HPF1 remodels PARP1 active site for ADP-ribosylation of histones on serine

Upon binding to DNA breaks, poly(ADP-ribose) polymerase 1 (PARP1) ADP-ribosylates itself and other factors to initiate DNA repair. Serine is the major residue for ADP-ribosylation upon DNA damage, which strictly depends on HPF1. Here we report the crystal structures of human HPF1/PARP1-CAT ΔHD complex at 1.98 Å resolution and mouse and human HPF1 at 1.71 Å and 1.57 Å resolution, respectively. These structures and mutagenesis data conrm that the structural insights obtained in the HPF1/PARP2 study apply to PARP1. Moreover, we quantitatively characterize the key residues for HPF1/PARP1 binding. This information claries the confusion regarding HPF1/PARP1 assembly and can direct following functional studies. Our data show that through salt-bridging to Glu284/Asp286, Arg239 positions Glu284 for catalyzing serine ADP-ribosylation, maintains the local conformation of HPF1 to limit PARP1 automodication and facilitates HPF1/PARP1 binding through neutralizing the negative charge of Glu284. These ndings and the high-resolution structural data may facilitate drug discovery targeting PARP1. catalyze ADP-ribosylation of a serine by promoting deprotonation of its sidechain hydroxyl.


Introduction
Covalently attaching the ADP-ribose moiety from NAD + to substrate molecules (ADP-ribosylation, ADPr) is a reversible post-translational modi cation involved in various biological processes including DNA damage response (DDR), DNA replication, transcription, chromatin modulation, host-pathogen interactions, RNA metabolism, unfolded protein response and mitosis 1,2,3,4 . The poly(ADP-ribose) polymerases (PARPs) family, also known as ADP-ribosyltransferase diphtheria toxin-like proteins (ARTDs), are the major enzymes catalyzing ADPr in eukaryotes 5 . Although 17 PARPs family members have been discovered, 80-90% of cellular NAD + consumption during DDR is related to the founding member PARP1, which plays an essential role in DNA damage repair 6,7 . When DNA damage occurs, PARP1 can sense and attach to DNA chain breaks in milliseconds, which in turn activates its ADP-ribosyl transferase activity for auto-modi cation or modi cation of other factors including histones. The electronegative modi cations further recruit other DDR machines to launch DNA repair. Due to its extensive and pivotal role in DDR, PARP1 has emerged to be a promising target in cancer treatment 8,9,10,11,12 . This is a highly active eld of research, because of PARP1's relevance in many tumor types. Since 2014, several PARP1 inhibitors (such as olaparib and rucaparib) have been approved for the treatment of ovarian cancer, prostate cancer and breast cancer, based on the synthetic lethality strategy, and more inhibitors are being developed 13,14,15 .
PARP1 consists of six domains 16,17 (Supplemental Figure S1). Three zinc nger domains (Zn1, Zn2 and Zn3) are responsible for recognizing and binding to DNA breaks. The auto-modi cation domain (AD) contains a BRCA-C-terminus (BRCT) fold and bears multiple poly(ADP-ribosylation) sites. The Trp-Gly-Arg (WGR) domain interacts with Zn1, Zn3, CAT and DNA and is essential for coupling DNA binding to ADPribosyl transferase activation. The catalytic domain (CAT) contains two subdomains, HD and ART: The helical domain (HD) acts as an autoinhibitory domain in the folded state and undergoes local unfolding upon PARP1 activation to enable NAD + binding and enzyme activation, while the ART domain catalyzes the transfer of ADP-ribose and is highly conserved in other ADP-ribosyl transferases (ARTs). In the last decade, several studies elegantly showed that an inter-domain interaction network is induced by the binding of DNA breaks to the zinc nger domains, which triggers conformational changes of the catalytic domain, especially the local unfolding of HD followed by complete exposure of the NAD + binding pocket in the catalytic core and PARP1 activation 18,19,20,21,22 .
Besides the understanding of the intramolecular activation mechanism of PARP1, it is also necessary to clarify how other factors modulate PARP1 enzymatic output. A previously uncharacterized protein C4orf27 was demonstrated to be a cofactor of PARP1, which physically interacts with PARP1, restricts PARP1 hyper-automodi cation, promotes histone modi cation and changes the amino-acid speci city of ADP-ribosylation from aspartate/glutamate to serine 23,24,25 . This protein was therefore named histone PARylation factor 1 (HPF1). Soon afterwards, serine residue was proved to be the major target of PARP1 upon DDR 26 . These studies implied an important role played by HPF1 to regulate PARP1 function.
Recently, Suskiewicz et al. determined the crystal structures of Nematostella vectensis and Homo sapiens HPF1 both at 2.09 Å resolution, and Homo sapiens HPF1/PARP2-CAT ΔHD complex at 2.96 Å resolution, which provided important insights into how HPF1 binding to PARP2 and promoting serine ADPr 27 . However, a high resolution structure of HPF1 in complex with PARP1, the most important member of PARPs was unavailable, and the resolution of the reported structures, especially HPF1/PARP2-CAT ΔHD complex, were modest. There also remain some unresolved questions related to the assembly and function of the complex. For example, why is Arg239 indispensable for the interaction between HPF1 and PARPs but this residue does not interact with any residue of PARPs? What functional roles does the conserved HPF1 Arg239 residue play? We have determined the crystal structures of mouse and human HPF1, and human HPF1/PARP1 complex at 1.71 Å, 1.57 Å and 1.98 Å resolution, respectively, and carefully studied the function of the key residues participating in HPF1/PARP1 interaction through extensive site-directed mutagenesis, ADPr activity assays, isothermal titration calorimetry (ITC) and mass spectrometry. Our work sheds light on the so far obscure role of Arg239 in regulating complex assembly and function, and the high resolution HPF1/PARP1 complex structure will facilitate future drug design on this important target.

Results
HPF1 binds to activated ART domain of PARP1 HPF1 has been shown to bind to the CAT domain (HD-ART) of PARP1 in cells in response to DNA damage 23 . Since HD is an autoinhibitory domain that blocks productive binding to NAD + in the resting state, and it undergoes local unfolding to enable NAD + binding and ADP-ribose transferase activity when PARP1 binds to DNA breaks 19 , we asked if unfolding or removing the HD subdomain is the prerequisite for HPF1 binding to PARP1. To answer this question, we tested HPF1 binding to PARP1 CAT domain in the fulllength autoinhibited form or in the constitutively activated form through removing the majority (residues 679-786) of the autoinhibitory domain HD (CAT ΔHD) using size-exclusion chromatography (SEC).
Interestingly, it is observed that CAT ΔHD, but not full-length CAT, forms a complex with HPF1 ( Figure 1a). The interaction was further quantitatively characterized by isothermal titration calorimetry (ITC), which showed that CAT ΔHD binds to HPF1 with a dissociation constant (K d ) of about 1.5μM and 1:1 stoichiometry (Figure 1b, Table 1, Supplemental Figure S2,S3), while full-length CAT showed no binding to HPF1 (Figure 1c). Since these observations con rmed that HPF1 binds to activated CAT, we also tested the binding of HPF1 to full-length PARP1 activated by DNA (Table 1, Supplemental Figure S2). HPF1 binds to full-length PARP1 with a K d of about 2.8μM, slightly weaker than its binding a nity for CAT ΔHD, probably related to transient folding of the HD domain in solution.
The human HPF1/PARP1-CAT ΔHD Complex crystal structure In order to understand the assembly and function of the HPF1/PARP1 complex, the crystal structure of human HPF1/PARP1-CAT ΔHD complex bound by benzamide (introduced when expressing the protein) was determined and re ned to 1.98 Å (Figure 2a, Supplemental Figure S4 and Table S1). HPF1 and PARP1 were con rmed to form a 1:1 complex since there were one HPF1 and one PARP1-CAT ΔHD molecule in the asymmetric unit of the complex crystal structure.
The high resolution structure of the HPF1/PARP1-CAT ΔHD complex reveals an extensive interface between the two proteins that envelopes the region of the PARP1 active site (see interface I in Figure 2a). The hetero-dimer binding interface was con rmed through extensive mutagenesis combined with ADPribosylation assay and ITC assay. It has been well established that HPF1 binding to PARP1 restricts hyper-automodi cation of PARP1 23,26 . Indeed, hyper-automodi cation of PARP1 showed up clearly on SDS-PAGE as smeared bands above the un-modi ed PARP1 band, while wild-type HPF1 binding abolished this effect (Figure 2b, see lanes 3 and 4). On the designated hetero-dimer interface, HPF1 Phe268, Phe280, Asp283, Cys285 and Lys307 are found to directly interact with PARP1 residues ( Figure  2a, top-right and bottom-right insets), and these residues are conserved in HPF1 from different species (Supplemental Figure S5). Indeed, mutating most of these HPF1 residues (F268S, F280A, D283H, C285H and K307S) signi cantly restored hyper-automodi cation of PARP1 (Figure 2b, see lanes 9, 11, 12, 14 and 15), indicating loss of binding between the HPF1 mutants and PARP1. ITC assays further con rmed that these mutations dramatically reduced the binding a nity between HPF1 and PARP1-CAT ΔHD (Table 1, Supplemental Figure S2, S3). Importantly, HPF1 F268S and D283H mutations completely abolished the interaction between these two proteins, which may be used in following functional studies on the HPF1/PARP1 system. Interestingly, although PARP1, PARP2 and PARP3 share pretty high sequence homology in the CAT domain (25.41% of the sequences are identical and the other 18.85% are strongly similar), PARP3 lacks the key residues corresponding to His826, Leu985, Ser1012, Leu1013 and Trp1014 in PARP1, which according to the HPF1/PARP1-CAT ΔHD complex structure and the mutagenesis data are important for HPF1-PARP interaction, while these key residues are largely conserved between PARP1 and PARP2 (Figure 2c, Supplemental Figure S6). This explains why HPF1 can bind to PARP1 and PARP2, but not to PARP3 23 .
In examining the structure, we noted another interface between a different HPF1 and PARP1 molecule in the crystal lattice (see interface II in Figure 2a, top-left inset). However, mutating key residues on this interface (Glu138, Phe139 and Lys216) did not interfere with the binding (Figure 2b, see lanes 5, 6, 7; Table 1, Supplemental Figure S2, S3), and these residues are not conserved in HPF1 (Supplemental Figure S5). Thus this contact does not appear to be relevant to the catalytic function of the HPF1/PARP1 complex.
We also determined structures of human and mouse HPF1 alone, and that the conformation does not change upon formation of the complex with PARP1 (see Supplemental Figure S7). Notably, analysis of the surface electrostatic potential showed that positively and negatively charged areas are evenly distributed on the surface of this protein, except for part of helix α9/α10 and the linker loops connecting helices α6 and α7, where the protein surface is dominantly negatively charged (Supplemental Figure  S7B).

HPF1 binding remodels the PARP1 active site for histone binding and serine ADPr
Remarkably, our functional analysis of the structure shows that HPF1 and PARP1 form a composite active site, with HPF1 contributing key active site residues.
We rst analyzed the surface electrostatic potential of HPF1, PARP1-CAT ΔHD and the complex ( Figure   3a). Though PARP1 has been shown to play a key role in DDR involving the modi cation of histones 16,28 , the surface of ART subdomain, especially the active site, is positively charged, which is clearly incompatible with the positively charged nucleosome (Supplemental Figure S8). Interestingly, HPF1 binding to PARP1-CAT dramatically changes the situation. As is mentioned above, HPF1 surface covering helix α9, α10 and the loops connecting helices α6 and α7 is negatively charged (Supplemental Figure S7). Many solvent-exposed acidic amino-acid residues are located in this region, including Asp235, Glu240, Glu243, Asp245, Glu273, Asp283, Glu284, Asp286 and Glu292, and Arg239 turns out to be the only basic residue in this region ( Figure 3a). Notably, Arg239 and most of these acidic residues are conserved in HPF1 from different species (Supplemental Figure S5). When binding to PARP1, this negatively charged region of HPF1 merges to the active site of PARP1-ART and changes the surface potential of the latter, creating an overall negatively charged joint active site that would accommodate the access of the positively charged histones (Figure 3a). The surface of the HD subdomain facing the active site is also negatively charged due to three acidic residues Glu756, Asp766 and Asp770 and lack of any basic residue here, hence seems to form a negatively charged narrow tunnel with the joint active site ( Figure   3b). However, the HD subdomain is believed to undergo partial unfolding upon PARP1 activation 19 , and therefore it is unknown whether the tunnel still maintains. We speculate that if the HD subdomain does not completely unfold, the negatively charged tunnel may still maintain (but likely undergoes some conformational changes) that is presumably most accessible for positively charged extended peptide substrate (say histone H3 N-terminal tail), while the folded and/or negatively charged proteins may not readily access this joint active site to get ADP-ribosylated.
A benzamide molecule was found to bind in the active center of the HPF1/PARP1-CAT ΔHD complex structure (Supplemental Figure S4). When superimposed with the previously reported PARP1-CAT ΔHD/BAD complex crystal structure (PDB 6BHV) 22 , the benzamide molecule in our HPF1/PARP1-CAT ΔHD complex structure well overlaps the benzamide moiety of BAD, the NAD + analog (Figure 3c). Importantly, we noted that the HPF1 Glu284 side-chain carboxyl, being positioned/locked by the Arg239-Glu284 salt-bridge, is located beside the ribose of the ADP moiety of carba-NAD + (representing the acceptor for ADP-ribosylation 29 , see Figure 3d), and is about 1.9Å farther away than the latter from the C1" atom of the nicotinamide ribose moiety of BAD/NAD + (representing the donor for ADP-ribosylation, see Figure 3c), which from the structural biological view would strongly suggest an important functional role of this residue rather as a catalyzer for ADP-ribose transfer to a third party (say a serine residue) than as a direct acceptor itself for ADP-ribose. In NAD + , the positively charged nicotinamide nitrogen trends to deprive an electron from the connecting ribose C1" carbon, rendering the latter an oxacarbenium ion that would then be attacked by a nucleophile such as a negatively charged/deprotonated amino-acid sidechain. Since acidic amino-acids glutamate and aspartate may spontaneously lose a proton to become negatively charged at physiological pH, these residues can be readily ADP-ribosylated once they access the active center of ART (for example, in the absence of HPF1, see Figure 3e). However, in the HPF1/PARP1-CAT ΔHD complex structure the Glu284 side-chain carboxyl is too far away (~4.6Å) from the NAD + to attack the ribose C1" carbon ( Figure 3c). Therefore it seems unlikely that Glu284 itself can get ADP-ribosylated. On the other hand, since the non-acidic serine residue does not spontaneously lose a proton at physiological pH, it must be activated (deprotonated) by a third party prior to ADP-ribosylation.
The Glu284 side-chain carboxyl may ful ll this task when it becomes negatively charged ( Figure 3f Why does HPF1 binding hinder PARP1 automodi cation and poly-ADP-ribosylation? Based on the veri ed interaction mode between HPF1 and PARP1-CAT ΔHD (corresponding to interface I shown in Figure 2a), an overall model of HPF1 binding to full-length activated PARP1 except the Zn2 and AD domains is proposed through overlapping the PARP1-ART domain with the corresponding domain in the PARP1/DNA complex structure (PDB 4DQY) (Figure 4a) 18 . HPF1 binds to ART at the active site of the enzyme, while in the meanwhile it is well accommodated with other domains of PARP1 except for HD that undergoes local unfolding upon PARP1 activation. This model immediately hints why HPF1 binding restricts automodi cation of PARP1. According to the previous model proposed by Langelier et al., the AD/BRCT domain carrying most of the automodi cation sites is located close to the CAT domain presumably for being ADP-ribosylated 18 (Supplemental Figure S1). This location is roughly the same place where HPF1 binds (compare Figure 4a to Supplemental Figure S1). We therefore speculate that HPF1 binding would dislodge AD from this location, hence limits the automodi cation of PARP1, at least in the folded BRCT domain.
The complex structure hints that HPF1 not only restricts hyper-automodi cation of PARP1, but may limit the modi cation from poly-ADP-ribosylation to mono-ADP-ribosylation. An early study by Ruf et al. observed the ADP moiety of carba-NAD + binding in the active site, and this ADP moiety was believed to represent the acceptor for the next ADP-ribose during poly-ADP-ribose chain elongation and branching 29 . Comparing to the HPF1/PARP1-CAT ΔHD complex structure reported here, we noted that the ADP (i.e. the acceptor for next ADP-ribose in poly-ADP-ribosylation) binding site is partly occupied by HPF1 residues Asp283 and Glu284 (Figure 3d), which would implicate that HPF1 binding is likely to prevent poly-ADPribose chain elongation and branching. However, we speculate that poly-ADP-ribosylation may still take place, likely at lower probability, due to the relatively weak and thus transient binding between HPF1 and PARP1. This explains the long smeared bands observed in our ADP-ribosylation assay (Figure 2b, see histone H3/H4 ADPr ).
The unique role of HPF1 Arg239 Arg239 is highly conserved in HPF1 (Supplemental Figure S5), and has been thought to be important for the interaction between HPF1 and PARPs. Mutating this residue to alanine (R239A) did restore automodi cation of PARP1 (Figure 2b, lane 8), which was previously interpreted to be loss of binding between HPF1 and PARP1 23 .
However, in the human HPF1/PARP1-CAT ΔHD and HPF1/PARP2-CAT ΔHD complex structures, HPF1 Arg239 was found to only interact with Glu284 and Asp286, but not with any residue from PARP1/2 ( Figure 4b). Furthermore, ITC assays con rmed that this mutation only mildly weakens the binding a nity between HPF1 and PARP1, indicating that without Arg239 HPF1 still binds pretty well to PARP1 (Table 1). Curiously, in the ADP-ribosylation assays, the band of HPF1 R239A, but not of any other HPF1 mutants, clearly shifted on the SDS-PAGE (Figure 2b, lane 8, black arrow), which looks very likely that HPF1 R239A itself was ADP-ribosylated. We then analyzed the shifted HPF1 R239A band by Mass Spectrometry, which con rmed that HPF1 R239A did get ADP-ribosylated on Asp235 and Glu240, two acidic residues located on both sides of residue 239 in the long loop connecting helices α6/α7 (Figure 4b,  4c). The long loop regions are often di cult to resolve in crystal structure due to their intrinsic exibility, but in HPF1 the long loop connecting helices α6 and α7 (residues 217-244) is ordered and well resolved. As is shown in Figure 4b, the C-terminal half of this long loop is stabilized by two groups of interactions: one is the salt-bridge network including Arg239-Asp286 and Arg239-Glu284, which also acts to position the catalytically important Glu284 side-chain; the other is the hydrophobic interaction between Tyr238 phenyl and Val218 side-chain, and hydrogen bond between Tyr238 hydroxyl and Glu292 carboxyl. It is inferred that when Arg239 was mutated to alanine that lost the interlocking with Glu284 and Asp286, the loop became so exible that Asp235 and Glu240 gained the freedom to attack NAD + in the active center and became ADP-ribosylated. Importantly, although HPF1 R239A still bound PARP1 pretty well, it also restored the automodi cation of PARP1 (Figure 2b, lane 8). The PARP1 automodi cation domain (AD) contains a folded subdomain BRCT (residues 380-480) and an unstructured automodi cation peptide (residues 481-530). In a previous in vitro ADP-ribosylation assay, the serine residues in the Lys-Ser motif within the automodi cation peptide (Ser499, Ser507and Ser519), but not that in the BRCT subdomain (Lys467-Ser468), were ADP-ribosylated 24 in the presence of HPF1, indicating that the Lys-Ser motif in a exible peptide, but not in a folded domain may access the remodeled HPF1/PARP1 active center under certain circumstances. In our assays, HPF1 WT binding robustly limited automodi cation of PARP1, while R239A, although still bound to PARP1, restored the automodi cation. Taken together, our data indicate that the rigidity of the loop region strengthened by Arg239-Glu284/Arg239-Asp286 interactions is essential for limiting PARP1 automodi cation in the unstructured automodi cation peptide region.
In the ADP ribosylation assay we also noted that the HPF1 E284A band mysteriously became weaker (Figure 2b, lane 13, gray arrow), indicating losing a small amount of the un-modi ed protein (or ADPribosylation of a small portion of the protein). However, the modi cation of this mutant HPF1 protein must be very minor since no shifted band was obviously seen on the gel. We speculate that in this mutant, after losing the Arg239-Glu284 salt-bridge but still maintaining the Arg239-Asp286 salt-bridge, the loop region gained some exibility but the acidic residues (say Asp239 and Glu240) did not gain enough freedom to attack NAD + to be ADP-ribosylated. Accordingly, HPF1 E284A only mildly restored PARP1auto-modi cation, which again supported the notion that the rigidity of this loop region strengthened by Arg239-Glu284/Arg239-Asp286 interactions is essential for limiting PARP1 automodi cation.
The above observations do not explain why mutating Arg239 to alanine weakens the interaction between HPF1 andPARP1, since no residue in the loop region (including Arg239 itself) directly interacts with PARP1. We noted that the HPF1 E284A mutation showed a very unique behavior in our assays. It is the only mutation that enhanced the binding between HPF1 and PARP1 (Table 1). Taken into account the negative electrostatic property of the joint active site discussed above, we propose that the existence of the negatively charged Glu284 in the active center is not favorable for HPF1/PARP1 binding, while the salt-bridge with the positively charged Arg239 side-chain may act to neutralize the negative charge on Glu284 carboxyl, thus facilitate the binding. Mutating Arg239 to directly break Arg239-Glu284 interaction, or mutating Tyr238 to destroy the local conformation supporting Arg239-Glu284 interaction, would restore the negative charge of Glu284, thus weaken HPF1/PARP1 binding. This explains why HPF1 R239A (and/or Y238A, as was found in the previous study 23 ) mutation weakens the binding between the two proteins, and indicates that although mutating Arg239 (and/or Tyr238) may apparently weaken HPF1/PARP1 binding, they may not be the right choices to act as the non-binding control in functional studies because the mutation(s) lead to much more complicated consequences rather than simply abolishing the binding -and, they do not really abolish the binding.
Taken together, our study indicates that Arg239 is a key residue acting to 1) position Glu284 at the right location for catalyzing serine ADP-ribosylation; 2) stabilize the conformation of the long loop region across Arg239 to limit auto-modi cation of PARP1; 3) facilitate HPF1/PARP1 binding through neutralizing the negative charge of Glu284 side-chain.

Discussion
The sea anemone and human HPF1 (both at 2.09 Å resolution) and human HPF1/PARP2-CAT ΔHD complex (at 2.96 Å resolution) structures determined by Suskiewicz et al. provided the rst glance at the interaction between HPF1 and a PARP family member. This work depicted the mechanism about how this binding completes the PARP2 active site and promotes ADP-ribosylation of histone serine, which were believed to also apply to PARP1 based on the NMR and functional studies 27 . Our studies independently determined at higher resolution the crystal structures of mouse and human HPF1 (at 1.71 Å and 1.57 Å resolution, respectively) and human HPF1/PARP1-CAT ΔHD complex (at 1.98 Å resolution), which provided direct structural insights into how HPF1 binds to PARP1, why HPF1 prefers binding to PARP1/2 but not PARP3, why this interaction limits hyper-automodi cation of PARP1/2 and may also limit poly-ADP-ribosylation. The HPF1/PARP1-CAT ΔHD complex also provided insights into why HPF1 binding promotes histone peptide recognition and serine ADP-ribosylation. Since PARP1 and PARP2 share pretty high homology in the CAT domain (45.7% identity), the Suskiewicz et al. structures and ours look similar (Supplemental Figure S9), and our structural and functional studies con rmed most of the proposals made by Suskiewicz et al.
Besides con rming the key ndings by Suskiewicz et al., our work provides novel and fundamental insights into the structure-function relationships of the HPF1/PARP1 complex. Since there are always multiple contacting interfaces due to protein molecules packing in crystals, and the packing may distort the protein complex architecture, caution must be taken to carefully verify which observed interface in the complex crystal structure represents the true dimer interface in solution. As a common practice in crystallography, this is usually done by extensive mutagenesis combined with functional studies, which were absent in the previous study, but are ful lled in our study. When studying the function of HPF1 mutants to verify the true dimer interface, we not only conducted the traditional ADP-ribosylation assays (Fig. 2b), an indirect and non-quantitative way to ask if the HPF1/PARP1 complex is formed, but also carefully measured the K d values using ITC assays (Table 1). Our data quantitatively and directly described the binding behavior between HPF1 and PARP1, and showed that only certain mutations (F268S and D283H) can completely abolish the binding, which may be used in following functional studies, especially in the cell-based studies, as non-binding control. While if using other mutations, the experimental readout may be confusing or even misleading.
One such confusing and misleading result is related to HPF1 Arg239. In the previous studies, this residue was thought to mediate HPF1 and PARP1/2 interaction, but this idea is not supported by the structures. The same confusion applies to Tyr238, too, a residue located inside the HPF1 molecule. To resolve this problem, we rst asked if Arg239 interacts with some other PARP1 residue not presenting in our HPF1/PARP1-CAT ΔHD complex structure. Based on the structural model (Fig. 4a), the PARP1 HD subdomain seems to be the only domain that could directly interact with HPF1 Arg239, and Glu756 in the HD subdomain seems to be the residue most likely to form a salt-bridge with HPF1 Arg239. However, it has been proved that the HD subdomain hinders the binding of HPF1, and our mutagenesis studies on PARP1 E756A showed that this mutation had no effect on HPF1/PARP1 binding (Supplemental Figure  S2). On the other side, the ITC data con rmed that HPF1 R239A still binds to PARP1, and Mass Spectrometry con rmed that this mutant protein got ADP-ribosylated in the ADP-ribosylation assay. Taken into account the observation that HPF1 R239A signi cantly restored PARP1 automodi cation, we propose that Arg239 (and presumably also Tyr238) acts at the central stage to stabilize the local conformation of the α6-to-α7 loop region through interlocking with Glu284 and Asp286 (and through intramolecular interactions for Tyr238), which is essential for limiting PARP1 automodi cation, but dispensable for histone serine modi cation. To our surprise, HPF1 E284A mutant binds to PARP1-CAT ΔHD with about 3 times higher a nity than HPF1 WT, which hints that the acidic property of Glu284 is not favored for HPF1/PARP1 binding and that neutralizing the negative charge of Glu284 by Arg239 facilitates the binding. In accordance with the observation on HPF1 R239A that loses the Arg239-Glu284 and Arg239-Asp286 salt-bridges, the HPF1 E284A mutant, losing only the Arg239-Glu284 salt-bridge, only partly restored PARP1 automodi cation (Fig. 2b), which once again supports our proposal about the Arg239 function of stabilizing the local conformation to limit PARP1 automodi cation. Since the function of Arg239 is much more complicated than it was previously proposed, if one use this mutation as a control to abolish HPF1 and PARP1/2 binding in cell-based studies, the result may be misleading.
Taken together, our work provided the direct glance of the complex between HPF1 and PARP1, the most important member of the PARP family. The higher resolution structures, along with the extensive mutagenesis studies and quantitative functional study data in our work, completed the previous studies done by Suskiewicz  Since histone serine ADP-ribosylation is a key step in DDR 8,9,13,16,26 Figure S2 and S3 for raw data.
2 The assessment was made based on the ADP-ribosylation assays (see Figure 2b). Number of "+" indicates the relative intensity of observed ADP-ribosylation. Page (Table 1, Supplemental Figure S2). (c) Typical ITC data where human HPF1 was titrated into full-length PARP-1 CAT. No binding was detected in this assay.

Figure 2
Crystal structure of human HPF1/PARP1-CAT ΔHD and the binding mode between these two proteins. (a) Overall structure and the interaction details between human HPF1 and PARP1-CAT ΔHD. The overall structure of human HPF1/PARP1-CAT ΔHD was shown as cartoons. HPF1 and PARP1-CAT ΔHD are colored in cyan and orange, respectively. The gray-colored molecule on the left side indicates another PARP1-CAT ΔHD molecule from a neighboring crystallographic asymmetric unit that also contacts the HPF1 molecule shown here due to close packing of molecules in the crystal. Therefore, the complex crystal structure actually indicated two possible interaction modes between HPF1 and PARP1-CAT ΔHD.
The three insets show the key residues on the two interaction interfaces that mediate the interactions between the HPF1 and PARP1 in the crystal. The dashes indicate polar interactions between the key residues. (b) Mutagenesis/ADP ribosylation studies to verify which key residues indicated in crystal structure do mediate HPF1 and PARP1 binding in solution. Hyper-automodi cation of PARP1 (shown by the smeared PARP1ADPr bands on SDS-PAGE) was recognized as the sign of losing the interaction between HPF1 mutants and PARP1. Surprisingly, we noted that HPF1 R239A, but not any other HPF1 mutants, was robustly ADP-ribosylated in the assay (black arrow, and see Figure 4c), while E284A seemed to be mildly ADP-ribosylated -The HPF1 E284A band became weaker, indicating loss of un-modi ed protein in the assay (gray arrow). (c) Amino-acid sequence alignment of human PARP1/2/3. The veri ed key residues to mediate HPF1 binding in PARP1 in solution are indicated by red dots. The corresponding residues in PARP2 and PARP3, if they are conserved (same to those in PARP1), are indicated by green and blue dots, respectively. here) ribose. This distance is too far away for Glu284 itself to accept the ADP-ribose, but would work to catalyze ADP-ribosylation of a serine by promoting deprotonation of its sidechain hydroxyl.

Figure 4
Depicting the function of HPF1 Arg239. (a) Comparison of the HPF1/PAPR1-CAT ΔHD complex structure and the DNA-bound PARP1 crystal structure (PDB 4DQY) 18 by superimposing the ART domains from both (orange -ART determined in this report; green -ART of PDB 4DQY). HPF1 binding to PARP1 would presumably dislodge PARP1 automodi cation domain (AD) that was previously proposed to locate roughly in the same position as HPF1 binds (see Supplemental Figure S1), thus prevents automodi cation of PARP1. (b) A close view of the long α6-α7 loop region structure and its relative position to the active center of the HPF1/PARP1 complex. (c) Mass Spectrometry analysis revealed that HPF1 R239A mutation resulted ADP-ribosylation on Asp235 and Glu240, a unique phenomenon only seen for this mutant (see Figure 2B, black arrow). The mutated residue 239 is shown in blue in the sequence, while the modi ed residues Asp235 and Glu240 are shown in red.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.