Patent Application
20140018253
The Sequence Listing written in file SEQTXT 77429-871826-010220US.txt, created on Apr. 4, 2013, 332,697 bytes, machine format IBM-PC, MS-Windows operating system, is hereby incorporated by reference in its entirety for all purposes.
Large randomized trials have shown that chemotherapy administered in the perioperative setting (e.g., adjuvant chemotherapy) can cure patients otherwise destined to recur with systemic, incurable cancer (1). Once this recurrence has happened, the same chemotherapy is not curative. Therefore, the adjuvant window is a privileged period of time, when the decision to administer additional therapy or not, as well as the type, duration and intensity of such therapy takes center stage. Node-negative, estrogen receptor (ER)-positive, HER2-negative patients generally show a favorable prognosis when treated with adjuvant hormonal therapy only. However, because an unknown subset of these patients develops recurrences, most are currently treated not only with hormonal therapy but also cytotoxic chemotherapy, even though it is probably unnecessary for most. Our goal was to stratify these patients into those that are most or least likely to develop a recurrence within 10 years after surgery. Our approach was to develop a multi-gene transcription-level-based classifier of 10-year-relapse (disease recurrence within 10 years) using a large database of existing, publicly available microarray datasets. The probability of relapse and relapse risk score group reported by our method can be used to assign systemic chemotherapy to only those patients most likely to benefit from it.
The present invention is based, in part, on the identification of a panel of gene expression markers for node-negative, ER-positive, HER2-negative breast cancer patients. The probability of relapse and relapse risk score group using the panel of gene expression markers of the invention can be used to assign systemic chemotherapy to only those patients most likely to benefit from it.
The invention can be used on tissue from LN−, ER+, HER2−breast cancer patients by any assay where transcript levels (or their expression products) of primary genes (or their alternate genes) in the Random Forest Relapse Score (RFRS) signature are measured. These measurements can be used to assign an RFRS value and to determine the likelihood of breast cancer relapse. Those breast cancer patients with tumors at high risk of relapse can be treated more aggressively whereas those at low risk of relapse can more safely avoid the risks and side effects of systemic chemotherapy. Thus, this method can provide rapid and useful information for clinical decision making.
Thus, in one embodiment, the invention relates to a method of evaluating the likelihood of a relapse for a patient that has a lymph node-negative, estrogen receptor-positive, HER2-negative breast cancer, the method comprising: providing a sample comprising breast tumor tissue from the patient; determining the levels of expression of the 17 genes, or one or more corresponding alternates thereof, identified in Table 1; or of the 8 genes, or one or more corresponding alternates thereof, identified in Table 2; in the sample; and correlating the levels of expression with the likelihood of a relapse. In some embodiments, the method further comprises detecting the level of expression of one or more reference genes, e.g., one or more reference genes selected from the genes identified in Table 3, or one or more reference genes selected from the genes ARPC2, ATF4, ATP5B, B2M, CDH4, CELF1, CLTA, CLTC, COPB1, CTBP1, CYC1, CYFIP1, DAZAP2, DHX15, DIMT1, EEF1A1, FLOT2, GADPH, GUSB, HADHA, HDLBP, HMBS, HNRNPC, HPRT1, HSP90AB1, MTCH1, MYL12B, NACA, NDUFB8, PGK1, PPIA, PPIB, PTBP1, RPL13A, RPLPO, RPS13, RPS23, RPS3, S100A6, SDHA, SEC31A, SET, SF3B1, SFRS3, SNRNP200, STARD7, SUMO1, TBP, TFRC, TMBIM6, TPT1, TRA2B, TUBA1C, UBB, UBC, UBE2D2, UBE2D3, VAMP3, XPO1, YTHDC1, YWHAZ, and 18S rRNA. In some embodiments, the step of determining the levels of expression of the gene comprises detecting the level of expression of RNA. In some embodiments, the determining step comprises detecting the level of expression of protein. The RNA may be detected using any known methods, e.g., a method comprising a quantitative PCR reaction. In some embodiments, detecting the level of expression of the RNA comprises hybridizing a nucleic acid obtained from the sample to an array that comprises probes to the 17 genes set forth in Table 1, and/or one or more corresponding alternates thereof; or hybridizing a nucleic acid obtained from the sample to an array that comprises probes to the 8 genes set forth in Table 2, and/or one or more corresponding alternates thereof.
In a further aspect, the invention provides a kit for detecting RNA expression comprising primers and/or probes for detecting the level of expression of the 17 genes set forth in Table 1, and/or one or more corresponding alternates thereof; or for detecting the level of expression of the 8 genes set forth in Table 2, and/or one or more alternates thereof. In some embodiments, the kit further comprises primers and/or probes for detecting the level of RNA expression of one or more reference genes, e.g., one or more reference genes selected from the genes identified in Table 3, or one or more reference genes selected from the genes ARPC2, ATF4, ATP5B, B2M, CDH4, CELF1, CLTA, CLTC, COPB1, CTBP1, CYC1, CYFIP1, DAZAP2, DHX15, DIMT1, EEF1A1, FLOT2, GADPH, GUSB, HADHA, HDLBP, HMBS, HNRNPC, HPRT1, HSP90AB1, MTCH1, MYL12B, NACA, NDUFB8, PGK1, PPIA, PPIB, PTBP1, RPL13A, RPLPO, RPS13, RPS23, RPS3, S100A6, SDHA, SEC31A, SET, SF3B1, SFRS3, SNRNP200, STARD7, SUMO1, TBP, TFRC, TMBIM6, TPT1, TRA2B, TUBA1C, UBB, UBC, UBE2D2, UBE2D3, VAMP3, XPO1, YTHDC1, YWHAZ, and 18S rRNA.
In a further aspect, the invention relates to a microarray comprising probes for detecting the level of expression of the 17 genes set forth in Table 1, and/or one or more corresponding alternates thereof; or for detecting the level of expression of the 8 genes set forth in Table 2, and/or one or more alternates thereof. In some embodiments, the microarray further comprises probes for detecting the level of expression of one or more reference genes, e.g., one or more reference genes selected from the genes identified in Table 3, or one or more reference genes selected from the genes ARPC2, ATF4, ATP5B, B2M, CDH4, CELF1, CLTA, CLTC, COPB1, CTBP1, CYC1, CYFIP1, DAZAP2, DHX15, DIMT1, EEF1A1, FLOT2, GADPH, GUSB, HADHA, HDLBP, HMBS, HNRNPC, HPRT1, HSP90AB1, MTCH1, MYL12B, NACA, NDUFB8, PGK1, PPIA, PPIB, PTBP1, RPL13A, RPLPO, RPS13, RPS23, RPS3, S100A6, SDHA, SEC31A, SET, SF3B1, SFRS3, SNRNP200, STARD7, SUMO1, TBP, TFRC, TMBIM6, TPT1, TRA2B, TUBA1C, UBB, UBC, UBE2D2, UBE2D3, VAMP3, XPO1, YTHDC1, YWHAZ, and 18S rRNA.
In an additional aspect, the invention relates to a computer-implemented method for evaluating the likelihood of a relapse for a patient that has a lymph node-negative, estrogen receptor-positive, HER2-negative breast cancer, the method comprising: receiving, at one or more computer systems, information describing the level of expression of the 17 genes set forth in Table 1, or one or more corresponding alternates thereof; or information describing the level of expression of the 8 genes set forth in Table 2, or one or more corresponding alternates thereof; in a breast tumor tissue sample obtained from the patient; performing, with one or more processors associated with the computer system, a random forest analysis in which the level of expression of each gene in the analysis is assigned to a terminal leaf of each decision tree, representing a vote for either “relapse” or no “relapse”; generating, with the one or more processors associated with the one or more computer systems, a random forest relapse score (RFRS). In some embodiments in which the level of expression of the 17 genes, or at least one alternate, set forth in Table 1 is determined, if the RFRS is greater than or equal to 0.606 the patient is assigned to a high risk group, if greater than or equal to 0.333 and less than 0.606 the patient is assigned to an intermediate risk group and if less than 0.333 the patient is assigned to low risk group. In some embodiments in which the level of expression of the 8 genes, or at least one alternate, set forth in Table 2 is determined, if the RFRS is greater than or equal to 0.606 the patient is assigned to a high risk group, if greater than or equal to 0.333 and less than 0.606 the patient is assigned to an intermediate risk group and if less than 0.333 the patient is assigned to a low risk group.
In some embodiments, the computer-implemented method further comprises generating, with the one or more processors associated with the one or more computer systems, a likelihood of relapse by comparison of the RFRS score for the patient to a loess fit of RFRS versus likelihood of relapse for a training dataset.
In another aspect, the invention relates to a non-transitory computer-readable medium storing program code for evaluating the likelihood of a relapse for a patient that has a lymph node-negative, estrogen receptor-positive, HER2-negative breast cancer, the computer-readable medium comprising:
code for receiving information describing the level of expression of the 17 genes identified in Table 1, or one or more corresponding alternates thereof; or information describing the level of expression of the 8 genes identified in Table 2, or one or more corresponding alternates thereof; in a breast tumor tissue sample obtained from the patient;
code for performing a random forest analysis in which the level of expression of each gene in the analysis is assigned to a terminal leaf of each decision tree, representing a vote for either “relapse” or no “relapse”; and
code for generating a random forest relapse score (RFRS). In some embodiments in which the level of expression of the 17 genes, or one or more designated alternates, identified in Table 1 is determined, if the RFRS is greater than or equal to 0.606 the patient is assigned to a high risk group, if greater than or equal to 0.333 and less than 0.606 the patient is assigned to an intermediate risk group and if less than 0.333 the patient is assigned to low risk group. In some embodiments in which the level of expression of the 8 genes, or one or more designated alternates, identified in Table 2, is determined, if the RFRS is greater than or equal to 0.606 the patient is assigned to a high risk group, if greater than or equal to 0.333 and less than 0.606 the patient is assigned to an intermediate risk group and if less than 0.333 the patient is assigned to a low risk group. In some embodiments, the non-transitory computer-readable medium storing program code further comprises code for generating a likelihood of relapse by comparison of the RFRS score for the patient to a loess fit of RFRS versus likelihood of relapse for a training dataset.
An “estrogen receptor positive, lymph node-negative, HER2-negative” or “ER+N− HER2−” patient as used herein refers to a patient that has no discernible breast cancer in the lymph nodes; and has breast tumor cells that express estrogen receptor and do not show evidence of HER2 genomic (DNA) amplification or HER2 over-expression. LN− status is typically determined when the sentinel node is surgically removed and examined by microscopy for cytological evidence of disease. Patients are considered LN− (N0) if zero positive nodes were observed. Patients are considered LN+ if one or more lymph nodes were considered positive for disease (1-2 positive=N1; 3-6 positive=N2, etc). ER+ status is typically assessed by immunohistochemistry (IHC) where a positive determination is made when greater than a small percentage (typically greater than 3%, 5% or 10%) of cells stain positive. ER status can also be tested by quantitative PCR or biochemical assays. HER2− status is generally determined by either IHC, fluorescence in situ hybridization (FISH) or some combination of the two methods. Typically, a patient is first tested by IHC and scored on a scale from 0 to 3 where a “3+” score indicates strong complete membrane staining on >5-10% of tumor cells and is considered positive. No staining (score of “0”) or a “1+” score, indicating faint partial membrane staining in greater than 5-10% of cells, is considered negative. An intermediate score of “2+”, indicating weak to moderate complete membrane staining in more than 5-10% of cells, may prompt further testing by FISH. A typical HER2 FISH scheme would consider a patient HER2+ if the ratio of a HER2 probe to a centromeric (reference) probe is more than 4:1 in ˜5% or more of cells after examining 20 or more metaphase spreads. Otherwise the patient is considered HER2−. Quantitative PCR, array-based hybridization, and other methods may also be used to determine HER2 status. The specific methods and cutoff points for determining LN, ER and HER2 status may vary from hospital to hospital. For the purpose of this invention, a patient will be considered “ER+LN−HER2−” if reported as such by their health care provider or if determined by any accepted and approved methods, including but not limited to those detailed above.
In the current invention, a “gene set forth in” a table or a “gene identified in” a table are used interchangeably to refer to the gene that is listed in that table. For example, a gene “identified in” Table 4 refers to the gene that corresponds to the gene listed in Table 4. As understood in the art, there are naturally occurring polymorphisms for many gene sequences. Genes that are naturally occurring allelic variations for the purposes of this invention are those genes encoded by the same genetic locus. The proteins encoded by allelic variations of a gene set forth in Table 4 (or in any of Tables 1-3 or Table 4) typically have at least 95% amino acid sequence identity to one another, i.e., an allelic variant of a gene indicated in Table 4 typically encodes a protein product that has at least 95% identity, often at least 96%, at least 97%, at least 98%, or at least 99%, or greater, identity to the amino acid sequence encoded by the nucleotide sequence denoted by the Entrez Gene ID number (Apr. 1, 2012) shown in Table 4 for that gene. For example, an allelic variant of a gene encoding CCNB2 (gene: cyclin B2) typically has at least 95% identity, often at least 96%, at least 97%, at least 98%, or at least 99%, or greater, to the CCNB2 protein sequence encoded by the nucleic acid sequence available under Entrez Gene ID no. 9133). A “gene identified in” a table, such as Table 4, also refers to a gene that can be unambiguously mapped to the same genetic locus as that of a gene assigned to a genetic locus using the probes for the gene that are listed in Appendix 3. Similarly, a “gene identified in Table 1” or a “gene identified in Table 2” refers to a gene that can be unambiguously mapped to a genetic locus using the probes for that gene that are listed in Appendix 4 (panel of 17 genes from Table 1, which includes the genes for the 8 gene panel identified in Table 2); and a “gene identified in Table 3” refers to a gene that can be unambiguously mapped to a genetic locus using the probes for that gene that are listed in Appendix 5.
The terms “identical” or “100% identity,” in the context of two or more nucleic acids or proteins refer to two or more sequences or subsequences that are the same sequences. Two sequences are “substantially identical” or a certain percent identity if two sequences have a specified percentage of amino acid residues or nucleotides that are the same (i.e., 70% identity, optionally 75%, 80%, 85%, 90%, or 95% identity, over a specified region, or, when not specified, over the entire sequence), when compared and aligned for maximum correspondence over a comparison window, or designated region as measured using known sequence comparison algorithms, e.g., BLAST using the default parameters, or by manual alignment and visual inspection.
A “gene product” or “gene expression product” in the context of this invention refers to an RNA or protein encoded by the gene.
The term “evaluating a biomarker” in an LN−ER+HER2− patient refers to determining the level of expression of a gene product encoded by a gene, or allelic variant of the gene, listed in Table 4. Preferably, the gene is listed in Table 1 or Table 2 as either a primary or alternate gene. Typically, the RNA expression level is determined.
The invention is based, in part, on the identification of a panel of at least eight genes whose gene expression level correlates with breast cancer prognosis. In some embodiments, the panel of at least eight genes comprises at least eight genes, or at least 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 25, 30, 35, 40, 45, or 50, or more genes, identified in Table 4 with the proviso that the gene is one of those also listed in Table 5. In some embodiments, the panel of genes comprises at least 8 primary genes, or at least 9, 10, 11, 12, 13, 14, 15, 16, or all 17 primary genes identified in Table 1; or the 8 primary genes set forth in Table 2. Table 1 also shows alternate genes for each of the seventeen that can replace the specific primary gene in the analysis. At least one alternate gene can be evaluated in place of the corresponding primary gene listed in Table 1, or can be evaluated in addition to the corresponding primary gene listed in Table 1. Similarly, Table 2 shows alternate genes for each of the eight that can replace, or be assayed in addition to, the specific primary gene in the analysis. The results of the expression analysis are then evaluated using an algorithm to determine breast cancer patients that are likely to have a recurrence, and accordingly, are candidates for treatment with more aggressive therapy, such as chemotherapy.
The invention therefore relates to measurement of expression levels of a biomarker panel, e.g., a 17-gene expression panel, or an 8-gene expression panel, in a breast cancer patient prior to the patient undergoing chemotherapy. In some embodiments, probes to detect such transcripts may be applied in the form of a diagnostic device to predict which LN−ER+HER2− breast cancer patients have a greater risk for relapse.
Typically, the methods of the invention comprise determining the expression levels of all seventeen primary genes, and/or at least one corresponding alternate gene shown in Table 1. However, in some embodiments, the expression level of fewer genes, e.g., 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, or 16 genes, may be evaluated. In some embodiments, the methods of the invention comprise determining the expression level of all eight gene and/or at least one corresponding alternate gene shown in Table 2. Gene expression levels may be measured using any number of methods known in the art. In typical embodiments, the method involves measuring the level of RNA. RNA expression can be quantified using any method, e.g., employing a quantitative amplification method such as qPCR. In other embodiments, the methods employ array-based assays. In still other embodiments, protein products may be detected. The gene expression patterns are determined using a sample obtained from breast tumor.
In the context of this invention, an “alternate gene” refers to a gene that can be evaluated for expression levels instead of, or in addition to, the gene for which the “alternate gene” is the designated alternate in Table 1. For example, one of the genes in Table 1 is CCNB2. MELK and GINS1 are both alternatives that can be evaluated for expression instead of CCNB2 or in addition to CCNB2, when evaluating the gene expression levels of the 17 genes set forth in Table 1. With respect to Table 2, an “alternate gene” refers to a gene that can be evaluated for expression levels instead of, or in addition to, the gene for which the “alternate gene” is the designated alternate in Table 2. For example, one of the genes in Table 2 is CCNB2. MELK and TOP2A are both alternatives that can be evaluated for expression instead of CCNB2 or in addition to CCNB2 when evaluating the gene expression levels of the 8 genes set forth in Table 2.
The quantity of RNA encoded by a gene set forth in Table 1 or Table 2 and optionally, a gene set forth in Table 3 or an alternative reference gene, can be readily determined according to any method known in the art for quantifying RNA. Various methods involving amplification reactions and/or reactions in which probes are linked to a solid support and used to quantify RNA may be used. Alternatively, the RNA may be linked to a solid support and quantified using a probe to the sequence of interest.
An “RNA nucleic acid sample” analyzed in the invention is obtained from a breast tumor sample obtained from the patient. An “RNA nucleic acid sample” comprises RNA, but need not be purely RNA, e.g., DNA may also be present in the sample. Techniques for obtaining an RNA sample from tumors are well known in the art.
In some embodiments, the target RNA is first reverse transcribed and the resulting cDNA is quantified. In some embodiments, RT-PCR or other quantitative amplification techniques are used to quantify the target RNA. Amplification of cDNA using PCR is well known (see U.S. Pat. Nos. 4,683,195 and 4,683,202; PCR PROTOCOLS: A GUIDE TO METHODS AND APPLICATIONS (Innis et al., eds, 1990)). Methods of quantitative amplification are disclosed in, e.g., U.S. Pat. Nos. 6,180,349; 6,033,854; and 5,972,602, as well as in, e.g., Gibson et al., Genome Research 6:995-1001 (1996); DeGraves, et al., Biotechniques 34(1):106-10, 112-5 (2003); Deiman B, et al., Mol Biotechnol. 20(2):163-79 (2002). Alternative methods for determining the level of a mRNA of interest in a sample may involve other nucleic acid amplification methods such as ligase chain reaction (Barany (1991) Proc. Natl. Acad. Sci. USA 88:189-193), self-sustained sequence replication (Guatelli et al. (1990) Proc. Natl. Acad. Sci. USA 87:1874-1878), transcriptional amplification system (Kwoh et al. (1989) Proc. Natl. Acad. Sci. USA 86:1173-1177), Q-Beta Replicase (Lizardi et al. (1988) Bio/Technology 6:1197), rolling circle replication (U.S. Pat. No. 5,854,033) or any other nucleic acid amplification method, followed by the detection of the amplified molecules using techniques well known to those of skill in the art.
In general, quantitative amplification is based on the monitoring of the signal (e.g., fluorescence of a probe) representing copies of the template in cycles of an amplification (e.g., PCR) reaction. One method for detection of amplification products is the 5′-3′ exonuclease “hydrolysis” PCR assay (also referred to as the TaqMan™ assay) (U.S. Pat. Nos. 5,210,015 and 5,487,972; Holland et al., PNAS USA 88: 7276-7280 (1991); Lee et al., Nucleic Acids Res. 21: 3761-3766 (1993)). This assay detects the accumulation of a specific PCR product by hybridization and cleavage of a doubly labeled fluorogenic probe (the “TaqMan™” probe) during the amplification reaction. The fluorogenic probe consists of an oligonucleotide labeled with both a fluorescent reporter dye and a quencher dye. During PCR, this probe is cleaved by the 5′-exonuclease activity of DNA polymerase if, and only if, it hybridizes to the segment being amplified. Cleavage of the probe generates an increase in the fluorescence intensity of the reporter dye.
Another method of detecting amplification products that relies on the use of energy transfer is the “beacon probe” method described by Tyagi and Kramer, Nature Biotech. 14:303-309 (1996), which is also the subject of U.S. Pat. Nos. 5,119,801 and 5,312,728. This method employs oligonucleotide hybridization probes that can form hairpin structures. On one end of the hybridization probe (either the 5′ or 3′ end), there is a donor fluorophore, and on the other end, an acceptor moiety. In the case of the Tyagi and Kramer method, this acceptor moiety is a quencher, that is, the acceptor absorbs energy released by the donor, but then does not itself fluoresce. Thus, when the beacon is in the open conformation, the fluorescence of the donor fluorophore is detectable, whereas when the beacon is in hairpin (closed) conformation, the fluorescence of the donor fluorophore is quenched. When employed in PCR, the molecular beacon probe, which hybridizes to one of the strands of the PCR product, is in “open conformation,” and fluorescence is detected, while those that remain unhybridized will not fluoresce (Tyagi and Kramer, Nature Biotechnol. 14: 303-306 (1996)). As a result, the amount of fluorescence will increase as the amount of PCR product increases, and thus may be used as a measure of the progress of the PCR. Those of skill in the art will recognize that other methods of quantitative amplification are also available.
Various other techniques for performing quantitative amplification of nucleic acids are also known. For example, some methodologies employ one or more probe oligonucleotides that are structured such that a change in fluorescence is generated when the oligonucleotide(s) is hybridized to a target nucleic acid. For example, one such method involves a dual fluorophore approach that exploits fluorescence resonance energy transfer (FRET), e.g., LightCycler™ hybridization probes, where two oligo probes anneal to the amplicon. The oligonucleotides are designed to hybridize in a head-to-tail orientation with the fluorophores separated at a distance that is compatible with efficient energy transfer. Other examples of labeled oligonucleotides that are structured to emit a signal when bound to a nucleic acid or incorporated into an extension product include: Scorpions™ probes (e.g., Whitcombe et al., Nature Biotechnology 17:804-807, 1999, and U.S. Pat. No. 6,326,145), Sunrise™ (or Amplifluor™) probes (e.g., Nazarenko et al., Nuc. Acids Res. 25:2516-2521, 1997, and U.S. Pat. No. 6,117,635), and probes that form a secondary structure that results in reduced signal without a quencher and that emits increased signal when hybridized to a target (e.g., Lux Probes™).
In other embodiments, intercalating agents that produce a signal when intercalated in double stranded DNA may be used. Exemplary agents include SYBR GREEN™ and SYBR GOLD™. Since these agents are not template-specific, it is assumed that the signal is generated based on template-specific amplification. This can be confirmed by monitoring signal as a function of temperature because melting point of template sequences will generally be much higher than, for example, primer-dimers, etc.
In other embodiments, the mRNA is immobilized on a solid surface and contacted with a probe, e.g., in a dot blot or Northern format. In an alternative embodiment, the probe(s) are immobilized on a solid surface and the mRNA is contacted with the probe(s), for example, in a gene chip array. A skilled artisan can readily adapt known mRNA detection methods for use in detecting the level of mRNA encoding the biomarkers or other proteins of interest.
In some embodiments, microarrays, e.g., are employed. DNA microarrays provide one method for the simultaneous measurement of the expression levels of large numbers of genes. Each array consists of a reproducible pattern of capture probes attached to a solid support. Labeled RNA or DNA is hybridized to complementary probes on the array and then detected by laser scanning Hybridization intensities for each probe on the array are determined and converted to a quantitative value representing relative gene expression levels. See, U.S. Pat. Nos. 6,040,138, 5,800,992 and 6,020,135, 6,033,860, and 6,344,316. High-density oligonucleotide arrays are particularly useful for determining the gene expression profile for a large number of RNA's in a sample.
Techniques for the synthesis of these arrays using mechanical synthesis methods are described in, e.g., U.S. Pat. No. 5,384,261. Although a planar array surface is often employed the array may be fabricated on a surface of virtually any shape or even a multiplicity of surfaces. Arrays may be peptides or nucleic acids on beads, gels, polymeric surfaces, fibers such as fiber optics, glass or any other appropriate substrate, see U.S. Pat. Nos. 5,770,358, 5,789,162, 5,708,153, 6,040,193 and 5,800,992. Arrays may be packaged in such a manner as to allow for diagnostics or other manipulation of an all-inclusive device.
Primer and probes for use in amplifying and detecting the target sequence of interest can be selected using well-known techniques.
In some embodiments, the methods of the invention further comprise detecting level of expression of one or more reference genes that can be used as controls to determine expression levels. Such genes are typically expressed constitutively at a high level and can act as a reference for determining accurate gene expression level estimates. Examples of control genes are provided in Table 3 and the following list: ARPC2, ATF4, ATP5B, B2M, CDH4, CELF1, CLTA, CLTC, COPB1, CTBP1, CYC1, CYFIP1, DAZAP2, DHX15, DIMT1, EEF1A1, FLOT2, GADPH, GUSB, HADHA, HDLBP, HMBS, HNRNPC, HPRT1, HSP90AB1, MTCH1, MYL12B, NACA, NDUFB8, PGK1, PPIA, PPIB, PTBP1, RPL13A, RPLPO, RPS13, RPS23, RPS3, S100A6, SDHA, SEC31A, SET, SF3B1, SFRS3, SNRNP200, STARD7, SUMO1, TBP, TFRC, TMBIM6, TPT1, TRA2B, TUBA1C, UBB, UBC, UBE2D2, UBE2D3, VAMP3, XPO1, YTHDC1, YWHAZ, and 18S rRNA genes. Accordingly, a determination of RNA expression levels of the genes of interest, e.g., the gene expression levels of the panel of genes identified in Table 1, and/or an alternate; or the gene expression levels of the panel of genes identified in Table 2, and/or an alternate; may also comprise determining expression levels of one or more reference genes set forth in Table 3 or one or more reference genes selected from the genes ARPC2, ATF4, ATP5B, B2M, CDH4, CELF1, CLTA, CLTC, COPB1, CTBP1, CYC1, CYFIP1, DAZAP2, DHX15, DIMT1, EEF1A1, FLOT2, GADPH, GUSB, HADHA, HDLBP, HMBS, HNRNPC, HPRT1, HSP90AB1, MTCH1, MYL12B, NACA, NDUFB8, PGK1, PPIA, PPIB, PTBP1, RPL13A, RPLPO, RPS13, RPS23, RPS3, S100A6, SDHA, SEC31A, SET, SF3B1, SFRS3, SNRNP200, STARD7, SUMO1, TBP, TFRC, TMBIM6, TPT1, TRA2B, TUBA1C, UBB, UBC, UBE2D2, UBE2D3, VAMP3, XPO1, YTHDC1, YWHAZ, and 18S rRNA.
In the context of this invention, “determining the levels of expression” of an RNA of interest encompasses any method known in the art for quantifying an RNA of interest.
In some embodiments, e.g., where the expression level of a protein encoded by a biomarker gene set forth in Table 1 or Table 2 is measured. Often, such measurements may be performed using immunoassays. Protein expression level is determined using a breast tumor sample obtained from the patient.
A general overview of the applicable technology can be found in Harlow & Lane, Antibodies: A Laboratory Manual (1988) and Harlow & Lane, Using Antibodies (1999). Methods of producing polyclonal and monoclonal antibodies that react specifically with an allelic variant are known to those of skill in the art (see, e.g., Coligan, Current Protocols in Immunology (1991); Harlow & Lane, supra; Goding, Monoclonal Antibodies: Principles and Practice (2d ed. 1986); and Kohler & Milstein, Nature 256:495-497 (1975)). Such techniques include antibody preparation by selection of antibodies from libraries of recombinant antibodies in phage or similar vectors, as well as preparation of polyclonal and monoclonal antibodies by immunizing rabbits or mice (see, e.g., Huse et al., Science 246:1275-1281 (1989); Ward et al., Nature 341:544-546 (1989)).
Polymorphic alleles can be detected by a variety of immunoassay methods. For a review of immunological and immunoassay procedures, see Basic and Clinical Immunology (Stites & Terr eds., 7th ed. 1991). Moreover, the immunoassays can be performed in any of several configurations, which are reviewed extensively in Enzyme Immunoassay (Maggio, ed., 1980); and Harlow & Lane, supra. For a review of the general immunoassays, see also Methods in Cell Biology: Antibodies in Cell Biology, volume 37 (Asai, ed. 1993); Basic and Clinical Immunology (Stites & Ten, eds., 7th ed. 1991).
Commonly used assays include noncompetitive assays, e.g., sandwich assays, and competitive assays. Typically, an assay such as an ELISA assay can be used. The amount of the polypeptide variant can be determined by performing quantitative analyses.
Other detection techniques, e.g., MALDI, may be used to directly detect the presence of proteins correlated with treatment outcomes.
As indicated above, evaluation of protein expression levels may additionally include determining the levels of protein expression of control genes, e.g., of one or more genes identified in Table 3.
In a further aspect, the invention provides diagnostic devices and kits for identifying gene expression products of a panel of genes that is associated with prognosis for a LN− ER+HER2− breast cancer patient.
In some embodiments, a diagnostic device comprises probes to detect at least 8, 9, 10, 11, 12, 13, 14, 15, 16, or all 17 gene expression products set forth in Table 1, and/or alternates. In some embodiments, a diagnostic device comprises probes to detect the expression products of the 8 genes set forth in Table 2, and/or alternates. In some embodiments, the present invention provides oligonucleotide probes attached to a solid support, such as an array slide or chip, e.g., as described in DNA Microarrays: A Molecular Cloning Manual, 2003, Eds. Bowtell and Sambrook, Cold Spring Harbor Laboratory Press. Construction of such devices are well known in the art, for example as described in US Patents and Patent Publications U.S. Pat. No. 5,837,832; PCT application WO95/11995; U.S. Pat. No. 5,807,522; U.S. Pat. Nos. 7,157,229, 7,083,975, 6,444,175, 6,375,903, 6,315,958, 6,295,153, and 5,143,854, 2007/0037274, 2007/0140906, 2004/0126757, 2004/0110212, 2004/0110211, 2003/0143550, 2003/0003032, and 2002/0041420. Nucleic acid arrays are also reviewed in the following references: Biotechnol Annu Rev 8:85-101 (2002); Sosnowski et al, Psychiatr Genet 12(4):181-92 (December 2002); Heller, Annu Rev Biomed Eng 4: 129-53 (2002); Kolchinsky et al, Hum. Mutat 19(4):343-60 (April 2002); and McGail et al, Adv Biochem Eng Biotechnol 77:21-42 (2002).
An array can be composed of a large number of unique, single-stranded polynucleotides, usually either synthetic antisense polynucleotides or fragments of cDNAs, fixed to a solid support. Typical polynucleotides are preferably about 6-60 nucleotides in length, more preferably about 15-30 nucleotides in length, and most preferably about 18-25 nucleotides in length. For certain types of arrays or other detection kits/systems, it may be preferable to use oligonucleotides that are only about 7-20 nucleotides in length. In other types of arrays, such as arrays used in conjunction with chemiluminescent detection technology, preferred probe lengths can be, for example, about 15-80 nucleotides in length, preferably about 50-70 nucleotides in length, more preferably about 55-65 nucleotides in length, and most preferably about 60 nucleotides in length.
A person skilled in the art will recognize that, based on the known sequence information, detection reagents can be developed and used to assay any gene expression product set forth in Table 1 or Table 2 (or in some embodiments Table 3 or another reference gene described herein) and that such detection reagents can be incorporated into a kit. The term “kit” as used herein in the context of detection reagents, are intended to refer to such things as combinations of multiple gene expression detection reagents, or one or more gene expression detection reagents in combination with one or more other types of elements or components (e.g., other types of biochemical reagents, containers, packages such as packaging intended for commercial sale, substrates to which gene expression detection reagents are attached, electronic hardware components, etc.). Accordingly, the present invention further provides gene expression detection kits and systems, including but not limited to, packaged probe and primer sets (e.g., TaqMan probe/primer sets), arrays/microarrays of nucleic acid molecules where the arrays/microarrays comprise probes to detect the level of RNA transcript, and beads that contain one or more probes, primers, or other detection reagents for detecting one or more RNA transcripts encoded by a gene in a gene expression panel of the present invention. The kits can optionally include various electronic hardware components; for example, arrays (“DNA chips”) and microfluidic systems (“lab-on-a-chip” systems) provided by various manufacturers typically comprise hardware components. Other kits (e.g., probe/primer sets) may not include electronic hardware components, but may be comprised of, for example, one or more biomarker detection reagents (along with, optionally, other biochemical reagents) packaged in one or more containers.
In some embodiments, a detection kit typically contains one or more detection reagents and other components (e.g. a buffer, enzymes such as DNA polymerases) necessary to carry out an assay or reaction, such as amplification for detecting the level of transcript. A kit may further contain means for determining the amount of a target nucleic acid, and means for comparing the amount with a standard, and can comprise instructions for using the kit to detect the nucleic acid molecule of interest. In one embodiment of the present invention, kits are provided which contain the necessary reagents to carry out one or more assays to detect one or more RNA transcripts of a gene disclosed herein. In one embodiment of the present invention, biomarker detection kits/systems are in the form of nucleic acid arrays, or compartmentalized kits, including microfluidic/lab-on-a-chip systems.
Detection kits/systems for detecting expression of a panel of genes in accordance with the invention may contain, for example, one or more probes, or pairs or sets of probes, that hybridize to a nucleic acid molecule encoded by a gene set forth in Table 1 or Table 2. In some embodiments, the presence of more than one biomarker can be simultaneously evaluated in an assay. For example, in some embodiments probes or probe sets to different biomarkers are immobilized as arrays or on beads. For example, the same substrate can comprise probes for detecting expression of at least 8, 9, 10, 11, 12, 13, 14, 15, 16, or 17 or more of the genes set forth in Table 1, and/or alternates to the genes. In some embodiments, the same substrate can comprise probes for detecting expression of 8 or more genes set forth in Table 2, and/or alternates to the genes.
Using such arrays or other kits/systems, the present invention provides methods of identifying the levels of expression of a gene described herein in a test sample. Such methods typically involve incubating a test sample of nucleic acids obtained from a breast tumor from a LN−ER+HER2−patient with an array comprising one or more probes that selectively hybridizes to a nucleic acid encoded by a gene identified in Table 1 or Table 2. Such an array may additionally comprise probes to one or more reference genes identified in Table 3, or one or more reference genes selected from the genes ARPC2, ATF4, ATP5B, B2M, CDH4, CELF1, CLTA, CLTC, COPB1, CTBP1, CYC1, CYFIP1, DAZAP2, DHX15, DIMT1, EEF1A1, FLOT2, GADPH, GUSB, HADHA, HDLBP, HMBS, HNRNPC, HPRT1, HSP90AB1, MTCH1, MYL12B, NACA, NDUFB8, PGK1, PPIA, PPIB, PTBP1, RPL13A, RPLPO, RPS13, RPS23, RPS3, S100A6, SDHA, SEC31A, SET, SF3B1, SFRS3, SNRNP200, STARD7, SUMO1, TBP, TFRC, TMBIM6, TPT1, TRA2B, TUBA1C, UBB, UBC, UBE2D2, UBE2D3, VAMP3, XPO1, YTHDC1, YWHAZ, and 18S rRNA. In some embodiments, the array comprises probes to all 17 genes identified in Table 1, and/or alternates; or all 8 genes identified in Table 2, and/or alternates. Conditions for incubating a gene detection reagent (or a kit/system that employs one or more such biomarker detection reagents) with a test sample vary. Incubation conditions depend on such factors as the format employed in the assay, the detection methods employed, and the type and nature of the detection reagents used in the assay. One skilled in the art will recognize that any one of the commonly available hybridization, amplification and array assay formats can readily be adapted to detect a gene set forth in Table 1 or Table 2.
A gene expression detection kit of the present invention may include components that are used to prepare nucleic acids from a test sample for the subsequent amplification and/or detection of a gene transcript.
In some embodiments, a gene expression kit comprises one or more reagents, e.g., antibodies, for detecting protein products of a gene identified in Table 1 or Table 2 and optionally Table 3.
Correlating Gene Expression Levels with Prognostic Outcomes
The present invention provides methods of determining the levels of a gene expression product to evaluate the likelihood that a LN-ER+HER2−breast cancer patient will have a relapse. Accordingly, the method provides a way of identifying LN−ER+HER2− breast cancer patients that are candidates for additional treatment, e.g., chemotherapy.
In step 1020, information is received describing one or more levels of expression of one or more predetermined genes in a sample obtained from a subject. For example, the level of a gene expression product associated with a prognostic outcome for a LN−ER+HER2− breast cancer patient may be recorded. In one embodiment, input data includes a text file (e.g., a tab-delimited text file) of normalized expression values for 17 transcripts from primary genes (or an indicated alternative) from Table 1. In one embodiment, input data includes a text file (e.g., a tab-delimited text file) of normalized expression values for 8 transcripts from the primary genes (or an indicated alternative) from Table 2. For example, the text file may have the gene expression values for the 17 transcripts/genes as columns and patient(s) as rows. An illustrative patient data file (patient_data.txt) is presented in Appendix 1.
In step 1030, a random forest analysis is performed on the information describing the one or more levels of expression of the one or more predetermined genes in the sample obtained from the subject. A Random Forest (RF) algorithm is used to determine a Relapse Score (RS) when applied to independent patient data. A sample R program for running the RF algorithm is presented in Appendix 2. A Random Forest Relapse Score (RFRS) algorithm as used herein typically consists of a predetermined number of decision trees suitably adapted to ensure at least a fully deterministic model. Each node (branch) in each tree represents a binary decision based on transcript levels for transcripts described herein. Based on these decisions, the subject is assigned to a terminal leaf of each decision tree, representing a vote for either “relapse” or no “relapse”. The fraction of votes for “relapse” to votes for “no relapse” represents the RFRS—a measure of the probability of relapse. In some embodiments, if a subject's RFRS is greater than or equal to 0.606, the subject is assigned to one or more “high risk” groups. If an RFRS is greater than or equal to 0.333 and less than 0.606, the subject is assigned to one or more “intermediate risk” group. If an RFRS is less than 0.333, the subject is assigned to one or more “low risk” groups. In further embodiments, a subject's RFRS value is also used to determine a likelihood of relapse by comparison to a loess fit of RFRS versus likelihood of relapse for a training dataset. A subject's estimated likelihood of relapse is determined, added to a summary plot, and output as a new report.
In step 1040, information indicative of either “relapse” or no “relapse” is generated based on the random forest analysis. In some embodiments, information indicative of either “relapse” or no “relapse” is generated to include one or more summary statistics. For example, information indicative of either “relapse” or no “relapse” may be representative of how assignments to a terminal leaf of each decision tree, representing a vote for either “relapse” or no “relapse”, are made. In further embodiment, information indicative of either “relapse” or no “relapse” is generated for the fraction of votes for “relapse” to votes for “no relapse” as discussed above to represent the RFRS.
In step 1050, information indicative of one or more additional therapies is generated based on indicative of “relapse”. For example, if an RFRS is greater than or equal to 0.606, the subject is assigned to a “high risk” group from which the one or more additional therapies may be selected. If an RFRS score is greater than or equal to 0.333 and less than 0.606, the subject is assigned to an “intermediate risk” group from which all or none of the one or more additional therapies may be selected. If an RFRS is less than 0.333, the subject is assigned to a “low risk” group. In various embodiments, a subject's RFRS value is also used to determine a likelihood of relapse by comparison to a loess fit of RFRS versus likelihood of relapse for a training dataset described in
In step 1120, training data is received. For example, training data was generated as discussed below in the Examples section. In step 1130, variables on which to base decisions at tree nodes and classifier data are received. In one embodiment, classification was performed on training samples with either a relapse or no relapse after 10yr follow-up. In one example, a binary classification (e.g., relapse versus no relapse) is specified. However, additional classifier data may be included, such as a probability (proportion of “votes”) for relapse which is termed the Random Forests Relapse Score (RFRS). Risk group thresholds can be determined from the distribution of relapse probabilities using mixed model clustering to set cutoffs for low, intermediate and high risk groups.
In step 1140, a random forest model is generated. For example, a random forest model may be generated with at least 100,001 trees (i.e., using an odd number to ensure a substantially fully deterministic model).
The invention thus includes a computer system to implement the algorithm. Such a computer system can comprise code for interpreting the results of an expression analysis evaluating the level of expression of the 17 genes, or a designated alternate gene) identified in Table 1; or code for interpreting the results of an expression analysis evaluating the level of expression of the 8 genes, or a designated alternate gene, identified in Table 2. Thus in an exemplary embodiment, the expression analysis results are provided to a computer where a central processor executes a computer program for determining the propensity for relapse for a LN−ER+HER2− breast cancer patient.
The invention also provides the use of a computer system, such as that described above, which comprises: (1) a computer; (2) a stored bit pattern encoding the expression results obtained by the methods of the invention, which may be stored in the computer; and, optionally, (3) a program for determining the likelihood of relapse.
The invention further provides methods of generating a report based on the detection of gene expression products for a LN−ER+HER2− breast cancer patient. Such a report is based on the detection of gene expression products encoded by the 17 genes, or one of the designated alternates, set forth in Table 1; or detection of gene expression products encoded by the 8 genes, or one of the designated alternates, set forth in Table 2.
Computer system 1200 can include hardware and/or software elements configured for performing logic operations and calculations, input/output operations, machine communications, or the like. Computer system 1200 may include familiar computer components, such as one or more one or more data processors or central processing units (CPUs) 1205, one or more graphics processors or graphical processing units (GPUs) 1210, memory subsystem 1215, storage subsystem 1220, one or more input/output (I/O) interfaces 1225, communications interface 1230, or the like. Computer system 1200 can include system bus 1235 interconnecting the above components and providing functionality, such connectivity and inter-device communication. Computer system 1200 may be embodied as a computing device, such as a personal computer (PC), a workstation, a mini-computer, a mainframe, a cluster or farm of computing devices, a laptop, a notebook, a netbook, a PDA, a smartphone, a consumer electronic device, a gaming console, or the like.
The one or more data processors or central processing units (CPUs) 1205 can include hardware and/or software elements configured for executing logic or program code or for providing application-specific functionality. Some examples of CPU(s) 1205 can include one or more microprocessors (e.g., single core and multi-core) or micro-controllers. CPUs 1205 may include 4-bit, 8-bit, 12-bit, 16-bit, 32-bit, 64-bit, or the like architectures with similar or divergent internal and external instruction and data designs. CPUs 1205 may further include a single core or multiple cores. Commercially available processors may include those provided by Intel of Santa Clara, Calif. (e.g., x86, x86—64, PENTIUM, CELERON, CORE, CORE 2, CORE ix, ITANIUM, XEON, etc.) or by Advanced Micro Devices of Sunnyvale, Calif. (e.g., x86, AMC—64, ATHLON, DURON, TURION, ATHLON XP/64, OPTERON, PHENOM, etc). Commercially available processors may further include those conforming to the Advanced RISC Machine (ARM) architecture (e.g., ARMv7-9), POWER and POWERPC architecture, CELL architecture, and or the like. CPU(s) 1205 may also include one or more field-gate programmable arrays (FPGAs), application-specific integrated circuits (ASICs), or other microcontrollers. The one or more data processors or central processing units (CPUs) 1205 may include any number of registers, logic units, arithmetic units, caches, memory interfaces, or the like. The one or more data processors or central processing units (CPUs) 1205 may further be integrated, irremovably or moveably, into one or more motherboards or daughter boards.
The one or more graphics processor or graphical processing units (GPUs) 1210 can include hardware and/or software elements configured for executing logic or program code associated with graphics or for providing graphics-specific functionality. GPUs 1210 may include any conventional graphics processing unit, such as those provided by conventional video cards. Some examples of GPUs are commercially available from NVIDIA, ATI, and other vendors. In various embodiments, GPUs 1210 may include one or more vector or parallel processing units. These GPUs may be user programmable, and include hardware elements for encoding/decoding specific types of data (e.g., video data) or for accelerating 2D or 3D drawing operations, texturing operations, shading operations, or the like. The one or more graphics processors or graphical processing units (GPUs) 1210 may include any number of registers, logic units, arithmetic units, caches, memory interfaces, or the like. The one or more data processors or central processing units (CPUs) 1205 may further be integrated, irremovably or moveably, into one or more motherboards or daughter boards that include dedicated video memories, frame buffers, or the like.
Memory subsystem 1215 can include hardware and/or software elements configured for storing information. Memory subsystem 1215 may store information using machine-readable articles, information storage devices, or computer-readable storage media. Some examples of these articles used by memory subsystem 1270 can include random access memories (RAM), read-only-memories (ROMS), volatile memories, non-volatile memories, and other semiconductor memories. In various embodiments, memory subsystem 1215 can include data and program code 1240.
Storage subsystem 1220 can include hardware and/or software elements configured for storing information. Storage subsystem 1220 may store information using machine-readable articles, information storage devices, or computer-readable storage media. Storage subsystem 1220 may store information using storage media 1245. Some examples of storage media 1245 used by storage subsystem 1220 can include floppy disks, hard disks, optical storage media such as CD-ROMS, DVDs and bar codes, removable storage devices, networked storage devices, or the like. In some embodiments, all or part of breast cancer prognosis data and program code 1240 may be stored using storage subsystem 1220.
In various embodiments, computer system 1200 may include one or more hypervisors or operating systems, such as WINDOWS, WINDOWS NT, WINDOWS XP, VISTA, WINDOWS 7 or the like from Microsoft of Redmond, Wash., Mac OS or Mac OS X from Apple Inc. of Cupertino, Calif., SOLARIS from Sun Microsystems, LINUX, UNIX, and other UNIX-based or UNIX-like operating systems. Computer system 1200 may also include one or more applications configured to execute, perform, or otherwise implement techniques disclosed herein. These applications may be embodied as breast cancer prognosis data and program code 1240. Additionally, computer programs, executable computer code, human-readable source code, shader code, rendering engines, or the like, and data, such as image files, models including geometrical descriptions of objects, ordered geometric descriptions of objects, procedural descriptions of models, scene descriptor files, or the like, may be stored in memory subsystem 1215 and/or storage subsystem 1220.
The one or more input/output (I/O) interfaces 1225 can include hardware and/or software elements configured for performing I/O operations. One or more input devices 1250 and/or one or more output devices 1255 may be communicatively coupled to the one or more I/O interfaces 1225.
The one or more input devices 1250 can include hardware and/or software elements configured for receiving information from one or more sources for computer system 1200. Some examples of the one or more input devices 1250 may include a computer mouse, a trackball, a track pad, a joystick, a wireless remote, a drawing tablet, a voice command system, an eye tracking system, external storage systems, a monitor appropriately configured as a touch screen, a communications interface appropriately configured as a transceiver, or the like. In various embodiments, the one or more input devices 1250 may allow a user of computer system 1200 to interact with one or more non-graphical or graphical user interfaces to enter a comment, select objects, icons, text, user interface widgets, or other user interface elements that appear on a monitor/display device via a command, a click of a button, or the like.
The one or more output devices 1255 can include hardware and/or software elements configured for outputting information to one or more destinations for computer system 1200. Some examples of the one or more output devices 1255 can include a printer, a fax, a feedback device for a mouse or joystick, external storage systems, a monitor or other display device, a communications interface appropriately configured as a transceiver, or the like. The one or more output devices 1255 may allow a user of computer system 1200 to view objects, icons, text, user interface widgets, or other user interface elements.
A display device or monitor may be used with computer system 1200 and can include hardware and/or software elements configured for displaying information. Some examples include familiar display devices, such as a television monitor, a cathode ray tube (CRT), a liquid crystal display (LCD), or the like.
Communications interface 1230 can include hardware and/or software elements configured for performing communications operations, including sending and receiving data. Some examples of communications interface 1230 may include a network communications interface, an external bus interface, an Ethernet card, a modem (telephone, satellite, cable, ISDN), (asynchronous) digital subscriber line (DSL) unit, FireWire interface, USB interface, or the like. For example, communications interface 1230 may be coupled to communications network/external bus 1280, such as a computer network, to a FireWire bus, a USB hub, or the like. In other embodiments, communications interface 1230 may be physically integrated as hardware on a motherboard or daughter board of computer system 1200, may be implemented as a software program, or the like, or may be implemented as a combination thereof.
In various embodiments, computer system 1200 may include software that enables communications over a network, such as a local area network or the Internet, using one or more communications protocols, such as the HTTP, TCP/IP, RTP/RTSP protocols, or the like. In some embodiments, other communications software and/or transfer protocols may also be used, for example IPX, UDP or the like, for communicating with hosts over the network or with a device directly connected to computer system 1200.
As suggested,
Many hardware and/or software configurations of a computer system may be apparent to the skilled artisan, which are suitable for use in implementing a RFRS algorithm as described herein. For example, a computer system or data processing device may include desktop, portable, rack-mounted, or tablet configurations. Additionally, a computer system or information processing device may include a series of networked computers or clusters/grids of parallel processing devices. In still other embodiments, a computer system or information processing device may use techniques described above as implemented upon a chip or an auxiliary processing board.
Various embodiments of an algorithm as described herein can be implemented in the form of logic in software, firmware, hardware, or a combination thereof. The logic may be stored in or on a machine-accessible memory, a machine-readable article, a tangible computer-readable medium, a computer-readable storage medium, or other computer/machine-readable media as a set of instructions adapted to direct a central processing unit (CPU or processor) of a logic machine to perform a set of steps that may be disclosed in various embodiments of an invention presented within this disclosure. The logic may form part of a software program or computer program product as code modules become operational with a processor of a computer system or an information-processing device when executed to perform a method or process in various embodiments of an invention presented within this disclosure. Based on this disclosure and the teachings provided herein, a person of ordinary skill in the art will appreciate other ways, variations, modifications, alternatives, and/or methods for implementing in software, firmware, hardware, or combinations thereof any of the disclosed operations or functionalities of various embodiments of one or more of the presented inventions.
The experiments outlined in the initial examples that identified markers for prognosis stratified node-negative, ER-positive, HER2-negative breast cancer patients into those that are most or least likely to develop a recurrence within 10 years after surgery. A multi-gene transcription-level-based classifier of 10-year-relapse (disease recurrence within 10 years) was developed using a large database of existing, publicly available microarray datasets. The probability of relapse and relapse risk score group using the panel of gene expression markers of the invention can be used to assign systemic chemotherapy to only those patients most likely to benefit from it.
Methods:
Literature Search and Curation:
Studies were collected which provided gene expression data for ER+, LN−, HER2− patients with no systemic chemotherapy (hormonal-therapy allowed). Each study was required to have a sample size of at least 100, report LN status, and include time and events for either recurrence free survival (RFS) or distant metastasis free survival (DMFS). The latter were grouped together for survival analysis where all events represent either a local or distant relapse. If ER or HER2 status was not reported, it was determined by array, but preference was given to studies with clinical determination first. A minimum of 10 years follow up was required for training the classifier. However, patients with shorter follow-up were included in survival analyses. Patients with immediately postoperative events (time=0) were excluded. Nine studies1-9 meeting the above criteria were identified by searching Pubmed and the Gene Expression Omnibus (GEO) database10. To allow combination of the largest number of samples, only the common Affymetrix U133A gene expression platform was used. 2175 breast cancer samples were identified. After filtering for only those samples which were ER+, node-negative, and had not received systemic chemotherapy, 1403 samples remained. Duplicate analysis removed a further 405 samples due to the significant amount of redundancy between studies (
Pre-Processing:
All data processing and analyses were completed with open source R/Bioconductor packages. Raw data (Cel files) were downloaded from GEO. Duplicate samples were identified and removed if they had the same database identifier (e.g., GSM accession), same sample/patient id, or showed a high correlation (r>0.99) compared to any other sample in the dataset. Raw data were normalized and summarized using, the ‘affy’ and ‘gcrma’ libraries. Probes were mapped to Entrez gene symbols using both standard and custom annotation files11. ER and HER2 expression status was determined using standard probes. For the Affymetrix U133A array we and others have found the probe “205225_at” to be most effective for determining ER status12. Similarly a rank sum of the best probes for ERBB2 (216835_s_at), GRB7 (210761_s_at), STARD3 (202991_at) and PGAP3 (55616_at) was used to determine HER2 amplicon status. Cutoff values for ER and HER2 status were chosen by mixed model clustering (‘mclust’ library). Unsupervised clustering was performed to assess the extent of batch effects. Once all pre-filtering was complete, data were randomly split into training (⅔) and test (⅓) data sets while balancing for study of origin and number of relapses with 10 year follow-up. The test data set was put aside, left untouched, and only used for final validation, once each for the full-gene, 17-gene and 8-gene classifiers. Probes sets were then filtered for a minimum of 20% samples with expression above background threshold (raw value>100) and coefficient of variation between 0.7 and 10. A total of 3048 probesets/genes passed this filtering and formed the basis for the ‘full-gene set’ model described below.
Classification:
Classification was performed on only training samples with either a relapse or no relapse after 10yr follow-up using the ‘randomForest’ library. Forests were created with at least 100,001 trees (odd number ensures fully deterministic model) and otherwise default settings. Performance was assessed by area under the curve (AUC) of a receiver operating characteristic (ROC) curve, calculated with the ‘ROCR’ package, from Random Forests internal out-of-bag (00B) testing results. By default, RF performs a binary classification (e.g., relapse versus no relapse). However it also reports a probability (proportion of “votes”) for relapse which we term Random Forests Relapse Score (RFRS). Risk group thresholds were determined from the distribution of relapse probabilities using mixed model clustering to set cutoffs for low, intermediate and high risk groups (
Determination of Optimal 17-Gene and 8-Gene Sets:
Initially an optimal set of 20 genes was selected by removing redundant probe sets and extracting the top 100 genes (by reported Gini variable importance), k-means clustering (k=20) these genes and selecting the best gene from each cluster (again by variable importance). Additional genes in each cluster serve as robust alternates in case of failure to migrate primary genes to an assay platform. A gene might fail to migrate due to problems with prober/primer design or differences in the sensitivity of a specific assay for that gene. The top 100 genes/probesets were also manually checked for sequence correctness by alignment to the reference genome. Seven genes/probesets with ambiguous or erroneous alignments were marked for exclusion. Three genes/probesets were also excluded because of their status as hypothetical proteins (KIAA0101, KIAA0776, KIAA1467). After these removals, a set of 17 primary genes and 73 alternate genes remained. All but two primary genes have two or more alternates (TXNIP is without alternate, and APOC 1 has a single alternate). Table 1 lists the final gene set, their top two alternate genes (where available) and their variable importance values (See Table 4 for complete list). The above procedure was repeated to produce an optimal set of 8 genes, this time starting from the top 90 non-redundant probe-sets (excluding the 10 genes with problems identified above), k-means clustering (k=8) these genes and selecting the best gene from each cluster. All 8 genes were also included in the 17-gene set and have at least two alternates (Table 2, Table 5). Using the final optimized 17-gene and 8-gene sets as input, new RF models were built on training data.
Validation (testing and survival analysis): Survival analysis on all training data, now also including those patients with less than 10 years of follow-up, was performed with risk group as a factor, for the full-gene, 17-gene, and 8-gene models, using the ‘survival’ package. Note, the risk scores and groups for samples used in training were assigned from internal 00B cross-validation. Only those patients not used in initial training (without 10 year follow-up) were assigned a risk score and group by de novo classification. Significance between risk groups was determined by Kaplan-Meier logrank test (with test for linear trend). However, to directly compare relapse rates per risk group to that reported by Paik et al., N Engl J Med 351: 2817-2826 (2004)13, the overall relapse rates in our patient cohort were randomly down-sampled to the same rate (15%) as in their cohort13 and results averaged over 1000 iterations. To illustrate, the training data set includes 572 samples with 143 relapse events (I.e., 25.0% relapse rate). Samples with relapse events were randomly eliminated from the cohort until only 15% of remaining samples had relapse events (76/505=15%). This “down-sampled” dataset was then classified using the RFRS model to assign each sample to a risk group and the rates of relapse determined for each group. The entire down-sampling procedure was then repeated 1000 times to obtain average estimated rates of relapse for each risk group given the overall rate of relapse of 15%. Setting the overall relapse rate to 15% is also useful because this more closely mirrors the general population rate of relapse. Without this down-sampling, expected relapse rates in each risk group would appear unrealistically high. See
Next, the full-gene, 17-gene and 8-gene RF models along with risk group cutoffs were applied to the independent test data. The same performance metrics, survival analysis and estimates of 10 year relapse rates were performed as above. The 17-gene model was also tested on the independent test data, stratified by treatment (untreated vs hormone therapy treated), to evaluate whether performance of the signature was biased towards one patient subpopulation or the other. These independent test data were not used in any way during the training phase. However, these samples represent a random subset of the same patient populations that were used in training Therefore, they are not as fully independent as recommended by the Institute of Medicine (IOM) ‘committee on the review of omics-based tests for predicting patient outcomes in clinical trials’18. Therefore, an additional independent validation was performed against the NKI dataset19 obtained from the http address bioinformatics.nki.nl/data.php. These data represent a set of 295 consecutive patients with primary stage I or II breast carcinomas. The dataset was filtered down to the 89 patients who were node-negative, ER-positive, HER2-negative and not treated by systemic chemotherapy19. Relapse times and events were defined by any of distant metastasis, regional recurrence or local recurrence. Expression values from the NKI Agilent array data were re-scaled to the same distribution as that used in training using the ‘preprocessCore’ package. Values for the 8-gene and 17-gene-set RFRS models were extracted for further analysis. If more than one Agilent probe set could be mapped to an RFRS gene then the probe set with greatest variance was used. The full-gene-set model was not applied to NKI data because only 2530/3048 Affymetrix-defined genes (probe sets) in the full-gene-set could be mapped to Agilent genes (probe sets) in the NKI dataset. However, the 17-gene and 8-gene RFRS models were applied to NKI data to calculate predicted probabilities of relapse. Patients were divided into low, intermediate, and high risk groups by ranking according to probability of relapse and then dividing so that the proportions in each risk group were identical to that observed in training ROC AUC, survival p-values and estimated rates of relapse were then calculated as above. It should be noted that while the NKI clinical data described here (N=89) had an average follow-up time of 9.55 years (excluding relapse events), 34 patients had a follow-up time less than 10 years (range 1.78-9.83 years). These patients would not have met our criteria for inclusion in the training dataset and likely represent some events which have not occurred yet. If anything, this is likely to reduce the AUC estimate and underestimate p-value significance in survival analysis.
Selection of Control Genes:
While not necessary for Affymetrix, migration to other assay technologies (e.g., RT-PCR approaches) may employ highly expressed and invariant genes to act as a reference for determining accurate gene expression level estimates. To this end, we developed two sets of reference genes. The first was chosen by the following criteria: (1) filtered if not expressed above background threshold (raw value>100) in 99% of samples; (2) filtered if not in top 5th percentile (overall) for mean expression; (3) Filtered if not in top 10th percentile (remaining genes) for standard deviation; (4) ranked by coefficient of variation. The top 30 control genes from set #1 are listed in Table 3. Control genes underwent the same manual checks for sequence correctness by alignment to the reference genome as above and five genes were marked for exclusion. The second set of control genes were chosen to represent three ranges of mean expression levels encompassed by genes in the 17-gene signature (low: 0-400; medium: 500-900; high: 1200-1600). For each mean expression range, genes were (1) filtered if not expressed above background threshold (raw value>100) in 99% of samples; (2) ranked by coefficient of variation. The top 5 genes from each range in set #2 are listed in Table 3 along with previously reported reference genes (Paik et al., supra)13
Internal OOB cross-validation for the initial (full-gene-set) model on training data reported an ROC AUC of 0.704. This was comparable or better than reported by Johannes et al (2010) who tested a number of different classifiers on a smaller subset of the same data and found AUCs of 0.559 to 0.67114. It also compares favorably to the AUC value of 0.688 when the OncotypeDX algorithm was applied to this same training dataset. Mixed model clustering analysis identified three risk groups with probabilities for low risk<0.333; 0.333≦intermediate risk<0.606; and high risk≧0.606 (
Validation of the models against the independent test dataset also showed very similar results to training estimates. The full-gene-set model had an AUC of 0.730 and the 17-gene and 8-gene optimized models had minimal reduction in performance with AUC of 0.715 and 0.690 respectively. Again, this compared favorably to the AUC value of 0.712 when the OncotypeDX algorithm was applied to the same test dataset. Survival analysis again found very significant differences between the risk groups for the full-gene (p=6.54E-06), 17-gene (p=9.57E-06) and 8-gene (p=2.84E-05;
The linear relationship between risk group and rate of relapse continues if groups are broken down further (using training data) into five equal groups instead of the three groups defined above (
In order to maximize the total size of our training dataset we allowed samples to be included from both untreated patients and those who received adjuvant hormonal therapy such as tamoxifen. Since outcomes likely differ between these two groups, and they may represent fundamentally different subpopulations, it is possible that performance of our predictive signatures is biased towards one group or the other. To assess this issue we performed validation against the independent test dataset, stratified by treatment status, using the 17-gene model. Both groups were found to have comparable AUC values with the slightly better value of 0.740 for hormone-treated versus 0.709 for untreated. Survival curves were also highly similar and significant with p-value of 0.004 and 3.76E-07 for treated and untreated respectively (
The genes utilized in the RFRS model have only minimal overlap with those identified in other breast cancer outcome signatures. Specifically, the entire set of 100 genes (full-gene set before filtering) has only 6/65 genes in common with the gene expression panel proposed by van de Vijver, et al. N Engl J Med 347, 1999-2009 (2002)15, 2/21 with that proposed by Paik et al., supra, and 4/77 with that proposed by Wang et al. Lancet 365:671-679 (2005)20. The 17-gene and 8-gene optimized sets have only a single gene (AURKA) in common with the panel proposed by Paik et al., a single gene (FEN1) in common with Wang et al., and none with that of van de Vijver et al. A Gene Ontology analysis using DAVID16,17 revealed that genes in the 17-gene list are involved in a wide range of biological processes known to be involved in breast cancer biology including cell cycle, hormone response, cell death, DNA repair, transcription regulation, wound healing and others (
While methods such as those proposed by Paik et al., and de Vijver, et al. (both supra)13,15 exist to predict outcome in breast cancer, the RFRS is advantageous in several respects: (1) The signature was built from the largest and purest training dataset available to date; (2) Patients with HER2+ tumors were excluded, thus focusing only on patients without an existing clear treatment course; (3) The gene signature predicts relapse with equal success for both patients that went on to receive adjuvant hormonal therapy and those who did not (4) The gene signature was designed for robustness with (in most cases) several alternate genes available for each primary gene; (5) probe set sequences have been manually validated by alignment and manual assessment. These features, particularly the latter two, make this signature an especially strong candidate for efficient migration to multiple low-cost platforms for use in a clinical setting. Development of a panel for use in the clinic could take advantage of not only primary genes but also some number of alternate genes to increase the chance of a successful migration. Given the small but significant number of discrepencies observed between clinical and array based determination of ER status we also recommend inclusion of standard biomarkers such as ER, PR and HER2 on any design. Finally, we provide a list of consistently expressed genes, specific to breast tumor tissue, for use as control genes for those platforms that require them.
The RFRS algorithm is implemented in the R programming language and can be applied to independent patient data. Input data is a tab-delimited text file of normalized expression values with 17 transcripts/genes as columns and patient(s) as rows. A sample patient data file (patient_data.txt) is presented in Appendix 1. A sample R program (RFRS_sample_code.R) for running the algorithm is presented in Appendix 2. The RFRS algorithm consists of a Random Forest of 100,001 decision trees. This is pre-computed, provided as an R data object (RF_model—17gene_optimized) based on the training set and is included in the working directory. Each node (branch) in each tree represents a binary decision based on transcript levels for transcripts described above. Based on these decisions, the patient is assigned to a terminal leaf of each decision tree, representing a vote for either “relapse” or no “relapse”. The fraction of votes for “relapse” to votes for “no relapse” represents the RFRS—a measure of the probability of relapse. If RFRS is greater than or equal to 0.606 the patient is assigned to the “high risk” group, if greater than or equal to 0.333 and less than 0.606 the patient is assigned to “intermediate risk” group and if less than 0.333 the patient is assigned to “low risk” group. The patient's RFRS value is also used to determine a likelihood of relapse by comparison to a loess fit of RFRS versus likelihood of relapse for the training dataset. Pre-computed R data objects for the loess fit (RelapseProbabilityFit.Rdata) and summary plot (RelapseProbabilityPlot.Rdata) are loaded from file. The patient's estimated likelihood of relapse is determined, added to the summary plot, and output as a new report (see,
All publications, patents, accession numbers, and patent applications cited in this specification are herein incorporated by reference as if each individual publication or patent application were specifically and individually indicated to be incorporated by reference.
Although the foregoing invention has been described in some detail by way of illustration and example for purposes of clarity of understanding, it will be readily apparent to those of ordinary skill in the art in light of the teachings of this invention that certain changes and modifications may be made thereto without departing from the spirit or scope of the appended claims.
This application claims priority benefit of U.S. provisional application No. 61/789,071, filed Mar. 15, 2013 and U.S. provisional application No. 61/620,907, filed Apr. 5, 2012, which applications are herein incorporated by reference.
This invention was made with government support under Contract No. DE-ACO2-05CH11231 awarded by the U.S. Department of Energy. The government has certain rights in this invention.
| Number | Date | Country | |
|---|---|---|---|
| 61620907 | Apr 2012 | US | |
| 61789071 | Mar 2013 | US |
