US20110224908A1

US20110224908A1 - Gene signature for diagnosis and prognosis of breast cancer and ovarian cancer

Info

Publication number: US20110224908A1
Application number: US12/661,286
Authority: US
Inventors: Nancy Lan Guo
Original assignee: Individual
Current assignee: Individual
Priority date: 2007-03-22
Filing date: 2010-03-15
Publication date: 2011-09-15
Also published as: US20090197259A1; US20120123105A1; US8030060B2

Abstract

A first embodiment is a breast cancer prognosticator comprising a detection mechanism consisting a 15-gene signature. In addition there are embodiments comprised of 23-gene signatures and 28-gene signatures. The 28-gene signature may also be used for the prognosis of ovarian cancer. A second embodiment is a method to determine metastatic potential, relapse potential, or both in breast cancer patients comprising collecting a sample from an individual, removing marker-derived polynucleotide from said sample, using a detection mechanism to search for positive matches of said polynucleotides and either the 15, 23, or 28-gene signatures, and developing a quantitative expression profile. Utilizing risk analysis the individual can be placed into one of two or more groups predicting risk and/or clincopathogic variables. Another embodiment is a method to determine relapse free potential in breast cancer patients comprising collecting a sample from an individual, removing marker-derived polynucleotide from said sample, using a detection mechanism to search for positive matches of said polynucleotides and a 24-gene signature, and developing a quantitative expression profile.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

This application is a Divisional of United States patent application 12/077,992 filed on Mar. 24, 2008.

REFERENCE TO SEQUENCE LISTING, A TABLE, OR A COMPUTER PROGRAM LISTING COMPACT DISC APPENDIX

This application contains a Sequence Listing submitted on compact disk containing file name 387.Seq. The sequence listing on the compact disc is incorporated by reference herein in its entirety.

BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWINGS

The following figures are not drawn to scale and are for illustrative purposes only.

FIG. 1 is a Time dependent ROC analyses of the 28-gene signature in disease-free survival prediction in three breast cancer patient cohorts. FIG. 1 is a Time dependent ROC (t=5 years) curve of the 28-gene signature on the training set from Sotiriou et al. (8) The area under the ROC curve (AUC)=0.983.

FIG. 2 is a Time dependent ROC analyses of the 28-gene signature in disease-free survival prediction in three breast cancer patient cohorts. FIG. 2 is a AUC in year 1 to year 11 during follow-up after surgery in the patient cohort from Sotiriou et al. (8).

FIG. 3 is a Time dependent ROC analyses of the 28-gene signature in disease-free survival prediction in three breast cancer patient cohorts. FIG. 3 is a Time-dependent ROC (t=5 years) curves of the 28-gene signature on two validation sets. AUC=0.843 with 25 overlapping genes on data from van't Veer et al.(27) AUC=0.764 with 8 overlapping genes on data from Sorlie et al. (10).

FIG. 4 is a Time dependent ROC analyses of the 28-gene signature in disease-free survival prediction in three breast cancer patient cohorts. FIG. 4 is a AUC in year 1 to year 13 during follow-up after surgery on two independent patient cohorts (10;28).

FIG. 5 is a Time-dependent ROC analyses of the 28-gene signature in overall survival prediction in there breast cancer patient cohorts. FIG. 5 is a Time-dependent ROC curves at time=5 years. AUC=0.927 on data from Sotiriou et al.(8) AUC=0.808 on data from Sorlie et al.(10)

FIG. 6 is a Time-dependent ROC analyses of the 28-gene signature in overall survival prediction in there breast cancer patient cohorts. FIG. 6 is the area under the ROC curve (AUC) of overall survival prediction during the follow-up after surgery.

FIG. 7 is a Time-dependent ROC analyses of 15 genes within the 28-gene signature in relapse-free survival prediction in three breast cancer patient cohorts. FIG. 7 are Time-dependent ROC curves at time=5 years. AUC=0.92 on data from Sotiriou et al. (8)

FIG. 8 is a Time-dependent ROC analyses of 15 genes within the 28-gene signature in relapse-free survival prediction in three breast cancer patient cohorts. FIG. 8 are Time-dependent ROC curves at time=5 years. AUC=0.87 on data from Sorlie et al.(10)

FIG. 9 is a Time-dependent ROC analyses of 15 genes within the 28-gene signature in relapse-free survival prediction in three breast cancer patient cohorts. FIG. 9 are Time-dependent ROC curves at time=5 years. AUC=0.79 on data from van't Veer et al. (26).

FIG. 10 is a Time-dependent ROC analyses of 24 genes within the 28-gene signature in relapse-free survival prediction in one ovarian cancer patient cohort from Bild et al. (29)

DETAILED DESCRIPTION OF THE INVENTION

A first embodiment in this application can be an expression profile-defined prognostic model able to predict the recurrence and metastases of breast cancer and ovarian cancer by using unique gene expression patterns in tumors. Additionally, the expression profile-defined prognostic model may be used to predict the relapse-free interval and metastases-free interval. The expression based profile-defined prognostic model has been developed and is a highly accurate predictor of disease-free survival as well as overall survival in individual breast cancer patients. The expression based profile-defined prognostic model can be a gene signature such as a 15-gene signature, a 23-gene signature, or a 28-gene signature comprised of a combination of the following genes (Table 1).

TABLE 1

28 genes that quantifies disease-free survival and overall
survival of breast cancer

		UniGene
Gene	Clone_IMAGE	Cluster ID	Sequence ID

FAM134C	198917	Hs.463079	NM_178126
TOMM70A	198312	Hs.227253	NM_014820
MCF2	268412	Hs.387262	NM_001099855
			NM_005369
RAD52 Pseudogene	1377154	Hs.552577	NM_134424
MCM2	239799	Hs.477481	NM_004526
C18B11	131988	Hs.173311	NM_152260
SEC13L	757210	Hs.301048	NM_031216
			NM_001013437
SLC25A5	291660	Hs.522767	NM_001152
PLSCR1	268736	Hs.130759	NM_021105
TXNRD1	789376	Hs.434367	NM_003330
			NM_001093771
			NM_182742
			NM_182729
			NM_182743
RAD50	261828	Hs.242635	NM_005732
			NM_133482
—	46196		BX100884
			H09243
			H09242
INPPL1	703964	Hs.523875	NM_001567
—	501651	Hs.439445	AK025546
PBX2	80549	Hs.509545	NM_002586
SSBP1	125183	Hs.490394	NM_003143
—	34396	Hs.448229	BE870371
PDGFRA	376499	Hs.74615	NM_006206
ACOT4	488202	Hs.49433	NM_152331
DDOST	50666	Hs.523145	NM_005216
IGHA1	182930	Hs.497723	AK128652
S100P	135221	Hs.2962	NM_005980
FAT	591266	Hs.481371	NM_005245
FGF2	324383	Hs.284244	NM_002006
INSM1	22895	Hs.89584	NM_002196
IRF5	260035	Hs.521181	NM_001098629
			NM_002200
			NM_001098627
			NM_001098630
			NM_001098628
			NM_032643
			NM_001098631
SMARCD2	741067	Hs.250581	NM_001098426
			NM_003077
MAP2K2	769579	Hs.465627	NM_030662

There is no overlap between the disclosed gene signature and previously reported gene signatures. Of the 28 genes in Table 1, 17 are related to tumorigenesis (Table 2) and 9 genes are linked to breast cancer pathogenesis (Table 3). Furthermore, among the nine breast cancer-related genes, five genes are established breast cancer biomarkers ((MCM2, Rad50, PDGFRA, S100P, and FGF2) (Table 3)).

TABLE 2

Genes that are related to tumorigenesis

Gene	Gene Name	Function

MCF2	Mcf.2 cell line derived	Guanine nucleotide exchange
	transforming sequence	factor
MCM2	Mcm2 minichromosome	DNA replication
	maintenance
	deficient 2, mitotin
SEC13L	Seh1-like	mRNA export, nuclear pore
		distribution and cell division
PLSCR1	Phospholipid scramblase	1	Lipid transfer signaling
RAD50	RAD50 homolog	DNA repair
INPPL1	Inositol polyphosphate	Lipid metabolism
	phosphatase-like 1
TXNRD1	Thioredoxin reductase 1	Antioxidant and redox
		regulator
PBX2	Pre-b-cell leukemia	Transcriptional repressor and
	transcription factor 2	tumor suppressor
SSBP1	Single-stranded dna	DNA binding protein
	binding protein
1
PDGFRA	Platelet-derived growth	Growth factor receptor
	factor receptor
S100P	S100 calcium binding	Cell differentiation
	protein p
FAT	Fat tumor suppressor	Cell signaling suppressor
	homolog
1
FGF2	Fibroblast growth factor 2	Signaling tranduction
INSM1	Insulinoma-associated 1	Transcriptional repressor
IRF5	Interferon regulatory	Tumor suppressor gene
	factor
5
SMARCD2	Swi/snf related, matrix	chromatin remodelling
	associated, actin dependent
	regulator of chromatin,
	subfamily d, member 2
MAP2K2	Mitogen-activated protein	Signaling transduction
	kinase kinase
2

TABLE 3

Genes that are linked to breast cancer pathogenesis

			Breast Cancer
Gene	Gene Name	Function	Involvement

MCF2	Mcf.2 cell line derived	Guanine nucleotide	(+)
	transforming sequence	exchange factor
MCM2	Mcm2 minichromosome	DNA replication	(+)
	maintenance deficient 2, mitotin		biomarker (1)
RAD50		DNA repair	(+)
			biomarker (2)
TXNRD1	Thioredoxin reductase 1	Antioxidant and	(+)
		redox regulator
PDGFRA	Platelet-derived growth factor	Growth factor	(+)
	receptor	receptor	biomarker (3; 4)
S100P	S100 calcium binding protein p	Cell differentiation	(+)
			biomarker (5; 6)
FGF2	Fibroblast growth factor 2	Signaling	(+)
		tranduction	biomarker (7)
SMARCD2	Swi/snf related, matrix associated,	chromatin	(+)
	actin dependent regulator of	remodelling
	chromatin, subfamily d, member 2
MAP2K2	Mitogen-activated protein kinase	Signaling	(+)
	kinase 2	transduction

Based upon the expression profiles of these 28 genes in the data from Sotiriou et al. (8), a Linear Discriminant Analysis function classified 5-year relapse status for patients provided an accuracy of 0.92, a sensitivity of 1.90, and a specificity of 0.95. To evaluate relapse-free survival prediction, a Cox proportional hazards model was built on the 28-gene signature and the risk score was used to construct the time-dependent receiver operating curve (ROC). The area under the ROC curve (AUC) during year five was 0.983 (FIG. 1), and remained 0.92 between years 8 and 11 during the follow up (FIG. 2).
To evaluate the prognostic power of the identified gene signature, two independent validation sets were used (9;10). Using the signature genes, time-dependent ROC analyses were performed to evaluate relapse/metastases prediction on two independent patient cohorts (FIGS. 3 and 4). The area under the ROC (5-year) curve on the data from van't Veer et al. (11) was 0.843 with 25 signature genes in predicting metastatic potential. The AUC (5-year) was 0.764 on the data from Sorlie et al. (10) with eight overlapped genes in the relapse-free survival prediction (FIG. 3).
Time dependent ROC analysis showed that the 28-gene signature was also predictive of overall survival (P<0.001; FIGS. 5 and 6). In the prediction of overall survival , the AUC (5-year) was 0.927 on data from Sotiriou et al. (Sotiriou C. et al., Breast Cancer classification and prognosis based on gene expression profiles from a population-based study, Proc. Natl. Acad. Sci., USA 2003; 100:10393-8) and 0.808 on data from Sorlie et al.(Sorlie T. et al., Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications, Proc. Natl. Acad. Sci., USA 2001; 98:10869-74).
Among the 28-gene signature, 11 genes had significant association with relapse-free survival in Cox modeling (Table 4).

TABLE 4

Genes that are significantly associated with breast
cancer relapse.

	GENE	P-value

	FGF2	0.0039
	SLC25A5	0.0051
	C18B11	0.0062
	SMARCD2	0.0087
	TOMM70A	0.0250
	PBX2	0.0330
	SEC13L	0.0350
	Clone ID: 501651	0.0350
	IRF5	0.0350
	DDOST	0.0470
	Clone ID: 182930	0.0520

Among the 28-gene set, 15 genes (Table 5) predict disease-free survival with an accuracy ranging from 0.79 to 0.92 in three patient cohorts from Sotiriou et al. (8), van't Veer et al. (12), and Sorlie et al. (10) (FIGS. 7, 8, and 9). These 15 genes can be used as a 15-gene signature prognostic model for breast cancer. In addition, the 8 unique genes from Table 4 may be added to form a 23-gene signature prognostic model for breast cancer. The remaining 5 unique genes from Table 1 form a 28-gene signature prognostic model for both breast and ovarian cancer. Together, genes in Tables 4 and 5 can predict both breast cancer relapse and metastases.

TABLE 5

Genes that predicts breast cancer relapse.

	GENE	CLONE ID

	MAP2K2	769579
	SMARCD2	741067
	S100P	2060823
	FAT	591266
	DDOST	50666
	SSBP1	125183
	PDGFRA	1643186
	INPPL1	703964
	RAD50	261828
	PLSCR1	268736
	RAD52	140004
	C18B11	131988
	MCM2	239799
	MCF2L	1781388
	TXNRD1	630625

To assess a breast cancer patient's relapse and metastatic potential, risk scores can be generated by using a Cox model of the 28-gene signature, independent of clinical-pathological parameters although any standard risk evaluation could be used. In this appication large value of the risk scores indicates a high risk of relapse/metastases, while a small value indicates a lower risk of breast cancer relapse. The 28-gene signature obtained from the training set (8) was fitted into a Cox regression model as covariates. To avoid overfitting, the data set are randomly partitioned into two subsets—one was used to define risk groups by fitting the model and obtaining the risk score cutoffs; the other subset was used to validate the cutoffs for defining the risk groups. The distribution of the risk scores can be categorized into groups of two or more. If two groups, patients could be labeled as high risk at the 65^thpercentile or above and low risk at 64^thpercentile and below. Alternatively, the patients could be categorized into high, low, or intermediate risk group is 39%, 26%, and 35%, respectively in the training set. The cutoffs defined in the training subset can be used to separate the patients in the test subset into high, low and intermediate risk groups.
A further embodiment is the ability to evaluate clincopathogic variables for cancer patients. Clincopathogic variable includes, but is not limited to, average metastases-free days, ER and PR status, age, tumor size, and tumor grade. Table 6 displays the clinical characteristics of each risk group, including average relapse-free days, ER status, Her2/neu overexpression, nodal status, age, tumor size, and treatment received on the data from Sotiriou et al.(8).. Risk scores were generated for patients in Cox modeling using the gene expression profiles, without including clinicopathologic parameters. The 39^thand 65^thpercentile of the risk scores were used to partition patients into high, intermediate, and low risk groups. Same analysis is applied to the two validation sets. Table 7 summarizes the clinical characteristics of each risk group, including average metastases-free days, ER and PR status, age, tumor size, and tumor grade on the data from van't Veer et al. (13). Table 8 summarizes the clinical characteristics of each risk group, including average relapse-free days, ER status, age, and tumor grade on the data from Sorlie et al.(10).

TABLE 6

Clinical characteristics of each risk group on the patients from Sotiriou et al. (8)

			# of
			Her-		% of
	Average	% of	2\neu	% of	Positive
Risk	RFS	Age ≧	positive	Tumor	Nodal	% of	% of	% of
Group	(days)	50 yrs	cases	Size > 2 cm	Status	Chemo	Hormone	ER+

High	969	82%	6	82%	67%	38%	79%	54%
Inter.	2407	73%	1	58%	50%	35%	85%	58%
Low	2781	65%	0	47%	41%	24%	74%	85%

TABLE 7

Clinical characteristics of each risk group on the patients
from van't Veer et al. (14)

		Average			% of	% of
Risk	% of	RFS	% of	% of	Tumor	T3/T4
Group	Patients	(days)	Age ≧ 50	ER+	Grade	3	Tumors

High	28%	553	50%	69%	81%	94%
Intermediate	32%	801	84%	89%	26%	89%
Low	40%	1376	70%	73%	32%	77%

TABLE 8

Clinical characteristics of each risk group (Sorlie et al. (10))

Clinical variables such as nodal status, tumor size, tumor grade, ER status and HER2/neu overexpression in breast cancer patients affect the disease outcomes. The clinical characteristics of each risk group in the studied cohorts are analyzed including average disease-free survival days, ER and PR status, HER2/neu overexpression, nodal status, age, tumor size, grade, and treatment received. The 28-gene signature is strongly associated with the clincopathogic variables, including tumor size, tumor grade, ER and PR status, and HER2/neu overexpression (P<0.05; Table 9).

TABLE 9

Association of gene expression-defined risk groups and
clinicopathologic parameters

P-Values

		van't Veer
Risk Groups vs.	Sotiriou et al. (8)	et al. (15)	Sorlie et al. (10)

Age¹	0.243	0.458	0.095
(<50 yrs or ≧50 yrs)
Tumor size	0.006*	0.047*
(<2 cm or >2 cm)
Tumor grade	0.041*	0.004*	0.001*
(½ vs. 3)
ER status	0.011*	0.004*	0.296
PR status		0.001*
Her2/neu	0.020*

¹The percentage of patients who were at least 50 years old was 74%, 28%, and 69% in the cohorts from Sotiriou et al. (8), van't Veer et al. (16), and Sorlie et al. (10), respectively.

The 28-gene signature also predicts tumor recurrence in ovarian cancer with an accuracy of 0.89 (FIG. 10). Table 10 listed the genes that are predictive of ovarian cancer relapse.

TABLE 10

24 genes that quantifies relapse-free survival of breast cancer

		UniGene
Gene	Clone_IMAGE	Cluster ID	Sequence ID

FAM134C	198917	Hs.463079	NM_178126
TOMM70A	198312	Hs.227253	NM_014820
MCF2	268412	Hs.387262	NM_001099855
			NM_005369
RAD52 Pseudogene	1377154	Hs.552577	NM_134424
MCM2	239799	Hs.477481	NM_004526
C18B11	131988	Hs.173311	NM_152260
SEC13L	757210	Hs.301048	NM_031216
			NM_001013437
SLC25A5	291660	Hs.522767	NM_001152
PLSCR1	268736	Hs.130759	NM_021105
TXNRD1	789376	Hs.434367	NM_003330
			NM_001093771
			NM_182742
			NM_182729
			NM_182743
RAD50	261828	Hs.242635	NM_005732
			NM_133482
INPPL1	703964	Hs.523875	NM_001567
PBX2	80549	Hs.509545	NM_002586
SSBP1	125183	Hs.490394	NM_003143
PDGFRA	376499	Hs.74615	NM_006206
DDOST	50666	Hs.523145	NM_005216
IGHA1	182930	Hs.497723	AK128652
S100P	135221	Hs.2962	NM_005980
FAT	591266	Hs.481371	NM_005245
FGF2	324383	Hs.284244	NM_002006
INSM1	22895	Hs.89584	NM_002196
IRF5	260035	Hs.521181	NM_001098629
			NM_002200
			NM_001098627
			NM_001098630
			NM_001098628
			NM_032643
			NM_001098631
MAP2K2	769579	Hs.465627	NM_030662

In the present invention, target polynucleotide molecules are extracted from a sample taken from an individual afflicted with breast cancer or ovarian cancer. The sample may be collected in any clinically acceptable manner, but must be collected such that marker-derived polynucleotides (i.e., RNA) are preserved. mRNA or nucleic acids derived there from (i.e., cDNA or amplified DNA) can be labeled distinguishably from standard or control polynucleotide molecules, and both are simultaneously or independently hybridized to a detection mechanism. A detection mechanism can be any standard comparison mechanism such as a microarray or an assay of reverse transcription polymerase chain reaction (RT-PCR) comprising some or all of the markers or marker sets or subsets described above. This process identifies positive matches. Alternatively, mRNA or nucleic acids derived therefrom may be labeled with the same label as the standard or control polynucleotide molecules to identify positive matches, wherein the intensity of hybridization of each at a particular probe or primer is compared for such an identification. A sample may comprise any clinically relevant tissue sample, such as a tumor biopsy or fine needle aspiration, or a sample of bodily fluid, such as blood, plasma, serum, lymph, ascetic fluid, cystic fluid, urine, or nipple exudate. The sample may be taken from a human, or from non-human animals such as horses; mice, ruminants, swine or sheep. Patients' gene expression levels may be quantified by any means known in the art based on the marker sets defined above. Patients may be classified based on the quantitative expression profiles using any means known in the art. For example, the risk scores of a patient cohort may be generated using a Cox proportional hazard model. Patients with a risk score greater than the median is defined as high risk, whereas patients with a risk score less than the median is classified as low risk. Alternatively, a patient may be classified as high risk if this patient's gene expression profile is correlated with the high risk signature, or classified as low risk if this patient's gene expression profile is correlated with the low risk signature. A patient's prognostic categorization can also be determined by using a statistical model or a machine learning algorithm, which computes the probability of recurrence based on this patient's gene expression profiles. Cutoffs can be defined for patient stratification based on specific clinical setting. In addition, patients may be defined into three risk groups in the prognostic categorization based on the marker sets defined above.
Methods for preparing total and poly(A)+RNA are well known and are described in (17). RNA may be isolated from eukaryotic cells by procedures that involve cell lysis and denaturation of the proteins contained therein. Cells of interest include wide-type cells (i.e., no mutation), drug-treated wild-type cells, tumor- or tumor-derived cells, modified cells, normal or tumor cell lines cells, and drug-treated modified cells. Total RNA may also be extracted from samples using commercially available kits such as the RNeasy mini kit according the manufacturer's protocol (Qiagen, USA).
Additional steps may be performed to remove DNA (17). If desired, RNase inhibitors may be added to the lysis buffer. Likewise, a protein denaturation/digestion step may be added to the protocol. mRNA may be purified by means such as magnetic separation using Dynabeads (Dynal) or the Invitrogen FastTrack 2.0 kit (10).
For many applications, it is desirable to preferentially enrich mRNA with respect to other cellular RNAs, such as transfer RNA (tRNA) and ribosomal RNA (rRNA). Total RNA may also be linearly amplified using the original or modified Eberwine method (18) and be used as a reference for cDNA analysis (8).
The sample of RNA can comprise a plurality of different mRNA molecules, each different mRNA molecular having a different nucleotide sequence. In a specific embodiment, the RNA sample has not been functionally annotated.
The present invention provides a set of biomarkers for the identification of conditions of indications associated with breast cancer. Generally, the markers sets were identified by determining which of ˜25,000 human genes had expression patterns that correlated with the conditions or indications.
In one embodiment, the expression of all markers in a sample X is compared to the expression of all markers in the 28-gene signature or subsets as described above derived from tumor samples. The comparison may be accomplished by any means known in the art. The expression level may be determined by isolating and determining the level (i.e., the abundance) of nucleic acid transcribed from each marker gene. Alternatively, or additionally, the level of specific proteins translated from mRNA transcribed from a marker gene may be determined. For example, expression levels of various markers may be measured by separation of target nucleotide molecules (e.g., RNA or cDNA) derived from the markers in agarose or polyacrylamide gels, followed by hybridization with marker-specific oligonucleotide probes. Alternatively, the comparison may be accomplished by the labeling of target polynucleotide molecules followed by separation on a sequence gel. The comparison may also be accomplished by measuring the gene expression level using real-time reverse transcription polymerase chain reaction with marker-specific primers/probes. Patients may be classified based on the quantitative expression profiles using any means known in the an For example, the risk scores of a patient cohort may be generated using a Cox proportional hazard model. Patients with a risk score greater than the median is defined as high risk, whereas patients with a risk score less than the median is classified as low risk. Alternatively, a patient may be classified as high risk if this patient's gene expression profile is correlated with the high risk signature, or classified as low risk if this patient's gene expression profile is correlated with the low risk signature. A patient's prognostic categorization can also be determined by using a statistical model or a machine learning algorithm, which computes the probability of recurrence based on this patient's gene expression profiles. Cutoffs can be defined for patient stratification based on specific clinical setting. In addition, patients may be defined into three risk groups in the prognostic categorization based on the marker sets defined above.
A marker is selected based on its predictive power of breast cancer recurrence, including local recurrence and distant metastasis. A combination of Random Forests (19) and Linear Discriminant Analysis (LDA) is used to identify gene signatures for predicting breast cancer recurrence/metastases. Random forests of software R is first used to identify a small subset of genes from the original microarray data. Linear Discriminant Analysis of software SAS is used to further refine the gene signature.
Random forests are a generalization of the standard tree algorithms (20). The basic step of random forests is to form diverse tree classifiers from a single training set. Each tree is built upon a bootstrap sample from the training set. The variables used for splitting the tree nodes are a random subset of the whole variables set. The classification decision of a new case is obtained by majority voting (unless the cutoff value is user defined) over all trees. In random forests, about one-third of the cases in the bootstrap sample are not used in growing the tree. These cases are called “out-of-bag” (OOB) cases and are used to evaluate the algorithm performance. A very important function of random forests is variable importance evaluation. The importance of a variable is defined in terms of its contribution to classification accuracy. Based on the variable importance measure, backward elimination was used to identify the gene subset with the smallest OOB error rate. Here, the OOB error rate was not used to assess the prediction accuracy of the identified gene subsets. Instead, it served as a stopping rule for feature selection. The varSelRF package of software R (21) was used according to the following steps:
1. Build a forest with N trees and obtain a ranking of variable importance
2. Remove 20% of the least important variables
3. Construct a new forest with K trees
4. Repeat steps 2 and 3 until two genes are left
5. Select the gene subset with the smallest OOB error rate
In the experiments, N=3,000 and K=1,000 are chosen because the large number of trees in the initial forests are likely to produce stable importance measures (21). The “0-Standard Error (0-SE) rule” is used, which identifies the gene subset with the smallest 00B error rate. The “0-SE rule” usually selects more genes than the “1-SE rule” does. Since further gene filtering would be performed by using Linear Discriminant Analysis, the gene subsets are selected with the lowest prediction error using random forests.
Discriminant analysis is used to determine which variables discriminate two or more naturally occurring groups in prognosis. Given a number of variables as the data representation, each class is modeled as multivariate normal distribution with a covariance matrix and a mean vector. Instances are classified to the label of the nearest mean vector based on Mahalanobis distance. The decision surfaces between classes become linear if the classes have a common covariance matrix. When the distribution within each group is assumed to be multivariate normal, a parametric method can be used to develop a discriminant function. Such function is determined by a measure of generalized square distance which is based on the pooled covariance matrix as well as the prior probabilities of group membership. The generalized squared distance D_i ²(x) from input x to class i is:
D _i ²(x)=d _i ²(x)+g(i)
where d_i ²(x)=(x−m)′V⁻¹(x−m_i) is the squared distance from x to group I; m_iis the p-dimensional mean vector for group I; V is the pooled covariance matrix and g(i) depends on the prior probability of class i. In practice, the prior probability can be assumed as equal for all groups (refer to SAS Users' Manual). In this study, we assumed equal prior probability and thus g(i)=0. x is classified into class I, if D_i ²(x) is the smallest among all the distance measures. We selected the gene markers using backward selection of stepwise discriminant analysis with software SAS.
Linear Discriminant Analysis (LDA) is used to refine the gene signature obtained from random forests and assess the classification accuracy of models in predicting 5-year relapse-free survival based on the identified gene signatures. Leave-one-out cross-validation is used in the evaluation to identify the optimal marker subset (22).
Once a marker set is identified, validation of the marker set may be accomplished by a survival analysis. To evaluate the accuracy of survival prediction, time-dependent receiver operating characteristic (ROC) analysis for censored data (23;24) was performed with software R. Time-dependent ROC analysis extends the concepts of sensitivity, specificity, and ROC curves for time-dependent binary disease variables in censored data. In this embodiment, the binary disease variable R_i(t)=1, if patient i has recurrent or metastatic breast cancer prior to time t; otherwise, R_i(t)=0. For a diagnostic marker M, both sensitivity and specificity are defined as a function of time t:
sensitivity(c,t)=P{M>c|R(t)=1}
sensitivity(c,t)=P{M≦c|R(t)=0}
A ROC(t) is a function of t at different cutoffs c. A time-dependent ROC curve is a plot of sensitivity(c, t) vs. 1−specificity(c, t). The area under the ROC curve (AUC) can be used as an accuracy measure of the ROC curve. A higher prediction accuracy is evidenced by a larger AUC(t) (23;24).
The prediction of patient outcome may be accomplished with any means known in the art. For example, to estimate a patient's recurrent and metastatic potential, risk scores are generated by fitting the identified gene predictors in a Cox proportional hazard model as covariates. A higher risk score represents a higher probability of tumor recurrence. The distribution of the risk scores can be used to classify the patients into three groups: high-risk, low-risk, and intermediate-risk. Alternatively, patients may be stratified into two groups: high- or low-risk. Kaplan-Meier analysis may be used to assess the disease-free survival probability of three risk groups in the studied patient cohorts (8;10;25). Similarly, a Cox proportional hazard model may be developed to estimate a patient's overall survival probability. A higher survival risk score represents a higher risk for death from breast cancer. Alternatively, a Linear Discriminant Analysis (LDA) function may be determined by a measure of generalized square distance which is based on the pooled covariance matrix based on the marker sets described above as well as the prior probabilities of group membership for prognostic categorization.
For prognostic predictions in clinic, the expression levels of the markers can be measured with any means known in the art such as cDNA microarrays (8;10;26), various generations of Affymetrix gene chips (Affymetrix, Santa Clara, Calif.), and real-time reverse transcription polymerase chain reactions. The present invention further provides for kits comprising the marker sets above. The analytical methods described above can be implemented by use of following computer systems. For example, a computer system can be an Intel 8086-, 80386-, 80486-, or Pentium-based process with preferably 64 MB or more of main memory. The computer system can be linked to an external component, including mass storage. This mass storage can be one or more hard disks, preferably of 1 GB or more storage capacity. Other external components include regular accessories for a computer such as a monitor, a mouse, or a printer.
The software program described in above sections can be implemented with software packages R and SAS. The software to be included in the kit comprises the data analysis methods for this invention as disclosed herein. In particular, the software algorithms may include mathematical procedures for biomarker discovery, including the computation of the Mahalanobis distance between clinical categories (i.e., relapse status) and marker expression. The software may also include mathematical procedures for computing the regression coefficients between the marker expression and patient survival.
Alternative computer systems and software for implementing the analytical methods of this invention will be apparent to one of skill in the art and are intended to be comprehended within the accompanying claims.
These terms and specifications, including the examples, serve to describe the invention by example and not to limit the invention. It is expected that others will perceive differences, which, while differing from the forgoing, do not depart from the scope of the invention herein described and claimed. In particular, any of the function elements described herein may be replaced by any other known element having an equivalent function.

REFERENCE LIST

1. Gonzalez M A, Pinder S E, Callagy G, Vowler S L, Morris L S, Bird K, Bell J A, Laskey R A, Coleman N. Minichromosome maintenance protein 2 is a strong independent prognostic marker in breast cancer. J.Clin.Oncol. 2003 Dec. 1;21(23):4306-13.
2. Tommiska J, Seal S, Renwick A, Barfoot R, Baskcomb L, Jayatilake H, Bartkova J, Tallila J, Kaare M, Tamminen A, et al. Evaluation of RAD50 in familial breast cancer predisposition. Int.J.Cancer 2006 Jun. 1;118(11):2911-6.
3. Carvalho 1, Milanezi F, Martins A, Reis R M, Schmitt F. Overexpression of platelet-derived growth factor receptor alpha in breast cancer is associated with tumour progression. Breast Cancer Res. 2005;7(5):R788-R795.
4. Jechlinger M, Sommer A, Moriggl R, Seither P, Kraut N, Capodiecci P, Donovan M, Cordon-Cardo C, Beug H, Grunert S. Autocrine PDGFR signaling promotes mammary cancer metastasis. J.Clin.Invest 2006 June;116(6):1561-70.
5. Guerreiro D S, I, Hu Y F, Russo I H, Ao X, Salicioni A M, Yang X, Russo J. Sloop calcium-binding protein overexpression is associated with immortalization of human breast epithelial cells in vitro and early stages of breast cancer development in vivo. Int.J.Oncol. 2000 February;16(2):231-40.
6. Schor A P, Carvallo F M, Kemp C, Silva I D, Russo J. SLOOP calcium-binding protein expression is associated with high-risk proliferative lesions of the breast. Oncol.Rep. 2006 January;15(1):3-6.
7. Granato A M, Nanni O, Falcini F, Folli S, Mosconi G, De Paola F, Medri L, Amadori D, Volpi A. Basic fibroblast growth factor and vascular endothelial growth factor serum levels in breast cancer patients and healthy women: useful as diagnostic tools? Breast Cancer Res. 2004;6(1):R38-R45.
8. Sotiriou C, Neo S Y, McShane L M, Korn E L, Long P M, Jazaeri A, Martiat P, Fox S B, Harris A L, Liu ET. Breast cancer classification and prognosis based on gene expression profiles from a population-based study. Proc.Natl.Acad.Sci.U.S.A 2003 Sep. 2;100(18):10393-8.
9. 't Veer L J, Dai H, van de Vijver M J, He Y D, Hart A A, Mao M, Peterse 11L, van der K K, Marton M J, Witteveen A T, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002 Jan. 31;415(6871):530-6.
10. Sorlie T, Perou C M, Tibshirani R, Aas T, Geisler S, Johnsen H, Hastie T, Eisen M B, van de R M, Jeffrey S S, et al. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc.Natl.Acad.Sci.U.S.A 2001 Sep. 11;98(19): 10869-74.
11. 't Veer L J, Dai H, van de Vijver M J, He Y D, Hart A A, Mao M, Peterse H L, van der K K, Marton M J, Witteveen A T, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002 Jan. 31;415(6871):530-6.
12. 't Veer L J, Dai H, van de Vijver M J, He Y D, Hart A A, Mao M, Peterse H L, van der K K, Marton M J, Witteveen A T, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002 Jan. 31;415(6871):530-6.
13. 't Veer L J, Dai H, van de Vijver M J, He Y D, Hart A A, Mao M, Peterse H L, van der K K, Marton M J, Witteveen A T, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002 Jan. 31;415(6871):530-6.
14. 't Veer L J, Dai H, van de Vijver M J, He Y D, Hart A A, Mao M, Peterse H L, van der K K, Marton M J, Witteveen A T, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002 Jan. 31;415(6871):530-6.
15. 't Veer L J, Dai H, van de Vijver M J, He Y D, Hart A A, Mao M, Peterse H L, van der K K, Marton M J, Witteveen A T, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002 Jan. 31;415(6871):530-6.
16. 't Veer L J, Dai H, van de Vijver M T, He YD, Hart A A, Mao M, Peterse H L, van der K K, Marton M J, Witteveen A T, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002 Jan. 31;415(6871):530-6.
17. Sambrook J, Russell D W. Molecular Cloning: A Laboratory Manual. 3rd ed. Cold Spring Harbor Laboratory Press; 2001.
18. Eberwine J, Yeh H, Miyashiro K, Cao Y, Nair S, Finnell R, Zettel M, Coleman P. Analysis of Gene Expression in Single Live Neurons. PNAS 1992 Apr. 1;89(7):3010-4.
19. Breiman L Random Forests. Machine Learning 2001;45:5-32.
20. Bioinformatics and Computational Biology Solutions. 1st ed. New York: Springer; 2005.
21. Diaz-Uriarte R, Alvarez d A. Gene selection and classification of microarray data using random forest. BMC.Bioinformatics. 2006;7:3.
22. Kohavi R. A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model Selection. Proceedings.of International Joint Conference on Artificial Intelligence (IJCAI) 1995;1137-43.
23. Heagerty P J, Lumley T, Pepe M S. Time-dependent ROC curves for censored survival data and a diagnostic marker. Biometrics 2000 June;56(2):337-44.
24. Heagerty P J, Zheng Y. Survival model predictive accuracy and ROC curves. Biometrics 2005;61:92-105.
25. 't Veer L J, Dai H, van de Vijver M J, He Y D, Hart A A, Mao M, Peterse H L, van der K K, Marton M J, Witteveen A T, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002 Jan. 31;415(6871):530-6.
26. van 't Veer L J, Dai H, van de Vijver M J, He Y D, Hart A A, Mao M, Peterse H L, van der K K, Marton M J, Witteveen A T, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002 Jan. 31;415(6871):530-6.
27. 't Veer L J, Dai H, van de Vijver M J, He Y D, Hart A A, Mao M, Peterse H L, van der K K, Marton M J, Witteveen A T, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002 Jan. 31;415(6871):530-6.
28. 't Veer L J, Dai H, van de Vijver M J, He Y D, Hart A A, Mao M, Peterse H L, van der K K, Marton M J, Witteveen A T, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002 Jan. 31;415(6871):530-6.
29. Bild A H, Yao G, Chang J T, Wang Q, Potti A, Chasse D, Joshi MB, Harpole D, Lancaster J M, Berchuck A, et al. Oncogenic pathway signatures in human cancers as a guide to targeted therapies. Nature 2006 Jan. 19;439(7074):353-7.

Claims

1. A method to determine metastatic potential, relapse potential, or both in breast cancer patients comprising collecting a sample from an individual, removing marker-derived polynucleotide from said sample, using a detection mechanism to search for positive matches of said polynucleotides and the markers in Table 5, and developing a quantitative expression profile.

2. The method to determine metastatic potential, relapse potential, or both in breast cancer patients of claim 1 further comprising the addition of unique markers in Table 4 for said search of positive matches.

3. The method to determine metastatic potential, relapse potential, or both in breast cancer patients of claim 2 further comprising the addition of unique markers in Table 1 for said search of positive matches.

4. The method to determine metastatic potential, relapse potential, or both in breast cancer patients of claim 3 further comprising evaluating said quantitative expression profile using risk analysis.

5. The method to determine metastatic potential, relapse potential, or both in breast cancer patients of claim 4 wherein said risk analysis is a statistical model or machine learning algorithm.

6. The method to determine metastatic potential, relapse potential, or both in breast cancer patients of claim 4 further, comprising placing an individual in two or more categories.

7. The method to determine metastatic potential, relapse potential, or both in breast cancer patients of claim 6 wherein said categories are high risk or lower risk based on said statistical model or machine learning algorithm.

8. The method to determine metastatic potential, relapse potential, or both in breast cancer patients of claim 6 wherein said categories are high risk, intermediate risk, or lower risk based on said statistical model or machine learning algorithm.

9. The method to determine metastatic potential, relapse potential, or both in breast cancer patients of claim 4 wherein said risk analysis is a Cox proportional hazard model.

10. The method to determine metastatic potential, relapse potential, or both in breast cancer patients of claim 14 wherein said risk analysis is a Kaplan Meier analysis for disease free survival.

11. The method to determine metastatic potential, relapse potential, or both in breast cancer patients of claim 4 wherein said risk analysis is a Linear Discriminate Analysis.

12. The method to determine metastatic potential, relapse potential, or both in breast cancer patients of claim 4 further comprising assessing clincopathogic variables.

13. A method to determine relapse free potential in breast cancer patients comprising collecting a sample from an individual, removing marker-derived polynucleotide from said sample, using a detection mechanism to search for positive matches of said polynucleotides and the markers in Table 10, and developing a quantitative expression profile.

14. The method to determine relapse free potential in breast cancer patients of claim 13 further comprising evaluating said quantitative expression profile using risk analysis.

15. The method to determine relapse free potential in breast cancer patients of claim 14 wherein said risk analysis is a statistical model or machine learning algorithm.