Poly(ADP-ribose): A Dynamic Trigger for Biomolecular Condensate Formation

Abstract

Poly(ADP-ribose) (PAR) is a nucleic acid-like protein modification that can seed the formation of microscopically-visible cellular compartments that lack enveloping membranes, recently termed biomolecular condensates. These PAR-mediated condensates are linked to cancer, viral infection and neurodegeneration. Recent data have shown therapeutic possibilities based on PAR modulation, as PAR polymerase (PARP) inhibitor can modulate the formation and dynamics of these condensates as well as the trafficking of their components—many of which are key disease factors. However, the way in which PAR facilitates these functions remains unclear, partly due to our lack of understanding of the fundamental parameters of PAR in cells, such as conjugated sites, length, structure and physicochemical properties. This review first introduces the role of PAR in regulating biomolecular condensates, followed by discussions on the current knowledge gaps, potential solutions and therapeutic applications.

PAR is enriched in specific biomolecular condensates

Cells are organized into compartments stably concentrating macromolecules for biochemical processes while segregating other cellular components from unwanted reactions. Although classic compartments are enveloped with membranes such as nucleus and mitochondria, some cellular compartments are not bound by any physical barriers; yet these so-called biomolecular condensates are enriched with specific proteins, nucleic acids and other macromolecules [1–5]. Many of these condensates, such as stress granules, behave like liquid where they fuse and exchange components rapidly with the surrounding cellular fluid. Because of the liquid-like behavior observed in these condensates and their formation and size is dependent on the concentration of macromolecules, these cellular compartments are currently regarded as being formed through a physical process known as liquid-liquid phase separation.

Macromolecules are miscible with cellular fluid until their concentration exceeds the solubility limit. At that threshold, macromolecules phase separate from solution, resulting in the stable coexistence of one liquid phase highly concentrated with the macromolecules (condensed phase) and the other relatively more dilute (bulk phase). The solubility limit is determined by the balance between the entropy for macromolecules to be distributed homogeneously within the cellular fluid and the affinity between macromolecules with the cellular fluid versus those amongst the macromolecules. The stronger the affinity amongst the macromolecules over those with the fluid, the lower the concentration of the macromolecules is required to trigger phase separation. In cells, phase separation is often facilitated by multivalent interactions, which increase affinity, amongst proteins and nucleic acids. Excellent reviews have extensively covered the physical process of liquid-liquid phase separation in biology [1–5]. This review focuses on the phase separation mediated through the lesser-known nucleic acid PAR, which is enriched in time- and location-specific biomolecular condensates [6], such as DNA repair foci [7,8], stress granules [9–11], mitotic spindles [12] and nucleoli [13] (Table 1). Because these condensates are linked to various diseases from cancer to viral infection to neurodegeneration (such as amyotrophic lateral sclerosis, ALS), and the enzymes responsible for its synthesis PARPs are druggable [14], understanding how various properties of PAR influences phase separation could be translated into novel therapies.

Table 1 |. Examples of disease-linked biomolecular condensates enriched with PAR and PARPs.

PARPs targeted by specific inhibitors are highlighted in red.

Biomolecular Condensates	Diseases	PAR-adding PARPs	MAR-adding PARPs
DNA repair foci	Cancer	1,2	3
Stress granules	Cancer, viral infection and neurodegeneration	5a	12,14,15
Mitotic Spindle	Cancer	1,2,5a	3,4
Nucleolus	Cancer	1,2

Uniqueness of PAR as a trigger of phase separation

Multivalency of PAR promotes liquid-liquid phase separation

PAR is a polymer of ADP-ribose units, comprising the same constituents of the classic nucleic acids: nucleobases (adenine), ribose sugars and phosphates (Fig. 1, Key Figure). But unlike DNA and RNA, PAR is usually covalently conjugated to proteins as part of a broader class of post-translational modification called ADP-ribosylation, where single units (mono ADP-ribose, MAR) or polymeric chains of ADP-ribose, i.e. PAR, are conjugated to proteins, known as MARylation and PARylation, respectively (Box 1). The synthesis of these modifications is regulated by 17 ADP-ribosyltransferases, commonly known as PARPs, in humans (Box 1).

Fig. 1 |. Poly(ADP-ribose) code (varying sites, length and structure) directs protein interactions within biomolecular condensates.

(A) PAR can be conjugated to different amino acids with varying numbers of ADP-ribose (blue pentagon) and the ADP-ribose can be connected with two possible configurations, resulting in linear and branched chain formation. Analogous to the ubiquitin code, the varying site, length and structure may comprise a PAR code directing biological outcomes. (B) Multivalency can be achieved by multiple PARylated sites from a single protein or by a PAR chain comprising multiple ADP-ribose units for binding to proteins (magenta). Recent data also indicate that the length and structure of PAR are determinants of protein binding (e.g., orange proteins in the illustration require the binding of three ADP-ribose units whereas the red protein binds to the branchpoint of PAR. Therefore, the formation of different PAR may increase the number of multivalent interactions critical for phase separation. (C) PAR can serve a scaffold, where its length and structure specify which client proteins to recruit, resulting in compositional control of biomolecular condensates.

Box 1. Synthesis and Degradation of PAR.

PARylation is part of a family of post-translational modifications called ADP-ribosylation, which refers to the addition of one or more ADP-ribose units onto proteins [30–34]. The synthesis of ADP-ribosylation is regulated by ADP-ribosyltransferases, which transfer ADP-ribose units from NAD⁺ onto proteins or to another ADP-ribose. ADP-ribosylation in cells is primarily regulated by a subset of transferases commonly known as PARPs. In humans, there are 17 PARPs; four conjugate the initial and subsequent ADP-ribose units (PARylation), 11 conjugate only single ADP-ribose units (mono-ADP-ribosylation or MARylation) and two are catalytically inactive [94] (Figure I, ADP-ribosylation Dynamics). Systematic analyses revealed that PARPs localize to specific subcellular compartments [95]. All PAR-adding PARPs can shuttle between the nucleus and cytoplasm, with PARPs 1 and 2 primarily localized in the nucleus and PARPs 5a and 5b in the cytoplasm. Therefore, PAR can be theoretically synthesized anywhere in cells.

The complete removal of PAR from ADP-ribosylated proteins requires two steps (Figure I): the degradation of the polymeric chain down to single ADP-ribose units conjugated to proteins by PAR-specific degrader poly(ADP-ribose) glycohydrolase (PARG), followed by the hydrolysis of the final, proximal ADP-ribose groups from proteins [30–34]. Several degraders that remove the last ADP-ribose have known specificity: ARH1 for arginine, ARH3 for serine as well as MacroD1, MacroD2 and TARG1 for glutamate and aspartate [30–34]. This amino acid specificity indicates that the reversal of ADP-ribosylation is highly regulated. As in the case of PARPs, these degradative enzymes localize in specific compartments [30–34], allowing local control of the concentration and length of ADP-ribose units in cells.

Figure I |.

ADP-ribosylation Dynamics

PAR is made up of 2–200 ADP-ribose units in cells, and the repetitive surface of this nucleic acid may therefore act as a multivalent platform for non-covalent binding of proteins. As complexes form between PAR molecules and proteins, the entropic cost of confining the PAR-protein complexes into the condensed phase is lower than the cost of confining individual components. This difference in entropic cost due to multivalency partly drives phase separation. Phase separation occurs when the affinity between PAR and protein is higher than the affinity between cellular fluid with either macromolecules.

PAR binding to protein can be mediated through stereospecific interactions with structurally-defined modular PAR-binding domains [15,16]. Several modular domains recognize different moieties of PAR. Macrodomain recognizes the terminal ADP-ribose, the PAR-binding zinc finger (PBZ) domain recognizes adjacent ADP-ribose groups, and the WWE domain binds to the structural unit that bridges monomers [17]. Many DNA-binding domains, such as the Oligonucleotide/oligosaccharide-binding (OB) fold, or RNA-binding domains, such as the RNA recognition motif (RRM), can bind to PAR. Notably, several PAR-binding proteins have two or more binding domains in tandem. For example, the DNA repair factors APLF and BRCA2 have two PBZ domains and three OB-folds, respectively, whereas PARPs 12 and 13 have two WWE domains and PARPs 9, 14 and 15 have two or three macrodomains. In the case of APLF, although the individual PBZ domains are sufficient to bind one to two ADP-ribose groups of PAR, the tandem construct binds to PAR with at least 1,000-fold higher affinity than their counterparts [18]. These types of multivalent interactions further promote PAR-protein complex formation and increase the propensity of phase separation. Besides increasing affinity, tandem domains may also specify the types of interactions within the PAR-protein complex. For example, as the macrodomain interacts only with the terminal ADP-ribose [19], the tandem macrodomains of PARP9 may engage with distinct PAR molecules, thus potentially forming a complex with PAR to protein in a 2:1 stoichiometry.

Nucleic acid-binding proteins are also enriched with intrinsically disordered regions (IDRs) that lack defined 3D structures [20]. These regions are often of low sequence complexity (i.e., repeated amino acid) that mediate weak adhesive intermolecular interactions, which in turn allows for protein condensation [1–5]. Informatics analyses revealed that many biomolecular condensates accumulate PAR-conjugated proteins and PAR-binding proteins [21,22], and both classes of proteins are enriched with low-complexity regions [6,8,22,23]. Recently, several groups showed that purified PAR induces phase-separation of proteins with low-complexity regions in vitro [7,8,10,11,24,25]. Therefore, PAR may trigger phase separation by recruiting low-complexity region-containing proteins to form biomolecular condensates in cells [6,8]. For example, RG/RGG motif not only binds RNA but also PAR [8,21,25,26], where the arginine side chain may contribute to favorable π-stacking, cation–π and hydrogen bond interactions with the ribose sugar and bases of the nucleic acids [27]. Peptide screens revealed a linear PAR-binding motif (PBM) that is composed of a basic KR–KR doublet flanked by hydrophobic amino acids [28,29]. These positive charge residues of proteins may provide favorable electrostatic interactions with PAR. Complex formation of these highly but oppositely charged macromolecules results in a net lowering of the charge, driving a type of phase separation commonly known as complex coacervation [3,4].

Depending on how ADP-ribose units are conjugated to their substrates and each other, PAR polymers can be generated with diverse moieties. As a result, different modes of multivalent interactions can be generated to drive phase separation based on the differences in the site, length and structure of PAR (Fig. 1A–C).

Site

First, ADP-ribose can be conjugated onto amino acids with distinct chemistry (e.g., Glu, Asp, Arg, Ser) [30–35], and these polar and charged residues are commonly found in intrinsically disordered regions of proteins [1–5]. Such addition may positively or negatively affect the existing protein-protein and protein-nucleic acid interactions critical for the formation of biomolecular condensates. If a single protein is modified at multiple sites, it is also possible to create a high valency scaffold for PAR-binding proteins to increase the propensity of phase separation (Fig. 1B).

Length

Second, PAR can be elongated to different lengths in cells. ADP-ribose chain length is highly regulated by enzymes and cofactors [30–35]. The enzymatic control of PAR length may modulate the multivalency critical for phase separation. For example, a longer ADP-ribose chain length of PAR increases its valency for protein binding (Fig. 1B). As PAR length increases, more PAR-binding proteins are selectively enriched locally, resulting in the condensation of proteins and the formation of cellular compartments (Fig. 1C).

Structure

Third, PAR is formed by linking adjacent ADP-ribose units with ribose-ribose bonds. As there are two ribose groups in each ADP-ribose, two configurations for linking adjacent units are possible, resulting in linear and branched structures (Fig. 1A–B). Recent data indicate that the polymer length is a critical determinant for protein binding and the branched structure recruits specific proteins [36–40]. Therefore, by recruiting different proteins, the length and structure of PAR potentially direct the compositional control of biomolecular condensates (Fig. 1C). Taken together, these varying site, length and branching may comprise a PAR code, which may in turn be critical for directing biological outcomes, including the formation and composition of biomolecular condensates.

PAR as a scaffold

Remarkably, PAR can be synthesized and degraded within seconds [30–35] (Box 1). Upon DNA damage, PAR polymers containing on average 67 ADP-ribose units (~0.5 kDa each) per chain are rapidly synthesized [41]. The resulting PAR polymer could present itself as a sizeable, multivalent structure with a large surface area. This size and rapid formation kinetics enable PAR to serve as a temporary “scaffold” of biomolecular condensates to recruit proteins on demand [6]. Such behavior is analogous to the preferential condensation of RNA-binding proteins at nascent transcription sites upon the local production of RNA [42]. The difference is that PAR allows the formation of a multivalent scaffold for phase separation at conjugated proteins theoretically anywhere in cells, not restricted to specific genome foci. This rapid, local, enzyme-based addition or degradation of a scaffold thereby enables a sharp phase transition in response to environmental changes.

The localization of PARylated proteins can be determined by the protein-targeting signal and by the modification itself. For example, p53 is a nucleo-cytoplasmic shuttling protein, but once this transcription factor is PARylated, it can no longer export to the cytoplasm, resulting in nuclear accumulation that facilitates its roles in DNA damage responses [43]. The modification can also be regulated in biological contexts through the action of enzymes and cofactors in modifying and de-modifying at specific classes of amino acids. For example, Ser-linked substrates are induced upon DNA damage [44,45] and certain human disease-causing viruses remove Glu- and Asp-linked substrates [46]. This enzymatic regulation of the synthesis and degradation of PAR thus enables tight control of phase separation in time and space.

Based on the current framework describing condensate composition, PAR may be considered as a scaffold where its length and structure dictate what client proteins to be recruited into biomolecular condensates [2,3]. Here, we focus on two PAR-mediated condensates, DNA repair foci and stress granules, as examples to illustrate the principles of how PAR modulates the formation and compositional control of biomolecular condensates:

Two case studies of PAR-mediated biological condensates in cells

DNA repair foci

PAR-adding PARPs 1, 2 and MAR-adding PARP3 are recruited within seconds to the site of DNA damage, followed by the recruitment of PAR-binding proteins that are responsible for repairing DNA [7,8,40,47,48]. Activated by DNA breaks, PARP1 modifies itself as the major target as well as other proteins [49]. Published proteome data revealed that PARP1 has at least 10 sites modified during DNA damage [22,50–52], thereby potentially serving as a scaffold emanating multiple PAR chains for binding and condensing proteins.

The length of the PAR polymer is determined by the opposing action of PARPs and degradative enzymes, such as PARG [34] (Box 1). Recent studies indicate that PAR length can also be regulated by proteins, such as HPF1 and MacroH2A1.1 [53,54]. An intriguing aspect of PAR-protein recognition is that the polynucleotide chain length plays a vital role. DNA repair factors XPA and DEK exhibit a strong preference for a long PAR chain length (>40mer), whereas WRN and histone H1 efficiently bind short polymers [36–39]. Such length-selective PAR binding also plays a regulatory role, as demonstrated by Chk1 kinase, for which PAR of >65mers, but not 30mers, activates its phosphorylation [38]. During DNA damage, a distinct shift of polymer length is observed, where long polymers are first synthesized rapidly, but degraded slowly to shorter lengths [35,55]. It is therefore possible that PAR chain length controls the type of protein binding and serves as a timer when recruiting specific proteins in a defined temporal order.

Upon binding to PAR, PARP2 increases its branching activities, suggesting that branching is the second wave of PAR formation at DNA repair foci [40]. These branched structures are recognized by the PAR-binding protein APLF, which regulates chromatin remodeling during DNA damage responses [40]. Therefore, regulating the type of PAR structure provides another means to recruit defined proteins within biomolecular condensates.

Several pieces of evidence support the notion that PAR-mediated phase separation occurs at DNA repair foci in vitro and in cells. PARP1 knockdown or PARP1/2-specific inhibitors (e.g., FDA-approved anticancer drug Olaparib) prevent the recruitment of low-complexity region-containing proteins, such as the abundant DNA repair factor and ALS disease factor FUS, to the foci [7,8,26,56]. Conversely, PARG inhibition or depletion prolongs their presence [7,8,56]. Local production of PAR at a DNA damage site provides a focal point for recruiting other proteins, such as MDC1 or FUS-related proteins EWS and TAF15, but also selectively excluding others (e.g., 53BP1) [7,8]. Consistent with these observations in cells, biochemically purified PAR directly interacts with FUS and forms phase-separated condensates in vitro [7,8,25] (Table 2). The addition of NAD⁺ to purified PARP1 allows the direct recruitment of FUS onto aggregates visible under atomic force microscopy [25]. Importantly, FUS can form condensates at a much lower concentration in the presence of PAR than by itself in vitro [7]. Therefore, PAR facilitates the phase separation of DNA repair factors.

Table 2 |.

Effect of PAR in phase separation observed in vitro.

Proteins	Effect of PAR in phase separation	Methods	References
FUS	PAR induces FUS to form liquid droplets in the presence of the crowding reagent dextran.	Fluorescence microscopy	Patel 2015 [7]
FUS, EWS TAF15	PAR induces FUS to form aggregates.	Transmission electron microscopy, formaldehyde cross-linking experiments	Altmeyer 2015 [8]
FUS	Addition of NAD+ to PARP1 induces FUS to form aggregates. Similarly, PAR also induces FUS to form aggregates	Atomic force microscopy	Singatulina 2019 [25]
TDP-43	PAR induces TDP-43 to form liquid droplets in the presence of the crowding reagent dextran.	Light microscopy	McGurk 2018a [10]
TDP-43	PAR inhibits TDP-43 to form aggregates.	Optical turbidity assay, Transmission electron microscopy	McGurk 2018b [57]
TDP-43, hnRNPA1	PAR induces hnRNPA1 to form liquid droplets, which can be co-mixed with TDP-43.	Light and fluorescence microscopy	Duan 2019 [11]
α-synuclein	PAR induces α-synuclein to form aggregates.	Transmission electron microscopy, thioflavin T fluorescence	Kam 2018 [24]

Stress granules

Stress granules are a class of cytoplasmic condensates of RNA and proteins formed in response to a variety of stressors and are involved in post-transcriptional gene regulation, antiviral activities and apoptosis [58–60]. Dysregulation of stress granule formation is implicated in the pathogenesis of viral infection, cancer and neurodegeneration [60–62]. Stress granules are enriched with PAR, five PARPs (5a, 12, 13, 14, 15) and PARG [9–11,63]. Overexpression of any of these PARPs, including those adding MAR only, induces the formation of stress granules [9]. Moreover, the overexpression and knockdown of PARG suppress stress granule formation and disassembly, respectively [9,11]. Many stress granule proteins are PARylated (e.g., G3BP1, FUS, hnRNPA1) [9–11] or bind PAR (e.g., PARP12, G3BP1, FUS, TDP-43, hnRNP A1) [8,10,11,21,63]. As in the case of DNA repair foci, purified PAR facilitates the formation of condensates of stress granule proteins, such as FUS, TDP-43 and hnRNPA1, in vitro at a much lower concentration than by themselves [7,8,10] (Table 2). These data suggest that PAR is critical for the structural integrity of and protein targeting to stress granules.

Given that FUS, TDP-43 and hnRNPA1 are disease factors implicated in ALS and other related diseases such as frontotemporal dementia and stress granule-like pathological aggregates are observed in patients of these diseases [61,62], there are considerable interest to examine whether the formation and dynamics of stress granules or targeting of their components to the condensates can be pharmacologically modulated by inhibiting PARPs. Chemical inhibition of PARP5a/b does not suppress stress granule formation [10]. However, inhibition of the only remaining PAR-adding PARPs 1,2 can block the localization of PARP12, G3BP1 and TIAR to stress granules [21,63,64], suggesting that other PARPs are redundant in PAR formation in stress granules. As in the case with PARP5a/b inhibitors, PARP1/2 inhibitors alone do not inhibit the formation of stress granules [21,63]. Therefore, it is possible that all PAR-adding PARPs need to be completely inhibited to suppress stress granule formation, or only very small amounts of PAR are required to trigger the phase separation of stress granule proteins. Alternatively, MAR-adding PARPs may play a role in coordinating with these PAR-adding PARPs in the synthesis of PAR that is critical for stress granule formation (Box 1).

Knowledge gaps related to PAR-mediated phase separation in cells

Several fundamental parameters (site, length, structure) critical for evaluating how PAR regulates these biomolecular condensates in cells are currently unknown, partly due to a lack of tools. In the following section, we focus on tools that are required to investigate these parameters and address how physiologically-relevant concentrations and types of PAR regulate the formation, material state and compositional control of biomolecular condensates.

Identify PARylated sites and PAR-binding proteins to determine how PAR regulates the composition of PAR-mediated condensates

Considering PAR as a scaffold of biomolecular condensates, defining the PARylated proteins will indicate the possible starting points of the scaffold for recruiting client proteins. A PARylated protein can be experimentally defined through the identification of the sites of modification [65], and this site information in turn provides the theoretical valency for each scaffold protein. Given that the PAR length may specify which proteins it binds [36–39], it is also critical to identify the length of PAR associated with each site. Existing proteomics approaches to identify modified sites all require the trimming of the polymer to a unique mass tag and thus cannot distinguish whether the modification sites are MARylated or PARylated [50,51,66,67]. Innovative approaches are therefore needed to simultaneously identify both the modified sites and the number of ADP-ribose groups that are attached, similar to what is currently being explored in the field of glycosylation to identify the structure of sugar polymers at modified sites (e.g., [68]).

For identifying the client PAR-binding proteins, current approaches involve pulldowns using antibodies or proteins that bind to PAR under native conditions (e.g., [28]). However, these methods do not distinguish PARylated proteins and proteins directly bound to PAR from those associated with these two classes of proteins. Therefore, methods are needed to identify direct PAR-binding proteins, similar to the UV or chemical cross-linking methods that have been developed for identifying RNA-binding proteins [69]. In addition, probes should be developed to globally identify length-selective PAR-binders from these condensates.

Combining proteomics analyses of both PARylated proteins and PAR-binding proteins will therefore uncover the PAR-protein interaction networks for individual biomolecular condensates. In addition, determining which PARPs are responsible for the modification will pave the path for modulating the formation of these biomolecular condensates pharmacologically. Quantification of the protein copy number of PAR-binding proteins and PARylated proteins in these structures will further indicate whether the in vitro conditions of phase separation are physiologically relevant. These systems analyses will reveal what are the key PARylated proteins for the phase separation of which PAR-binding proteins in the condensates. Intriguingly, two recent studies demonstrated that the PARylation of certain substrates, such as the DNA repair factor p53 and the nucleolar helicase protein DDX21, is dependent on their ability to bind to PAR [23,70]. Therefore, one or more PAR-binding-independent PARylation events may serve as a trigger for a wave of PAR-binding-dependent PARylation in their vicinity, forming extensive PAR-protein interaction networks in biomolecular condensates [71].

Determine the length and concentration of PAR in biomolecular condensates in cells

Phase separation depends on the multivalency and concentration of the nucleic acid scaffold for client proteins to bind and condense. However, the length and concentration of PAR remain uncharacterized for individual compartments in cells. DNA repair foci and stress granules can be detected by multiple PAR-specific antibodies that have specificity biases of at least 6–20mers [8,9,72,73]. Profiling the length distribution and concentration measurement will be facilitated by isolating PAR from biochemically purified fractions containing these structures using novel techniques [55,74]. For example, assuming the average chain length of 10mer, it has been estimated that there are ~3,000 PAR molecules/cell at the basal level, with undetectable branching [74]. The PAR level significantly increases to >150,000 molecules/cell with 2% branching upon DNA damage [74]. However, it remains unclear the local concentration of PAR in DNA repair foci. Similar analyses in stress granules-isolated fractions will help determine physiologically relevant PAR concentrations, length and structures. These parameters are critical for building physiologically-relevant in vitro models for mechanistic investigation (Box 2). A faithful model will ultimately use defined PAR-binding proteins and proteins PARylated with appropriate populations of PAR lengths and structures that are relevant to physiological and pathological conditions.

Box 2: Develop physiologically-relevant models of PAR-mediated phase separation.

In vitro models of phase separation have been instrumental for investigating the properties of PAR-mediated condensates in cells (Table 2). When considering the physiological relevance of in vitro models of PAR-mediated condensates, attention should therefore be paid to the source of the PAR. First, all published studies have used PAR derived from the automodification of PARP1 (e.g., Trevigen #4336–100-01) [7,8,10,11,24,57], which does not localize to stress granules but is present in DNA repair foci. As PARP1 automodification makes both linear and branched polymers [41], it is unclear how branching affects phase separation. Second, this PAR polymer mixture ranges from 2–300 units in length, making it impossible to evaluate the actual concentration of PAR and thereby its potency in phase separation. As in the case of RNA [88], PAR may have differential effects on phase separation based on the chain length. Therefore, it is important to systematically evaluate the effect of PAR of single-chain lengths in phase separation. Third, the use of PAR with mixed chain lengths may obscure the length-selective effect of protein binding [36–39]. Therefore, it remains unclear how the observed PAR-mediated phase separation in vitro (using PARs of variable sizes, branching and concentration) relates to what occurs in biomolecular condensates under physiological conditions.

Though PAR facilitates the phase separation of proteins with low-complexity regions [7,8,10,11,24,25], some proteins could form different material states of phase-separated condensates depending on the experimental conditions (Table 2). For example, FUS in the presence of PAR and the crowding agent dextran forms spherical liquid droplets [7]. With PAR alone, FUS condensates appear as irregularly shaped aggregates [8]. Similarly, TDP-43 forms liquid droplets in the presence of PAR and the crowding reagent dextran [10], but, in the absence of dextran, PAR prevents the formation of TDP-43 condensates [57]. The latter phenomenon may be similar to observations in RNA, where phase separation is triggered by a low molar ratio of RNA:protein but inhibited by a high molar ratio [88]. Future studies would thus benefit from using a population of PAR of defined length and structure that mimics relevant physiological concentrations.

Examine how length, structure and other physicochemical properties of PAR affects complex coacervation

As a highly negatively charged molecule, PAR likely serves as a counterion for positively charged regions of proteins to undergo complex coacervation. The electrostatic interaction between PAR and protein is critical, as phase separation can be disrupted by negative charges such as phosphorylation on peptides [8]. Based on recent examples demonstrated by proteins, RNA and other macromolecules, the charge density and patterning are critical for triggering complex coacervation [75,76]. Compared with RNA or DNA, PAR has an additional phosphate per monomer where the two phosphates are flanked by two ribose groups, with interspersing hydrophobic adenine base for π-stacking (Fig. 1). This additional negative charge and intermittent spacing likely result in an electrostatic environment for proteins to condense in a manner distinct from other nucleic acids. As a case in point, high concentrations of RNA in the nucleus prevent FUS from forming condensates [77], yet PAR induced at the site of DNA damage still allows the formation of FUS-enriched DNA repair foci [7,8]. These data indicate that the condensation of FUS is not only determined by the amount of negatively charged polymers available in cells but also by the type of the polymer.

The dynamics and material states of condensates are clinically relevant as it may underlie the cause of devastating protein aggregation diseases [3,78]. Emergent data revealed that these two properties of biomolecular condensates can be controlled by the charge density, length and secondary structure of the nucleic acids or other charged polymers that trigger phase separation [2,3]. For example, Alzheimer’s disease factor Tau forms solid-like fibrils in the presence of heparin (charge density ~4.0/nm) but forms liquid-like droplets in the presence of RNA (charge density ~ 3.0/nm) [75]. Therefore, it is critical to understand how physicochemical properties and charge segregation of PAR may impact charge neutralization and various interactions (e.g., π–π or cation–π) during complex coacervation.

Though different lengths of PAR allow selective protein binding [36–39], the molecular basis behind this phenomenon remains unclear. One possibility is that different chain lengths confer distinct secondary structures. Molecular simulation suggests that a 5mer-PAR adopts a compact structure whereas a 25mer-PAR adopts a multi-globular conformation [79]. As in other nucleic acids, PAR has the potential to form a regular helical structure, driven by hydrogen bonding and base stacking. Though no regular secondary structure is observed with PAR at physiological salt concentrations [80,81], NMR studies revealed that adenine bases of PAR are in anti-conformation [80], which makes them accessible for possible interactions with other macromolecules in cells. Indeed, circular dichroism studies revealed that PAR adopt helical conformations under various buffer conditions, including 0.1 mM spermine, 0.5 mM CaCl₂, 0.5 mM MgCl₂, NaCl > 3M or pH >5 [82]. It would be of interest to test whether or not helical conformation could be formed through binding to appropriate counterion ligands (e.g., protein or nucleic acids) that mimic some of these conditions. On the other hand, recent infrared spectroscopy studies revealed that the length-dependent affinity for p53 to PAR is not because of any change in PAR structure but is due to the increased amounts of disordered regions present in the protein [81]. These data suggest that different PAR polymer lengths may induce differential changes in the conformation of binding proteins.

Finally, physicochemical properties of PAR could be altered through branched linkage or chemical modifications, which are mostly unexplored. Branched PAR structures confer distinct conformations, which are observable by electron microscopy [83] or atomic force microscopy [25], thereby endowing distinct multivalent interactions compared to its linear counterpart. The nucleoside adenosine in PAR can also potentially be modified with methyl groups [84], whereby this modification in the case of RNA potentiates phase separation [85]. On the other hand, ADP-ribose itself can be modified with AMP variants at the 2’OH terminus [55,86], which may alter the protein-binding partners of the PAR terminus [87].

Concluding Remarks and Future Perspectives: Three Takeaways

PAR code to direct composition and properties of biomolecular condensates

As an organizer of cellular architecture, PAR serves as a multivalent platform that can be generated de novo with diverse moieties (site, length and structure). The PAR code provides a molecular grammar for cells to specify interactions with proteins (Fig. 1), which may in turn direct the formation of time- or location-specific biomolecular condensates where PAR serves as a scaffold (see Outstanding Questions). As in the case of RNA [88], the physicochemical properties of PAR may have important ramifications on the size, dynamics and material state of the resulting condensates. As all PAR formation is mediated via MAR and NAD⁺, their roles in coordinating the formation of PAR-mediated condensates in cells should also be explored (Box 3). Given that PARylation is enriched in nucleic acid-binding proteins [50,51,65] and many PAR-binding proteins also bind RNA or DNA [15,16], the presence of PAR in the mix of RNA- or DNA-mediated biological condensates may create competition for protein binding and thus provides a means for compositional control of biomolecular condensates. Notably, in vitro condensates of the RNA-binding protein hnRNPA1 have slower dynamics in the presence of PAR [11]. Therefore, PARylation may provide a trigger to modulate the material property of or create a subcompartment within the RNA- or DNA-enriched biomolecular condensates.

Outstanding Questions Box.

• Fundamental properties of PAR

  • What are the fundamental physiochemical properties (e.g., flexibility/stiffness, structures) of PAR?
  • How does PAR bind to proteins in a length-selective manner?
  • Do length and structure (e.g., branching) of PAR matter in phase separation?
  • What is the critical threshold of PAR to trigger phase separation?
  • What determine PAR to be an inducer or inhibitor of phase separation?
• Fundamental properties of PAR-mediated condensates in cells

  • What are the concentration, length distribution and structure of PAR in biomolecular condensates?
  • Does PAR alter the size, dynamics and material states of existing DNA- or RNA-enriched biomolecular condensates? If so, how?
  • What are the constituents of PAR-protein interaction networks within biomolecular condensates?
  • Are there key nodes of PAR-protein interaction networks? If so, which PARPs modify these key node proteins?
  • Is PAR required for the formation and maintenance of biomolecular condensates?
  • Could PAR be required for initiating the formation but not for the maintenance of the organelles?
  • Are these PAR-enriched cellular structures indeed formed by phase separation? or by other mechanisms?
  • What are the roles of MARylation and NAD⁺ in PAR-mediated phase separation?
• Therapeutic opportunities in modulating PAR-regulated phase separation

  • How do PAR levels correlate with aging-related conditions and diseases?
  • Would an abnormal increase in cellular PAR level trigger aberrant phase separation?
  • What are the potential therapeutic targets in modulating PAR-mediated condensates?

Box 3: Things to watch regarding biomolecular condensates from PAR-related fields.

MARylation

Though in vitro data clearly indicate that protein-free MAR does not promote the phase separation of low-complexity region-containing proteins [10], the role of MARylation in phase separation remains unclear. In Drosophila, the formation of membrane-less Sec bodies can be inhibited by depleting the MAR-adding PARP16 [96]. Many biomolecular condensates contain both MAR- and PAR-adding PARPs (Table 1). In the case of stress granules, overexpression of any of the localized MAR-adding PARPs is sufficient to induce granule formation [9,63]. As all PAR formation is mediated through MAR, it may be possible that the initial conjugation of ADP-ribose is a rate-limiting step in polymer formation. Consistent with this possibility, recent data revealed that stress granule formation can be suppressed by an enzyme that removes protein-conjugated MAR but not PAR [97]. Combinatorial genetic depletion of MAR- and PAR-adding PARPs may reveal how these different classes of ADP-ribosyltransferases cooperate in condensate formation. Future studies should aim to determine the mechanistic roles of MARylation in PAR formation and phase separation. These studies will be greatly facilitated through developing reagents that specifically detect MAR or site-specific MARylation as well as inhibitors that target MAR-adding PARPs.

NAD⁺

ATP is a major modulator of phase separation, either serving as a hydrotrope to solubilize aggregated proteins or as an energy source for various proteins, such as helicases, chaperones and disaggregases, which are critical for biomolecular condensate formation [98,99]. Compared with other nucleic acids and protein modifications, PAR is special in that its synthesis requires NAD⁺ instead of ATP. Therefore, PAR provides an ATP-independent track to modulate phase separation in cells, which may be important in cellular contexts when the pool of free ATP is limited. In addition, the physiological concentration of NAD⁺ is ∼100 μM, and the K_m for NAD⁺ of different PARPs ranges from <1 μM (PARP1) to ∼600 μM (PARP16). Therefore, the local supply of NAD⁺ through enzymes, such as nicotinamide mononucleotide adenylyl transferases (NMNATs), may regulate the rate of ADP-ribosylation by individual PARPs. Given that NMNAT-1 localizes in the nucleus, NMNAT-2 in the cytoplasm and NMNAT-3 in mitochondria [100], these enzymes may help regulate local PARP activities and thus phase separation.

Innovative tools to reveal fundamental parameters of PAR

The field of PAR-mediated condensates cannot move forward without determining the feasibility of phase separation in these cellular structures. Rigorous quantitative analyses should be performed to determine whether these structures are formed by phase separation or other mechanisms [89]. Therefore, characterization of the critical parameters of PAR (site, length, structure and concentration) in cells is urgently needed. For example, we need to define how different lengths and structures of PAR regulate the size and dynamics of phase-separated condensates in vitro and in cells. Given that chemical methods have been developed to synthesize linear PAR chains and the core of the branched structure [90–92], it may be possible to perform systematic analyses on how phase separation is regulated by specific PAR chain length and structure in the future. Finally, we need to develop approaches for comprehensively mapping the PAR-protein interaction networks in these biomolecular condensates. Though tools are emerging to identify the conjugation sites and concentrations of PAR in cells [50,51,65,66,74], techniques for monitoring the length, structure or possible modifications of PAR are lacking. Particularly, innovative approaches are needed to monitor PAR chain formation and branching in cells. With this new information, we could then revise our in vitro models with physiologically relevant PAR-binding proteins and PARylated proteins to investigate the underlying mechanisms and properties of PAR-mediated condensates.

Therapeutic strategies to modulate PAR-mediated biological condensates

PAR-mediated condensates are linked to cancer, virus infection and neurodegeneration (Table 1), and PAR can potentiate the formation of pathological aggregates. PAR levels in cerebrospinal fluid are increased in Parkinson’s patients, and PAR accelerates the formation of the pathologic α-synuclein aggregates in vitro (Table 2) [24]. Because the enzymes involved in ADP-ribosylation are druggable [14], many diseases may be treated with inhibitors to modulate the formation and dynamics of these physiological and pathological condensates. Four PARP inhibitors are FDA-approved to treat various cancers based on the role of PARP1 in DNA repair [14,93]. Notably, these PARP inhibitors also reduce the neurotoxicity in models of Parkinson’s disease, ALS and other neurodegenerative diseases [24,64]. In addition, PARP or PARG inhibitors alter the localization of disease factors, such as FUS or TDP-43, to condensates [7,8,10,56], thus providing yet another therapeutic strategy. Lastly, besides focusing on inhibiting the enzymes that regulate PAR formation and degradation, therapeutic strategies might also be developed to target effectors that recognize or regulators that control the site, length and structure of PAR. Taken together, repurposing existing drugs or developing new inhibitors may allow a new avenue for therapeutics by targeting the formation and dynamics of condensates or trafficking of their components.

Highlights.

Poly(ADP-ribose) (PAR) is a nucleic acid and a protein modification enriched in biomolecular condensates, such as DNA repair foci and stress granules.

Informatics analyses revealed that PAR-conjugated proteins and PAR-binding proteins are enriched with low-complexity regions that support phase separation in biomolecular condensates.

In vitro purified PAR facilitates the phase separation of low-complexity region-containing proteins, such as FUS, TDP-43, hnRNPA1, alpha-synuclein—many of which are disease factors.

PAR polymerase (PARP) inhibitors affect the formation and dynamics of these organelles or trafficking of their components.

Fundamental parameters, such as the conjugated site, length and structure, of PAR in cells are needed to evaluate the underlying roles, mechanisms and possibly therapeutic modulation of PAR-mediated phase separation.