Integrated mutation, copy number and expression profiling in resectable non-small cell lung cancer.

BACKGROUND: The aim of this study was to identify critical genes involved in non-small cell lung cancer (NSCLC) pathogenesis that may lead to a more complete understanding of this disease and identify novel molecular targets for use in the development of more effective therapies.
METHODS: Both transcriptional and genomic profiling were performed on 69 resected NSCLC specimens and results correlated with mutational analyses and clinical data to identify genetic alterations associated with groups of interest.
RESULTS: Combined analyses identified specific patterns of genetic alteration associated with adenocarcinoma vs. squamous differentiation; KRAS mutation; TP53 mutation, metastatic potential and disease recurrence and survival. Amplification of 3q was associated with mutations in TP53 in adenocarcinoma. A prognostic signature for disease recurrence, reflecting KRAS pathway activation, was validated in an independent test set.
CONCLUSIONS: These results may provide the first steps in identifying new predictive biomarkers and targets for novel therapies, thus improving outcomes for patients with this deadly disease.

Background

Non-small cell lung cancer (NSCLC) is the commonest cause of cancer death in Western communities. Current treatments offer the potential of cure only to the small number of patients who present with early stage NSCLC, whilst outcomes for those with advanced disease remain poor. Recent advances including adjuvant chemotherapy and targeted biological therapies have lead to modest improvements in survival for small subgroups of patients. Clearly new treatment approaches are required to substantially improve outcome. As has been the case for other tumour types, molecular profiling techniques have the potential to provide benefit through improved understanding of disease pathogenesis, identification of subgroups in whom current therapies are most likely to be effective and in the development of novel therapies.

Genetic heterogeneity is a feature of NSCLC, with varying combinations of multiple molecular alterations contributing to tumour development [1]. A key challenge for high-throughput molecular profiling techniques is to distinguish between genes whose expression is altered directly by heritable changes in gene function and those where changes are an inevitable down-stream consequence of primary changes to genes directly involved in disease pathogenesis. Correlation of transcriptional and genomic data allows more focussed analysis of the large number of genetic alterations identified by molecular profiling. This study incorporates the results of both transcriptional and genomic profiling for clinically relevant subgroups of NSCLC to identify genes of potential predictive or pathogenic importance in this deadly disease.

Methods

Samples

After obtaining institutional ethics approval, patients with stage I-IIIA NSCLC seen by the St Vincent's Hospital Combined Lung Service between February 2004 and July 2006 and planned for curative resection were invited to participate. Exclusion criteria included age <18 years, administration of neoadjuvant chemotherapy, and inability to provide informed consent. Integrated demographic, radiological, pathological and outcome data was collected for all consenting patients.

In addition, a small number of samples collected earlier and stored in the Peter MacCallum (PeterMac) tissue bank were utilised after approval by the PeterMac Tissue Management Committee.

Microarray analyses

Samples of tumour (≥1 cm 3) were selected from fresh specimens, and then stored whole at -180°C. Only those specimens containing >75% tumour cells and <25% necrosis were used in molecular studies. Both RNA and DNA were isolated from each sample for analysis using established protocols (see additional files 1 and 2). Transcriptional profiling was performed using 10,500 element cDNA microarrays (PeterMac, Melbourne, Australia) [2]. Genomic profiling using 2400 element bacterial artificial chromosome (BAC) arrays was completed at the University of San Francisco, California, USA [3]. Detailed description of transcriptional and genomic profiling is included in additional files 3 and 4.

Mutation analyses

All samples were tested for TP53 mutations and all adenocarcinoma (AC) and large cell carcinoma (LCC) samples were screened for KRAS mutations using high resolution melting analysis [4,5] with or without DNA sequencing.

Bioinformatic analyses

The effect of histology, presence or absence of KRAS or TP53 mutation, tumour size and metastasis status, recurrence within 12 months of surgery and survival on gene expression was explored. After removing control genes, analysis was conducted in CRAN, R Bioconductor using the LIMMA [6,7] package to generate detailed lists of gene expression differences with significance p values between each subgroup of interest. To account for multiple testing, p values of <0.005 were considered statistically significant. Gene lists were then interrogated using publicly available programs (Intelligent Systems and Bioinformatics Laboratory, Wayne State University, Detroit, MI, USA, http://vortex.cs.wayne.edu/projects.htm) to identify gene ontology and molecular pathway patterns.

Changes in the normalized and smoothed genomic data [8] were assigned stepwise copy change levels from -2 to 3 (-2 = homozygous loss, -1 = heterozygous loss, 1 = single copy gain, 2 = gain of two copies, 3 = high-level gain, 0 = normal copy number). Using these standardised copy number values, it was possible to make comparisons of the frequency of each level of copy change at each BAC location for specific groups of interest.

To compare our data with that of other groups of patients with early stage NSCLC, we performed comparisons with publicly available external data sets obtained via the NCBI Gene Expression Omnibus (GEO) website. Different platforms were reconciled using HUGO approved gene symbols, and extracted gene expression data were log2 transformed, centred and scaled across samples in order to emphasise relative expressions as opposed to absolute values.

Integration of transcriptional and genomic profiles

Integration of transcriptional and genomic datasets was performed by investigating levels of expression for genes located in regions of copy number variation between groups of interest. Genes whose differences in expression varied in the same direction as differences in copy number between two groups (e.g. relative over-expression of genes in a region of increased copy number) were viewed as genes of interest.

Results

Molecular and clinical data was available for 69 patients who underwent surgery for NSCLC between May 1999 and July 2006. Demographic and pathologic data are included in table 1. Median follow-up for surviving patients exceeds 35 months. After a median follow-up of 36 months (1 - 80) for all patients, 28/68 patients developed recurrent disease, and 23/68 patients died of NSCLC (one patient with disseminated disease at diagnosis excluded from analysis). Comparable to other series of early stage NSCLC [9], five year overall survival rates approximated 55%.

Table 1

Demographic and pathologic details for 69 NSCLC patients

Variable (number of patients assessed)	Number	Percentage (95% CI)
Gender (69)
Male	42	61 (49-72)
Female	27	39 (28-51)
Age at definitive treatment (years) (69)
Median (range)	70 (29 - 85)
Histology (69)
AC	30	43 (32-55)
SCC	23	33 (23-45)
LCC	16	23 (14-34)
Smoking status (69)
Never	16	23 (14-34)
Current/Ex	53	77 (66-86)
Stage (69)
IA	16	23 (14-34)
IB	26	38 (27-49)
IIA	3	4 (1-11)
IIB	7	10 (5-19)
IIIA	8	12 (6-20)
IIIB	3	4 (1-11)
IV	5*	12 (6-20)
Unknown	1#	1 (0-7)
Primary tumour size (mm) (69)
0 - 20	23	33 (23-45)
21 - 30	10	15 (8-24)
31 - 50	22	32 (22-43)
51 - 70	5	7 (3 - 16)
>70	7	10 (5 - 20)
Unknown	2^	3 (0-9)
T < 40 mm, N1 &/or M1 (69)	13	19 (11-29)
T > 40 mm, N0M0 (69)	11	16 (9-26)
T > 70 mm, N0M0 (69)	4	6 (2-13)
K-Ras mutation (46)**	10	21 (11-34)
EGFR mutation (46)**	5	11 (4-22)
P53 mutation (69)	26 +/- 17
Definite	26	38 (27-49)
Possible	17	25 (16-36)
Recurrence (68)+	28	41 (30-53)
Death from NSCLC (68)+	23	34 (23-45)
Time from surgery to death (months) (23)
Median (range)	19 (7 - 73)

AC = adenocarcinoma; LCC = large cell carcinoma; SCC = squamous cell carcinoma

*3 patients had solitary cerebral metastases resected at diagnosis, 1 patient had 2 separate lung lesions (unclear if synchronous primary lesions or pulmonary metastasis), 1 patient had disseminated disease

# Stage unable to be determined in 1 patient after wedge resection without nodal dissection.

^ Tumour size unable to be assessed in 1 patient who underwent incomplete tumour resection after combined chemoradiation, and one who had incomplete resection of an obstructing stage IIIA tumour followed by definitive radiotherapy.

** Only AC and LCC samples were tested for mutations of K-Ras and EGFR.

+One patient with disseminated disease at diagnosis excluded from recurrence and survival analysis

Genomic Analysis

Aneuploidy

All samples demonstrated significant chromosomal instability with an average of 43.6 chromosomal breakpoints per sample (defined by a change in the stepwise copy number along a chromosome), with over 100 and 150 regions of high-level gain (+3) and loss (-3) respectively. There was also a very high rate of low-level genomic alteration (both gains and losses). On average, over 10 whole arm losses or duplications were seen per sample, with a rate of isochromosome formation of 1.6 per genome (duplication of one arm with loss of the opposing arm of the same chromosome). These results are consistent with the highly disordered nature of lung cancer genomes. Comparisons between clinical subgroups of interest revealed remarkably similar degrees of aneuploidy and chromosomal disorder in almost all groups (Table 2). Specifically, neither prognosis, degree of histologic differentiation, K-Ras or TP53 status was associated with evidence of greater aneuploidy.

Table 2

Comparisons of average number of breakpoints for NSCLC subgroups

Subgroup	Ave. No.Breakpoints	Ratio of comparison (p value)
Histotype
AC	45.7	ACC:SCC = 1.1 (1.0)
SCC	43.1
LCC	39.5
Gender
Male (M)	44.9	M:F = 1.1 (1.0)
Female (F)	41.2
Smoking
Smokers (S)	44.5	S:N = 1.1 (1.0)
Never smokers (N)	41.1
Recurrence
Recurrent (R)	44.7	R:NR = 1.0 (1.0)
Non-recurrent (NR)	43.3
Survival
Non-survivors (NS)	46.5	NS:S = 1.1 (0.9)
Survivors (S)	42.8
Metastasis
Metastatic (M)	40.0	M:NM = 0.9 (0.9)
Non-metastatic (NM)	46.8
EGFR status
Mutant (m)	38.6	m:wt = 0.9 (0.9)
Wild-type (wt)	44.6
K-Ras status
Mutant (m)	42.3	m:wt = 1.0 (1.0)
Wild-type (wt)	43.9
p53 status
Mutant (m)	46.8	m:wt = 1.1 (0.9)
Wild-type (wt)	40.8

Histotype comparisons

Charts documenting the averaged standardised copy numbers at each BAC location were generated to enable visual comparisons of genomic profiles between AC (23) and SCC (17) (figure 1). Regions of shared change (amplification of 5p15.33-p13.13, loss of 3p26.3-p13) and several regions of difference (SCC: gains at 3q12.1-q28 and 12p13.33-p12.1, losses of chromosome 4 (4pter-4qter); AC: gains of 6p25.1-p21.31, losses of 6q13-27, and a greater magnitude of change in several regions of shared loss (9p, 13q, 17p, 18q, 19p, 19q and 22q) or gain (7p, 7q, 8q)) were identified. These findings are consistent with previous studies [9-16].

Figure 1

Plot of average standardised copy number at each BAC for AC and SCC samples. X axis: BAC location along genome. Y axis: average standardised copy number.

Associations with mutation status in TP53, KRAS and EGFR pathways

Clinical data demonstrated a trend to greater rate of TP53 mutation in SCC than AC (TP53 mutation in 9/12 (75%) SCC and 9/18 (50%) AC, p = 0.083). Amplification of 3q was also more frequent in SCC than AC samples (p = 0.004). When analysing all samples, no relationship was found between 3q amplification and TP53 mutation (p = 0.99). However, when analysing SCC and AC separately, a statistically significant relationship between TP53 mutation and 3q amplification was detected in AC samples, with amplification of 3q being significantly more common in TP53 mutant cancers (1/8 and 4/10 samples with 3q amplification in TP53 wild-type (wt) and mutant AC groups respectively, p = 0.027). Both TP53 mutant and wt samples more frequently demonstrated copy number loss at the TP53 locus (17p13) than gain (11/18 and 9/12 samples in TP53 mutant and wt groups respectively).

Adenocarcinomas were screened for EGFR and KRAS mutations. Small numbers of EGFR mutant tumours limited detailed analysis. Good quality genomic profiles were available for only 5 tumours with EGFR mutation and no significant differences were seen between the profiles of EGFR mutant and non-mutant tumours.

Genomic profiles were available for 8 KRAS mutant tumours, 31 wt and 16 tumours with unknown KRAS mutation status (untested SCC samples). In averaged copy number charts, KRAS mutant tumours showed a predilection for losses of 1p36.32-p13.2, 6q11.1-q27, 11p13-11q13.2 and 11q21-12p13.1 and gains of 1q21.1-q43 (Figure 2).

Figure 2

Chart of average standardised copy number at each BAC location for . X axis: BAC location along genome. Y axis: average standardised copy number.

Associations with metastasis, tumour recurrence and NSCLC-specific survival

To investigate the notion of inherent metastatic potential, molecular profiles of large (>4 cm) non-metastatic tumours and small (<2 cm) metastatic (nodal or distant) tumours were compared. Genomic profiles of 3 'metastatic' and 8 'non-metastatic' tumours revealed some differences in the magnitude of copy number changes, without regions of clear difference between the two groups. There was no correlation between genomic changes and tumour recurrence or survival. There were differences in the magnitude of gene copy number changes at 7p, 8q, 9p, 15q and 17p in recurrent compared to non-recurrent tumours. Contained within these regions are the MYC oncogene (8q), as well as TP53 (17p), and the CDKN2A locus (containing p14(ARF) and p16 tumour suppressor genes (9p) (TSG's)).

Transcriptional Analysis

Ranking the genes by moderated t-statistics and selecting a p value cut-off of <0.005, 310 genes with differential expression between 16 SCC and 25 AC samples were identified, representing the biological processes of cell adhesion, epidermis development, keratinisation and keratinocyte differentiation. A significant proportion of these genes had roles in antigen processing and presentation, and the phosphatidylinositol signalling pathway. Thirty of 310 genes in the differentiating gene list were located on chromosome 3 (p = 0.0098), implicating genomic changes at this locus in determining NSCLC phenotype. This is consistent with the genomic data, which indicates gain of 3q is associated with SCC histology.

Associations with mutation status in TP53, K-Ras and EGFR pathways

Expression levels of 67 genes differed significantly between TP53 mutant (17) and wt (21) tumours. Many of the biological functions represented by these genes were also strongly represented by the genes differentially expressed between AC and SCC. In addition, 20/67 discriminating genes were also included in the gene list differentiating SCC from AC. Hierarchical clustering based on the expression of these 67 differentially expressed genes not only segregated TP53 mutant from wild-type tumours, but also resulted in clustering of SCC samples with the TP53 mutant tumours. Our results suggest that the gene expression signature observed for TP53 mutant tumours may be at least in part related to SCC histology rather than TP53 biology.

Transcriptional profiles of AC and LCC tumours with (8) and without (31) KRAS mutation were compared. Biological processes represented by 108 differentiating genes included cell growth, second-messenger mediated signalling, chromosome organisation and biogenesis, and gene regulation (mediated via histones and their effects on biosynthesis and nucleosome assembly) (table 3). These findings are consistent with other published studies linking KRAS mutation to increased translation of cancer related proteins, and chromosome instability [17,18]. The low frequency of EGFR mutant cancers precluded statistically meaningful analysis of transcriptional data according to EGFR genotype.

Table 3

Details of genes over-expressed in K-Ras mutant tumours

Symbol	Gene Name	Cytoband	P value	Gene Function
ACOX2	Acyl-Coenzyme A oxidase 2, branched chain	3p14.3	4.9 E-3	Lipid metabolism
ALDH2	Aldehyde dehydrogenase 2 family (mitochondrial)	12q24.2	9.8 E-4	Alcohol metabolism
ALDH3B1	Aldehyde dehydrogenase 3 family, member B1	11q13	3.9 E-3	Alcohol metabolism
AREG	Amphiregulin (schwannoma-derived growth factor)	4q13-q21	1.4 E-4	Autocrine growth factor family member
CD55	CD55 molecule, decay accelerating factor for complement (Cromer blood group)	1q32	6.9 E-4	Immune response, protection of cells from complement mediated damage
CLDN10	Claudin 10	13q31-q34	2.9 E-3	Intercellular tight junctions
DGKD	Diacylglycerol kinase, delta 130kDa	2q37.1	1.0 E-3	Intercellular signalling, cell growth.
EGR1	Early growth response 1	5q31.1	2.5 E-3	Transcriptional regulation of genes involved in mitogenesis
MSLN	Mesothelin	16p13.3	4.61 E-5	Possible role in cell adhesion
MST1R	Macrophage stimulating 1 receptor (c-met-related tyrosine kinase)	3p21.3	4.5 E-3	Cell motility, positive regulation of cell cycle
NPC2	Niemann-Pick disease, type C2	14q24.3	4.6 E-4	Regulation of cholesterol transport and storage
NR4A1	Nuclear receptor subfamily 4, group A, member 1	12q13	6.2 E-4	Transcription factor
PFDN1	Prefoldin subunit 1	5q31	1.5 E-3	Cell cycle, transcription factor activity, protein folding
PTGS2	Prostaglandin-endoperoxide synthase 2 (cyclo-oxygenase 2)	1q25.2-q25.3	3.1 E-3	Prostaglandin biosynthesis
ST3GAL5	ST3 beta-galactoside alpha-2,3-sialyltransferase 5	2p11.2	1.4 E-3	Cell differentiation, cell proliferation, signal transduction.
TFPI2	Tissue factor pathway inhibitor 2	7q22	4.5 E-4	Matrix remodelling, coagulation
UPP1	Uridine phosphorylase 1	7p12.3	1.0 E-4	Nucleotide catabolism

Transcriptional profiles identified 39 genes that differentiated between 19 'metastatic' and 35 'non-metastatic' tumours, with molecular pathways involved in protein translation most strongly represented (MRPL33, RPL12, RPL27A, RPS5, RPS9). Comparison of expression profiles of 14 tumours recurring within 12 months of surgery to remaining samples identified 60 genes with differential expression between the two groups, with a common theme of RAS activation represented in ontological and single gene analyses. Included in the differentiating gene list were MAPK1, DUSP11 and DUSP13, PTPN11, and PIK3CB, all having roles in signal transduction and the MAPK pathway. The phosphatidylinositol signalling pathway was also significantly over-represented in ontology analysis.

Expression levels of only 38 genes differed significantly between deceased and surviving patients. 18 of these genes were shared with gene lists of recurrent vs. non-recurrent tumours. Few biological processes were represented by more than one gene, and clear patterns of gene ontology were not apparent.

Correlation with External Data Sets

Comparison of our differential gene list for recurrence with the discriminating gene list for survival in GSE11117 (transcriptional and survival data for 41 NSCLC samples, using Novachip Human 34.5k microarray interrogating ~34,500 transcripts for each sample; http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE11117) identified 40 matched transcripts (additional file 5). Log-transformed expression values of the 40 transcripts were used to classify the samples from GSE11117 into two subgroups using a correlation, average-linkage hierarchical clustering R package. Kaplan-Meier curves for these external samples using the 40 transcripts matched to our recurrence gene list demonstrated statistically significant survival prediction (figure 3, p < 0.0153)), with 21 and 20 samples in each group.

Figure 3

Survival curves based on recurrence differential gene list. A. Recurrence free survival curve for our dataset grouped by recurrence differential gene list. B: Kaplan Meier survival curve for external dataset GSE11117 using 40 transcripts matched to our recurrence differential gene list to classify.

Integration of Genomic and Transcriptional Profiles

To determine whether integration of genomic and expression data added to the predictive value of these datasets for histologic classification, we identified 34 genes whose copy number and expression varied between AC and SCC, 24 of which demonstrated concordant differences in copy number and expression (table 4). Notably, 17 of 24 differentiating genes were located on chromosome 3q.

Table 4

Details of genes located in regions of genomic difference with concordant differences in expression levels between AC and SCC

Symbol	Gene Name	Cytoband	Difference in expression (log2 SCC-AC)	P value
TP73L	Tumour protein 73-like	3q28	2.94	1.69 E-10
CSTA	Cystatin A (Stefin A)	3q21	2.27	3.42 E-8
ATP1B3	ATPase, Na+/K+ transporting, beta 3 polypeptide	3q23	1.46	1.06 E-6
VNN2	Vanin 2	6q23-q24	1.34	1.00 E-4
ABCC5	ATP-binding cassette, subfamily C, member 5	3q27	1.28	2.2 E-4
UPK1B	Uroplakin 1B	3q13.3-q21	1.12	3.3 E-3
RBP1	Retinol binding protein 1	3q23	1.12	1.98 E-3
SGK	Serum/glucocorticoid regulated kinase	6q23	1.06	1.3 E-4
TNFSF10/TRAIL	Tumour necrosis factor (ligand) superfamily, member 10/TNF-related apoptosis-inducing ligand	3q26	1.05	1.7 E-4
TNFAIP3	Tumour necrosis factor, alpha-induced protein 33	6q23	0.84	5.0 E-4
PFN2	Profilin 2	3q25.1-25.2	0.83	8.1 E-3
PLOD2	Procollagen-lysine, 2-oxoglutarate 5-dioxygenase 2	3q23-q24	0.80	6.7 E-3
B3GNT5	UDP-GlcNAc:betaGal beta-1,3-N-acetylglucosaminyltransferase 5	3q28	0.80	2.8 E-3
EIF4A2	Eukaryotic translation initiation factor 4A, isoform 2	3q28	0.66	4.1 E-3
MAP3K4	Mitogen activated protein kinase kinase kinase 4	6q26	0.59	1.8 E-3
PDCD10	Programmed cell death 10	3q26.1	0.58	2.9 E-3
SEP1	5'-3' exoribonuclease 1	3q25-q26.1	0.54	8.2 E-3
NCK1	NCK adaptor protein	3q21	0.52	7.3 E-3
GPR126	G-protein coupled receptor 126	6q24.1	0.52	1.7 E-3
LSAMP	Limbic system-associated membrane protein	3q13.2-q21	0.49	3.4 E-3
DZIP3	Zinc finger DAZ interacting protein	3q13.13	0.45	4.9 E-3
PIK3A2	Phosphoinositide-3-kinase, catalytic, alpha	3q26.3	0.44	9.2 E-3
WASF-1	WAS protein family, member 1	6q21-q22	0.44	8.3 E-3
CITED2/MRG1	CBP/P300-interacting transactivator, with GLU/ASP rich carboxy-terminal domain 2/Melanocyte specific gene 1 related gene 1	6q23.3	-0.63	8.0 E-3

Associations with mutation status

Examination of transcriptional data from 1p, 1q, 6q, 11p, 11q and 12p (regions of genomic difference) identified 25 genes with concordant differences in copy number and expression between K-Ras mutant and wt tumours (table 5). A number of genes demonstrated reduced copy number and expression in KRAS mutant tumours, including putative tumour suppressor genes (FOXO3, EXTL2, PPP2R1B), negative regulators of the receptor tyrosine kinase oncogenic pathways (PTPRK, DGKZ, NCAM1), and negative regulators of Ras (EPHB2). Several over-expressed genes located in regions of amplification play roles in constitutive KRAS activation (PTGS2/COX2), enhanced transactivation of the EGFR (RGS2), enhanced invasive potential (ECM1), and MAPK/ERK activation (PTGS2/COX2).

Table 5

Details of genes with differential copy number and expression between K-Ras mutant and wild-type tumours

Symbol	Gene name	Cytoband	Difference in expression(log2 mutant - wild type)	P value
RGS2	Regulator of G-protein signaling 2, 24kDa	1q31	1.99	5.9 E-3
NCAM1	Neural cell adhesion molecule 1	11 ( multiple clusters)	-1.63	1.5 E-3
CD55	CD55 molecule, decay accelerating factor for complement (Cromer blood group)	1q32	1.49	6.9 E-4
PTGS2/COX-2	Prostaglandin-endoperoxide synthase 2 (prostaglandin G/H synthase and cyclooxygenase)/Cyclooxygenase-2	1q25.2-q25.3	1.32	3.1 E-3
ECM1	Extracellular matrix protein 1	1q21	1.21	5.9 E-3
EXTL2	Exostoses (multiple)-like 2	1p21	-1.08	4.2 E-3
HIST2H2BE	Histone cluster 2, H2be	1q21-q23	1.07	2.7 E-3
H2AA	Histone 2	1q21	0.95	1.2 E-3
CHEK1	CHK1 checkpoint homolog (S. pombe)	11q24-q24	-0.94	6.1 E-3
GPX7	Glutathione peroxidase 7	1p32	-0.88	5.6 E-3
RWDD2A	RWD domain containing 2A	6q14.2	-0.80	8.7 E-4
HSF2	Heat shock transcription factor 2	6q22.31	-0.78	1.6 E-3
BLR1	Burkitt lymphomas receptor 1, GTP binding protein (chemokine (CXC motif) receptor 5)	11q23.3	-0.69	7.4 E-3
MIZF	MBD2-interacting zinc finger	11q23.3	-0.69	5.9 E-3
KPNA5	Karyopherin alpha 5 (importin alpha 6)	6q22.2	-0.66	8.7 E-3
LMO4	LIM domain only 4	1p22.3	-0.66	3.2 E-3
MUTYH	MutY homolog (E. coli)	1p34.3-p32.1	-0.65	4.9 E-4
REV3L	REV3-like, catalytic subunit of DNA polymerase zeta (yeast)	6q21	-0.65	8.8 E-3
EPHB2	Ephrin B2	1p36.1-p35	-0.64	6.7 E-3
DGKZ	Diacylglycerol kinase, zeta 104kDa	11p11.2	-0.64	9.2 E-3
TSPYL4	TSPY-like 4	6q22.1	-0.63	1.6 E-3
DBT	Dihydrolipoamide branched chain transacylase E2	1p31	-0.58	5.8 E-3
FOXO3	Forkhead box O3	6q21	-0.53	5.7 E-3
PPP2R1B	Protein phosphatase 2, regulatory subunit A, beta isoform	11q23.2	-0.51	2.7 E-3
PTPRK	Protein tyrosine phosphatase, receptor type, kappa	6q22.2-q22.3	-0.08	4.3 E-4

Investigation of genomic and transcriptional data identified only 2 genes (ARFGEF1 and PENK) whose copy number and expression differentiated 'metastatic' from 'non-metastatic' tumours, neither of which have been previously implicated in malignancy. Similarly, correlation of transcriptional and genomic data identified only 3 genes with concordant differences in recurrence and survival comparisons - SMARCA2, MINK and RECK.

Discussion

The clinical, demographic and pathologic characteristics of this NSCLC cohort are consistent with the published literature. The transcriptional and genomic profiles identified in this study should therefore be generalisable to other patients with early-stage NSCLC. The tumour samples analysed demonstrated substantial genomic instability, with comparisons between subgroups failing to demonstrate any significant difference. Previous studies of copy number changes in NSCLC have found no association between age, gender, histology, stage or tumour grade and the degree of genomic instability [19-21]. The absence of difference in the degree of genomic abnormalities between KRAS mutant and wt tumours is interesting, as both our transcriptional and genomic data imply enhanced activity of genes involved in chromosome structure and organisation in KRAS mutant tumours. We recognise that there were a small number of KRAS mutant tumours available for comparison and this may have limited our analysis.

Consistent with previously reported studies [9-16], the major differences in copy number and gene expression profiles between AC and SCC of the lung involved chromosome 3q. The strong independent correlation with amplification and over-expression at this locus suggests a causal relationship in SCC for genes in this region which warrant further investigation. These include TP73L, a gene extensively implicated in SCC, whose expression was most strongly correlated with the SCC phenotype, and which has been previously reported to be a putative oncogene [22-32]. While the role of TP73L in squamous cell pathogenesis remains unclear, a recent study of SiRNA mediated TP73L inhibition in SCC resulted in reduced cell survival with maintenance of squamous characteristics [28]. These results suggest that TP73L is important in SCC cell survival. Other genes previously shown to be over-expressed in SCC were included in our differentiating list (CSTA [33,34], FGFBP1), and warrant functional validation.

The differential copy number and expression levels between AC and SCC of TNFSF10/TRAIL and ABCC5, which have roles in apoptosis and chemoresistance respectively, may have implications for treatment of NSCLC. Recently published clinical data [35] suggest there are histotype-specific differences in response to systemic therapies. Validation of the differential activity of the roles of these genes and sensitivity to conventional and novel chemotherapeutic agents may be an area for future research.

Published data on the relationship between TP53 mutations and histotype in NSCLC is conflicting [36,37]. SCC were associated with more frequent TP53 mutations than AC in our dataset. Cigarette smoking is a causal factor for both SCC phenotype and TP53 mutation [37]. However, we also observed a correlation between 3q amplification and TP53 mutations in AC samples. This suggests that the apparent association between TP53 mutations and SCC may be mediated by the relationship between TP53 mutations and amplification of regions of 3q. We caution that this study is underpowered to draw strong conclusions regarding the role of TP53 in NSCLC pathogenesis.

Tumours possessing mutations of KRAS express genes playing key roles in cell growth, chromosome organisation and gene regulation. As previously reported, we identified amplification and over-expression of COX2 in KRAS mutant tumours. KRAS mutant tumours did not demonstrate mutations in EGFR consistent with previous reports in both NSCLC and colon carcinoma which suggest that KRAS mutations predict resistance to EGFR antagonists [29,38-41]. Several reports link NCAM1 to Ras-dependent activation of ERK MAPK's [42,43]. Reduced copy number and expression of NCAM1 in tumours bearing KRAS mutations, as seen in our data, has not previously been reported. Further research into a KRAS mutation profile may yield simple and reliable immunohistochemical markers of KRAS mutation, thereby significantly reduce the cost of determining KRAS status in clinical practice.

The gene expression profile observed in 'metastatic' tumours is consistent with a growing body of literature implicating deregulated protein synthesis in the development and metastatic potential of human cancers [44,45]. Increased mRNA translation is a critical downstream function of many cancer related genes, and many gene products with roles in metastasis are not mutated but inappropriately expressed in malignant cells (e.g. VEGF, c-Myc, fos, Her2Neu, PDGF) [18]. Opportunities for therapeutic intervention currently in development include oncolytic viruses that require deregulated protein translation for their replication [18], or agents that inhibit mTOR, an integral factor in protein translation (eg. temsirolimus (CCI-779), everolimus (RAD001) and deforolimus (AP23573)).

While the small number of recurrences and deaths due to NSCLC in our tumour-set makes it difficult to draw strong conclusions, transcriptional profiles linked to tumour recurrence suggest KRAS pathway activation. This may be due to a higher proportion of AC and LCC vs. SCC in the 'recurrent' group. Other regions of copy number change demonstrate genomic gains in the region of c-Myc and losses in the region of p16 (INK4a, CDNK2A) in recurrent or non-survivor tumours, supporting a prognostic association of the Myc:CDNK2A ratio in NSCLC, as has been described in head and neck SCC [46]. Specific genes linked to recurrence or survival include SMARCA2 (implicated in the regulation of gene expression cell cycle control and oncogenesis), MINK (linked to the JNK MAP kinase pathway) [47] and RECK, which has putative roles in the suppression of tumour growth, invasion, angiogenesis and metastasis [48]. KRAS mutation has been associated with reduced expression of RECK in NSCLC [49], consistent with the clinical observation of poor outcome in patients with KRAS mutation bearing NSCLC. Activation of the Ras pathway may reduce RECK expression and thereby increase tumour recurrence. Importantly, our prognostic gene signature was validated in an independent test set, suggesting that these findings may eventually yield prognostic markers in resected early-stage NSCLC to better select patients for adjuvant treatments.

Conclusions

Several molecular alterations have been identified in association with NSCLC histotype, KRAS mutation, TP53 mutation, metastatic potential, disease recurrence and survival. Although the size of the current study is small, our findings are in many cases consistent with those of previous studies, and have been validated in the case of the prognostic classifier in an independent test set. In addition, several novel molecular changes associated with clinically relevant endpoints have been demonstrated. It is hoped that these results will contribute to identifying new predictive markers and targets for novel therapies to improve treatment selection and better outcomes for patients with this deadly disease.

Competing interests

The authors declare that they have no competing interests.

Authors' contributions

GN performed DNA and RNA extraction and transcriptional microarray studies, and drafted the manuscript. MC, SM and GW participated in the design of the study and co-ordinated collection of clinical data. AD performed and supervised mutation analyses. HD performed mutational analyses. JL and NT performed statistical analysis. KO performed pathology review. DT and MC conceived the idea of the study. DT supervised molecular studies and assisted in the preparation of the manuscript. All authors read and approved the final copy of the manuscript.

Pre-publication history

The pre-publication history for this paper can be accessed here:

http://www.biomedcentral.com/1471-2407/11/93/prepub

Additional file 1

Protocol for extraction of RNA

Click here for file

Additional file 2

Protocols for extraction of DNA

Click here for file

Additional file 3

Transcriptional profiling

Click here for file

Additional file 4

Genomic profiling

Click here for file

Additional file 5

40 matched transcripts between GSE11117 and our differential gene list for recurrence

Click here for file