Phase separation properties of RPA combine high-affinity ssDNA binding with dynamic condensate functions at telomeres

RPA has been shown to protect single-stranded DNA (ssDNA) intermediates from instability and breakage. RPA binds ssDNA with sub-nanomolar affinity, yet dynamic turnover is required for downstream ssDNA transactions. How ultrahigh-affinity binding and dynamic turnover are achieved simultaneously is not well understood. Here we reveal that RPA has a strong propensity to assemble into dynamic condensates. In solution, purified RPA phase separates into liquid droplets with fusion and surface wetting behavior. Phase separation is stimulated by sub-stoichiometric amounts of ssDNA, but not RNA or double-stranded DNA, and ssDNA gets selectively enriched in RPA condensates. We find the RPA2 subunit required for condensation and multi-site phosphorylation of the RPA2 N-terminal intrinsically disordered region to regulate RPA self-interaction. Functionally, quantitative proximity proteomics links RPA condensation to telomere clustering and integrity in cancer cells. Collectively, our results suggest that RPA-coated ssDNA is contained in dynamic RPA condensates whose properties are important for genome organization and stability. Here, the authors show that replication protein A (RPA) tends to self-assemble into dynamic condensates, in a manner that is stimulated by ssDNA and regulated by RPA2 phosphorylation. RPA condensates are functionally important for telomere clustering and RAD52-dependent telomere maintenance.

Article https://doi.org/10.1038/s41594-023-00932-w RPA2 optoDroplets, once induced, remained stable for several minutes in the absence of blue light, with a half-life of around 5 minutes (Fig. 1c). The RPA2 optoDroplets showed a high degree of mobility (Supplementary Video 1) and frequently underwent droplet fusions (Fig. 1d). Furthermore, RPA2 optoDroplets could be induced multiple times sequentially in the same cell in a switch-like manner (Fig. 1e). Together, these results suggest that RPA2 has an intrinsic and hitherto unexplored propensity for dynamic self-assembly. As RPA2 forms a stable trimeric protein complex together with RPA1 and RPA3, we generated a polycistronic construct to express all three RPA subunits in stoichiometric amounts in cells. When expressed as a trimeric complex (tRPA), light-controlled clustering was observed within seconds after induction with blue light (Extended Data Fig. 2a-e), indicating that RPA condensation was not due to isolated expression of one of its subunits.
To exclude that light-induced clustering of RPA was due to protein over-expression, we analyzed nuclear RPA levels in transfected cells by multicolor high-content microscopy (Extended Data Fig. 3a). By differentiating cells according to their Cry2-mCherry-RPA2 expression levels into transfected (positive) and untransfected (negative) cells, we identified cells in both categories with similar RPA2 expression on the basis of RPA2 antibody staining. We observed that the transfected cells formed RPA2 optoDroplets despite the RPA2 concentration being close to the endogenous level (Extended Data Fig. 3b). Moreover, a short interfering RNA (siRNA) against RPA1 lowered endogenous RPA levels (measured by RPA1 and RPA2 co-staining) and allowed us to look at Cry2-mCherry-RPA2-transfected cells, in which RPA2 levels (after RPA depletion by siRNA) were comparable to endogenous RPA2 levels (cells transfected with non-targeting negative control siRNA (siControl) served as a reference). These cells also showed optoDroplet formation, excluding supraphysiological RPA expression as a cause of light-induced RPA clustering (Extended Data Fig. 3c).
Next, we analyzed intracellular RPA condensation in cells stably expressing fluorescently labeled tRPA 20 . RPA formed nuclear foci in these cells, and time-lapse experiments revealed fusion and occasional fission events after replication stress (Extended Data Fig. 4a). Similarly, replication-stress-induced RPA foci fusions and rare fission events were observed at physiological protein concentrations in cells in which the endogenous RPA1 locus had been engineered, using CRISPR-Cas9, to express an mScarlet-RPA1 fusion protein (Extended Data Fig. 4b-d) 21 .
RPA foci appeared to be of spherical shape in widefield and confocal microscopy images, a feature that was also observed when nuclear RPA ensembles were imaged at higher resolution by stimulated emission depletion (STED) microscopy (Extended Data Fig. 4e). Fluorescence recovery after photobleaching (FRAP) experiments to evaluate RPA exchange at individual RPA foci showed fast recovery rates, with half-recovery times below 10 seconds, both in unchallenged cells and in cells exposed to replication-stress-and DNA-damage-inducing treatments (Extended Data Fig. 4f-h). These half-recovery times at endogenous RPA foci and at ATR-kinase-inhibitor-and ionizing radiation (IR)-induced DNA lesions are in the same range as was measured for other proteins with a high propensity to phase separate, including FUS, hnRNPA1, and DDX4 (ref. 22 ). Moreover, endogenous and replication-stress-induced RPA foci were sensitive to a short treatment with low concentrations of the aliphatic alcohol 1,6-hexanediol, which interferes with weak hydrophobic interactions (Extended Data Fig. 4i,j).

Phase separation of the RPA complex is stimulated by ssDNA
Collectively, these findings raised the possibility that RPA forms condensates owing to protein properties linked to phase separation. To further explore this possibility, we expressed and purified the untagged human tRPA complex over three sequential columns [23][24][25] , resulting in highly pure tRPA fractions ( Fig. 2a and Supplementary Fig. 1a,b). The purified tRPA complex readily formed dynamic liquid droplets in vitro (Fig. 2b, Supplementary Fig. 1c, and Supplementary Video 2), and their formation was enhanced by increasing concentrations of irrespective of their sequence and genomic location, and owing to its universal role in DNA metabolism, RPA is essential for cell survival. Non-lethal RPA mutations result in DNA-repair defects and genome instability, and RPA haploinsufficiency causes greatly enhanced tumor formation and shortened lifespan 8 .
One RPA heterotrimer binds to approximately 30 nucleotides of ssDNA with ultrahigh affinity 2,6,9 . RPA coating of ssDNA has been considered to resemble stoichiometric binding, like tightly packed 'beads on a string.' However, individual RPA domains dynamically exchange on ssDNA, and diffusion-driven sliding of RPA along ssDNA has been observed [10][11][12] . RPA, when bound to ssDNA, also functions as an interaction and activation platform for a variety of signaling molecules and genome caretakers, including the replication stress response factors ATRIP-ATR, ETAA1, and PRIMPOL 2 . In the context of HR, the tumor suppressor BRCA2 is needed to displace RPA from ssDNA in exchange for RAD51. RPA not only has up to 1 × 10 5 -fold higher affinity to ssDNA than RAD51, but it is also highly abundant: there are approximately 3-5 million molecules per cell 13 , equivalent to a concentration in the 10 μM range in the nucleoplasm, and its local concentration is likely even higher when it is bound to ssDNA. Considering RPA's high abundance and high affinity for ssDNA, how ssDNA is handed off from RPA to different downstream effectors remains poorly understood.
In bacteria, the essential ssDNA-binding protein SSB has recently been reported to phase separate in vitro and in bacterial extracts 14 . In yeast, however, the ssDNA-binding protein Rfa1 showed less dynamic behavior and tighter ssDNA binding than did Rad52, for which condensation behavior consistent with phase separation had been proposed 15,16 . Here, we reveal that mammalian RPA has a strong intrinsic propensity to form highly dynamic condensates both in vitro and in the nucleus of living cells. In vitro, phase separation of the purified RPA trimer is specifically enhanced by ssDNA, which in turn gets selectively enriched in RPA droplets. We show that the RPA subunit RPA2 is critical for phase separation and, using structure prediction and systematic site-directed mutagenesis in conjunction with light-inducible RPA optoDroplet formation, reveal that the intrinsically disordered amino terminus of RPA2 regulates condensation properties. This disordered sequence stretch of RPA2 is subjected to multi-site phosphorylation, and mutations that mimic RPA hyper-phosphorylation disrupt RPA phase separation in vitro and abolish cellular RPA optoDroplet formation. Finally, using label-free quantitative proteomics, we identify heterotypic interactions that respond to intracellular RPA condensation and link RPA's clustering capacity to telomere maintenance by alternative lengthening of telomeres (ALT) in cancer.

RPA forms optoDroplets and dynamic DNA-repair condensates
Using an optogenetic tool to interrogate protein properties associated with phase separation in living cells 17,18 , we identified surprisingly strong light-induced clustering of human RPA2 (Fig. 1a,b and Extended Data Fig. 1a-c). The prion-like N-terminal domain of FUS and an oligomerization-prone Cry2 mutant 19 , which were included as two positive controls, also showed strong optoDroplet formation (Extended Data Fig. 1a-c). Other DNA-damage response (DDR) factors showed weaker or no light-induced clustering when over-expressed as Cry2-mCherry fusions, and neither expression of the Cry2-mCherry module alone (empty) nor as a fusion with dimerization-prone glutathione S-transferase (GST) resulted in optoDroplet formation (Extended Data Fig. 1a-c). We conclude that light-induced seeding requires self-assembly-driven amplification of protein condensation to cause discernable optoDroplet formation, and that RPA2 carries these features. Although other DDR factors did not show these features in the optoDroplet system, we do not exclude the possibility of their dynamic clustering at endogenous expression levels and in other cellular contexts.
Article https://doi.org/10.1038/s41594-023-00932-w the crowding agent PEG-8000 ( Supplementary Fig. 1d). RPA droplets displayed gravity flow resulting in surface wetting at plate bottoms ( Fig. 2b and Supplementary Video 2), and frequently underwent droplet fusions (Fig. 2c), behaviors consistent with phase separation 26,27 . Freshly formed RPA droplets were initially mobile, but then settled and only partially formed again upon remixing ( Supplementary Fig. 1e). RPA droplets that had formed in vitro were dissolved by 1,6-hexanediol ( Supplementary Fig. 1f) and by millimolar concentrations of ATP, which has previously been reported to act as a biological hydrotrope that solubilizes aggregation-prone proteins in aqueous solutions 28 (Supplementary Fig. 1g). Taken together, these results demonstrate that the human trimeric RPA complex has a propensity to phase separate and form liquid droplets in vitro.
Given the prominent role of RPA as a ssDNA-binding protein, we went on to test whether ssDNA affects the phase-separation behavior of RPA. Strikingly, ssDNA, but not sequence-matched annealed double-stranded DNA (dsDNA), greatly enhanced RPA phase separation ( Fig. 2d and Extended Data Fig. 5a,b). Although short ssDNA oligomers of 10-12 nucleotides in length did not measurably induce RPA phase separation, ssDNA molecules that were 20-50 nucleotides long were sufficient to robustly trigger RPA droplet formation and render RPA solutions turbid (Extended Data Fig. 5c,d). These findings are consistent with the biochemically characterized ssDNA binding preference of RPA 2,6 and suggest that RPA-ssDNA binding regulates RPA phase-separation behavior. Of note, titration of both RPA and ssDNA demonstrated that sub-stoichiometric amounts of 40-nucleotide-long ssDNA are sufficient to trigger RPA phase separation at physiological RPA concentrations (Fig. 2e,f). Labeling of purified RPA and ssDNA with fluorescent dyes revealed that ssDNA not only enhanced RPA phase separation, but also was specifically enriched in RPA-containing droplets (Fig. 2g,h and Extended Data Fig. 5e). Matched RNA, on the other hand, did not induce RPA phase separation and was not enriched in RPA droplets (Extended Data Fig. 5f,g). Without RPA, ssDNA did not form discernible liquid droplets under these conditions (Extended Data Fig. 5h), and ssDNA-containing RPA droplets were dissolved by 1,6-hexanediol (Extended Data Fig. 5i). Taken together, ssDNA, but not dsDNA or ssRNA, induces RPA phase separation in vitro and is enriched in RPA condensates.

Multi-site phosphorylation regulates RPA self-interaction
Next, to elucidate the mechanism and regulation of RPA phase separation, we turned again to the cellular optoDroplet system for the following reasons: the induction of optoDroplets by blue light not only provides a means to reversibly trigger protein condensation in the natural environment of living cells, but also to compare individual proteins, protein domains, and point mutants with respect to their propensity to form biomolecular condensates. When we compared the individual RPA subunits RPA1, RPA2, and RPA3 side by side, RPA2 showed the most pronounced light-induced clustering in cells (Fig. 3a-c), suggesting that RPA2 is the main driver of RPA assembly into light-induced condensates. To address whether RPA self-assembly occurs in the absence of blue-light-induced seeding, we immunoprecipitated green fluorescent protein (GFP)-RPA2 from benzonase-treated lysates of a stable GFP-RPA2 cell line and tested whether other RPA subunits are enriched together with GFP-RPA2. Remarkably, GFP-RPA2 pulled down not only its constitutive interaction partner RPA1, but also endogenous RPA2 (Fig. 3d).
RPA2 has two globular domains connected by a short linker, and an extended N-terminal intrinsically disordered region (N-IDR) with multiple phosphorylation sites 29 (Fig. 3e-h). AlphaFold predicted the N-IDR to be flexible and unconstrained by the two globular domains of RPA2 (Fig. 3e), with a high expected position error for the first 40-45 amino acids and a low expected position error for the two globular domains (Fig. 3f). Two sequence-based prediction tools for protein folding and disorder, GlobPlot 30 and PONDR (Predictor of Natural Disordered Regions) 31 , agreed on the N-IDR being unstructured (Fig. 3g,h). The N-IDR of human RPA2 and the N-IDR of its yeast homolog Rfa2 share little homology (Extended Data Fig. 6a) and, unlike human RPA2, yeast Rfa2 showed no light-induced clustering in optoDroplet experiments in mammalian cells (Extended Data Fig. 6b-d). We therefore focused our attention on the human N-IDR of RPA2 with its multiple phosphorylation sites. While alanine mutations did not abolish the condensation behavior of RPA2 in the optoDroplet system, phospho-mimicking aspartate mutations completely abrogated light-inducible RPA2 clustering ( Fig. 4a-c). In a series of single, double, triple, and quadruple S/T→D mutants, there was a gradual suppression of light-inducible RPA2 clustering, with the most pronounced clustering defect being observed when multiple phosphorylation sites (S8D, S11D, S12D, S13D, Article https://doi.org/10.1038/s41594-023-00932-w T21D, S23D, S29D, and S33D) 32 were altered ( Fig. 4d and Extended Data Fig. 6e-h). In accordance, when we expressed and purified trimeric wild-type (WT) RPA and the complete phospho-mimicking S/T→D mutant, only WT tRPA was able to form liquid droplets in vitro (Fig. 4e,f). When purified tRPA was phosphorylated in vitro by the DDR kinase DNA-PK, which was previously shown to antagonize the self-assembly of FUS into hydrogels through phosphorylation of its low complexity domain 33 , tRPA phase separation was reduced (Fig. 4g,h). Moreover, in the presence of ssDNA, the S/T→D mutant formed ssDNA-containing aggregates in vitro, unlike the dynamic ssDNA-containing droplets formed by WT tRPA (Fig. 4i), and the dynamic exchange of ssDNA, measured by FRAP, was reduced in these aggregates (Extended Data Fig. 6i,j). These results are consistent with previous data showing that phosphosite mutations in RPA2 are compatible with ssDNA binding 32,34 ,  with an anti-RPA2 antibody. e, AlphaFold 66 protein structure prediction for RPA2. Consistent with X-ray crystallography data on the RPA trimerization core, two globular domains, a flexible linker, and an N-IDR were predicted. Putative and confirmed phosphorylation sites within the N-IDR are indicated. f, AlphaFold Predicted Aligned Error (PAE) heatmap depicting the distance error for every pair of residues as an estimate of position error for the structure prediction shown in e. RPA2 residue numbers run along the vertical and horizontal axes, and the color indicates PAE values for the corresponding pair of residues. g, RPA2 protein disorder prediction by GlobPlot 2.3, an online tool plotting the tendency within query proteins for order/globularity and disorder 30 . Disorder propensity is shown on the y axis, and amino acid position of RPA2 is on the x axis. h, RPA2 protein disorder prediction by PONDR (Predictor of Natural Disordered Regions) 31 . PONDR score is shown on the y axis, and amino acid position of RPA2 is on the x axis. Highly disordered regions are marked by the black bar.
Article https://doi.org/10.1038/s41594-023-00932-w but they also suggest that the rapid exchange between RPA and ssDNA may be facilitated by dynamic condensate properties.

RPA condensation properties are linked to telomere maintenance
To explore biological functions potentially associated with dynamic RPA condensation, we generated stable cell lines in which light-induced RPA2 optoDroplet formation, without induction of DNA damage and independent of RPA's known roles in genome maintenance and repair, is coupled to TurboID-mediated proximity labeling (Fig. 5a). Negative control cells expressed TurboID fused to mCherry and the blue-light sensor Cry2, whereas matched test cells expressed TurboID-RPA2 fused to mCherry-Cry2. As anticipated, only the RPA2-expressing cells formed light-inducible condensates (Fig. 5b). Using a comparatively short biotin labeling time of 15 minutes, with and without simultaneous blue-light induction, we performed label-free quantitative proteomics (TurboID-RPA2 light, TurboID-RPA2 dark, TurboID light; six replicates per condition). The proximity labeling coupled to mass spectrometry revealed 186 proteins common to the TurboID-RPA2 light and TurboID-RPA2 dark conditions (Fig. 5c), including all three RPA subunits and several known interactors of RPA (Supplementary  Table 1). Importantly, the proximity labeling proteomics also revealed 53 proteins that were significantly enriched in the TurboID-RPA2 light versus TurboID-RPA2 dark conditions (P ≤ 0.05, fold change ≥ 1.5). With only four exceptions, these proteins were also specifically enriched in the TurboID-RPA2 light versus TurboID light comparison (Supplementary Table 2). Of note, among the RPA clustering-induced interactions were components of the BLM-TOP3A-RMI (BTR) complex together with its associated partners FANCM and RAD52 (Fig. 5d).
The BTR complex dissolves double Holliday junctions to prevent crossovers during homologous recombination and is recruited to stalled replication forks to promote fork restart 35,36 . Moreover, the BTR complex localizes to telomeres and, together with FANCM and RAD52, is involved in alternative lengthening of telomeres (ALT), a recombination-dependent telomere maintenance pathway used by telomerase-negative cancers to achieve replicative immortality [37][38][39] . We have previously found that ALT is associated with replication-stress-induced post-mitotic DNA synthesis (post-MiDAS) in G1 cells at genomic regions marked by RPA 21 . Parallel work has shown that ALT is a self-perpetuating process, which involves telomere clustering and RAD52-dependent telomere synthesis in PML bodies [40][41][42] . We therefore hypothesized that RPA self-assembly into nuclear condensates could be involved in telomere clustering and RAD52-mediated telomere elongation. In support of this possibility, liquid droplets formed in vitro by purified unlabeled trimeric RPA selectively enriched labeled WT RPA, but not a RPA mutant lacking the RPA2 N-IDR (ΔN), and also resulted in partitioning of purified RAD52 into the preformed RPA droplets (Extended Data Fig. 7a-c).
RPA and RAD52 co-localized at the telomeres of ALT-positive cancer cells marked by telomere repeat binding factor 2 (TRF2), and their association with telomeres was further enhanced by replication stress or when ALT activity was increased by FANCM depletion, as has been demonstrated previously 43,44 (Extended Data Fig. 8a-j). Next, we generated stable cell lines expressing siRNA-resistant WT RPA2 or mutated RPA2 encoding the phase-separation-impaired but ssDNA-binding-proficient S/T→D mutant. This allowed for efficient depletion and replacement of endogenous RPA2, despite its essentiality ( Supplementary Fig. 2a). The levels of GFP-RPA2 and RAD52 in these cells were unaffected by the RPA2 mutations ( Supplementary  Fig. 2b), and both cell lines showed comparable cell cycle profiles, similar responses to camptothecin (CPT)-induced DNA damage, and recruitment of RPA to telomeres (Extended Data Fig. 9a-g). However, the accumulated intensity of RPA in CPT-induced nuclear foci at sites of replication-fork stalling was lower in RPA S/T→D cells (Extended Data Fig. 9h). Similarly, the accumulated intensity of RPA in IR-induced foci was reduced in RPA S/T→D cells compared with cells expressing WT RPA (Extended Data Fig. 9i), indicating that RPA assembly was affected. Moreover, RPA S/T→D showed impaired self-interaction in co-immunoprecipitation (co-IP) experiments (Extended Data Fig. 9j), and RPA S/T→D cells had elevated markers of replication-stress-induced telomere fragility, including increased numbers of ALT-associated PML bodies (APBs), more DNA synthesis at telomeres outside S-phase in G2 and in G1, higher levels of ssDNA at telomeres, and more extrachromosomal C-circles ( Fig. 6a-e and Supplementary Fig. 3a-e). On the other hand, RAD52 enrichment at stressed telomeres was reduced in RPA S/T→D cells (Fig. 6f,g and Supplementary Fig. 3d). Similar defects were observed in conditions of unrestrained ALT activity due to FANCM depletion ( Supplementary  Fig. 4a-i).

Discussion
Principles of phase separation by associative polymers provide a framework to investigate liquid unmixing behavior of proteins in vitro and complex biomolecular condensates in different physiological and pathological conditions in vivo [46][47][48][49] . RPA is the main ssDNA-binding protein in mammalian cells and is essential for most DNA transactions, including replication, recombination, and repair. RPA-coated ssDNA is protected from unscheduled nucleolytic cleavage and degradation, secondary structure formation, and spontaneous breakage. Moreover, RPA bound to ssDNA serves as a binding platform for several genome caretakers and coordinates the handover of ssDNA to proteins involved in replication-fork processing, recombination, and repair 2 . Although RPA-coated ssDNA has often been viewed and depicted as linear 'beads on a string,' a dynamic exchange of RPA is required to prevent the formation of rigid filaments and allow other proteins to gain access to ssDNA 5,8 . In accordance with previous biochemical and structural studies 50-54 , our results confirm that the RPA complex is flexible and binds ssDNA in a highly dynamic manner. Additionally, we found that RPA also self-interacts dynamically, both in vitro and in the cell nucleus. RPA self-interaction is concentration-dependent and can be triggered by molecular seeds, such as in the light-controllable Cry2 system, at endogenous expression levels, or by its physiological binding substrate ssDNA. Of note, sub-stoichiometric amounts of ssDNA were sufficient to trigger RPA phase separation in vitro. Considering that nuclear RPA is in large excess over ssDNA under physiological conditions, these results indicate that ssDNA-seeded self-assembly of RPA into dynamic condensates is likely favored over stoichiometric RPA-ssDNA binding in vivo.
Biochemical experiments using single-molecule ssDNA curtains previously demonstrated that free RPA in solution was required for dynamic exchange of RPA with ssDNA 55,56 . Our results are consistent with these observations and suggest that nuclear RPA condensates provide a reservoir of highly concentrated free RPA in excess over the bound ssDNA, which enables rapid exchange of RPA molecules on the enclosed ssDNA. Dynamic condensate formation, initiated by ssDNA binding and amplified by the dynamic self-interaction properties of RPA, can thus explain how ultrahigh-affinity binding to ssDNA and rapid RPA exchange are both achieved simultaneously (Fig. 6j).
Mechanistically, we found that RPA2 is critical for RPA phase separation in the context of the functional RPA heterotrimer. Within RPA2, we identified the N-IDR to be involved in RPA phase separation. Bacterial SSB was recently reported to form phase-separated liquid condensates in vitro and in Escherichia coli cell extracts, dependent on a, Schematic of proximity-labeling mass spectrometry to identify RPA condensation-dependent interactions. A TurboID-mCherry-Cry2 cell line was used as negative control. A corresponding TurboID-RPA2-mCherry-Cry2 cell line was used with and without blue-light induction (15 minutes). Biotin proximity labeling was performed for 15 minutes. Six replicates per condition were analyzed by mass spectrometry for label-free quantification. b, Representative fluorescence microscopy images of the indicated cell lines, with and without blue-light induction. Scale bar, 10 μm. c, Venn diagram of specifically enriched proteins from proximity labeling mass spectrometry (fold change ≥ 1.5; P ≤ 0.05) in conditions without light-induced RPA condensation (RPA2 dark) versus with light-induced RPA condensation (RPA2 light). d, Volcano plot of differentially enriched proteins in RPA2 light versus dark conditions (fold change ≥ 1.5; P ≤ 0.05). RPA, BTR complex components, and BTR-associated proteins are highlighted in red. Note that BLM scored just below the fold change threshold. c,d, P values were assessed by moderated t-test from n = 6 replicates. Article https://doi.org/10.1038/s41594-023-00932-w its internal intrinsically disordered linker region 14 . In yeast, however, the RPA subunit Rfa1 (corresponding to human RPA1) exhibits tighter binding to ssDNA and slower motion inside subcellular foci than does the yeast DNA-repair protein Rad52 (BRCA2 in human), suggesting that yeast Rad52, but not Rfa1, displays phase-separation behavior 15,16 . Consistently, yeast Rfa2 (corresponding to human RPA2) failed to form optoDroplets in our experiments, although it remains possible that this would be different in a yeast cell environment. The yeast Rfa2 N-IDR shows only little sequence homology with the human RPA2 N-IDR, and the larger size of the human genome and the bigger volume of the cell nucleus may have required additional layers of control for the spatial and temporal regulation of subcellular compartmentalization. This requirement could be one of the reasons why human cells have acquired and preserved condensation properties of RPA2 that can be modulated by kinase-dependent phosphorylation of its N-IDR.
RPA phosphorylation has been studied in the context of its ability to bind and melt ssDNA structures and regulate DNA resection, and multiple DDR kinases and CDKs are involved in RPA2   phosphorylation and functionally co-operate in human cells to achieve RPA2 hyper-phosphorylation [57][58][59][60][61][62] . Our results suggest that multi-site phosphorylation of RPA2, without abolishing ssDNA binding 32,34 , can gradually reduce the phase-separation capacity of RPA, indicating that phosphorylation may fine-tune RPA condensation properties and condensate-related functions in a context-dependent manner. Functionally, unbiased proximity proteomics revealed that RPA condensation leads to an enrichment of the BTR complex and its associated proteins FANCM and RAD52. As part of the BTR complex, RAD52 is involved in ALT in cancer 63 . We found that RAD52 readily partitions into phase-separated RPA droplets in vitro, and that RAD52 enrichment at sites of ALT activation is reduced in cancer cells expressing phase-separation-impaired, phosphomimetic RPA2. Rather than loss of APB formation, RPA S/T→D cells with reduced RAD52 recruitment to telomeres showed elevated C-circles, in accordance with a shift towards RAD52-independent ALT associated with C-circle formation and progressive telomere shortening 64 . In line with this, we observed more frequent telomere loss in cells expressing phosphomimetic RPA2. Considering that we also observed signs of impaired telomere clustering, homology search and donor template usage might be altered by RPA hyper-phosphorylation, for example, towards more intratelomeric recombination. Further studies employing dedicated assays to interrogate ALT sub-pathway usage will be needed to investigate this hypothesis. In summary, our findings suggest that the self-assembly and condensation properties of RPA functionally contribute to telomere maintenance in ALT-positive cancer cells, consistent with an emerging implication of phase separation at telomeres 65 and specifically in ALT [40][41][42] . Given that telomere maintenance by ALT represents a vulnerability of certain cancers, understanding the condensation properties of the involved molecules may help to improve targeted therapies against ALT-dependent tumors. Beyond ALT, RPA condensation properties may also be involved in other cellular contexts, including DNA-repair compartments formed at DSBs that undergo resection for repair by HR and stressed replication factories. We observed fast recovery in FRAP experiments at these regions, and although RPA foci formation was not abrogated in RPA S/T→D cells, consistent with the mutant being proficient in ssDNA binding, the accumulated intensity of RPA in these regions was reduced. These findings are consistent with a more general role of RPA condensation in genome function, although testing the implications of RPA's phase-separation properties in additional cellular contexts will require further studies.

Cell culture
All cell lines were grown at 37 °C under standard cell culture conditions (humidified atmosphere, 5% CO 2 ) in Dulbecco's modified Eagle's medium (DMEM) containing 10% fetal bovine serum (Gibco) and 1% penicillin-streptomycin antibiotics. Stable U-2 OS GFP-RPA2 WT and S/T→D cell lines (siRNA-resistant) were maintained in the presence of 400 μg/ml Geneticin (Gibco). For experiments with siRNA-resistant GFP-RPA2 WT and S/T→D cells, the endogenous RPA2 was transiently depleted by siRNA. Stable U-2 OS cells expressing GFP-RPA2 and TRF2-RFP were maintained in the presence of 400 μg/ml Geneticin (Gibco) and 100 μg/ml hygromycin B (Thermo Fisher Scientific). Stable U-2 OS TurboID-RPA2-mCherry-Cry2 and TurboID-mCherry-Cry2 cells were maintained in the presence of 100 μg/ml hygromycin B (Thermo Fisher Scientific) and expression of the constructs was induced by the addition of 1 μg/ml doxycycline (Sigma-Aldrich) for 24 hours. Stable U-2 OS GFP-RPA cells expressing a polycistronic GFP-RPA construct for stoichiometric expression of the three RPA subunits RPA1, RPA2, and RPA3 (ref. 20 ) were maintained in presence of 400 μg/ml Geneticin (Gibco). U-2OS mScarlet-RPA1 cells, stably expressing endogenously tagged RPA1 and ectopic GFP-tagged 53BP1 (ref. 21 ), were maintained in the presence of 0.5 μg/ml puromycin (Sigma-Aldrich) and 5 μg/ml blasticidin (InvivoGen). All cell lines used in this study were grown in sterile conditions and routinely tested for mycoplasma contamination and are listed in Supplementary Table 4. The generation of stable cell lines is described in the Supplementary Methods.

Cloning
Cloning was performed using chemically competent DH5α generated in house, derived from Library Efficiency DH5α Competent Cells (Thermo Fisher Scientific). All constructs were generated by two-piece Gibson assembly or by site-directed mutagenesis. All primers and the cloning strategy used for expression constructs are provided in the Supplementary Methods (sections: cloning of Cry2-mCherry-fusion constructs; cloning of TurboID-RPA2-mCherry-Cry2, cloning of GFP-RPA2 mutants, cloning of TRF2-RFP, cloning of p11d-tRPA-32ΔN) and Supplementary Table 5. Correct cloning and integration into target vectors were confirmed by sequencing. All plasmids used in this study are listed in Supplementary Table 6.

siRNA and plasmid transfections
Individual siRNA transfections (siRPA1, s12127; siRPA2, s12130; siFANCM, s33621; final concentration of 25 nM) were performed with Ambion Silencer Select siRNAs using Lipofectamine RNAiMAX (Thermo Fisher Scientific). Negative Silencer Select control Neg2 from Ambion was used as a non-targeting control. Plasmid transfections for transient expression were performed with TransIT-LT1 (Mirus Bio), according to the manufacturer's instructions.

EdU labeling
For pulsed EdU (5-ethynyl-2′-deoxyuridine, Thermo Fisher Scientific) labeling, cells were incubated for 20 minutes in medium containing 10 μM EdU. The Click-iT EdU Alexa Fluor Imaging Kit (Thermo Fisher Scientific) was used for EdU detection.

Immunofluorescence staining
Immunofluorescence staining was performed as previously described 21,67 . Specifically, cells were grown on sterile glass coverslips, fixed in 3% formaldehyde in PBS for 15 minutes at room temperature, and permeabilized for 5 minutes in 0.2% Triton X-100 (Sigma-Aldrich) in PBS. Primary and secondary antibodies were diluted in filtered DMEM containing 10% FBS and 0.02% sodium azide, and antibody incubations were performed for 1-2 hours at room temperature. Coverslips were incubated for 10 minutes with PBS containing 4′,6-diamidino-2-phenylindole dihydrochloride (DAPI, 0.5 μg/ml) at room temperature, subsequently washed three times in PBS and briefly submerged in distilled water before being mounted on glass slides with Mowiol-based mounting medium (Mowiol 4.88 in glycerol/TRIS). To stain RAD52, cells were fixed in ice-cold methanol for 20 minutes at −20 °C and processed as described above without the permeabilization step. Antibody information is provided in the Reporting Summary.

Quantitative image-based cytometry
Automated multichannel widefield microscopy for quantitative image-based cytometry (QIBC) was performed as described previously 18,21,67 on an Olympus ScanR Screening System (ScanR Image Acquisition 3.01) equipped with an inverted motorized Olympus IX83 microscope, a motorized stage, IR-laser hardware autofocus, a fast emission filter wheel with one set of bandpass filters for multi-wavelength acquisition (DAPI (ex BP 395/25, em BP 435/26), FITC (ex BP 470/24, em BP 511/23), TRITC (ex BP 550/15, em BP 595/40), Cy5 (ex BP 640/30, em BP 705/72)), and a Hamamatsu ORCA-FLASH 4.0 V2 sCMOS camera (2048 × 2048 pixel, pixel size 6.5 × 6.5 μm) with a ×20 UPLSAPO (NA 0.75) air objective. Image information of cell populations was acquired under non-saturating conditions, and identical settings were applied to all samples within one experiment. Images were analyzed with the Olympus ScanR Image Analysis Software (version 3.0.1), a dynamic background correction was applied, and nuclei segmentation was performed using an integrated intensity-based object-detection module based on the DAPI signal. Foci segmentation was performed using an integrated spot-detection module. Downstream analyses were focused on properly detected interphase nuclei containing a 2N-4N DNA content as measured by total and mean DAPI intensities with comparable quantified GFP expression. Fluorescence intensities are depicted as arbitrary units. Color-coded scatter plots of asynchronous cell populations were generated with Spotfire data visualization software (version 7.9.1 and 10.10.1, TIBCO). For visualization of discrete data in scatter plots, mild jittering (random displacement of data points along the discrete data axes) was applied to demerge overlapping data points. Representative scatter plots are shown.

Cry2 optoDroplet experiments
Cry2 optoDroplet experiments were performed as described previously 18 . Specifically, U-2 OS cells were seeded into a 96-well plate (Greiner μclear), and 24 hours prior to imaging, they were transfected with plasmid DNA. During live-cell microscopy, FluoroBrite DMEM supplemented with 10% fetal bovine serum (Gibco) and Glutamax (Thermo Fisher Scientific) was used. Time-lapse microscopy of optoDroplet formation upon blue-light exposure was carried out in temperature-and CO 2 -controlled conditions (37 °C, 5% CO 2 ) on a GE Healthcare IN Cell Analyzer 2500HS (V7.4) with a PCO sCMOS 16-bit camera (2048 × 2048 pixels, pixel size 6.5 × 6.5 μm) using a CFI Plan Apo Lambda (NA 0.75) ×20 air objective at 15-second intervals for 6 minutes (25 ms ex BP 475/28, em BP 526/52; 100 ms ex BP 575/25, em BP 607.5/19). For optoDroplet quantification in Figure 1c and Extended Data Figure 6h, images from time-lapse microscopy were analyzed with the Olympus ScanR Image Analysis Software (version 3.0.1), a dynamic background correction was applied, and single-cell segmentation was performed using an integrated intensity-based object-detection module based on the mCherry signal. Droplet segmentation was performed using an integrated spot-detection module. Upon blue-light exposure, transfected cells were either kept in the dark to serve as negative controls or were exposed to 20 cycles of 5 seconds of blue light and 15 seconds of dark in a custom-made blue-light box equipped with eight 1-W LED lamps with a power of 500 Lm at 10-cm distance from the cells. Cells were then fixed in 3% formaldehyde in PBS for 15 minutes at room temperature, and stained with DAPI. Imaging and image analyses were performed on the Olympus ScanR Screening System, as described above. Expression levels were normalized between samples for each experiment. The relative optoDroplet formation per construct was defined as the

Trimeric RPA purification
The trimeric human RPA complex was expressed in chemocompetent BL21 cells using p11d-tRPA(123) for expression of the WT protein (Addgene plasmid no. 102613, kindly provided by M. Wold 25 ), p11d-tRPA-32Asp8 for expression of the phosphomimetic RPA2 S/T→D mutant (Addgene plasmid no. 102617, kindly provided by M. Wold 32 ), and p11d-tRPA-32ΔN for the expression of RPA2 lacking the disordered N-terminal domain. The multi-step purification procedure was performed as described previously 24,25 . Transformed BL21 cells were grown in LB medium supplemented with 100 μg/ml ampicillin (Sigma-Aldrich) at 37 °C to an optical density of 0.6. Expression was induced by 0.4 mM IPTG following overnight incubation at 18 °C. Bacteria were collected by centrifugation, resuspended in lysis buffer (30 mM HEPES pH 7.5, 0.01% NP-40, 25 μM EDTA, 1 mM DTT, 10% glycerol, 1× complete protease inhibitor cocktail (Roche)) and lysed using a French press. Protein purification was performed on an ÄKTA Purifier (GE Healthcare). The lysate was first loaded onto a HiTrap Blue column (GE Healthcare), then washed sequentially with lysis buffer at different salt concentrations (50 mM KCl; 800 mM KCl; 400 mM NaSCN), and finally eluted with lysis buffer containing 1.5 M NaSCN. Peak fractions were pooled and loaded onto a HiTrap Desalting column (GE Healthcare). After desalting, the peak fractions were loaded onto a HiTrap Q column (GE Healthcare) and washed sequentially with lysis buffer at different salt concentrations (50 mM KCl; 87.5 mM KCl; 200 mM KCl), before final fractions were eluted in lysis buffer using a salt gradient from 200 mM KCl to 500 mM KCl. The collected protein fractions were subjected to SDS-PAGE using a 12% gel for Coomassie staining. Amicon Ultra 3 kDa MWCO centrifugal tubes (Millipore) were used to concentrate the peak fractions and for buffer exchange to diluted Sørensen buffer (43.4 mM Na 2 HPO 4 , 6.6 mM KH 2 PO 4 , pH 7.6) supplemented with 150 mM KCl, 25 μM EDTA, and 1 mM DTT. Protein concentrations were measured by Bradford assay. Freshly purified protein, kept at 4 °C, was used for in vitro assays.

Preparation of RAD52
The sequence encoding human RAD52 was ordered from GenScript as codon-optimized for E. coli, and was cloned into pMALT-P (a kind gift of the Kowalczykowski laboratory, UC Davis) using BamHI and PstI restriction sites. Additionally, a His-tag was inserted before the MBP-tag to yield the final construct pMALT-His-MBP-PP-hRAD52co. RAD52 was then expressed in E. coli, upon induction with 0.5 mM IPTG and incubation overnight at 18 °C. RAD52 was purified by affinity chromatography using amylose resin, the MBP-tag was subsequently cleaved by PreScission protease. RAD52 was then applied on HiTrap Heparin column in 20 mM Tris-HCl pH 7.5, 1 mM EDTA, 0.5 mM DTT, 10% glycerol, and 100 mM NaCl. The column was eluted using a salt gradient from 100 mM NaCl to 600 mM NaCl. The pooled eluted fractions were briefly incubated with NiNTA resin to remove uncleaved His-MBP-RAD52 from the cleaved RAD52 protein, and the final sample was dialyzed into 20 mM Tris-HCl pH 7.5, 1 mM DTT, 10% glycerol, and 100 mM NaCl. Four liters of bacterial culture yielded 1 ml of 39 μM RAD52.

Protein labeling
For fluorescence labeling of trimeric RPA complexes, the purified protein complexes were incubated with Sulfo-Cyanine3 maleimide (Luminoprobe) using an excess molar ratio of 40:1 (label:protein) overnight at 4 °C. The labeled protein complex was then mixed 1:9 with the unlabeled protein complex for in vitro droplet assays. For Extended Data Figure 7a,b, proteins (RPA WT, RPA ΔN, RAD52) were incubated with Sulfo-Cyanine3 maleimide (Luminoprobe) using an excess molar ratio of 100:1 (label:protein) overnight at 4 °C. Then, 1 μM labeled protein was added to preformed unlabeled RPA droplets.

Oligonucleotide hybridization
To generate a double-stranded DNA oligonucleotide, the 40-nucleotide ssDNA (Supplementary Table 3) was mixed with its reverse and complementary oligonucleotide at a 1:1 molar ratio. For annealing of the two strands, the mixture was incubated for 10 minutes at 80 °C and then slowly cooled down. To verify oligonucleotide annealing, the product and the two ssDNA oligonucleotides were subjected to 2% agarose gel electrophoresis and RedSafe (Lucerna-Chem) detection. Redsafe signals were acquired with Infinity ST5 Xpress.

In vitro droplet experiments
In vitro droplet experiments were performed in a 384-well imaging plate (Greiner μclear) at room temperature in diluted Sørensen buffer (43.4 mM Na 2 HPO 4 , 6.6 mM KH 2 PO 4 , pH 7.6) supplemented with 150 mM KCl, 25 μM EDTA, and 1 mM DTT. Images were acquired using a ×63 HC PL APO corr CS2 oil objective (NA 1.4) on a Leica SP5 UV-VIS or Leica SP8 inverse FALCON confocal laser scanning instruments (Leika Application Suite X 3.5.  Table 3. For quantification of droplet formation, labeled tRPA (9:1 unlabeled:labeled protein) was incubated with nucleotides, and condensate formation was assessed in quadruplicates by automated droplet quantification of the Cy3-labeled tRPA signal at three matched time-points per condition and per replicate using Olympus ScanR Analysis Software (version 3.2).

Turbidity measurements
For turbidity measurements, the absorbance at 600 nm was measured in quadruplicates using a Tecan microplate reader (Tecan i-control 2.0). Protein mixtures were prepared in diluted Sørensen buffer (43.4 mM Na 2 HPO 4 , 6.6 mM KH 2 PO 4 , pH 7.6) supplemented with 150 mM KCl, 25 μM EDTA, 1 mM DTT, and 4% PEG-8000 in a 384-well plate (Greiner μclear). The single-stranded oligonucleotides used for turbidity measurements are listed in Supplementary Table 3. Turbidity measurements were normalized to the negative control condition.

Co-immunoprecipitation
U-2 OS and U-2 OS GFP-RPA2 WT and S/T→D cells were washed twice with PBS and directly lysed on ice in 500 μl TNE buffer (50 mM Tris-HCL pH 8.0, 150 mM NaCl, 0.1% Igepal CA630, 1 mM EDTA) supplemented with 2 mM MgCl 2 , cOmplete inhibitor cocktail (Roche), phosphoSTOP (Roche), and 25 U/ml benzonase. Cell lysates were incubated for 5 minutes at room temperature and then centrifuged at 15,000g for 15 minutes. Then, 600 μg of cell lysate was incubated with 0.8 μg of rabbit anti-GFP antibody (Torrey Pines biolabs, TP401) for 3 hours at 4 °C. A 20-μl slurry of protein G-sepharose beads (GE Healthcare, 17-061801) was added per sample for a 1-hour incubation period at 4 °C. The beads were collected by centrifugation, washed five times in TNE buffer, and eluted by boiling in 10× SDS-PAGE loading buffer. Samples were subjected to SDS-PAGE and immunoblotting.

Immunoblotting
Proteins were separated by standard SDS-PAGE and transferred onto PVDF membranes. Membranes were incubated for 1 minute at room temperature with Ponceau solution, washed three times with PBS-T Article https://doi.org/10.1038/s41594-023-00932-w (PBS + 0.1% Tween-20) and then blocked with 5% milk in PBS-T for 2 hours at room temperature. Primary antibodies were incubated overnight at 4 °C in blocking solution. Membranes were then washed three times with PBS-T and incubated with HRP-conjugated secondary antibodies for 1 hour at room temperature. Membranes were washed again three times with PBS-T, and protein signals were detected using ECL Western Blotting Detection Reagent (Amersham) and acquired with Canon MP Navigator EX. Antibody information is provided in the Reporting Summary.
Additional methods are provided in the Supplementary Information.

Statistics and reproducibility
No samples were measured repeatedly for statistical analysis. Two-tailed unpaired t-test, one-way analysis of variance (ANOVA) with Dunnett`s test, one-way ANOVA with Tukey test, or two-way ANOVA with Šidák's test was performed as indicated in the figure legends using GraphPad Prism (Versions 5, 8 and 9). Moderate t-test 68 was calculated using R package limma 69 . Sample sizes and statistical tests used are specified in the figure legends. All experimental findings were confirmed by independent repetitions. Data shown in Figures 1b-e, 2a-h, 3a-d, 4a-i, 5b, and 6a-i, Extended Data Figures 1a, 2c, 4a,b,i, 5a,b,f-h, 6i,j, 8a-j, 9a,b,g,i,j, and 10a-c,f, Supplementary Figures 1b,c, 2a, 3a-d, and 4a-d,f-i were confirmed in at least three independent experiments; data shown in Extended Data Figures 1b,c, 2b,d,e, 3b,c, 4c-h,j, 5c-e,i, 6b-h, 7a-c, 9c-f,h, 10d,e and Supplementary Figures 1d-g, 2b, 3e, and 4e were confirmed in at least two independent experiments. Label-free proximity labeling mass spectrometry was performed once with six technical replicates.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
The mass spectrometry proteomics data were analyzed using Homo Sapiens UniProt reference proteome database (taxonomy 9606; canonical version from 20190709), reversed decoy-database, and database of common protein contaminants. The mass spectrometry proteomics data have been deposited together with the reversed decoy-database and the database of common protein contaminants to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD036935. All other data are available from the corresponding author upon request. Source data are provided with this paper. Fig. 2 | The trimeric RPA complex forms dynamic intracellular optoDroplets. a, Schematic of the polycistronic Cry2-mCherry-tRPA optoDroplet construct possessing P2A and T2A cleavage sites for stoichiometric expression of all three RPA subunits. b, Western blot analysis of U-2 OS cells transiently transfected with Cry2-mCherry or Cry2-mCherry-tRPA as indicated. c, Timeresolved optoDroplet formation of trimeric RPA. Representative stills from live-cell microscopy are shown. d, Accumulated optoDroplet intensity per nucleus of trimeric RPA or RPA2 fused to Cry2-mCherry was analyzed and normalized to the average accumulated optoDroplet intensity of the corresponding dark condition. Two-way ANOVA with Šidák`s test, tRPA **** p < 0.0001, RPA2 *** p = 0.0003. e, Nuclear mean intensities of Cry2-mCherry in cells analyzed in (d). d-e, Averages and standard deviations for n = 3 independent samples are shown (cell number: RPA2 dark n 1 = 429, n 2 = 507, n 3 = 433; RPA2 light n 1 = 361, n 2 = 452, n 3 = 411; tRPA dark n 1 = 305, n 2 = 412, n 3 = 304; tRPA light n 1 = 498, n 2 = 589, n 3 = 469). Scale bars 10 μm.

Extended Data
Article https://doi.org/10.1038/s41594-023-00932-w Extended Data Fig. 7 | RPA condensates concentrate RAD52 in vitro. a, Purified RAD52 partitions into RPA droplets in vitro. Purified trimeric unlabeled RPA was allowed to form liquid droplets prior to incubation with Cy3-labeled purified RAD52. Time-resolved partitioning of RAD52 into RPA droplets was followed by time-lapse microscopy. Cy3-labeled RPA served as positive control, confirming homotypic RPA interactions in phase separated RPA droplets. The Cy3 label added to preformed RPA droplets served as negative control. Cy3-labeled RPAΔN, lacking the N-IDR of RPA2, served as additional negative control and confirmed the involvement of the N-IDR in homotypic RPA interactions. Representative stills from time-lapse microscopy are shown. b, Purified RAD52 alone is not forming phase separated droplets in the conditions used in (a). c, Coomassie staining of the purified human RAD52 after SDS-PAGE. Scale bars 10 μm.