Cancer-Testis Genes Are Coordinately Expressed and Are Markers of Poor Outcome in Non–Small Cell Lung Cancer

Cancer-testis gene expression is observed at different frequencies in all tumors regardless of tissue of origin. Of the 40 or so cancer-testis genes/gene families, more than half map to chromosome X (http://www.cancerimmunity.org/CTdatabase/). These X chromosome cancer-testis genes (CT-X) and BAGE genes that map to juxtacentromeric regions are typically multigene families that have arisen through chromosomal duplications. Most cancer-testis genes have been initially identified through immunologic assays (1). In normal tissues, CT-X genes are consistently expressed in spermatogonia, oogonia, and the trophoblast cells (1). In testicular germ cells, the expression of most CT-X genes decrease as cells enter meiosis (2), and regain genomic methylation (3, 4), coinciding with the loss of both Suv39h2 expression and H4 hyperacetylation (5, 6). Cancer-testis genes mapping to somatic chromosomes, on the other hand, tend to have fewer homologues and can be expressed in meiotic gametes (7–9). Based on these differences and the common epigenetic mechanisms associated with the regulation of their expression, it has been proposed that CT-X genes might constitute a distinct group (10).

Ectopic hypomethylation of genomic DNA has been associated with CT-X gene expression and the demethylation of critical CpG residues within their promoter regions (11). All CT-X genes that are expressed in tumors or testis can be induced in vitro by DNA demethylation or by inhibitors of histone deacetylation (1). Despite the evidence suggesting that CT-X genes share common epigenetic regulatory mechanisms, the evidence regarding CT-X coexpression and whether CT-X expression associates with clinical variables of disease and outcome is inconclusive.

In this study, we analyzed tumors from 523 non–small cell lung cancer (NSCLC) patients for the expression of nine CT-X genes. We show coordinate expression among all CT-X genes tested. In 447 patients for whom clinical data were available, we evaluated the association of CT-X expression with variables of disease and outcome. CT-X expression was found to be sporadically associated with advanced disease and other variables that indicate worse prognosis in all major histologic types of NSCLC. In addition, Cox regression analysis revealed that expression of NY-ESO-1 and MAGE-A3 were independent markers of worse outcome in adenocarcinomas of the lung. CT-X expression was not found to affect outcome in squamous carcinomas, bronchioloalveolar carcinoma, or adenocarcinoma with bronchioloalveolar features.

Patients. A total of 523 patients undergoing curative surgical resection for primary NSCLC at the Department of Cardio-Thoracic Surgery, Weill Medical College of Cornell University, from 1991 to July 2004, were included in this study. Informed consent was obtained from all patients. The study was approved by the Institutional Review Board of Weill Medical College of Cornell University.

Expression analysis. Tumor tissues were obtained during surgery. Following gross dissection, tissues were immediately frozen on dry ice. Total RNA was prepared following homogenization by the guanidium isothiocyanade method followed by CsCl gradient centrifugation. Alternatively, the Ribopure kit (Ambion, Austin, TX) was used according to manufacturer's instructions. Remaining tumor tissue was maintained either frozen or in RNAlater (Ambion). Total RNA (2 μg) was reverse-transcribed with 200 units Moloney murine leukemia virus reverse transcriptase (Invitrogen, Carlsbad, CA), according to the manufacturer's instructions, in the presence of 2 μg random hexamers (Applied Biosystems, Foster City, CA), 20 units RNaseOUT (Invitrogen), and 5 mmol/L DTT in a total volume of 20 μL. Each reverse transcription reaction was done in the presence or absence of Moloney murine leukemia virus-RTase to test for contamination. For individual PCR reactions, 250 ng of cDNA were amplified with gene-specific oligonucleotides (2 ng per 25 μL reaction) in the presence of 1 unit AmpliTaq Gold (Applied Biosystems) and 5 μmol/L of each deoxynucleotidetriphosphates (Applied Biosystems). Gene-specific primers used to amplify individual CT-X transcripts are shown in Supplementary Table S1. The integrity of cDNA obtained was tested by amplification of p53 transcripts (p53-F, 5′-TACTCCCCTGCCCTCAACAAG; p53-R, 5′-CTCAGGCGGCTCATAGGG). For semiquantitative PCR analysis, reverse transcription-PCR products were categorized after separation on ethidium bromide–stained agarose gels as either −, +/−, +, ++, or +++, reflecting the intensity of the product when compared with a standardized testis sample (Fig. 1). Real-time reverse transcription-PCR analysis (ABI Prism, Applied Biosciences) of a representative number of tumor RNA suggested the different intensities corresponded to ∼0.1-1 fg (+/−), 1-5 fg (+), 5-100 fg (++), and <100 fg (+++) of transcript per 2 μg of mRNA (12). Expression levels of <5 fg (−, +/−, +) were classified as low (CT-X_[low]), and those with >5 fg (++ and +++) as high (CT-X_[high]).

Statistical analysis. To analyze coordinate cancer-testis gene expression, the frequency of the expression of a given cancer-testis in a tumor sample expressing a second cancer-testis gene was compared with the frequency observed among all tumors. Additionally, expression data was analyzed as a categorical value where the strength of coexpression was represented by κ (13). The strength of coexpression is proportionate to κ, with higher values corresponding to more frequent coexpression. κs can range from −1.0 to 1.0. Values of >0 indicate agreement better than chance. For example, κ = 0.5 means that the expression of two given CT-X genes occurs simultaneously 50% of the time over and above that expected by chance alone (14).

To evaluate whether cancer-testis gene expression was related to sex, smoking status, tumor size, pleural invasion, disease stage, and other clinical variables, univariate analysis using the Wilcoxon rank sum test (or t test) and Fisher's exact test (or χ² test) were used for continuous and categorical variables, respectively. The effect of cancer-testis gene expression on survival was evaluated using the Kaplan-Meier method, and differences between two groups were compared using the log-rank test. All survival curves were calculated from the date of surgery. All statistical analyses were two sided with a 5% type I error rate and were computed using SAS (version 9.0) software (SAS Institute, Cary, NC). P < 0.05 was considered statistically significant. Covariates with P < 0.05 by univariate analysis were subjected to multivariate analysis. Cox regression analysis was done to assess the effects of CT-X expression on survival while controlling for confounding clinical covariates.

Frequency of X chromosome cancer-testis genes in non–small cell lung cancer. A total of 523 cases of NSCLC were typed for CT-X expression. The comparative level of CT-X expression was estimated by semiquantitative PCR and recorded as +/−, +, ++, and +++ based on the intensity of the amplification product (Fig. 1). Among tumors tested, 377 (72.1%) expressed at least one of the nine CT-X genes tested. The most frequently observed CT-X was MAGE-A3, present in 55.2% of samples followed by MAGE-A1 (46.3%), MAGE-A4 (34.7%), LAGE-1 (32.1%), MAGE-A10 (27.2%), NY-ESO-1 (26.6%), MAGE-C1/CT7 (18.8%), SSX4 (13.5%), and SSX2 (9.6%; Table 1). The expression frequency of CT-X genes showed striking differences between histologic subtypes. Sixty-two percent (59 of 95) of the bronchioloalveolar carcinoma cases (including adenocarcinomas with bronchioloalveolar features) expressed at least one CT-X gene. This frequency was 67% (148 of 221) for adenocarcinomas and 90% (87 of 97) for squamous cell carcinomas; the frequency in the latter group being significantly different than in that of the other two (P < 0.0001). This difference was preserved when the frequency of only tumors of the different histologic types expressing high levels of CT-X genes (CT-X_[high]) were compared.

X chromosome cancer-testis gene expression is coordinated. Of the NY-ESO-1 positive tumors, 64% also expressed LAGE-1, which is twice the frequency of 32% expected if the expression of these genes were independent events. Similarly, the possibility of finding any one of the nine CT-X genes expressed in a tumor with a second CT-X gene was consistently and significantly higher than expected (Table 2). Coexpression of CT-X genes was tested using κ, a statistical measure that indicates the degree of coexpression above that expected due to chance. When applied to NY-ESO-1 and LAGE-1 expression, κ statistics result in a value of 0.43, indicating coexpression is 43% above that expected by chance (P < 0.0001). Only 2 of the 36 CT-X combinations showed no statistically significant coexpression (SSX4 and LAGE-1, and SSX4 and CT7/MAGE-C1). However, these coexpression rates are still higher than expected, and their κ values are similar to those of the other CT-X combinations analyzed. Overall, our results conclusively show coordinate expression of all nine CT-X genes in NSCLC.

Associations of X chromosome cancer-testis expression with prognostic indicators in non–small cell lung cancer. The patient characteristics of this series are shown in Table 3. Univariate analyses confirmed associations between overall survival and tumor stage, nodal stage, metastasis stage, and pathologic stage. We then investigated the possible correlation between CT-X expression and these clinical variables (Table 4). “Any CT-X” included patients whose tumors expressed at least one CT-X gene. In addition, tumors with high levels of CT-X expression (CT-X_[high]), as judged by semiquantitative PCR, were compared with tumors with no or low cancer-testis expression (CT-X_[low]). Clinical associations of CT-X expression were studied either as a group (Any CT-X), or individually. CT-X expression generally correlated with larger tumor size, presence of pleural invasion, a lack of ground glass opacity, and advanced-stage disease, by univariate analysis. Any CT-X expression was found to be associated with male sex (P = 0.02) and advanced tumor stage (P = 0.04). In addition, high level expression of any CT-X gene (Any-CT_[high]) associated with male sex (P = 0.0009), smoking history (P = 0.003), larger tumors (>3 cm; P = 0.0005), tumor stage (P = 0.004), the presence of pleural invasion (P = 0.009), and the absence of ground glass opacity (P = 0.014). Expression of individual cancer-testis genes showed similar trends (Supplementary Table S2).

When the patients were stratified according to histology similar associations were observed. These correlations included those listed above, but in addition, CT-X expression was found to be associated with later nodal and pathologic stages in all histologic groups. Statistically significant associations are shown in Supplementary Table S3. CT-X gene expression consistently associated with advanced disease or indicators of worse outcome except for SSX2 expression in adenocarcinomas, which was more frequent among nonsmokers.

X chromosome cancer-testis gene expression is an independent prognostic indicator in non–small cell lung cancer. The expression of CT-X was then compared with patient survival in various histologic groups. Kaplan-Meier analysis showed that CT-X expression conferred worse survival primarily in patients with adenocarcinomas (Table 5; Fig. 2). “Any CT-X_[high]” patients had worse survival when compared with “Any CT-X_[low]” patients with adenocarcinoma (P = 0.06). CT-X expression had no effect on the survival of patients with bronchioloalveolar carcinoma, adenocarcinoma with bronchioloalveolar characteristics, or for squamous cell carcinoma. When individually analyzed, the expression of NY-ESO-1, MAGE-A1, SSX2, and high levels of MAGE-A3 was found to be associated with shorter survival in adenocarcinoma. When stratified according to stage, CT-X expression was associated with worse survival at all stages. Examples for MAGE-A3, NY-ESO-1, and SSX2 are shown in Fig. 3.

Cox regression analysis was done to assess whether CT-X expression was prognostic of survival independent of confounding criteria, including stage, histology, and adjuvant therapy (Table 5). This revealed that high level expression of NY-ESO-1 or MAGE-A3 were predictors of worse outcome in adenocarcinoma independent of confounding factors.

The evaluation a large series of patients has made it possible to perform a rigorous analysis of CT-X expression and also to conduct correlative studies using multivariate analysis with greater reliability. We show that all the CT-X genes examined are expressed in a coordinate manner in NSCLC. Previous studies have argued both for and against coexpression (15–18). Such discrepancies might have been due to smaller sample sizes or due to the heterogeneous expression of CT-X genes within tumors, which might lead to false negative results in as many as 30% of samples, especially in immunohistochemical analyses (19). To avoid such caveats, this study was conducted with a very large sample size. We chose reverse transcription-PCR–based analysis due to its high sensitivity and because we could prepare total RNA from samples that were large enough to reflect CT-X gene expression in case the genes were heterogeneously expressed in the tumor tissues evaluated.

The best documented epigenetic aberrations in tumors, regardless of tissue of origin, are genome-wide hypomethylation, region-specific hypermethylation, and loss of imprinting (20–23). Ectopic methylation has been associated with larger tumors, extensive disease, and ultimately poor prognosis (24–30). Current data suggests that the transcriptional regulation of all CT-X genes is governed by common epigenetic mechanisms (1). CT-X genes have in common a TATA-less promoter, which is heavily methylated and thus silent in normal tissues. Their expression is associated with genome wide hypomethylation, and they can be up-regulated by DNA methyl transferase inhibitors, histone deacetylase inhibitors, or by the absence of histone methyl transferase G9a (11, 31–35). Genome-wide hypomethylation has been shown to increase progressively in parallel to advanced grade in breast, ovarian, cervical, and neural cancers (26–30) and has been associated with lung cancer progression (36). Therefore, it is not surprising that CT-X gene expression has also been reported to be associated with less-differentiated, higher-grade tumors; later stages of cancer; and worse outcome (37–50). In fact, MAGE-A3, MAGE-A10, and MAGE-A1 promoters were shown to undergo progressive demethylation in gastric cancer parallel to disease progression (43).

We found significant correlations between CT-X expression and larger tumors, pleural invasion, later stages of disease, and also male sex and a history of smoking. Moreover, multivariate analysis showed CT-X expression to be a marker of worse outcome independent of the known clinical prognostic indicators. We observed that high-level CT-X expression was even more closely related to later stages and worse disease outcome. This is similar to previous reports (37, 50, 51) and is likely to reflect more severe hypomethylation in these cancers. However, we have not tested for genome-wide hypomethylation in tumor tissues and therefore can not conclude whether these are indeed parallel events. Although our observations are limited to CT-X genes, the expression of two other cancer-testis genes, SCP-1 on chromosome 1 and PRAME that maps to chromosome 22, have been associated with worse survival in ovarian cancers and neuroblastoma, respectively, suggesting that cancer-testis genes located on somatic chromosomes might also show similar clinical associations (52, 53). Although CT-X expression could be a passive bystander in a more hypomethylated genome, recent evidence suggests these genes can have antiapoptotic qualities, thus conferring a survival advantage to some tumors (54, 55).

Although the expression of a number of CT-X genes did not reach the statistical significance to qualify as markers of worse outcome in this study, hazard ratios for most CT-X genes in adenocarcinoma or squamous cell carcinoma were >1, suggesting an association between the expression of almost all CT-X genes with worse prognosis.

In this series, squamous cell carcinomas were found to express all CT-X genes at significantly higher rates then adenocarcinoma and bronchioloalveolar carcinomas. In other studies, CT-X expression frequency has also been shown to be higher in squamous cell carcinomas of the lung, bladder, and esophagus compared with tumors of other histologic types in the same organs (50, 56, 57). Whether this directly reflects the degree of hypomethylation in these tumors is not known; however, squamous and spindle carcinoma of the skin have been shown to have more extensive hypomethylation compared with papilloma (58). The fact that expression of CT-X genes should associate with particular tumor types suggests that the underlying mechanisms controlling CT-X gene expression are complex and vary among tumor types. In this context, both aberrant CpG hypermethylation and imprinting defects have been shown to be tumor type specific (23, 59).

Most CT-X antigens are immunogenic, and their use as therapeutic cancer vaccines is being systematically evaluated (60–63). The DNA methyltransferase inhibitor 5-aza-2′ deoxycytidine (Desitabine) and various histone deacetylase inhibitors, including trichostain A, SAHA, and valproic acid, have been used as part of chemotherapy protocols, primarily to reverse hypermethylation of tumor suppressor gene promoters, and have been shown to be beneficial in hematologic and solid tumors (64–67). These drugs, if capable of inducing CT-X expression in vivo, could increase patient eligibility and treatment effectiveness in CT-X-targeted immunotherapy. Our results suggest that as tumors progress, the level and number of CT-X genes they express are likely to increase. This in turn suggests a potentially important therapeutic role for CT-X-based cancer vaccines in the management of later stages of malignancy. Because there is strong coexpression among CT-X genes, a multi-CT-X targeting strategy would be predicted to be most effective. Moreover, it could be argued that CT-X antigen immunization might be beneficial even to patients with CT-X -negative tumors, as an induced anti-CT-X immune response might prevent the emergence of CT-X expressing tumor cells, thus halting or slowing tumor progression.