Strong Correlation of Gene Counts and Differentially Expressed Genes Between a 3' RNA-Seq and an RNA Hybridization Platform in Transcriptome Analyses from Canine Archival Tissues

Alexander F. H. Haake^1*Alina K. Loriani Fard^1*Vladimir M. Jovanovic²Sandro Andreotti³Achim D. Gruber^1*

• ¹Institute of Veterinary Pathology, Department of Veterinary Medicine, Freie Universität Berlin, Germany, Berlin, Germany
• ²Human Biology and Primate Evolution, Institute of Biology, Freie Universität Berlin, Germany, Berlin, Germany
• ³Department of Mathematics and Computer Science, Freie Universität Berlin, Bioinformatics Solution Center, Berlin, Germany, Berlin, Germany

Analyses of nucleic acids from archival tissues offer invaluable prospects for numerous fields of veterinary medicine, such as the study of differential gene expression in rare or historic diseases. The establishment of modern methodologies, however, raises questions regarding the comparability and reproducibility of data obtained from unlike tools. 3' RNA-Seq and direct RNA hybridization are such conceptually different approaches for high-throughput transcriptome analysis. Since both are applicable to short, partially degraded mRNA fragments, they in principle allow investigations of formalin-fixed, paraffin-embedded (FFPE) tissues that are abundantly available in pathology archives. Here, we compared the two methods in several relevant details using the RNA from the same set of 35 FFPE canine tumors as input, including sample-and gene-wise count levels, gene expression strengths and directions, as well as the overlaps of differentially expressed genes (DEGs). Both methods proved suitable for their use on archival tissues with moderately to very strong overall count correlations, as indicated by a range of Pearson and Spearman means between 0.66 and 0.87. Of note, the gene-wise count correlations depended on gene expression strength. In an entity-contrasting comparison, expression directions correlated very strongly ranging from 0.88 to 0.91, but DEGs overlapped only moderately with a Jaccard index of 0.53. Finally, we contrasted the different practically relevant aspects of the two technologies with their distinct advantages that depend on the objectives and design of the study. This comparison will guide and help to select the appropriate method and to validate and interpret the data obtained.