Bayesian modeling of reproducibility and robustness of RNA reverse transcription and quantitative real-time polymerase chain reaction

Abstract

Gene expression measurements with quantitative real-time polymerase chain reaction (PCR) have a major hindrance of not directly measuring RNA and, thus, needing a reverse transcription (RT) step to first produce complementary DNA (cDNA). However, few studies have assessed the robustness of the RT step, and almost none has compared the factual performance of RT and real-time PCR enzymes. Here, we examined the variability of RT and PCR reactions and compared enzyme reproducibility by reverse transcribing identical RNA with eight RT enzymes and then amplifying the cDNA produced by one of them with six real-time PCR enzymes. The same four reference genes were measured in both experiments, and the data were analyzed with Bayesian multilevel models. Reproducibility and the efficiency of the RT enzymes, excluding one, varied moderately, but RT was always less precise than PCR; four PCR enzymes performed in an excellent manner. The transcription efficiencies of two of the measured reference genes (Actb and Sdha) lacked covariance with the general RT efficiency and showed poor reproducibility. In conclusion, most variation in quantitative real-time RT–PCR (RT–qPCR) emanates from the RT phase, and gene-related factors seem to be the primary determinants of this variation, discouraging the use of control genes in RT–qPCR normalization without prior information of their RT robustness.