An assay system for gene expression analysis of paraffin-embedded patient samples is predictive of lung cancer histologic type, according to results published in the Journal of Molecular Diagnostics.
More than 40 histologic types of lung cancer are known to exist. Treatment choice is heavily based on obtaining accurate assessment of this histology (eg, bevacizumab is not to be used for squamous cell carcinoma). The current standard method for determining histologic type is light microscope-based analysis of stained slides from a tumor. Therefore, observer variability can lead to inconclusive or inaccurate diagnosis. Theoretically, immunohistochemistry can be used to remove ambiguity associated with histologic determination; however, standard markers using this technique are not yet available. Gene expression has been shown to correlate to histologic type, but analyzing genes has traditionally required the use of DNA microarrays. While useful, these microarrays require fresh frozen samples to ensure quality of the extracted DNA. This necessity is not amenable to the clinical setting, where formalin-fixed, paraffin-embedded samples are routinely collected and stored.
In this report, researchers obtained paraffin-embedded samples of 442 lung cancer tumors in order to analyze gene expression using quantitative reverse-transcriptase PCR (RT-qPCR). This technique allows researchers to isolate, amplify, and measure the expression of RNA molecules in a sample. The authors mined a previous report characterizing genes that help delineate histologic types. This analysis led to a set of 57 genes to be quantified. Using RT-qPCR data, the researchers arrived at a histologic expression predictor that was able to discern adenocarcinoma, carcinoid, small-cell carcinoma, and squamous cell carcinoma. Slides from the same tumor samples were independently evaluated using traditional light microscopy by pathologists in order to assess accuracy.
The mean accuracy of the predictor was shown to be 84% when compared with pathologist evaluation. Reproducibility of the diagnosis was also established using Monte Carlo cross-validation (κ = 0.77). This observation suggested to the researchers that gene expression may be a viable means of predicting histologic type. Interestingly, the expression analysis of poor-quality samples (ie, those samples lacking cellularity) was not significantly worse, since the mean accuracy was 81%. Remarked the authors, “The implication of this result is a greater number of patients can be eligible for RT-qPCR histology prediction because not all patient tumor specimens have high tumor cellularity.”
The principle limitation of these data was the lack of validation in the prospective setting. In addition, pathologists were provided with one slide to determine type, which may have limited the ability to diagnose and could impact the results by making the relative accuracy of gene expression higher. In all, while gene expression is promising, it remains to be seen whether it will see future value in the clinic.