Comparison of transcriptomics technologies for assessment of Staphylococcus aureus gene expression

Saturday, May 10 • P 0196 POSTERS

D. Hernandez1, D.Baud 1, J. Schrenzel 1, P. François 1, A. Fischer 1, M. Girard 1, J.B. Veyrieras 2, C. Le Priol 2, G. Gervasi 2, F. Reynier 2 1Genomic Research Laboratory, Infectious Diseases Service, Geneva University Hospitals, Geneva, Switzerland, 2bioMérieux, Technology Research Department, Innovation and Systems Unit, Marcy l’Etoile, France.

Objectives: Today, high-throughput of transcriptomes (RNA-seq) in prokaryotes seems to be an interesting alternative to well- established transcriptomics technologies such as microarray. While this later technology provides an analogical quantification of gene transcription (via the fluorescent intensity measuring the amount of hybridization between capture probes and their complementary cDNA fragments), RNA-seq methods make it possible to obtain a comprehensive digital quantification of transcribed regions (by counting the number of sequenced reads that map onto the corresponding genomic regions). Furthermore, contrary to existing digital technologies like the NanoString nCounter platform (and contrary to microarrays too), RNA-seq approaches do not require the prior design of probes and can then be used to simultaneously determine the transcriptomic profile of prokaryote strains at both known and unknown transcribed regions (Baume, et al., 2010). Nevertheless, the analytical performance of RNA-seq approaches in prokaryotes has not yet been investigated. Here, we compared two RNA-seq solutions (Illumina MiSeq and Ion Torrent PGM) with Agilent microarrays and the NanoString nCounter system on Staphylococcus aureus total RNA samples. Methods: We extracted four total RNA samples from the Staphylococcus aureus strain NCTC 8325. Samples were obtained at 2h and 4h of growth from a wild-type strain, as well as from a GdpS mutant (Fischer, et al., 2013). Each sample was depleted from structural RNAs using the MicrobEnrich method (Ambion). The samples were then subjected to the different methods. RNA-seq data were mapped onto the reference sequence using BWA (Li and Durbin, 2009) and converted to gene counts using Bedtools (Quinlan and Hall, 2010) software applications. Statistical analysis was performed using the software R, with both the DESeq R package (Anders and Huber, 2010), as well as home-made scripts. Results: Both Illumina and Ion-Torrent RNA-seq experiments displayed an average variation coefficient of about 25% between individual replicates. However, at the gene level, the variation was strongly correlated with the individual coverage. Microarray and NanoString nCounter showed better reproducibility with Pearson correlation coefficients > 0.99. Conclusions: RNA-seq, which is likely to become the standard approach in prokaryote transcriptomics, require sufficient coverage for the results to be reliable. Since individual gene counts are not independent, highly expressed genes are detected at the expense of weakly covered genes for which reads counts may be insufficient for a reliable expression measurement. Both sequencing technologies are affected by sequence-related biases (such as %GC content), which may prevent comparison of expression levels between individual genes. However, the sequence bias is strongly correlated for each sequencing technology, which allows for differential expression measurements. The probe-based NanoString nCounter system provides the most accurate expression measure and remarkable correlation between replicates. However, it only allows querying for a number of 800 targeted genes corresponding to known genomic sequences.

27