Confounding factors in profiling of locus-specific human endogenous retrovirus (HERV) transcript signatures in primary T cells using multi-study-derived datasets

Abstract

Abstract Background Human endogenous retroviruses (HERV) are repetitive sequence elements and a substantial part of the human genome. Their role in development has been well documented and there is now mounting evidence that dysregulated HERV expression also contributes to various human diseases. While research on HERV elements has in the past been hampered by their high sequence similarity, advanced sequencing technology and analytical tools have empowered the field. For the first time, we are now able to undertake locus-specific HERV analysis, deciphering expression patterns, regulatory networks and biological functions of these elements. To do so, we inevitable rely on omics datasets available through the public domain. However, technical parameters inevitably differ, making inter-study analysis challenging. We here address the issue of confounding factors for profiling locus-specific HERV transcriptomes using datasets from multiple sources. Methods We collected RNAseq datasets of CD4 and CD8 primary T cells and extracted HERV expression profiles for 3220 elements, resembling most intact, near full-length proviruses. Looking at sequencing parameters and batch effects, we compared HERV signatures across datasets and determined permissive features for HERV expression analysis from multiple-source data. Results We could demonstrate that considering sequencing parameters, sequencing-depth is most influential on HERV signature outcome. Sequencing samples deeper broadens the spectrum of expressed HERV elements. Sequencing mode and read length are secondary parameters. Nevertheless, we find that HERV signatures from smaller RNAseq datasets do reliably reveal most abundantly expressed HERV elements. Overall, HERV signatures between samples and studies overlap substantially, indicating a robust HERV transcript signature in CD4 and CD8 T cells. Moreover, we find that measures of batch effect reduction are critical to uncover genic and HERV expression differences between cell types. After doing so, differences in the HERV transcriptome between ontologically closely related CD4 and CD8 T cells became apparent. Conclusion In our systematic approach to determine sequencing and analysis parameters for detection of locus-specific HERV expression, we provide evidence that analysis of RNAseq datasets from multiple studies can aid confidence of biological findings. When generating de novo HERV expression datasets we recommend increased sequence depth ( > = 100 mio reads) compared to standard genic transcriptome pipelines. Finally, batch effect reduction measures need to be implemented to allow for differential expression analysis.