Expression signatures of metastatic capacity in a genetic mouse model of lung adenocarcinoma

Abstract

Background: Non-small cell lung cancer (NSCLC) is the foremost cause of cancer-related death in Western countries, which is due partly to the propensity of NSCLC cells to metastasize. The biologic basis for NSCLC metastasis is not well understood.

Methodology/principal findings: Here we addressed this deficiency by transcriptionally profiling tumors from a genetic mouse model of human lung adenocarcinoma that develops metastatic disease owing to the expression of K-ras(G12D) and p53(R172H). We identified 2,209 genes that were differentially expressed in distant metastases relative to matched lung tumors. Mining of publicly available data bases revealed this expression signature in a subset of NSCLC patients who had a poorer prognosis than those without the signature.

Conclusions/significance: These findings provide evidence that K-ras(G12D); p53(R172H) mice recapitulate features of human NSCLC metastasis and will provide a useful platform on which to study the biologic basis for lung adenocarcinoma metastasis and its prevention by novel agents.

Figure 1. A gene expression signature of spontaneous metastasis in a K-ras/p53 mutant mouse model.

(A) Gene expression profiles of tumor metastases were compared to the corresponding primary tumor to define the metastasis gene signature (P<0.01, paired t-test). Each row of the expression matrix represents a gene and each column represents a profiled sample; relative gene expression (metastasis: primary) is represented using a yellow–blue color scale. Genes defined as cell cycle-related by either the Whitfield signature or by Gene Ontology (GO) are indicated. (B) The expression patterns of the mouse model metastasis signature in a panel of human lung tumors from Bhattacharjee et al. . Tumors showing “activation” of the metastasis signature (as measured by the “met signature t-score”) tend to have high expression of the genes high in the mouse metastases and low expression of the genes low in the mouse metastases. (C) Kaplan-Meier analysis of the human lung tumors comparing the differences in risk between tumors showing activation (yellow line, t-score>0) and tumors showing deactivation (blue line, t-score<0) of the mouse model metastasis signature. Log rank test evaluates whether there are significant differences between the two arms. Univariate Cox test evaluates the association of the met signature t-score with patient outcome, treating the coefficient as a continuous variable. (D) Same as for part C, except that cell cycle-associated genes (as defined by either Whitfield et al or GO) were first removed from the mouse model metastasis signature prior to deriving the met signature t-score.

Figure 2. Verification of mRNA expression levels by Q-PCR.

Total RNA from spontaneous tumors for gene expression profiling was reverse-transcribed with the First-Strand cDNA Synthesis Kit (Amersham Bioscience). Real-time PCR reactions were prepared in duplicate in a 96- or 384-well clear optical reaction plate (Applied Biosystems), using SybrGreen master mix (Applied Biosystems), and run on an ABI PRISM 7900HT Sequence Detection System (Applied Biosystems). Normal lung tissue from wild-type mice was used for calibration. Glyceraldhyde-3-phosphate dehydrogenase (GAPDH) was used as the endogenous control. The relative level of a gene was determined by calculating ΔΔCt, based on the formula ΔΔCt = (sample Ct [gene]−sample Ct [GAPDH])−(normal lung Ct [gene]−normal lung Ct [GAPDH]).

Figure 3. The spontaneous mouse metastasis signature is associated with poor prognosis in human lung tumor profile datasets from the Director's Challenge Consortium.

Kaplan-Meier analysis comparing the differences in risk between human lung tumors showing activation (yellow line, t-score>0) and tumors showing deactivation (blue line, t-score<0) of the mouse model metastasis signature. Datasets from the study by Shedden et al. and represent four independent cohorts from (A) Memorial Sloan-Kettering Cancer Center (MSK), (B) Moffitt Cancer Center (HLM), (C) Dana-Farber Cancer Institute (CAN/DF), and (D) University of Michigan Cancer Center (MICH). (E) Kaplan-Meier analysis of tumors combined from all four datasets (N = 362).