The present invention relates to a development of a prognostic gene and a method for predicting prognosis of the early stage breast cancer.
As human genome information has been utilizing actively, cancer researches tens to identify mechanism at genome level. In particular, microarray is able to identify cancer cell properties in a macroscopic view, on the basis of information for tens of thousands of microarray gene expression patterns or increase or decrease of the number of genes. Analyzing this genome level information is a very innovative way to understand intimately complicated life phenomena and it will be more activated. Particularly, in the case of complex diseases such as cancer, analysis for a small number of specific genes is likely to obtain a blinkered result. In addition, since capturing whole behavior patterns for oncogenesis and development of cancer is important, genome information analysis is necessary. Like this, most of genome information which is a basis for cancer research is produced using genome chip such as microarray. Technologies able to get information for tens of thousands of genes are evolving day by day. In spite of the disadvantages of high cost, researches using microarray are actively developed such that amounts of related information are explosively increasing. From the middle of in the year 2000, this genome information began to be collected into a database, whereby secondary and tertiary analysis using the information are becoming a focal point for researches of phenomenon of life.
In general expression gene chip, tens of thousands of probes representing approximately 20,000 to 30,000 genes are immobilized. In microarray measuring precisional information such as SNP, more than one million of probes are immobilized. Method of microarray is very efficient since it is relatively simple, standardized and a large amount of information may be obtained in a short time at once. However, analyzing results are key point as well as drawback. Comprehensive analysis for tens of thousands of genes incomparable to existing analysis for a small number of genes must be supported by wide knowledge and statistical analysis techniques for genome, whereby information eventually becomes useful. Besides, high-performance computing equipment is required to store and analyze large amounts of information, and the related computational techniques are also required. However, since it is difficult to perform to the researchers who familiar with traditional biological research range and experimental methods, it is not utilized usefully. Therefore, utilizing the released genome information serves a very valuable purpose. Particularly, researches for cancer have been actively introduced, whereby a considerable amount of related-information has been accumulated.
Breast cancer is possible for self-diagnosis and the importance of self-diagnosis is promoted such that breast cancer may be found in the early-stage. It is difficult to decide whether this early-stage breast cancer patients after surgery receive treatments for anticancer. Although it is possible to predict for an approximate prognosis by pathological observation, the observation result is difficult to standardize and quantify, and its reliability is low. For these reasons, it is recommended that most of patients with the early-stage breast cancer should have clinical anticancer treatments. Due to the nature of anticancer treatments, patients suffer from a huge pain and financial expenditure. It is supposed that the patients not required anticancer treatments are to be more than half of the early-stage breast cancer patients. Therefore, if unnecessary anticancer treatments are reduced by predicting prognosis of patient with analysis to the characteristics of the early-stage breast cancer, it may be a great help to life quality of patients.
As the information about tens of thousands of breast cancer gene expression patterns may be obtained using microarray at once, researches for classifying breast cancer at molecular level and revealing mechanism to cancer occurrence and development are actively carried out. It is important to predict prognosis of the early stage breast cancer patient in clinic. To develop gene for predicting prognosis using microarrays has been began already in the early 2000s.
Although researches using microarray are required high-costs, a significant number of breast cancer tissue expression profiles have been produced and opened to researchers. Starting from the development of 70 genes for predicting prognosis by analyzing the early stage breast cancer tissues and survival data followed for 10 years of 78 people in 2002, a dozen genes for predicting prognosis were published. Among them, several genes have already commercialized and used in clinic (1-13). As representatives of those, MammaPrint® (Agendia) and Oncotype DX® (Genomic Health) have been currently used in clinic. However, they have been still used as one of references to prognosis (2, 7).
Throughout this application, various patents and publications are referenced and citations are provided in parentheses. The disclosure of these patents and publications in their entities are hereby incorporated by references into this application in order to more fully describe this invention and the state of the art to which this invention pertains.
The present inventors have made intensive researches to develop gene diagnostic systems for predicting breast cancer prognosis with reliability in order to decide anticancer treatment for patients with the early-stage breast cancer. As a result, microarray data and clinical information obtained from the early-stage breast cancer tissue were collected and analyzed to develop genes related prognosis, whereby they have developed a prognostic model.
Accordingly, it is an object of this invention to provide a gene selection method for predicting cancer prognosis.
It is another object of this invention to provide a gene for predicting cancer prognosis.
It is still another object of this invention to provide a method for predicting cancer prognosis.
Other objects and advantages of the present invention will become apparent from the detailed description to follow taken in conjugation with the appended claims and drawings.
In one aspect of the present invention, there is provided a method for selecting a gene for predicting the prognosis of a cancer, comprising:
(a) collecting cancer tissues from a patient group of which clinical information has been known;
(b) classifying the patient group into a poor prognosis group and a good prognosis group; wherein the poor prognosis group comprises patients in which metastasis occurs before a reference time point, and the good prognosis group comprises patients in which metastasis does not occur after the reference time point;
(c) obtaining expression profiles of genes from the collected cancer tissues;
(d) selecting genes showing difference in expression levels between the poor prognosis group and the good prognosis group;
(e) classifying the selected genes by expression patterns into gene clusters by use of a clustering analysis for expression patterns;
(f) selecting an expression pattern having significant correlation with certain function by performing a function analysis to the gene clusters classified by expression patterns; and
(g) selecting a gene in genes belonging to the selected expression pattern; wherein the selected gene shows not only high expression level but also large difference in expression levels between the poor prognosis group and the good prognosis group.
The present inventors have made intensive researches to develop gene diagnostic systems for predicting breast cancer prognosis with reliability in order to decide anticancer treatment for patients with the early-stage breast cancer. As a result, microarray data and clinical information obtained from the early-stage breast cancer tissue were collected and analyzed to develop genes related prognosis, whereby they have developed a prognostic model.
The term “prognosis” as used herein refers symptoms in the future or prospects of progress determined by diagnosing a disease. Prognosis in patients with cancer means occurrence of metastasis or survival period after surgical procedures within certain period. Since prediction of prognosis in patients with the early-stage breast cancer, especially chemotherapy as well as the future direction of the clues for breast cancer treatment, it is a very important clinical challenge.
According to a preferred embodiment, the clinical information in the present step (a) comprises information for cancer metastasis state.
The term “metastasis” as used herein refers to a proliferating state that the primary cancer is transplanted to other parts of body by various routes to set and grow. Since occurrence of metastasis is determined by specific characteristics of the cancer, and it is an important clue in determining the prognosis of cancer, it is considered as the most important clinical information associated with the survival of cancer patients. According to the present invention, the cancer tissues are collected to obtain information about the patient's metastasis and differences of gene expression profiles between different groups in metastasis occurrence are analyzed such that prognostic marker genes may be selected.
In the present step (b), the reference time point is a period for determining the prognosis to cancer patients, it refers to a period to generation of cancer before the onset of metastasis. The reference time point is preferably 3 to 12 years after onset, and more preferably 5 to 10 years. In addition, it may be same or different period to classify a group as poor prognosis and good prognosis. Most preferably, the patients in metastasis occurred within 5 years after the onset in the patient population is classified as poor prognosis group, and the patients in metastasis no- occurred more than 10 years after the onset is classified as a good prognosis group.
In the present step (c), the term “expression profile” refers that activities of a lot of genes are simultaneously measured to obtain information about cell, tissue or organ function. Activity of gene includes transcription activity, translation activity, expression level of a protein produced and, and its activity in vivo.
Steps to collect the gene expression profiles may be carried out using, for example, microarray analysis, multiplex PCR (polymerase chain reaction), quantitative RT-PCR (reverse transcription polymerase chain reaction), transcriptome analysis using tiling array and short read sequencing, but not limited thereto, in various ways known in the art. Preferably it may be carried out by microarray analysis. To statistical analysis of the collected microarray expression profiles, various normalization methods conventionally used in the art may be used, and preferably RMA (Robust Multi-array Average) normalization method.
In the present step (d), the term “a difference in the expression level” means that specific gene expression level between the prognostic groups is significantly different when specific gene expression level between the prognostic groups are compared using microarray expression profile analysis (FDR<0.01).
The analysis of differences in expression levels may use a variety of methods commonly used in the art, preferably SAM (Significant Analysis of Microarray) analysis.
SAM analysis is an analysis using the microarray analysis algorithm SAM. The difference in expression levels between groups is calculated in the similar manner with T-test, and the significance of the difference in expression level is represented by FDR (false discovery rate, q-value). The smaller q-values, it is more significant in the difference of gene expression.
According to a preferred embodiment, the cancer is breast cancer, and more preferably the early-stage breast cancer.
According to a preferred embodiment, the method further includes, between the steps (a) and (b), the step of classifying the patient group into a patient group showing less than a reference expression level of estrogen receptor (ER) and a patient group showing more than a reference expression level of estrogen receptor (ER).
Expression occurrence of estrogen receptor is the most commonly used standard for classifying subtypes of breast cancer patients. It has been known that the lower expression level of estrogen receptors in breast cancer leads to be higher risk of metastasis of breast cancer. In clinic, pathologist divides into ER+ or ER− by reading the results of ER IHC (immuno-histochemistry). According to the present invention, the subject patient groups are classified according to the expression levels of the estrogen receptor, and classified as the ER positive group and the ER negative group to the good prognosis group and the poor prognosis group, respectively to analyze, whereby the genes showing significantly differential expression between the prognostic groups may be selected with more reliability.
The most preferably, the reference expression level to classify subtype (ER+ or ER−) for estrogen receptor (ER) is determined ROC (receiver-operating characteristics) analysis using ER IHC (estrogen receptor immuno-histochemistry) results or mRNA expression level of ESR 1 (estrogen receptor 1).
The term “clustering analysis” as used herein refers to a multivariate analysis method classifying subjects of analysis to cluster to verify the structural relationship between them.
In the present step (e), the clustering analysis may be carried out using various methods commonly used in the art, and preferably PCA (Principal Component Analysis). PCA analysis generates a small number of novel super-genes recombinated by linear combination with information of various gene variables. In other words, it is a method reducing dimensions by reducing the number of variables while loss of the original data is minimized
The term “function analysis” as used herein refers verifying biological functions for the genes highly associated with the principal component selected in the step (e).
In the present step (f), the function analysis may be carried out using various method commonly used in the art, and preferably a GO (Gene Ontology) analysis.
In the present step (g), a prognostic gene selection may be selected according to the statistical significance, and preferably, it was selected by additionally considering the correlation with the selected principal component, the average expression level and the range of quartile besides the difference of the average expression level between prognostic groups. The term “higher expression levels” as used herein refers to a case that the average expression levels in gene groups belonging to the selected expression pattern is sufficiently high to allow statistical analysis, and preferably, it is selected in the order of the top-ranked gene in the expression level among the selected gene group. The term “large differences in expression levels” as used herein refers to a case that the differences of the average expression levels in gene groups belonging to the selected expression pattern is sufficiently distinct to allow experimental analysis, preferably, it is selected in the order of the top-ranked gene in the differences of the expression level between the prognostic groups among the selected gene group, and most preferably, it is selected in the order of the top-ranked gene in the expression level among the selected gene group and the top-ranked gene in the differences of the expression level between the prognostic groups among the selected gene group.
Preferably, the present invention may further include a step developing mathematical model for survival probability using the selected prognostic gene after the present step (g). This model development may be performed by mathematizing the relationship between times as long metastasis to occur and prognostic genes through survival regression analysis in which the selected prognostic gene is covariates. The relationship between times as long metastasis to occur and prognostic genes may be verified using a variety of survival models, and preferably parametric survival analysis AFT model. Preferably, the survival model developed using the selected prognostic genes may be verified in independent dataset. The validation method may compare to the survival probability and the actual observed survival probability. Moreover, where the prognostic groups are classified using the survival model, an accuracy of the validation method may be evaluated by comparing with the actual observed prognostic group.
In another aspect of the present invention, there is provided a kit for predicting metastasis risk of a breast cancer patient comprising a primer or a probe which is specifically hybridized with a nucleotide sequence selected from a group consisting of nucleotide sequences as set forth in SEQ ID NOs:1-9.
The term “nucleotide” as used herein refers to as deoxyribonucleotide or ribonucleotide that is present in a single-strand or double-strand form, and includes natural nucleotide analogues, unless stated otherwise (Scheit, Nucleotide Analogs, John Wiley, New York (1980); Uhlman and Peyman, Chemical Reviews, 90:543-584 (1990)).
The term “primer” as used herein refers to an oligonucleotide, which is capable of acting as a point of initiation of synthesis when placed under conditions in which synthesis of primer extension product which is complementary to a nucleic acid strand (template) is induced, i.e., in the presence of nucleotides and an agent for polymerization, such as DNA polymerase, and at a suitable temperature and pH. Preferably, the primer is deoxyribonucleotide and single stranded. The primer of this invention may be comprised of naturally occurring dNMP (i.e., dAMP, dGM, dCMP and dTMP), modified nucleotide, or non-natural nucleotide. The primer may also include ribonucleotides.
The present primer may be an extension primer forming a complementary sequence to a target nucleic acid by annealing to a target nucleic acid by template-dependent nucleic acid polymerase. The immobilized probe is extended to a site annealed probe to occupy.
The extension primer used in the present invention includes a hybridizing nucleotide sequence complementary to first site of a target nucleic acid sequence. The term “complementary” is used herein to mean that primers or probes are sufficiently complementary to hybridize selectively to a target nucleic acid sequence under the designated annealing conditions or hybridization conditions, encompassing the terms “substantially complementary” and “perfectly complementary”, preferably perfectly complementary.
The term “substantially complementary” in conjunction with the primer sequence, the sequences of primers are not required to have perfectly complementary sequence to templates. The sequences of primers may comprise some mismatches, so long as they can be hybridized with templates and serve as primers.
The primer must be sufficiently long to prime the synthesis of extension products in the presence of the agent for polymerization. The suitable length of primers will depend on many factors, including temperature, application and source of primer, generally, 15-30 nucleotides in length. In general, shorter primers need lower temperature to form stable hybridization duplexes to templates. The term “annealing” or “priming” as used herein refers to the apposition of an oligodeoxynucleotide or nucleic acid to a template nucleic acid, whereby the apposition enables the polymerase to polymerize nucleotides into a nucleic acid molecule which is complementary to the template nucleic acid or a portion thereof.
The sequences of primers are not required to have perfectly complementary sequence to templates. The sequences of primers may comprise some mismatches, so long as they can be hybridized with templates and serve as primers. Therefore, the primers of this invention are not required to have perfectly complementary sequence to the nucleotide sequence as described above; it is sufficient that they have complementarity to the extent that they anneals specifically to the nucleotide sequence of the gene for acting as a point of initiation of synthesis. The primer design may be conveniently performed with referring to the above-described nucleotide sequences. For instance, the primer design may be carried out using computer programs for primer design (e.g., PRIMER3 program).
The term “nucleic acid molecule” as used herein refers to a comprehensive DNA (gDNA and cDNA) and RNA molecule, and a nucleotide as a basic unit in the nucleic acid includes not only natural nucleotides but also analogues which a sugar or base are modified (Scheit, Nucleotide Analogs, John Wiley, New York (1980); Uhlman and Peyman, Chemical Reviews, 90:543-584(1990)).
Where a gRNA is employed as starting material in the present kit, an isolation of gDNA may be carried out according to conventional methods known in the art (see: Rogers & Bendich (1994)).
Where a mRNA is employed as starting material in the present kit, an isolation of total RNA may be carried out according to conventional methods known in the art (see: Sambrook, J. et al., Molecular Cloning. A Laboratory Manual, 3rd ed. Cold Spring Harbor Press(2001); Tesniere, C. et al., Plant Mol. Biol. Rep., 9:242(1991); Ausubel, F. M. et al., Current Protocols in Molecular Biology, John Willey & Sons(1987); and Chomczynski, P. et al., Anal. Biochem. 162:156(1987)). The isolated total RNA is synthesized to cDNA using reverse transcriptase. Since total RNA molecules used in the present invention are isolated from human samples, mRNA molecules have poly-A tails and converted to cDNA by use of dT primer and reverse transcriptase (see: PNAS USA, 85: 8998 (1988); Libert F, et al., Science, 244: 569 (1989); and Sambrook, J. et al., Molecular Cloning. A Laboratory Manual, 3rd ed. Cold Spring Harbor Press (2001)).
The investigation of the certain sequence in the present kit may be carried out according to the various methods known in the art. For example, techniques that may be used in the present invention includes, but is not particularly limited to, fluorescence in situ hybridization (FISH), direct DNA sequencing, PFGE analysis, Southern blotting analysis, single-strand conformation analysis (SSCA, Orita et al., PNAS, USA 86:2776 (1989)), RNase protection assay (Finkelstein et al., Genomics, 7:167 (1990)), dot-blot assay, denaturing gradient gel electrophoresis (DGGE, Wartell et al., Nucl. Acids Res., 18:2699 (1990)), a method using proteins (e.g, mutS protein from E. coli) which recognize nucleotide mismatches (Modrich, Ann. Rev. Genet., 25:229-253 (1991)), and allele-specific PCR.
The changes in sequences lead to the difference in the binding of single-stranded intracellular bases, resulting in appearance of bands with different mobility. At this time, the bands are detected using the SSCA. The sequences having mobility different from that of a wild-type sequence are also detected using the DGGE analysis or TDGS (Two-Dimensional Gene Scanning) analysis.
Other techniques are generally carried out using probes or primers which are complementary to the sequence including the nucleotides of the present invention.
For example, a riboprobe that is complementary to the sequence including the nucleotide of the present invention is used in the case of the RNase protection assay. The isolated DNA or mRNA is hybridized with the riboprobe, and then digested with an RNase A enzyme that can detect nucleotide mismatches. Smaller bands are observed if the nucleotide mismatches are recognized by the RNase A.
A probe complementary to the nucleotide of the present invention is used in the case of the analysis using a hybridization signal. Hybridization signals of the probe and a target sequence are detected to directly determine DM or MS in this technique.
As used herein, the term “probe” means a natural or modified monomer, or a linear oligomer having a bond(s), wherein the natural or modified monomer includes deoxyribonucleotides and ribonucleotides that can be hybridized with a specific nucleotide sequence. Preferably, the probe is present in a single-strand form for the purpose of the maximum efficiency in hybridization. The probe is preferably deoxyribonucleotide.
A nucleotide sequence that is perfectly complementary to the nucleotide sequence may be used as the probe used in the present invention, but nucleotide sequences that are substantially complementary to the nucleotide sequence may be used without obstructing the specific hybridization. Generally, stability of a duplex formed through the hybridization tends to be determined by the consensus of the terminal sequences, and therefore the duplex may be broken down under stringent conditions if the terminal region of the probe having a base complementary to the present nucleotide sequence is not hybridized with the 3′- or 5′-terminus of the probe.
The condition that is suitable for the hybridization may be determined with reference to the context disclosed in Joseph Sambrook, et al . , Molecular Cloning, A Laboratory Manual, Cold Spring Harbor Laboratory Press, Cold Spring Harbor, N.Y. (2001) and Haymes, B. D., et al . , Nucleic Acid Hybridization, A Practical Approach, IRL Press, Washington, D.C. (1985). The stringent condition used in the hybridization may be determined by adjusting the temperature, the ionic strength (concentration of buffer) and the presence of compounds such as organic solvent. This stringent condition may be differently determined, depending on the sequences to be hybridized.
According to a preferred embodiment, the nucleotide sequence selected from a group consisting of nucleotide sequences as set forth in SEQ ID NO:1 to SEQ ID NO:4 is high-expressed in a patient with high-risk of metastasis of breast cancer, and the nucleotide sequence selected from a group consisting of nucleotide sequences as set forth in SEQ ID NO:5 to SEQ ID NO:9 is low-expressed in a patient with high-risk of metastasis of breast cancer which shows no significant differences in expression levels of the nucleotide sequence selected from a group consisting of nucleotide sequences as set forth in SEQ ID NO:1 to SEQ ID NO:4.
According to the present invention, as a result of function analysis to each gene showing a difference of an expression level between the prognostic groups, the nucleotide sequence selected from a group consisting of nucleotide sequences as set forth in SEQ ID NO:1 to SEQ ID NO:4 is a gene involved in the proliferation of cancer cells, and the nucleotide sequence selected from a group consisting of nucleotide sequences as set forth in SEQ ID NO:5 to SEQ ID NO:9 is a gene involved in the immune response.
In still another aspect of the present invention, there is provided a method for predicting metastasis risk of breast cancer patient comprising measuring the expression of the nucleotide sequence selected from a group consisting of nucleotide sequences as set forth in SEQ ID NO:1 to SEQ ID NO:9. Since the nucleotide sequence and its expression level measurement method used in the present invention is described as above, the common descriptions between them are omitted in order to avoid undue redundancy leading to the complexity of this Specification.
The features and advantages of this invention will be summarized as follows:
(a) The present invention provides a gene selection method for predicting prognosis of cancer, the selected gene for predicting the prognosis of cancer, and a kit and a method for predicting metastasis risk of breast cancer patient using thereof.
(b) The present invention may predict to the patient's prognosis by analyzing for the genetic characteristics of the early stage breast cancer, whereby the present invention may be used to advantage in prognosis diagnosis which may reduce unnecessary anticancer therapy.
The present invention will now be described in further detail by examples. It would be obvious to those skilled in the art that these examples are intended to be more concretely illustrative and the scope of the present invention as set forth in the appended claims is not limited to or by the examples.
Expression profiles and clinical information obtained with frozen tissue of the early stage breast cancer were collected from the Gene Expression Omnibus (GEO, at the World Wide Web: ncbi.nlm nih.gov/geo). Each of total nine independent expression profile sets was consisting of more than 100 samples. It was made to perform prognosis related-researches for the early stage breast cancer patients (2, 4, 9, 10, 13, 25, 32, 33). Eight dataset of them were generated with same platform, Affymetrix® U133A, and last dataset was generated by Agilent Hu25K. Most of them, the patient's clinical information (age, sex, tumor size, state of tumor metastasis and degree of tumor differentiation) and survival information were collected together. Six datasets of eight data sets generated by Affymetrix® U133A information was about distant-metastasis free survival, and last two datasets were about overall survival. Agilent data included survival information for metastasis. Metastasis is the most important event to decide prognosis, and it is determined by the unique characteristics of cancer. In addition, the most frequent patients had metastasis information in the collected data. Based on this metastasis occurrence, survival analysis was performed. From the precise investigation on the expression data and clinical information, about 186 cases turned out to be overlapped in datasets, and the duplicated cases were removed in order to prevent double-counting in statistical analysis such that total 1,861 unique cases were researched. Seven dataset generated with same platform (Affymetrix® U133A) were pooled into a grand dataset, and their raw expression files (.CEL) were combined together by re-preprocessing with RMA algorithm were pooled into a grand dataset, and their raw expression files (.CEL) were normalized with rma (background correction:rma, normalization:quantile, summarization:medianpolish) and ENTREZG version 13 of custom CDF (at the World Wide Web brainarray.mbni.med.umich edu/Brainarray/) developed by Manhong Dai et al (34). After normalization, the 1-color expression values of the discovery dataset were transformed to 2-color-like values by subtracting the mean of each probe in the discovery dataset. The last dataset generated by Agilent Hu25K was used as another validation set. Separately, five datasets of total eight normalized dataset were hold together to use as the discovery dataset, two datasets were hold together to use a validation dataset 1, and the remaining one was used as a validation dataset 2. Agilent dataset was used as a validation dataset 3.
To determine the genes showing significantly differential expression between good prognosis group and poor prognosis group, the present inventors defined clinical outcomes in a strict manner to minimize the biases arising from censored survival data. According to the distribution of the distant-meta in the discovery dataset, about 73% of the metastatic patients developed distant-meta within 5 years, and only 7% of them developed distant-meta after 10 years. The present inventors defined good prognosis as ‘no distant-meta for more than 10 years’, and poor prognosis as ‘distant-meta within 5 years’. Following the definitions, 281 patients were assigned to ‘good prognosis group’, and 217 patients were assigned to ‘poor prognosis group’. The median distant-meta survival of the poor prognosis group was 2.4 years. The median distant-meta survival of the good prognosis group was 12.9 years. Estrogen receptor (ER) expression is the most commonly used classification criterion for subtypes of breast cancers. In clinic, pathologist divides into ER+ or ER− by reading the results of ER IHC (immuno-histochemistry). To determine the ER status of each sample in the discovery dataset, we used mRNA expression levels of ESR1 on the expression profiles, because about 200 of patients missed ER IHC results and even the available ER IHC results would not be incompatible across independent datasets in which independent decision rules for ER IHC status were applied to.
ROC (region of convergence) analysis was performed using ER IHC results and mRNA expression levels of ESR1 in the patients having ER IHC results. Under the gold standard of ER IHC results, the cutoff for ER status was determined at the point where the accuracy was highest (accuracy=0.88). According to the cutoff, the cases showing higher expression than the cutoff were classified as ER+, and the other cases were classified as ER−. In the discovery dataset, 864 peoples were placed in ER+, and 240 peoples were placed in ER−.
The present inventors defined good prognosis group as ER+, and poor prognosis group as ER− in the discovery dataset. Following the definitions, 281 patients were assigned to good prognosis group, and 217 patients were assigned to poor prognosis group. Through SAM (Significant Analysis of Microarray), the present inventors selected the genes showing significantly differential expression between two prognosis groups. Using q-value of SAM analysis, 182 of overexpressed genes were selected in good prognosis group, and 120 of overexpressed genes were selected in poor prognosis group. As a result, set of total 302 genes not duplicated has been created. To identify expression patterns of these genes, cluster analysis was performed using PCA (Principal Component Analysis) method. Two principal components were selected, and each cluster was performed by GO function analysis in order to investigate biological functions related to each principal component (Tables 1 to 3).
As a result of GO analysis, it is determined that the principal component 1 was concentrated in the proliferation, and the principal component 2 was concentrated in the immune response. In genes belonging to the two principal components involved in the proliferation and the immune response, 4 genes and 5 genes showing the highest expression level between two prognosis groups were selected respectively. Each gene set is named as p-gene representing expression patterns of the proliferation and i-gene representing expression patterns of the immune response.
The present inventors performed regression analysis in which expression levels of p-gene and i-gene were covariates using the accelerated failure time (AFT) models based on various parametric distributions. 4 p-genes were converted to p.mean, and 5 i-genes to i.mean by calculating the average per patient to apply. AFT models are specified as
T
i
=T
0exp(β1χ1+β2χ2+ . . . +βqχq)·ϵi (1)
wherein T, is the survival time of the i th object, To is the baseline survival time, χi is the vector of the covariates (j=1,2, . . . ,q), β is the coefficient of the corresponding covariates and E is the error. Since the covariates synergistically influence the baseline survival time in this model, it is called the accelerated failure time (AFT) model in the industry frequently used this. Synergistic effect on the survival time φ=β1χ1+β2χ2+ . . . +βqχq is called the acceleration factor.
Where the natural logarithm in Equation (1) is calculated, AFT models are specified as
logTi=logT0+β1χ1+β2χ2+ . . . +βqχq+ϵ* (2)
whereby AFT model has the same form as the general linear regression model.
However, since the dependent variable logT not only normally distributes but also there must always exist censored cases in survival analysis data, Equation (2) cannot be processed as the linear regression model. It is cumbersome to process practical statistics because distribution in ϵ* of Equation (2) may differ depending on each dataset. To overcome this, logT0 and ϵ* are modified and expressed by as follows.
logT=β0+β1χ1+β2χ2+ . . . +βqχq+σW (3)
wherein W follows in the distribution of LogT, and the distribution is fixed as the value of the standardized distribution. The scale parameter σ is a constant, and its value is determined by the dataset. The present inventors selected the best model from a set of candidate prognostic models by fitting them to the Weibull, loglogistic and lognormal distributions. The shape of hazard function was explored by fitting the hazard rate calculated from the life table of the discovery dataset to risk distribution for AFT model. Since the hazard function obtained by the life table shows a form of unimodal, it was predicted that the Weibull, loglogistic and lognormal distributions would appropriate. The final model was chosen considering Akaike's information criterion (AIC) and the R square (R2).
The performance of the prognostic model was assessed in terms of ‘calibration’ and ‘discrimination’. ‘Calibration’ is the degree of correspondence between the estimated probability produced by the model and the actual observed probability. ‘Discrimination’ is the ability of the model to correctly separate the subjects into different groups. Calibration was evaluated with the calibration curves plotted as observed versus predicted survival. ‘Predicted’ means predicted survival probability by a model, and ‘observed’ refers to the corresponding Kaplan-Meier survival estimate. For each patient in the data set, the model was used to predict survival probabilities for all time points up to the date of the occurrence of distant-meta. The predicted survival probabilities were averaged over all cases in the sub set at each time point (from 0 to 25 by 0.1) to give a survival curve representative of the sub group. Using this method allows comparison of observed and predicted survival in groups of patients. Together comparison of survival probability for the overall survival time, the survival probability for 5-year was also compared. A prognostic index of all patients in the given dataset was divided into four areas, and survival probabilities of patients belonging to each area were compared with KM graph to discriminate. The prognostic index is the dependent variable in the model. The more the KM graph to the four prognostic groups is clearly divided, the better model has discrimination power.
‘Calibration’ and ‘determination’ for discovery dataset and three independent validation datasets was investigated.
All statistical analyses were performed using the open source statistical language R with the packages below.
affy: preprocessing of .CEL files using rma algorithm
samr: mining of the genes showing differentially expressed between two prognosis groups
GOstats: GO analysis for the identified expression patterns
KMsurv: creating of a life table of the discovery dataset
rms: fitting prognostic models to various parametric distribution and calibration of the model using AFT model
Selection of the p-Genes and i-Genes
Five independent breast cancer datasets were pooled into a grand discovery data set consisted of 1,072 unique cases. All patients did not receive chemotherapy, and they have no metastasis of axillary lymph node (NO or N-), or have the early stage breast cancer (1st stage or 2nd stage). Among them, 1072 peoples with information about metastasis were performed to target statistical analysis. To search for genes associated with prognosis, the present inventors divided into good prognosis as ‘no distant-meta for more than 10 years’ and poor prognosis as ‘distant-meta within 5 years’ to compare expression profiles. 182 of overexpressed genes were selected in good prognosis group, and 120 of overexpressed genes were selected in poor prognosis group (FDR<0.001). PCA (Principal Component Analysis) were performed to the selected 302 genes. GO function analysis was performed to principal Component 1 and 2. It is determined that the principal component 1 was concentrated in the proliferation, and the principal component 2 was concentrated in the immune response. In genes belonging to the two principal components involved in the proliferation and the immune response, 4 genes and 5 genes showing the highest expression level between two prognosis groups were selected respectively. Based on this, the present inventors selected 4 genes from principal component 1 (proliferation) and 5 genes from principal component 2 (immune response) to develop a prognostic model.
The 9 genes were selected genes that are not only associated with prognosis, but also have the largest expression difference between the groups. Each gene is named as p-gene representing expression patterns of the proliferation and i-gene representing expression patterns of the immune response.
Occurrence of estrogen receptor (ER) expression is known to be closely related to generation and development of breast cancer. Two functions (i.e., proliferation and immune response) representing the genes selected with regard to the prognosis are interesting functions in mechanism of cancer. Using the expression levels of the p-genes and i-genes as a measure of biological activity, we compared ER+ tumors and ER− tumors in terms of the activity of proliferation and immune response. We stratified the pooled data set into 3 subgroups according to the expression levels of either the p-genes or the i-genes, and the subgroups represented 3-steps intensity of proliferation or immune response (p1, p2, p3 or il, i2, i3). p1 was a group of the lowest expression level of the p-gene and it was considered to be the slowest proliferation. p3 was a group of the highest expression levels of the p-genes and it was considered to be the most active proliferation. p2 showing a moderate expression level of the p-gene was considered to be a moderate proliferation. i1 was a group of the lowest expression level of the i-gene and it was considered to be the weakest immune response. i3 was a group of the highest expression levels of the i-genes and it was considered to be the strongest immune response. i2 showing a moderate expression level of the i-gene was considered to be a moderate immune response.
About 62% of ER− tumors were highly proliferating tumors (p3), while only 18% of ER+ tumors were highly proliferating (p3) supporting that ER− tumors tend to be more aggressive than ER+ tumors. In the same sense, about 35% of ER+ tumors were slowly growing tumors, but there were only 9% in ER− tumors. Predominant immune response was another characteristic of ER− tumors. About 38% of ER− tumors showed high activity of immune response (i3), while only 21% of ER+ tumors showed highly active immune response (i3) supporting the inhibition of ER on immune response (
Beside, a good correlation between histologic grade and proliferation was observed from the concentration of high grade (G3) along highly proliferating cases (p3) in both ER statuses. Deleterious effect of active proliferation on clinical prognosis was also observed in both ER statues from the high concentration of poor prognosis (development of distant-meta within 5 years) along highly proliferating cases (p3).
Overall, both of proliferation and immune response in ER− breast cancer were very active as compared to ER+ breast cancer, and it was supposed that ER expression levels influences in mechanisms of the generation and the development of breast cancer.
Development of a Prognostic Model with the p-Genes and i-Genes
Using the expression levels of the p-genes and i-genes, we developed a prognostic model for early breast cancer patients using the accelerated failure time (AFT) models. To reduce the number of candidate models, we roughly checked the distribution of hazard and the linearity of the selected variables. We fitted the hazard calculated from the life-table of the discovery data set to various distributions.
Since the hazard function obtained by the life table shows a form of unimodal, it was predicted that the Weibull, loglogistic and lognormal distributions would appropriate. Covariates included in the prognostic model are the p.mean and i.mean. p.mean is the average of the p-genes, and i.mean is the average of the i-genes.
As a result of applying the Weibull, loglogistic and lognormal distributions to three models, the lognormal distribution showed the best fit. Using AIC (Akaike's information criterion), the final model (3) was selected.
log (T)=−0.689×p.mean+0.274×i.mean+3.219
According to the model, the p.mean showed a predominantly negative correlation (−0.689, p value=2.47×e−17) with survival time (T) indicating that high activity of proliferation was corresponding to short survival time. In contrast to the p.mean, the i.mean showed positive correlation (0.274, p value=3.69×e−11) with survival time (T) indicating that high activity of proliferation was corresponding to long survival time. Therefore, it could be understood that immune response act as the defense mechanism by immune response against high proliferation activity, whereas the proliferation play a pivotal role in breast cancer prognosis, and its high activity shows bad prognosis.
The present inventors assessed the performance of the prognostic model in two ways, its ‘calibration’ and ‘discrimination’ aspects, using the expression profile to 1,072 of the early stage breast cancer patients in discovery dataset. ‘Calibration’ is the degree of correspondence between the estimated probability produced by the model and the actual observed probability. The actual observed probability refers to the corresponding Kaplan-Meier survival estimate. ‘Discrimination’ is the ability of the model to correctly separate the subjects into different groups. Verifications for calibration and determination to discovery dataset and three independent validation datasets were carried out.
4 prognostic groups were classified by dividing prognostic index (PI) into 4 areas to the discovery dataset developing prognostic model. 4 prognostic groups classified by prognostic index were compared using the KM graph as the observed survival probability. As a result, it could be determined that 4 prognostic groups were very well classified, and the predicted survival probability for each prognostic group corresponded well with the observed survival probability for each prognostic group.
KM survival probability and the survival probability predicted by the prognostic model were compared using graph. For each patient in the data set, the model was used to predict survival probabilities for all time points up to the date of the occurrence of distant-meta. The predicted survival probabilities were averaged over all cases in the sub set at each time point (from 0 to 25 by 0.1) to give a survival curve representative of the sub group. Using this method allows comparison of observed and predicted survival in groups of patients. The predicted survival probability was slightly higher than survival probability by KM, but they were similar overall. Together comparison of survival probability for the overall survival time, the survival probability for 5-year was also compared. The 5-year survival probability predicted by the model was similar to the actual observed 5-year survival probability. Particularly, the higher the predicted survival probability, it was showed the greater agreement with the observed survival probability.
Three independent validation datasets was subject to more objective verification for the prognostic model.
The first validation dataset is a pooled data set combined with two independent datasets generated with Affymetrix® U133A platform. The second validation dataset was the dataset generated by Affymetric® U133A platform, in which all patients were ER+ patients taking tamoxifen for five years. The third validation dataset was generated by Agilent Hu25K platform to use for developing and validating 70 prognostic genes (commercialized as MammaPrint®). The validation datasets 1 and 2 were generated with same platform, Affymetrix® U133A. The expression levels for the validation datasets 1, 2 and the discovery dataset were standardized. The validation datasets 1 and 2 assessed in terms of calibration and discrimination aspects. The validation datasets 3 assessed in terms of discrimination aspect, since it has a problem in the expression level of standardization.
In the validation dataset 1, 4 prognostic groups were clearly classified, and the predicted survival probability for each prognostic group corresponded well with the observed survival probability for each prognostic group. The predicted survival probability to overall time showed a good agreement with the observed KM graph, and the predicted survival probability at 5 years was higher than the observed survival probability by about 2%.
In the validation dataset 2, 4 prognostic groups were not clearly classified. However, overall, it was showed that the higher the predicted survival probability, the higher the observed survival probability. The predicted survival probability to overall time showed a good agreement with the observed KM graph, and the predicted survival probability at 5 years was higher than the observed survival probability by about 2%.
Having described a preferred embodiment of the present invention, it is to be understood that variants and modifications thereof falling within the spirit of the invention may become apparent to those skilled in this art, and the scope of this invention is to be determined by appended claims and their equivalents.
1. Chang, H. Y., et al., Gene expression signature of fibroblast serum response predicts human cancer progression: similarities between tumors and wounds. PLoS Biol 2(2): p. E7(2004).
2. van de Vijver, M. J., et al., A gene-expression signature as a predictor of survival in breast cancer. N Engl J Med 347(25):1999-2009(2002).
3. van 't Veer, L. J., et al., Gene expression profiling predicts clinical prognosis of breast cancer. Nature 415(6871): 530-536(2002).
4. Wang, Y., et al., Gene-expression profiles to predict distant metastasis of lymph-node-negative primary breast cancer. Lancet 365(9460): 671-679(2005).
5. Buyse, M., et al., Validation and clinical utility of a 70-gene prognostic signature for women with node-negative breast cancer. J Natl Cancer Inst, 98(17):1183 -92(2006).
6. Paik, S., Development and clinical utility of a 21-gene recurrence score prognostic assay in patients with early breast cancer treated with tamoxifen. Oncologist 12(6):631-635(2007).
7. Paik, S., et al., A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer. N Engl J Med 351(27) :2817-2826(2004).
8. Sotiriou, C., et al., Gene expression profiling in breast cancer: understanding the molecular basis of histologic grade to improve prognosis. J Natl Cancer Inst 98(4):262-72(2006)
9. Pawitan, Y., et al., Gene expression profiling spares early breast cancer patients from adjuvant therapy: derived and validated in two population-based cohorts. Breast Cancer Res 7(6):R953-964(2005).
10. Miller, L. D., et al., An expression signature for p53 status in human breast cancer predicts mutation status, transcriptional effects, and patient survival. Proc Natl Acad Sci USA, 102(38):13550-13555(2005).
11. Bild, A. H., et al., Oncogenic pathway signatures in human cancers as a guide to targeted therapies. Nature 439(7074):353-357(2006).
12. Teschendorff, A. E., et al., A consensus prognostic gene expression classifier for ER positive breast cancer. Genome Biol 7(10):R101(2006).
13. Desmedt, C., et al., Strong time dependence of the 76-gene prognostic signature for node-negative breast cancer patients in the TRANSBIG multicenter independent validation series. Clin Cancer Res 13(11): 3207-3214(2007).
14. Kim, S. Y., Effects of sample size on robustness and prediction accuracy of a prognostic gene signature. BMC Bioinformatics 10:147(2009).
15. Hummel, M., et al., Association between a prognostic gene signature and functional gene sets. Bioinform Biol Insights 2:329-341(2008).
16. Pfeffer, U., et al., Prediction of breast cancer metastasis by genomic profiling: where do we stand? Clin Exp Metastasis 26(6): 547-558(2009).
17. Ein-Dor, L., O. Zuk, and E. Domany, Thousands of samples are needed to generate a robust gene list for predicting prognosis in cancer. Proc Natl Acad Sci USA, 103(15):5923-5928(2006).
18. van Vliet, M. H., et al., Pooling breast cancer datasets has a synergetic effect on classification performance and improves signature stability. BMC Genomics, 9:375(2008).
19. Yasrebi, H., et al., Can survival prediction be improved by merging gene expression data sets? PLoS One 4(10):e7431(2009).
20. Fan, C., et al., Concordance among gene-expression-based predictors for breast cancer. N Engl J Med 355(6):560-569(2006).
21. Reyal, F., et al., A comprehensive analysis of prognostic signatures reveals the high predictive capacity of the proliferation, immune response and RNA splicing modules in breast cancer. Breast Cancer Res 10(6):R93(2008).
22. Yu, J. X., et al., Pathway analysis of gene signatures predicting metastasis of node-negative primary breast cancer. BMC Cancer 7:182(2007).
23. Kim, S. Y. and Y. S. Kim, A gene sets approach for identifying prognostic gene signatures for prognosis prediction. BMC Genomics 9:177(2008).
24. Thomassen, M., Q. Tan, and T.A. Kruse, Gene expression meta-analysis identifies metastatic pathways and transcription factors in breast cancer. BMC Cancer 8:394(2008).
25. Schmidt, M., et al., The humoral immune system has a key prognostic impact in node-negative breast cancer. Cancer Res 68(13):5405-13(2008).
26. Schmidt, M., et al., Coordinates in the universe of node-negative breast cancer revisited. Cancer Res 69(7):2695-2698(2009).
27. Calabro, A., et al., Effects of infiltrating lymphocytes and estrogen receptor on gene expression and prognosis in breast cancer. Breast Cancer Res Treat 116(1):69-77(2009).
28. Finak, G., et al., Stromal gene expression predicts clinical prognosis in breast cancer. Nat Med 14(5):518-27(2008).
29. Ma, X. J., et al., Gene expression profiling of the tumor microenvironment during breast cancer progression. Breast Cancer Res 11(1):R7(2009).
30. Rutqvist, L. E., A. Wallgren, and B. Nilsson, Is breast cancer a curable disease? A study of 14,731 women with breast cancer from the Cancer Registry of Norway. Cancer 53(8):1793-1800(1984).
31. Mould, R. F. and J. W. Boag, A test of several parametic statistical models for estimating success rate in the treatment of carcinoma cervix uteri. Br J Cancer 32(5):529-550(1975).
32. Loi, S., et al., Predicting prognosis using molecular profiling in estrogen receptor-positive breast cancer treated with tamoxifen. BMC Genomics, 9:239(2008).
33. Zhang, Y., et al., The 76-gene signature defines high-risk patients that benefit from adjuvant tamoxifen therapy. Breast Cancer Res Treat 116(2):303-309(2009).
34. Dai, M., et al., Evolving gene/transcript definitions significantly alter the interpretation of Gene Chip data. Nucleic Acids Res 33(20):e175(2005).
35. Tusher, V. G., R. Tibshirani, and G. Chu, Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci USA, 98(9):5116-21(2001).
Number | Date | Country | Kind |
---|---|---|---|
10-2011-0000521 | Jan 2011 | KR | national |
This application is a divisional application of U.S. application Ser. No. 13/935,502 filed on Jul. 4, 2013, which is a continuation of International Application No. PCT/KR2012/000021 filed on Jan. 2, 2012, which claims priority to Korean Application No. 10-2011-0000521 filed on Jan. 4, 2011. The entire contents of the aforementioned patent applications are incorporated herein by reference. The instant application contains a Sequence Listing which has been submitted in ASCII format via EFS-Web with regard to U.S. application Ser. No. 13/935,502 and is hereby incorporated by reference in its entirety. Said ASCII copy, created on Nov. 23, 2016, is named 48116-540D01US_ST25.txt and is 15,305 bytes in size.
Number | Date | Country | |
---|---|---|---|
20170159126 A1 | Jun 2017 | US |
Number | Date | Country | |
---|---|---|---|
Parent | 13935502 | Jul 2013 | US |
Child | 15285194 | US |
Number | Date | Country | |
---|---|---|---|
Parent | PCT/KR2012/000021 | Jan 2012 | US |
Child | 13935502 | US |