Functional identification of cancer-specific methylation of CDO1, HOXA9, and TAC1 for the diagnosis of lung cancer

Abstract

Purpose: Non-small cell lung cancer (NSCLC) is the leading cause of cancer mortality in the world. Novel diagnostic biomarkers may augment both existing NSCLC screening methods as well as molecular diagnostic tests of surgical specimens to more accurately stratify and stage candidates for adjuvant chemotherapy. Hypermethylation of CpG islands is a common and important alteration in the transition from normal tissue to cancer.

Experimental design: Following previously validated methods for the discovery of cancer-specific hypermethylation changes, we treated eight NSCLC cell lines with the hypomethylating agent deoxyazacitidine or trichostatin A. We validated the findings using a large publicly available database and two independent cohorts of primary samples.

Results: We identified >300 candidate genes. Using The Cancer Genome Atlas (TCGA) and extensive filtering to refine our candidate genes for the greatest ability to distinguish tumor from normal, we define a three-gene panel, CDO1, HOXA9, and TAC1, which we subsequently validate in two independent cohorts of primary NSCLC samples. This three-gene panel is 100% specific, showing no methylation in 75 TCGA normal and seven primary normal samples and is 83% to 99% sensitive for NSCLC depending on the cohort.

Conclusion: This degree of sensitivity and specificity may be of high value to diagnose the earliest stages of NSCLC. Addition of this three-gene panel to other previously validated methylation biomarkers holds great promise in both early diagnosis and molecular staging of NSCLC.

Figure 1. Cancer Specific DNA Methylation Discriminates NSCLC Tumors from Normal Lung Samples

Methylation data are derived from 636 NSCLC in the Cancer Genome Atlas representing 227 lung squamous carcinomas with 43 matched normal samples and 409 lung non-squamous carcinomas with 32 matched normal samples. Columns represent tumor or normal tissue samples. Rows represent individual methylation probes from the Infinium methylation array. The ability of each probe to discriminate tumor versus normal and an associated t-statistic was estimated by a linear model for each CpG island promoter probe. Only probes with significant p-values are included in the heatmap. Rows are ordered from top-to-bottom by p-value. All p-values are < 0.0001. Probes with mean Beta-values >0.2 in normal samples were excluded from the analysis. Of the 305 genes exhibiting DAC-specific upregulation, 63 genes represented by 172 methylation probes met the preceding criteria. Columns are ordered by unsupervised hierarchical clustering. A few tumors cluster with normal samples. This is consistent with prior TCGA analyses that demonstrate “normal-like” methylation patterns in a subset of tumors.

Figure 2. DNA Methylation of CDO1, HOXA9, and TAC1 is Highly Sensitive for NSCLC in the Cancer Genome Atlas

A single Infinium methylation probe with the best discriminative capacity between tumor and normal samples was selected for each of the 3 genes. A sample is considered methylated for a gene if its β-value was larger than three times the standard deviation of the mean of β-values of normal samples. Methylation of at least one gene-promoter among CDO1, HOXA9, and TAC1 by Infinium array identifies 98.9% of NSCLC cases in 636 cases in The Cancer Genome Atlas.

Figure 3. Validation of the Sensitivity of Methylation-Specific PCR for 3 Genes in Two Independent Cohorts

Methylation of at least one gene-promoter among CDO1, HOXA9, or TAC1 by methylation-specific PCR identifies 94.9% of NSCLC cases in 59-patient United States cohort A and 83.3% of NSCLC cases from the independent 30-patient Japanese cohort B.