Regulation of PARP1/2 and the tankyrases: emerging parallels

The post-translational modification (PTM) of biomolecules is a key currency in signal transduction, driving decision making in virtually every aspect of cell function [1]. The impact of signalling enzymes responsible for creating PTMs is profound, and their activities need to be tightly controlled to avoid aberrant signalling. This control encompasses three basic levels: (1) maintaining the basal (inactive) state, in which signalling enzymes remain responsive to activation cues, (2) their acute and robust activation following a positive-regulatory input, and (3) the re-establishment of the basal state once the signalling event has concluded.

ADP-ribosylation is a versatile and prevalent PTM that controls cellular events in diverse biological processes and subcellular compartments [2,3]. ADP-ribosyltransferases (ARTs) cleave off nicotinamide from nicotinamide adenine dinucleotide (NAD⁺) and transfer the resulting ADP-ribose moiety onto protein and nucleic acid targets [2,3]. ADP-ribosylation can be considerably complex, occurring as both mono- and poly-ADP-ribosylation (MARylation and PARylation; Figure 1A). Common to both, ADP-ribose is attached to amino acid side chains of either aspartate, glutamate, serine, arginine or cysteine, amongst others [2]. In PARylation, the priming ADP-ribose is next extended through ribose-ribose glycosidic bonds, primarily 2′-1″ linkages (i.e. between an adenine-linked and previously nicotinamide-linked ribose), which gives rise to linear PAR chains. Connecting two previously nicotinamide-linked riboses (2″-1′″) introduces branches into the PAR chain (Figure 1A).

The ART superfamily includes a subfamily of proteins whose catalytic fold resembles that of Diphtheria toxin, the so-called ARTDs [2]. These encompass 17 members which share the catalytic domain but diverge in their accessory modules, which link these proteins to different cellular processes. Four members, namely PARP1, PARP2, tankyrase 1 (TNKS) and tankyrase 2 (TNKS2) produce PAR (Figure 1B), whilst others either produce MAR or are catalytically inactive [4]. PARP1 and 2 are best known for their functions in the DNA damage response, as early responders to DNA single- and double-strand breaks (DSBs) [5], and are targets of clinical PARP inhibitors in breast, ovarian, fallopian tube, peritoneal, prostate and pancreatic cancer [6]. Clinical PARP inhibitors exploit the concept of synthetic lethality whereby cancer cells deficient in breast cancer type 1/2 susceptibility protein (BRCA1/2), or bearing related defects in homology-directed DNA damage repair, are exquisitely sensitive to PARP inhibition [7–9]. TNKS and TNKS2 (from here on collectively referred to as ‘tankyrase’ for simplicity) fulfil a wide range of cellular functions [10–12], modulating several key cellular processes, including telomere length homeostasis [13–15], mitosis [15–19], and Wnt/β-catenin signalling [20,21]. Because of their central role in these disease-relevant pathways, tankyrases have attracted attention as potential drug targets, particularly in the context of colorectal cancer [11,22]. Whilst our knowledge of PARP1/2 regulation is relatively advanced, following extensive mechanistic studies over the past few decades [23], how tankyrase is regulated had largely remained unknown. Recent new insights into tankyrase regulation [24,25] indicate that overall regulatory principles between these seemingly divergent proteins may be shared, whilst the details of the molecular mechanisms are unique. Here, rather than offering a comprehensive review of ART regulation, we compare selected mechanistic aspects that underpin the control of PARP1/2 and tankyrase, and point to open questions in our understanding of tankyrase regulation.

Tankyrase and PARP1/2 are multi-domain proteins and, like other signalling enzymes, undergo quaternary structure rearrangements as they progress through their activation cycles. PARP1 bears three N-terminal zinc finger domains (Zn1, Zn2, Zn3), followed by a BRCA1 C-terminal (BRCT) domain, a WGR domain, named after a conserved Trp-Gly-Arg motif, and the catalytic module, composed of a helical domain (HD) and the ART catalytic domain [23] (Figure 1B). PARP2 is substantially shorter and lacks the zinc fingers and the BRCT domain [23]. The tankyrases share five N-terminal ankyrin repeat clusters (ARCs), followed by a polymerising sterile alpha motif (SAM) domain and the catalytic ART domain [12]; additionally, TNKS bears an N-terminal, presumably disordered region enriched in homopolymeric stretches of histidine, proline and serine residues (Figure 1B). Interactions of the ART accessory domains control catalytic activity and function of these enzymes and render them responsive to activation cues in their respective cellular environment.

The ARTD catalytic domain fold consists of two β-sheets with adjacent helical regions and loops [26,27] (Figure 1C). The binding site for the co-substrate NAD⁺ consists of two sub-sites: the adenine subsite (A-site) accommodates the adenine moiety of NAD⁺ whilst the nicotinamide subsite (N-site) harbours its nicotinamide portion (Figure 1C). The adjacent acceptor site engages the substrate for modification: either the polypeptide containing the target amino acid or the growing PAR chain [28]. The ART domain folds of PARP1/2 and TNKS/TNKS2 are highly similar, but each also have unique features (Figure 1C). First, the ART domain of PARP1/2, as well as the mono-ARTs PARP3 and PARP4, is preceded by a HD that is absent from the tankyrases [2,29]. Second, the ART domain of the tankyrases is unique in that it co-ordinates Zn²⁺ through three cysteines and one histidine side chain, all residing within a short zinc-binding loop [26] (Figure 1C). Zinc does not directly contribute to catalysis but plays a stabilising structural role, consistent with its removal disrupting or destabilising the ART fold [30]. The zinc binding site has been proposed to respond to oxidation by releasing zinc, pointing towards a potential regulatory mechanism [31]. Whilst PARP1/2 produce branched chains of poly-ADP-ribose (PAR), only linear and substantially shorter PAR chains have been observed for tankyrase (Figure 1A) [28,32].

PARP1 is highly abundant: based on mass spectrometry, the average cellular concentration of PARP1 (in HeLa cells) was estimated to be ≈2 µM [33]. Considering the nuclear localisation of PARP1, its actual concentration may be about one to two magnitudes higher [34,35]. Paired with the high local concentration of DNA [36], the high affinity of PARP1 for damaged DNA (in the nM range) [37], and its high catalytic activity [38], this illustrates the critical importance of keeping PARP1 in check to prevent aberrant activation. In the absence of damaged DNA, PARP1 displays low activity and is thought to adopt a beads-on-string architecture of domains [39]. Recent research has revealed that this basal activity state is maintained by autoinhibition through the HD, which in both PARP1 and PARP2 sterically blocks access of NAD⁺ to the donor site [29]. HD deletion dramatically increases catalytic activity, in agreement with its autoinhibitory function [40]. A co-crystal structure of the catalytic domain of PARP1 lacking the adjacent HD, bound to the non-hydrolysable NAD⁺ analogue benzamide adenine dinucleotide (BAD), provided the first direct insights into co-substrate binding by PARP1 and explained the inhibitory function of the HD [29]. Three acidic residues in the HD would clash with the adenine base and phosphate groups of BAD in the donor site (Figure 2A). Binding assays confirmed this mechanism: mutation of these residues increased BAD binding and DNA-independent PARP1 catalytic activity [29]. NAD⁺ binding assays suggested that a similar autoinhibitory mechanism operates in the other DNA-dependent ARTs, PARP2 and PARP3 [29]. Preventing pre-loading of PARP1 with NAD⁺ might reduce aberrant activation and off-target PARylation [29]. Nonetheless, it is well appreciated that PARP1 displays low basal activity that is functionally relevant, for example in transcriptional regulation [41]. This basal activity may be brought about by dynamic transitions of the HD into the partially unfolded state (see below), which would stochastically relieve autoinhibition and provide windows of relatively low but physiologically relevant activity [29]. Interestingly, whilst the HD is also present in PARP4 (also known as vault PARP or vPARP), it does not fully inhibit catalytic activity in this context [42]. The crystal structure of the PARP4 HD-ART was recently determined, showing that the HD is positioned away from the ART active site, thereby not interfering with NAD⁺ access. The PARP4 HD-ART unit can bind BAD and is catalytically active. However, deletion of the PARP4 HD nonetheless increases catalytic activity, showing that it still contributes to PARP4 regulation [42].

Compared with PARP1, the cellular concentrations of TNKS and TNKS2 are much lower (≈5 and 2 nM, respectively, in HeLa cells) [33]; nevertheless, tankyrase's enzymatic activity is safeguarded. Tankyrase displays low basal activity in its non-polymeric state and is fully activated only once it assembles into filamentous polymers through its SAM domain. This activity switch, first reported 20 years ago [43] and confirmed in later studies [24,44–46], pointed to a potential allosteric regulation mechanism. The precise mechanistic basis for tankyrase's ground state still remains unclear, but recent structural insights provided a first picture of how tankyrase undergoes the polymerisation-dependent catalytic switch [24]. As for PARP1, NAD⁺ binding appears to be regulated, although this awaits formal experimental conformation. Structural data suggest that in tankyrase, NAD⁺ binding is impeded by the so-called donor or D-loop within the ART domain itself [24]. The D-loop sits atop the NAD⁺ binding site (Figures 1C and 2B). X-ray crystal structures have indicated that the D-loop is mobile and conformationally responds to small-molecule inhibitors binding to the A-site [47]. An extensive comparison of published X-ray crystal structures of tankyrase ART domains [24] suggests that in the basal state, the N-terminal portion of the D-loop, termed the ‘D-loop base’, adopts a closed conformation in which a well-conserved histidine side chain (H1201^TNKS, H1048^TNKS2) partially occupies the NAD⁺ binding site, thereby blocking binding of the NAD⁺ adenine group to the A-site (Figure 2B). Tankyrase's low basal activity [24,44] may reflect the mobility inherent to the D-loop, which allows it to sample the open conformation compatible with NAD⁺ binding.

PARP1 binds undamaged DNA during DNA surveillance but is not substantially activated unless it encounters a DNA lesion [48,49]. Its interaction with intact DNA involves all three N-terminal zinc fingers and importantly the BRCT domain [50]. This binding mode prevents the conformational changes required for PARP1 activation following its encounter with DNA lesions.

Interaction of PARP1 and PARP2 with damaged DNA generates striking foci visible by fluorescence light microscopy [51,52] and triggers dramatic activation of PARP1/2 by three orders of magnitude [53]. PARP1 is extraordinarily versatile in the types of DNA damage it recognises [54]. Over the past decade, numerous structural studies, including X-ray crystallography, cryo-EM and NMR spectroscopy, have shed light onto the molecular details of PARP1/2:DNA complexes [23]. In PARP1, damaged DNA is sensed by two of its three zinc finger domains, whereby Zn2 binds first, enabling the subsequent binding of Zn1 [37]. This creates a nucleus for the stepwise assembly of active PARP1 on the DNA lesion, which involves Zn3, then the WGR domain, and finally the unit of the HD and catalytic ART domain. This mechanism ensures tight regulation of PARP1, whilst at the same time enabling PARP1 to rapidly and sensitively respond to DNA damage [37]. The chief consequence of DNA binding is an extensive allosteric change, primarily in the HD, which relieves the autoinhibited state and enables PARP1/2 to bind NAD⁺ for subsequent target ADP-ribosylation.

Biochemical studies, X-ray crystallography and hydrogen-deuterium exchange coupled to mass spectrometry (HDX-MS) have revealed that DNA binding induces distortion and mobility within the HD, thereby releasing autoinhibition [23]. The first crystal structure of a multi-domain assembly of PARP1 on damaged DNA showed a compact organisation with extensive inter-domain contacts that communicate DNA binding to the catalytic domain via the HD (Figure 3A,B) [55]. A series of recent crystal structures of PARP1 mutant variants has shed further light on the active state of PARP1 [56] and showed that the complex crystallised earlier [55] likely represents a semi-activated transition state [56]. CRISPR mutagenesis screening of PARP1 identified a mutant variant in which two residues (V687, E688 in HD helix αB) were deleted. PARP1 ΔVE, although showing elevated DNA-independent activity, could still be substantially activated by damaged DNA, indicating a degree of retained regulation [57]. Its crystal structure in the DNA-damage bound state provided a snapshot of the fully active state (Figure 3A,B) [56]. Whilst the ART conformation does not change compared with the equivalent, incompletely activated wild-type PARP1 complex [55], there are dramatic changes to the HD conformation and interdomain communication within PARP1 (Figure 3A,B). Restructuring of helix αB and the αB-αD linker enables the HD to engage a hydrophobic pocket on the WGR domain. Relative to the remaining domains, the entire ART domain and HD helix αF rotate by ≈30°, and helix αF pulls away from the ART domain. Together with the other conformational changes in the HD, this creates an entry path for the NAD⁺ co-substrate [56], which can now bind unhindered. This was indeed confirmed by a co-crystal structure of PARP1 ΔVE with the NAD⁺ analogue EB-47, which is accommodated without any need for further conformational changes (Figure 3A) [56].

Early biochemical studies, particularly size exclusion chromatography showing a concentration-dependent putative dimer peak, suggested that full-length PARP1 operates as a dimer [58]. The dimerisation model is compatible with, but not unequivocally proven by, observed kinetic reaction parameters [59], fluorescence-spectroscopy studies of PARP1 Zn1–Zn2 DNA binding [60], zinc finger domain interactions within a PARP1 Zn1–Zn2:DNA crystal structure [61], the intriguing observation that PARP1 domains from different polypeptide chains can complement each other to form a functional PARP1 assembly [53,61–63], and the assumption that auto-PARylation is more readily achievable within the context of a dimer given steric considerations. However, intramolecular complementation using purified proteins only achieves incomplete activation [63], and analogous complementation studies in cells do not restore PARylation in response to induced DNA damage [63]. Mutating the Zn3 dimer interface does not affect DNA-dependent PARP1 catalysis [53], although this does not rule out other contributors to potential dimerisation.

Conversely, there is ample direct evidence that PARP1 regulation occurs in the context of a monomer, as visualised by X-ray crystallography [55]. Analytical ultracentrifugation (AUC) showed that full-length PARP1 is monomeric [64] and engages a single-strand break, a gapped DNA dumbbell lacking any further DNA termini, as a monomer [37]. Full-length PARP1 was also shown by atomic force microscopy to bind DNA ends, nicks and abasic sites as a monomer [65]. This was confirmed by single-molecule fluorescence studies [65]. NMR spectroscopy confirmed that a DNA single-strand break is recognised by a monomer of the PARP1 Zn1-Zn2 unit [37], previously proposed to act as a dimer [61]. The two zinc fingers occupy distinct positions with a defined directionality on the DNA, interacting with each other in a DNA-dependent manner [37,66]. Steric clashes between the domains are avoided by extensive, Zn2-induced distortion of the continuous DNA strand, which accommodates Zn1 [37]. The Zn1–Zn2 linker, proposed to act as an ‘entropic spring’, imposes directionality as the linker would need to bridge a three-fold longer distance if zinc finger positions were reversed [37]. Moreover, a reverse order of Zn1 and Zn2 on the break would require more extensive DNA distortions [37]. In line with these considerations, it seems more plausible that Zn1 and Zn2 from the same rather than different polypeptide chains engage the break. In an earlier study, shortening of the linker by nine amino acids did not impair DNA damage recruitment of PARP1 Zn1–Zn2 [61]. This is entirely compatible with the NMR-based model of PARP1 Zn1–Zn2 [37]. A single, full-length PARP1:DNA complex can be modelled based on the NMR and crystal structures without any domain clashes [37]. Directionality of Zn1–Zn2 is critical, as only in the observed DNA binding mode Zn1 can interact with Zn3 and WGR for allosteric communication [37]. NMR suggests weaker interactions of Zn3 with the Zn1–Zn2 unit, and these require Zn3 to be part of the same polypeptide chain [37], again supporting the monomer model. Subsequent binding of the WGR domain cooperatively stabilises the pre-formed interdomain interactions. Multi-domain assembly of monomeric PARP1 with subsequent activation is further supported by its ability to modify itself in cis, with trans-auto-PARylation only observed when activating DNAs contained two proximal PARP1 binding sites [37]. Biophysical studies of an N-terminal fragment of PARP1 encompassing Zn1, Zn2, Zn3 and the BRCT domain, including size exclusion chromatography coupled with multi-angle light scattering (SEC-MALS), small-angle X-ray scattering, AUC and fluorescence spectroscopy, also indicate that PARP1 operates as a monomer on a range of DNA damage model species [39]. Dimeric, or even higher-order PARP1:DNA assemblies were observed where the DNA contained more than one binding site [64]. In these instances, PARP1 may occupy the duplex termini as well as interact along the DNA [34,64].

Recently, Chappidi et al. [34] proposed that PARP1 dimerisation drives the formation of higher-order species (‘condensates’) of PARP1 that enable synapsis of DNA DSBs. Whilst SEC-MALS studies leave the possibility that DNA bridges two end-associated PARP1 molecules, single-molecule Förster resonance energy transfer (FRET) experiments support PARP1 dimerisation. However, the latter was performed under low-salt conditions (10 mM KCl) [34], which limits comparability with other studies. Studies proposing the assembly of PARP1 into functional higher-order biomolecular ‘condensates’, potentially assisted by PAR-binding downstream effector proteins such as Fused in Sarcoma (FUS), have re-opened the case for intermolecular interactions governing PARP1 activation and generating spatially confined DNA damage repair compartments [34,67–69].

In conclusion, whilst there is convincing direct evidence for a model in which PARP1 is regulated in the context of a monomer, a dimerisation-dependent mechanism and the formation of higher-order assemblies driven by protein:protein interactions requires further investigation. Methods that directly visualise monomeric vs. higher-order species, including biochemical approaches and single-molecule imaging, under standardised experimental conditions, will be particularly informative when paired with high-resolution live-cell microscopy.

Tankyrase recognises its effector proteins (substrates and binders) through its N-terminal ARCs. Effector proteins contain peptide motifs termed tankyrase binding motifs (TBMs), which are recognised by four out of five ARCs, namely ARCs 1, 2, 4 and 5 [70,71] (Figure 7B). From an amino acid scan of a known tankyrase effector, 3BP2, the TBM was determined as an octameric peptide with a consensus pattern of: R-[any]-[any]-[small hydrophobic or G]-[D/E/I/P]-G-[no P]-[D/E] [71]. The TBM binding pocket within the ARC is pre-formed, ready to engage with effectors [71]. The presence of multiple peptide-binding ARCs contributes to the avidity of tankyrase:effector interactions, but can also enable the penta-ARC unit to conformationally respond to the binding partner, potentially facilitating substrate positioning for PARylation [72]. So, similarly to PARP1, which distinguishes between intact and damaged DNA and adapts its conformation accordingly, tankyrase may adopt different conformations and assemblies depending on its binding partner.

The degenerate TBM is thought to be present in dozens to hundreds of proteins [71], and the complement of known effectors keeps growing. The diversity of tankyrase effectors reflects the diversity of cellular processes tankyrase modulates (see [12,73] for a comprehensive picture of the range of effectors). Some of these effectors, such as AXIN1, AMOT, and NOTCH2, are involved in cell signalling processes which are frequently dysregulated in cancer, namely Wnt/β-catenin, Hippo/YAP, and Notch signalling [15,20,74]. Other tankyrase effectors are involved in the maintenance of genome integrity. BABAM1/MERIT40 is a member of both BRISC and BRCA1-A complexes, which are involved in the DNA damage response, and recruits tankyrase to DNA DSBs [71,75]. Similarly, the recruitment of tankyrase to DSBs by MDC1 promotes homologous recombination (HR) [76]; tankyrase is also able to bind RAD54, another key protein in HR [71]. Originally discovered at telomeres in human cells, tankyrase controls both telomere length and sister telomere resolution during mitosis by targeting TRF1, a member of the shelterin complex [13,14,18,19,77]. Tankyrase is recruited to several cellular locations by its effectors, including the Golgi by LNPEP/IRAP and GOLGIN-45/BLZF1 [78,79], spindle poles by NUMA1 [17,80], and peroxisomes by PEX14 [81], amongst other proteins at some of these sites. Proteomic analyses further revealed that tankyrase-interacting proteins are enriched in diverse cellular compartments and pathways, many of which still require further characterisation [15,81]. Many of tankyrase's cellular functions are evolutionarily well-conserved, as indicated by TBMs in effectors such as AXIN1, GOLGIN-45 and NOTCH2 being present across both vertebrates and invertebrates [12]. In contrast, tankyrase's role in telomere length regulation has emerged more recently and is absent in mice [12,82,83].

Several of tankyrase's effector proteins, including BABAM1, FAT4, and AXIN1, present multiple TBMs in a single polypeptide chain [71,72,84]. This likely facilitates the engagement of multiple ARCs at once, either within a single tankyrase molecule or spanning multiple tankyrase protomers within a filamentous polymer. Similarly, effector proteins with a single TBM can oligomerise, including TRF1, GMDS/GMD, and RAD54, presenting multiple TBMs in close proximity which enhances tankyrase binding and can promote ADP-ribosylation [85]. The combination of multiple TBMs in an effector protein that can also oligomerise, such as AXIN1, may provide a strong clustering effect, promoting tankyrase polymerisation and thereby activation. However, not all tankyrase effector proteins are also PARylated upon binding [71,85–87]. Whether or not an effector is PARylated may depend on several factors, including the solvent accessibility of target residues and whether the location and geometry of TBMs allows for the productive presentation of the effector to the ART domain [72,85]. Interestingly, several of tankyrase's effectors are present in larger multi-protein complexes, including BABAM1, TRF1, and AXIN1. Whether tankyrase is able to PARylate proteins other than the TBM-containing factor following recruitment to such complexes remains to be seen.

Analogous to PARP1/2, structural information supports a model in which interdomain interactions activate the ART domain of tankyrase. In the case of tankyrase, these contacts are driven by filamentous polymerisation, potentially around effector proteins acting as polymerisation nuclei [24,44,45]. A recent cryo-EM study revealed the architecture of a SAM-ART filament of TNKS2 [24] (Figure 4A). The centre of the filament is formed by SAM domains arranged in two mutually attached, antiparallel protofilaments. Contacts between the protofilaments give rise to a minor and major groove of the resulting double-helix. The ART domains protrude outwards, across the major groove of the SAM domain filament core. ART domains extending from opposing protofilaments on either side of the major groove intercalate, forming two distinct homotypic interfaces termed head-to-head and tail-to-tail contacts, respectively (Figure 4B). Head contacts occur close to the catalytic site and are responsible for the polymerisation-induced activation. Similarly to the HD in PARPs1, 2 and 3, which mediates both the inactive state by blocking NAD⁺ binding and the active state by contributing to intramolecular domain communication, the D-loop in tankyrase either appears to block NAD⁺ binding or contributes to the head interface, which in turn stabilises tankyrase polymers [24]. A search of the Protein Data Bank (PDB) using the PISA server revealed 167 TNKS or TNKS2 ART domains in crystal contacts that closely resemble the ART:ART head contact [24], correctly identified previously to regulate tankyrase activity [46]. Remarkably, all these ART chains display an open D-loop base in which the proposed inhibitory H1201^TNKS or H1048^TNKS2 side chain is moved out of the A-site, just as in the cryo-EM structure of the TNKS2 SAM-ART filament (Figure 4C) [24]. Only ≈50% of the ART domains in these crystal structures harbour a small molecule in the A-site, which could reasonably ‘induce’ this pocket (see below). This strongly suggests that the head interactions are responsible for the observed open conformation and enable co-substrate binding. Another change triggered by ART:ART head interactions is the structuring of an otherwise disordered loop, the β6–β7 loop, sitting atop the D-loop and additionally communicating with the D-loop across an ART:ART head pair [24] (Figure 4B,C; Figure 7D below). Mutations disrupting the head contacts potently impair catalytic activity of tankyrase [24]. However, biochemical and biophysical evidence for increased NAD⁺ binding upon head contact formation remains limited. Fluorescence polarisation assays using the fluorescently labelled NAD⁺ analogue BAD showed that polymerisation-competent or -deficient TNKS2 SAM-ART were able to bind co-substrate with similar K_d values[24], but under these conditions, only short filaments without many head-to-head pairs may form. Importantly, formation of a head-to-head ART:ART pair requires a minimum of five protomer pairs to assemble before a head interface can be established, which sets a threshold of polymerisation that must be surpassed before full catalytic activation can occur (Figure 4D) [24]. This likely prevents the spurious activation of tankyrase by the formation of very short transient filaments.

A recent study on PARP15 showed that the PARP15 ART domain can dimerise in solution and that dimerisation is required for ADP-ribosylation activity [88]. HDX-MS and mutagenesis revealed that the dimerisation interface can be found in existing crystal structures of the PARP15 ART domain [88], and closely resembles the head interface seen TNKS2 (Figure 4E). Mutation of residues at this interface abolished PARP15 ART automodification [88]. We conducted a search of ART domain crystal structures in the PDB, showing that in addition to PARP15, head-to-head-like ART dimers are found for PARP10 and PARP14 [24] (Figure 4E), but their tendency to dimerise in solution appears weak [88]. Whether head dimer formation is necessary for catalytic activity in these proteins, and indeed in other ART family proteins which are less well studied and currently lack structural information, remains to be seen. For several other ART family members (PARP7, 11, 12, and 13), AlphaFold 3 [89] predicts dimers in a head-like arrangement, with predictions for PARP7, 11, and 12 being made with high confidence (Figure 4F). Experimental validation of these predictions will be needed to reveal whether such interfaces exist in solution, and their effects on catalytic activity.

PARP1's primary substrate is PARP1 itself, and an extensive body of work has shown that PARP1 auto-PARylation, in addition to mediating downstream effector functions, counteracts its retention on sites of DNA damage [52,90–94] (Figure 5A,B). The mechanism behind this observation is poorly understood: simple charge repulsion between DNA and the highly negatively charged PAR chains likely contributes, but the negative feedback mechanism may be more complex and involve allosteric effects (see below) and additional protein factors. Irrespective of the precise mechanism, PAR-induced dissociation of PARP1 from DNA provides an effective negative feedback loop ensuring clearance of the DNA lesion following its detection and the recruitment of DNA repair factors. PAR is further proposed to trigger the formation of biomolecular ‘condensates’, which may involve the recruitment of PAR-binding proteins and impact downstream molecular events [34,67–69]. A deeper understanding of the molecular impact of auto-PARylation requires the identification of PAR acceptor sites within PARP1. Early efforts to pinpoint the automodification sites identified a stretch including and surrounding the BRCT domain, and this region was termed ‘automodification domain’ [95] (Figure 5A). Within recent years, the ability to unequivocally identify ADP-ribose acceptor sites has increased extensively, owing to the development of tools for trimming PAR chains prior to mass spectrometry, the enrichment of ADP-ribosylated peptides using affinity reagents, ionisation techniques that preserve the linkage between the acceptor amino acid and ADP-ribose, and protein engineering to interrogate the activity of selected ARTs [96–101]. Whilst ADP-ribosylation of aspartate, glutamate, arginine, cysteine and asparagine had been appreciated for some time, and ADP-ribosylation on lysines suggested as well [62,96,102,103], the field experienced a surprise with the discovery of serine as an ADP-ribose acceptor site [97]. Serine ADP-ribosylation sites are frequently found within Lys-Ser (or Arg-Ser) motifs, which suggests a possible reason for previous attribution of ADP-ribosylation to lysines [96,97,104,105]. Lysine ADP-ribosylation can also proceed non-enzymatically [106], and this mechanism was proposed to account for some of the lysine-linked ADP-ribosylation observed by mass spectrometry [107]. With continued mapping efforts, ADP-ribose modification has been further expanded to or confirmed for tyrosine, threonine, histidine and cysteine [98,108,109].

Initially discovered in histones, it quickly emerged that histone PARylation factor 1 (HPF1, see below for more details) is required for PARP1 to switch its substrate preference from acidic amino acids to serines, and away from abundant auto-modification on aspartates and glutamates towards other substrates, including histones [38,104]. Nonetheless, PARP1:HPF1 also auto-PARylates on serines, and some of the most abundant of these sites map to the previously identified ‘automodification domain’ [98]. The mutation of three serine residues within this region (S499, S507, S519, with the first two being particularly critical), but not mutation of six glutamates (E471, E484, E488, E491, E513, E514) (Figure 5A), appreciably reduced automodification, both under basal conditions and upon induction of DNA damage [110]. The triple-serine mutant variant displayed increased PARP-inhibitor induced trapping on DNA. Its residency time on microirradiation-induced DNA damage sites increased, but experiments in an HPF1-null background pointed towards additional sites or factors that determine DNA release of PARP1 following DNA damage [110]. Serine is now thought to be the major acceptor of ADP-ribose following DNA damage [111]. However, the O-glycosidic linkage found in serine-linked ADP-ribosylation is substantially more stable than the chemically labile ester linkage found in aspartate- or glutamate-linked ADP-ribosylation. A recent publication described a workflow to preserve ester-linked ADP-ribosylation during proteomics sample preparation [112]. This revealed that aspartate and glutamate residues are rapidly mono-ADP-ribosylated by PARP1 following DNA damage, with such modifications found both on PARP1 and other target proteins [112].

As PARP1, tankyrase also is subject to auto-PARylation [13]. Given the polymeric nature of active tankyrase, it is likely that tankyrase, unlike PARP1, modifies itself in trans, although this requires formal experimental proof. Tankyrase auto-PARylation regulates tankyrase in several distinct and competing ways [25,113–115]. The E3 ubiquitin ligase RNF146 binds to tankyrase as well as PAR. PAR binding allosterically activates RNF146 to poly-ubiquitylate both tankyrase and its effectors, which in turn initiates their subsequent proteasomal degradation [113–115]. This process is known as PAR-dependent ubiquitylation (PARdU) [116]. Recent work provided insights into an additional layer of regulation of PARdU: two E3 ubiquitin ligases from the RING-UIM family, RNF114 and RNF166, counteract PAR- and RNF146-induced poly-ubiquitylation, acting on the mono-ubiquitin-modified intermediate initially produced by PARdU [25]. RNF114/166 bind mono-ubiquitylated tankyrase and catalyse K11-linked di-ubiquitylation, which stabilises tankyrase as it competes with K48-linked poly-ubiquitylation by RNF146. Unlike all other known tankyrase binders, which are recruited by tankyrase's ARC domains, RNF114/166 bind to the SAM-ART unit [25]. This allows for the formation of a ternary complex between tankyrase, RNF114/166, and tankyrase effectors, enabling K11-linked di-ubiquitylation and stabilisation of tankyrase effectors.

As well as regulating tankyrase stability through downstream E3 ligases, auto-PARylation counteracts tankyrase polymerisation, as indicated by ultracentrifugation sedimentation [43]. Confocal microscopy of fluorescently tagged tankyrase also supports that auto-PARylation negatively regulates polymerisation. Tankyrase displays a punctate cytoplasmic distribution, likely corresponding to tankyrase filaments as introduction of polymer-breaking mutations in the SAM domain renders tankyrase localisation diffuse [24,44,45]. Pharmacologic catalytic inhibition or the introduction of inactivating mutations in the ART domain cause these puncta to increase in both size and number [24,44], likely reflecting larger polymers in the absence of auto-PARylation (Figure 7A). Similarly, tankyrase inhibition stabilises cytoplasmic puncta in SW480 colorectal cancer cells; these puncta increase over time in both number and size after inhibitor addition and rapidly dissolve upon inhibitor washout [117]. Thus, similar to PARP1, tankyrase may be reined in by negative feedback regulation through auto-PARylation. Polymer breakage by auto-PARylation would inactivate catalysis and promote downstream proteasomal degradation by PARdU (Figure 5C, Figure 7A). Whilst the auto-ADP-ribosylation sites responsible for negative feedback regulation remain unidentified, current evidence points towards tankyrase ADP-ribosylation on acidic residues [85,118]. In one study, tankyrase was shown to modify glutamate residues on the effector LKB1 [118]. PARylation of LKB1 by tankyrase was carried out, followed by mass spectrometry which identified several putative acidic modification sites. Subsequent mutation of two glutamate residues substantially decreased the extent of PARylation. In another study, the nature of modification sites of both tankyrase and its effector TRF1 was investigated using the differing specificities of (ADP-ribosyl)hydrolases [85]. PAR chains on tankyrase and TRF1 were first trimmed to MAR by poly(ADP-ribose) glycohydrolase (PARG), and then incubated with either terminal ADP-ribose protein glycohydrolase 1 (TARG1), which cleaves the terminal MAR moiety from aspartate and glutamate residues, or ADP-ribosyl hydrolase 3 (ARH3), which cleaves MAR from serine residues. ARH3 was unable to remove MAR from either tankyrase or TRF1, suggesting that serine is not a major acceptor of PARylation under these conditions [85]. Incubation with TARG reduced the MAR signal from both tankyrase and TRF1, suggesting that aspartate or glutamate residues are modified, although residual MAR signal on tankyrase suggests additional acceptor amino acids [85]. It is possible that ADP-ribose acceptor sites reside directly within interfaces required for polymerisation. In this case, mono-ADP-ribosylation may suffice for negative feedback. Alternatively, PAR chains may non-covalently weaken polymer interactions, which may require a critical PAR chain length to be achieved for the inhibitory effect. Future studies will explore whether such feedback inhibition occurs, and its precise mechanism.

The ability of a PARP inhibitor (PARPi) to promote the retention of PARP1 on DNA was first observed with 3-amino benzamide, which countered PARP1 release from DNA in a human cell-free system [91]. Further evidence that the effect of PARP1/2 inhibitors (PARPis) extends beyond simply blocking ART activity and subsequent DNA repair pathways came from observations that PARPis ‘trapped’ PARP1/2 in the chromatin fraction upon DNA damage but that inhibitors with comparable potencies in terms of catalytic inhibition conferred different extents of trapping [93,119,120]. The ability of inhibitors to trap correlated with their cytotoxicity [93,119,120]. Subsequent mechanistic studies led to the proposal of a mechanism now termed ‘reverse allostery’, which describes the communication of inhibitor binding from the catalytic site to the DNA-binding domains. HDX-MS showed that binding of the NAD⁺ analogue BAD to the ART domain limited solvent accessibility in distal domain interfaces that are engaged upon DNA binding [29]. Co-crystal structures of PARP1 bound to EB-47, UKTT15 (a derivative of veliparib), or rucaparib, alongside HDX-MS, enabled mechanistic insights into reverse allostery [94]. Pro-retention inhibitors, including EB-47, contact residues D766 and D770 in the HD, leading to subsequent HD destabilisation [94] (Figure 6A,B). This change is next communicated through the WGR domains to the Zn domains to promote DNA retention [94]. The resulting model proposes that PARP inhibition not only counters the PAR-induced release from DNA by preventing charge repulsion but additionally induces reverse allostery to further promote PARP retention at damaged sites and promote cytotoxicity by blocking downstream DNA repair in HR-deficient cancer cells (Figure 6C) [121]. Thus, these studies have highlighted reverse allostery as an important factor in the design of next-generation PARPis.

Supporting the notion that trapped PARP constitutes the toxic lesion generated by PARPis, loss of PARP1 confers PARPi resistance [51,57]. High-resolution Cas9-mediated mutagenesis uncovered PARP1 mutant variants that confer PARPi resistance, notably affecting residues D45^Zn1, H742^HD, D743^HD, and E688^HD [57]. These residues establish hydrogen bonds responsible for interdomain communication between the catalytic ART domain and DNA-binding domains [55]. Quantitative live-cell imaging with fluorescence recovery after photobleaching (FRAP) revealed that GFP-PARP1 molecules within the characteristic DNA damage foci exchange rapidly, even in the presence of the PARPi niraparib [51]. Thereby, PARP1 molecules ‘trapped’ by niraparib are not physically stalled at these sites and remain dynamic, despite niraparib increasing the half-life of PARP1 foci. Inactivation by mutation of PARP1's catalytic glutamate (E988A^ART) gives rise to a similar phenotype [122], yet physical stalling of PARP1 was achieved by the H862D^ART mutation, which leads to tighter binding of nicked DNA (compared with wild-type PARP1) as measured in vitro by competitive fluorescence polarisation [51]. HDX-MS studies enabled the classification of PARPis into three categories according to their impact on DNA retention or release: type I (pro-retention), II (no allosteric effect) and III (pro-release) (Table 1) [94]. Surprisingly, clinical PARPis demonstrated either a type-II (olaparib and talazoparib) or pro-release type III (rucaparib, niraparib) effect on PARP1 allostery and DNA retention [94]. In contrast to PARP1, clinical PARP inhibitors exhibit a type-I pro-retention effect on PARP2 (Table 1) [123]. The magnitude of this effect is striking considering the conservation of the PARP1 and 2 ART domains and their shared feature of HD-dependent regulation [29]. As for PARP1, the PARPi-induced HD destabilisation triggers allostery in PARP2, as was demonstrated using a WGR:HD contact mutant variant which was no longer subject to allosteric communication [123]. Modelling the HD from PARP1 bound to DNA and niraparib with a structure of PARP2 on DNA revealed subtle differences in the positioning of the N terminal portion of HD helix αF, which could explain why niraparib destabilises the HD of PARP2, but not PARP1 (Figure 6A,B) [123]. Specifically, niraparib, talazoparib and rucaparib show opposing effects on DNA retention between PARP1 and PARP2, and all contact the divergent N-terminal part HD helix αF (Figure 6B) [123]. Olaparib abuts the middle portion of helix αF and the E322^HD side chain of PARP2. PARP1 bears a shorter aspartate (D766^HD) at this position, providing olaparib with more space to occupy the pocket without contacting helix αF of PARP1 (Figure 6B). The NAD⁺ mimetic EB-47 is the only PARPi to display a type-I effect on both PARP1 and PARP2. Structural modelling demonstrates that EB-47 would contact the C-terminal portion of the helix αF, a region of higher alignment congruence between PARP1 and 2 (Figure 6B) [94].

Recently, olaparib was converted to a type-I PARPi by modulating its A-site binding moiety [124]. This type-I PARPi (AZ0108), but not olaparib, was shown to drive replication stress and tumorigenic cell killing [124]. Further modulation of the A-site substituent of AZ0108 identified Pip6, which was ≈90-fold more potent at killing ES8 tumour cells. Detailed mechanistic experiments demonstrated increased ART residency time for Pip6 [124]. There is a need for PARP1-selective PARP inhibitors to avoid the hematological toxicity associated with PARP2 inhibition [125]. These toxicities may result from a PARP2-induced reverse allostery mechanism, as most clinical PARPis cause type-I pro-retention effects in PARP2 (Table 1) [123].

To untangle the mechanisms behind PARPi-induced PARP1 retention (i.e. the extended persistence of PARP1 in DNA damage foci) on chromatin in cells, Kanev et al. [52] developed mathematical models based on quantitative live-cell imaging of PARP1 following laser micro-irradiation, FRAP and imaging the extent of PAR produced. The models combine two distinct mechanisms that contribute to PARP1 chromatin retention: (1) the reduced exchange of PARP1 on damaged chromatin (allosteric trapping) as measured by FRAP, and (2) the increased apparent retention of PARP1 due to catalytic inhibition, which leads to repeated ‘unproductive’ cycles of DNA association and dissociation. PARP1 molecules retained by veliparib, a type-III inhibitor without allosteric trapping effects in vitro [94], showed increased retention in live-cell imaging, yet no modulation of exchange [52]. Compared with these PARPis, inhibitors that decreased PARP1 exchange rates (i.e. increased allosteric trapping), such as niraparib and talazoparib, displayed enhanced cytotoxicity. Increased PARP1 retention was shown to lead to delayed recruitment of various DNA damage repair factors, including RFC4, PCNA, POLD2, and POLH [52].

Taken together, PARPi-induced PARP1/2 retention at damaged sites occurs through multiple mechanisms and is much enhanced by allosteric trapping. Increased PARP1 retention could delay DNA damage repair factor recruitment and repair, leading to toxicity in repair-deficient tumour cells [52]. Thus, considering allostery-induced PARP trapping will be critical to elucidating and optimising the next generation of PARPis.

Given tankyrase's control of disease-relevant cellular processes, numerous academic and commercial small-molecule inhibitor programmes have been initiated. Disease links include various cancer types, primarily colorectal cancer [22], diabetes and obesity [126–128], neurodegeneration and brain injury [129–133], fibrosis [134] and viral infections [135]. For a systematic overview of recent developments, we refer the reader to a recent comprehensive review article [22]. Despite long-standing drug discovery and development efforts, thus far only three tankyrase inhibitors, two of which also target PARP1/2, found their way into clinical trials [22]. The limited pursuit of other inhibitors is primarily due to challenges posed by intestinal toxicity observed in pre-clinical studies, at least for earlier inhibitors [136,137]. The mechanism of toxicity remains poorly defined but may reflect on-target effects either involving inhibition of Wnt/β-catenin signalling or other tankyrase-regulated processes. Interestingly, recent inhibitor development suggests that inhibitor-induced toxicity can be overcome [138,139]. Tankyrase catalytic inhibition has complex consequences: it is unlikely to fully inhibit tankyrase function given that it turns tankyrase from a dynamic enzyme into an abundant ‘super-scaffold’, following the blockage of PARdU and stabilisation of filamentous polymers that could act as scaffolding hubs (Figure 7A) [11,44]. Therefore, tankyrase's accessory domains have come into the focus of small-molecule discovery efforts [140–144], given that both N-terminal ARCs and the polymerising SAM domain are required for tankyrase function [11,44].

Unlike catalytic inactivation, mutational disruption of the SAM:SAM interface or the TBM pockets in the ARCs fully inhibit tankyrase function in a β-catenin-dependent transcriptional reporter system [44]. Additionally, ARC-directed inhibitors are not predicted to result in the accumulation of tankyrase as NAD⁺ can still bind to the ART domain, triggering auto-PARylation and destabilising tankyrase through PARdU (Figure 7A). However, the SAM domain is challenging to target, as the SAM:SAM polymerisation interaction is mainly electrostatic in nature and involves a flat interface rather than a targetable pocket [44,141]. Conversely, since the initial proposal to target the ARCs, four ARC-binding small-molecules have been identified [140,141,143,144] (Table 2). This includes fragment screening, which identified molecules interacting with multiple ARCs of both tankyrase paralogues [140]. Two of these studies specifically focussed on disrupting the interaction between ARCs and ubiquitin-specific protease 25 (USP25), resulting in the identification of the ARC-binding small molecule C44 by virtual screening and subsequent validation [143], and the biosynthetically derived neoantimycin analogue UAT-B, initially identified on the basis of its selective cytotoxicity towards cancer cells and later linked to Wnt/β-catenin signalling and tankyrase [144,145]. USP25 is a deubiquitinating enzyme that binds tankyrase via a C-terminal TBM (RTPADG) and stabilises tankyrase through removal of ubiquitin [146]. Hence, the ability of ARC binders to block USP25 recruitment was proposed to destabilise tankyrase. Whilst both studies postulate that the observed decrease in β-catenin levels arises from inhibiting ARC:USP25 interactions, it is likely that inhibition of other tankyrase effectors such as AXIN1/2 also contributes as these compounds are predicted to bind to the TBM binding pocket shared by all known tankyrase effectors [71]. The TBM binding pocket is highly conserved across the four TBM-binding ARCs 1, 2, 4, and 5 from both TNKS and TNKS2 (Figure 7B), suggesting that pan-ARC inhibitors can be developed. Whilst X-ray crystal structures of catalytic inhibitors bound to the ART domain have guided the design of more selective and potent compounds [22], no direct structural insights into the binding modes have yet been obtained for ARC-directed small molecules. Obtaining such structures will undoubtedly accelerate the development of compounds into potent effector competitors in the future. Further to small-molecule studies, a peptidomimetic approach (Table 2) also provided proof of concept: after achieving sufficient cell permeability and stability, a macrocyclised peptide derived from an optimised TBM [71], inhibited a β-catenin-dependent reporter in HEK293 T cells [142].

Compared with tankyrase catalytic inhibitors, ARC-directed small molecules, although targeting the same protein, are likely to have distinct downstream consequences (Figure 7A). Treatment of APC-truncated colorectal epithelial cells with tankyrase catalytic inhibitors leads to the accumulation of cytoplasmic AXIN1/2 puncta, so called β-catenin degradasomes [117,147–151]. These structures are thought to represent microscopic correlates of the β-catenin destruction complex and are proposed sites of restored β-catenin turnover, which otherwise is impaired upon APC truncation [117,151]. However, silencing of both tankyrase paralogues, which like tankyrase catalytic inhibition increases AXIN1/2 levels [20], results in the loss of degradasomes [150]. This suggests that tankyrase is a critical scaffold of degradasomes, and predicts that inhibition of effector binding to ARCs, which drive scaffolding, may give rise to distinct biological effects as degradasomes may not be stabilised. Future work will establish the molecular and cellular consequences of effector competition by small molecules. Given the anticipated unique mechanism of inhibition, effector binding antagonists may open new opportunities for the development of better-tolerated tankyrase inhibitors.

The characteristic cytoplasmic puncta that accumulate upon tankyrase inhibition [117] may loosely be regarded as equivalents of PARP1 foci induced by type-I PARPi that induce ‘trapping’ on DNA (see above). In addition to stabilising tankyrase filaments, first by increasing the cellular concentration of tankyrase and second by abrogating the auto-PARylation-dependent filament disassembly, a third possibility, akin to ‘reverse allostery’ in PARP1, is suggested by structural studies [24,152]. TNKSi binding to the A-site would invariably induce an open D-loop conformation (Figure 7C). This may promote the formation of ART:ART head contacts, which in turn would stabilise tankyrase filaments (Figure 7D) [24]. If this were the case, TNKSis, similar to PARPis, may fall into different classes depending on their ability to ‘trap’ tankyrase in its polymeric form. Superposition of tankyrase ART domain crystal structures, selected for the absence of head-like crystal contacts constraining D-loop conformation, revealed that 93% displayed a closed D-loop base, indicative of the proposed basal (inactive) state where NAD⁺ is thought to be excluded from the active site [24]. 91% of the closed D-loop structures contained small molecules bound to the N-site, yet none contained an A-site or a dual-site inhibitor, indicating the closed D-loop base conformation is compatible with the binding of an N-site inhibitor (Figure 7C). Of the ART structures in an open D-loop base conformation, 72% presented an A-site or a dual-site inhibitor, strongly suggesting that occupation of the A-site by a small molecule can force open the D-loop base [24]. For structures with an open D-loop and an N-site inhibitor, electron densities were often observed in the A-site and could primarily be attributed to buffer components [24]. We propose that prying the D-loop base open by an A-site binder could facilitate ART:ART head interactions in the context of the tankyrase filament, and thereby increase polymer stability and avidity-dependent effector recruitment. A metabolic increase in NAD⁺ levels in insulin-secreting β-cells was reported to promote tankyrase (but not PARP1) auto-PARylation [153], We speculate that NAD⁺ co-substrate-induced conformational changes in the D-loop may contribute to tankyrase activation. These hypotheses will require rigorous testing through biophysical techniques at tankyrase concentrations that support robust polymerisation.

Most ADP-ribosylation events in the DNA damage response occur on serine residues [97,111]. The PARP1/2 accessory factor HPF1 is a critical determinant of target selection and modification [104]. In addition to directing ADP-ribosylation towards histones rather than PARP1/2 itself and eliciting a switch of the acceptor amino acid from aspartate/glutamate to serine (see above), HPF1 limits PARylation processivity, giving rise to mono-ADP-ribosylation [104]. The crystal structure of human HPF1 bound to the PARP2 catalytic domain in complex with the NAD⁺ analogue EB-47 [154] provided the structural basis for these observations. NMR spectroscopy and structure-informed mutagenesis revealed that PARP1 and HPF1 form a similar complex. This was later confirmed by a crystal structure of the human PARP1 ART:HPF1 complex [155]. PARP1/2-dependent ADP-ribosylation on aspartate and glutamate side chains can proceed without the need for an extrinsic nucleophilic residue as the acidic acceptor side chain itself would be de-protonated at neutral pH [154]. Serine ADP-ribosylation, however, requires a catalytic nucleophile, equivalent to the catalytic base in protein kinases, which also modify hydroxyl groups [156]. Intriguingly, this nucleophile is provided in the form of a ‘glutamate finger’ within HPF1, which complements the catalytic site of PARP1/2 [154,155]. This model finds support in the cholera-toxin like ARTs that require two conserved catalytic glutamates to ADP-ribosylate arginine residues in target proteins [154]. The catalytic dyad strikingly aligns with the composite active site formed by the PARP1/2:HPF1 complex. Interestingly, the equivalent of a histidine residue (H381^PARP2) previously suggested to bind the acceptor ADP-ribose in the PAR elongation cycle [28] interacts with HPF1, which would render the PARP1/2:HPF1 complex incapable of extending mono-ADP-ribose to form PAR chains [154]. Rudolph and colleagues further showed that HPF1 increases NAD⁺ hydrolysis by PARP1 and proposed that once available ADP-ribose acceptor sites are depleted, water is used as nucleophile instead [38].

No such accessory factor has yet been identified for tankyrase. Although the PARP domain surface equivalent to the HPF1 binding site remains accessible in the tankyrase filament [24] and is highly conserved (Figure 8A–D), the infrastructure required for HPF1 binding in PARP1/2 does not appear sufficiently conserved in tankyrase to accommodate HPF1 (Figure 8E). Therefore, it is unlikely that HPF1 will regulate tankyrase, in line with experimental observations [85]. This does not preclude the potential existence of alternative factors functionally equivalent to HPF1 that could act in conjunction with tankyrase. Current in vitro evidence for tankyrase substrate specificity points to glutamate and aspartate as the primary targets for ADP-ribosylation [85,118]. Whether these represent the main residues targeted in vivo, where potential accessory factors may be present, requires further investigation.

Extensive functional and structural studies have revealed a comprehensive picture of PARP1 regulation, encompassing autoinhibition, subcellular targeting, activation, negative feedback regulation, and the control of substrate and ADP-ribose acceptor site selection. For some of these aspects, similar regulatory principles are emerging for tankyrase (Table 3). Despite recent structure-function insights into the basis of polymerisation-induced regulation of tankyrase, the fundamental question of what triggers tankyrase activation remains unanswered. Similarly, to understand auto-PARylation-dependent negative feedback regulation in tankyrase, and to appreciate the mechanistic basis of substrate ADP-ribosylation in various systems, we need to know the precise sites modified by tankyrase. Tankyrase's substrate binding modules have been investigated in isolation, but how they fit into the picture of full-length tankyrase in its inactive or active state, and whether they serve regulatory roles beyond effector recruitment, remains to be studied. Whilst PARP1 inhibitors have been successful in the clinic, numerous advanced tankyrase inhibitors displayed intestinal toxicity, which led to the discontinuation of some tankyrase inhibitor programmes. However, toxicity was not observed for all inhibitors, despite their potency. Understanding the mechanism of toxicity, and further exploring alternative modalities of tankyrase inhibition, such as harnessing potential reverse allostery or developing effector binding antagonists, will be important future pursuits. The confident use of any potential tankyrase inhibitor, either as a tool compound or a future drug molecule, will require a deep mechanistic understanding of tankyrase and its biological roles.

The authors declare that there are no competing interests associated with the manuscript.

Sebastian Guettler: Conceptualisation, Supervision, Formal analysis, Visualisation, Writing — original draft, Writing — review & editing, Funding acquisition. Matthew Jessop: Formal analysis, Investigation, Visualisation, Writing — original draft, Writing — review & editing. Benjamin J. Broadway: Formal analysis, Visualisation, Writing — original draft, Writing — review & editing. Katy Miller: Formal analysis, Visualisation, Writing — review & editing.

We thank members of the Guettler group for stimulating discussions and Dragomir Krastev for feedback. Work in S.G.’s laboratory was funded by a Cancer Research UK Programme Foundation Award (C47521/A28286), a Wellcome Trust Investigator Award (214311/Z/18/Z), and by The Lister Institute of Preventive Medicine through a Lister Institute Research Prize Fellowship. B.J.B. and K.M. were supported by MRC iCASE studentships.

AMOT

angiomotin

ARC

ankyrin repeat cluster

ARH3

ADP-ribosyl hydrolase 3

ART

ADP-ribosyltransferase

AUC

analytical ultracentrifugation

AXIN1

axis inhibition protein 1

BABAM1

BRISC and BRCA1-A complex member 1

BAD

benzamide adenine dinucleotide

BLZF1

basic leucine zipper nuclear factor 1

BRISC

BRCC36-containing isopeptidase complex

BRCT

BRCA1 C-terminal

DSB

double-strand break

EM

electron microscopy

FAT4

protocadherin Fat 4

FRAP

fluorescence recovery after photobleaching

FRET

Förster resonance energy transfer

FUS

RNA-binding protein FUS (“fused in sarcoma”)

GMDS/GMD

GDP-mannose 4,6 dehydratase

HD

helical domain

HDX-MS

hydrogen-deuterium exchange coupled to mass spectrometry

HPF1

histone PARylation factor 1

HR

homologous recombination

IRAP

insulin-responsive aminopeptidase

LNPEP

leucyl-cystinyl aminopeptidase

MALS

multi-angle light scattering

MAR

mono-ADP-ribose

MERIT40

mediator of RAP80 interactions and targeting subunit of 40 kDa

NAD⁺

nicotinamide adenine dinucleotide

NOTCH2

neurogenic locus notch homolog protein 2

NUMA1

nuclear mitotic apparatus protein 1

PAE

predicted aligned error

PAR

poly-ADP-ribose

PARdU

PAR-dependent ubiquitylation

PARG

poly(ADP-ribose) glycohydrolase

PARPi

PARP inhibitor

PCNA

proliferating cell nuclear antigen

PDB

Protein Data Bank

PEX14

peroxisomal membrane protein PEX14

POLD2

DNA polymerase delta subunit 2

POLH

DNA polymerase eta

PTM

post-translational modification

RCF4

replication factor C subunit 4

RING

really interesting new gene

RNF

RING finger nuclear factor

SAM

sterile alpha motif

SEC

size exclusion chromatography

SPR

surface plasmon resonance

TARG1

terminal ADP-ribose glycohydrolase 1

TBM

tankyrase binding motif

TNKSi

tankyrase inhibitor

TRF1

telomeric repeat-binding factor 1

UIM

ubiquitin-interacting motif

USP25

ubiquitin-specific protease 25

WGR

Trp-Gly-Arg

Zn

zinc finger