cancer from the microsatellite status and the prognostic marker.
The cancer may be any cancer known to be microsatellite instable in at least a fraction of the cases, such as colon cancer, uterine cancer, ovary cancer, stomach cancer, cancer in the small intestine, cancer in the biliary system, urinary tract cancer, brain cancer or skin cancer. These cancers are part of the spectrum of cancers that belong to the hereditary non-polyposis colon cancer syndrome, but the invention is not limited to this syndrome.
Gene expression patterns may be formed by only a few genes, but it is also a preferred embodiment that a multiplicity of genes form the expression pattern whereby information for classification of cancer can be obtained.
Furthermore, the invention relates to a method for classification of cancer in an individual having contracted cancer, wherein the microsatellite status is determined by a method comprising the steps of
in a sample from the individual having contracted cancer, said sample comprising a plurality of gene expression products the presence and/or amount of which forms a pattern that is indicative of the microsatellite status of said cancer,
determining the presence and/or amount of said gene expression products forming said pattern,
obtaining an indication of the microsatellite status of said cancer in the individual based on the step above.
Yet another aspect of the invention relates to a method for classification cancer in an individual having contracted cancer, wherein the hereditary or sporadic nature is determined by a method comprising the steps of
in a sample from the individual having contracted cancer, said sample comprising a plurality of gene expression products the presence and/or amount of which forms a pattern that is indicative of the hereditary or sporadic nature of said cancer,
determining the presence and/or amount of said gene expression products forming said pattern,
obtaining an indication of the hereditary or sporadic nature of said cancer in the individual based on the step above.
The present invention further concerns a method for treatment of an individual comprising the steps of
selecting an individual having contracted a colon cancer, wherein the microsatellite status is stable, determined according to any of the methods as defined herein
treating the individual with anti cancer drugs.
Another aspect of the present invention relates to a method for treatment of an individual comprising the steps of
selecting an individual having contracted a colon cancer, wherein the microsatellite status is instable, determined according to any of the methods as defined herein
treating the individual with anti cancer drugs.
Yet another aspect of the present invention relates to a method for reducing malignancy of a cell, said method comprising
contacting a tumor cell in question with at least one peptide expressed by at least one gene selected from genes being expressed at least two-fold higher in tumor cells than the amount expressed in said tumor cell in question.
Additionally, the present invention concerns a method for reducing malignancy of a tumor cell in question comprising,
obtaining at least one gene selected from genes being expressed at least two fold lower in tumor cells than the amount expressed in normal cells
introducing said at least one gene into the tumor cell in question in a manner allowing expression of said gene(s).
The invention also relates to a method for reducing malignancy of a cell in question, said method comprising
obtaining at least one nucleotide probe capable of hybridising with at least one gene of a tumor cell in question, said at least one gene being selected from genes being expressed in an amount at least two-fold higher in tumor cells than the amount expressed in normal cells, and
introducing said at least one nucleotide probe into the tumor cell in question in a manner allowing the probe to hybridise to the at least one gene, thereby inhibiting expression of said at least one gene.
In a further aspect the invention relates to a method for producing antibodies against an expression product of a cell from a biological tissue, said method comprising the steps of
obtaining expression product(s) from at least one gene said gene being expressed as defined herein
immunising a mammal with said expression product(s) obtaining antibodies against the expression product.
The present invention also concerns a method for treatment of an individual comprising the steps of
selecting an individual having contracted a colon cancer, wherein the microsatellite status is stable, determined according to any of the methods as defined herein
introducing at least one gene into the tumor cell in a manner allowing expression of said gene(s).
The present invention further relates to a pharmaceutical composition for the treatment of a classified cancer comprising at least one antibody as defined herein.
In yet another aspect the invention concerns a pharmaceutical composition for the treatment of a classified cancer comprising at least one polypeptide as defined herein.
Further, the invention relates to a pharmaceutical composition for the treatment of a classified cancer comprising at least one nucleic acid and/or probe as defined herein.
In an additional aspect the present invention relates to an assay for classification of cancer in an individual having contracted cancer, comprising
at least one marker capable of determining the microsatellite status in a sample and at least one marker in a sample determining the prognostic marker, wherein the microsatellite status and the prognostic marker is determined simultaneously or sequentially.
Unsupervised Hierarchical Clustering of Colorectal Tumors Based on the 1239 Genes with the Highest Variation Across all Tumors.
The phylogenetic tree shows the spontaneous clustering of tumor samples and normal biopsies. Germline mutation indicates samples with hereditary mutations in either MLH1 or MSH2 genes. In columns referring to results of immunohistochemistry a plus indicates a positive antibody staining. Tumor location indicates right-sided or left-sided location in the colon of the tumor.
Summary of the Performance of the Microsatellite Instability Classifier Based on Microarray Data.
Panel A shows the number of classification errors as a function of the number of genes used. Panel B shows log2 of the ratio of the distance between a tumor to the centers of the microsatellite instable group and the microsatellite stable tumors. A value of +2 indicates that the distance of a tumor to the microsatellite instable group is 4 times the distance to the microsatellite stable group. Open bars are MSI tumors and solid bars are MSS tumors. Panel C shows the result of the permutation analysis for estimation of the stability of the classifier. This was estimated by generating one hundred new classifiers based on randomly chosen datasets from the 101 tumors each consisting of 30 microsatellite stable and 25 microsatellite instable samples. In each case the classifier was tested with the remaining 46 samples. The performance for each set was evaluated and averaged over all 100 training and test sets.
Classification of MSI Tumors as Hereditary or Sporadic Cases Based on Two Genes.
Panel A shows the number of classification errors as a function of the number of genes used. In crossvalidation we found a minimum number of one error using two genes and adding more genes increased the number of errors to a maximum number of twelve. Both genes were used in at least 36 of the 37 crossvalidation loops. Panel B shows log2 of the ratio of the distance between a tumor to the centers of the sporadic microsatellite instable group and the hereditary microsatellite instable group. Panel C shows microarray signal values for MLH1 and PIWIL1 genes for all tumors. Asterisk indicates the misclassified tumor
Classification of Microsatellite-Instability Status Based on Real-Time PCR.
Panel A shows a cluster analysis of 18 of the 101 tumors samples and 9 genes based on the microarray data and compared to real-time PCR data from same samples and genes. Dark colors indicate relative low expression and light/light grey color palette high expression. Panel B shows the result of 47 new independent samples based on PCR data from 7 of the 9 genes. Relative distances are explained in the legend to
Kaplan-Meier estimates of crude survival among patient with Stage II and Stage III colorectal cancer according to microsatellite status of the tumor, determined by gene expression. Open triangles indicate censored samples. The patients left at risk are denoted in brackets. The P values were calculated with use of the log-rank test.
Phylogenetic tree resulting from unsupervised hierarchical clustering. Cluster analysis of colon specimens with associated clinicopathological features.
Multidimentional scaling plot showing distances between groups of tumors.
Performance of prediction of survival before and after separation in MSI-H and MSS
Performance of the classifier for identification of hereditary disease.
Kaplan Meier estimates of overall survival among patients with Dukes' B and Dukes' C colon cancer according to microsatellite-instability status of the tumor, determined by gene expression.
The present inventors have, using large-scale array-based screenings, found a pool of genes, the expression products of which may be used to classify cancer in an individual. The presence of expression products and level of expression products provides an expression pattern which is correlated to a specific status and/or prognostic marker of the cancer. Characterization of the genes or functional analysis of the gene expression products as such is not required to classify the cancer based on the present method. Thus, the expression products of the plurality of genes can be used as markers for the classification of disease.
One aspect of the present invention concerns a method for classifying cancer in an individual having contracted cancer by determining the microsatellite status and a prognostic marker in a sample. Determination of the microsatellite status and the prognostic marker may be performed simultaneously or sequentially. In one embodiment of the present invention the microsatellite status is determined. The prognostic marker is determined in a sample, wherein the presence and/or the amount of a number of gene expression products form a pattern wherefrom the prognostic marker is determined. Based on the information gathered from the microsatellite status and the prognostic marker the cancer can be classified. In a preferred embodiment the prognostic marker is the hereditary or sporadic nature of the cancer. The hereditary or sporadic nature of the cancer can be determined through a number of steps comprising determining the presence and/or amount of gene expression products forming a pattern in a sample. The sample comprises a number of gene expression products the presence and/or amount of which forms a pattern that is indicative of the hereditary or sporadic nature of the cancer. Hereby, an indication of the hereditary or sporadic nature of the cancer is obtained.
In one embodiment of the invention the microsatellite status is determined using conventional analysis of microsatellite status as described elsewhere herein.
In another embodiment of the present invention the microsatellite status is determined by gene expression patterns wherein the presence and/or the amount of the gene expression products form a pattern that is indicative of the microsatellite status.
Classification of cancer provides knowledge of the survival chances of an individual having contracted cancer. In case of cancer which according to the present invention has been classified as a hereditary cancer, screening programmes of family members to the individual having the classified cancer can be initiated. Such screening programmes can comprise conventional screening programmes employing sequencing and other methods as described elsewhere. Thus, individuals at risk of developing cancer may be identified and action taken accordingly to detect developing cancer at an early stage of the disease greatly improving the chances of successful intervention and thus survival rates.
Classification of cancer also provides insights on which sort of treatment should be offered to the individual having contracted cancer, thus providing an improved treatment response of the individual. Likewise, the individual may be spared treatment that is inefficient in treating the particular class of cancer and thus spare the individual severe side effects associated with treatment that may even not be suitable for the class of cancer.
The use of highly variable repetitive sequences found in microsatellite regions adjacent to genes or other areas of interest may be used as markers for linkage analysis, DNA fingerprinting, or other diagnostic application.
Microsatellites are defined as loci (or regions within DNA sequences) where short sequences of DNA are repeated in tandem repeats. This means that the sequences are repeated one right after the other. The lengths of sequences used most often are di-, tri-, or tetra-nucleotides. At the same location within the genomic DNA the number of times the sequence (ex. AC) is repeated often varies between individuals, within populations, and/or between species. Due to the many repeats the microsatellites are prone to alter if there is a reduced repair of mismatches in the genome. In the present invention the traditional method of determining microsatellite status by employing microsatellite markers is replaced by determination of gene expression patterns.
An important factor in multi-step carcinogenesis is genomic instability. The development of some cancer forms is known to follow two distinct molecular routes. One route is the microsatellite stable, MSS, (and chromosomal instable pathway) which is often associated with a high frequency of allelic losses, cytogenetic abnormalities and abnormal DNA tumor contents. The second route is the microsatellite instable pathway MSI that is characterized by defects in the DNA mismatch repair system which leads to a high rate of point mutations and small chromosomal insertions and deletions. The small chromosomal insertions and deletions can be detected as mono and dinucleotide repeats (Boland C R, Thibodeau S N, Hamilton S R, et al., Cancer Res 1998; 58(22):5248-57).
One aspect of the present invention relates to the classification of cancer in an individual having contracted cancer by determining the microsatellite status and a prognostic marker. One embodiment of the invention relates to microsatellite status determined by conventional methods employing microsatellite analysis as described above. Another embodiment of the invention relates to establishing the microsatellite status by determining the presence and/or amount of gene expression products of a sample which comprises a plurality of gene expression products forming a pattern which is indicative of the microsatellite status.
The expression products of genes according to the present invention are not necessarily identical to the genes that are analysed by microsatellite markers in conventional methods of determining microsatellite status. The pattern of the gene expression products according to the present invention however correlates with information on microsatellite status that can be obtained using traditional methods.
The determination of the microsatellite status and the prognostic marker of the cancer may be performed sequentially. However, the determinations may also be performed simultaneously.
Together with knowledge of the microsatellite status in a sample of an individual having contracted cancer a prognostic marker is employed for classifying the cancer. The prognostic marker may be any marker that provides knowledge of the cancer type when combined with knowledge of microsatellite status. Consequently the prognostic marker may provide additional information on the cancer type when the microsatellite status is stable and similarly when the microsatellite status is instable. In a preferred embodiment of the present invention the prognostic marker is the hereditary or sporadic nature of a cancer given that the microsatellite status is instable. The prognostic marker may in another embodiment be a prognostic marker for any feature or trait that provides further possibilities of classifying cancer. The prognostic marker is determined in a sample comprising a number of gene expression products wherein the presence and/or amounts of gene expression products form a pattern that is indicative of the prognostic marker.
Hereditary nonpolyposis colon cancer (HNPCC) is a hereditary cancer syndrome which carries a very high risk of colon cancer and an above-normal risk of other cancers (uterus, ovary, stomach, small intestine, biliary system, urinary tract, brain, and skin). The HNPCC syndrome is due to mutation in a gene in the DNA mismatch repair system, usually the MLH1 or MSH2 gene or less often the MSH6 or PMS2 genes. Families with HNPCC account for about 5% of all cases of colon cancer and typically have the following features (called the Amsterdam clinical criteria):
Three or more first relative family members with colorectal cancer; affected family members in two or more generations; and at least one person with colon cancer diagnosed before the age of 50.
The highest risk with HNPCC is for colon cancer. A person with HNPCC has about an 80% lifetime risk of colon cancer. Two-thirds of these tumors occur in the proximal colon. Women with HNPCC have a 20-60% lifetime risk of endometrial cancer. In HNPCC, the gastric cancer is usually intestinal-type adenocarcinoma. The ovarian cancer in HNPCC may be diagnosed before age 40. Other HNPCC-related cancers have characteristic features: the urinary tract cancers are transitional carcinoma of the ureter and renal pelvis; the small bowel cancer is most common in the duodenum and jejunum; and the most common type of brain tumor is glioblastoma. The diagnosis of HNPCC may be made on the basis of the Amsterdam clinical criteria (listed above) or on the basis of molecular genetic testing for mutations in a mismatch repair gene (MLH1, MSH2, MSH6 or PMS2). Mutations in MLH1 and MSH2 account for 90% of HNPCC. Mutations in MSH6 and PMS2 account for the rest.
HNPCC is inherited in an autosomal dominant manner. Each child of an individual with HNPCC has a 50% chance of inheriting the mutation. Most people diagnosed with HNPCC have inherited the condition from a parent. However, not all individuals with an HNPCC gene mutation have a parent who had cancer. Prenatal diagnosis for pregnancies at increased risk for HNPCC is possible.
In tumors that are microsatellite instable it is often found that the DNA mismatch repair proteins that are encoded by the MLH1 or MSH2 genes are inactivated. In case of microsatellite instable hereditary non-polyposis colorectal cancers germline mutation in MLH1 and MSH2 and somatic loss of function of the normal allele has been found to be associated with the disease.
For most sporadic MSI tumors epigenetic hypermethylation of the MLH1 promoter can be found to be associated with the cancer (Cunningham J M, Christensen E R, Tester D J, et al., Cancer Res 1998; 58(15):3455-60., Kane M F, Loda M, Gaida G M, et al., Cancer Res 1997; 57(5):808-11., Herman J G, Umar A, Polyak K, et al., Proc Natl Acad Sci USA 1998; 95(12):6870-5., Kuismanen S A, Holmberg M T, Salovaara R, de la Chapelle A, Peltomaki P., Am J Pathol 2000; 156(5):1773-9).
Cancer leads to a change in the expression of one or more genes. The methods according to the invention may be used for classifying cancer according to the microsatellite status and/or the hereditary or sporadic nature of the cancer. Thus, the cancer may be any malignant condition in which genomic instability is involved in the development of cancer, such as cancers related to hereditary non-polyposis colorectal cancer, such as endometrial cancer, gastric cancer, small bowel cancer, ovarian cancer, kidney cancer, pelvic renal cancer or tumors of the nervous system, such as glioblastoma.
One particular form of cancer according to the present invention is that of the colon/rectum.
The cancer may be of any tumor type, such as an adenocarcinoma, a carcinoma, a teratoma, a sarcoma, and/or a lymphoma.
In relation to the gastrointestinal tract, the biological condition may also be colitis ulcerosa, Mb. Crohn, diverticulitis, adenomas.
The data presented herein relates to colorectal tumors and therefore the description has focused on the gene expression level as one manner of identifying genes involved in the prediction of survival in cancer tissue. The malignant progression of cancer of colon or rectum may be described using Dukes stages where normal mucosa may progress to Dukes A superficial tumors to Dukes B, slightly invasive tumors, to Dukes C that have spread to lymph nodes and finally to Dukes D that have metastasized to other organs.
The grade of a tumor can also be expressed on a scale of I-IV. The grade reflects the cytological appearance of the cells. Grade I cells are almost normal, whereas grade II cells deviate slightly from normal. Grade III appear clearly abnormal, whereas grade IV cells are highly abnormal.
The phrase colon cancer is in this application meant to be equivalent to the phrase colorectal cancer. Colon cancers may be located in the right side of the colon, the left side of the colon, the transverse part of the colon and/or in the rectum.
The samples according to the present invention may be any cancer tissue. The sample may be in a form suitable to allow analysis by the skilled artisan, such as a biopsy of the tissue, or a superficial sample scraped from the tissue. In one embodiment of the invention it is preferred that the sample is from a resected colon cancer tumor. In another embodiment the sample may be prepared by forming a suspension of cells made from the tissue. The sample may, however, also be an extract obtained from the tissue or obtained from a cell suspension made from the tissue. The sample may be fresh or frozen, or treated with chemicals.
Expression of one gene or more genes in a sample forms a pattern that is characteristic of the state of the cell. In a sample from an individual having contracted cancer a plurality of gene expression products are present. By expression pattern is meant the presence of a combination of a number of expression products and/or the amount of expression products specific for a given biological condition, such as cancer. The pattern is produced by determining the expression products of selected genes that together reveals a pattern that is indicative of the biological condition. Thus, a selection of the genes that carry information about a specific condition is developed. Selection of the genes is achieved by analyzing large numbers of genes and their expression products to find the genes that will enable the desired differentiation between various conditions, such as microsatellite status (MSS or MSI) and/or prognostic marker, such as for example the sporadic or hereditary nature of a given cancer sample. The criteria for selection of the best genes for the pattern to be indicative of given biological conditions include confidence levels i.e. how accurate are the selected genes forming an expression pattern in giving correct information of the biological condition. Thus, in one aspect of the present invention a specific pattern of gene expression profiles can be used to determine the microsatellite status in the sample. In a second aspect of the present invention the microsatellite status is determined and a specific pattern of the presence of a plurality of gene expression products and/or amount wherefrom a prognostic marker is determined.
One aspect of the invention specifically relates to a method for determining the microsatellite status in a sample of an individual having contracted cancer based on determination of the expression pattern of at least two genes, such as at least three genes, such as at least four genes, such as at least 5 genes, such as at least 6 genes, such as at least 7 genes, such as at least 8 genes, such as at least 9 genes, such as at least 10 genes, such as at least 15 genes, such as at least 20 genes, such as at least 30 genes, such as at least 40 genes, such as at least 50 genes, such as at least 60 genes, such as at least 70 genes, such as at least 80 genes, such as at least 90 genes, such as at least 126 genes selected from the group of genes listed in Table 1 below
One embodiment of the invention concerning the determination of microsatellite status is based on the expression pattern of at least 2 genes, such as at least 3 genes, such as at least 4 genes, such as at least 5 genes, such as at least 6 genes, such as at least 7 genes, such as at least 8 genes, such as at least 9 genes, such as at least 10 genes, such as at least 15 genes, such as at least 20 genes, such as at least 25 genes selected from the group of genes listed in Table 2.
or from
or from
or from
Homo sapiens metallothionein 1H-like protein
or from
Another embodiment of the invention concerning the determination of microsatellite status is based on the expression pattern of at least 2 genes, such as at least 3 genes, such as at least 4 genes, such as at least 5 genes, such as at least 6 genes, such as at least 7 genes, such as at least 8 genes, such as at least 9 genes selected from the group of genes listed in Table 7 below.
RNA purification Colon specimens were obtained fresh from surgery and were immediately snap frozen in liquid nitrogen either as was, in OCD-compound or in a SDS/guadinium thiocyanate solution. Total RNA was isolated using RNAzol (WAK-Chemie Medical) or spin column technology (Sigma) following the manufactures' instructions.
Gene expression analysis These procedures were performed at described in detail elsewhere (Dyrskødt et al). Briefly, ten μg of total RNA was used as starting material for the target preparation as described. First and second strand cDNA synthesis was performed using the SuperScript II System (Invitrogen) according to the manufacturers' instructions except using an oligo-dT primer containing a T7 RNA polymerase promoter site. Labelled aRNA was prepared using the BioArray High Yield RNA Transcript Labelling Kit (Enzo) using Biotin labelled CTP and UTP (Enzo) in the reaction together with unlabeled NTP's. Unincorporated nucleotides were removed using RNeasy columns (Qiagen). Fifteen μg of cRNA was fragmented, loading onto the Affymetrix HG_U133A probe array cartridge and hybridized for 16 h. The arrays were washed and stained in the Affymetrix Fluidics Station and scanned using a confocal laser-scanning microscope (Hewlett Packard GeneArray Scanner G2500A). The readings from the quantitative scanning were analyzed by the Affymetrix Gene Expression Analysis Software (MAS 5.0) and normalized using RMA (robust multi array normalisation, Irizarry et al. 2002) in the statistical application R. Redundant probesets (as defined form Unigene build 168) with high correlation (>0.5) over all samples were removed, which reduced the dataset to approximately 14.400 probesets. This dataset was used a source for all further calculations in this manuscript.
For hierarchical cluster analysis 1239 genes with a variation across all samples greater than 0.5 were median-centred to a magnitude of 1. Samples and genes were then clustered using average linkage clustering with a modified Person correlation as similarity metric (Eisen et al., PNAS 95: 14863-14868, 1998). The cluster dendrogram was visualized with TreeView (Eisen).
We make a statistical test where the p-value is evaluated through permutations. For each group and gene we calculate the average and the sum of squared deviations from the average. We then sum these over the genes and the groups:
This expression is calculated for joining DK with SF and MSI with MSS such that we end up with two groups. The sum of squared deviations is denoted S2. As a test statistic we use S1/S2. A small value indicates that there is a real reduction in the deviations when going from 2 to 4 groups and thus the groups have a real significance. To judge if a value is significantly small we use permutations. For each of the four groups left when joining DK and SF we randomly allocate the members to a pseudo DK and pseudo SF in such a way that the number of members in each group are as in the original data.
To get an understanding of this separation we performed a test to see if this is caused by few genes or if many genes are involved. For this test we calculated S1=Σgenes S1(gene) and similarly with S2=Σgenes S2(gene). For each gene j we used the test statistic S1(j)/S2(j) (Table 3).
We carried out multidimentional scaling on median-centered and normalized data using CMD—scale in the statistical application R and visualized in a two-dimensional plot.
The readings from the quantitative scanning were analyzed by the Affymetrix Gene Expression Analysis Software (MAS 5.0) and normalized using RMA (robust multi array normalisation, Irizarry et al. 2002) in the statistical application R. Redundant probesets (as defined form Unigene build 168) with high correlation (>0.5) over all samples were removed, which reduced the dataset to approximately 14.400 probesets.
The microsatellite instability status classifier was based on a dataset of 4.266 genes. These genes result from the removal of genes with a variance over all tumor samples smaller than 0.2 and genes that separate Danish from Finnish samples with a t-value numerically greater than 2. We used a normal distribution with the mean dependent on the gene and the group (MSI, MSS). For each gene, we calculated the variation between the groups and the variation within the groups to select genes with a high ratio between these. To classify a sample, we calculated the sum over the genes of the squared distance from the sample value to the group mean, standardized by the variance and assigned the sample to the nearest group. The sample to be classified was excluded when calculating group means and variances.
We validated the performance of the classifier by permutation. One hundred datasets consisting of 30 MSS samples and 25 MSI samples were randomly chosen by permutation for training of the classifier with the remaining samples in each case being assign to a testset. Averages over the 100 data sets of the number of errors in the cross-validation of the training set and in the test set were used as a measure of the precision of the classifier.
Real-time PCR (RT-PCR). The procedures were as described (Birkenkamp-Demtroder) except that we used short LNA (Locked Nucleic Acid) enhanced probes from a Human Probe Library (Exiqon™). In short, cDNA was synthesized from single samples some of which were previously analyzed on GeneChips. Reverse transcription was performed using Superscript II RT (Invitrogen). Real-time PCR analysis was performed on selected genes using the primers (DNA Technology) and probes (Exiqon, DK) described in figure legend X. All samples were normalized to GAPDH as described previously (Birkenkamp-Demtroder et. al. Cancer Res., 62: 4352-4363, 2002).
The 79 tumors samples that were not analysed by real-time PCR were transformed into log ratios using one of the tumor samples as reference and used for training of the classifier. Then 23 samples of which 18 were also analyzed on arrays were equally transformed into log ratios using the same tumor sample as above as reference and tested. The idea behind this translation is that we expect the normalized PCR values to be proportional to the normalized array values, and on a log scale this becomes an additive difference. The difference is gene specific and is therefore estimated for each gene separately. The variation obtained from the microarray data, and used in the classifier, can be used directly on the PCR platform.
The clinical specimens used in this study were collected in two different countries from 14 different clinics in the period 1994 to 2001. The samples were selected to keep a balanced representation of microsatellite instable (MSI) and microsatellite stable (MSS) tumors from both the right- and left-sided colon. The MSI class was represented both by sporadic MSI and hereditary MSI (HNPCC) tumors. Only Dukes' B and Dukes' C tumor samples were included were selected (table 19). Before any attempt to divide a diverse sample collection into distinct classes analyzed the data for systematic bias that may have been introduces during the experimental procedures. A fast and easy way to discover both true distinct classes as well as systematic biases in the data is to perform a hierarchical clustering.
The phylogenetic tree resulting from hierarchical clustering on 1239 genes (
Inspection of the gene cluster dendrogram shows that the two groups of MSS tumors are mainly separated by a large cluster of genes being upregulated in the Danish samples (data not shown) indicating that a systematic difference between Danish and Finnish samples.
Based on these observations, we performed a series of test to evaluate if the observed separation of tumors into MSS and MSI as well as DK and SF are significant. For these tests the tumor samples were grouped into four virtual tumor-groups labelled, i.e. Danish MSI (MSI-DK), Danish MSS (MSS-DK), Finnish MSI (MSI-SF) and Finnish MSS (MSS-SF). Based on 5082 genes with a variance above 0.2, we tested if all four groups are significant or if some of the groups can be joined. We considered the two possibilities of joining DK and SF, and of joining MSI and MSS and made a statistical test where the p-value is evaluated through permutations. In 100 permutations of each group combination our test value S1/S2 is considerably smaller than in all permutation (Table 20) demonstrating a very clear separation between DK and SF and between MSI and MSS.
Such a clear distinction between groups may rely on a few highly separating genes or a general difference in the gene expression profile including many genes. For both the DK-SF and MSI-MSS the effect are caused by many genes even at very criteria, i.e. low test statistic S1(j)/S2(j) values (Table 21).
When a property is present that influences a large proportion of the genes this may obscure separation of clinical relevant features in unsupervised clustering. To visualize the effect of such properties, we calculated distances by multidimensional scaling between samples with and without of 816 genes separating DK from SF with a t-value numerically greater than 2 (
For the construction of a classifier we used the expression profiles from 97 tumors for which no ambiguity had been identified in relation to microsatellite status. The 816 genes separating DK from SF were excluded, as these would be unreliable for MS classification. We built a maximum likelihood classifier in order to select a minimum of genes giving the largest possible separation of the two groups. We tested the performance of the classifier using 1-1000 genes and found that it was stable showing 3-6 errors when using 4-400 genes. Of these 106 genes were especially suited for discrimination of MSS from MSI (table 22).
The minimum of three errors was found even using only 7 genes (Table 23).
Application of the 7-gene classifier to the four samples showing ambiguity in the microsatellite analyses assigns all four to be microsatellite stable tumor class. Notably, all four showed expression levels of Tumor Growth Factor β induced protein (TFGBI), MLH1 and thymidylate synthase (TYMS) that are atypical for MSI tumors. Furthermore, these tumors were all from the left colon. Thus the misclassified tumors are clearly truly MSS or they belong to a yet undefined class of MSI tumors.
To estimate the stability of the classifier based on all 97 tumor samples, we generated one hundred new classifiers based on randomly chosen datasets consisting of 30 MSS and 25 MSI samples. In each case the classifiers were tested with the remaining samples. The performance for each set was evaluated and averaged over all 100 training and test sets (Table 24). The mean error rate for MSS tumors was 0.52% and 1.38% for MSI tumors. The seven genes defined above were found to be those genes that were most frequently used in the crossvalidation loop. More than 50% of the errors were related to three tumors of which two were wrongly classified in all permutation and one in 94%. The remaining errors were mainly caused by four tumors with error rates of 40-47% showing that the former three samples are truly assigned contradictory to result from the microsatellite analysis and that four samples could not be assigned with confidence too any of the classes.
Using the same classification methods described above, we build classifiers for survival based on either all samples or the above defined groups of MSI-H and MSS. As seen in
In order to identify a gene set for identification of hereditary microsatellite instable tumors we applied 19 sporadic microsatellite instable samples and 18 microsatellite instable samples to supervised classification as described above. We found ten genes we high scored for separation of sporadic MSI-H from hereditary MSI-H tumours (Table 26). In crossvalidation we found a minimum number of one error using two genes (
Real time PCR was applied both to verify the array data and examine if the 7-gene classifier would also perform on this platform. We chose 23 samples of which 18 were also analyzed on arrays. The correlation between the two platforms was high (data not shown). In order to test the performance of classification using PCR data we re-build our classifier with a 79 samples array dataset including only those tumors that were not analyzed with PCR. Two samples were classified in discordance with the microsatellite instability test of which one of them was ambiguously classified by the 7-gene array classifier.
Based on the 7-gene classifier, classification of 36 patients with Dukes' B tumors receiving no adjuvant chemotherapy, 18 were classified as MSI tumors and 18 as MSS tumors. The overall survival was highly significantly related to the classification since all nine patients that died within five years of follow-up were belonged to the MSS group (P=0.0014) (
Among 65 patients with Dukes' C tumors receiving adjuvant chemotherapy, 17 were classified as MSI tumors and as 48 MSS tumors. Of these, 6 MSI and 27 MSS patients died within five years of follow-up meaning no significant difference in overall survival between these groups (P=0.55) (
In the clinic the 106 or less genes described can be used for predicting outcome of colorectal cancer when examined at the RNA level and also on the protein level as each gene identified is the project is transcribed to RNA that is further translated into protein. The genes can also be used determine which patient should be treated with chemotherapy as only non-microsatellite instable tumors will respond to 5-FU based therapy. Building classifiers can achieve a further stratification of patient with god and bad prognosis after stratification into microsatellite instable and stable tumors. The genes used to identify hereditary disease can be used to decide which patient should enter into sequencing analysis of mismatch repair genes.
The RNA determination can be made in any form using any method that will quantify RNA. The proteins can be measured with any method quantification method that can determine the level of proteins.
| Number | Date | Country | Kind |
|---|---|---|---|
| PA 2003 01940 | Dec 2003 | DK | national |
| PA 2004 00096 | Jan 2004 | DK | national |
| PA 2004 00586 | Apr 2004 | DK | national |
| PA 2004 01843 | Nov 2004 | DK | national |
| Filing Document | Filing Date | Country | Kind | 371c Date |
|---|---|---|---|---|
| PCT/DK2004/000914 | 12/23/2004 | WO | 00 | 12/3/2008 |
