Quality-controlled R-loop meta-analysis reveals the characteristics of R-loop consensus regions

Abstract

R-loops are three-stranded nucleic acid structures formed from the hybridization of RNA and DNA. While the pathological consequences of R-loops have been well-studied to date, the locations, classes, and dynamics of physiological R-loops remain poorly understood. R-loop mapping studies provide insight into R-loop dynamics, but their findings are challenging to generalize. This is due to the narrow biological scope of individual studies, the limitations of each mapping modality, and, in some cases, poor data quality. In this study, we reprocessed 810 R-loop mapping datasets from a wide array of biological conditions and mapping modalities. From this data resource, we developed an accurate R-loop data quality control method, and we reveal the extent of poor-quality data within previously published studies. We then identified a set of high-confidence R-loop mapping samples and used them to define consensus R-loop sites called ‘R-loop regions’ (RL regions). In the process, we identified a stark divergence between RL regions detected by S9.6 and dRNH-based mapping methods, particularly with respect to R-loop size, location, and colocalization with RNA binding factors. Taken together, this work provides a much-needed method to assess R-loop data quality and offers novel context regarding the differences between dRNH- and S9.6-based R-loop mapping approaches.

Exploration and analysis of R-loop mapping data with RLBase

“…The processed data stored in RLBase was generated using a long-running computational pipeline available in its entirety in the RLBase-data GitHub repository (see Code Availability ). The steps for upstream data processing were described in detail in our recent work ( 12 ). In addition to those details previously described, the present work will also describe the process of analyzing RNA-Sequencing datasets.…”

“…In cases where R-loop mapping studies also provided matched RNA sequencing (RNA-Seq) datasets, they were downloaded and quantified to assess gene expression with RLPipes via the following procedure: raw reads were downloaded and pre-processed using the same procedure described in our recent work ( 12 ) up until the genomic alignment step and then reads were quantified using the salmon pseudo aligner ( 16 ) to generate read counts.…”

“…This involved (i) RLFS analysis, (ii) quality model building, (iii) sample classification, (iv) R-loop consensus analysis, (v) peak annotation and enrichment testing, (vi) Expression matrix generation, (vii) R-loop region annotation, (viii) sample-level correlation analysis, (ix) R-loop region abundance matrix generation, (x) calculating R-loop/expression correlation, (xi) updating the RLBase genome browser trackhub, (xii) RLSeq analysis of every sample and (xiii) upload of all data to the RLBase Amazon Web Services (AWS) S3 bucket. Notably, many of these steps (i-v,viii) are described in detail in our previous work ( 12 ). Therefore, the present work will describe only those (vi,vii,ix-xiii), which have not been described elsewhere.…”

“…introduced the first R-loop mapping technique, DNA:RNA immunoprecipitation sequencing (DRIP-Sequencing) ( 11 ). Since that time, >40 R-loop mapping studies have been conducted, yielding hundreds of publicly available R-loop samples ( 12 ). In our recent work, we mined 810 public R-loop mapping datasets, building the largest standardized R-loop data resource to date ( 12 ).…”

“…Since that time, >40 R-loop mapping studies have been conducted, yielding hundreds of publicly available R-loop samples ( 12 ). In our recent work, we mined 810 public R-loop mapping datasets, building the largest standardized R-loop data resource to date ( 12 ). From this work, we defined consensus sites of R-loop formation and developed a framework for the analysis of R-loop data.…”