Several DNA microarray based expression signatures for the different clinically relevant thyroid tumor entities have been described over the past few years. However, reproducibility of these signatures is generally low, mainly due to study biases, small sample sizes and the highly multivariate nature of microarrays. While there are new technologies available for a more accurate high throughput expression analysis, we show that there is still a lot of information to be gained from data deposited in public microarray databases. In this study we were aiming (1) to identify potential markers for papillary thyroid carcinomas through meta analysis of public microarray data and (2) to confirm these markers in an independent dataset using an independent technology.
We adopted a meta analysis approach for four publicly available microarray datasets on papillary thyroid carcinoma (PTC) nodules versus nodular goitre (NG) from N2-frozen tissue. The methodology included merging of datasets, bias removal using distance weighted discrimination (DWD), feature selection/inference statistics, classification/crossvalidation and gene set enrichment analysis (GSEA). External Validation was performed on an independent dataset using an independent technology, quantitative RT-PCR (RT-qPCR) in our laboratory.
From meta analysis we identified one gene (SERPINA1) which identifies papillary thyroid carcinoma against benign nodules with 99% accuracy (n = 99, sensitivity = 0.98, specificity = 1, PPV = 1, NPV = 0.98). In the independent validation data, which included not only PTC and NG, but all major histological thyroid entities plus a few variants, SERPINA1 was again markedly up regulated (36-fold, p = 1:3*10-10) in PTC and identification of papillary carcinoma was possible with 93% accuracy (n = 82, sensitivity = 1, specificity = 0.90, PPV = 0.76, NPV = 1). We also show that the extracellular matrix pathway is strongly activated in the meta analysis data, suggesting an important role of tumor-stroma interaction in the carcinogenesis of papillary thyroid carcinoma.
We show that valuable new information can be gained from meta analysis of existing microarray data deposited in public repositories. While single microarray studies rarely exhibit a sample number which allows robust feature selection, this can be achieved by combining published data using DWD. This approach is not only efficient, but also very cost-effective. Independent validation shows the validity of the results from this meta analysis and confirms SERPINA1 as a potent mRNA marker for PTC in a total (meta analysis plus validation) of 181 samples.