3.3. DNA:RNA Immunoprecipitation (DRIP) and DRIP-Like Methods

The use of the S9.6 antibody to map DNA:RNA hybrids at specific loci was first developed by the Tollervey and the Proudfoot labs in yeast [44] and in human cells [27], respectively. A similar strategy was then implemented by the Chédin lab to obtain genome-wide maps of RNase H-sensitive R-loops in human cells [25]. The Chédin lab coined the now widely-used name DNA:RNA immunoprecipitation (DRIP) to describe this approach. The principle of DRIP is very simple: nucleic acids are extracted and sheared, and the DNA:RNA hybrids are immuno-precipitated using the S9.6 antibody. Pre-treatment of half the sample with exogenous RNase H validates the specificity of the immunoprecipitation, although this crucial control is not always shown (see below). When studying a small set of loci, the enrichment of DNA:RNA hybrids is classically estimated using quantitative polymerase chain reaction (DRIP-qPCR). To better demonstrate the presence of R-loops, a few studies have used reverse transcription qPCR (RT-qPCR) on DRIP material (DRIP-RT-qPCR) to definitely demonstrate that the immuno-precipitated material did indeed contain RNase H-sensitive RNA molecules [4,8]. At the genome-wide scale, various sequencing techniques have been implemented (Figure 1): DRIP-seq sequences the DNA present in the immuno-precipitated material [25]; for increased resolution, DRIPc-seq sequences that the RNA molecules present in the immuno-precipitated fraction [4,38]; ssDRIP-seq sequences the template strand hybridized to the R-loop RNA [8]; bisDRIP-seq combines S9.6 immunoprecipitation with bisulfite footprinting to map the R-loop-associated ssDNA [7]. DRIPc-seq and bisDRIP-seq map R-loops in a strand-specific manner at near nucleotide resolution. Although DRIP has considerably improved our understanding of R-loops in recent years, it suffers from an apparent lack of robustness between studies, which begs some questions.

Genome-wide mapping of DNA:RNA hybrids using DNA:RNA immunoprecipitation (DRIP)-like approaches.

After DNA extraction and shearing, the S9.6 monoclonal antibody is used to precipitate DNA:RNA hybrids. After purification, the precipitated DNA:RNA hybrids are sequenced. DRIP-seq ([25], not stranded) and ssDRIP-seq ([8], stranded) sequences the associated DNA. With these strategies, DNA:RNA hybrids are mapped with a resolution that is dependent on the size of the starting DNA fragments, which could be bigger than the R-loop. DRIPc-seq sequences the RNA moiety in the DNA:RNA hybrid in a strand-specific manner ([4,38]). Paired-end sequencing of the RNA ensures that the exact borders of the sequenced RNA are known. As a result, DRIPc-seq maps DNA:RNA hybrids in a strand-specific manner at near nucleotide resolution. Because it requires working with RNA, its implementation and the building of the sequencing libraries is technically more challenging than DRIP-seq. bisDRIP-seq [7] maps R-loops by combining the precipitation of the DNA:RNA hybrids with S9.6 with the bisulfite footprinting of the R-loop associated ssDNA. It is stranded (see text), but its resolution depends on the number of unmethylated cytosine residues that are present in the non-template strand and as a result it might under-estimate the size of R-loops (green line). However, with bisDRIP-seq, the R-loop associated ssDNA is modified before DNA extraction, which could limit the loss of unstable R-loops that might happen during DNA extraction. Of those four approaches, bisDRIP-seq is probably the hardest to implement and to analyze.

DRIP in different studies sometimes gives inconsistent results. For example, in S. cerevisiae, R-loops were shown by some [50] but not others [51] to accumulate at PMA1 when the DNA&RNA helicase Sen1 is deficient. The pattern of R-loop formation at the commonly-used gene model Actß varies also unexpectedly between studies: R-loops would be as likely to form within the first intron and the terminator of the gene for some [28,52], much more likely to form within the terminator for others [26], and finally far less likely for a fourth study [10]. Similarly, R-loops were shown by some to accumulate at terminator but not promoter regions when BRCA1 is deficient [10,52], whilst others detected R-loop accumulation particularly at promoter regions in those conditions [9]. In addition, DRIP enrichment for a given locus sometimes varies significantly from one study to the next or even within one study from one figure to another. At the genome-wide scale, there is also apparent variability in the number of R-loop forming regions that were identified using DRIP: for example, after DRIPc-seq, one study reported 8112 promoters forming R-loops forming regions in human embryonic carcinoma Ntera2 cells [25], whilst another DRIP study only identified 3257 R-loop forming promoters in human primary fibroblasts [26]. It is conceivable that at least some of these apparent discrepancies could be explained by the fact that cell types and hence transcriptional patterns often differ between studies, although there is evidence that patterns of R-loop formation are conserved between cell-types [4]. Moreover, different sequencing depths or peak calling algorithms could conceivably account for the inconsistencies in genome-wide DRIP studies. However, until it is firmly established that such parameters do indeed explain those differences, one is left feeling that, despite its very simple principle, DRIP is either not that straightforward to implement or not robust enough for its purpose. What could explain this apparent lack of robustness?

Many different DRIP protocols have been set up. Some of those protocols contain steps that are considered as very detrimental by others: for example some protocols include fixation and sonication steps [40], whilst both of these steps are to be absolutely avoided according to other protocols [36]. Although the aim of this review is not to discuss the different DRIP protocols, but rather to highlight the intrinsic strengths and weaknesses of the DRIP approach as compared to other R-loop mapping methods, it is likely that the different protocols and in particular the way the DNA is sheared and extracted could explain part of the variability that is observed. As explained below, other parameters could also contribute to this variability.

As discussed above, it is highly likely that the way the DNA is extracted has a very significant impact on the DRIP signal. In addition, the percentage of cells in S-phase in the starting material and the specificity of the antibody are also key factors to take into account to explain those discrepancies.

The cell-cycle profile. The recent observation that DNA replication is a modulator of R-loop accumulation [19,20] demonstrates that the cell-cycle profile of the starting material, and in particular, the proportion of cells in S-phase and in mitosis, will greatly influence the DRIP results. One should therefore either synchronize the samples of interest or endeavor to compare cultures with similar proportions of cells in S-phase and mitosis.

The specificity of the antibody. As mentioned above, although the greater affinity of the S9.6 antibody for DNA:RNA hybrids is very clear on in vitro substrates, several observations indicate that it can also recognize other structures when used on more complex material. For example, we recently showed that the known affinity of S9.6 for RNA:RNA hybrids, albeit weaker than its affinity of DNA:RNA hybrids, remained a confounding parameter for the accurate mapping of R-loops in fission yeast using DRIPc-seq [38]. In addition, the use of bisDRIP-seq has shown that S9.6 could efficiently immuno-precipitate single-stranded promoter regions that did not contain R-loops [7]. These observations strongly suggest that S9.6 is able to recognize a variety of nucleic acids that do not adopt the conformation of the ordinary B form of DNA. It is therefore conceivable that the way the genomic DNA is extracted will greatly influence the extent to which these other non-B forms of DNA are preserved, and hence the level of RNase H-resistant signal in DRIP experiments (Figure 2). This could at least partly explain the discrepancies between studies. DRIP enrichments might result from a combination of features recognized by S9.6 and might not necessarily only reflect the presence of R-loops. For example, the strong DRIP signal at the termination site (TES) of the TEFM gene in human fibroblasts is fully resistant to RNase H treatment [10]. This is why the RNase H-treated sample is such an important control, which cannot be replaced by complicated normalization methods as it is sometimes done. These observations strengthen the argument that the best way to demonstrate the presence of R-loops at a particular locus is to show the presence of strand-specific and RNase H-sensitive RNA using RNA-based methods, such as DRIP-RTqPCR or DRIPc-seq [4,8,38].

The S9.6 antibody is likely to recognize also ssDNA-containing non-B DNA forms that do not correspond to R-loops [7].

Scheme illustrating the fact that DRIP signals could result from R-loop and non R-loop structures and the importance of the RNase H treatment to accurately estimate DNA:RNA hybrid formation at a locus of interest.