WO2012125712A2 - Lung tumor classifier for current and former smokers - Google Patents

Abstract

The invention provides methods for the diagnosis of lung cancer in a subject based detecting abnormal gene expression patterns. The invention further provides methods of treating lung cancer.

Description

LUNG TUMOR CLASSIFIER FOR CURRENT AND FORMER SMOKERS

RELATED APPLICATION(S)

This application claims the benefit of U.S. Provisional Application No. 61/452,481, filed on March 14, 2011.

The entire teachings of the above application are incorporated herein by reference.

BACKGROUND OF THE INVENTION

Lung cancer has an extremely high mortality rate of up to 85% within five years and is consequently the number one cause of cancer death in the US and the rest of the world. Current or former smokers can often be suspect for lung cancer on the basis of unusual radiographic imaging results or the presence of clinical symptoms associated with the disease. The technique of flexible bronchoscopy is used as a noninvasive diagnostic test in these cases; however the sensitivity for identifying individuals who actually have lung cancer can vary from 30-80%.

Thus, there is a need for better tests to diagnose lung cancer in current and former smokers.

SUMMARY OF THE INVENTION

The present invention provides, inter alia, diagnostic methods for lung cancer and associated methods of treatment. The invention is based, at least in part, on the discovery that a support vector machine (SVM) index, based on the expression level of 51 genes, is a significant diagnostic tool for lung cancer.

Significant diagnostic power can be achieved with smaller subsets of genes, such as three of the 51.

Accordingly, in one aspect, the invention provides a method of predicting the risk of lung cancer in a subject. The method comprises the step of determining {e.g. , by testing by any means), in an isolated sample from a patient {e.g. , a biological sample), whether the sample exhibits an abnormal expression pattern of one or more genes selected from ABCF2, AGMAT, BBS9, BLK, C16orf42, CAT, CDK14, CLCN3, CNPY4, DIP2A, ENPP4, EPX, EXOC6B, EZH1, H2AFV, HAUS2, IL1F6, INHBC, IVD, MCF2L, MRPL52, MYL12B, MZT2B, OLFML2B, ORC6L, PDE4C, PGF, PGRMC1, POLR1B, PRDX2, PRKAR1A, RIOK3, RNASE4, RPL27A, RPL37A, RPL38, RRAGD, SBFl, SH2B2, SLC35E1, SLC39A6, SNX17, SPG21, TAOK1, TEX13B, TYMP, USP34, WDR6, XRCC2, ZNF701, and ZNF721, where an abnormal expression pattern predicts and increased risk of lung cancer in the patient. In particular embodiments, the method comprises determining the expression level of at least three of these genes, e.g. , at least 3, 5, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 25, 30, 35, 40, 45, 50, or all 51 genes. In still more particular embodiments, the expression level of at least 3 genes is determined. In other particular embodiments, the expression level of at least 7 genes is determined.

In some embodiments, at least one of the genes is selected from ABCF2, AGMAT,

BBS9, BLK, C16orf42, CAT, CDK14, CNPY4, DIP2A, ENPP4, EPX, EXOC6B, EZH1, H2AFV, HAUS2, IL1F6, IVD, MCF2L, MYL12B, MZT2B, OLFML2B, ORC6L, PDE4C, PGRMC1, RPL27A, RRAGD, SBFl, SH2B2, SNX17, SPG21, TAOKl, TEX13B, TYMP, WDR6, XRCC2, ZNF701, and ZNF721.

In certain embodiments, at least one of the genes is selected from CAT, ENPP4,

RRAGD, MCF2L, SLC39A6, SNX17, PGRMC1, ZNF721, PRKARIA, CLCN3, RNASE4, BBS9, H2AFV, POLR1B, DIP2A, EZH1, MRPL52, HAUS2, RIOK3, MZT2B, e.g. , the expression level of 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, or all 20 genes is determined. In more particular embodiments, at least one of the genes is selected from CAT, ENPP4, RRAGD, MCF2L, SNX17, PGRMC1, ZNF721, BBS9, H2AFV, DIP2A, EZH1, HAUS2, and MZT2B, e.g., the expression level of 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, or all 13 genes is determined. In still more particular embodiments, the expression levels of CAT, ENPP4, and RRAGD are determined.

In some embodiments an abnormal expression pattern comprises increased expression of one or more of BBS9, CAT, CLCN3, ENPP4, EZH1, H2AFV, MCF2L, PGRMCl, PRKARIA, RNASE4, RRAGD, SLC39A6, or SNX17. In certain embodiments, an abnormal expression pattern comprises decreased expression of one or more of ABCF2, AGMAT, BLK, C16orf42, CDK14, CNPY4, DIP2A, EPX, EXOC6B, FAM128B, HAUS2, IL1F6, INHBC, IVD, MRPL52, MYL12B, OLFML2B, ORC6L, PDE4C, PGF, POLR1B, PRDX2, RIOK3, RPL27A, RPL37A, RPL38, SBFl, SH2B2, SLC35E1, SPG21, TAOKl, TEX13B, TYMP, USP34, WDR6, XRCC2, ZNF701 or ZNF721. In more particular embodiments, an abnormal expression pattern comprises increased expression of one or more of BBS9, CAT, CLCN3, ENPP4, EZH1, H2AFV, MCF2L, PGRMCl, PRKARIA, RNASE4, RRAGD, SLC39A6, or SNX17 and decreased expression of one or more of ABCF2, AGMAT, BLK, C16orf42, CDK14, CNPY4, DIP2A, EPX, EXOC6B, FAM128B, HAUS2, IL1F6, INHBC, IVD, MRPL52, MYL12B, OLFML2B, ORC6L, PDE4C, PGF, POLR1B, PRDX2, RIOK3, RPL27A, RPL37A, RPL38, SBF1, SH2B2, SLC35E1, SPG21, TAOK1, TEX13B, TYMP, USP34, WDR6, XRCC2, ZNF701 or ZNF721.

The expression level of a gene may be measured at the nucleic acid (e.g., mRNA) or protein level. In particular embodiments, the expression level of one or more genes is measured at the nucleic acid level and in more particular embodiments, the expression levels of two or more genes are measure simultaneously, for example, using a microarray. In still more particular embodiments, the expression level of one or more genes is measured at the nucleic acid level using a microarray, such as an Exon 1.0 ST, Gene 1.0 ST, U 95, U133, U133A 2.0, or U133 Plus 2.0 AFYYMETRIX™ microarray. In other embodiments, the expression levels of one or more genes is measured at the nucleic acid level by rtPCR followed by qPCR, or serial analysis of gene expression (SAGE).

Expression levels can be analyzed by any means known in the art. In certain

embodiments, for example, expression levels can transformed before evaluation, e.g. , expressed as a fold-induction, log normalized, and, optionally, percentile ranked. In particular

embodiments, a sample is classified using by an SVM. The SVM may be based on the expression level of any of the gene combinations described above. In still more particular embodiments, the SVM may use weights substantially similar (i.e. , within about 1 , 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 30, 40, or 50%) to MCF2L (-1.4179), CLCN3 (-1.3756), EZH1 (-1.3011), TYMP (1.0144), CDK14 (0.8952), IVD (0.8668), EXOC6B (0.8071), ABCF2 (0.7865), SBF1 (0.7841), RRAGD (-0.7208), MRPL52 (0.7176), CAT (-0.6615), INHBC (0.6389), RIOK3

(0.6296), C16orf42 (0.5584), TEX13B (0.5532), RPL37A (-0.5478), SNX17 (-0.4909), AGMAT (0.4859), BBS9 (-0.4511), WDR6 (0.4187), SPG21 (0.4007), MZT2B (-0.3857), RPL38 (- 0.3581), OLFML2B (-0.3557), H2AFV (0.348), ENPP4 (-0.3385), ZNF721 (0.2899), PRDX2 (0.2413), MYL12B (0.2217), RPL27A (-0.2036), ZNF701 (0.1925), SLC39A6 (-0.1778), EPX (- 0.1525), PDE4C (-0.1521), DIP2A (0.152), IL1F6 (0.1517), POLR1B (0.1501), RNASE4

(0.1278), ORC6L (-0.1093), XRCC2 (-0.1071), PRKAR1A (-0.0991), PGF (-0.0949), USP34 (- 0.0769), PGRMC1 (-0.0763), CNPY4 (0.0726), SLC35E1 (0.0713), TAOK1 (0.0602), SH2B2 (0.0529), BLK (-0.0373), and HAUS2 (-0.0301).

A patient to be tested by the methods provided by the invention may be symptomatic or asymptomatic for lung cancer. In certain embodiments, the patient is a current or former smoker. In more particular embodiments, the patient is a former smoker. In still more particular embodiments, the patient is a former smoker who quit smoking within 10 years of isolation of the sample to be tested. In some embodiments, the subject is a non-smoker but may be considered to be at an increased risk for developing lung cancer— for example, the subject may have a history of exposure to second hand tobacco smoke, other environmental exposure, or genetic predisposition.

A sample to be tested for abnormal gene expression can be obtained from a patient by any means known in the art from any suitable source in the oral, esophageal, nasal, and/or pulmonary system. In particular embodiments, the sample is obtained by bronchoscopy or from nasal epithelial tissue.

In certain embodiments, the methods provided by the invention may further comprise the step of follow-on diagnosis including one or more of sputum cytology, flexible bronchoscopy (FB), transthoracic needle aspiration (TTNA), 18F-Fluorodeoxyglucose-positron emission tomography (FDG-PET), magnetic resonance imaging (MRI), endobronchial ultrasound and conventional or low-dose spiral computed tomography (LDCT), chest X-ray, or any biopsy. In more particular embodiments, subsequent to testing by the methods provided by the invention, the subject is determined to have lung cancer. In still more particular embodiments, the subject is determined to have early stage lung cancer, such as stage la or lb.

Accordingly, in another aspect the invention provides a method of treating lung cancer comprising administering a suitable prophylaxis to a patient determined to have lung cancer by any of the methods provided by the invention. In particular embodiments, the prophylaxis includes, e.g. , chemotherapy, hormonal therapy, immunotherapy, radiotherapy, surgery, targeted gene therapies (e.g. , epidermal growth factor receptor-tyrosine kinase inhibitors, such as gefitinib; and agents targeting ALK mutations and rearrangements, such as crizotinib, etc.) and combinations thereof.

BRIEF DESCRIPTION OF THE DRAWINGS

The patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.

FIG. 1 is a representation of a hierarchical clustering of the 51 gene signature. Patients are represented by columns (1 = diagnosed with lung cancer, 0 = cancer-free) and genes are represented by rows. FIGS. 2a-2e summarize significantly enriched biological function categories in the 51- gene signature identified by Ingenuity Pathway Analysis.

FIG. 3 summarizes the results of a ROC analysis of SVM index for the 90 patient training series.

FIG. 4 summarizes the gene expression data and clinical variables for the 90 patient training series. The yellow horizontal line across the figure represents the classification threshold of -0.16. Patients below this line are classified as 'non-tumor' while patients above the line are classified as 'tumor'. To the left of the gene expression data (red = high relative expression, green = low relative expression, in tumor vs. non-tumor) the patient ID, tumor/non- tumor status and SVM index is shown. To the right of the figure is each patient's bronchoscopy status, age, sex and cigarette pack-years. Patient data is also summarized in Table 5a.

FIG. 5 is a graph summarizing classifier LOOCV performance for SVM algorithms calculated on gene subsets of the 51 gene signature.

FIG. 6 is a graph of the mean of all LOOCV classification performance criteria shown in FIG. 5, generated for variations of the SVM classification algorithm using gene subsets of the 51 gene signature.

FIG. 7 summarizes the 51 gene signatures in the 60 patient validation, series 1. Patients ' represented by rows and genes are represented by columns. Red = high expression, green = low expression. The yellow horizontal line corresponds to a classification threshold of -0.75. Patient data is also summarized in Table 5b.

FIG. 8 is a graphical summary of a ROC analysis of the 51 -gene SVM classifier applied to the 60 patient validation series 1. AUC = 0.78 (95% CI: 0.65 to 0.87) PO.0001.

FIG. 9 is a graphical summary of a ROC analysis of the 51 -gene SVM and a previously- published [1] 80-gene algorithm for predictions of validation series 1.

FIG. 10 is a box plot of the 51 gene SVM indices calculated from lung tumor and normal lung tissue gene expression profiles. The difference is statistically significant (P<=0.001).

FIG. 1 1 summarizes a ROC analysis of the 51 -gene SVM applied to unpaired specimens from independent validation series 2, consisting of 41 gene expression profiles from either lung tumor or normal lung tissue. The AUC of 0.76 was statistically significant (P=0.0006).

FIG. 12 is a box plot of 51 -gene SVM indices calculated from gene expression data generated from histologically-normal BEC's from individuals of varying smoking status. P=0.25. FIG. 13 is a general linear model analysis of the 51 gene SVM index measured in BEC of histologically normal, current and former smokers.

FIG. 14 summarizes exemplary steps involved in applying the 51 -gene SVM to gene expression data generated from lung BEC and making a prediction of cancer or non-cancer. DETAILED DESCRIPTION OF THE INVENTION

The invention provides methods for the diagnosis of lung cancer. The methods provided by the invention may be used to classify a patient as having lung cancer or not having lung cancer based on an abnormal gene expression pattern. The invention further provides identifying patients for treatment of lung cancer. The following definitions are will be adhered to throughout this application.

"Patient" refers to a human at any stage of development. Examples of suitable patients include, but are not limited to, both female and male adult patients that have, or are at risk for developing, a lung cancer.

"Gene expression" refers to both nucleic acid level (e.g., mR A or cDNA derived from it) and protein level expression of a gene. Genes expressed as nucleic acids may or may not encode for and/or be translated into protein.

"Level of expression," "expression level," "gene expression level" and the like, refers to the amount of a gene expression product (e.g. , mRNA or protein). Expression levels may be transformed (e.g. , normalized) or analyzed "raw."

"Expression pattern" means at least two expression levels. For example, in some embodiments, the two or more expression levels may be the expression level of one gene at two or more time points or the expression levels of two or more different genes at the same, or different, times.

"Abnormal expression pattern" refers to a significant statistical and/or practical deviation in the expression level of one or more genes, relative to a suitable control. "Suitable controls" include, for example, paired biopsy samples from a single patient (e.g. , tissue samples obtained at different times from a patient, e.g. , before and after developing cancer; as well as a pair of biopsies from morphologically and/or histologically normal and morphologically and/or histologically suspect samples from the patient, which may be obtained at the same or different times) as well as reference values previously compiled from samples determined— by any means— to be cancerous or non-cancerous. For example, reference values for one or more genes may be compiled and used to develop a binary or probabilistic classification algorithm that is then used to classify a sample from a patient as cancerous or non-cancerous.

"Highly stringent hybridization" means hybridization conditions comprising about 6X SSC and 1% SDS at 65°C, with a first wash for 10 minutes at about 42°C with about 20% (v/v) formamide in 0.1X SSC, and with a subsequent wash with 0.2 X SSC and 0.1% SDS at 65°C.

Genes for use in the methods provided by the invention

The invention provides a useful set of genes that are differentially expressed in cancerous and non-cancerous lung tissues that can be used to diagnose cancer in a variety of patient samples from the oral, esophageal, nasal, and/or pulmonary system. The genes identified by the present invention are listed in Table 1. Table 1 further provides reference GenelDs, mRNA sequence accession numbers, protein sequence accession numbers, and AffymetrixID. These identifiers may be used to retrieve, inter alia, publicly-available annotated mRNA or protein sequences from sources such as the NCBI website, which may be found at the following uniform resource locator (URL): http://www.ncbi.nlm.nih.gov. The information associated with these identifiers, including reference sequences and their associated annotations, are all incorporated by reference. Additional useful tools for converting IDs or obtaining additional information on a gene are known in the art and include, for example, DAVID, Clone/GenelD converter and SNAD. See Huang et al, Nature Protoc. 4(l):44-57 (2009), Huang et at, Nucleic Acids Res. 37(1)1-13 (2009), Alibes et al, BMC Bioinformatics 8:9 (2007), Sidorov et al, BMC

Bioinformatics 10:251 (2009).

Table 1

Affymetrix Entrez

Symbol Name Unigene ID Cytoband RefSeq RNA ID RefSeq peptide

ProbeSetID GenelD

interleukin 1 family, member 6

IL1F6 Hs.278910 2ql2-ql4.1 221404_at 27179 NM_014440.1 NP_055255.1

(epsilon)

INHBC inhibin, beta C Hs.632722 12ql3.1 207688_s_at 3626 NM_ 005538.2 NP_005529.1

NM 004116 NP 004107.1

IVD isovaleryl-CoA dehydrogenase Hs.513646 15ql4-ql5 216495_x_at 3712

NM_054033 NP_473374.1

MCF.2 cell line derived

MCF2L Hs.170422 13q34 212935_^at 23263 NM_024979 NPJ)79255.2 transforming sequence-like

NM 178336.1 NP 848026.1 NM 180982 NP 851313.1 mitochondrial ribosomal protein NM 181304.1 NP 851821.1

MRPL52 Hs.355935 14ql l.2 221997_s_at 122704

L52 NM 181305.1 NP 851822.1

NM 181306.1 NP 851823.1 NM_181307 NP_851824.1

NM 001144944.1 NP 001138416.1

MYL12B myosin, light chain 12B, regulatory Hs.190086 18pl l.31 221474_at 103910 NM 001144945.1 NP 001138417.1

NMJ)33546 NP_291024.1 mitotic spindle organizing protein

MZT2B Hs.469925 2q21.1 220720_x at 80097 NMJ)25029.3 NP_079305.2

2B

OLFML2B olfactomedin-like 2B Hs.507515 lq23.3 213125_at 25903 NM_015441 NP 056256.1

Affymetrix Entrez

Symbol Name Unigene ID Cytoband RefSeq RNA ID RefSeq peptide

ProbeSetID GenelD

NM 002937 NP 002928.1

RNASE4 ribonuclease, RNase A family, 4 Hs.283749 14ql 1.1 205158_at 6038

NM_194431 NP_919412.1

RPL27A ribosomal protein L27a Hs.523463 l lpl5 212044_s_at 6157 NM 000990 NP 000981.1

RPL37A ribosomal protein L37a Hs.433701 2q35 214041 x_at 6168 NM 000998 NP 000989.1

NM 000999 NP 000990.1

RPL38 ribosomal protein L38 Hs.380953 17q25.1 221943_x_at 6169

NM_001035258 NP_001030335.1

RRAGD Ras-related GTP binding D Hs.31712 6ql5-ql6 221524_s_at 58528 NM_021244 NP 067067.1

SBFl SET binding factor 1 Hs.589924 22ql3.33 213383_at 6305 NM_002972.2 NP_002963.2

SH2B2 SH2B adaptor protein 2 Hs.489448 7q22 205367_at 10603 NM_020979.2 NP_066189.2 solute carrier family 35, member

SLC35E1 Hs.620596 19pl3.11 220796_x_at 79939 NM_024881.4 NP 079157.3

El

solute carrier family 39 (zinc NM 001099406.1 NP 001092876.1

SLC39A6 Hs.725276 18ql2.2 202089_s_at 25800

transporter), member 6 NM_012319.3 NP 036451.3

SNX17 sorting nexin 17 Hs.278569 2p23-p22 20099 l s at 9784 NMJ 14748 NP__055563.1 spastic paraplegia 21 (autosomal

SPG21 Hs.242458 15q21-q22 21 383_x_at 51324 NM_016630 NP_057714.1 recessive, Mast syndrome)

TAOK1 TAO kinase 1 Hs.631758 17ql 1.2 216310_at 57551 NM_020791 NP_065842.1

TEX13B testis expressed 13B Hs.333130 Xq22.3 221034_s_at 56156 NM_031273 NP_1 12563.1

Affymetrix Entrez

Symbol Name Unigene ID Cytoband RefSeq RNA ID RefSeq peptide

ProbeSetID GenelD

NM 001 113755.1 NP 001107227.1

TYMP thymidine phosphorylase Hs.592212 22ql3 217497_at 1890 NM 001113756.1 NP 001107228.1

NM_001953.3 NP_001944.1

USP34 ubiquitin specific peptidase 34 Hs.644708 2pl5 207365_x_at 9736 NM_014709.3 NP_055524.3

WDR6 WD repeat domain 6 Hs.654815 3p21.31 217734_s_ at 11180 NM_018031.3 NP_060501.3

X-ray repair complementing

XRCC2 defective repair in Chinese hamster Hs.647093 7q36.1 207598_x_at 7516 NM_005431 NP 005422.1 cells 2

NM 018260 NP 060730.1

ZNF701 zinc finger protein 701 Hs.235167 19ql3.41 220242_x_at 55762

NM 001172655.1 NP 001 166126.1

ZNF721 zinc finger protein 721 Hs.428360 4pl6.3 215978_x_at 170960 NMJ33474.2 NP_597731.2

An expression pattern for a patient can be obtained by determining the expression level of one or more of the genes in Table 1. In particular embodiments, the expression level of at least 3 of the genes in Table 1 are determined, e.g., at least 3, 5, 7 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 25, 30, 35, 40, 45, 50, or all 51 genes in Table 1. In particular embodiments, the expression level of at least one of ABCF2, AGMAT, BBS9, BLK, C16orf42, CAT, CD 14, CNPY4, DIP2A, ENPP4, EPX, EXOC6B, EZH1, H2AFV, HAUS2, IL1F6, IVD, MCF2L, MYL12B,

MZT2B, OLFML2B, ORC6L, PDE4C, PGRMC1, RPL27A, RRAGD, SBF1, SH2B2, SNX17, SPG21, TAOK1, TEX13B, TYMP, WDR6, XRCC2, ZNF701, and ZNF721 is determined.

In certain embodiments, at least one of the genes is selected from CAT, ENPP4, RRAGD, MCF2L, SLC39A6, SNX17, PGRMC1, ZNF721, PRKAR1A, CLCN3, RNASE4, BBS9, H2AFV, POLR1B, DIP2A, EZH1, MRPL52, HAUS2, RIOK3, MZT2B, e.g. , the expression level of 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, or all 20 genes is determined. In more particular

embodiments, at least one of the genes is selected from CAT, ENPP4, RRAGD, MCF2L, SNX17, PGRMC1, ZNF721, BBS9, H2AFV, DIP2A, EZH1, HAUS2, and MZT2B, e.g. , the expression level of 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, or all 13 genes is determined. In still more particular embodiments, the expression levels of CAT, ENPP4, RRAGD, and optionally MCF2L, are determined.

In some embodiments, an abnormal expression pattern comprises increased expression of one or more of BBS9, CAT, CLCN3, ENPP4, EZH1, H2AFV, MCF2L, PGRMC1, PRKAR1A, RNASE4, RRAGD, SLC39A6, or SNX17 (e.g., 1, 2, 3, 4, 5, 67, 8, 9, 10, 11, 12, 13, or all 14), relative to a suitable control. An abnormal expression pattern may also comprise decreased expression of one or more of ABCF2, AGMAT, BLK, C16orf42, CDK14, CNPY4, DIP2A, EPX, EXOC6B, FAM128B, HAUS2, IL1F6, INHBC, IVD, MRPL52, MYL12B, OLFML2B, ORC6L, PDE4C, PGF, POLR1B, PRDX2, RIOK3, RPL27A, RPL37A, RPL38, SBF1, SH2B2, SLC35E1, SPG21, TAOK1, TEX13B, TYMP, USP34, WDR6, XRCC2, ZNF701 or ZNF721 (e.g. , 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13,14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, or all 37). Detection methods

To determine a gene expression pattern, one or more gene expression levels are determined. Expression levels can be measured at the nucleic acid or protein level, or a combination thereof. Any means of determining gene expression levels can be employed when practicing the methods provided by the invention.

For example, nucleic acid expression levels can be determined in a number of ways including polymerase chain reaction (PCR), including reverse transcriptase (rt) PCR, real-time and quantitative PCR methods (including, e.g., TAQMAN, molecular beacon, LIGHTUP, SCORPION, SIMPLEPROBES; see, e.g. , U.S. Pat. Nos. 5,538,848; 5,925,517; 6,174,670; 6,329,144; 6,326,145 and 6,635,427));

Northern blotting; Southern blotting of reverse transcription products and derivatives; array based methods, including blotted arrays or in szYw-synthesized arrays; and sequencing, e.g., sequencing by synthesis, pyrosequencing, dideoxy sequencing, and sequencing by ligation, or any other methods known in the art, such as discussed in Shendure et al. , Nat. Rev. Genet. 5:335-44 (2004) or Nowrousian Euk. Cell, 9(9): 1300-1310 (2010), including such specific platforms as

HELICOS™, ROCHE™ 454, ILLUMINA™ /SOLEXA™, ABI SOLiD™, and POLONATOR™ sequencing.

Expression levels can be measured by detecting the reference nucleic acid sequences listed in Table 1 , as well as complements, fragments, and similar nucleic acid sequences of the reference nucleic acid sequences listed in Table 1. "Similar nucleic acid sequences" can be naturally occurring (e.g., allelic variants or homologous sequences from other species) or engineered variants to the reference nucleic acid sequences in Table 1 and will be at least about 60, 65, 70, 75, 80, 85, 90, 95, 96, 97, 98, 99% or more identical (or hybridize under highly stringent hybridization conditions to a complement of a nucleic acid sequence listed in Table 1) over a length of at least about 10, 20, 40, 60, 80, 100, 150, 200 or more nucleotides or over the entire length of the reference nucleic acid sequences in Table 1. Fragments of the reference nucleic acid sequences in Table 1— or similar nucleic acid sequences— can be of any length sufficient to distinguish the fragment from other sequences expected to be present in a mixture, e.g. , at least 5, 10, 15, 20, 40, 60, 80, 100, 150, 200 or more nucleotides or at least about 5, 10, 20, 30, 40, 50, 60, 70, 80, 90, 95 % of the length of the reference nucleic acid sequences in Table 1.

In particular embodiments, the expression levels of one or more of the genes in Table 1 are measured simultaneously, for example, on a nucleic acid microarray. Various microarray platforms may be adapted for use in the methods provided by the invention, including both spotted and in szYu-synthesized arrays as well as both commercially available standard arrays as well as custom arrays, e.g. , custom arrays capable of detecting the expression level of one or more of the genes in Table 1. Specific microarray platforms that are useful in the methods provided by the invention are available from AFFYMETRIX™, AGILENT™ and ILLUMINA™ In more particular embodiments, the expression levels may be measured on a U133 2.0 Plus or U133 A 2.0 microarray from AFFYMETRIX™.

Protein levels can be measured by quantitative cytochemisty or

histochemisty, ELISA (including direct, indirect, sandwich, competitive, multiple and portable ELISAs (see, e.g. , U.S. Patent No. 7,510,687)), western blotting (including one, two or higher dimensional blotting or other chromatographic means— optionally including peptide sequencing), peptide sequencing {e.g. , coupled to HPLC), and microarray adaptations of any of the foregoing (including antibody or protein-protein (i.e., non-antibody) arrays).

Protein techniques typically, but not necessarily, employ antibodies (e.g. , direct sequencing). Antibodies for use in the methods provided by the invention can be directed to any of the reference peptide sequences listed in Table 1 , as well as fragments of these sequences, similar peptide sequences, and fragments of similar peptide sequences. "Similar peptide sequences" can be naturally occurring (e.g. , allelic variants or homologous sequences from other species) or engineered variants to the reference peptide sequences in Table 1 and will exhibit substantially the same biological function and/or will be at least about 60, 65, 70, 75, 80, 85, 90, 95, 96, 97, 98, 99% or more homologous (i.e., conservative substitutions (see, e.g., Heinkoff and HeinkoffPNAS 89 (22): 10915-10919 (1992) and Styczynski et at, Nat.

Biotech. 26 (3): 274-275 (BLOSUM, e.g. , BLOSUM 45, 62 or 80) or Dayhoffet al, Atlas of protein sequence and structure (volume 5, supplement 3 ed.). Nat. Biomed. Res. Found., pp. 345-358 (PAM, e.g., PAM 30 or 70)) or identical at the amino acid level over a length of at least about 10, 20, 40, 60, 80, 100, 150, 200 or more amino acids or over the entire length of the reference peptide sequences in Table 1.

Fragments of the reference protein sequences in Table 1— or similar peptide

5 sequences— can be of any length sufficient to distinguish the fragment from other sequences expected to be present in a mixture, e.g. , at least 5, 10, 20, 40, 60, 80, 100, 150, 200 or more amino acids or at least about 5, 10, 20, 30, 40, 50, 60, 70, 80, 90, 95 % of the length of the reference peptide sequences in Table 1.

The term "antibody," as used herein, refers to an immunoglobulin or a part

10 thereof, and encompasses any polypeptide comprising an antigen-binding site

regardless of the source, species of origin, method of production, and characteristics. As a non-limiting example, the term "antibody" includes human, orangutan, mouse, rat, goat, sheep, and chicken antibodies. The term includes but is not limited to polyclonal, monoclonal, monospecific, polyspecific, non-specific, humanized,

15 camelized, single-chain, chimeric, synthetic, recombinant, hybrid, mutated, and CDR-grafted antibodies. For the purposes of the present invention, it also includes, unless otherwise stated, antibody fragments such as Fab, F(ab')2, Fv, scFv, Fd, dAb, VHH (also referred to as nanobodies), and other antibody fragments that retain the antigen-binding function. Antibodies also include antigen-binding molecules that

20 are not based on immunoglobulins, as further described below.

Antibodies can be made, for example, via traditional hybridoma techniques (Kohler and Milstein, Nature 256: 495-499 (1975)), recombinant DNA methods (U.S. Patent No. 4,816,567), or phage display techniques using antibody libraries (Clackson et al., Nature 352: 624-628 (1991); Marks et al, J Mol. Biol. 222:

25 581-597 (1991)). For various other antibody production techniques, see Antibodies:

A Laboratory Manual, eds. Harlow et al., Cold Spring Harbor Laboratory, 1988.

In some embodiments, the term "antibody" includes an antigen-binding molecule based on a scaffold other than an immunoglobulin. For example, non- immunoglobulin scaffolds known in the art include small modular

30 immunopharmaceuticals {see, e.g., U.S. Patent Application Publication Nos.

20080181892 and 20080227958 published My 31, 2008 and September 18, 2008, respectively), tetranectins, fibronectin domains (e.g., AdNectins, see U.S. Patent Application Publication No. 2007/0082365, published April 12, 2007), protein A , lipocalins (see, e.g., U.S. Patent No. 7,118,915), ankyrin repeats, and thioredoxin. Molecules based on non-immunoglobulin scaffolds are generally produced by in vitro selection of libraries by phage display (see, e.g., Hoogenboom, Method Mol. Biol. 178:1-37 (2002)), ribosome display (see, e.g. , Hanes et al., FEBS Lett.

450: 105-110 (1999) and He and Taussig, J. Immunol. Methods 297:73-82 (2005)), or other techniques known in the art (see also Binz et al., Nat. Biotech. 23: 1257-68 (2005); Rothe et al., FASEB J. 20:1599-1610 (2006); and U.S. Patent Nos.

7,270,950; 6,518,018; and 6,281 ,344) to identify high-affinity binding sequences.

Programs for sequence alignments and comparisons include FASTA

(Lipman and Pearson, Science, 227: 1435—41 (1985) and Lipman and Pearson, PNAS, 85: 2444^18), BLAST (McGinnis & Madden, Nucleic Acids Res., 32:W20- W25 (2004) (current BLAST reference, describing, inter alia, MegaBlast); Zhang et al. , J. Comput. Biol, 7(l-2):203-14 (2000) (describing the "greedy algorithm" implemented in MegaBlast); Altschul et al, J. Mol. Biol, 215:403-410 (1990) (original BLAST publication)), Needleman-Wunsch (Needleman and Wunsch, J. Molec. Bio., 48 (3): 443-53(1970)), Sellers (Sellers, Bull. Math. Biol, 46:501-14 (1984), and Smith- Waterman (Smith and Waterman, J Molec. Bio., 147: 195-197 (1981)), and other algorithms (including those described in Gerhard et al, Genome Res., 14(10b):2121-27 (2004)), which are incorporated by reference. In particular embodiments, sequences are compared by BLAST using default parameters for nucleic acid or protein queries.

Analysis

Gene expression levels can be analyzed by any means in the art. Before further analysis, raw gene expression data can be transformed, e.g. , log-normalized, expressed as an expression ratio, et cetera. In particular embodiments, data may further be percentile-ranked or quantile-scaled, or modified by any nonparametric data scaling approaches.

Expression patterns can be evaluated and classified by a variety of means such as general linear model (GLM), ANOVA, regression (including logistic regression), support vector machines (SVM), linear discriminant analysis (LDA), principal compnant analysis (PCA), k-nearest neighbor (kNN), neural network (N ), nearest mean/centroid (NM), and baysian covariate predictor (BCP). A model, such as SVM, can be developed using any of the subsets and combinations of genes described herein based on the teachings of the invention, including reference 10, below. In a particular embodiment, the SVM utilizes the expression levels of CAT, ENPP4, and RRAGD.

Suitable cutoffs for evaluating an expression pattern {e.g., for classification as abnormal (cancer) or normal (non-cancer)) can be determined using routine methods such as ROC (receiver operating characteristic) analysis and adjusted to achieve the desired sensitivity (e.g., at least about 50, 52, 55, 57, 60, 62, 65, 67, 70, 72, 75, 77, 80, 82, 85, 87, 90, 92, 95, 97, or 99% sensitivity) and specificity (e.g. , at least about 50, 52, 55, 57, 60, 62, 65, 67, 70, 72, 75, 77, 80, 82, 85, 87, 90, 92, 95, 97, or 99% specificity). Patients

A patient tested and/or treated by the methods of the invention can be any human. The patient may be symptomatic or asymptomatic for lung cancer. In certain embodiments, the patient is a current or former smoker. In more particular embodiments, the patient is a former smoker. In still more particular embodiments, the patient is a former smoker who quit smoking within 10 years of isolation of the sample to be tested, for example, within about 9, 8, 7, 6, 5, 4, 3, 2.5, 2, 1.5, 1.0, 0.5 years or within about 18, 16, 14, 12, 10, 8, 6, 4, 2 or 1 months. In some

embodiments, the subject is a non-smoker but may be considered to be at an increased risk for developing lung cancer— for example, the subject may have a history of exposure to second hand tobacco smoke, other environmental exposure (e.g., asbestos, commercial or industrial exhaust, dust, fire and/or smoke, radiation, et cetera), occupational exposures, or have an increased familial risk for developing lung cancer. In certain embodiments, the patient is determined to be at risk by clinical symptoms and/or due to abnormal radiography. A sample to be tested for abnormal gene expression can be obtained from a patient by any means known in the art from any suitable source in the oral, esophageal, nasal, and/or pulmonary system— such as lung or nasal tissues, such as epithelium. A sample can be isolated by swabbing, scraping, aspiration, lavage, sputum collection, brochoscopy, or biopsy. The sample isolation may be

nondiagnostic, inconclusive, or suggestive of lung cancer— for example, based on abnormal gross tissue morphology, cytogenetic abnormalities, initial histological analysis, and/or abnormal gene expression.

Subsequent diagnostics and treatment

In certain embodiments, the methods provided by the invention may further comprise the step of follow-on diagnosis, for example where the subject has been identified as having an increased risk for lung cancer by methods provided by the invention. For example, in certain embodiments, the patient may be subject to techniques such as sputum cytology, flexible bronchoscopy (FB), transthoracic needle aspiration (TTNA), 18F-Fluorodeoxyglucose-positron emission tomography (FDG-PET), magnetic resonance imaging (MRI), endobronchial ultrasound and conventional or low-dose spiral computed tomography (LDCT), chest X-ray, or any biopsy. In more particular embodiments, subsequent to testing by the methods provided by the invention, the subject is determined to have lung cancer, e.g. , stage 0, la, lb, 2a, 2b, 3a, 3b, or stage 4, as defined in the American Joint Committee on Cancer 7^th Edition staging poster (2009), which is incorporated by reference. In still more particular embodiments, the subject is determined to have early stage lung cancer, such as stage la or lb. For additional descriptions of lung cancer classifications, see, for example, Travis et ah, J. Thorac Oncol. 6(2); 244-285 (2011), Ravenel et al, J. Thorac Imaging 25(4) W107-111 (2010).

Where a patient tested by the methods provided by the invention is determined to have lung cancer, suitable prophylaxis may be administered. The skilled artisan will be familiar with appropriate prophylaxis following a diagnosis of lung cancer. For example, the prophylaxis can include, e.g., chemotherapy, hormonal therapy, immunotherapy (including both immunization of a patient as well as administering antibodies to, e.g. , a. tumor antigen or other suitable anti-neoplastic target), radiotherapy, surgery, targeted gene therapies (e.g. , epidermal growth factor receptor-tyro sine kinase inhibitors, such as gefitinib; and agents targeting ALK mutations and rearrangements, such as crizotinib, etc.) and combinations thereof. In 5 particular embodiments, prophylaxis may include, in order, radiotherapy, surgery, and adjuvant therapy, such as chemotherapy.

EXEMPLIFICATION

Example 1 : Initial results and materials and methods:

1.1. Gene expression data and clinical annotations

10 In order to identify genes with expression patterns associated with the

malignant status if a patient's lung tissue, 192 Affymetrix U133a CEL files and clinical annotations were downloaded from the Boston University website:

http://pulm.bumc.bu.edu/CancerDx/. The data were generated from bronchoscopy epithelial cells (BEC's) obtained from current and former smokers undergoing

15 flexible bronchoscopy for clinical suspicion of lung cancer at four tertiary medical centers between January 2003 and April 2005. This dataset was previously published [1] and is also available in the NCBI Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/) with accession number GSE4115 and is referred to as the training series or validation series 1 in this report. After excluding

20 GeneChips with fewer than 30% the total probe set detected, 150 gene expression profiles were available for analysis.

To further investigate the function of the genes selected from analysis of the BEC data, gene expression data representing lung tumor tissue and normal, noncancerous, lung tissue was obtained from NCBI GEO with accession number

25 GSE 10072 (validation series 2). This series represented 135 fresh frozen tissue samples of adenocarcinoma and paired noninvolved lung tissue from current, former and never smokers, with biochemically validated smoking information [5]. After quality control analysis, as described in section 1.2, below, 107 gene expression profiles were available; 58 tumor and 49 non-tumor tissues, from 20 never smokers,

30 26 former smokers, and 28 current smokers. Two additional validation series were gene expression profiles generated from BEC of histologically-normal individuals of varying smoking status

(Validation series 3) and of individuals with dysplastic lesions, with and without lung cancer (Validation series 4). All four series used are summarized in Table 2 Table 2: Overview of patient series used

1.2. Data processing and quality control

Raw CEL files were obtained for each series and processed with the MAS 5 algorithm (Affymetrix Inc, Santa Clara SF) before being median centered using the

Affymetrix internal 100 housekeeping/reference probe set. Individual probes with intensity measurements less than 10 were excluded. Multiple probes/probe sets were reduced to one per gene symbol by using most variable probe, measured by

IQR (interquartile range) across all samples and probes absent across 25% or more arrays were excluded.

Profiles from the BEC series with less than 30% of the microarray features detected were excluded from further analysis, leaving 150 which were randomly divided into a training series of 90 individuals and a validation series comprised of the remaining 60. All 107 GeneChip profiles from the LT (lung tissue ) series (series 3 and 4) were used, as these had been previously screened for quality control issues.

All data processing was performed using Microsoft Excel (Microsoft Corporation, Redmund, WA), R (www.r-project.org), BRB Array Tools [6] and Bioconductor [7]. Statistical analyses were performed using MedCalc 11.4

(MedCalc Software, Mariakerke, Belgium)

1.3. Gene selection & data scaling to minimize batch effects

To identify genes differentially expressed between patients with and without cancer, data from the BEC training series were analyzed with ANOVA. This method was chosen to identify individual genes expression patterns associated with cancer/no-cancer, independent to other clinical variables that may potentially impact on gene expression in BEC. The formula applied to the normalized training series dataset was:

Equation 1 : ANOVA formula used to select individual genes with differential expression patterns in BEC from smokers with and without lung cancer.

log expression ~ [cancer/no-cancer] + [pack years] + [age] + [smoking status]

Where smoking status refers to a binary variable of whether an individual patient quit smoking more or less than 10 years ago and pack-years and age are also binary, depending on whether the individual is above or below the median value of these measurements.

Genes were selected based on two criteria: 1) a P-value for the difference between cancer and no cancer of < 0.001 by ANOVA and 2) an absolute fold- change of > 1.5. This ensures that each gene is both statistically and 'biologically' differentially expressed between patients with and without cancer.

Two-dimensional hierarchical clustering was performed using the gene expression data associated with the genes selected by this approach in order to visualize the differences between the two patient phenotypes described by the molecular signature. Median centering was performed on the genes, prior to application of the average-linkage hierarchical clustering function using Cluster 3.0 [8].

Gene expression data can be affected by a number of non-biological parameters, including environmental conditions, subtle differences in wash station and scanner settings and operator error. In order to minimize the impact of systematic variation on the expression values of the selected genes, data

corresponding to those passing the ANOVA selection criteria were converted to percent-rank values, which range from 0.00 to 100.00. This conversion can be performed with the function percentrank in Microsoft Excel 2007 or later (Microsoft Corporation, Redmond, WA), or with the ecdf function in the R statistical programming language.

1.4. Algorithm training & analysis of independent validation series

In order to create a diagnostic algorithm capable of predicting the malignant status of an individual smoker, a Support Vector Machine (SVM) algorithm was trained on those genes identified as being differentially expressed using ANOVA as described above [9]. A support vector machine (SVM) is a class prediction algorithm whose effectiveness has been demonstrated in many areas of machine learning research. SVMs were developed by V. Vapnik [10]

The SVM predictor is a linear function of the log-intensities or percent-rank values that best separates the data subject to penalty costs on the number of specimens misclassified. To apply the trained algorithm to a test sample, each gene expression measurement is multiplied by a specific weight, before the sum of the resulting values is taken. This value is referred to as the SVM index and is used to assign the test sample or patient to a class of tumor or no-tumor, depending upon which side of a predetermined classification threshold it lies. This threshold can be adjusted to achieve a certain sensitivity and specificity, based on the indices generated by cross validation of the training series. For the purposes of this diagnostic application, the classification threshold was tuned to give a (non-tumor result) sensitivity of >90%. The trained SVM algorithm was then applied to the four independent validation series and the resulting SVM index and binary classification of tumor/non-tumor was recorded for each patient.

1.5. Statistical evaluation of predictions

The SVM index generated for each validation series was analyzed using

Receiver Operator Curve (ROC) and Area Under the Curve (AUC) analysis. T-test, chi-square tests and General Linear Model (GLM) analysis was performed on the binary predictions of tumor/non-tumor. All T-tests were two-sided and a P<0.05 was considered to be statistically significant. 1.6. Results

1.6.1. Analysis of genes differentially expressed between tumor and non-tumor gene expression profiles.

After pre-processing and data normalization, ANOVA was used to compare the gene expression BEC's from patients with and without a diagnosis of lung cancer. By applying a selection criteria of P<0.001 and fold-change >1.5, a set of 51 unique genes was identified (Table 3). The number of genes returned by this equation by applying other selection criteria is also shown Table 3.

Table 3: Training series ANOVA results

Hierarchical clustering was performed on the 51 genes in the 90 patient training series is shown in FIG.l . The separation of individuals with and without cancer can be seen by the structure of the horizontal dendrogram. Approximately ¾ of the 51 genes are more highly expressed (red) in individuals without cancer, compared to those with cancer.

Functional characterization of the 51 gene set was carried out using the Database for Annotation, Visualization and Integrated Discovery (DAVID ) v6.7 [11]. This method uses gene ontology, molecular pathway, chromosomal location, disease association and other data to identify significantly enriched categories of genes within a specific set. This analysis showed that the 51 gene set is enriched for genes involved in cellular response to hydrogen peroxide (enrichment score: 1.77), ribosomal structure (1.24), iron binding (0.93), hormone stimulus (0.61) and apoptosis regulation (0.59).

Ingenuity Pathway Analysis (INGENUITY SYSTEMS®;

www.ingenuity.com) was also performed. As shown in FIG.2 and Table 4, the 51- gene signature exhibits significant overlap with molecular networks involved in cell- to-cell signaling, cellular growth, proliferation, tissue morphology, small molecule biochemistry, cellular development.

Table 4 Ingenuity Pathway Analysis - Significantly enriched molecular networks present in the 51 -gene signature.

Number

Enrichment Top Network

Molecules in IPA Network of genes in

Score Functions

common

AGMAT, ATM, CHI3L2, CLCN3, CLEC5A,

DPP7, EPX, GBP6, GBP4 (includes Cell-To-Cell EG: 17472), GLIPR2, IFNG, IGF1R, IKBKE, Signaling and IL4, MAPK1 , MYHS, MZT2B, NR3C1, Interaction, OLFML2B, PDE4C, PTRH2, RAET1B, 28 13 Cellular Growth RNASE4, RPL37A (includes EG:6168), and Proliferation, RRAGD, SBF1, SLC28A1, SREBF1, STYK1, Gastrointestinal TAOK2 (includes EG:381921), TNF, TREM3, Disease

TYMP, USP34, WDR6 (includes EG: 11180)

ABLIM, Akt, BLK, CAT, CDK14, CREBL2,

Tissue

DHRS3, DOK5, ERK1/2, ethylene glycol,

Morphology, EZH1, FSH, GK7P, IL1F6, Lh, LOC81691,

Small Molecule Mapk, NFkB (complex), NRG1, NRG3, PGF, 23 1 1

Biochemistry, PGRMCl, PI4K2A, POP5, PRDX2,

Cellular

PRKAR1A, PRKX, QRFP, RIOK3, RIP,

Development SH2B2, STK17A, TMEFF2, TP53I11, TRIB1 ABCF2, ABCF3, BBS9, BMP2K (includes

EG:55589), C140RF147, C20ORF111, CDC6,

Genetic

CDIPT, CWC25, DGKQ, EID1, FXYD6,

Disorder, HAUS2, HNF4A, INHBC, INHBE, KDM5B,

Neurological MCM8, MDM2, MT1H, MT1X, ORC6L, 21 1 1

Disease,

POLR1B, Prereplicative Complex, RBL2,

Ophthalmic ROB03, RPL27A, SLC35E1, SLC39A6,

Disease

SN 17, SRP19, TMEM87A, ZNF133,

ZNF701, ZNHIT6

ABL1, ARHGAP26, ARHGEF4, ARHGEF25,

CHCHD2, CNTNAP1, CRK/CRKL, DEF6,

Cellular

DOCK3, Integrin alpha 6 beta 1, LCK, LIMK2,

Assembly and MAPK8, MAPK15, MATK, MCF2L, MICALl ,

Organization, MYL12B, NET1, PTPN18, PTPRH, PXN, 10 6

Cellular

RAD52, RHOA, RhoGap, RHOU, RIPK4,

Movement, Cell RPL38 (includes EG:6169), SEMA4D,

Morphology SH2D2A, SPG21 , TAOK1 , TUB, Vegf

Receptor, XRCC2

Genetic

Disorder, Neurological

ENPP4, MECP2 3 1

Disease,

Psychological Disorders

Cardiac

Infarction,

DIP2A, FSTL1 3 1 Cardiovascular

Disease, Genetic Disorder

Molecular Transport, Protein

ARL4D, CNPY4 3 1

Trafficking, Cancer

Gene Expression,

EXOC6B (includes EG:23233), GTP, MTERF, Energy

2 1

RALA Production, Lipid

Metabolism

1.6.2. Optimization of the 51 gene SVM classification threshold

ROC analysis was performed on the cross-validated SVM indices generated for the 90 patient training series, as shown in FIG.3. An AUC of 0.891 was observed, with a 95% CI of 0.808 and 0.947 (PO.0001).

To determine the optimal classification threshold value in order necessary to achieve the desired sensitivity of >90%, the ROC output shown in Table 5 was inspected. This represents the performance of the multi-gene classifiers at different threshold cutoff levels. The threshold of -0.16 gives a sensitivity of 90.91 (95% CI: 78.3 to 97.5) and specificity of 73.91 (95% CI: 58.9 to 85.7). Table 5: Classification threshold variations for the 51 -gene SVM and corresponding performance criteria when applied to the 90 patient training series.

To visualize this classification threshold in the context of the 90 patient training series and associated clinical variables, a gene expression heatmap was created in which hierarchical clustering was used to order the gene expression data (columns) and patients (rows) were ordered by increasing value of the SVM index FIG.4. The subject data is also summarized in Table 5a, below. Years

Pack Years since

Cancer or SVM diagnostic age year pack cancer cancer smoki quittin

GWEIGHT normal Index bronch median sex Median years type stage age race status ng s

Adenocar

s215p309Bl _l -3.04 N b F b 46 cinoma 3b 86 CAU CC 47 1 s222p311Bl 1 -2.75 N b M b 105 Squamous 2a 70 CAU CC 65 0

Non Small

s237p340Bl _l -2.43 N b M b 48 Cell 3a 73 CAU CC 50 0

Non Small

s258p365Bl _l -2.43 N b M b 100 Cell 4 75 AFA FC 50 10 s212p306Bl 1 -2.37 Y a M b 78 Squamous 3a 55 CAU FC 39 4 s213p307Bl 1 -2.29 Y a M b 64 Small Cell Lim 55 CAU CC 44 0 s257p364Bl 1 -2.21 N a M a 25 Large Cell 3a 55 AFA CC 35 0

Adenocar

sl63p244Bl ! -2.18 N b M b 60 cinoma 3b 65 CAU FC 42 6 sl93p280Bl 1 -2.17 Y a M a 20 Squamous 3a 52 AFA FC 38 1

Adenocar

s252p360Bl _l -2.01 Y a M b 75 cinoma 55 CAU CC 46 0 s206p299Bl 1 -1.94 N b M b 60 Squamous 3a 75 CAU CC 61 0 s220p313Bl -1.89 Y a M a 22 None None 58 CAU CNC 45 0 s233p341Bl 1 -1.87 N b M b 90 Small Cell Lim 79 CAU CC 61 0

Adenocar

sl33pl86Bl _l -1.72 N a M b 50 cinoma 2a 55 CAU FC 37 2 sl38pl95B l 1 -1.65 N b M a 4.5 Small Cell Ext 67 CAU FC 6 42 sl75p260Bl 1 -1.60 Y b F b 75 Squamous 3b 76 CAU FC 49 7 s325p450Bl 1 -1.56 Y b M b 57 Squamous 83 CAU FC 57 8 sl73p255Bl 0 -1.56 N b M a 0.5 None None 63 ASI FNC 2 43 s201p294Bl 1 -1.55 N b M b 108 ? 70 CAU FC 55 7

Years

Pack Years since

Cancer or SVM diagnostic age year pack cancer cancer smoki quittin

GWEIGHT normal Index bronch median sex Median years type stage age race status ng g s226p319Bl 0 -1.53 N b M a 25 None None 69 CAU CNC 51 0

Adenocar

s216p310Bl 1 -1.50 N a F a 32 cinoma 4 49 CAU FC 32 1

Adenocar

s321p446Bl 1 -1.50 N b F a 9 cinoma la 70 CAU FC 18 36 sl42p201Bl 1 -1.50 Y b M b 80 Squamous 4 61 CAU CC 48 0

Non Small

s228p321B l 1 -1.49 N b F a 20 Cell lb 67 CAU FC 20 20

Adenocar

s328p453Bl 1 -1.49 N a M a 19.5 cinoma 57 AFA CC 39 0 sl95p283Bl 0 -1.49 N a F a 5 None None 50 CAU CNC 26 0 s301p427Bl 0 -1.49 N a M b 70 ? 53 CAU FNC 43 3 s271p386Bl 1 -1.42 N b F a 22 Small Cell Lim 72 CAU FC 43 11 s270p385Bl 1 -1.40 Y a M a 29 Squamous lb 54 CAU CC 45 0

Adenocar

sl48p211Bl 1 -1.38 Y b F b 50 cinoma 4 65 HIS FC 51 2

Adenocar

sl54p219Bl 1 -1.37 Y a M a 20 cinoma 3b 59 CAU FC 20 26 sl61p231Bl 0 -1.34 Y a M b 70.5 None None 59 CAU CNC 48 0 s283p373Bl 0 -1.30 N a M a 17 None None 34 AFA CNC 17 0 s275p394Bl 1 -1.27 Y b M b 71 Small Cell Lim 65 CAU CC 53 0

Non Small

sl77p262Bl 1 -1.18 Y b M b 50 Cell 3b 65 CAU FC 57 1 sl39pl96Bl 1 -1.18 Y a M a 20 Squamous 3a 49 AFA FC 27 3 s209p302Bl 1 -1.13 Y b M a 23 Squamous 4 80 CAU FC 45 16 sl36pl93Bl 1 -1.12 N b M b 80 Large Cell 4 69 AFA CC 51 0

Years

Pack Years since

Cancer or SVM diagnostic age year pack cancer cancer smoki quittin

GWEIGHT normal Index bronch median sex Median years type stage age race status ng

s l22pl71Bl 1 -1.09 N a M b 120 Small Cell Ext 59 CAU CC 52 0 s l23pl72Bl 1 -1.00 N b M b 70.5 Squamous lb 64 CAU cc 49 0 sl59p238Bl 1 -0.97 N b F b 100 Squamous 3a 69 CAU FC 50 7 s97pl46Bl 0 -0.96 N a M a 20.5 None None 54 AFA CNC 43 0 s244p353Bl 0 -0.90 N b M b 105 ? 68 CAU CNC 53 0

Adenocar

s302p428Bl 1 -0.67 Y a M a 33 cinoma 4 46 CAU CC 30 0 sl l4pl63Bl 1 -0.62 N b F b 46 Large Cell 4 60 CAU FC 46 2 sl45p205B l 1 -0.57 Y a F b 35 Large Cell 4 55 AFA FC 38 3 sl51p215Bl 0 -0.56 N a F b 34 None None 49 AFA CNC 36 0

Non Small

s255p362Bl 1 -0.40 Y a F b 37 Cell 4 42 CAU CC 25 0 s274p391Bl 1 -0.28 N b M b 51 Squamous 3b 77 CAU FC 52 8 sl72p254Bl 0 -0.25 N a F a 5 None None 59 CAU FNC 5 40 s300p423Bl 0 -0.22 N b M a 4 ? 66 AFA CNC 54 0

Adenocar

sl44p204Bl 1 -0.16 N b M b 70 cinoma lb 68 CAU FC 24 28 s l40pl97B l 0 -0.13 N b M b 35 None None 76 CAU FNC 34 27 s77pl27B l 0 -0.04 N a M a 10 None None 43 AFA FNC 20 3 s235p344Bl 1 0.00 Y b M b 53 Squamous lb 75 CAU FC 52 6 s263p377Bl 0 0.09 N b F a 24.25 ? 64 CAU FNC 34 9 sl50p212Bl 1 0.11 Y b F b 60 Small Cell Ext 66 CAU FC 30 17 s68pl l8B l 0 0.22 N a M a 28 None None 38 AFA CNC 30 0 s57pl l3B l 1 0.26 Y b M b 75 Small Cell Ext 63 CAU CC 52 0

Years

Pack Years since

Cancer or SVM diagnostic age year pack cancer cancer smoki quittin

GWEIGHT normal Index bronch median sex Median years type stage age race status ng

s292p400B l 0 0.31 N a M a 16 ? 45 AFA CNC 31 0 sl l3pl62Bl 0 0.33 N a F a 10.5 None None 36 AFA CNC 23 0 s290p404Bl 1 0.35 Y b M b 35 Squamous 3b 63 CAU CC 50 0 sl24pl73Bl 0 0.35 N a M b 90 None None 54 CAU CNC 42 0 s86pl03Bl 0 0.36 N a M a 4 None None 30 HIS CNC 14 0 sl07pl56Bl 0 0.48 N a M a 14 None None 52 AFA FNC

s92pl41Bl 0 0.51 N a M a 3.5 None None 33 ASI CNC 16 0 s67pl l7Bl 0 0.51 N a M a 21 None None 48 AFA CNC 23 0 s240p345B l 0 0.51 N b M a 31 None None 76 CAU FNC 21 31 sl47p207Bl 0 0.51 N a M a 13 None None 59 HIS FNC 13 31 s81pl31Bl 0 0.53 N a M a 9.75 None None 49 AFA CNC 37 0 sl l2pl61B l 0 0.59 N b M a 30 None None 69 AFA CNC 62 0 s238p323Bl 0 0.69 N b M b 64 None None 79 CAU CNC 65 0 s56p l l2Bl 0 0.72 N b M b 51 None None 73 CAU CNC 66 0 s307p433Bl 0 0.87 N a M b 39.5 ? 47 AFA CNC 35 0 s89pl38B l 0 0.92 N a F a 7 None None 43 AFA FNC 14 15 s82pl32Bl 0 1.01 N a F a 22.5 None None 26 AFA CNC 17 0 s83pl33B l 0 1.03 N a M a 28.25 None None 41 CAU CNC 31 0 s317p443Bl 0 1.07 Y b M b 67 ? 71 CAU FNC 48 6 sl09pl58Bl 0 1.08 N a M a 0.5 None None 49 ASI CNC 7 0 s61pl08B l 0 1.20 N b M b 56 None None 71 CAU FNC 56 3 s262p376Bl 0 1.24 N a M a 5.5 ? 59 CAU FNC 7 39 sl06pl55Bl 0 1.25 N a F a 13 None None 37 HIS CNC 28 0

Years

Pack Years since

Cancer or SVM diagnostic age year pack cancer cancer smoki quittin

GWEIGHT normal Index bronch median sex Median years type stage age race status ng £ sl l0pl59Bl 0 1.27 N a F b 90 None None 54 CAU FNC 30 14 s309p424Bl 0 1.35 N b F b 53 ? 68 AFA CNC 53 0 s88pl37Bl 0 1.44 N a M a 20 None None 39 AFA CNC 22 0 s250p358B l 0 1.57 N a M a 24 None None 46 AFA FNC 26 2 s75pl25Bl 0 1.64 N a M a 19 None None 37 AFA CNC 29 0 s98pl47Bl 0 1.75 N a M a 1 None None 42 AFA CNC 9 0 sl l9pl68Bl 0 1.81 N a F b 64 ? 48 CAU FNC 32 3 s84pl34B l 0 2.04 N a F a 2 None None 45 AFA FNC 4 3

Table 5a

1.6.3. Training series: Comparison of gene number vs. classifier performance To determine whether a smaller number of genes could be used for classification, variations of the SVM algorithm trained on 2 to 51 genes were generated and applied to the 90-sample training series using LOOCV. The overall percent accuracy, sensitivity, specificity, positive predictive value and negative predictive values were recorded for each version of the classifier and plotted as shown in FIG.5. The mean of these five measurements was also calculated and plotted against gene number FIG.6).

This analysis revealed that with three (3) genes only, excellent LOOCV performance was achieved. Therefore acceptable performance can be achieved with minimum three gene subset of the 51 gene algorithm.

Example 2: Validation series 1 : GSE4115 (BEC)

A random selection of 60 gene expression profiles of BEC s from

individuals suspected of having lung cancer were partitioned from validation series 1 [1 ] and set aside for validation of the predictive algorithm. The 51 -gene SVM was applied to the validation set, as shown in FIG.7. The results of FIG.7 are also summarized in Table 5b.

Cancer

SVM or diagnostic age Pack year cancer

NAME index normal bronch median sex race Median cancer type stage age smoking status

quit less than 10 sl76p261Bl -2.87 0 N b M CAU b None None 65 years

quit for 10 years or sl88p273Bl -2.29 ! N b M AFA b Squamous 3b 73 greater

quit less than 10 sl97p285Bl -2.28 ! Y a M CAU a Adenocarcinoma lb 54 years

quit less than 10 s266p380B l -2.23 ! Y a M CAU b Small Cell Lim 59 years

quit for 10 years or s219p292Bl -2.23 _l Y b M CAU b Non Small Cell 3b 74 greater

quit less than 10 sl99p290Bl -2.19 _l Y a M CAU b Non Small Cell 4 57 years

quit less than 10 sl60p239Bl -2.11 ! N a M CAU b Squamous la 59 years

quit less than 10 s241p346Bl -2.10 N a M CAU b None None 51 years

quit less than 10 s236p339B l -2.02 ! Y b M CAU b Small Cell Lim 65 years

quit for 10 years or s203p296Bl -1.95 ! N b M CAU b Adenocarcinoma lb 74 greater

quit less than 10 s285p398Bl -1.89 _l Y a M CAU b Adenocarcinoma 3b 53 years

quit for 10 years or s312p435B l -1.82 _l N b M CAU b Small Cell Ext 78 greater

quit less than 10 s269p384Bl -1.79 Y b M CAU b Squamous lb 65 years

quit less than 10 s234p343Bl -1.78 1 N a M CAU a Squamous lb 51 years

quit less than 10 s223p312Bl -1.75 0 Y b M CAU b None None 69 years

quit less than 10 sl96p284Bl -1.74 0 N a F CAU a None None 53 years

quit for 10 years or s322p447Bl -1.68 1 N b F CAU a Adenocarcinoma 3b 78 greater

quit less than 10 s239p331Bl -1.61 1 Y a M CAU a Squamous 3a 58 years

quit less than 10 sl34pl87Bl -1.53 0 N a M ASI a None None 23 years

quit less than 10 s286p406Bl -1.44 0 N a F AFA a ? 45 years

quit less than 10 sl56p225Bl -1.41 1 N b M CAU b Squamous 4 68 years

quit less than 10 s214p308Bl -1.32 1 Y b F CAU b Squamous 3a 71 years

quit less than 10 s202p295Bl -1.24 1 N a M CAU b ? 4 56 years

quit for 10 years or sl41pl98Bl -1.22 0 N a M CAU a None None 55 greater

quit less than 10 s273p390Bl -1.08 1 N b M CAU a Squamous 70 years

quit less than 10 sl90p277Bl -0.94 1 N b M CAU b Squamous lb 75 years

quit less than 10 s308p434Bl -0.92 1 Y a M CAU b Non Small Cell 4 55 years

quit less than 10 sl62p242Bl -0.91 0 N a M CAU a None None 40 years

quit less than 10 s297p418Bl -0.80 1 Y b M CAU a Small Cell Lim 65 years

quit less than 10 sl35pl92B l -0.60 1 N b F CAU b Squamous 4 68 years

quit less than 10 sl 52p216B l -0.43 1 Y b F CAU b Non Small Cell 4 65 years

quit less than 10 s207p300Bl -0.21 0 N b M CAU b None None 71 years

quit less than 10 sl68p249Bl -0.14 1 Y a M CAU a Small Cell Lim 54 years

quit for 10 years or s246p355Bl -0.10 0 Y b M CAU b None None 74 greater s231p328Bl -0.07 1 Y b M CAU b Squamous 4 74

quit less than 10 s315p440Bl -0.06 1 Y a F CAU b Squamous 4 57 years

quit less than 10 s304p430Bl 0.02 0 N b M CAU b ? 63 years

quit less than 10 s282p370Bl 0.08 0 N b M CAU b ? 66 years

quit less than 10 s221p314Bl 0.14 1 Y a M AFA b Small Cell Lim 60 years

quit for 10 years or sl30pl83B l 0.18 1 N b M CAU b Adenocarcinoma la 69 greater

quit less than 10 s71pl21Bl 0.30 0 N a M HIS a None None 23 years

quit less than 10 s l l6pl65B l 0.31 0 N a M AFA a None None 38 years

quit less than 10 s243p349B l 0.34 1 Y b M CAU b Non Small Cell 4 83 years

quit for 10 years or sl 74p257Bl 0.53 1 N b M CAU a Adenocarcinoma 4 79 greater

quit less than 10 s299p422Bl 0.55 1 N a M CAU a Non Small Cell 4 58 years

quit less than 10 sl02pl51B l 0.61 0 N a M CAU a None None 27 years

quit for 10 years or s247p356Bl 0.64 0 N b M CAU b None None 74 greater

quit less than 10 s296p413Bl 0.74 0 N b M CAU b ? 71 years

quit less than 10 s72pl22Bl 0.79 0 N a M AFA a None None 38 years

quit for 10 years or sl81p266Bl 0.86 1 Y b M CAU b Squamous 4 62 greater

quit less than 10 s58pl l4B l 0.90 0 N a M HIS a None None 32 years

quit less than 10 s79pl29Bl 0.98 0 N a F AFA a None None 39 years

quit less than 10 s225p318Bl 1.01 0 N b M CAU b None None 71 years

quit less than 10 s63p78Bl 1.07 0 N a F AFA a None None 27 years

quit less than 10 s73pl23Bl 1.24 0 N a F AFA a None None 42 years

quit for 10 years or s95 l44Bl 1.38 0 N a M AFA a None None 50 greater

quit for 10 years or sl04pl53Bl 1.42 0 N a F AFA a None None 40 greater

quit less than 10 s85pl02Bl 1.68 0 N a M CAU a None None 39 years

quit less than 10 sl20pl69Bl 1.72 0 N a F CAU a None None 35 years

quit less than 10 s87pl36Bl 2.26 0 N a F AFA a None None 40 years

Table 5b

Overall, 24/33 (73%) patients predicted to have lung cancer were correctly classified, i. e. received a clinical diagnosis of cancer during the length of the time in which the original study was performed. Of the 27 individuals predicted to be tumor-free, 17 were diagnosed as tumor-free (70%>).

Chi-square analysis showed the association between SVM predictions and clinical diagnosis to be highly significant (P=0.001). A sensitivity of 75.00% (95% CI: 56.60% to 88.54%) and specificity of 67.86% (95% CI: 47.65% to 84.12%) were achieved.

2.1. : ROC analysis of validation series 1 and comparison to 80 gene algorithm.

ROC analysis was performed on the SVM indices generated for independent validation series 1. The association between this continuous variable and the tumor/non- tumor status of each patient was investigated with ROC analysis (FIG.8). A statistically significant AUC of 0.78 was observed, with a 95% CI of 0.65 to 0.87. This corresponds to a P-value of < 0.0001.

Comparison of ROC curves for the 51 gene SVM and the 81 gene 'biomarker' classifier of previously published [1] was then performed. The difference in AUC was (P=0.89).

2.2. Subset analysis of validation series 1

A number of important clinical cofactors exist within for patients included in this validation series, including smoking status (length of time between cessation of smoking and bronchoscopy procedure) and the ability of each patient to be diagnosed during the procedure itself.

To compare the performance of the 51 gene SVM in the subsets of the training series closely resembling the actual clinical need of a multi-gene diagnostic assay, performance data shown in Table 6 were calculated.

Table 6: Performance of the 51 gene SVM in subsets of validation series 1.

Sensitivity: probability that a test result will be positive when the disease is present (true positive rate).

"Specificity: probability that a test result will be negative when the disease is not present (true negative rate).

The 51 -gene SVM performed optimally in the subset of individuals who quit smoking less than 10-years prior to the diagnostic investigation and who could not be diagnosed by the procedure itself. In these patients the sensitivity was 88%, specificity was 73% with a negative predictive value of 94%.

The one individual incorrectly classified in the non-tumor category of subset (iv) was diagnosed with advanced (stage IV) lung cancer. A prediction of 'tumor' is less accurate in this subset, with 54% of cases in this category being diagnosed with lung cancer.

A noteworthy aspect of results from validation series 1 is the fact that 6/7 (86%) of the patients diagnosed with early-stage (la/lb) lung cancer, were correctly assigned to the 'tumor' category. Of the 12 patients diagnosed with stage 4 cancer, six were assigned to the 'tumor' category (50%), suggesting that the 51 -gene algorithm may be more accurate at correctly diagnosing patients with early-stage tumors, compared to individuals with advanced disease.

2.3. Multivariate analysis of validation series 1

General linear model analysis was performed on the 60 patients from validation series

1 to evaluate the significance of the 51 gene SVM in the context of other clinical covariates known to affect cancer risk and gene expression of BEC's from smokers. As shown in Table 7, the SVM prediction index is significantly associated with malignancy (P=0.013) in this validation series, independent to age, pack years, smoking status and sex. For comparison purposes, the same analysis was applied to predictions made with the previously published [1] 80-gene signature. The P-value for this algorithm was also significant (P=0.029), independent to the clinical covariates; however the P-value was larger than observed with the 51 gene signature.

Table 7: General linear model comparison of association between gene expression classifiers and malignant status of lung tissue

In both general linear model analyses, the age of the individual was also significantly associated with malignancy, independent to the other covariates included in the model. Example 3: Validation series 2: GSE 10072 (Lung tissue)

This dataset consists of lung tumor and normal lung gene expression profiles, generated from lung tissue biopsies. This is a different tissue type than that used in the training and validation series 1, and was included in the analysis to explore the characteristics of the 51 -gene signature in lung tissue directly. 3.1. Application of the 51 -gene SVM to lung tumor and normal lung gene expression data.

The 51 -gene SVM indices for the 25 lung tumors and 16 normal tissues (non-paired, i. e, from different individuals) were compared with standard Students T-Test. As shown in FIG.10, the SVM index is (on average) significantly lower in normal lung tissue, compared to lung tumors (mean indices: -1.27, -0.72 respectively; P=0.005). The dynamic range of SVM indices is also smaller in data generated from normal tissues, compared to the range of indices observed in lung tumors.

3.2. Validation Series 2: ROC Analysis of SVM index

The area under the curve was calculated for SVM indices from non-parried specimens from validation series 2 (41 gene expression profiles) as shown in FIG.l 1. The result of 0.76 (95% CI: 0.60 to 0.88) was statistically significant (P=0.0006), indicating the strength of the relationship between the 51 -gene SVM and the malignant or non-malignant status of lung tissue.

3.3. Comparison of the 51 -gene SVM index in paired lung tumor and normal tissue

Next it was investigated whether a significant difference exists in the S VM51 generated from paired lung and normal tissue gene expression profiles (i. e. both gene expression profiles generated from tissue material obtained from the same patient). A paired t-test was performed on the 33 pairs of tumor/normal tissue, resulting in a statistically significant difference of P=0.01 1. The mean index value of the normal tissues was -1.03, compared to a mean value of -0.72 from the matched tumor tissues (difference: 0.31).

Example 4: Validation series 3 : GSE7895 (BEC)

This dataset consists of gene expression data generated from histologically normal BEC's of Never Smokers (n=21), Former Smokers (n=31), and Current Smokers (n=52). This series was selected in order to investigate the variation in the 51 -gene SVM index between normal epithelium of individuals of varying smoking status.

A box plot and t-test of indices was performed for the three smoking-status categories (FIG.12), however no statistically significant difference was detected (P=0.25). There was a trend towards higher indices in former-smokers, compared to current smokers. As the 51 - gene SVM index is positively correlated with the probability of a patient being cancer-free, the trend observed in this dataset in agreement with results observed in the training and validation series 1. It may also reflect the time-dependant, lung tissue damage repair process that occurs after an individual ceases smoking [12].

4.1. General linear model analysis of SVM index and current/former smoking status

The relationship between smoking status of the histologically-normal individuals and the value of the 51 -gene SVM index was further explored using multivariate, general linear model analysis. As shown in FIG.13, when analyzed in a setting where variation in age and pack-years are taken into account, a statistically significant difference between current and former smokers is observed (P=0.042). The mean (adjusted) difference between the indices of former and current smokers is 0.54 (standard error: 0.27), with the former smokers in this validation series exhibiting a small but significantly larger (i.e. more 'tumor-like') SVM index compared to current smokers.

Example 5: Validation Series 4: GSE12815 (BEC)

This validation set consisted of gene expression profiles generated from BEC's from two smokers with COPD (no lung cancer), one smoker without COPD (no lung cancer), two patients with lung cancer and two individuals without lung cancer. As this validation series is too small for meaningful statistical analysis, the results of the 51 -gene SVM analysis are shown in Table 8.

Table 8: Patient summary and 51 -gene SVM predictions for validation series 5

Two current-smoking patients (ID's: 80 and 327) were classified as 'non-tumor' by the 51 -gene SVM, which agreed with the clinical diagnosis. Three other currently-smoking patients (BT5, BT16, XXX) were classified as 'tumor', however only BT5 was diagnosed with lung cancer.

Two other patients (BT17 and BT18) quit smoking 30 and 20 years prior to their bronchoscopy procedures, respectively. Patient BT17 was predicted to have lung cancer, in agreement with their clinical diagnosis. Patient BT18 was also predicted to have lung cancer, however had not been diagnosed with this condition.

The results of this validation series are agreement with the previous analyses which show the 51 -gene SVM to be more accurate in diagnosing lung tumors in patients who quit smoking less than 10 years prior to the diagnostic investigation. Furthermore, both patients predicted to be cancer-free, were actually free of lung cancer, in keeping with the high negative predictive value of the algorithm observed in validation series 1.

A multi-gene diagnostic assay has been developed using AFFYMETRIX™ whole- genome profiles generated from BEC of individuals who were suspected of having lung cancer. By selecting genes with robust patterns of differential expression between individuals who were confirmed (with traditional diagnostic methods) to have lung cancer and those who were not, a 51 -gene signature was created. These genes were used to train a SVM algorithm, capable of classifying a new sample as either tumor or non-tumor, optimized for high sensitivity and negative predictive value.

The classifier was applied to four separate independent validation series and was shown to be highly accurate at identifying individuals without lung cancer, independent to other clinical variables known to influence a smoker's risk of lung cancer, such as age, pack- years, current smoking status and gender. The assay is also accurate at identifying patients with early stage cancer, who quit smoking within 10 years prior to collecting the BEC's for diagnostic analysis. In this subset of individuals, the 51 -gene signature is able to identify patients who have a clinical suspicion of lung cancer who are free of lung cancer with 94% accuracy (NPV).

The 51 -gene SVM index was also shown to be significantly associated with malignancy when calculated using gene expression data generated from biopsies of lung cancer and normal lung tissue, indicating the utility of the assay in a different tissue-type than it was developed from.

In a multivariate analysis of histologically-normal BEC gene expression data, the 51- gene SVM index was significantly different between current and former smokers, potentially reflecting a reversal of tissue damage that occurs when an individual ceases smoking. Table 9 Summarizes the SVM results Table 9: Genes present in the 51 gene SVM algorithm. Probe ID and gene annotation obtained from BioConductor Affymetrix annotation packages.

Affymetrix Gene name Accession Unigene ID Gene Cytoband SWM ProbeSet ID Symbol Weight

200605_s_at protein kinase, cAMP-dependent, NM_002734 Hs.280342 PRKAR1A 17q23- -0.0991 regulatory, type 1, alpha (tissue q24 specific extinguisher 1)

20099 l_s_at sorting nexin 17 NM_014748 Hs.278569 SNX17 2p23-p22 -0.4909

201120_s_at progesterone receptor membrane AL547946 Hs.90061 PGRMC1 Xq22-q24 -0.0763 component 1

201432_at catalase NM_001752 Hs.502302 CAT llpl3 -0.6615

201732_s_at chloride channel 3 AF029346 Hs.481186 CLCN3 4q33 -1.3756

202089_s_at solute carrier family 39 (zinc NM_012319 Hs.725276 SLC39A6 18ql2.2 -0.1778 transporter), member 6

203249_at enhancer of zeste homolog 1 AB002386 Hs.194669 EZH1 17q21.1- -1.3011

(Drosophila) q21.3

204160_s_at ectonucleotide AW 194947 Hs.643497 ENPP4 6p21.1 -0.3385 pyrophosphatase/phosphodiestera

se 4 (putative)

205158_at ribonuclease, RNase A family, 4 NM_002937 Hs.283749 RNASE4 14qll.l 0.1278

205367_at SH2B adaptor protein 2 NM_020979 Hs.489448 SH2B2 7q22 0.0529

206792_x_at phosphodiesterase 4C, cAMP- NM_000923 Hs.132584 PDE4C 19pl3.11 -0.1521 specific

207365_x_at ubiquitin specific peptidase 34 NM_014709 Hs.644708 USP34 2pl5 -0.0769

207598_x_at X-ray repair complementing NM_005431 Hs.647093 XRCC2 7q36.1 -0.1071 defective repair in Chinese

hamster cells 2

207688_s_at inhibin, beta C NM_005538 Hs.632722 INHBC 12ql3.1 0.6389

209246_at ATP-binding cassette, sub-family F AF261091 Hs.654958 ABCF2 7q36 0.7865

(GCN20), member 2

209958_s_at Bardet-Biedl syndrome 9 AF095771 Hs.372360 BBS9 7pl4 -0.4511

210934_at B lymphoid tyrosine kinase BC004473 Hs.146591 BLK 8p23-p22 -0.0373

211502_s_at cyclin-dependent kinase 14 AF119833 Hs.430742 CDK14 7q21-q22 0.8952

212044_s_at ribosomal protein L27a BE737027 Hs.523463 RPL27A llpl5 -0.2036

212206_s_at H2A histone family, member V BF343852 Hs.488189 H2AFV 7pl3 0.348

212935_at MCF.2 cell line derived AB002360 Hs.170422 MCF2L 13q34 -1.4179 transforming sequence-like

213105_s_at chromosome 16 open reading AI799802 Hs.134846 C16orf42 16pl3.3 0.5584 frame 42

213125_at olfactomedin-like 2B AW007573 Hs.507515 OLFML2B lq23.3 -0.3557

213383_at SET binding factor 1 AW593269 Hs.589924 SBF1 22ql3.33 0.7841

214041_x_at ribosomal protein L37a BE857772 Hs.433701 RPL37A 2q35 -0.5478

214627_at eosinophil peroxidase X14346 Hs.279259 EPX 17q23.1 -0.1525

215067_x_at peroxiredoxin 2 AU 147942 Hs.432121 PRDX2 19pl3.2 0.2413

215179_x_at placental growth factor AK023843 Hs.252820 PGF 14q24.3 -0.0949

215383_x_at spastic paraplegia 21 (autosomal AL137312 Hs.242458 SPG21 15q21- 0.4007 recessive, Mast syndrome) q22

Table 10 summarizes additional data (log normalized and percentile-ranked) about the genes used to generate the SVM.

Table 10

Non-

Parametric p- Tumor Tumor Fold-

ProbeSet Symbol value t-value (Mean) (Mean) change

204160 s at ENPP4 0.0000001 -5.722 1.6 1.77 0.9

201432 at CAT 0.0000001 -6.195 1.82 1.95 0.94

221524 s at RRAGD 0.0000002 -5.634 1.42 1.59 0.89

212935 at MCF2L 0.0000013 -5.195 1.1 1.27 0.87

202089 s at SLC39A6 0.0000046 -4.884 1.27 1.44 0.88 Non-

Parametric p- Tumor Tumor Fold-

ProbeSet Symbol value t-value (Mean) (Mean) change

200991 s at SNX17 0.0000064 -4.8 1.34 1.54 0.87

201 120 s at PGRMC1 0.0000098 -4.698 1.3 1.52 0.85

215978 x at ZNF721 0.0000106 4.674 1.78 1.69 1.05

200605 s at PRKAR1A 0.0000111 -4.662 1.73 1.88 0.92

201732 s at CLCN3 0.000012 -4.681 1.11 1.25 0.89

205158 at RNASE4 0.0000211 -4.495 1.47 1.65 0.89

209958 s at BBS9 0.0000329 -4.386 1.11 1.26 0.88

212206 s at H2AFV 0.0000396 -4.328 1.12 1.26 0.89

220113 x at POLR1B 0.0000511 4.26 1.55 1.49 1.04

215529 x at DIP2A 0.0000664 4.189 1.38 1.28 1.08

203249 at EZH1 0.0000988 -4.108 1.21 1.35 0.9

221997 s at MRPL52 0.0001043 4.065 1.31 1.23 1.06

215588 x at RIOK3 0.0001322 3.999 1.79 1.72 1.04

220071 x at HAUS2 0.0001322 3.999 1.78 1.72 1.04

220720 x at MZT2B 0.0001648 3.937 1.6 1.49 1.07

205367 at SH2B2 0.000218 3.857 1.6 1.54 1.04

220242 x at Z F701 0.000279 3.788 1.17 1.11 1.05

213105 s at C16orf42 0.000366 3.716 1.36 1.27 1.08

207688 s at INHBC 0.0003682 3.705 1.24 1.17 1.06

215067 x at PRDX2 0.0004099 3.674 1.55 1.46 1.06

207598 x at XRCC2 0.0004774 3.629 1.39 1.32 1.06

216495 x at IVD 0.000605 3.558 1.45 1.35 1.07

216310 at TAOK1 0.0006626 3.53 1.53 1.45 1.06

220796 x at SLC35E1 0.0009844 3.409 1.86 1.8 1.03

215383 x at SPG21 0.0015874 3.26 1.57 1.5 1.05

206792 x at PDE4C 0.001972 3.19 1.96 1.93 1.02

219792 at AGMAT 0.0038212 2.972 1.21 1.16 1.04

221404 at IL1F6 0.0043026 2.935 1.17 1.11 1.05

215417 at EXOC6B 0.0045969 2.933 1.09 1.05 1.04

213383 at SBF1 0.0047047 2.901 1.25 1.19 1.05

217497 at TYMP 0.0061321 2.826 1.09 1.05 1.04

210934 at BLK 0.0063738 2.796 1.2 1.14 1.05

221034 s at TEX13B 0.0077024 2.73 1.2 1.14 1.05

211502 s at CDK14 0.0081815 2.726 1.09 1.05 1.04

215179 x at PGF 0.0089887 2.672 1.77 1.74 1.02

219105 x at ORC6L 0.0145461 2.494 1.2 1.15 1.04

217734 s at WDR6 0.0165359 2.444 1.55 1.49 1.04

217700 at CNPY4 0.0191607 2.388 1.17 1.12 1.05

207365 x at USP34 0.0264526 2.257 1.79 1.75 1.02

214627 at EPX 0.032782 2.172 1.23 1.17 1.04 Non-

Parametric p- Tumor Tumor Fold-

ProbeSet Symbol value t-value (Mean) (Mean) change

209246 at ABCF2 0.0362803 2.13 1.16 1.1 1 1.05

214041 x at RPL37A 0.0450646 2.033 1.81 1.74 1.04

213125 at OLFML2B 0.0681089 1.848 1.12 1.09 1.03

212044 s at RPL27A 0.0762717 1.794 1.58 1.51 1.04

221474 at MYL12B 0.2770767 1.094 1.98 1.97 1

221943 x at RPL38 0.5045344 0.67 1.88 1.87 1.01

References:

1. Spira, A., et al., Airway epithelial gene expression in the diagnostic evaluation of smokers with suspect lung cancer. Nat Med, 2007. 13(3): p. 361-6.

2. Gordon, G.J., et al., Translation of microarray data into clinically relevant cancer diagnostic tests using gene expression ratios in lung cancer and mesothelioma. Cancer Res, 2002. 62(17): p. 4963-4967.

3. Cerutti, J.M., et al., A preoperative diagnostic test that distinguishes benign from malignant thyroid carcinoma based on gene expression. J Clin Invest, 2004. 113(8): p. 1234-1242.

4. Bridgewater, J., et al., Gene expression profiling may improve diagnosis in patients with carcinoma of unknown primary. Br J Cancer, 2008. 98(8): p. 1425-30.

5. Landi, M.T., et al., Gene Expression Signature of Cigarette Smoking and Its Role in Lung Adenocarcinoma Development and Survival. PLoS ONE, 2008. 3(2): p. el 651.

6. Simon, R., et al., Analysis of Gene Expression Data Using BRB-Array Tools.

Cancer Inform, 2007. 3: p. 11-7.

7. Gentleman, R.C., et al., Bioconductor: open software development for computational biology and bioinformatics. Genome Biol, 2004. 5(10): p. R80.

8. Eisen, M.B., et al., Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci U S A, 1998. 95(25): p. 14863-14868.

9. Yeang, C.H., et al., Molecular classification of multiple tumor types.

Bioinformatics, 2001. 17 Suppl 1: p. S316-S322.

10. Vapnik, V., Statistical Learning Theory. 1998, New York: John Wiley. 11. Huang, D.W., B.T. Sherman, and R.A. Lempicki, Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat. Protocols, 2008. 4(1): p. 44-57.

12. Beane, J., et al., Reversible and permanent effects of tobacco smoke exposure on airway epithelial gene expression. Genome Biol, 2007. 8(9): p. R201.

It should be understood that for all numerical bounds describing some parameter in this application, such as "about," "at least," "less than," and "more than," the description also necessarily encompasses any range bounded by the recited values. Accordingly, for example, the description at least 1, 2, 3, 4, or 5 also describes, inter alia, the ranges 1-2, 1-3, 1-4, 1-5, 2-3, 2-4, 2-5, 3-4, 3-5, and 4-5, et cetera.

For all patents, applications, or other reference cited herein, such as non-patent literature and reference sequence information, it should be understood that it is incorporated by reference in its entirety for all purposes as well as for the proposition that is recited.

Where any conflict exits between a document incorporated by reference and the present application, this application will control. All information associated with reference gene sequences disclosed in this application, such as GenelDs or accession numbers, including, for example, genomic loci, genomic sequences, functional annotations, allelic variants, and reference mRNA (including, e.g. , exon boundaries or response elements) and protein sequences (such as conserved domain structures) are hereby incorporated by reference in their entirety.

While this invention has been particularly shown and described with references to example embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the scope of the invention encompassed by the appended claims.

Claims

CLAIMS What is claimed is:

1. A method of predicting the risk of lung cancer in a patient comprising:

a) testing an isolated sample from a patient who is a former cigarette smoker and ceased smoking within 10 years of isolation of the sample, wherein the sample is tested for an abnormal expression pattern of three or more genes selected from the group consisting of: ABCF2, AGMAT, BBS9, BLK, C16orf42, CAT, CDK14, CLCN3, CNPY4, DIP2A, ENPP4, EPX, EXOC6B, EZH1, H2AFV, HAUS2, IL1F6, INHBC, IVD, MCF2L, MRPL52,

MYL12B, MZT2B, OLFML2B, ORC6L, PDE4C, PGF, PGRMC1, POLR1B, PRDX2, PRKAR1A, RIOK3, RNASE4, RPL27A, RPL37A, RPL38, RRAGD, SBFl, SH2B2, SLC35E1, SLC39A6, SNX17, SPG21, TAOKl, TEX13B, TYMP, USP34, WDR6, XRCC2, ZNF701, and ZNF721,

wherein an abnormal expression pattern of the three or more genes predicts an increased risk of lung cancer in the patient.

2. The method of Claim 1 , wherein at least one of the three or more genes is selected from the group consisting of ABCF2, AGMAT, BBS9, BLK, C16orf42, CAT, CDK14, CNPY4, DIP2A, ENPP4, EPX, EXOC6B, EZH1, H2AFV, HAUS2, IL1F6, IVD, MCF2L, MYL12B, MZT2B, OLFML2B, ORC6L, PDE4C, PGRMC1, RPL27A, RRAGD, SBFl, SH2B2, SNX17, SPG21, TAOKl, TEX13B, TYMP, WDR6, XRCC2, ZNF701, and ZNF721.

3. A method of predicting the risk of lung cancer in a patient comprising:

a) testing an isolated sample from the patient, wherein the sample is tested for an abnormal expression pattern of three or more genes selected from the group consisting of: ABCF2, AGMAT, BBS9, BLK, C16orf42, CAT, CDK14, CLCN3, CNPY4, DIP2A, ENPP4, EPX, EXOC6B, EZH1, H2AFV, HAUS2, IL1F6, INHBC, IVD, MCF2L, MRPL52,

MYL12B, MZT2B, OLFML2B, ORC6L, PDE4C, PGF, PGRMC1, POLR1B, PRDX2, PRKAR1A, RIOK3, RNASE4, RPL27A, RPL37A, RPL38, RRAGD, SBFl, SH2B2, SLC35E1, SLC39A6, SNX17, SPG21, TAOK1, TEX13B, TYMP, USP34, WDR6, XRCC2, ZNF701, and ZNF721,

wherein an abnormal expression pattern of the three or more genes predicts an increased risk of lung cancer in the patient, wherein at least one of the at least three genes is selected from the group consisting of ABCF2, AGMAT, BBS9, BLK, CI 6orf42, CAT, CDK14, CNPY4, DIP2A, ENPP4, EPX, EXOC6B, EZHl, H2AFV, HAUS2, IL1F6, IVD, MCF2L, MYL12B, MZT2B, OLFML2B, ORC6L, PDE4C, PGRMCl, RPL27A, RRAGD, SBF1, SH2B2, SNX17, SPG21, TAOK1, TEX13B, TYMP, WDR6, XRCC2, ZNF701, and ZNF721.

4. The method of Claim 3, wherein the patient is a former cigarette smoker and ceased smoking within 10 years of isolation of the sample.

5. The method of any one of Claims 1-4, wherein the patient had a nondiagnostic bronchoscopy for lung cancer.

6. The method of any one of Claims 1-5, wherein the abnormal expression pattern of the three or more genes comprises increased expression of one or more of BBS9, CAT, CLCN3, ENPP4, EZHl, H2AFV, MCF2L, PGRMCl , PRKARIA, R ASE4, RRAGD, SLC39A6, or SNX17.

7. The method of any one of Claims 1-6, wherein the abnormal expression pattern of the three or more genes comprises decreased expression of one or more of ABCF2, AGMAT, BLK, C16orf42, CDK14, CNPY4, DIP2A, EPX, EXOC6B, FAM128B, HAUS2, IL1F6, INHBC, IVD, MRPL52, MYL12B, OLFML2B, ORC6L, PDE4C, PGF, POLR1B, PRDX2, RIOK3, RPL27A, RPL37A, RPL38, SBF1, SH2B2, SLC35E1, SPG21, TAOK1, TEX13B, TYMP, USP34, WDR6, XRCC2, ZNF701 or ZNF721.

8. The method of any one of Claims 1-5, wherein the abnormal expression pattern of the three or more genes comprises:

a) increased expression of one or more of BBS9, CAT, CLCN3 , ENPP4, EZH 1 , H2AFV, MCF2L, PGRMCl, PRKARIA, RNASE4, RRAGD, SLC39A6, or SNX17; and b) decreased expression of one or more of ABCF2, AGMAT, BLK, C16orf42, CDK14, CNPY4, DIP2A, EPX, EXOC6B, FAM128B, HAUS2, IL1F6, INHBC, IVD, MRPL52, MYL12B, OLFML2B, ORC6L, PDE4C, PGF, POLR1B, PRDX2, RIOK3, RPL27A, RPL37A, RPL38, SBFl , SH2B2, SLC35E1, SPG21, TAOKl, TEX13B, TYMP, USP34, WDR6, XRCC2, ZNF701 or ZNF721.

9. The method of any one of Claims 1-8, wherein at least one of the three or more genes is selected from CAT, ENPP4, RRAGD, MCF2L, SLC39A6, SNX17, PGRMCl, ZNF721 , PRKARIA, CLCN3, RNASE4, BBS9, H2AFV, POLR1B, DIP2A, EZH1,

MRPL52, HAUS2, RIOK3, MZT2B.

10. The method of Claim 9, wherein at least one of the three or more genes is selected from CAT, ENPP4, RRAGD, MCF2L, SNX17, PGRMCl, ZNF721, BBS9, H2AFV, DIP2A, EZH 1 , HAUS2, and MZT2B .

1 1. The method of any one of Claims 1-10, wherein the abnormal expression pattern of the three or more genes is measured at the protein level.

12. The method of any one of Claims 1-10, wherein the abnormal expression pattern of the three or more genes is measured at the nucleic acid level.

13. The method of any one of Claims 1-12, wherein the abnormal expression pattern of the three or more genes is measured simultaneously.

14. The method of any one of Claims 1-13, wherein the abnormal expression pattern of the three or more genes is measured simultaneously on a microarray.

15. The method of any one of Claims 1-14, wherein the abnormal expression of the three or more genes is measured at the nucleic acid level on a microarray and the microarray is a Exon 1.0 ST, Gene 1.0 ST, U 95, U133, U133A 2.0, or U133 Plus 2.0.

16. The method of any one of Claims 1-15, wherein the expression level of at least 5, 7, 10, 1 1, 12, 13, 14, 15, 16, 17, 18, 19, 20, 25, 30, 35, 40, 45, 50, or all 51 genes is determined.

17. The method of any one of Claims 1-16, wherein the expression levels are log- normalized.

18. The method of Claim 17, wherein the log-normalized expression levels are converted to percent-rank values.

19. The method of any one of Claims 1-18, further comprising the step of determining a disease index for the individual using the expression pattern of three or more genes.

20. The method of Claim 19, wherein the disease index is a support vector machine (SVM) index.

21. The method of Claim 20, wherein the SVM index is determined using SVM weights substantially similar to MCF2L (-1.4179), CLCN3 (-1.3756), EZH1 (-1.3011), TYMP (1.0144), CDK14 (0.8952), IVD (0.8668), EXOC6B (0.8071), ABCF2 (0.7865), SBF1 (0.7841), RRAGD (-0.7208), MRPL52 (0.7176), CAT (-0.6615), INHBC (0.6389), RIOK3 (0.6296), C16orf42 (0.5584), TEX13B (0.5532), RPL37A (-0.5478), SNX17 (- 0.4909), AGMAT (0.4859), BBS9 (-0.4511), WDR6 (0.4187), SPG21 (0.4007), MZT2B (- 0.3857), RPL38 (-0.3581), OLFML2B (-0.3557), H2AFV (0.348), ENPP4 (-0.3385), ZNF721 (0.2899), PRDX2 (0.2413), MYL12B (0.2217), RPL27A (-0.2036), ZNF701 (0.1925), SLC39A6 (-0.1778), EPX (-0.1525), PDE4C (-0.1521), DIP2A (0.152), IL1F6 (0.1517), POLR1B (0.1501), RNASE4 (0.1278), ORC6L (-0.1093), XRCC2 (-0.1071), PRKARIA (-0.0991), PGF (-0.0949), USP34 (-0.0769), PGRMCl (-0.0763), CNPY4 (0.0726), SLC35E1 (0.0713), TAOK1 (0.0602), SH2B2 (0.0529), BLK (-0.0373), and HAUS2 (-0.0301).

22. The method of any one of Claims 1-21, wherein the isolated sample comprises lung epithelial tissue.

23. The method of Claim 22, wherein the sample is obtained by bronchoscopy.

24. The method of any one of Claims 1-21, wherein the isolated sample comprises nasal epithelial tissue.

25. The method of any one of Claims 1-24, further comprising the step of one or more of non-invasive imaging, lung mass biopsy, or histological and/or cytological analysis of an isolated sample from the oral, esophageal, nasal, and/or pulmonary system of the subject.

26. A kit comprising reagents for detecting the expression level of three or more genes selected from the group consisting of: ABCF2, AGMAT, BBS9, BLK, C16orf42, CAT, CDK14, CLCN3, CNPY4, DIP2A, ENPP4, EPX, EXOC6B, EZHl, H2AFV, HAUS2, IL1F6, INHBC, IVD, MCF2L, MRPL52, MYL12B, MZT2B, OLFML2B, ORC6L, PDE4C, PGF, PGRMC1, POLR1B, PRDX2, PRKAR1A, RIOK3, RNASE4, RPL27A, RPL37A, RPL38, RRAGD, SBFl, SH2B2, SLC35E1, SLC39A6, SNX17, SPG21, TAOK1, TEX13B, TYMP, USP34, WDR6, XRCC2, ZNF701, and ZNF721, for detecting lung cancer in a patient; the kit optionally containing suitable positive controls.

27. A set of oligonucleotide primers for detecting the nucleic acid expression level three or more genes selected from the group consisting of: ABCF2, AGMAT, BBS9, BLK, C16orf42, CAT, CDK14, CLCN3, CNPY4, DIP2A, ENPP4, EPX, EXOC6B, EZHl, H2AFV, HAUS2, IL1F6, INHBC, IVD, MCF2L, MRPL52, MYL12B, MZT2B, OLFML2B, ORC6L, PDE4C, PGF, PGRMC1, POLR1B, PRDX2, PRKAR1A, RIOK3, RNASE4, RPL27A, RPL37A, RPL38, RRAGD, SBFl , SH2B2, SLC35E1, SLC39A6, SNX17, SPG21, TAOK1, TEX13B, TYMP, USP34, WDR6, XRCC2, ZNF701, and ZNF721, for detecting lung cancer in a patient.

28. A method of treating lung cancer, comprising administering suitable treatment and/or prophylaxis to a patient determined to have lung cancer by the method of any one of Claims 1-25.

29. A treatment and/or prophylaxis for lung cancer in a patient determined to have lung cancer by the method of any one of Claims 1-25.

30. Use of a treatment and/or prophylaxis for lung cancer in a patient determined to have lung cancer by the method of any one of Claims 1-25

31 The method of Claim 28, treatment and/or prophylaxis of Claim 29, or use of Claim 30, wherein the treatment and/or prophylaxis is selected from the group consisting of chemotherapy, hormonal therapy, immunotherapy, radiotherapy, surgery, and combinations thereof.