Quality assessment and control of tissue specific RNA-seq libraries of Drosophila transgenic RNAi models

Andreia J. Amaral, Francisco F. Brito, Tamar Chobanyan, Seiko Yoshikawa, Takakazu Yokokura, David Van Vactor, Margarida Gama-Carvalho

RNA-sequencing (RNA-seq) is rapidly emerging as the technology of choice for whole-transcriptome studies. However, RNA-seq is not a bias free technique. It requires large amounts of RNA and library preparation can introduce multiple artifacts, compounded by problems from later stages in the process. Nevertheless, RNA-seq is increasingly used in multiple studies, including the characterization of tissue-specific transcriptomes from invertebrate models of human disease. The generation of samples in this context is complex, involving the establishment of mutant strains and the delicate contamination prone process of dissecting the target tissue. Moreover, in order to achieve the required amount of RNA, multiple samples need to be pooled. Such datasets pose extra challenges due to the large variability that may occur between similar pools, mostly due to the presence of cells from surrounding tissues. Therefore, in addition to standard quality control of RNA-seq data, analytical procedures for control of "biological quality" are critical for successful comparison of gene expression profiles. In this study, the transcriptome of the central nervous system (CNS) of a Drosophila transgenic strain with neuronal-specific RNAi of an ubiquitous gene was profiled using RNA-seq. After observing the existence of an unusual variance in our dataset, we showed that the expression profile of a small panel of marker genes, including GAL4 under control of a tissue specific driver, can identify libraries with low levels of contamination from neighboring tissues, enabling the selection of a robust dataset for differential expression analysis. We further analyzed the potential of profiling a complex tissue to identify cell-type specific changes in response to target gene down-regulation. Finally, we showed that trimming 5' ends of reads decreases nucleotide frequency biases, increasing the coverage of protein coding genes with a potential positive impact in the incurrence of systematic technical errors.