R-loop proximity proteomics identifies a role of DDX41 in transcription-associated genomic instability

Transcription poses a threat to genomic stability through the formation of R-loops that can obstruct progression of replication forks. R-loops are three-stranded nucleic acid structures formed by an RNA–DNA hybrid with a displaced non-template DNA strand. We developed RNA–DNA Proximity Proteomics to map the R-loop proximal proteome of human cells using quantitative mass spectrometry. We implicate different cellular proteins in R-loop regulation and identify a role of the tumor suppressor DDX41 in opposing R-loop and double strand DNA break accumulation in promoters. DDX41 is enriched in promoter regions in vivo, and can unwind RNA–DNA hybrids in vitro. R-loop accumulation upon loss of DDX41 is accompanied with replication stress, an increase in the formation of double strand DNA breaks and transcriptome changes associated with the inflammatory response. Germline loss-of-function mutations in DDX41 lead to predisposition to acute myeloid leukemia in adulthood. We propose that R-loop accumulation and genomic instability-associated inflammatory response may contribute to the development of familial AML with mutated DDX41.

T ranscription by RNA polymerase II (RNAPII) is essential to all cellular processes and hence to the adaptive response of cells to internal and external stimuli. Dysregulated transcription resulting in high transcription rates and increased frequency of transcription-replication conflicts is observed in many tumors. Accordingly, targeting different mechanisms that enable tumor cells to cope with transcription stress is being explored as a therapeutic strategy 1 . Co-transcriptional R-loops are three-stranded nucleic acid structures formed by an RNA-DNA hybrid with a displaced non-template DNA strand. R-loops are most prevalent at gene promoters where they regulate transcription initiation 2,3 . R-loop formation at CpG islands (CGIs) that are present at~60% of gene promoters protects these regions from DNA methylation by repelling DNA methyltransferases (DNMTs) 2,4 . In addition to reducing the binding of DNMTs, R-loop-dependent recruitment of GADD45A and active de-methylation by TET enzymes has been proposed as mechanism for loss of CGIs methylation 3 . Furthermore, R-loops are present at the 3′ end of genes where they facilitate transcription termination by stalling RNAPII downstream of the polyadenylation sequence 5,6 . R-loops are prevalent in highly transcribed genes and accumulate in repeats such as centromeres, telomeres, and retrotransposons [7][8][9][10][11] . Genome-wide approaches for mapping R-loops in human cells revealed that R-loops occupy up to 5% of unique sequences 12 . The presence of G-quadruplexes (G4s) on the displaced DNA contributes to the stabilization of R-loops 13 . Modification of RNA in the hybrid with N6methyladenosine (m6A) provides an additional layer of regulation through the recruitment of m6a reader proteins [14][15][16] . In addition to the regulatory functions of R-loops in transcription, DNA repair, telomere maintenance, and chromosome segregation, these non-B DNA structures can be drivers of genomic instability [17][18][19][20][21][22] . The single-stranded DNA in the R-loops is more prone to DNA damage 23,24 . Stalled transcription complexes at R-loops can trigger their processing by nucleotide excision repair endonucleases ERCC4 (also known as XPF) and ERCC5 (also known as XPG) into double-strand breaks (DSBs) 19 . In cycling cells, R-loops can be formed as a consequence of head-on transcription-replication conflicts [25][26][27] .
Different proteins regulate R-loop levels in human cells either by preventing their formation or by assisting their resolution. RNA-binding proteins that bind to nascent RNAs and are involved in the maturation or export of mRNA, such as the THO complex or the nuclear exosome, oppose R-loop formation 27,28 . Furthermore, negative supercoiling of DNA that is normally relaxed by Topoisomerase 1 (TOP1) favors R-loop formation 25,29 . Once formed, R-loops can be removed by the action of Ribonuclease H1 (RNaseH1) and H2 (RNaseH2)-conserved endonucleases that hydrolyze the phosphodiester backbone of the RNA moiety in RNA-DNA hybrids 30 . RNaseH1 acts as a monomer and harbors an N-terminal hybrid-binding domain (HBD) and a C-terminal catalytic domain. In vitro, the HBD (residues 27-76) displays at least 25-fold higher affinity for RNA-DNA hybrids as compared to dsRNA 31,32 . Recombinant GFP-tagged, catalytically inactive (D210N) human RNaseH1 was recently reported as a sensitive and specific tool for in situ imaging of RNA-DNA hybrids in fixed cells 33 . In addition to RNase H enzymes, different helicases including SETX, DDX5, and DDX39B have also been implicated in the unwinding of RNA-DNA hybrids and resolution of R-loops 5,18,34 .
Dead box helicase 41 (DDX41) is a tumor suppressor that is conserved in D. melanogaster, C. elegans, D. rerio, and plants, and is considered essential for cell growth and viability [35][36][37][38] . Somatic and germline mutations in DDX41 are present in 0.5 to 4% of adult myelodysplastic syndrome (MDS)/acute myeloid leukemia (AML) cohorts and are considered as oncogenic drivers 39 .
Pathogenic germline variants in DDX41 predominantly lead to frameshifts and production of truncated protein forms, whereas somatic mutations are mostly located within the DEAD box and helicase domain likely resulting in compromised helicase activity 40 . DDX41 has been reported to interact with components of the spliceosome, and DDX41 deletion or mutations led to splicing defects and faulty RNA processing 41 . The role of DDX41 in RNA processing appears to be conserved as the C. elegans orthologue, SACY-1, was recently shown to associate with the spliceosome, and to impact the transcriptome through splicingdependent and -independent mechanisms 36 . Despite the relevance of DDX41 in cancer, the cellular and molecular functions of DDX41 remain poorly understood.
We employed quantitative mass spectrometry (MS)-based proteomics to identify proteins that regulate R-loops in human cells. To this end, we developed RNA-DNA Proximity Proteomics (RDProx) that enables mapping of the R-loop-proximal proteome using the fusion protein of the hybrid-binding domain (HBD) of RNaseH1 and an engineered variant of ascorbate peroxidase (APEX2). We implicated proteins with different cellular functions in R-loop regulation and characterized the role of the tumor suppressor DDX41 in opposing R-loops and DSBs in promoters. We demonstrate that DDX41 preferentially associates with promoter regions and that recombinant DDX41 can bind and unwind RNA-DNA hybrids in vitro. We propose that the accumulation of co-transcriptional R-loops, and consequently replication stress, DSBs, and an inflammatory response may collectively contribute to the development of familial AML and MDS with mutated DDX41.

Results
RDProx identifies the R-loop proximal proteome. Tight regulation of R-loop levels across the genome is essential for their function in promoting chromatin-associated processes and for preventing R-loop-dependent genomic instability. To gain insights into protein-based mechanisms that regulate R-loop homeostasis, we probed the R-loop-proximal protein networks using RNA-DNA proximity proteomics (RDProx). We fused the HBD (residues 27-76) of RNaseH1 to an engineered variant of soybean ascorbate peroxidase (APEX2) 42 . As a negative control, we employed a construct harboring three point mutations in the HBD (HBD-WKK) that led to a loss of affinity towards RNA-DNA hybrids ( Supplementary Fig. 1a) 31 . In accordance, only GFP-tagged HBD and not the WKK mutant associated with chromatin under pre-extraction conditions ( Supplementary  Fig. 1b). The preferential binding behavior of the HBD for RNA-DNA hybrids was confirmed in vitro using purified domains and different nucleic acid substrates. As expected, the WKK mutant displayed dramatically reduced affinity for hybrids ( Supplementary Fig. 1c, d). APEX2-HBD or APEX2-HBD-WKK fusion proteins were expressed in HEK293T cells and labeling of proximal proteins was induced in vivo ( Fig. 1a and Supplementary Fig. 1e). Biotinylated proteins were enriched using streptavidin and analyzed by liquid chromatography (LC)-tandem mass spectrometry (MS/MS). Stable isotope labeling with amino acids in cell culture (SILAC) was used to distinguish proteins that are proximal to the wild-type HBD compared to the WKK mutant (Fig. 1a). We performed three replicate experiments that showed excellent reproducibility (r > 0.85) and identified 312 proteins enriched with high confidence by RDProx (log 2 FC > 2; FDR < 0.01) ( Fig. 1b and Supplementary Fig. 1f Fig. 1 RDProx-Mapping R-loop-proximal proteome on native chromatin. a Schematic representation of the RDProx workflow for identification of R-loopproximal proteins. HBD or HBD-WKK fused N-terminally to APEX2 were transiently expressed in light or heavy SILAC-labeled HEK293T cells. Biotinylation was induced upon the addition of 500 µM biotin-phenol for 2 h at 37°C and 1 mM H 2 O 2 for 2 min at room temperature. Samples were pooled after cell lysis and biotinylated proteins purified using NeutrAvidin beads. Denatured proteins were separated by SDS-PAGE and in-gel digested before LC-MS/MS analysis. b Volcano plot of protein groups identified by RDProx in n = 3 biologically independent experiments. Mean log 2 ratios of all replicates between HBD and HBD-WKK are plotted against the −log 10 FDR. The FDR and enrichment were calculated using Limma 103 . Significantly enriched proteins are highlighted in blue (FDR < 0.01). Light blue indicates proteins in Tier 2 (300 proteins) above 2-fold change of the mean ratio and dark blue indicates proteins in Tier 1 (312 proteins) with a 4-fold change or higher. c Functional interaction network of proteins identified by RDProx. Genes were manually annotated based on literature and corresponding GO terms (Biological Process and Molecular Function). Clusters were generated based on the manual annotation. Edges between the nodes indicate interactions based on STRING with a confidence score equal or above 0.7. NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27530-y ARTICLE NATURE COMMUNICATIONS | (2021) 12:7314 | https://doi.org/10.1038/s41467-021-27530-y | www.nature.com/naturecommunications (EXOSC7, EXOSC10). R-loop proximal proteins showed functional interactions as demonstrated by the identification of different protein clusters involved in splicing, m 6 A regulation, mRNA 3′ end processing, mRNA export, transcription regulation, chromatin organization, and DNA replication/repair ( Fig. 1c and Supplementary Fig. 1h). These proteins were enriched with  domains typical for RNA-and DNA-binding proteins including  RRM, helicase, DEAD/DEAH, CID domain, MCM N-terminal  domain, RNA polymerase II binding domain, SAP domain, MCM OB domain, and K Homology domain ( Supplementary Fig. 1g).
DDX41 loss leads to replication stress and R-loop-dependent genomic instability. To assess the possible function of the identified DEAD box helicases in the regulation of R-loops, we monitored the intensity of Ser139 phosphorylation on the histone variant H2AX (γH2AX; proxy for DSBs) upon depletion of DDX27, DDX41, DDX42, DHX37, and DDX39A. Only depletion of DDX41 and AQR led to a notable increase in nuclear γH2AX intensity in unchallenged conditions (Fig. 2a). Dead box helicase 41 (DDX41) is a poorly characterized tumor suppressor and pathogenic variants in DDX41 cause familial MDS and AML, which prompted us to investigate its potential role in R-loop metabolism 41 . We confirmed that the proximity of DDX41 to R-loops is reduced when R-loops are suppressed by overexpression of RNaseH1 under the control of a doxycyclineinducible promoter ( Supplementary Fig. 2a). Increased levels of γH2AX in DDX41 knockdown cells were confirmed by western blotting (Supplementary Fig. 2b, c). Overexpression of RNaseH1 partially rescued the effect of DDX41 knockdown on γH2AX pointing to R-loop-dependent genomic instability (Fig. 2b). Overexpression of RNaseH1 also rescued the increased formation of DSBs in DDX41 knockdown cells measured by neutral comet assay (Fig. 2c). Increased formation of DSBs in DDX41 knockdown cells was confirmed by monitoring TP53BP1 (also known as 53BP1) foci formation (Fig. 2d, Supplementary Fig. 2d). Knockdown of DDX41 resulted in increased phosphorylation of RPA on Ser33 and significantly reduced DNA fiber length similarly to mild replication stress induced by DNA polymerase inhibition with 100 nM aphidicolin (Fig. 2e, f). Accordingly, an increase in γH2AX and pRPA intensity was predominant in the S phase ( Supplementary Fig. 2e, f). Moreover, the phosphorylation of RPA was, to some extent, mediated by the replication stress kinase ATR, since ATR inhibition with VE-821 significantly reduced the increase in pRPA intensity after DDX41 knockdown ( Supplementary Fig. 2g). To corroborate that these cells depend on ATR activity to respond to replication stress, we treated U2OS cells after DDX41 knockdown, and OCI-AML3 cells expressing DDX41 disease variants (L237F/P238T and R525H), with ATR inhibitors and monitored their viability. Both DDX41 knockdown and disease variants-expressing cells displayed sensitivity to ATR inhibition suggesting that AML cells with pathogenic DDX41 variants also display replication stress ( Supplementary Fig. 2h, i).
DDX41 unwinds RNA-DNA hybrids in vitro and its loss results in R-loop accumulation. We confirmed that DDX41 is proximal to R-loops in human cells using proximity ligation assays with antibodies against endogenous DDX41 and GFPtagged HBD or HBD-WKK ( Fig. 3a and Supplementary Fig. 3a). The proximity to R-loops and the occurrence of spontaneous DNA damage and replication stress in DDX41 knockdown cells, prompted us to investigate whether DDX41 opposes the accumulation of R-loops. To test this, we performed dot-blot analysis with the S9.6 RNA-DNA hybrid antibody. Indeed, knockdown of DDX41 resulted in the accumulation of RNA-DNA hybrids that were sensitive to RNaseH1 overexpression (Fig. 3b). Using chromatin-bound GFP-tagged HBD as a proxy for R-loops, we could confirm increased levels of R-loops upon depletion of DDX41 and AQR as a positive control ( Fig. 3c and Supplementary Fig. 3b, c). Inhibition of transcription by treating cells with DRB partially rescued the effect of DDX41 knockdown on R-loop accumulation ( Fig. 3c and Supplementary Fig. 3b). As a consequence of inhibiting transcription elongation, DRB also induces R-loop formation, which could explain the incomplete rescue of DRB on R-loop accumulation in DDX41 knockdown cells 43 .
Our experiments indicated that DDX41 opposes the accumulation of R-loops and R-loop-dependent genomic instability. To address whether DDX41 can directly bind and unwind RNA-DNA hybrids in R-loops, we purified full-length DDX41, DDX41 lacking the helicase domain (153-410), and the AMLassociated R525H variant in the C-terminus of the RecA-like helicase core domain ( Supplementary Fig. 3d, e). Full-length DDX41 was incubated with five different fluorophore-conjugated oligonucleotide substrates to determine the binding affinity. Binding of DDX41 to the substrates resulted in a change of fluorescence polarization. We found that recombinant DDX41 possesses the strongest affinity (K d = 2.5 µM ± 1.4 µM) for RNA-DNA hybrids in vitro compared to other nucleic acid substrates (Fig. 3d). The lack of known K d values for other RNA-DNA helicases make it difficult to draw comparisons, but we note that HBD's affinity for RNA-DNA hybrids is~10× higher (K d = 190 nM ± 30) ( Supplementary Fig. 1c). Furthermore, we employed an ATPase assay to determine whether the ATPase domain of DDX41 hydrolyzes ATP when encountering an RNA-DNA hybrid substrate. We found that that ATP hydrolysis by DDX41 was stimulated by RNA-DNA hybrids with a singlestranded DNA overhang ( Supplementary Fig. 3f). To test whether the ATPase activity was accompanied by the unwinding of this substrate, we established a FRET-based displacement assay. The separation of a fluorophore-labeled DNA strand and a quencherconjugated RNA strand resulted in increased fluorescence intensity. Importantly, DDX41 not only bound but also unwound the RNA-DNA hybrids in vitro in a concentration-dependent manner, whereas DDX41 lacking the helicase domain was not able to separate the two strands (Fig. 3e, f). Also, DDX41 harboring the disease-associated R525H variant showed decreased efficiency in RNA-DNA hybrid unwinding ( Fig. 3f and Supplementary Fig. 3h). Taken together, recombinant full-length DDX41 preferentially binds RNA-DNA hybrids compared to other nucleic acids and can unwind RNA-DNA hybrids in vitro.

DDX41 opposes R-loop accumulation in promoters.
To test whether DDX41 can associate with chromatin in vivo, we generated U2OS cells that express GFP-tagged DDX41 under a doxycycline-inducible promoter. We confirmed that these cells show pan-nuclear DDX41 staining that mirrored the localization of endogenous DDX41 ( Supplementary Fig. 4a). We used a GFP nanobody to target micrococcal nuclease to chromatin regions bound by DDX41 using greenCUT&RUN 44,45 . We detected 19,327 DDX41 peaks in 2 biologically independent experiments, among which 6,363 were consistently found in both experiments (Supplementary Data 2). Interestingly, DDX41 displayed a preference to bind promoters with 41% of DDX41 peaks mapping to promoter regions (TSS ± 3 kb) (Fig. 4a, b). We identified 6,441 promoters bound by DDX41, and a comparison of DDX41 binding sites with RNA-sequencing from U2OS cells revealed that DDX41 association with promoters depends on gene expression levels with the association being stronger at highly expressed genes ( Supplementary Fig. 4b).
To quantify R-loops genome-wide in wild-type U2OS cells and upon knockdown of DDX41, we performed MapR that uses a catalytically-dead E. coli Ribonuclease H to target micrococcal nuclease to R-loops, which are subsequently cleaved, released, and identified by sequencing 46,47 . 47% of R-loops were identified in promoter regions (TSS ± 3 kb) and the levels of R-loops in promoters positively correlated with gene expression (Fig. 4c and Supplementary Fig. 4c, d, Supplementary Data 3, 4). Inhibition of transcription by Actinomycin D resulted in a dramatic decrease of R-loops in promoters ( Fig. 4c and Supplementary Fig. 4d). Importantly, knockdown of DDX41 led to a significant increase of R-loops in promoters ( Fig. 4c and Supplementary Fig. 4d, e). This was also reflected by the increased number of MapR peaks detected in DDX41 knockdown cells in comparison to wild-type . Reactome pathway over-representation analysis of genes with accumulated R-loops revealed chromatin organization, NOTCH, and TGFβ signaling (Fig. 4g). Nearly 40% of promoters that displayed accumulation of R-loops also associated with GFP-tagged DDX41 ( Fig. 4h and Supplementary Fig. 4f). This is likely an underestimation since CUT&RUN was performed under very mild crosslinking conditions, and DDX41 loosely associates with chromatin. Accumulation of R-loops at promoter regions was not accompanied by a global defect in nascent transcription based on 5-Ethynyl Uridine (EU) incorporation ( Supplementary  Fig. 4g). However, we observed a mild but significant decrease in serine 5-and an increase in serine 2-phosphorylation of the RBP1 C-terminal domain (CTD), suggesting that RNAPII initiation and possibly elongation are perturbed by accumulated R-loops in promoter regions ( Supplementary Fig. 4h).
DDX41 loss increases DNA fragility in promoters and induces inflammatory response. To investigate whether the genomic instability observed upon loss of DDX41 derives from DSBs and whether the sites of DSBs coincide with sites of R-loop accumulation, we performed sBLISS (Break Labeling In Situ and Sequencing) in wild type and DDX41 knockdown HCT116 cells 48,49 . In accordance with previous studies [49][50][51][52] , the majority of endogenous DNA fragility hotspots in unchallenged cells were mapped to promoters and this phenotype was even more pronounced in DDX41 knockdown cells where 63% or 5,307 out of 8,381 of DNA fragility hotspots were mapped to promoters (Fig. 5a-c). In total 8,381 DNA fragility hotspots were identified in DDX41 knockdown cells, of which 5,958 were not present in wild-type HCT116 cells ( Fig. 5d and Supplementary Data 5). We found 3,108 DSB gains (FC > 2) in DDX41 KD cells, 54% of which mapped to promoter regions (Supplementary Fig. 5a and Supplementary Data 5). To investigate whether and to which extent R-loops correlate with DSBs and whether DSBs in DDX41 knockdown cells overlap with promoters displaying R-accumulation, we performed MapR in HCT116 cells. We compared the MapR with previously published GRO-sequencing data in HCT116 cells to test the dependency of R-loops on transcription 53 . Similar to U2OS cells, loss of DDX41 in HCT116 cells led to a dramatic accumulation of R-loops in promoters of active genes ( Supplementary Fig. 5b, c). From 15,177 MapR peaks identified in all 3 replicate experiments in DDX41 knockdown cells, 7,275 showed a gain in R-loops (FC > 1.5) (Supplementary Fig. 5d and Supplementary Data 6). Using sBLISS, we identified 1,642 promoter regions that showed increased fragility in DDX41 knockdown cells, and 53% of those promoters with DSB gains also displayed R-loops upon DDX41 loss (Fig. 5e, f and Supplementary Fig. 5e). This suggests that a large proportion of DNA fragile sites in promoters coincides with R-loops in DDX41 knockdown cells.
Interestingly, RNA-sequencing revealed upregulation of genes involved in inflammatory signaling in DDX41 knockdown cells, in particular NF-kB signaling, which was confirmed also by increased nuclear localization of NF-kB subunit p65 in these cells ( Fig. 5h and Supplementary Fig. 5f). Inflammatory signaling genes did not display an accumulation of R-loops nor were bound by DDX41, suggesting that these genes are not directly regulated by DDX41, but are likely an indirect consequence of DDX41-dependent R-loop accumulation, and genomic instability. To test whether DDX41 also opposes DSBs and genomic instability in hematopoietic stem and progenitor cells (HSPCs) in which loss-of-function mutations of DDX41 result in AML, we depleted DDX41 in human CD34 + HSPCs, and monitored DSBs using 53BP1 foci. Importantly, we found that depletion of DDX41 using two different shRNAs led to spontaneous DSB formation (Fig. 6a). Notably, overexpression of DDX41 variants found in AML patients in either the DEAD (L237F/P238T) or helicase domain (R525H) resulted in the accumulation of 53BP1 foci in CD34 + cells (Fig. 6b). These results suggest that also in human HSPCs DDX41 functions in opposing transcription-associated genomic instability. leads to replication stress and genomic instability. a Immunofluorescence analysis of yH2AX in U2OS cells 48 h after indicated knockdowns. Center lines of boxplots indicate the median, the limits the 25th-75th percentile, whiskers the 10th-90th percentile, dots outliers. Representative data of n = 3 biologically independent experiments; p-values (p < 0.0001, p = 0.1788, p < 0.0001, p = 0.8259, p > 0.9999, p = 0.9569) were derived from >1000 cells using one-way ANOVA with Tukey correction for multiple comparisons. Representative images of yH2AX (red) staining and Hoechst33342 (blue). Scale bars-20 µm. b Immunofluorescence analysis of yH2AX in U2OS cells ± doxycycline-inducible GFP-tagged M27-RNaseH1. Quantification of cells with medium GFP intensity (medium M27-RNaseH1 expression). Representative boxplots of n = 2 biologically independent experiments. Center of boxplots indicates the median, limits the 25th-75th percentile, whiskers the 10th-90th percentile, dots outliers. p < 0.0001 derived from n > 500 cells using a two-sided Mann-Whitney test. c Single-cell electrophoresis of U2OS cells 48 h after knockdown ± doxycycline-inducible expression of HA-tagged M27-RNaseH1. Representative images are displayed (right). Scale bars-40 µm. Dots depict individual tail moments, black line the median. Representative results from n = 2 biologically independent experiments. p-values (p = 0.0001, p = 0.0031) were derived from n > 50 cells using one-way ANOVA with Tukey correction for multiple comparisons. d Immunofluorescence analysis of 53BP1 foci in U2OS cells 48 h after indicated knockdowns. Whiskers of the box plot represent the 10th-90th percentile, the center line the median, the limits the 25th-75th percentile, and the dots depict outliers. Representative of n = 3 biologically independent experiments. p-value < 0.0001 was derived from n > 1000 cells using an unpaired, twosided Student's t-test. e DNA fiber spreading assay of U2OS cells after 48 h knockdown of DDX41. Controls were either treated with DMSO or 100 nM aphidicolin for 1.5 h. Representative images (white line indicates 10 µm scale) and quantifications of fiber tract length. Dots represent individual values and the black line the median. At least 260 fibers were quantified across n = 1 experiment. p-values (p < 0.0001, p < 0.0001, p < 0.0001, p = 0.5794) were derived using one-way ANOVA with Tukey correction for multiple comparisons. f Immunofluorescence analysis of pRPA (Ser33) in U2OS cells 48 h after indicated knockdowns. Representative images (right): Hoechst33342 (blue), pRPA (Ser33) (green). Center of boxplots indicates the median, limits the 25th-75th percentile, whiskers the 10th-90th percentile, dots outliers. Representative data of n = 3 biologically independent experiments are displayed. p-values (p < 0.0001, p > 0.9999, p < 0.0001, p < 0.0001, p = 0.9384, p < 0.0001) were derived from n > 1000 cells using one-way ANOVA with Tukey correction for multiple comparisons. Scale bars-20 µm. Source data are provided as a Source Data file.

Discussion
R-loop levels across the genome need to be balanced to ensure the regulation of chromatin-associated processes without inflicting DNA damage and genomic instability. Here, we developed RDProx that provides a snapshot of the R-loop-proximal proteome in human cells. We identified 612 R-loop-proximal proteins and divided them in two categories (Tier 1 and Tier 2) depending on the probability of their presence at R-loops, providing a rich resource for further functional investigations. The advantages of RDProx are manifold: (1) labeling of R-loop-proximal proteins is performed in vivo, which ensures that R-loops, chromatin, and cellular compartments remain intact; (2)  Thereby, it provides a methodological framework to answer outstanding questions, including how R-loop regulation differs between cell cycle stages or in response to stress that impacts transcription or co-transcriptional processes. Previous studies have employed the S9.6 RNA-DNA hybrid antibody for immunoprecipitation of proteins that associate with R-loops or used an in vitrogenerated RNA-DNA hybrid to pull down interacting proteins from cell extracts 54,55 . Recent reports showed that the S9.6 antibody in fixed human cells predominantly recognizes ribosomal RNA and not RNA-DNA hybrids 56 . Unbiased inspection of the previously reported S9.6-based proteomics data set by GO term enrichment analysis revealed "rRNA processing" and "ribosome biogenesis" as the most significantly enriched terms ( Supplementary Fig. 1f). RDProx relies on the HBD of RNaseH1 and is therefore inherently biased to the R-loops that are recognized and bound by RNaseH1. Recently, the existence of different classes of R-loops-promoterpaused R-loops and elongation-associated R-loops-that each display unique characteristics was proposed 57,58 . Promoter-paused R-loops are short R-loops frequently forming during promoterproximal pausing of RNAPII 57,58 . R-loop mapping approaches based on RNaseH1 showed an enrichment of RNaseH1 at promoter-proximal sites 43,46 . It remains unclear whether and to which extent RNaseH1 binds to R-loops in other genomic regions. We therefore speculate that RDProx might be most sensitive in recovering proteins that associate with promoter-proximal R-loops. This could explain why some previously reported R-loop-associated proteins, such as the RNA/DNA helicase SETX and endonucleases XPG/XPF, were not identified in RDProx. For instance, SETX seems to primarily associate with R-loops at DSBs 59 . On the other hand, we would expect XPG and XPF in proximity to transcriptionassociated R-loops but it might be that these proteins are recruited only in occasions when R-loops are processed into DSBs and when not anymore bound by RNaseH1 19 . In addition, XPG and XPF are relatively low abundant in cells, which might preclude their identification by mass spectrometry. Similar might be true for the RNaseH2 complex, where we only identified RNaseH2A but not the B and C subunits of the complex.
Recent studies have shown that the RNA moiety in the hybrid can be modified with N6-methyladenosine (m6A) [14][15][16] . We now provide evidence that components of the m6A RNA machinery including the m6A writers (m6A-METTL-associated complex: VIRMA, ZC3H13, and RBM15), readers (hnRNPA2B1 and hnRNPC), and erasers (ALKBH5 and FTO) are indeed proximal to R-loops. This finding suggests that dynamic m6A deposition at RNA-DNA hybrids modulates the stability of R-loops in a contextdependent manner and through an interplay with other pre-mRNA processing factors. METTL3-dependent m6A deposition on RNA-DNA hybrids was shown to favor R-loop turnover during mitosis by recruiting the m6A reader YTHDF2 16 . A recent study identified a role for m6A RNA modification in stabilizing cotranscriptional R-loops forming at transcription termination sites, thereby ensuring faithful transcription termination and avoiding RNAPII read-through 15 .
Another large group of proteins identified by RDProx was components of the DNA replication machinery such as the MCM complex, WDHD1, RFC1, MSH6, and CHAF1B. Regulation of R-loop levels by RNase H enzymes is known to be necessary for unperturbed replication 60,61 . Conversely, replication can influence R-loop formation during transcription-replication conflicts depending on the mutual orientation of the transcription and replication machineries: the replisome reduces R-loop levels when traveling co-directionally with the transcription machinery but stabilizes R-loops during head-on transcription-replication collisions 62 . The MCM complex was demonstrated to possess RNA-DNA helicase activity in vitro and is therefore potentially involved in the removal of R-loops in S phase 63 . Timely unloading of PCNA, as well as the recruitment of DEAD/DExHbox helicases to the replication fork, were shown to prevent replication-associated R-loop accumulation 64 . The identification of additional replication-associated factors by RDProx implies unexplored details of the crosstalk between DNA replication and R-loops.
Identification of SMARCA4, ARID1A, SMARCC1, and SMARCE1 proximal to R-loops and their known association with active transcription sites marked by H3K27 acetylation suggests a role of the SWI/SNF chromatin-remodeling complex in balancing R-loop levels 65,66 . The yeast and human FACT complex have been reported to resolve R-loop-mediated transcription-replication conflicts by reshaping the chromatin environment 67 . It has been recently reported that the SWI/SNF complex functions in resolving R-loop-mediated transcription-replication conflicts 68 . Furthermore, ARID1A-containing BAF complexes can recruit TOP2A to R-loop-associated chromatin, thereby preventing excessive R-loop formation and replication stress 69 . RDProx identified the role of the evolutionary conserved Dead box helicase 41 (DDX41) in opposing transcription-associated Rloop accumulation and DSBs at promoters. We demonstrate that DDX41 localizes to chromatin, preferentially associates with promoters, and opposes R-loop-dependent replication stress, DSBs, and genomic instability. Loss of DDX41 leads to a dramatic accumulation of R-loops in promoter regions-30% of all R-loops mapped to promoter regions accumulate upon loss of DDX41 in U2OS cells and this was even more apparent in HCT116 cells. Therefore, we propose that DDX41 plays a prominent role in counteracting the accumulation of R-loops in promoter regions of active genes. Dysregulated R-loops that obstruct the progression of replication forks have been proposed as a major source of Rloop-dependent genomic instability 70 . Indeed, we found that DDX41 loss leads to genomic instability and increased fragility of DNA in promoter regions. We observed that DDX41 knockdown cells display slower replication fork progression and signatures of ATR-dependent signaling, suggesting that R-loop accumulation in DDX41 knockdown cells can obstruct the progression of replication forks and lead to the replication stress response. We also found that DDX41 knockdown cells show signs of perturbed transcription initiation and elongation. It is plausible that R-loop accumulation upon DDX41 loss leads to DSBs by interfering with replication and/or transcription machinery, which might explain why not all sites with R-loop accumulation display increased DNA fragility. It remains to be investigated by which mechanisms and under which conditions dysregulated R-loops are processed into DSBs. We also found sites of increased DNA fragility upon DDX41 loss that did not display R-loop accumulation and hence do not exclude a possibility that DDX41 safeguards actively transcribed genes through additional mechanisms.
RDProx also identified the DEAD-box helicase DDX39B/ UAP56 that was described to participate in nuclear mRNA export as part of the TREX complex 27,71,72 . A recent study revealed the role of DDX39B in resolving R-loops by demonstrating that DDX39B/UAP56 associates with active transcription complexes to resolve R-loops throughout the gene body until the transcription termination site, thereby ensuring faithful transcriptional elongation and transcript release 34 . In contrast, DDX41 shows a striking preference to associate with, and oppose, R-loop accumulation in promoter regions, suggesting that different RNA-DNA helicases are required to balance R-loop levels in different genomic regions or after replication or transcription stress.
Interestingly, pathogenic variants in DDX41 cause familial MDS/AML 39,41,[73][74][75] . A recent study performed in zebrafish suggested that R-loop accumulation caused by Ddx41 deficiency leads to upregulated inflammatory signaling and aberrant expansion of the HSPCs 35 . Also, accumulation of R-loops was recently proposed as a feature of myelodysplastic syndrome (MDS) harboring splicing mutations [76][77][78] . In this work, we demonstrate that pathogenic variants in DDX41 lead to the accumulation of DSBs in human HSPCs. Furthermore, we show that knockdown of DDX41 leads to the dependency of AML cells on ATR signaling (Fig. 6c). Genes that show R-loop accumulation are enriched for pathways frequently altered in AML such as chromatin organization, RUNX1 interactions as well as NOTCH and TGFβ signaling suggesting that DDX41 loss results in dysregulated transcription and aberrant cellular signaling through those pathways. These results suggest that pathogenic DDX41 variants in human familial MDS/AML contribute to disease development through the accumulation of R-loops and DSBs as well as provide incentives to explore ATR inhibition as a therapeutic strategy in these patients.
RDProx. SILAC-labeled cells were transfected with a construct expressing APEX2tagged HBD or HBD-WKK. After 48 h, cells were pre-treated with 500 µM biotin-phenol (Iris Biochem) for 2 h at 37°C, followed by a 2 min incubation with 1 mM H 2 O 2 (Sigma-Aldrich) at room temperature. Cells were washed twice with quenching solution (10 mM sodium azide, 10 mM sodium ascorbate, 5 mM Trolox (all from Sigma-Aldrich), and twice with PBS. Cells were lysed on ice using RIPA buffer (50 mM Tris, 150 mM NaCl, 0.1% SDS, 0.5% sodium deoxycholate, 1% Triton X-100). To release chromatin-bound proteins, cell lysates were sonicated using Bioruptor (Diagenode). For affinity purification of biotinylated proteins, equal amounts of differentially SILAC-labeled cell extracts, originating from either the HBD or the HBD-WKK condition, were combined prior to the pulldown and incubated with pre-equilibrated NeutrAvidin agarose beads (Thermo Scientific) for 2 h at 4°C on a rotation wheel. Beads were washed once with RIPA buffer, thrice with 8 M Urea (Sigma) in 1% SDS, and once with 1% SDS in PBS. Bound proteins were eluted in NuPAGE LDS Sample Buffer (Life Technologies) supplemented with 1 mM DTT and boiled at 95°C for 15 min. The eluates, after cooling down to room temperature, were alkylated by incubating with 5.5 mM chloroacetamide for 30 min in the dark and then loaded onto 4-12% gradient SDS-PAGE gels. Proteins were stained using the Colloidal Blue Staining Kit (Life Technologies) and digested in-gel using trypsin. Peptides were extracted from the gel and desalted on reversedphase C18 StageTips.
MS analysis. Peptide fractions were analyzed on a quadrupole Orbitrap mass spectrometer (Q Exactive or Q Exactive Plus, Thermo Scientific) equipped with a UHPLC system (EASY-nLC 1000, Thermo Scientific) as described 80,81 . Peptide samples were loaded onto C18 reversed-phase columns (15 cm length, 75 µm inner diameter, 1.9 µm bead size) and eluted with a linear gradient from 8 to 40% acetonitrile containing 0.1% formic acid in 2 h. The mass spectrometer was operated in data-dependent mode, automatically switching between MS and MS 2 acquisition. Survey full-scan MS spectra (m/z 300-1700) were acquired in the Orbitrap. The 10 most intense ions were sequentially isolated and fragmented by higher-energy C-trap dissociation (HCD) 82 . An ion selection threshold of 5000 was used. Peptides with unassigned charge states, as well as with charge states less than +2 were excluded from fragmentation. Fragment spectra were acquired in the Orbitrap mass analyzer.
Peptide identification. Raw data files were analyzed using MaxQuant (development version 1.5.2.8) 83 . Parent ion and MS 2 spectra were searched against a database containing 98,566 human protein sequences obtained from the Uni-ProtKB released in 04/2018 using Andromeda search engine 84 . Spectra were searched with a mass tolerance of 6 ppm in MS mode, 20 ppm in HCD MS2 mode, strict trypsin specificity, and allowing up to 3 miscleavages. Cysteine carbamidomethylation was searched as a fixed modification, whereas protein N-terminal Consensus regions were constructed using the intersection of peaks for the replicates in each condition (siCtrl and siDDX41). The union of these regions was used for further analysis and quantification of the coverage/FC. The mean log 2 fold change between siCtrl and siDDX41 is plotted against the log 2 average counts per million representing the coverage. Genomic regions that are differentially regulated (FC > 2) are highlighted in red (up) or in blue (down). e Genomic feature distribution of the regulated MapR regions in U2OS cells after DDX41 knockdown. Features are color-coded as indicated in the legend. f The proportion of genomic regions with R-loop gains or losses in U2OS cells overlapping CGIs or not-overlapping regions are depicted. g Reactome pathway over-representation analysis for genes with R-loop gains in U2OS cells. The adjusted p-values (Fisher's exact test with Bonferroni-Holm correction) are indicated. h Representative snapshot of a genomic region depicting R-loops and GFP-DDX41 binding profiled by MapR and greenCUT&RUN, respectively, in U2OS cells.
acetylation and methionine oxidation were searched as variable modifications. The data set was filtered based on posterior error probability (PEP) to arrive at a false discovery rate of below 1% estimated using a target-decoy approach 85 .
RDProx network analysis. Pearson correlations were calculated using RStudio (version 1.3.959). Functional protein interaction network analysis was performed using interaction data from the STRING database 86 . Only interactions with a score >0.7 are represented in the networks. Cytoscape (version 3.2.1) was used for the visualization of protein interaction networks 87 . Genes were manually annotated by literature research and clustered based on similarity. PFAM domain enrichment analysis was performed using EnrichR 88 . The respective terms with the lowest FDR based on Fisher's exact test and correction for multiple comparisons are highlighted next to each cluster.  Neutral comet assay. Neutral comet assay was performed according to the manufacturer's protocol (Trevigen). Briefly, cells were embedded in low melting agarose at 37°C on Comet Slides (Trevigen). Overnight cell lysis at 4°C was followed by equilibration in 1× Neutral Electrophoresis Buffer for 30 min at room temperature. Single-cell electrophoresis was performed at 4°C in 1× Neutral Electrophoresis buffer for 45 min with constant 21 V. After DNA precipitation with 1× DNA Precipitation Buffer, Comet Slides were dried with 70% EtOH at room temperature. In order to completely dry the samples, Comet Slides were transferred to 37°C for 15 min. DNA was stained with SYBR Gold solution for 30 min at room temperature. Images were taken with a Leica AF7000 microscope using a ×20 0.8NA air objective and a filter cube 480/40 nm, 505 nm, and 527/30 for excitation, dichroic, and emission wavelengths respectively. Tail moments of the comets were quantified using the CometScore (TriTek Corp.) software. At least 50 comets were quantified per condition.
Quantification of RNA-DNA hybrids using dot blot. Genomic DNA was extracted using the DNeasy mini kit (Qiagen). The isolated gDNA was treated with 1.2 U RNase III (produced in-house) for 2 h at 37°C. After enzyme deactivation at 65°C for 20 min, samples were split in half to digest control samples with 10 U RNaseH1 (NEB) overnight at 37°C. Enzyme deactivation was followed by spotting DNA in a serial dilution on a nitrocellulose membrane (NeoLab Migge GmbH) using a dot-blot apparatus (BioRad). DNA was cross-linked to the membrane by UV light and afterward blocked with 10% skimmed milk solution in PBS supplemented with 0.1% Tween-20. The membrane was incubated overnight at 4°C with the S9.6 antibody (produced in-house). After incubation of secondary antibodies conjugated to horseradish peroxidase (Jackson ImmunoResearch Laboratories) signal was detected using SuperSignal West Pico Chemiluminescent Substrate (Thermo Scientific). An antibody against dsDNA was probed as a loading control after stripping the membrane with β-mercaptoethanol (Sigma) and 0.1% SDS in PBS. The detected signal was quantified using Fiji/ImageJ (v1.51) and ratios between the signal resulting from S9.6 and dsDNA staining were calculated to quantify global R-loop levels 89 .
Proximity ligation assay. Proximity Ligation Assay was performed according to the manufacturer's protocol (Duolink®, Sigma-Aldrich). Cells were fixed with 4% paraformaldehyde in PBS and permeabilized with 0.25% Triton X-100. Samples were blocked with Duolink® Blocking Solution for 1 h at 37°C in a humidity chamber. After removal of the blocking solution, primary antibodies diluted in Duolink® Antibody Diluent were added on the coverslips for 2 h at room temperature in a humidity chamber. Coverslips were washed 2× with Washing Buffer A. PLA plus and minus probes were put on in a 1:5 dilution in Duolink® Antibody Diluent for 1 h at 37°C in a humidity chamber. Two washes with Washing Buffer A were followed by Ligase treatment in 1× Ligation Buffer for 30 min at 37°C in a humidity chamber. Ligation buffer was tapped off and coverslips were washed twice with Washing Buffer A. Amplification was achieved by adding the Polymerase in 1× Amplification buffer for 100 min at 37°C in a humidity chamber. After washing the samples 2× with 1× Washing Buffer B and 1× with 0.01× Washing Buffer B, coverslips were stained with 1 µg/ml Hoechst33342 and mounted using Dako mounting medium. Images were taken with a Leica SPE microscope using a ×63 1.4NA oil objective. The number of PLA spots per nucleus was quantified using Fiji/ImageJ (v1.51) 89 .
ATPase assay. The ADP-Glo Assay was performed according to the manufacturer's protocol (Promega). In brief, an ATP/ADP standard curve was prepared before each experiment in order to interpolate the measured values. Purified full-length DDX41 was incubated in a serial dilution together with 100 nM of RNA-DNA substrate with an ssDNA overhang and 5 µM ATP. After incubating the mix at 37°C for 60 min, the reaction was stopped by depleting unconsumed ATP with the ADP-Glo Reagent. The Kinase Detection Buffer was added to convert ADP to ATP and to add luciferase and luciferin to detect ATP. The resulting luminescence was measured with a Spark M200 (Tecan). The measured values were interpolated based on the values obtained by the ATP/ADP standard curve using GraphPad PRISM (v7.04, Graphpad Software, Inc.). Electrophoretic mobility shift assay. 20 nM of 6-FAM-conjugated single-and double-stranded oligonucleotides were incubated with 25 µM of purified HBD or HBD-WKK mutant for 10 min at room temperature in interaction buffer (20 mM Tris-Cl pH 7.5, 150 mM NaCl, 10% glycerol, 1 mM EDTA, 1 mM DTT). 6× loading buffer (60% Glycerol, 20 mM Tris-Cl pH 8.0, 60 mM EDTA) was added to the samples before loading them on a 20% Novex TBE gel (ThermoFisher Scientific). The gel was run for 45 min at 200 V in TBE buffer and scanned using a Typhoon FLA 9000 @ 473 nm to visualize the fluorescence of the 6-FAM-labeled probes.
FRET-based unwinding assay. RNA-DNA hybrid substrates with a single-stranded DNA overhang were generated by mixing an IBFQ-conjugated 38-mer DNA oligo (IDT) and a 6-FAM-conjugated 13-mer RNA oligo (IDT) and heating them to 95°C and gradually cooling them down to 4°C. Annealed substrates were incubated together with 5 µM ATP and either full-length DDX41 or mutant proteins. Increased fluorescence intensity upon addition of DDX41 after displacement of the quencher during unwinding was measured on a Spark M20 (Tecan) plate reader.
qPCR analysis. RNA was extracted using the RNeasy Plus Mini Kit (Qiagen). 500 ng of purified RNA was reverse-transcribed into cDNA by using the Quanti-Tect Reverse Transcription Kit (Qiagen). Purified cDNA was amplified during qPCR on a CFX384 BioRad instrument using 2× SYBR Green mix and 0.5 µM final primer mix.
Cell viability assay. Cell viability assay was performed using the Cell Titer-Blue Cell Viability Assay (Promega) according to the manufacturer's instructions.
RNA-sequencing and data analysis. RNA was extracted using the RNeasy Plus Mini Kit (Qiagen). In brief, cells were lysed and genomic DNA was depleted. Samples were treated with DNase to remove residual DNA. After purification using   97 with the parameters "--keepdup auto --broad --broad-cutoff 0.1 --bw 100 --min-length 100 --format BAMPE --g hs". The MapR samples were further analyzed using the FC between siDDX41 and siCtrl. A set of consensus regions was created using the intersection of peaks called per group (either in siDDX41 or siCtrl replicates). Then the union of these two peak sets was used to quantify the signal present in the samples. Using R/ Bioconductor 94 packages the fold change for the consensus regions was calculated using the average normalized coverage (rpkm) of the regions. Normalization was based on the total amount of sequenced reads. R-loop gains were determined based on the FC > 2 in siDDX41 compared to siCtr U2OS cells and FC > 1.5 in siDDX41 compared to siCtr HCT116 cells. Sequencing depth normalized coverage tracks for all samples and metagene/enrichment around the TSS or from TSS to TES were created using deepTools (v3.4.1) 95 and further processed using custom R scripts.
sBLISS and data analysis. sBLISS in HCT116 cells was performed as previously described 48 greenCUT&RUN and data analysis. CUT&RUN was performed in a stable U2OS cell line that expresses N-terminally GFP-tagged DDX41 under a doxycyclineinducible promoter. Expression of GFP-DDX41 was induced by adding 1 µg/ml doxycycline for 48 h or DMSO for un-induced control cells. Cells were mildly crosslinked with 1% formaldehyde for 2 min at room temperature. Quenching of the reaction with 125 mM glycine was followed by cell detachment using trypsin and two subsequent washes in Wash buffer (20 mM HEPES-KOH (pH 7.5), 150 mM NaCl and 0.5 mM spermidine and EDTA-free complete protease inhibitor). Concanavalin-A beads were activated in binding buffer (20 mM HEPES-KOH (pH 7.9), 10 mM KCl, 1 mM CaCl 2 , and 1 mM MnCl2) for 5 min at room temperature and afterward 1*10 6 cells immobilized on the beads for 10 min at room temperature. After cell permeabilization with 0.05% digitonin-Wash buffer, 1 µg of GFP-nanobody-MNase (GFP-nanobody LaG16 described in ref. 44   Model for DDX41 function in R-loop homeostasis. a Immunofluorescence analysis of HSPCs after 24 h of indicated knockdowns. Cells were nucleofected with plasmids encoding GFP and respective shRNAs. GFP-positive cells were sorted via FACS and seeded on coverslips. Representative images of 53BP1 (red) staining in HSPCs (left). DNA was counterstained with DAPI (blue). Quantification of nuclear 53BP1 intensity (right). Each dot represents a single measured value. The black line indicates the median. At least 80 cells across n = 2 biologically independent experiments were measured per condition. p-values (p < 0.0001, p < 0.0001, p = 0.2227) were calculated by one-way ANOVA with Tukey correction for multiple comparisons. Scale bars-20 µm. Source data are provided as a Source Data file. b Immunofluorescence analysis of HSPCs after expression of DDX41 WT, L237F + P238T or R525H mutants tagged with GFP. GFP-positive cells were sorted by FACS and used for the analysis. Representative images of 53BP1 (red) staining in HSPCs (left). DNA was counterstained with DAPI (blue). Quantification of nuclear 53BP1 intensity (right). Dots represent results from individual cells. The median is indicated by the black line. p-values (p = 0.0237, p = 0.624, p = 6.018) were derived from at least 30 cells across n = 1 experiment using one-way ANOVA with Tukey correction for multiple comparisons. Scale bars-20 µm. c Wild-type DDX41 associates with R-loops in promoters of active genes and balances R-loop levels by unwinding RNA-DNA hybrids. Pathogenic DDX41 variants found in acute myeloid leukemia (AML) display impaired RNA-DNA hybrid unwinding activity, leading to the accumulation of R-loops at promoters. Accumulation of R-loops at promoters results in increased replication stress, DSBs, and inflammatory signaling, rendering DDX41 mutated AML cells vulnerable to ATR inhibitors.
Immunofluorescence. Cells were washed 2× with PBS, incubated with 0.4% NP-40 for 20 or 40 min on ice, and washed 2× with PBS-T (0.1%). Cells were fixed with 4% paraformaldehyde in PBS for 15 min at room temperature, washed 2× with PBS-T (0.1%), and permeabilized with Triton X-100 (0.3%) for 5 min at room temperature, followed by 2× washes with PBS. Cells were blocked for 1 h with 5% fetal bovine serum albumin in PBS-T (0.1%) containing penicillin and streptomycin. Incubation with primary antibodies diluted in blocking buffer was performed overnight at 4°C and followed by 3× washes with PBS-T and 1 h incubation with Alexa Fluorcoupled secondary antibodies in a dark chamber at room temperature. Nuclei were counterstained with 1 µg/ml Hoechst33342 in PBS either simultaneously with secondary antibody incubation or for 30 min. For chromatin retention assay permeabilization, blocking and antibody incubations steps were omitted. Cells were washed 2× with PBS-T and kept at 4°C in PBS until imaging. Imaging was performed with an Opera Phenix (PerkinElmer) microscope using a ×40 1.1NA water objective. Image analysis was performed by using Harmony High-Content Imaging and Analysis Software (version 4.4, PerkinElmer). Standard building blocks allowed for nuclei segmentation based on the Hoechst signal and cells on the edges of the field were excluded. Mean intensity measurements were performed for maximum projections and spot detection was calculated by using algorithm B. Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support this study are available from the corresponding author upon reasonable request. The fasta file containing the human reference proteome (released in 04/2018) used for the analysis of raw MS data using MaxQuant was retrieved from UniprotKB UP000005640. HCT116 GRO-seq data to determine gene expression levels for MapR analysis was retrieved from GEO with access code GSM2296622. RNA-Seq, MapR, and BLISS data generated in this study have been deposited in the GEO database under accession code GSE168173. The mass spectrometry-based proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository 102 with the data set identifier PXD024517. Source data are provided with this paper.