Regulation of transcription by R loop-nucleated transcription condensate assembly

Abstract

Prevalence of promoter paused R loops formed from frequent RNA PolII pausing in eukaryotic cells not only pose a significant threat to genomic stability, but also a potent regulator of transcription. In this review, we discuss how R loops act as both a product and regulator of transcription. Direct effects of R loops on transcription output are not well established, as studies on R loops are mainly focused on genomic instability triggered by accumulation of R loops. We dissect the evidence showing how R loops regulate transcription through a new manner: transcription condensates. Transcription condensates are dynamic, membraneless assemblies enriched with components of the transcription machinery, transcription factors, coactivators, and RNA-binding proteins (RBPs) that form at regulatory loci of critical, highly expressed genes to promote transcription rapidly and robustly. Coinciding at these condensate-forming regulatory loci are R loop hotspots, but this relationship and how they impact transcription as a unit remain to be defined. Recent developments suggest that R loops may act as multivalent seeds for transcription condensates through its modular structure, and as a binding platform for numerous RBPs containing phase separation-prone disordered regions. Interestingly, R loop processing may act as a double-edged sword to transcription: with partitioning of some factors into the transcription condensate promoting transcription, while others terminating transcription as a trade-off for protecting genomic integrity against pervasive R loops. Finally, we also summarize pathological conditions involving alterations of factors implicated in both R loop homeostasis and condensate dysregulation.

Introduction

R loops are triple-stranded structures consisting of a DNA:RNA hybrid and a displaced ssDNA strand first visualized in 1976 [1]. Under native conditions, it is widely believed that the formation of R loops are the result of the “thread back” model, which suggests that the underwound DNA from a translocating RNA polymerase allows the nascently-synthesized RNA strand to anneal with the template strand, leaving a displaced single stranded DNA (ssDNA), especially since the hybridization of DNA:RNA is more thermodynamically stable compared to DNA:DNA [2] (see Fig. 1).

Fig. 1.

Disruption of R loop homeostasis leads to pathological conditions associated with genomic instability. R loops are formed at the promoter-proximal transcription start sites, where negative supercoiling from RNA PolII progression and the GC-rich sequences permit the binding of the nascently transcribed RNA (pink strand) onto the DNA template (blue strand) through a “thread back” model. Displaced ssDNA in the R loop is vulnerable to alterations and nicks, which can be converted into genotoxic DNA breaks. If unresolved by R loop regulators (green), pervasive R loops can act as physical barriers that hinder the progression of the replication machinery during S phase (orange) or subsequent rounds of transcription (blue). Altogether, DNA breaks, replication stress, and transcription stress contribute to genomic instability. However, the effects of pervasive R loops as a transcription roadblock and hence their direct effect on transcription is not as well established.

R loops are prevalent in regulatory regions and gene bodies of transcribed genes, while scarce (0.01–0.1%) at untranscribed and intergenic regions, averaging at 5% of the mammalian genome [3]. While R loop formation occurs during physiological processes like transcription, replication, class-switch recombination (CSR), etc. (“scheduled” R loops), human pathologies like neurodegeneration and cancer are linked to formation of pathological or “unscheduled” R loops that trigger replication stress or DNA damage [4, 5]. Formation of R loops per se does not pose a threat to genomic stability, rather it is when R loops are not resolved in a timely manner to allow accumulation and persistence (increased half-life) that leads to DNA damage and hence genome instability. However, this means that scheduled R loops, if not resolved, can be converted to unscheduled R loops. Due to the structure of R loops, the displaced ssDNA is susceptible to modifying enzymes, nucleases, and genotoxins, making these R loop-forming regions prone to transcription-associated mutagenesis or recombination [5, 6] (Fig. 1). R loops compromise genomic integrity primarily through replication stress, as R loop-mediated damage manifests primarily in the S phase [7] and is attenuated by RNaseH overexpression [8]. When the replication and transcription machinery collide (R-T conflicts), particularly in a head-on orientation, accumulation and stability of detrimental R loops can be enhanced [9]. Due to their pervasive status as a genomic threat, studies on R loops have mainly focused on the maintenance of genomic stability through the regulation of R loops, frequently overlooking its role in impacting transcription (Fig. 1). R loops indirectly influence transcription by acting as a binding substrate for transcription factors or epigenetic modulators [10–14] but it can also directly affect transcription due to its unique structure, acting as physical blockades [15–18].

In this review, we focus on co-transcriptional R loops as a direct regulator of transcription, and a new mechanism of R loop-dependent transcription regulation: formation of transcription condensates. At promoters, R loops act as binding platforms for a number of RNA-binding proteins (RBPs) and RNA helicases, which contain phase separation-prone disordered regions that interact with themselves (homotypic) and other factors (heterotypic) through weak, multivalent interactions. Furthermore, these RBPs and RNA helicases have been implicated in regulation of transcription, through a mechanism requiring condensate formation.

R loops as transcription intermediates

R loops are products of transcription. Genome-wide analysis of R loops using different methods like DNA: RNA immunoprecipitation (DRIP/DRIPc-seq) [3, 19], R loop ChIP (R-ChIP) [20], R loop CUT&Tag [21], and the newly developed R loop identification assisted by nucleases and sequencing (RIAN-seq) [22] indicate that R loops are enriched at promoter proximal regions adjacent to transcriptional start sites (TSS) and to a lesser extent, transcription termination sites (TTS). Castillo-Guzman and Chédin classifies co-transcriptional R loops into two classes: promoter-paused Class I R loops and elongation-associated Class II R loops [23].

Class I R loops, which are enriched using near-physiological conditions to capture native R loops such as the R-ChIP and CUT&Tag method, are prominent R loops formed at the promoter proximal region that overlap with RNA Polymerase II (PolII) and PRO-seq signals [21], strongly suggesting promoter proximal pausing of PolII. After initiation, PolII proceeds downstream of the TSS and begins synthesis of short ~30–50 bp nascent RNA transcripts before it pauses and awaits elongation cues [24]. A combination of the high GC skew at the promoter, nucleosome depletion, and underwound DNA due to the progression of the transcription machinery stabilizes the R loop [23].

Meanwhile, the widely used DRIP-seq method involves an ex vivo DNA extraction that involves treatment with strong detergents and deproteinization, followed by an overnight digestion with restriction enzymes, coupled with the size selection during library construction (> 100 bp), could attribute to the significant losses of R loop signals at the TSS region [23]. At the same time, these harsher conditions could have allowed the less abundant, but stable and long, Class II elongation-associated R loops to be detected more readily [23]. One major caveat however, is that results generated with the DRIP-seq method typically have high background and high inter-study heterogeneity [25].

Importantly, the frequency and the association with a paused or backtracked RNA PolII in Class I promoter-paused R loops suggest that these R loops are most likely to be converted into unscheduled, pathological R loops arising from replication-transcription conflicts (see Fig. 1). Indeed, mutation or depletion of R loop regulators leading to Class I R loop accumulation trigger replication stress-induced DNA damage [16, 26–30].

Due to the importance of R loop homeostasis, various factors play a role in its regulation. In general, R loop regulators either prevent or resolve R loops (see Fig. 2). R loop prevention is mediated by RBPs such as messenger ribonucleoprotein (mRNP) biogenesis and splicing factors that process nascent RNA and prevent its hybridization into the DNA template [26, 31–34], topoisomerases that resolve R loop-favoring negative supercoiling [19, 35–39], and chromatin remodelers that modulate DNA availability through nucleosome occupancy [31, 40]. R loop resolvers are either helicases that unwind the R loop helix or nucleases that digest R loops such as RNaseH1 [41–43], RNaseH2 [44], or XRN2 [45, 46]. R loop helicases consist of DEAD-box RNA helicases (DDX1 [47], DDX5 [48, 49], DDX21 [50, 51], DDX23 [28], DDX39B [52], DDX41 [53], etc.), DExH-box RNA helicases (BLM [54], WRN [55], DHX9 [32, 38], ARIP4 [15]), SETX [13, 45, 56], AQR [57, 58], and many others. DNA repair factors have also been shown to participate in the suppression of R loops, either directly like Fanconi Anemia (FA) proteins FANCI-FANCD2 [59] or Nucleotide Excision Repair (NER) proteins XPF and XPG [57], or indirectly like BRCA1 [30], BRCA2 [16], and MRE11-RAD50-NBS1 complex [60]. Furthermore, histone acetylation at H4K8 by PCAF has been shown to mediate R loop resolution by recruiting MRE11 and EXO1, which then recruits FA helicases BLM and FANCM [61], demonstrating that R loops and R loop-dependent genomic integrity can be regulated at the histone level. Recent evidence also indicates that R loops can be regulated through formation of biomolecular condensates by RBPs, which are discussed in the next sections.

Fig. 2.

Regulation of transcription at transcription start sites through direct R loop resolution. Pervasive R loop formation at the promoter proximal region acts as physical barriers to subsequent rounds of transcription. To prevent excessive stalling of PolII and ensure timely transcription output, R loops are prevented or resolved. R loop formation can be prevented either by maintaining DNA topology or direct binding by RNA binding proteins (RBPs) or processing (splicing, capping, etc) factors. R loops can also be resolved by enzymes such as helicases and nucleases. At the expense of transcription, genomic integrity may be preserved by premature termination of transcription at the promoter.

R loops as regulators of transcription

Indirect mechanisms

How R loops regulate transcription indirectly, mediated through different effectors has been reviewed previously [62]. Notably, R loops and transcription are inextricable from RNA PolII pausing sites, as both the TSS and the TTS where R loops tend to form are hotspots of R loop-mediated transcription regulation. At the TSS, R loops can maintain local open chromatin to facilitate transcription factor binding [10] and promote enrichment of activating chromatin modifiers [11] while preventing repressive marks [11–13]. Indeed, RIAN-seq, which allows for unbiased chemical labeling and antibody-free detection of genome-wide R loops, demonstrates that promoter R loops have a conserved motif that are bound by zinc finger transcription factors that promote transcription [22]. Additionally, recognition of R loops by RBPs has also been implicated in transcription regulation. Specific to the SEMA3F promoter, the mRNA-binding protein IGF2BP specifically recognizes m⁶A-modified R loops and its occupancy repels DNMT1 to promote transactivation [14]. At the TES, RNA PolII pauses downstream of the polyadenylation sequence, giving rise to R loops. Unwinding of these R loops by SETX and DHX9, and subsequent degradation by XRN2, leads to termination and release of PolII [45, 63].

With respect to RNA PolII pausing as discussed in the previous section, some studies suggest that PolII gives rise to promoter-paused R loops, but these R loops likely do not largely affect transcription through PolII pause release. For example, Chen and colleagues show that R-ChIP coupled with GRO-seq of samples treated with a reversible RNA PolII pause release inhibitor DRB indeed leads to the accumulation of R loops and nascent transcripts at the TSS [20]. However, when DRB is washed out to release paused PolII, nascent transcripts are rapidly depleted while R loop levels show a more gradual depletion, indicating that R loop resolution is not a prerequisite for pause release, at least not in all genes globally [20]. The same authors then studied myelodysplastic syndrome-associated mutations in splicing factors SRSF2, U2AF1, and U2AF2 and showed an upregulation in both promoter-associated R loops and GRO-seq signals. However, they showed that this is due to defects in extracting P-TEFB from the inhibitory 7SK complex hence suppressing PolII pause release [26]. Furthermore, depletion of BRCA1 and BRCA2 lead to hyperaccumulation of promoter-paused RNA PolII and R loops, manifesting from the disruption of interaction with elongation factors COBRA1 [30] and PAF1, [16] respectively.

Direct mechanisms

On the other hand, other studies imply that resolution of promoter R loops also directly affects transcription in cis. Belotserkovskii et al. showed that formation of cotranscriptional R loops in vitro blocks subsequent rounds of transcription [17], suggesting that the lack of R loop resolution, while not directly impacting the progression of the PolII molecule producing the R loop, could hinder the subsequently-loaded PolII progression and hence, transcriptional output. While Belotserkovskii and colleagues suggest that R loops formed from non-template G-rich sequences downstream of the TSS negatively affect successive rounds of transcription, Lee et al. proposed that R loops formed from a non-template potential G4 quadruplex forming sequence (PQS) downstream of the TSS conversely promotes subsequent rounds of transcription [18]. In this study, they show that the PQS ensures successive R loop formation by preventing the DNA from hybridizing through the G4 formation, with the R loop being formed and replaced with every round of transcription. The authors reason that the G4-mediated R loop stability maintains an open state or a more favorable structure for the invasion of RNA PolII, hence enhancing transcription output [18]. However, the conclusions in the two studies are not irreconcilable. Lee et al. demonstrated that the R loop-mediated transcription enhancement only occurs when it forms further downstream (> 40 bp) from the TSS. Both studies [17, 18] found a blockage in transcription when the R loop forms immediately downstream (< 10 bp) of the TSS, a region most prone to R loop formation [18], underlining the importance of R loop positioning on transcription modulation.

Furthermore, Sridhara et al. artificially induced the accumulation of R loops at a few genes by depleting THOC, which results in the recruitment of the RNA helicase DDX23 and the hyperaccumulation of paused PolII, both of which are sensitive to RNaseH digestion [28]. Therefore, the authors propose a model where R loop accumulation triggers the pausing of PolII, which then drives the accumulation of DDX23 at the R loop site where it resolves these R loops (Fig. 1). Recently, we also demonstrated that the elusive helicase ARIP4 suppresses promoter-paused R loops at highly transcribed AR genes, leading to defects in their androgen-dependent transcription [15] (Fig. 2). This has also been demonstrated in the gene body, where the lack of mRNP biogenesis from the lack of THO/TREX leads to the persistence of R loops, which then act as obstacles for the next elongating RNA PolII [64] (Fig. 2). At the TSS, the Sensor of Single-stranded DNA-Integrator-PP2A (SOSS-INTAC) complex recognizes promoter-paused R loops through the SSB1 subunit of SOSS, and the INTS11 subunit of INTAC digests R loops to promote termination of transcription at the promoter [65] (Fig. 2). Importantly, this R loop suppression is dependent on the liquid-like condensation of SOSS-INTAC [65]. This finding, together with other studies, link R loops to phase separation-driven condensate formation.

In the next sections, we further discuss a new mechanism by which R loops regulate transcription through assembly of the transcription condensate. In the next section, we introduce liquid-liquid phase separation (LLPS) and the assembly of biomolecular condensates, how transcription condensates are formed and regulated, and how transcription condensates promote the transcription process. Subsequently, we examine how the R loop structure, formation in the genome, and R loop-binding proteins drive condensate formation to regulate transcription.

Transcription condensates: how condensate formation promotes transcription

Fundamentals of biomolecular condensates

To perform specific functions, cellular components are compartmentalized into chemically distinct substructures. This is most evident in organelles enclosed in a definitive boundary like a lipid membrane, while membraneless organelles (MLOs) can sequester as biomolecular condensates through LLPS, which is akin to oil and water demixing [66–68]. These MLOs are micron-sized assemblies typically consisting of proteins and nucleic acids found in both the cytoplasm and the nucleus, like the nucleoli, Cajal bodies, splicing speckles, P granules, stress granules, and many others. LLPS is a process by which a homogeneous solution of biomolecules separates into two distinct liquid phases: a dense phase with high concentrations of the biomolecule and a dilute phase depleted of the biomolecule [66]. Phase separation into condensates can be tuned by various external factors like temperature, pH, and ionic strength [69], and internal factors such as interactions among proteins, DNA, and RNA [67, 68]. Typically, proteins containing intrinsically disordered regions (IDRs) mediate weak, multivalent self-interactions critical for phase separation. Past a threshold protein concentration called the saturation concentration (C_sat), it becomes energetically favorable for IDR-containing proteins to phase separate [67, 68]. For example, IDR-containing proteins that phase separate include components of the transcription machinery RNA PolII [70, 71], Mediator [71, 72], BRD4 [72], RBPs like FUS [73, 74], hnRNPA1, DDX4 [74, 75], TAF15 [76, 77], etc.

Due to recent developments of cellular tools to trigger condensate formation [74, 75], cellular phase behavior can be quantified by means of a phase diagram that maps condensate formation, C_sat, and dissolution based on temperature, interaction energy, and concentration [75, 78]. Nucleation marks the start of the condensate formation, which occurs either spontaneously through random diffusion by homogeneous nucleation, or at specific sites by heterogeneous nucleation [66]. In cells, it is unlikely that MLOs assemble through homogeneous nucleation, as they reliably form at specific cellular locations throughout cycles of assembly and disassembly during cell division. For example, the nucleoli are seeded or nucleated by actively transcribed rDNA [79], and suppression of transcription leads to extranucleolar droplets that fail to coarsen into nucleoli [80]. Once a liquid droplet is formed, whether it grows or dissolves depends on the interaction strengths between its components [78].

Transcription condensates: formation and key players

Transcription has also been posited to form nuclear condensates. The hypothesis of a stochastic transcription, where proteins and DNA freely diffuse and collide in the nucleus to mediate transcription, is unable to explain the rapid localization of transcription factors at their genomic targets [81, 82]. Advancement in microscopy has then led to the visualization of transcription-specialized zones and the formation of dynamic clusters of RNA PolII [83, 84]. This stable, long-lived clustering tendency is then observed in transcription regulatory proteins like Mediator and BRD4 at super enhancers, which is crucial for super enhancer-driven gene expression and exhibits liquid-like tendency of condensates [71, 72]. Currently, there is still ongoing debate on whether LLPS fully describes the formation of transcription condensates, so transcription condensates referred to here take on a broader definition of distinct, membraneless compartments consisting of high concentrations of transcription-related biomolecules held together by nonstoichiometric, multivalent interactions [85, 86].

Studies have shown that condensate formation is indispensable for transcription activity, as the liquid-like behavior allows for dynamic exchange of biomolecules yet sequestering critical factors within the vicinity of the condensate, hence enabling rapid transcriptional responses to a changing environment (reviewed extensively elsewhere [85–87]). Several observations in transcription can only be explained by transcription condensate formation, i.e. transcription bursts, which are the non-constant, episodic gene expression in eukaryotic cells. Upon serum activation, a “burst” of β-actin mRNA is synthesized for a period of time before returning to basal levels of expression, which is correlated with the observation of PolII clustering in situ of the β-actin gene [83]. At highly expressed, super enhancer-driven genes, transcription initiation forms condensates containing Mediator and hypophosphorylated RNA PolII [71, 72, 88]. As these condensates do not always overlap but are always within proximity of active transcription, Cho et al. propose a dynamic kissing model, in which the initiation condensate formed at the regulatory elements transiently contacts the gene to produce bursts of nascent transcription [71]. The same group recently updated this hypothesis into a three-way kissing model, where the transcription condensate (marked by PolII/Mediator), DNA regulatory element (super enhancer), and the gene locus transiently interact to result in gene bursting [89]. The authors found that the condensate is not necessarily formed at the super enhancer sequence, but the super enhancer is critical to maintain condensate size and gene burst frequency and size. Importantly, basal gene bursting occurs when condensates are distal, but when the condensate is proximal (< 1 μm), gene bursting increases in size and frequency, making condensate proximity a better indicator of gene bursting than enhancer proximity [89].

Aside from the Mediator complex as a coactivator, a number of transcription factors (TFs) have also been shown to undergo phase separation. While the DNA-binding domain of TFs allows targeting to specific genomic loci, the activation domain that mediates transcription activation contains phase separation-prone IDRs [90]. At super enhancer-driven genes, the depletion of OCT4 is indispensable for MED1 enrichment and condensate formation at the enhancer, despite OCT4 being unable to phase separate independently of MED1 in vitro. Other TFs such as NANOG, SOX2, GATA2, MYC, GCN4, and ligand-dependent receptors like ER and RARα also form droplets alone and are incorporated into Mediator droplets [90]. Unfortunately, the phase separation capacity lies within the activation domain, so the role of condensate formation cannot be discerned from the transactivation function.

The process of condensate assembly and chromatin organization are also inextricable, with various protein factors involved in both processes [91, 92]. Chromatin looping, particularly enhancer-promoter looping, is determined by factors such as cohesin, CTCF, Mediator, BRD4, PolII, YY1, and other transcription factors, and non-coding RNAs, all of which have been shown to undergo phase separation [71, 72, 90, 93–98]. Furthermore, cohesin has been shown to bring transcription condensates and gene loci into proximity to promote bursts of transcription [89]. Altogether, distal gene loci and regulatory elements can be brought into proximity not only through chromatin looping, but also phase separation. This drives active transcription hubs to form distinct domains enriched with transcription activating factors while excluding transcription inhibitory factors, enhancing transcription efficiency [92].

Regulation of transcription through condensate assembly

Due to the development of the optoDroplet system that enables light-induced phase separation in cells [74], fusion of a phase separation-prone FUS_IDR to a transcription factor indeed enables condensate-driven reporter gene expression [99]. The same has also been shown in endogenous gene expression facilitated by light-induced condensation of transcription activators [100]. Overall, this shows that IDR-containing factors cooperate at transcription initiation sites to coalesce RNA PolII, concentrating more transcription factors and coactivators into a large condensate through weak, multivalent interactions to drive transcription initiation. However, it is important to note that while the ability to form condensates through weak homotypic (self) interactions bestowed by the IDR is frequently highlighted in many studies, the IDR has also been shown to impart very valuable functions specific to that protein either through heterotypic (partner) interactions or sequence-specific targeting to the DNA [101]. Hence, the IDR of a specific transcription regulator may not be completely replaceable by another factor’s IDR fragment at specific gene loci it regulates.

Subsequently, phosphorylation of PolII by CDK7 and CDK9 transitions PolII into elongation. Concurrent with transcription, the synthesized mRNA is then bound and processed by RBPs and splicing factors, while hyperphosphorylated PolII is integrated into the splicing condensate, which is commonly observed as splicing speckles [88]. The C-terminal domain phosphorylation of RNA PolII is proposed to act as a switch that dictates the initiation/elongation condensate preference [88]. The phosphorylation state of PolII also determines interaction with other condensate-forming factors. For instance, the RBP TAF15, which is part of the FET family (FUS, EWS, and TAF15), has a unique tendency to more robustly condense in the nucleoplasm compared to the cytoplasm unlike other FET family proteins. The authors show that this is due to the enhanced interaction with unphosphorylated PolII CTD, which implicates TAF15 condensates as initiation condensates. Importantly, optogenetically induced TAF15 condensation further partitions PolII into the condensate, driving transcription [76]. This shows that apart from simply concentrating transcription regulators at the gene, some factors also influence RNA PolII partitioning at the initiation or the elongation condensate through distinct phase separating properties [76, 88, 102].

Later on, Henninger and colleagues proposed another model of transcription regulation mediated by condensate assembly and disassembly through changes in local RNA concentration. Electrostatic interactions between a low concentration of negatively charged RNA and positively charged IDRs initially favors the formation and growth of the initiation condensate, but high RNA:protein ratios arising from transcription bursts tip the charge balance, leading to condensate dissolution [103]. Furthermore, the authors simulated transcription by assembling an in vitro transcription mix consisting of RNA PolII, Mediator, general transcription factors, and a DNA template. They were able to show droplet formation that incorporated DNA and synthesized RNA, which potentially mirrors cellular conditions [103]. In the section below, we further discuss the role of RNA and R loops in transcription condensates.

R loops as seeds for transcription condensates

Biomolecular condensates consist of tens to hundreds of interacting molecules, but the structural integrity of the condensate hinges upon a few multivalent “scaffold” molecules, which act as seeds for a multitude of “client” molecules for assembly into the condensate [104]. Taking the nucleolus as an example, the nucleolar organizer region that is made up of rDNA, allows the targeting of scaffolding proteins like nucleolin and fibrillarin which facilitate multivalent interactions with rRNA and ribonucleoproteins [105]. In the case of transcription condensates, scaffold proteins directly bind to the chromatin through a DNA- or RNA-binding domain and nucleate client proteins like general transcription coactivators, PolII, and RNA processing factors.

RNA as a powerful nucleator in condensate assembly

In a review by Bevilacqua et al., the authors highlight the strength and frequency in which RNA binds in an intra- and interstrand manner, giving rise to multimerization (i.e. duplex, triplex, and quadruplexes), which acts as a robust, multivalent driver of condensate formation [106]. RNA has been shown to be a powerful “seed” that nucleates IDR-containing RBPs to promote condensation and hence MLO assembly [106, 107]. For example, rRNA is essential for nucleoli assembly by nucleating nucleolar proteins [108]. Non-coding RNA has also been shown to be a powerful nucleator of phase-separated compartments. For example, lncRNA NEAT1 interacts with a number of IDR-containing RBPs like NONO, SFPQ, FUS, RBM14, PSPC1, etc. to assemble paraspeckles, and also promote pri-miRNA processing through interacting with Microprocessor [95, 96]. Additionally, complementary ncRNA sequences of enhancer RNA and promoter upstream transcripts (PROMPTs) have recently been shown to drive enhancer-promoter looping and subsequent gene activation [97], which could promote RNA-enhanced condensate formation at loop-forming super enhancers. Damage-induced lncRNA has also been shown to stimulate phase separation of 53BP1 DNA damage foci formation [109]. The prion-like RBP FUS independently forms liquid-like droplets at high concentrations, but addition of RNA not only enhances FUS phase separation but further transitions FUS into fibrous assemblies akin to pathological aggregates in neurodegenerative diseases [110].

While RNA are powerful molecular seeds that nucleate phase separated condensates, RNA also limits condensate growth at high concentrations due to its reentrant behavior [103, 111, 112]. In a tunable cellular reporter system that tracks the effects of RNA on biophysical properties of a condensate, RNA is able to harden the condensate from liquid-like to a gel-like phase, but its enrichment at the condensate surface limits growth and coalescence [113].

R loops as an emerging seed for transcription condensate assembly

At transcription initiation condensates, nascently transcribed RNA is chromatin-bound. Approximately 40% of expressed genes harbor paused RNA PolII at the promoter proximal region [24], heavily favoring the formation of R loops. Owing to R loop association with many chromatin-bound RNA-binding proteins (chrRBPs), we ask if R loops can also seed transcription condensates, hence regulating transcription not only through maintaining R loop homeostasis but also through condensate assembly. Shao et al. recently reported that the chromatin-bound proteome of embryonic stem cells, excluding histones, is dominated by RBPs mainly mediating processes involving transcription and RNA processing [114]. Importantly, these chrRBPs tend to be more positively charged and contain more low-complexity sequences and IDR contents, implying a tendency for phase separation [114]. Of note, the affinity of chrRBPs for chromatin is comparable to epigenetic factors, which are sensitive to transcription inhibition and RNase treatment. Furthermore, the depletion of a panel of chrRBPs affects nascent transcription, suggesting that chrRBPs are not only dependent on transcription for chromatin targeting, but in turn also affects transcriptional programs [114]. In a study focused on R loop-associated interactome, Dettori et al. found that a majority of R loop-binding RBPs (~87% of readers and ~66% of enzymes) contain at least one long stretch of IDR, with a significant biased overrepresentation of low complexity and positively charged residues which support weak multivalent interactions and electrostatic interactions with negatively charged RNA [115], paralleling features of chrRBPs. As R loop interacting proteins typically consist of a folded domain and a large IDR, they further suggest that this modular nature imparts nucleic acid targeting specificity and condensation propensity respectively, to assemble an R loop-nucleated membraneless foci [115].

Research on R loop as a seed for condensate formation is only in its infancy, with only a handful of studies published in recent years suggesting this [65, 102, 115–117], but without visual proof of an R loop foci. In this section, we discuss the current evidence implicating R loops in condensate assembly and how this affects transcription.

R loop multivalency

The R loop itself is modular (Fig. 3), and its interacting proteins can bind to its substructures like the ssDNA, ssRNA, DNA: RNA hybrid, or the triplex structure in varying affinities (see Table 1 below for a few examples). Despite a paused PolII only forming one molecule of paused R loop at a time, this allows a number of R loop readers and enzymes to bind to a preferred substructure and potentially nucleating other binding partners. Evidence for tandem formation of a G4 structure on the displaced ssDNA of an R loop, also called G loops, has also been shown to regulate transcription [5, 18, 118, 119]. As both G4 DNA and RNA have been implicated in the phase separation properties of a number of proteins (summarized in [120]), the simultaneous G4 formation in an R loop can provide yet another R loop substructure as a condensate nucleator. Indeed, a number of RBPs and splicing factors have been shown to bind both R loops and G4 DNA, with some factors like U2AF1 displaying enhanced phase separation upon addition of G loops when compared to R loops [102].

Fig. 3.

R loops as potential multivalent nucleators for transcription condensates. The modular nature of R loops, containing many substructures such as ssDNA, ssRNA, DNA: RNA hybrids, the bubble structure, and G4 structures the ssDNA may take, allows binding of many factors which targets a preferred substructure (i.e. blue protein binding ssRNA, pink binding DNA: RNA hybrid, green binding the bubble structure, and yellow binding G4 DNA). Many R loop-binding factors are transcription-associated and contain condensate formation-prone intrinsically disordered regions (i.e. transcription factors (TFs), coactivators, and RNA-binding proteins (RBPs) denoted by differently colored protein), which also recruit components of the transcription machinery. By interacting with one another weakly and multivalently, they form a dynamic transcription condensate (yellow bubble). R loops form in clusters (Cluster 1 denoted by blue PolII, and Cluster 2 denoted by green PolII), which can multiply the number of available binding platforms for R loop-binding proteins to grow the condensate. Furthermore, activated PARP1 (purple protein with yellow branches indicating PAR chains) directly binds R loops and bridges multiple R loops together. This PARP1-mediated R loop bridging further enlarges the transcription condensate.

Table 1.

Summary of protein binding to R loops and R loop substructures

	R loop substructure binding affinity
Protein	ssDNA	ssRNA	dsDNA	DNA: RNA hybrid	R loop	Ref.
RNaseH1	−	−	−	++	+++	[21]
ARIP4	+	+	+	++	+++	[15]
SSB1	+++	+	+	N/A	+++	[65, 117]
PARP1	+++	N/A	++	+	++	[121]
FANCI/D2	+	+++	+	−	+++	[59]
FMRP	++	+++	−	−	++	[126]

A list of a few examples of R loop-binding proteins validated by EMSA, indicating their binding affinity (ranging from lowest to highest − < + < ++ < +++ ) to the indicated R loop substructures

N/A = not applicable, indicating that the cited reference did not test for that nucleic acid substructure

The robustness and high resolution of R loop detection by RIAN-seq also elucidates that R loops do not form broad peaks (as detected by DRIP-seq) or singular peaks per TSS (as frequently seen in R-ChIP and R loop CUT&Tag), but form in short (60–130 bp) clusters of multiple R loops [22]. Importantly, these R loop clusters are found only at promoter proximal regions, which suggests that R loop clustering plays a role in transcription. Indeed, the number of R loops in a cluster determines transcription activation and correlates with resistance towards transcription inhibition [22]. It would be interesting to investigate if the density of R loops clusters promotes transcription due to a propensity to nucleate transcription condensates.

Intriguingly, Laspata et al. recently showed direct binding of R loops by PARP1 not only through in vitro binding, but also visualized it through atomic force microscopy (AFM). AFM imaging reveals activated, ADP-ribosylated PARP1 bound to R loops that enables bridging to other R loop molecules [121]. If this can be proven physiologically, this suggests that PARylated PARP1 can bridge multiple R loop molecules to further nucleate RBPs to make up an R loop condensate. Additionally, PARP1 has been implicated as a transcription coactivator through PARylating histones and TFs and acting as a scaffold for TF binding (reviewed in [122]). In parallel, PARylation of RBPs like FUS, EWS, TAF15 [123, 124], hnRNAP1, and TDP43 [125] enhances phase separation properties. PARylation of a prion-like protein PARIS not only promotes phase separation into a liquid, but also into solid aggregates attributed to Parkinson’s Disease pathogenesis. Similarly, PARP1-dependent solidification has also been implicated in pathological aggregation of hnRNPs in neurodegenerative Amyotrophic Lateral Sclerosis (ALS) models [125]. Altogether, R loop substructures, R loop clustering at the promoter regions, and R loop bridges might contribute to the multivalency of R loops.

R loop-binding proteins form condensates and regulate transcription

Transcription factors, which promote transcription of specific genes through R loop detection [10, 22], are prone to phase separation and can nucleate transcription condensates [90]. Additionally, R loop-binding RBPs share features such as phase separation capacity and RNA binding, which are critical to promote transcription. Shao et al. suggest that the chromatin-bound RBP PSPC1 requires both these features to associate with the chromatin, particularly enriching together with both promoter proximal paused RNA and hypophosphorylated RNA PolII [114], which likely form Class I R loops. The addition of RNA severely increases the phase separation capacity and PolII CTD coalescence of PSPC1 in vitro, and the synergy between PSPC1 and RNA dramatically enhances CKD9-mediated PolII phosphorylation to promote elongation. Furthermore, the authors suggest that other chrRBPs cooperate at the PolII pause site to drive transcription. Acute degradation of PSPC1 by an auxin-inducible degron shows a notable decrease in nascent transcript synthesis within 6 hours, but drop in transcription output is restored after 24 hours of PSPC1 degradation, indicating that other factors act redundantly as a fail-safe. Indeed, analysis of a set of 14 chrRBPs in ESCs revealed a preference for regulatory regions such as promoter paused sites and super enhancers, majority of which are co-occupied by at least 4 out of the 14 chrRBPs analyzed. Overall, this suggests that a number of chrRBPs bind the Class I R loop-prone hotspots cooperatively through their affinity for RNA, and through weak interactions with one another form a large transcription initiation condensate that concentrates components of the transcription machinery (RNA PolII, CDK9, etc.) to promote elongation [114]. Another mode of transcription regulation is through the sequestration of splicing factors to the promoter proximal pause sites via phase separation with the transcription condensate. He et al. exemplify this with the splicing factor U2AF1, which phase separates only in the presence of R loops. Notably, the interaction between U2AF1 and Class I R loops coalesces U2AF1 into the initiation condensate and drives the redistribution of PolII away from the elongation/splicing condensate and into the initiation condensate [102]. A third manner in which RBPs cooperate with R loops to regulate transcription is through terminating transcription at the promoter proximal region to safeguard the genome from Class I R loop-induced genomic instability. The SOSS-INTAC complex can form condensates and accumulates at the promoter proximal pause sites by binding to Class I R loops through the SSB1 subunit. This complex is a large, multi-subunit complex that combines a DNA damage factor SOSS and a transcriptional regulator Integrator (RNA endonuclease)-PP2A (protein phosphatase) (INTAC). Deletion and degradation of SSB1 leads to the accumulation of R loops, but this is catalyzed by the endonuclease subunit of INTAC INTS11, which depends on its association to SSB1 to the ssDNA component of R loops for R loop resolution. Aside from imparting promoter targeting to the SOSS-INTAC complex, SSB1 also contains an IDR that allows SOSS-INTAC droplet formation, which permits SOSS-INTAC partitioning into the transcription condensate, hence resolving Class I R loops. INTAC terminates transcription prematurely at the promoter proximal sites. In line with this, deletion of SSB1 and an INTAC backbone component INTS2, leads to paused PolII accumulation at the TSS, indicating impairment in PolII eviction [65]. SOSS has also been shown to accumulate at DNA damage sites in an R loop-dependent manner, forming droplets together with PolII in a DNA repair compartment [117]. However, it is unclear whether SOSS also interacts with INTAC, or if it mediates repair through an R loop resolving mechanism, at this liquid-like DNA repair compartment. FUS has also been shown to depend on its IDR domain to promote transcription and suppress R loop formation, likely through sequestering RNA from the DNA template [127]. However, whether formation of FUS condensates is facilitated by, or required for R loop suppression and subsequent transcription activation remains to be determined.

Membraneless R loop foci: fact or fiction?

In the previous sections, we discussed the features of R loops that may nucleate the transcription condensate. The initiation condensate, which refers to the transcription machinery at both the initiation and paused states consisting of hypophosphorylated PolII, transcription factors and coactivators, and a low concentration of short nascently transcribed RNA, overlaps with Class I R loops which represent the majority of genomic R loops. With its many substructures, such as ssDNA, ssRNA, DNA: RNA hybrid, G quadruplex, or the branched bubble structure, R loops can be bound by numerous DNA/RNA-binding proteins. Importantly, these proteins mostly contain IDRs suited for condensate formation and coalescence with RNA PolII to modulate transcription through different methods (see Fig. 4). Clustering of R loops at the promoter proximal region suggests that a single gene may have multiple R loop “seeds” as a binding platform. R loop binding by proteins like PARP1, which bridges multiple R loop structures in a vicinity when activated, may further bring together several R loop-nucleated condensates to assemble an even larger condensate.

Fig. 4.

R loop-dependent transcription regulation through condensate formation. Recent evidence showing different modes of transcription regulation through an R loop-dependent transcription condensate formation mechanism. ① As discussed in Fig. 3, R loops concentrate numerous IDR-containing transcription-associated RNA-binding proteins (RBPs) that form the initiation condensate that coalesces into a larger condensate with more RBP partitioning. Short nascent transcript synthesis and aborted RNA products lead to the burst of the initiation condensate, and the elongating transcription machinery coalesces into splicing condensates, where stable, long nascent transcripts are processed. The next round of transcription begins, and the initiation condensate is formed again, and the cycle of transcription is repeated to suit cellular demands. ② Some factors that associate with the transcription machinery can tune its preference for the initiation or splicing condensates. Productive elongation may be hampered by factors that sequester the transcription machinery through an R loop- binding mechanism. ③ Other R loop targeting factors may also terminate transcription altogether in favor of preserving genomic integrity from R loop-mediated DNA damage. This R loop suppression and premature transcription termination areis dependent on partitioning into a condensate.

If an R loop-nucleated condensate assembly is possible in cells, detection methods using the S9.6 antibody [57, 128], overexpression of catalytically-dead RNaseH1 [20], overexpression of the RNaseH1 hybrid binding domain (HBD) [53], and staining with purified catalytically-dead RNaseH1 [129] are incapable of detecting distinct nuclear foci. Surprisingly, the most updated R loop detection method RHINO, consisting of a tandem array of three HBDs tagged with a fluorescent protein, exhibits distinct nuclear foci by confocal microscopy [130]. S9.6, full length RNaseH1, and a single copy of the HBD show a pan-nuclear staining with strong nucleolar staining, while the RHINO (3xHBD) is sensitive enough to detect R loop-enriched foci that are both susceptible to RNaseH1-digestion and transcription inhibition, with some foci found at the telomeres and within the nucleoli [130]. While it is well established that both the nucleoli and telomere (ALT-associated PML bodies/APBs) are assembled as membraneless nuclear bodies, it is unknown if nucleolar and telomeric R loops participate in condensate assembly. One molecule of RNaseH1 is capable of binding ~6 bp of DNA:RNA hybrid through one HBD [131], whereas the nascent RNA transcribed from a paused PolII is estimated to be ~20–60 bp [24], indicating that only one to three molecules of the 3xHBD RHINO plasmid is capable of binding at paused R loops, yet they appear as discrete foci. The idea that R loops commonly form as clusters consisting of multiple short R loop-forming sequences (with clusters consisting of 3–22 individual R loops), and that the density of R loops per cluster is associated with gene expression [22], can potentially explain how highly expressed, paused genes form visible R loop foci. Physiologically, long and extensive elongation-associated Class II R loops would be suppressed by RNA processing factors due to its threat to the genome, but the suppression of RHINO foci by the CDK9 inhibitor DRB implies that the RHINO foci also detects elongation-associated R loops. Altogether, whether or not paused R loops enable visual detection due to its potential to form larger condensates remain to be elucidated.

Overall, the combination of R loop features, such as the modular substructures, clustering tendency, binding to IDR-prone RBPs, and bridging by PARP1, can potentially provide the multivalency necessary to overcome C_sat and nucleate a distinct condensate, especially at actively transcribed regions.

Disease implications and future perspectives

Recently, Banani et al. mapped ~36,000 pathogenic mutations linked to dysregulation of condensate formation in Mendelian diseases and cancers. In both Mendelian diseases and cancers, these condensate-dysregulating pathogenic mutations are overrepresented by the transcription-related proteome, thus highlighting the importance of maintaining the integrity of transcription condensates [132]. Additionally, dysregulation in R loop regulators and IDR-containing proteins have intersecting disease manifestations in neurodegenerative disorders.

Neurodegenerative disorders are characterized by the progressive loss of motor and cognitive function due to degeneration in structure and function of the nervous system. Mutations in the R loop helicase SETX have been attributed to two neurodegenerative disorders, Ataxia with Oculomotor Apraxia type 2 (AOA2) and Amyotrophic Lateral Sclerosis type 4 (ALS4). AOA2 is linked to loss of SETX function while ALS4 is linked to gain of SETX function, thus highlighting the importance of R loop homeostasis. Loss of SETX function in AOA2 leads to DNA damage and genomic instability from R loop accumulation [133], while gain of SETX function disrupts R loop-dependent silencing of genes frequently altered in motor neuron diseases [13]. Aicardi Goutières syndrome (AGS) is a neuroinflammatory disease arising from mutations in R loop enzymes RNaseH2A/B/C and ADAR. Accumulation of RNaseH2-dependent cotranscriptional R loops is processed by endonucleases that elicit DNA damage and downstream immune responses associated with AGS [44]. Cockayne Syndrome (CS) is a genetic disorder due to mutations in DNA repair proteins CSA and CSB, characterized by growth retardation and severe neurological degeneration. Recently, CSB deficiency has been shown to elevate R loop levels and obstruct RNA PolII elongation at specific loci (intronic regions containing a G-rich sequence followed by T runs) that pauses and stalls PolII. At these transcription barriers, CSB is essential for propelling PolII forward to prevent the stabilization of R loops, which presents a threat to genomic stability. Notably, long, neuronal genes are most prone to this CSB-dependent pathological R loop accumulation due to the abundance of transcription barriers, which the authors suggest is the main culprit of genomic instability in neurodegenerative diseases [134].

Meanwhile, disease phenotypes arising from phase separation or condensation defects may result from alterations in environment (physicochemical changes), dysregulation of master regulators (epigenetic landscape, protein kinases, nucleator availability), or altered condensation mechanism (protein solubility, IDR mutations) [92]. These factors subsequently lead to alterations of C_sat, random or untargeted ectopic condensate formation, change in condensate content, or pathological phase transitions like aggregation or solidification, resulting in aberrant downstream responses mediated by condensation [92]. A central characteristic of all neurodegenerative disorders is the localized accumulation of cytosolic or nuclear proteins, which are thought to contribute to neurodegeneration. For instance, aggregation of the microtubule-associated protein Tau in Alzheimer’s Disease [135], ectopically cytoplasmic aggregation of the nuclear RBP TDP-43 [136], aberrant solidification of FUS in ALS [137], and poly-glutamine aggregation of Huntingtin in Huntington’s Disease [138].

Comprehensive summaries of disease-linked alterations and mutations of both R loop regulators [139, 140] and condensate formation [92, 141] have been reviewed elsewhere. Here, we focus on the few overlapping factors involved in both R loop homeostasis and condensate formation. The expansion of a hexanucleotide repeat (GGGGCC) in the C9ORF72 gene is the most common genetic cause of ALS and frontotemporal dementia. The expansion of these G and C-rich sequences has been implicated in stabilizing the formation of G4 DNA and R loops, which accumulate persistently in ALS patients and trigger genomic instability [142]. The precise mechanism behind how these R loops contribute to the progression of the disease and whether these R loops trigger changes in transcriptional programs are currently unknown [143]. How this repeat expansion leads to aberrant condensation is twofold: synthesis of repeat-containing RNA and subsequent synthesis of repeat-containing proteins, bolstering multivalent intermolecular interactions [92, 144]. FUS and TDP-43 are critical players in the pathology of familial and sporadic forms of ALS. ALS patient-derived mutations on the prion-like domain of FUS convert it from a liquid-like state into fibrous aggregates, implicating aberrant phase transitions as the crux of the pathology of ALS and other related diseases [137]. Insoluble TDP-43 from diseased brains seeds pathological aggregation in cultured cells, transmitting its pathogenic prion-like properties, which further propagates intracellularly when phosphorylated [136]. In ALS, both FUS and TDP-43 undergo mislocalization from the nucleus to the cytoplasm, where it forms toxic aggregates, a hallmark of most ALS cases. FUS and TDP-43 are RBPs involved mainly in RNA metabolism, such as transcription, RNA splicing and transport, hence FUS and TDP-43 suppress DNA damage and excessive PolII stalling by preventing R loop accumulation [145]. However, it remains to be investigated if R loops can tune FUS and TDP-43 condensation under normal conditions, and if pathological R loops are subsequently capable of influencing FUS and TDP-43 phase transitions to explain the pathogenesis of ALS.

Phase separation has become a new aspect in the understanding of cancer phenotypes and treatment development. One key feature of cancers is its alteration into a hyperproliferative state, which phase separation can modulate through different ways: (1) dysregulation of transcriptional programs to favor proliferation, advancement, and survival of cancers, (2) tumor-associated mutations may also alter cell signaling by constitutive condensate formation or including favorable factors and excluding inhibitory ones from a condensate, or (3) formation of oncogenic fusions between transcription factors or chromatin regulators and IDRs of prion-like proteins [92]. R loops are sources, products, and regulators of DNA damage, making them a prevailing source of genomic instability, a hallmark of cancer [4, 62, 146]. For example, in breast or ovarian cancers, estrogen treatment and mutation of BRCA1/BRCA2 lead to DNA damage from elevated Class I R loops [8, 16, 30]. Highly expressed androgen target genes implicated in prostate cancer proliferation and advancement accumulate Class I R loops resolved by the helicase ARIP4 [15]. In myelodysplastic syndromes, splicing factor mutations lead to replication stress induced by accumulation of Class I R loops [26].

As discussed in previous sections, R loop regulators have been shown to maintain normal transcription processes and genomic integrity through a phase separation mechanism. While this interplay is linked only recently, it would be an interesting relationship to pursue, as many therapies have been developed to target either pathological R loops or aberrant condensate formation without considering that one might affect the other in terms of mechanism of action. As R loops arise not only from routine cellular programs like transcription, but also from pathological conditions such as damaged DNA, both of which are processes facilitated by condensation, we ask if the condensation event can be triggered in a similar manner through R loops and PARylation. Conversely, in the event of transcription recovery after repair, whether R loops and PARylation can be a unifying mechanism that allows the exchange of components from a transcription condensate to DNA damage/repair condensate, and back into a transcription condensate. Furthermore, as R loop homeostasis hangs on a very fine balance, it would be exciting to see further investigations of how cells harmonize transcription schedules and maintain genomic stability through phase separation-mediated partitioning.

Finally, due to the value of targeting phase separation in pathological conditions such as neurodegenerative diseases and cancer, pharmacological interventions for augmenting condensation have been promptly developed. However, these drugs may have many unwanted non-specific targets and impact global phase separation events that assemble different MLOs (i.e. the use of 1,6-hexanediole, ATP depletion, or transcription inhibitors). For instance, as LLPS is an ATP-driven process and promotes protein solubility [147], depletion of ATP may influence the viscosity or aggregation properties of MLOs but may affect other ATP-dependent cellular processes [148, 149]. rRNA synthesis is the key driving process of nucleolar assembly [108], and inhibition of RNA PolI transcription by Actinomycin D indeed affects the nucleolar rigidity [150], but this predictably might affect nucleolar proteome and transcription by RNA PolII. PARP inhibitors have been used to treat certain cancers, which suppress the PARylation-dependent phase separation behaviors of prion-like RBPs [124, 125, 151, 152]. In a PARP1-bridged R loop condensate model, we wonder if PARP inhibitors could also ameliorate pathological conditions through disruption of R loop nucleated condensates. In a review of pharmacological interventions targeting phase separation, Wang and Zhang summarized that drugs like mitoxantrone can alter condensation properties of multiple MLOs such as stress granules, Cajal body, nucleoli and DNA damage foci, while DNA damaging chemotherapeutic drugs like daunorubicin and doxorubicin suppresses stress granule condensation, RNA and prion-like protein containing condensates, and biphasic drugs like Congo red or bis-ANS affects condensate formation (enhance or suppress) oppositely when applied at different concentrations (low or high) [149]. Overall, strategies targeting LLPS mainly involve chemicals with non-specific effects and are focused mainly on condensate assembly and disassembly, which may not address ectopic condensate formation (i.e. mislocalized FUS and TDP-43) or changes in condensate content [92]. The recent study by Patil and colleagues highlights that the IDRs of proteins impart not only phase separation capacity, but also mediate heterotypic interactions with partner proteins [101]. Hence, pathological conditions derived from mutated IDRs may not sufficiently be rescued with a universal cure that only ameliorates condensation but neglects the reconstruction of the condensate contents through heterotypic interactions, which may affect downstream processes. This implies that treating altered condensates might require more specific strategies.

Conclusions

Cotranscriptional R loops pose not only a threat to the genome, but also to normal transcription programs. A combination of its unique yet vulnerable structure, formation at negatively supercoiled and nucleosome-depleted regions, and association with paused or stalled RNA PolII, primes R loop-forming regions to DNA damage and genomic instability (Fig. 1). R loop levels are rigorously maintained by numerousregulatory proteins, which either sustain normal transcription levels by suppression of pervasive R loops as roadblocks to successive rounds of transcription, or forgo transcription in favor of maintaining genomic integrity (Fig. 2). Other R loop-binding factors such as DNA/RNA modifiers, epigenetic regulators and chromatin remodelers may also indirectly influence transcription outside of directly impacting PolII progression. Recent studies imply that R loops may also regulate transcription through a new mechanism: the formation of transcription condensates. Transcription condensate assembly enriches transcription-associated factors within a sequestered region in order to rapidly and efficiently regulate transcription in response to changes in cellular cues and environment. These condensate-forming transcription-associated factors, many of which are RBPs, have affinity for R loops and its multiple substructures, combined with R loop clustering at promoters and potential bridging by PARP1, make R loops a potent, multivalent nucleator for initiation condensates (Fig. 3). Aside from maintaining favorable conditions for active transcription through assembling transcription-associated factors, R loop-triggered partitioning of other factors may also negatively regulate transcription (Fig. 4). While further rigorous studies will be necessary to solidify R loops as a seed for transcription condensate formation, it is tempting to draw this conclusion based on the current evidence. Assembly of large, multivalent condensates may also explain the extensive catalogue of factors involved in R loop regulation. Partitioning of certain factors into specific transcription condensates not only through R loop detection but also other co-condensing proteins may confer a specificity towards particular genes, regulatory regions, or chromatin features, making individual R loop regulators functionally non-redundant. Furthermore, an interesting observation to dissect would be the conditions in which R loops are appropriately resolved as a prerequisite for transcription, and when transcription must be shut down in favor of catastrophic R loop accumulation (Fig. 4). It is also unknown if these two scenarios may occur in the same condensate or are mutually exclusive events. Overall, much remains to be understood in the emerging interplay between R loops and condensate formation, but advancements in the field may be beneficial not only to further the understanding of the complexities of transcription regulation, but also in the development of therapeutics.

Author contributions

RRN wrote the manuscript. ZL, HWT, Y-MX, and ATYL reviewed and edited the manuscript.

Funding

This work was supported by the grants from the National Natural Science Foundation of China (W2533080, 31771582 and 31271445), SUMC Research Initiation Grant awarded to Raissa Regina Ng, and the Natural Science Foundation of Guangdong Province of China (2025A1515012547).

Data availability

Not applicable.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing Interests

The authors declare no competing interests.